Clinical Effectiveness of Integrase Strand Transfer Inhibitor-Based Antiretroviral Regimens in a Pooled Cohort of North American Adults

CLINICAL EFFECTIVENESS OF INTEGRASE STRAND TRANSFER INHIBITOR-BASED ANTIRETROVIRAL REGIMENS IN A POOLED COHORT OF NORTH AMERICAN ADULTS

Haidong Lu

A dissertation submitted to the faculty at the University of North Carolina at Chapel Hill in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the Department of Epidemiology in the Gillings School of Global Public Health.

Chapel Hill 2020

Approved by:

Stephen R. Cole

Daniel Westreich

Adaora A. Adimora

Joseph J Eron

Michael G. Hudgens

ABSTRACT

Haidong Lu: Clinical Effectiveness of Integrase Strand Transfer Inhibitor-based Antiretroviral Regimens in a Pooled Cohort of North American Adults (Under the direction of Stephen Cole)

Integrase strand transfer inhibitor (InSTI)-based regimens are now widely recommended as first-line antiretroviral therapy (ART) for adults with human immunodeficiency virus. But evidence on long-term clinical effectiveness of InSTI-based regimens remains limited.

Using prospective pooled cohort data from the North American AIDS Cohort

Collaboration on Research and Design (NA-ACCORD) during 2009-2016, we identified the participants who initiated their first ART regimen, containing either InSTI (i.e., raltegravir, dolutegravir, and elvitegravir/cobicstat) or efavirenz (EFV) as an active comparator, and estimated the long-term effects of InSTI-based regimens on virologic and clinical outcomes compared with the EFV-based regimen. We used inverse probability of treatment and censoring weights to account for baseline confounding, differential loss to follow-up and treatment changes, and estimated observational analogs to the intention-to-treat and per-protocol effects.

For the virologic outcomes (i.e., viral suppression), the intention-to-treat analysis showed

81.3% of the patients in the InSTI group and 67.3% of those in the EFV group experienced viral suppression at 0.25 years after ART initiation, corresponding with a difference of 14.0% (95%

CI: 12.4, 15.6). By 1 year after ART initiation, the proportion virally suppressed was 89.5% in the InSTI group and 90.2% in the EFV group, corresponding with a difference of -0.7% (95%

CI: -2.1, 0.8). At 7 years after ART initiation, the proportion virally suppressed was 94.5% in the

iii

InSTI group, and 92.5% in the EFV group, corresponding with a difference of 2.0% (95% CI: -

7.3, 11.3). The results of per-protocol analyses were similar to the intention-to-treat analyses.

For the clinical outcomes (i.e., composite outcomes of AIDS-defining illnesses, death, and serious non-AIDS events), during the 6-year follow-up, 440 of the InSTI group and 1,097 of the EFV group incurred the composite outcome. The 6-year intention-to-treat risks were 14.6% for the InSTI group and 14.3% for the EFV group, corresponding with a RD of 0.3 percentage point (95% CI: -2.7, 3.3), and HR was 1.08 (95% CI: 0.97, 1.19); the 6-year per-protocol risks were 12.2% for the InSTI group and 11.9% for the EFV group, corresponding with a RD was 0.3 percentage point (95% CI: -3.0, 3.7), and HR was 1.09 (95% CI: 0.96, 1.25).

Our studies showed that although InSTI-based initial ART regimens had a more rapid viral response compared with the EFV-based initial ART regimen, the long-term effects on virologic and clinical outcomes were similar. Along with the increasing studies showing the potential adverse effects associated with InSTI, our results on relatively similar long-term effects bring into question whether InSTI-based regimens should be favored over other regimens, especially in the current context that InSTI/dolutegravir-based regimens are being rolled out globally.

ACKNOWLEDGEMENTS

First, I thank to thank my wife, Qiwen Sheng. Without her, I would not have been able to come to the US and start the epidemiology journey.

I am grateful for having Dr. Stephen Cole as my advisor, whose broad knowledge, passion in mentoring, keen insight, and resources led me to the 4-year rewarding journey at

UNC. I feel honored to be his student and advisee.

I also thank my dissertation committee members for providing deep insights for my dissertation projects. Drs. Joseph Eron and Adaora Adimora gave me suggestions from their clinical perspective. And Drs. Daniel Westreich and Michael Hudgens provided me with statistical and methodological guidance.

I am extremely thankful to my parents and my sister for the valuable support and substantial sacrifice they made for my education. Without them, I would not have completed my graduate degrees.

I thank those who supported and helped me throughout this PhD journey.

I am also grateful for the epidemiology education I received these years in the US, transforming me from a person with no directions to a man with a mission.

I was fortunately supported by the funding from Drs. Daniel Westreich, Stephen Cole, and Alexander Keil during this 4-year program, as well as the ViiV Healthcare Fellowship that funded my dissertation projects. I thank my collaborators from the North American AIDS Cohort

Collaboration on Research and Design (NA-ACCORD) for giving me the opportunity to use the fantastic data, and providing comments on my manuscripts.

TABLE OF CONTENTS

LIST OF TABLES ...... ix

LIST OF FIGURES ...... x

LIST OF ABBREVIATIONS ...... xi

CHAPTER 1: INTRODUCTION ...... 1

1.1 Brief overview of human immunodeficiency virus...... 1

1.2 Antiretroviral therapy ...... 3

CHAPTER 2: STATEMENT OF SPECIFIC AIMS ...... 14

CHAPTER 3: METHODS ...... 16

3.1 Study design ...... 16

3.2 Study sample ...... 16

3.3 Preliminary data ...... 17

3.4 Measurements...... 18

Initiation of InSTI-based or EFV-based regimens ...... 18

Outcomes of interest ...... 18

Covariates ...... 19

3.5 Analysis plans ...... 20

3.6 Statistical power ...... 22

3.7 Interpretation of results ...... 23

CHAPTER 4: VIROLOGIC OUTCOMES AMONG ADULTS WITH HIV USING INTEGRASE STRAND TRANSFER INHIBITOR BASED ANTIRETROVIRAL REGIMENS: A COLLABORATION OF COHORT STUDIES IN THE UNITED STATES AND CANADA ...... 26

vii

4.1 Introduction ...... 26

4.2 Methods ...... 26

4.3 Results ...... 31

4.4 Discussion ...... 33

CHAPTER 5: CLINICAL EFFECTIVENESS OF INTEGRASE STRAND TRANSFER INHIBITOR-BASED ANTIRETROVIRAL REGIMENS AMONG ADULTS WITH HUMAN IMMUNODEFICIENCY VIRUS ...... 43

5.1 Introduction ...... 43

5.2 Methods ...... 44

5.3 Results ...... 49

5.4 Discussion ...... 51

CHAPTER 6: CONCLUSION ...... 64

6.1 Overview ...... 64

6.2 Study findings ...... 65

6.3 Strengths ...... 67

6.4 Limitations ...... 69

6.5 Future directions ...... 72

6.6 Public health significance and conclusion of our work...... 73

REFERENCES ...... 75

viii

LIST OF TABLES

Table 1.1: Advantages and disadvantages of currently available InSTIs...... 11

Table 1.2: Summary of randomized trials comparing InSTIs with EFV...... 13

Table 4.2: Characteristics at ART initiation of 15,318 HIV-infected adults initiating an InSTI-based regimen or an EFV-based regimen between July 2009 and December 2016 in the NA-ACCORD, overall and by treatment regimens, the United States and Canada...... 39

Table 5.1: Incident type-specific AIDS-defining illnesses among 288 adults who initiated an InSTI-based regimen and incurred AIDS and 671 adults who initiated an EFV-based regimen and incurred AIDS as their first composite event between July 2009 and December 2016 in the NA-ACCORD, the United States and Canada...... 56

Table 5.2: Missing baseline covariates among 15,993 HIV-infected adults initiating an InSTI-based regimen or an EFV-based regimen between July 2009 and December 2016 in the NA-ACCORD, the United States and Canada...... 58

Table 5.3: Characteristics at ART initiation of 15,993 HIV-infected adults initiating an InSTI-based regimen or an EFV-based regimen between July 2009 and December 2016 in the NA-ACCORD, overall and by treatment regimens, the United States and Canada...... 59

Table 5.4: Estimated 6-year risk differences and hazard ratios of the composite outcome comparing adults initiating an InSTI-based regimen with those initiating an EFV-based regimen between July 2009 and December 2016 in the NA-ACCORD, the United States and Canada...... 60

Table 5.5: Secondary analyses for the effect on composite outcome comparing adults initiating an InSTI-based regimen with those initiating an EFV-based regimen between July 2009 and December 2016 in the NA-ACCORD, the United States and Canada...... 61

LIST OF FIGURES

Figure 3.1: The North American AIDS Cohort Collaboration of Research and Design (NA-ACCORD) ...... 25

Figure 3.2: Statistical power based on risk difference comparing EFV-based regimens to InSTI-based regimens across different sample sizes...... 26

Figure 4.1. Secular trend in proportion of initiating InSTI- versus EFV-based regimens between July 2009 and December 2016 among 15,318 HIV-infected adults in the NA-ACCORD, the United States and Canada...... 40

Figure 5.1. Secular trend in initiation of InSTI- versus EFV-based regimens between July 2009 and December 2016 in the NA-ACCORD, the United States and Canada...... 62

LIST OF ABBREVIATIONS

AIDS Acquired immune deficiency syndrome

ART Antiretroviral therapy

BMI Body mass index

COBI Cobicistat

DTG Dolutegravir

EFV Efavirenz

EVG Elvitegravir

FTC Emtricitabine

HBV Hepatitis B virus

HCV Hepatitis C virus

HIV Human immunodeficiency virus

InSTI Integrase strand transfer inhibitor

IQR Interquartile range

NA-ACCORD The North American AIDS Cohort Collaboration on Research and Design

NNRTI Non-nucleoside reverse transcriptase inhibitor

NRTI Nucleoside reverse transcriptase inhibitor

PI Protease inhibitor

RAL Raltegravir

TAF Tenofovir alafenamide

TDF Tenofovir disoproxil fumarate

WHO World Health Organization

CHAPTER 1: INTRODUCTION

1.1 Brief overview of human immunodeficiency virus

Human immunodeficiency virus (HIV), a member of the genus Lentivirus, is a causative agent of the spectrum of disease known as acquired immunodeficiency syndrome (AIDS). HIV is a retrovirus that infects the human immune system by binding to the CD4 receptor and chemokine coreceptors of host cells, which are primary found on CD4 T cells, but also on monocytes and macrophages, and dendritic cells.1

Once entering the host cell, the virus starts the replication process with the help of several virally encoded enzymes, including reverse transcriptase, protease, ribonuclease and integrase.

The reverse transcriptase transcribes the uncoating viral RNA into double-stranded proviral

DNA.2 The resulting proviral DNA then is then imported into the cell nucleus and integrated into the cellular RNA by the integrase, where it is transcribed by host cell to produce viral proteins and genomic RNA.2 Finally, the protease cleaves the viral proteins, and assembles them with genomic RNA into new virus particles, which is then released and begin the replication cycle anew.2

The major cause for AIDS progression in patients living with HIV is the decline of CD4

T-cells. As a consequence of progressive depletion of CD4 T cells, the immune system is damaged, and becomes incapable of controlling opportunistic infections that ultimately lead to life-threatening AIDS-defining diseases.1 HIV infection is also characterized by a notable increase in immune activation, which involves both the adaptive and innate immune systems.1

First, HIV directly triggers Toll-like receptors (TLR7 and TLR 8), leading to production of

interferon-α. Second, microbial translocation, due to the profound depletion of CD4 T cells in the gastrointestinal tract and enhanced gastrointestinal tract permeability, leads to increased plasma concentration of lipopolysaccharides as a potent activator of TLR4, which results in the production of pro-inflammatory cytokines such as interleukin 6 and tumor necrosis factor α.

Further, coinfection with viruses such as cytomegalovirus can induce marked expansion of active cytomegalovirus-specific T cells. There is a growing body of evidence of residual inflammation and increased immune activation even among patients with HIV on antiretroviral therapy.1

Residual chronic inflammation in patients with HIV on antiretroviral therapy has been associated with increased mortality3 and non-AIDS related diseases including (but not limited to) cardiovascular disease4, cancer5, neurological disease6, liver disease7, and depression8.

HIV is spread primarily through unprotected sex, mother to child transmission, contaminated blood transfusion, injection drug use, accidental inoculation, or organ transplantation.9 An important factor in the likelihood of HIV transmission is plasma HIV viral load. Acute HIV infection, which causes a high concentration of plasma viral load early after infection (while patients are asymptomatic, and even remain antibody-negative), is a significant driver of HIV epidemics.1 Thus the goal of HIV treatment is to reduce the HIV viral load to a consistently undetectable level. Other prognostic factors that increase the risk of sexual transmission of HIV include (but not limited to) sexually transmitted infections, pregnancy, multiple sexual partners, concurrent relationships, and injection drug use.1

The World Health Organization estimates that 37.9 million people were living with HIV globally and it resulted in 770,000 deaths in 2018.10 The African region remains most severely affected, which accounted for more than two-thirds of the people living with HIV worldwide.

The Centers for Disease Control and Prevention has reported that it is estimated that there were

1.1 million people living with HIV in the United States in 2016, of whom about 14% were not aware of their HIV status.

1.2 Antiretroviral therapy

The introduction of combination antiretroviral therapy (ART) in the mid-1990s brought about dramatic decline in AIDS diagnoses and deaths. Contemporary ART is highly effective in suppressing plasma viremia and prolonging survival, making HIV infection a manageable chronic disease.11 Contemporary ART is usually composed of two nucleoside reverse transcriptase inhibitors (NRTIs) plus a third active drug of different class (a nonnucleoside reverse transcriptase [NNRTI; e.g., efavirenz], a boosted protease inhibitor [PI; e.g., atazanavir or darunavir], or an integrase strand transfer inhibitor [InSTI; e.g., raltegravir]). Because there is no current strategy for HIV cure, people with HIV infection may be exposed to ART for decades.

Maximizing the safety profile and tolerability while maintaining strong potency of ART is a clinical priority.

The NNRTI-class efavirenz (EFV)-based regimen was recommended as the preferred first-line regimen for people with HIV for over fifteen years.12 Like other NNRTIs, EFV restricts

HIV viral replication by binding to and blocking the hydrophobic region of the HIV-1 reverse transcriptase via a mechanism that alters the enzyme conformation. Currently, EFV-based regimens are still one of the most commonly prescribed antiretroviral medications worldwide.12

Almost 70% of people on a first-line regimen, especially in low- and middle-income countries, were using EFV-based combinations by 2014.12 Numerous head-to-head randomized clinical trials have demonstrated the efficacy and short-term safety of EFV.13–18 One of the major advantages of EFV is its ability to be combined with two other NRTI agents - tenofovir

disoproxil fumarate (TDF) and emtricitabine (FTC) - into a convenient single-pill, once daily regimen, the first single tablet ART regimen approved in the US. Such formulation has potential for improved adherence to antiretroviral medications. Based on the high virologic efficacy and safety in clinical trials combined with ease of administration, EFV-based regimens have long been among the most frequently employed ART regimens to treat HIV-infection.

Despite its advantages, growing evidence has shown EFV is associated with several long- term adverse events, especially in terms of the central nervous system.19–21 EFV has been associated with neuropsychological side effects, including decreased concentration, dizziness, and vivid dreams. Some studies also found increased risks of depression and suicidal ideation for people taking EFV.22,23 Furthermore, EFV was related to neurocognitive impairment, although the evidence was mixed.22,24 In addition, it has been recognized that transmitted resistance is an important concerns for EFV. Because of these issues, the primacy of EFV has been challenged by newer potent drugs that have been tested in randomized trials to demonstrate a better toxicological profile and result in more desirable patient tolerance, a critical factor in the present era of lifelong HIV management. As a result, the current US Department of Health and Human

Services (DHHS) guidelines have removed EFV from the first-line treatment therapy recommendation list.

InSTI-based regimens are one of the most recently developed and promising classes of antiretroviral drugs, and have gained increasing popularity in the developed world since the FDA approved the first InSTI, raltegravir, in 2007. InSTI-class drugs inhibit the catalytic activity of

HIV-encoded viral enzyme integrase and block the insertion of HIV genome into the DNA of host cell, thus halting the spread of HIV virus. Randomized comparative trials have demonstrated that patients initiating InSTI-based regimens (raltegravir [RAL], dolutegravir

[DTG, FDA approved in 2013], elvitegravir/cobicistat [EVG/COBI, FDA approved in 2012], and bictegravir [BIC] FDA approved in 2018) experience more rapid and potent antiretroviral activity, and fewer adverse events and drug interactions compared with those initiating other regimens.25–33 In addition, the current available InSTIs, except RAL with no fixed-dose formulation, can be administered as a once-daily, single-tablet regimens, hence reducing the pill burden and promoting therapy adherence. An overview of each InSTI-class drug is described in

Table 1.1.

Because of these advantages, the US Department of Health and Human Services, the

International Antiviral Society-USA and the European antiretroviral treatment guidelines all now recommend InSTI-based regimens for adults.12,34 Particularly, treatment guideline panels have recommended InSTI-based regimens as initial ART therapy.34,35 The WHO added DTG to its first-line regimen list in 2016 consolidated guidelines for the use of antiretroviral drugs.12

Moreover, the US President’s Emergency Plan For AIDS Relief (PEPFAR) is rolling out DTG- based regimens to essentially replace EFV-based regimens worldwide. Although recent reports of neural tube defects have caused both US and WHO guidelines to caution providers about use of dolutegravir by individuals of childbearing potential,36 dolutegravir and other InSTI-class drugs remain the recommended therapy for most people with HIV.

Direct comparisons of efficacy and safety of InSTI-based regimens with EFV-regimen have been performed in the randomized clinical trials: STARTMRK (RAL vs. EFV)25,37–40,

SINGLE (DTG vs. EFV)28,41, and Study 102 (EVG vs. EFV)30,31,42,43. A brief summary of each trial is presented in Table 1.2.

The STARTMRK study25,37–40 was a multicenter, double blind, phase III, non-inferiority randomized controlled trial. A total of 566 treatment-naïve patients with HIV-1 infection were

enrolled, stratified by baseline screening HIV RNA levels (>50,000 vs. <= 50,000 copies/ml) and randomized (1:1 ratio) to RAL twice daily or EFV once daily, both combined with the TDF/FTC backbone. The primary endpoint was achievement of less than 50 viral RNA copies/ml at week

48. Secondary endpoints included the proportion of patients achieving viral load <400 copies/ml and change from baseline in CD4 counts both measured at week 48 and 96, as well as achieving viral load <50 copies/ml at week 96. The main analysis showed that, at week 48, 86.1% of the

RAL group and 81.9% of the EFV group achieved viral suppression <50 copes/ml. The time to achieve viral suppression was significantly shorter for patients on RAL than on EFV. The RAL group patients also presented significantly fewer drug-related clinical adverse events (44.1%) compared with those on EFV (77.0%). The results at week 96 and 156 also supported the non- inferiority of RAL, and these results further found that treatment effects were consistent across subgroups stratified by pre-specified demographic and prognostic characteristics. The final results at week 192 and 240 when the study maintained blinded showed that based on a subsequent test of superiority following the condition that non-inferiority was met, RAL induced a significantly better viral suppression than EFV (71.0% vs. 61.3% at week 240), and the RAL recipients also experienced significantly fewer drug-related clinical adverse events as well as neuropsychiatric side effects, proving the superiority of RAL/ TDF/FTC compared with

EFV/TDF/FTC.

The SINGLE study28,41 was a randomized, double-blind, multicenter phase III non- inferiority clinical trial directly comparing efficacy and safety of DTG plus abacavir/lamivudine

(ABC/3TC) fixed-dose combination with the co-formulated EFV/TDF/FTC among 844 treatment-naïve subjects with viral load >1,000 copies/ml and negative for the HLA-B*5701 allele. Eligible patients were stratified by plasma HIV RNA (< vs. >100,000 copies/ml) and CD4

cell count (> vs. ≤ 200/mm3), and randomly assigned (1:1) to either DTG/ABC/3TC or

EFV/FTC/TDF. The primary endpoint was the proportion of participants achieving viral suppression (viral load <50 copies/ml) at week 48, with a 10% non-inferiority margin. If both per-protocol and intention-to-treat showed non-inferiority, superiority test was then conducted.

At week 48, the DTG/ABC/3TC group had a significantly higher proportion of participants achieving viral suppression (88%) compared with the EFV/FTC/TDF group (81%). The

DTG/ABC/3TC also experienced a shorter median time to viral suppression than did the

EFV/FTC/TDF group (28 vs. 84 days), as well as a better immune recovery (267 vs. 208 CD4 count/mm3). Lower rate of treatment discontinuation due to adverse events was found in the

DTG/ABC/3TC group than in the EFV/FTC/TDF group (2% vs. 10%). Furthermore,

DTG/ABC/3TC has a more favorable tolerability profile than EFV/FTC/TDF with a lower rate of rash and neuropsychiatric events (whereas higher rate of insomnia was found in the

DTG/ABC/3TC group). Such treatment effect was found to be consistent in HIV-infected participants with a variety of baseline characteristics. In addition, no participant incurred emergence of resistant virus in the DTG/ABC/3TC arm, while five participants have virus with resistance detected in the EFV/TDF/FTC arm (one TDF-associated mutation and four

EFV/associated mutations). At week 96 and 144, DTG/ABC/3TC remained superior. 71% in the

DTG/ABC/3TC group and 63% in the EFV/TDF/FTC group maintained virally suppressed. In addition, before this phase 3 assessment, a phase-2b, dose-ranging randomized trials -- the

SPRING-1 study --44,45 was conducted to select a dose for DTG. Briefly, 205 patients were randomized (1:1:1:1) to receive 10 mg, 25 mg, and 50 mg DTG, respectively, or to receive EFV.

At week 48, 90% on DTG (all doses) achieved viral suppression compared with 82% on EFV.

The Study 10230,31,42,43 was a phase II clinical trial conducted among treatment-naïve patients comparing the safety and efficacy of co-formulated EVG/COBI/FTC/TDF versus co- formulated EFV/FTC/TDF. Eligible participants should have plasma HIV RNA ≥5000 copes/ml, an estimated glomerular filtration rate (eGFR) ≥70 ml/min, and be susceptible to EFV, TDF, and

FTC by HIV-1 genotype at screening. 700 eligible patients were randomized (1:1 ratio) to receive either EVG/COBI/FTC/TDF or EFV/FTC/TDF with matched tablets. The primary endpoint was the proportion of patients in the intention-to-treat population with viral suppression

(HIV RNA <50 copies/ml) at week 48 according to a snapshot analysis, with a 12% non- inferiority margin. At week 48, 87.6% in the EVG/COBI/FTC/TDF group and 84.1% in the

EFV/FTC/TDF group achieved viral suppression, suggesting non-inferiority of

EVG/COBI/FTC/TDF. Response rates were similar in the EVG/COBI/FTC/TDF group and in the EFV/FTC/TDF group among subgroups of patients stratified by demographic and prognostic factors. Significantly better immunological recovery was found in the EVG/COBI/FTC/TDF group than in the EFV/FTC/TDF group (CD4 increase: 239/mm3 vs. 206/mm3). FDA-defined time to loss of virologic response analysis in the intention-to-treat population confirmed that rate of achievement and maintenance of viral suppression was comparable between the two groups

(85.9% in the EVG/COBI/FTC/TDF group and 83.2% in the EFV/FTC/TDF group). Similar number of treatment discontinuations due to adverse events occurred in each group (3.7% in the

EVG/COBI/FTC/TDF group and 5.1% in the EFV/FTC/TDF group). The most frequent adverse events were nausea (more common with EVG/COBI/FTC/TDF [20.7%] than with

EFV/FTC/TDF [13.3%] and dizziness (6.6% vs. 24.4%). Other adverse events that were less common included abnormal dreams, insomnia, and rash. Results at week 96 and 144 continued

to show the comparable rates of viral suppression as well as treatment discontinuation in the two groups.

However, these randomized clinical trials that supported the current guidelines focused on short-term biomarkers including HIV viral suppression and CD4+ cell count change. HIV

RNA viral load, and to a lesser extent CD4+ cell count, are excellent HIV biomarkers and are often used in HIV trials, but it has been long appreciated that no surrogate biomarker is fully adequate to predict clinical HIV progression.46–50 According to Prentice’ criteria51, a “true” rare or distal clinical endpoint can be replaced by a more frequent or proximate surrogate endpoint in randomized trials to reduce the sample size or trial duration. But it is required that the surrogate biomarker should not only be strong predictor of clinical endpoint, but should also capture the net treatment effect on the clinical endpoint.51–53 That is, the surrogate biomarker should capture effect aggregated over all mechanisms of action of the treatment on the clinical endpoint. In practice, however, effects on surrogate biomarker often do not predict the true clinical effects.52

For instance, there may exist some causal pathways of the disease process that are not mediated through the surrogate biomarker but are influenced by intervention under unintended mechanisms. Further, short-term follow up using surrogate biomarkers in randomized trials may not be able to estimate the long-term clinical adverse events associated with switching or discontinuation of treatment that occurs after the study’s conclusion. Therefore, to more fully understand the effectiveness of InSTI-based regimens, evidence is needed concerning the impact of InSTIs on long-term clinical outcomes, including AIDS-defining illnesses and all-cause mortality. Existing randomized trials, however, are underpowered to study long-term clinical outcomes. Observational studies, especially prospective cohort studies, are the only feasible way to address this knowledge gap. However, only one observational study has explored the effect of

an InSTI-based regimen (RAL) on long-term clinical outcomes like incident AIDS and death.

This study had a relatively small number of patients initiating InSTI-regimens and was restricted to 4-year follow-up.54 Further, existing observational studies are focused on AIDS related conditions and death as the outcome, and have not characterized the impact of InSTI-based regimens on serious non-AIDS conditions.55–57 Given limitations of the existing randomized trials and cohort studies, large datasets with long-term clinical outcomes including AIDS- defining illnesses, serious non-AIDS events, and death on InSTI-based regimens are needed.

Additionally, recent publications indicate possible associations between InSTIs and other important adverse events in real-life settings, including muscular toxicity, neuropsychiatric toxicity, immune reconstitution syndrome, neural tube defects, weight gain, malignancy and mortality.36,58–69

Table 1.1: Advantages and disadvantages of currently available InSTIs

Drug Advantages Disadvantages Raltegravir  Few drug interactions  Higher pill burden  Longest safety record with more  No single-tablet complete clinical experience formulation with other NRTIs  Administration with or without  Lower genetic barrier than food bictegravir and dolutegravir  Rapid drop in viral load  Inferior to dolutegravir in  No interactions with anti-HCV treatment-experienced patients DAA agents with HIV RNA >100,000 copies/ml  Potential muscular toxicity  Potential neural tube defects Elvitegravir  Once-daily dosing  Require pharmacological boosting  Available as single-tablet complete of cobicistat, potential CYP3A regimen inhibitor  Rapid drop in viral load  Substantial drug interactions due to cobicistat  Cobicistat raises serum creatinine levels due to inhibition of tabular secretion of creatinine  Not recommended for patients with eGFR <70ml/min  Lower genetic barrier than dolutegravir  Must be administered with food  Interactions with anti-HCV DAA agents  Should be avoided in pregnant women because of inadequate plasma levels  Potential neural tube defects Dolutegravir  Once-daily dosing with or without  Rise in serum creatinine levels due food to inhibition of the OCT-2  Available as single-tablet complete transporter regimen  Higher risk of neuropsychiatric  Rapid drop in viral load adverse effects reported in some  Superior to raltegravir in treatment- observational studies experienced patients  Coformulated with  Long half-life: abcavir/lamivudine; abacavir  High genetic barrier; low risk of requires HLA-B*5701 testing, and resistance with virologic failure has been associated with  Relatively few drug interactions cardiovascular adverse events

 No interactions with anti-HCV  Drug interactions with rifampicin DAA agents (drug to treat tuberculosis), and metformin; need to adjust dosage.  Less clinical experience compared to RAL  Potential neural tube defects Bictegravir  Once-daily dosing with or without  Cannot be used with rifampin food  Limited clinical experience  Available as single-tablet complete  Concern for potential neural tube regimen defects  Non-inferior to dolutegravir  High genetic barrier; low risk of resistance with virologic failure  Relatively few drug interactions

Table 1.2: Summary of randomized trials comparing InSTIs with EFV

STARTMRK25,37–40 SINGLE28,41 Study 102 Study characteristics (RAL vs. EFV) (DTG vs. EFV) (EVG vs. EFV) ABC/3TC NRTI backbone TDF/FTC TDF/FTC (TDF/FTC for EFV) Total number of patients 566 833 700

Female, % 19 16 11 Per-protocol, Intention-to-treat, viral Intention-to-treat, viral suppression, suppression viral suppression Primary endpoint non-completer as failure Non-inferiority margin 12% 10% 12%

Viral suppression 86.1 vs. 81.9 88 vs. 81 87.6 vs. 84.1 proportion at week 48, %

Difference in viral suppression proportion 4.2 (-1.9, 10.3) 7 (2, 12) 3.6 (-1.6, 8.8) (95% CI) at week 48, %

Treatment discontinuation 2.6 vs. 6.0 2.4 vs. 10.0 3.7 vs. 5.1 due to adverse events, %

Viral suppression 81 vs. 79 80 vs. 72 84.2 vs. 81.5 proportion at week 96, %

Difference in viral suppression proportion 2 (-4, 9) 8 (2.3, 13.8) 2.7 (-2.9, 8.3) (95% CI) at week 96, % Viral suppression proportion at week 144 or 71.0 vs. 61.3 71 vs 63 82.2 vs 78.1 240, % Difference in viral suppression proportion 9.5 (1.7, 17.3) 8 (2.0, 14.6) 4.1 (-1.9, 10.0) (95% CI) at week 144 or 240, % RAL was non- DTG/ABC/3TC was EVG/COBI was inferior at week 48 superior to non-inferior to EFV and 96 among EFV/FTC/TDF among among treatment- Study Conclusion treatment-naïve treatment-naïve naïve patients patients (was superior patients at week 240)

CHAPTER 2: STATEMENT OF SPECIFIC AIMS

2.1 Specific Aims and Rationale

To explore the long-term comparative effectiveness of InSTI-based HIV treatment regimens, we use a pooled project of observational cohort data to address the following aims:

AIM 1

To replicate and extend randomized evidence by examining the effect of InSTI- based HIV treatment regimens on surrogate biomarker outcomes (i.e., HIV viral suppression) compared with an EFV-based regimen as an active comparator.70

Rationale: Previous randomized trials have demonstrated that InSTI-based regimens had a more rapid viral response compared with EFV-based regimens.27–29 However, the primary results of these trials were focused on short term viral outcomes (48 weeks). Long-term virologic effect of InSTI-based regimens have not been fully examined. With no randomized evidence forthcoming, observational studies provide a unique opportunity to examine the long-term virologic outcomes. The new user design along with the modern per-protocol analyses via causal inference methods enable us to better utilize the observational data to mimic a randomized trial.71–73 Availability of these outcome data allows us to examine the effect on these biomarkers with follow-up extended beyond the trial timelines. Based on previous randomized evidence27–29, we hypothesize that the InSTI-based regimens will have a more rapid effect on HIV viral suppression (proportion of patients with HIV viral load <50 copies/ml) and favorable mean changes in CD4+ cell count from baseline compared to the comparator EFV-based regimen..

AIM 2

To evaluate the comparative effectiveness of InSTI-based HIV treatment regimens on long-term clinical endpoints including AIDS-defining illnesses, serious non-AIDS events and all-cause mortality compared to EFV-based regimens.

Rationale: This study will include HIV-seropositive patients who started a first-observed antiretroviral regimen consisting of emtricitabine, tenofovir disoproxil fumarate (or tenofovir alafenamide), and either InSTI (i.e., raltegravir, dolutegravir and elvitegravir/cobicistat) or EFV between 2007 and 2017. We will estimate the observational analogs to the intention-to-treat effect as well as the per-protocol effect. We hypothesize that the InSTI-based regimens will produce fewer 10-year clinical outcomes than the comparator EFV-based regimen.

CHAPTER 3: METHODS

3.1 Study design

The North American AIDS Cohort Collaboration on Research and Design (NA-

ACCORD) includes 25 contributing single and multisite clinical and interval epidemiologic HIV cohorts of over 150,000 HIV-seropositive participants that encompass most HIV/AIDS cohort studies in the US and Canada, and is widely representative of HIV care in these two countries

(See Figure 3.1). We have approval from NA-ACCORD to conduct this study.

Participants of the NA-ACCORD are 29% female, 45% non-Hispanic white, 46%

African-American/black, and 11% Hispanic. The prescription of InSTI-based regimens increased substantially since the FDA approved RAL in 2007. In 2015, 40% of the participants in the NA-

ACCORD were prescribed InSTI-based regimens (RAL, DTG, and EVG/COBI) while 25% were prescribed NNRTI-based regimens (a majority EFV-based).

3.2 Study sample

For the study aims, we include HIV-seropositive adults aged 18 or older (NA-ACCORD cohorts primarily collect data on adults so children cannot be studied effectively here) who initiated an antiretroviral regimen consisting of 2 NRTIs (i.e., TDF [or tenofovir alafenamide,

TAF] and FTC) and either InSTI (i.e., RAL, DTG or EVG/COBI) or EFV while under follow-up between July 2009 and December 2016 (during 2007-2009, RAL was only recommended for those with drug-resistant HIV. Patients excluded are: 1) previously experienced with

antiretroviral treatment; 2) missing history of antiretroviral therapy; and 3) those with undetectable HIV RNA viral plasma viral concentration (<50 copies/ml) measured within 180 days before to 7 days after treatment initiation (because an undetectable HIV RNA viral load may indicate unreported prior use of antiretroviral therapy).

3.3 Preliminary data

Prior work by Cole et al, using data from the CFAR Network of Integrated Clinical

Systems (CNICS), assessed 4-year risk of AIDS and all-cause mortality among those initiating a

RAL-based regimen compared to those initiating an EFV-based regimen.54 In this study, 415 patients starting a RAL-based regimen and 2646 starting an EFV-based regimen were analyzed. PI During the 4-year follow-up period, 39 patients initiating a RAL-based regimen and 196 initiating an EFV-based regimen incurred an AIDS-defining illness or died. 24% of patients dropped out from the study (i.e., lost to follow up). Results showed similar 4-year risk of AIDS or death. The 4-year intention-to-treat risk difference was -0.9 (95% CI: -4.5, 2.7), while the per- protocol risk difference was -0.5 (95% CI: -3.8, 2.9) with no significant difference. However, this study was limited to RAL, was small (with only 39 events in the RAL group), was limited to

AIDS incidence and mortality, and was relatively short. Thus, the proposed aims using NA-

ACCORD data, with various InSTI (including DTG, whose genetic barrier to resistance exceeds those of EFV, EVG and RAL35), larger sample size, and longer follow-up will address existing uncertainties regarding clinical outcomes of InSTI-based regimens.

3.4 Measurements

Initiation of InSTI-based or EFV-based regimens

Initiation of InSTI-based regimens is defined as the date between July 2009 and

December 2016 at which a patient initiated ART regimens consisting of TDF (or TAF) and FTC backbones, and one InSTI (i.e., RAL, DTG, EVG/COBI), with all drugs started within 14 days of each other. To be consistent with the InSTI-based regimens in terms of calendar time, initiation of EFV-based regimens is defined as the date between July 2009 and December 2016 at which a patient initiated ART regimens consisting of TDF (or TAF) and FTC backbones, and EFV with all drugs started within 14 days of each other. All the antiretroviral regimens used will be abstracted from point-of-care electronic medical records.

Outcomes of interest

The primary outcome for Aim 2 is the combined clinical outcomes used by the START trial.74 Specifically, the AIDS-defining illnesses (e.g., non-Hodgkin lymphoma, progressive multifocal leukoencephalopathy, cryptococcosis, cerebral toxoplasmosis, AIDS dementia complex, and disseminated Mycobacterium avium complex) are defined based on 1993 criteria of the Centers of the Disease Control and Prevention75, and ascertained by the treating physicians. Dates of diagnoses of AIDS defining illnesses are recorded in the electronic medical record from each cohort. Serious non-AIDS events are defined as follows: cardiovascular disease

(myocardial infarction, stroke, or coronary revascularization), end-stage renal disease (initiation of dialysis or renal transplantation), liver disease (decompensated liver disease), non–AIDS- defining cancer (except for basal-cell or squamous-cell skin cancer).74 These endpoints are all noted as established diagnoses by ICD9/ICD10 in the electronic medical record from each

cohort. Some cohorts have active endpoint adjudication in place, and non-adjudicated endpoints will be termed presumptive. Primary analysis will include both presumptive endpoints and those confirmed by adjudication. The date of death from any cause is identified from queries to the US

Social Security Death Index and/or National Death Index. HIV RNA viral loads and CD4+ cell counts are measured using laboratory tests at study visits.

Covariates

Covariates include time-fixed and time-varying variables. Time-fixed covariates, measured at the start of follow-up (i.e., treatment initiation), are sex, race/ethnicity, age, history of male sex with men, history of injection drug use, smoking, BMI, diagnosis of depression or anxiety, baseline and nadir CD4+ cell count, baseline and peak HIV RNA viral load, cohort, calendar year, years since HIV diagnosis, and clinical diseases (e.g., hepatitis B and C virus infection, liver disease, diabetes mellitus, hypertension, cardiovascular and cerebrovascular diseases). Demographic variables including sex, age, race/ethnicity, history of male same sex activity and history of injection drug use are measured at baseline intake questionnaires. History of depression or anxiety diagnosis are typically recorded in the electronic medical record.

Missing information on important covariates including race/ethnicity, injection drug use, and baseline CD4 cell count will be imputed 50 times allowing for nonmonotone missing data.76

Time-varying covariates, used to account for possible bias due to informative dropout and treatment changes or switches, include CD4+ cell count, HIV RNA viral load, time since last laboratory measure, and diagnoses of clinical diseases (e.g., HCV, HBV, diabetes, and hypertension). HIV RNA viral loads and CD4+ cell counts are measured using laboratory tests

results obtained at baseline and at each visit during follow-up. In addition, dates of any regimen switches and changes are recorded in the electronic medical record.

3.5 Analysis plans

AIM 1: To examine the effect of InSTI-based HIV treatment regimens on biomarker outcomes.

Aim 1 will examine whether InSTI-based regimens cause more rapid viral suppression and more favorable changes in CD4+ cell count compared with the EFV-based regimen. The time zero for analyses, or study origin, is treatment initiation date. We will categorize data into

3-month intervals to accommodate discrepancy of timelines by different cohorts. Patients will be considered lost (i.e., dropouts) after 18 months without contact. The main study endpoint in this aim is achievement of a suppressed HIV RNA, i.e. an HIV concentration <200 copies/mL.

First, we will build inverse probability of treatment initiation weights77 by examining the pretreatment covariates that influence the decision to initiate InSTI-based HIV treatment regimens rather than the EFV-based regimen. We group factors into 5 categories: 1) demographic factors including age, gender (including transgender), race (white, black, Asian, and other), and ethnicity (Hispanic or not); 2) behavioral factors including injection drug use, men who have sex with men, heterosexual behavior; 3) clinical factors including baseline HIV viral load (copies/ml) and CD4+ count (cells/µl), history of depression diagnosis, hepatitis B and

C virus infection, liver disease, diabetes mellitus, hypertension, cardiovascular and cerebrovascular diseases; 4) other factors, e.g., year starting initial regimen.

The primary analysis is the observational analog of an intention-to-treat analysis. We will estimate logistic models to predict uptake of InSTI versus the EFV-based regimen conditioning on potential confounding factors.78 We will estimate the observational intention-to-treat risk

difference (and 95% confidence intervals) of achieving viral suppression comparing InSTI-based regimens with the EFV-based regimen at different follow-up times (i.e., 1 years, 3 years, and 7 years).

For the observational analog of the per-protocol effect, we will additionally censor patients with changes in treatment regimens.78,79 Treatment changes in both groups include treatment cessation and treatment switches. Treatment regimen changes that are considered allowable exceptions are a change from one InSTI regimen to another InSTI regimen (for instance, from a RAL-based regimen to a DTG-based regimen), and a change from the EFV- based regimen to a NNRTI-based regimen of rilpivirine, TDF and FTC. Inverse probability of treatment and censoring weights will be combined with additional weights for treatment changes.

Assuming no unmeasured confounding, we are able to consistently estimate the observational per-protocol risk differences for achieving viral suppression comparing InSTI- based regimens with the EFV-based regimen.

AIM 2: To evaluate the comparative effectiveness of InSTI-based HIV treatment regimens on long-term clinical endpoints over an EFV-based regimen.

Aim 2 will examine the effect of InSTI-based treatment regimen on long-term clinical endpoints (AIDS-defining illnesses, serious non-AIDS events and all-cause mortality). Similar to

Aim 1, the time origin is initiation of treatment. The primary analysis is again an observational analog of intention-to-treat effect. We will employ inverse probability weighting to account for potential confounding and informative dropout. We will estimate the 7-year risks, risk difference, and risk ratios, as well as 95% confidence intervals. We will explore the effect of treatment with InSTI versus the EFV-based regimen on components of the combined study

endpoint (i.e., AIDS incidence, incidence of serious non-AIDS events, and all-cause mortality, using methods appropriate for competing risks.80,81

For the observational analog of the per-protocol effect, we will additionally censor people with any change in treatment regimen. Exceptions to treatment regimen changes are the same as provided for Aim 1 above.

3.6 Statistical power

In the preliminary analysis of CNICS data comparing RAL-based to EFV-based regimens by Cole et al54, 415 (14%) of 3061 participants initiated the RAL-based regimens at study entry.

The 4-year risk of incident AIDS or death among those initiating EFV-based regimens was 6.5%.

Considering that the risk curves decelerated over follow-up, we estimate that the 8-year crude

ퟏퟎ−(ퟏퟎ−ퟒ)/ퟐ risk among those initiating EFV-based regimens is 12% (= ퟔ. ퟓ% × ). (Of course, ퟒ this is an underestimate because we will also have serious non-AIDS events in the proposed study.) We also estimate conservatively that, based on our previous experience, the sample size in this proposal using NA-ACCORD data is about 3 times that of the CNICS sample. That is, about 9000 participants who initiated EFV-based or InSTI-based regimens will be included in this study.

Although methods have not been developed to calculate statistical power of the inverse- probability weighted estimator directly, we can extend a standard normal approximation for the statistical power for the difference in two binomials to account for variance inflation due to inverse probability weights.82 Specifically, the statistical power is determined in terms of the

풑̅(ퟏ−풑̅) |풑ퟏ−풑ퟎ|−ퟏ.ퟗퟔ√휽 standard normal cumulative distribution function, 횽(풛), where 풛 = 풏ퟏ+풏ퟎ , where, 풑 (ퟏ−풑 ) 풑 (ퟏ−풑 ) √휽[ ퟏ ퟏ + ퟎ ퟎ ] 풏ퟏ 풏ퟎ

풑풋, 풏풋 represent the proportion and size of reference and treatment samples respectively denoted as 풋 = ퟎ, ퟏ, 풑̅ is average 풑풋, and 휽 is the variance inflation factor.

Based on our prior published works54, we estimate conservatively that the variance inflation factor will be 2. Then, combined with the expected sample size, the proportion of the study size initiating InSTI-based regimens (14%), and the estimated 10-year risk (12%) among those initiating the EFV-based regimen, the statistical power is shown in Figure 3.2 for different samples sizes and risk differences. With an 8-year risk difference of 4% comparing the EFV- based regimen to InSTI-based regimens (i.e., risk is 12% for those initiating the EFV-based regimen, and 8% for those initiating InSTI-based regimens), the statistical power is 86% for a sample size of 9000.

3.7 Interpretation of results

If we see long-term benefits and fewer side effects and discontinuations of the InSTI- based regimens over, or we see no differences with (as reported previously in limited data) the

EFV regimen, then we add assurance to current guidelines, and provide evidence of continued spread of InSTIs. Contra wise, if we see negative long-term clinical outcomes on InSTI-based regimes this calls into question the advantages of InSTI-based regimens and current guidelines, and calls for speeding the development of new ART. Given the large sample size, rich covariate set, and thorough analyses, we will have impactful results based on new knowledge regardless of whether we find InSTI-based regimens to perform better, the same, or worse than the well- understood reference group of EFV-based regimens.

Figure 3.1: The North American AIDS Cohort Collaboration of Research and Design (NA-ACCORD). (https://statepiaps7.jhsph.edu/naaccord/)

Figure 3.2: Statistical power based on risk difference comparing EFV- based regimens to InSTI-based regimens across different sample sizes

CHAPTER 4: VIROLOGIC OUTCOMES AMONG ADULTS WITH HIV USING INTEGRASE STRAND TRANSFER INHIBITOR BASED ANTIRETROVIRAL REGIMENS: A COLLABORATION OF COHORT STUDIES IN THE UNITED STATES AND CANADA

4.1 Introduction

Modern antiretroviral therapy (ART) is highly effective in suppressing plasma viremia

(i.e., HIV RNA viral load),1 which is a central biomarker for HIV prognosis. Integrase strand transfer inhibitors (InSTI)-based regimens have been recommended as first-line ART for adults.2–4 Randomized trials have demonstrated that patients initiating InSTI-based regimens experienced more rapid control of plasma viremia, as well as better tolerability, compared with those initiating other regimens.5–13 However, published trials had limited follow-up periods (48 weeks).5,10,14 Randomized evidence of long-term effectiveness of InSTI-based regimens remains unavailable.

Here using observational data from a large collaboration of North American cohort studies, we estimate the effect of InSTI-based regimens, compared to an efavirenz (EFV)-based regimen, on plasma viremia over 7 years.

4.2 Methods

Study Design

The North American AIDS Cohort Collaboration on Research and Design (NA-

ACCORD) is the largest multisite collaboration of clinical and interval HIV cohorts in the

United States and Canada. Details on this collaboration have been published previously.83

Briefly, the NA-ACCORD consists of >20 cohorts and 200 clinical care sites that prospectively collect data on more than 180,000 adults living with HIV. Cohorts securely transfer demographic, medication, laboratory, diagnostic, and vital status data annually to the central

Data Management Core (University of Washington, Seattle, WA, USA), where the data undergo quality control and are harmonized across cohorts for analyses by the Epidemiology/Biostatistics

Core (Johns Hopkins University, Baltimore, MD, USA). The human subjects research activities of the NA-ACCORD and each participating cohort have been approved by their respective local

Institutional review boards, the Johns Hopkins School of Medicine, and the University of North

Carolina School of Medicine.

Study Population and Eligibility Criteria

In the present study, we included HIV-seropositive and ART-naïve adults (≥18 years) who initiated an ART regimen consisting of two nucleoside reverse transcriptase inhibitors (i.e., tenofovir disoproxil fumarate [TDF] or tenofovir alafenamide [TAF], and emtricitabine [FTC]) and either InSTI (i.e., raltegravir [RAL], dolutegravir [DTG], elvitegravir/cobicstat

[EVG/COBI]) or EFV while under follow-up between July 2009 and December 2016 from 16 clinic cohorts in the NA-ACCORD. The follow-up began in 2009, rather than 2007 when RAL was first approved by the FDA, because RAL was recommended only for those with drug resistance from 2007 to 2009. Patients excluded were: 1) those having prior ART experience; 2) those with baseline HIV viral suppression (HIV RNA viral load <200 copies/mL) that is measured within 90 days before to 7 days after ART initiation; 3) those without any measurements of HIV RNA viral load since ART initiation.

Outcome Measurements

The primary outcome was the repeated measurements of being virally suppressed (HIV

RNA viral load <200 copies/mL vs. ≥200 copies/mL) since ART initiation. The timing and frequency of observed laboratory measurements for HIV RNA viral load since ART initiation varied by person. To standardize, we constructed time-updated HIV RNA viral load at 3-month intervals. If there were multiple HIV RNA measurements within a 3-month interval, an average of these measurements was adopted.

Covariates

We selected baseline covariates which we considered to be potential confounders for effect of ART initiation on the virologic outcome. Age, sex, race, ethnicity, and HIV-acquisition risk group (men who have sex with men, individuals reporting current or previous injection drug use, heterosexual behavior, and other) were self-reported at enrolment into the NA-ACCORD.

History of any clinical AIDS diagnosis, hepatitis C infection (having a positive antibody test, or a detectable RNA, or the presence of hepatitis C genotype test), hepatitis B infection (defined as positive surface antigen test, a positive e-antigen test, or a positive DNA test result), depression, anxiety, diabetes mellitus (a glycosylated hemoglobin ≥6.5%, diabetes-specific medication, or a diagnosis with a diabetes-related medication), hypertension (clinical diagnosis and prescription of anti-hypertensive medication), elevated total cholesterol (≥240 mg/dL) were recorded at ART initiation. Body mass index, which was captured as the closest date to ART initiation, was calculated as weight (kilograms) divided by height (meters) squared. Baseline CD4 cell count

(cells/μL) and HIV RNA viral load (copies/mL) were captured at the closest date to ART initiation within the window from 90 days before to 7 days after ART initiation. Calendar year at

initiation and individual cohorts were captured as indicator variables. Restricted quadratic splines with four knots at 5th, 35th, 65th, and 95th percentiles were used to model continuous covariates including age, body mass index, CD4 cell count and HIV RNA viral load.84

Time-varying covariates, which were used to account for ART treatment changes, included time-updated CD4 cell count, new occurrences of clinical diseases or conditions (i.e., diabetes mellitus, depression, anxiety, and hypertension), elevated total cholesterol after ART initiation.

Statistical Analyses

We restricted analyses to apparent new users of ART to avoid selection bias due to the inclusion of prevalent users, and to mimic a randomized trial where treatment-naïve patients were randomly assigned to either the InSTI-based regimen or the active comparator EFV-based regimen.72 Each participant was followed from date of ART initiation (study entry for individuals and time origin for our study design) until the date of the last HIV RNA measurement or administrative end of follow-up (at 7 years, or 31 December 2016).

Missing baseline covariates (see Table 4.1) were imputed 20 times using multiple imputation by chained equations.85,86 The imputation model included all baseline covariates, treatment regimen variable, and the repeatedly measured outcomes for HIV RNA viral load.87

Baseline CD4 cell count and repeatedly measured HIV viral load were log-transformed to avoid negative imputed values. Of note, missing measurements of HIV RNA viral load were also imputed.

For primary analyses, we estimated the observational analog of intention-to-treat effect of initiating an InSTI-based regimen compared with initiating the EFV-based regimen on virologic

outcome (being virally suppressed or not) regardless of ART treatment changes. For intention-to- treat analysis, in each imputed dataset, inverse probability of treatment weights were constructed to account for potential baseline confounding. These weights were assigned to each participant to create the intention-to-treat population, and then applied to repeated observed measures of being

HIV virally suppressed or not within each participant. Using the weighted observed measures of being HIV virally suppressed or not, we then calculated the proportion virally suppressed over follow-up by treatment groups in the intention-to-treat population. For intention-to-treat effect, we estimated the difference in the proportion virally suppressed in the InSTI-based regimen, versus the EFV-based regimen, at 0.25, 0.5, 0.75, 1, 3, 5, 7 years since ART initiation. The standard error for these differences was estimated from a nonparametric bootstrap with 200 random samples of participants with replacement for each imputed dataset.88 Rubin’s rule was applied to pool the results across imputed datasets to obtain the pooled differences with 95% confidence intervals (CIs).

We also estimated the per-protocol effect of initiating and remaining on an InSTI-based regimen compared with initiating and remaining on the EFV-based regimen on the virologic outcomes.73,89 For per-protocol analysis, participants were additionally censored when they deviated from their initial treatment strategy (i.e., ART treatment changes). ART treatment changes were those with ART treatment discontinuations and switches. Exceptions from ART treatment changes included: 1) a change from one InSTI-based regimen to another InSTI-based regimen (for instance, from a RAL-based regimen to a DTG-based regimen); 2) a change from the EFV-based regimen to a non-nucleoside reverse transcriptase inhibitor (NNRTI)-based regimen of rilpivirine, TDF and FTC; and 3) switch between TDF and TAF. We then constructed the stabilized inverse probability of censoring weights by conditioning on time-varying

covariates to account for treatment changes over follow-up.78,90 These censoring weights were combined with inverse probability of treatment weights, assigned to each participant to create the per-protocol population, and then applied to repeated observed measures of being HIV virally suppressed or not within each participant. Using the weighted observed measures of being HIV virally suppressed or not, we then calculated the proportion virally suppressed over follow-up by treatment groups in the per-protocol population. For the per-protocol effect, we also estimated the difference in proportions virally suppressed for initiating and remaining on an InSTI-based regimen versus initiating and remaining on the EFV-based regimen at 0.25, 0.5, 0.75, 1, 3, 5, 7 years since ART initiation.

In secondary analyses, we applied the inverse probability of treatment and censoring weights to both observed and imputed measures of being HIV virally suppressed or not, and estimated the difference in proportions virally suppressed for intention-to-treat and per-protocol effects.

SAS software version 9.4 (SAS Institute, Cary, North Carolina) was used for all analyses.

4.3 Results

A total of 15,318 participants in the NA-ACCORD were eligible and included in the study population. Among these participants, 5,519 (36.0%) initiated an InSTI-based regimen, and 9,799 (64.0%) initiated the EFV-based regimen between 2009 and 2016. Of 5,519 patients initiating an InSTI-based regimen, 1,783 (32.3%) initiated RAL, 3,137 (56.8%) initiated

EVG/COB, and 599 (10.9%) initiated DTG. Secular trends of ART initiation in Figure 4.1 show an increased proportion of InSTI-based regimen initiators across calendar year from 2009 to

2016. Baseline characteristics at ART initiation of the study population are described in Table

4.2. Compared with those in the EFV group, the participants in the InSTI group were more likely to be female, report male-to-male sexual contact, and have a previous diagnosis of depression or anxiety. During the 7-year follow-up period, there were a total of 124,822 HIV viral load measurements recorded in the study population (34,781 in the InSTI group and 90,041 in the

EFV group). The InSTI group had a median of 5 viral load measurements (interquartile range, 3–

8), and the EFV had a median of 9 viral load measurements (interquartile range, 5-13). The crude proportion virally suppressed among the study population is depicted in the upper panel of

Figure 4.2. The proportion virally suppressed was 81.9% in the InSTI group, and 65.8% in the

EFV group at 0.25 years after ART initiation, and then increased to 91.9% in the InSTI group and 92.3% in the EFV group at 7 years (shown in the upper panel of Figure 4.3).

In the intention-to-treat analyses, after accounting for baseline confounding, the proportion virally suppressed is shown in the middle panel of Figure 4.2. The intention-to-treat differences in the proportion virally suppressed in the InSTI-based regimen versus the EFV- based regimen, at 0.25, 0.5, 0.75, 1, 3, 5, 7 years since ART initiation are described in the middle panel of Figure 4.3. For example, at 0.25 years after ART initiation, the proportion virally suppressed was 81.3% in the InSTI group, and 67.3% in the EFV group, corresponding with a difference of 14.0% (95% CI: 12.4, 15.6). At 1 year after ART initiation, the proportion virally suppressed was 89.5% in the InSTI group and 90.2% in the EFV group, corresponding with a difference of -0.7% (95% CI: -2.1, 0.8). At 7 years after ART initiation, the proportion virally suppressed was 94.5% in the InSTI group, and 92.5% in the EFV group, corresponding with a difference of 2.0% (95% CI: -7.3, 11.3).

Sixty-eight percent of the participants initiating an InSTI-based regimen (3,750/5,519) and 71% of the participants initiating the EFV-based regimen (6,975/9,799) had a treatment

change before the last measurement of HIV viral load or completing the study. In the per- protocol analyses after accounting for baseline confounding and treatment changes, the model- fitted proportion virally suppressed is shown in the lower panel of Figure 4.2. The per-protocol differences in the proportion virally suppressed in the InSTI-based regimen versus the EFV- based regimen, at 0.25, 0.5, 0.75, 1, 3, 5, 7 years since ART initiation are described in the lower panel of Figure 4.3. At 0.25 years after ART initiation, the proportion virally suppressed was

81.3% in the InSTI group, and 67.3% in the EFV group, corresponding with a difference of

14.0% (95% CI: 8.7, 19.3). The proportion virally suppressed (comparing those in the InSTI group with those in the EFV group) became similar after 0.25 years. At 1 year after ART initiation, the proportion virally suppressed was 91.1% in the InSTI group, and 93.0% in the

EFV group, corresponding with a difference of -1.8% (95% CI: -4.9, 1.3). At 7 years after ART initiation, the proportion virally suppressed was 100.0% in the InSTI group, and 92.5% in the

EFV group, corresponding with a difference of 7.5% (95% CI: -0.3, 15.4).

The results of secondary analyses by including both the observed and imputed measures of being HIV virally suppressed were similar with the main analyses (see Figure 4.4).

4.4 Discussion

Using the observational data from this large collaboration of HIV cohorts in the United

States and Canada over 7 years of follow-up, we found that InSTI-based regimens had more rapid virologic response compared with EFV-based regimens among adults living with HIV, especially during the first quarter (0.25 years) since ART initiation. However, both the intention- to-treat and per-protocol analyses showed that, after 0.25 years since initiation, InSTI-based and

EFV-based regimens showed similar effects on virologic outcomes. Although the per-protocol

analyses showed a potential virologic benefit of InSTI at year 7 since ART initiation, this estimate was diluted by the substantial uncertainty due to the smaller per-protocol sample that remained on the initial ART regimens at the end of the study.

Our results, that patients initiating an InSTI-based regimen experienced more rapid virologic suppression than those initiating EFV-based regimens, align with the findings from prior randomized trials: STARTMRK (RAL vs. EFV)25, SINGLE (DTG vs. EFV)28, and Study

102 (EVG vs. EFV)91. The STARTMRK trial showed a shorter time to achieve viral suppression for patients on raltegarvir than on efavirenz within 48 weeks after randomization. At 48 weeks,

RAL-based regimen was found to be non-inferior to EFV-based regimen on viral suppression

(86.1% vs. 81.9% of patients with a virologic response to treatment).25 At week 240, the study showed RAL induced significantly better viral suppression, though much of this difference could be explained by more treatment discontinuations among patients on efavirenz.92 In the SINGLE study that compared DTG plus abacavir/lamivudine (ABC/3TC) with the co-formulated

EFV/FTC/TDF, at week 48, the proportion of being virally suppressed was higher in the

DTG/ABC/3TC group than in the EFV/FTC/TDF group.28 The DTG/ABC/3TC group also had a shorter median time to viral suppression and lower rate of treatment discontinuations than the

EFV/FTC/TDF group. The Study 102 trial, which compared co-formulated

EVG/COBI/FTC/TDF versus co-formulated EFV/FTC/TDF, showed non-inferiority of the

EVG-based regimen to the EFV-based regimen on viral suppression, and comparable number of treatment discontinuations.91

Furthermore, our findings on virologic outcomes within 1 year after ART initiation were also consistent with the several existing observational studies.70,93 Edwards et al. used the data from the Centers for AIDS Research Network of Integrated Clinical Systems (CNICS) and found

that patients initiating raltegravir spent more time alive and virally suppressed than those initiating efavirenz, which was primarily driven by more rapid viral response of raltegravir.70

The probability of being alive and virally suppressed at 2.5 years was similar among these two regimens. In a retrospective single-center study of 155 treatment-naïve patients, Jacobson and

Ogbuagu found patents on InSTI-based regimens experienced higher rates of virologic suppression within the 1 year after initiation, and shorter median time to viral suppression compared to patients on other ART regimens (including efavirenz).93

Our study is subject to limitations. First, in the analysis of observational data, a key limitation is that the treatment was not randomly assigned. To achieve comparability of treatment groups all confounders must be measured with appropriate adjustments made. Although we adjusted for a large number of likely confounders, it is still possible that uncontrolled confounding remains. Second, while most published randomized trials used the definition of viral suppression as HIV RNA viral load <50 copies/mL, we chose to use the cutoff of HIV

RNA <200 copies/mL as there were a portion of HIV viral load measurement with lower limit of detection greater than 50 copies/mL. Further, measurement error may occur during the measurement and classification of HIV viral suppression outcomes.

In the present study, we used a large observational pooling project to replicate findings from randomized trials and extend the evidence base to a larger and more representative sample of adults living with HIV in the United States and Canada over a prolonged follow-up.94 We adopted a new-user study design and modern statistical and causal inference approaches to perform intention-to-treat and per-protocol analyses to mitigate concerns about baseline confounding and treatment discontinuations or switches, and provided comprehensive evidence about the long-term effect of InSTIs on virologic outcomes.

In the present study, we used a large observational pooling project to replicate findings from randomized trials and the evidence base to a longer follow-up of 7-years, among a larger and more representative sample of treatment-naïve adults living with HIV in a real-world routine-care setting, which complements the findings from prior randomized trials and observational studies.94 We found that although InSTI-based regimens had more rapid response compared with efavirenz-based regimens, the long-term effect on virologic outcomes was similar for the InSTI-based regimens and efavirenz-based regimens. Our study also showed that the proportion of ART treatment changes including treatment switches and treatment discontinuations was high and comparable among patients initiating InSTI-based regimens and patients initiating efavirenz-based regimens, which contradicts the randomized evidence suggesting that InSTI-based regimens had fewer treatment discontinuations. Although the reasons for treatment changes were not captured in the NA-ACCORD, this discrepancy between clinical trials and our study might indicate the potential shortcomings of randomized trials that could not perfectly reflect the real-world settings. Combined with the increasing studies showing the potential adverse effects associated with InSTI, including neuropsychiatric toxicity64–66, neural tube defects36, weight gain59,61,95,96, and with our previous findings that there was no benefit of InSTI over efavirenz on clinical outcomes54, our results on a relatively similar long- term virologic effect of InSTI bring into question whether InSTI-based regimens should be favored over other regimens, especially in the current context that dolutegravir-based regimens are being rolled out globally.

Table 4.1: Missing baseline covariates among 15,318 HIV-infected adults initiating an InSTI- based regimen or an EFV-based regimen between July 2009 and December 2016 in the NA- ACCORD, the United States and Canada No. (%) of missingness InSTI-based EFV-based Characteristic regimena regimena Overall (n=5,519) (n=9,799) (n=15,318) Age 0 (0.0) 0 (0.0) 0 (0.0) Female 0 (0.0) 0 (0.0) 0 (0.0) Black race 123 (2.2) 300 (3.1) 423 (2.8) Hispanic ethnicity 123 (2.2) 300 (3.1) 423 (2.8) Body mass index 122 (2.2) 245 (2.5) 367 (2.4) Injection drug use 0 (0.0) 0 (0.0) 0 (0.0) Male-to-male sexual contact 0 (0.0) 0 (0.0) 0 (0.0) Heterosexual behavior 0 (0.0) 0 (0.0) 0 (0.0) Previous AIDS diagnosis 0 (0.0) 0 (0.0) 0 (0.0) Hepatitis B 20 (0.4) 27 (0.3) 47 (0.3) Hepatitis C 0 (0.0) 0 (0.0) 0 (0.0) Previous depression diagnosis 20 (0.4) 27 (0.3) 47 (0.3) Previous anxiety diagnosis 41 (0.7) 73 (0.7) 114 (0.7) Diabetes mellitus 1 (0.0) 0 (0.0) 1 (0.0) Hypertension 18 (0.3) 42 (0.4) 60 (0.4) Elevated total cholesterol 20 (0.4) 27 (0.3) 47 (0.3) Baseline CD4 countb 1,018 (18.4) 2,431 (24.8) 3,449 (22.5) Baseline viral loadc 1,171 (21.2) 2,813 (28.7) 3,984 (26.0) Calendar year at initiation 0 (0.0) 0 (0.0) 0 (0.0) Abbreviations: HIV, human immunodeficiency virus; InSTI, integrase strand transfer inhibitor; EFV, efavirenz; AIDS, acquired immunodeficiency syndrome a Both regimens included the same backbone of tenofovir disoproxil fumarate [or tenofovir alafenamide], and emtricitabine b Baseline CD4 count was captured at the closest date to ART initiation within the window from 90 days before to 7 days after ART initiation c HIV viral load was captured at the closest date to ART initiation within the window from 90 days before to 7 days after ART initiation

Table 4.2. Characteristics at ART initiation of 15,318 HIV-infected adults initiating an InSTI- based regimen or an EFV-based regimen between July 2009 and December 2016 in the NA- ACCORD, overall and by treatment regimens, the United States and Canada No. (%) of Participants InSTI-based EFV-based Characteristic regimena regimena Overall (n=5,519) (n=9,799) (n=15,318) 40.0 (30.0- Age, median (IQR), yrs 38.0 (28.0-49.0) 41.0 (31.0-51.0) 50.0) Female 836 (15.2) 1,055 (10.8) 1,891 (12.3) Black race 2,193 (39.7) 4,414 (45.1) 6,607 (43.1) Hispanic ethnicity 667 (12.1) 1,261 (12.9) 1,928 (12.6) 25.1 (22.3- Body mass index, median (IQR) 25.1 (22.3-28.7) 25.1 (22.4-28.6) 28.6) Injection drug use 533 (9.7) 1,036 (10.6) 1,569 (10.2) Male-to-male sexual contact 3,036 (55.0) 4,343 (44.3) 7,379 (48.2) Heterosexual behavior 1,273 (23.1) 1,931 (19.7) 3,204 (20.9) Previous AIDS diagnosis 459 (8.3) 705 (7.2) 1,164 (7.6) Hepatitis B 206 (3.7) 404 (4.1) 610 (4.0) Hepatitis C 559 (10.1) 1,120 (11.4) 1,679 (11.0) Previous depression diagnosis 831 (15.1) 1,024 (10.5) 1,855 (12.1) Previous anxiety diagnosis 667 (12.1) 722 (7.4) 1,389 (9.1) Diabetes mellitus 276 (5.0) 555 (5.7) 831 (5.4) Hypertension 814 (14.8) 1,843 (18.8) 2,657 (17.4) Elevated total cholesterol 218 (4.0) 483 (4.9) 701 (4.6) Baseline CD4 count, median 346.0 (169.0- 316.0 (174.0- 327.0 (173.0- (IQR), cells/μL 514.0) 451.0) 475.0) Baseline viral load, median (IQR), 44591.0.0 41772.5 42913.5 copies/mL (12712.5- (10700.0- (11358.0- 151552.5) 132824.0) 139000.0) Calendar year at initiation, median 2011 (2010- 2012 (2010- (IQR) 2014 (2013-2015) 2012) 2014) Abbreviations: ART, antiretroviral therapy; HIV, human immunodeficiency virus; InSTI, integrase strand transfer inhibitor; EFV, efavirenz; NA-ACCORD, the North American AIDS Cohort Collaboration on Research and Design; IQR, interquartile range; AIDS, acquired immunodeficiency syndrome a Both regimens included the same backbone of tenofovir disoproxil fumarate [or tenofovir alafenamide], and emtricitabine

Figure 4.1. Secular trend in proportion of initiatiating InSTI- versus EFV-based regimens between July 2009 and December 2016 among 15,318 HIV-infected adults in the NA-ACCORD, the United States and Canada. Abbreviations: InSTI, integrase strand transfer inhibitor; EFV, efavirenz; NA-ACCORD, the North American AIDS Cohort Collaboration on Research and Design; ART, antiretroviral therapy.

Figure 4.2. Proportion virally suppressed (<200 copies/mL) by ART group (InSTI- vs EFV- based regimens) across 7-year follow-up among 15,318 HIV-infected adults in the NA- ACCORD, the United States and Canada. The upper panel represents crude analyses, the middle panel represents intention-to-treat analyses that accounted for baseline confounding, and the lower panel represent per-protocol analyses that accounted for baseline confounding and ART treatment changes (ART treatment discontinuations and switches).

Figure 4.3. Proportion difference (PD) virally suppressed (<200 copies/mL) comparing InSTI- based regimen with EFV-based regimen across 7-year follow-up among 15,318 HIV-infected adults in the NA-ACCORD, the United States and Canada. The upper panel represents crude analyses, the middle panel represents intention-to-treat analyses that accounted for baseline confounding, and the lower panel represent per-protocol analyses that accounted for baseline confounding and ART treatment changes (ART treatment discontinuations and switches).

Figure 4.4. Results of secondary analyses that showed proportion difference (PD) virally suppressed (<200 copies/mL) comparing InSTI-based regimen with EFV-based regimen across 7-year follow-up among 15,318 HIV-infected adults in the NA-ACCORD, the United States and Canada. The upper panel represents crude analyses, the middle panel represents intention-to-treat analyses that accounted for baseline confounding, and the lower panel represent per-protocol analyses that accounted for baseline confounding and ART treatment changes (ART treatment discontinuations and switches).

CHAPTER 5: CLINICAL EFFECTIVENESS OF INTEGRASE STRAND TRANSFER INHIBITOR-BASED ANTIRETROVIRAL REGIMENS AMONG ADULTS WITH HUMAN IMMUNODEFICIENCY VIRUS

5.1 Introduction

Contemporary antiretroviral therapy (ART) is highly effective in suppressing plasma viremia and prolonging survival.11 Because there is no cure for human immunodeficiency virus

(HIV), people living with HIV may be exposed to ART for decades97, and therefore, maximizing the safety profile and tolerability while maintaining strong potency remains a clinical priority.

Integrase strand transfer inhibitor (InSTI)-based regimens are now recommended widely as first-line ART for adults.12,35,98 Randomized trials of InSTI-based regimens have demonstrated clear short-term evidence of strong potency as well as tolerability, compared with other regimens.25–29,32,33,91 However, most randomized trials have been focused on short-term (48 weeks) surrogate biomarkers. Limited data are available regarding longer-term clinical effectiveness of InSTI-based regimens. The few existing observational studies that have examined the clinical effects and safety endpoints of InSTI-based regimens have one or more limitations, such as insufficient sample size and limited follow-up.54,69 We aimed to examine in a collaboration of cohort studies in the United States (U.S.) and Canada by emulating a randomized trial whether those initiating a regimen of InSTI with tenofovir disoproxil fumarate

[TDF] or tenofovir alafenamide [TAF], and emtricitabine [FTC]) improved longer-term (6-year) clinical outcomes of AIDS-defining illnesses, all-cause mortality, and serious non-AIDS events when compared to those initiating a regimen of EFV with the same backbone.

5.2 Methods

Study Design

The North American AIDS Cohort Collaboration on Research and Design (NA-

ACCORD) is the largest consortium of clinical and interval HIV cohorts in North America and is one of the seven regional collaborations of the International Epidemiologic Databases to

Evaluate AIDS (IeDEA) project, supported by the National Institutes of Health. Details on this collaboration have been published previously.83 Briefly, the NA-ACCORD consists of >20 single-site and multisite cohorts that prospectively collect data on more than 180,000 adults living with HIV who had at least two care visits within 12 months at more than 200 clinical sites in the U.S. and Canada. Cohorts securely transfer demographic, medication, laboratory, diagnostic, and vital status data annually to the central Data Management Core (University of

Washington, Seattle, WA, USA), where the data undergo quality control and are harmonized across cohorts for analyses by the Epidemiology/Biostatistics Core (Johns Hopkins University,

Baltimore, MD, USA). The human subjects research activities of the NA-ACCORD and each participating cohort have been approved by their respective local Institutional review boards, the

Johns Hopkins School of Medicine, and the University of North Carolina School of Medicine.

Study Population and Eligibility Criteria

This prospective cohort study included HIV-seropositive and ART-naïve adults aged 18 years or older who initiated an ART regimen consisting of TDF (or TAF), FTC, and either

InSTI (i.e., raltegravir [RAL], dolutegravir [DTG], elvitegravir/cobicistat [EVG/COBI]) or EFV while under follow-up between July 2009 and December 2016 from 16 clinic cohorts in the NA-

ACCORD.. We started follow-up in 2009, rather than 2007 when RAL was first approved by the

FDA, because RAL was recommended only for those with drug resistance from 2007 through

2009. Patients excluded were: 1) those having evidence of prior ART; 2) those with undetectable

HIV RNA viral load measured within 90 days before to 7 days after ART initiation since undetectable viral load may indicate unreported or uncaptured treatment; 3) those having prior history of acute myocardial infarction (MI) or stroke, end-stage renal disease (ESRD) and end- stage liver disease (ESLD).

Outcome Assessment

The primary outcome was a composite of the first occurrence of an AIDS-defining illness, a serious non-AIDS event, or death from any cause after ART initiation. AIDS-defining illnesses are based on 1993 criteria of the Centers of the Disease Control and Prevention (see

Table 5.1 for specific AIDS-defining illnesses).75 Serious non-AIDS events were defined as follows: acute MI or stroke (based on diagnoses), ESRD (estimated glomerular filtration rate first consistently <30 mL/min per 1.73 m2 for at least 3 months with the second measurement date used for event), and ESLD (two FIB-4 [fibrosis-4] >3.25, more than 6 months apart with second measurement date used for event). ESRD and ESLD outcomes were based on laboratory tests as the date through which ESRD and ESLD events are validated in NA-ACCORD did not extend throughout the study follow-up. The diagnoses and laboratory tests were obtained from electronic medical records from each site. The date of death from any cause was identified from queries to the US Social Security Death Index, National Death Index, state (or provincial, for

Canadian sites) death certificates, and electronic medical records.

Covariates

We selected baseline covariates which we considered to be potential confounders for effect of ART initiation on the composite outcome. Age, sex, race, ethnicity, and HIV- acquisition risk group (men who have sex with men, individuals reporting current or previous injection drug use, heterosexual behavior, and other) were self-reported at enrolment into the

NA-ACCORD. Body mass index and history of any clinical AIDS diagnosis, hepatitis C infection (having a positive antibody test, or a detectable RNA, or the presence of hepatitis C genotype test), hepatitis B infection (defined as positive surface antigen test, a positive e-antigen test, or a positive DNA test result), depression or anxiety diagnosis, diabetes mellitus (a glycosylated hemoglobin ≥6.5%, diabetes-specific medication, or a diagnosis with a diabetes- related medication), hypertension (clinical diagnosis and prescription of anti-hypertensive medication), elevated total cholesterol (≥240 mg/dL), and statin prescription were recorded at

ART initiation. Baseline CD4 cell count (cells/μL) and HIV viral load (copies/mL) were captured at the closest date to ART initiation within the window from 90 days before to 7 days after ART initiation. Calendar year at initiation was captured as indicator variables.

Time-varying covariates, which were used to account for differential loss to follow-up and ART treatment changes, included time-updated CD4 cell count, HIV viral load, new occurrences of clinical diseases or conditions (i.e., diabetes mellitus, depression, anxiety, and hypertension), elevated total cholesterol, and new statin prescription after ART initiation.

Statistical Analyses

Our inclusion and exclusion criteria restricted our study population to those we believe initiated ART while under observation in the NA-ACCORD from 2009 to 2016 to avoid

selection bias due to the inclusion of prevalent ART users, and to mimic the setting of a randomized controlled trial where patients were randomly assigned to either the InSTI-based regimen or the active comparator EFV-based regimen through adjusting for baseline confounders measured in the NA-ACCORD.72 Each participant was followed from date of ART initiation

(study entry for individuals and time origin for our study design) until the earliest date of first occurrence of the composite outcome, date of loss to follow-up (defined by NA-ACCORD as 18 months after the date of last CD4 count or HIV viral load measurement), or administrative end of follow-up (at 6 years, cohort-specific end date, or 31 December 2016).

Missing baseline covariates were imputed 10 times using multiple imputation by chained equations.85,86 (See Table 5.2 for proportion of missing values for each baseline covariate). The imputation model included all baseline covariates, treatment regimen variable, the binary outcome indicator, and the cumulative reference hazard.86 Baseline CD4 cell count and HIV viral load were log-transformed to avoid negative imputed values.

For primary analyses, we estimated the observational analog of intention-to-treat effect of initiating an InSTI-based regimen compared with initiating the EFV-based regimen regardless of

ART treatment changes. For intention-to-treat analysis, in each imputed dataset, we accounted for potential baseline confounding and differential loss to follow-up by constructing inverse probability of treatment weights and inverse probability of censoring weights, respectively.

These weights were combined, and applied to Cox proportional hazard model for the composite outcome on initiating an InSTI-based regimen vs. initiating an EFV-based regimen. A robust standard error for hazard ratio was calculated for each imputed dataset. We also estimated the 6- year intention-to-treat risk of the composite outcome for each treatment regimen and the corresponding 6-year intention-to-treat risk difference using inverse probability-weighted

Kaplan-Meier estimators.99,100 We estimated the 6-year risks because there were few patients in the InSTI group followed for more than 6 years since ART initiation. The standard error for risk difference was estimated from a nonparametric bootstrap with 200 random samples with replacement for each imputed dataset. Rubin’s rule was applied to pool the results across imputed datasets to obtain the pooled hazard ratio and 6-year risk difference with 95% confidence intervals (CIs).

We also estimated the per-protocol effect of initiating and remaining on an InSTI-based regimen compared with initiating and remaining on the EFV-based regimen.73,89 For per- protocol analysis, participants were additionally censored when they deviated from their initial treatment strategy (i.e., treatment changes). Treatment changes in both groups included treatment discontinuations and switches. Treatment changes that were considered allowable exceptions and thus were not censored were: 1) a change from one InSTI-based regimen to another InSTI-based regimen (for instance, from a RAL-based regimen to a DTG-based regimen); 2) a change from the EFV-based regimen to a non-nucleoside reverse transcriptase inhibitor (NNRTI)-based regimen of rilpivirine, TDF and FTC; and 3) switch between TDF and TAF. The inverse probability of censoring weights were revised to censor at the minimum of loss to follow-up or treatment changes,78 and combined with inverse probability of treatment weights to estimate the

6-year per-protocol risks, risk difference as well as hazard ratio.

We assessed the robustness of our estimates using two secondary analyses: 1) we restricted the composite outcome only to the most HIV-relevant events – AIDS and death, as well as restricted to serious non-AIDS events; 2) we additionally adjusted for individual cohorts as a potential confounder; and 3) we reran the analyses stratified by baseline CD4 cell count

(≤200 vs. >200 cells/μL).

SAS software version 9.4 (SAS Institute, Cary, North Carolina) was used for all analyses.

5.3 Results

Of 15,993 eligible participants in the NA-ACCORD, 5,824 (36.4%) initiated an InSTI- based regimen, and 10,169 (63.6%) initiated the EFV-based regimen between 2009 and 2016.

Among 5,824 patients initiating InSTI, 1,840 (31.6%) initiated RAL, 3,361 (57.7%) initiated

EVG/COB, and 639 (10.7%) initiated DTG. Characteristics at initiation of the study population are described in Table 5.3. Compared to those in the EFV group, the participants in the InSTI group were more likely to be female, non-Black, report male-to-male sexual contact, and have a previous diagnosis of depression or anxiety. Secular trends showed an increased proportion of

InSTI-based regimen initiators across calendar year from 2009 to 2016 (see Figure 5.1).

During a median follow-up of 2.0 years (interquartile range [IQR] 1.2-3.2), 440 (7.6%) of the 5,824 participants who initiated an InSTI-based regimen incurred the composite outcome.

Among these, 288 (5.0%) had an incident diagnosis of AIDS, 16 (0.3%) had a diagnosis of acute

MI or stroke, 7 (0.1%) incurred end-stage renal disease, 40 (0.7%) incurred end-stage liver disease, and 89 (1.5%) died. Of the 10,169 participants initiating the EFV-based regimen with a median follow-up of 3.8 years (IQR 2.3-5.3), 1,097 (10.8%) experienced the composite outcome.

Among these, 671 (6.6%) received a diagnosis of AIDS, 61 (0.6%) had a diagnosis of acute MI or stroke, 26 (0.3%) had end-stage renal disease, 116 (1.1%) incurred end-stage liver disease,

223 (2.2%) died. A total of 3,104 (19.4%) among 15,993 participants were lost to follow-up with

13.4% in an InSTI-based regimen (782/5,824) and 22.8% in the EFV-based regimen

(2,322/10,169). Crude risk of loss to follow-up is shown in Figure 5.2. The 6-year risk of loss to follow-up is 38.8% for the InSTI group, and 34.9% for the EFV group.

After accounting for baseline confounding and differential loss to follow-up using inverse probability weighting, the intention-to-treat hazard ratio of the composite outcome comparing those who initiated an InSTI-based regimen with those who initiated the EFV-based regimen was

1.08 (95% CI: 0.97, 1.19). The intention-to-treat 6-year risk of the composite outcome was

14.6% among those initiating an InSTI-based regimen, and 14.3% among those initiating the

EFV-based regimen, corresponding with a 6-year risk difference of 0.3 percentage point (95%

CI: -2.7, 3.3) (Table 5.4). The Kaplan-Meier risk curve for adjusted intention-to-treat analyses across 6-year follow-up was depicted in the left panel of Figure 5.3.

Fifty-four percent of the participants initiating an InSTI-based regimen (3,140/5,824) and

68% of the participants initiating the EFV-based regimen (6,964/10,169) had a treatment change before incurring the composite outcome, loss to follow-up or completing the study. Crude risk of treatment changes is shown in Figure 5.2. The 6-year risk of treatment change was 85.9% for the

InSTI group, and 83.2% for the EFV group. After accounting for baseline confounding, differential loss-to-follow-up and treatment changes, the adjusted per-protocol hazard ratio of the composite outcome comparing those who initiated and remained on an InSTI-based regimen with those who initiated and remained on the EFV-based regimen was 1.09 (95% CI: 0.96, 1.25).

The adjusted per-protocol 6-year risk of the composite outcome was 12.2% among those initiating and remaining on an InSTI-based regimen, and 11.9% among those initiating and remaining on the EFV-based regimen, corresponding with a 6-year risk difference was 0.3 percentage point (95% CI: -3.0, 3.7) (Table 5.4). The Kaplan-Meier risk curve for adjusted per- protocol analyses across 6-year follow-up was depicted in the right panel of Figure 5.3.

Effect estimates were similar under the secondary analyses for restricting to most relevant clinical events, restricting to serious non-AIDS events, including confounding by cohort, and stratifying by baseline CD4 cell count (see Table 5.5).

5.4 Discussion

Using this large collaboration of North American HIV cohorts, we found similar effects on a composite clinical outcome between InSTI-based and EFV-based initial antiretroviral therapy in both intention-to-treat and per-protocol analyses. This finding is consistent with previous literature54, but with a broader spectrum of clinical outcomes, longer follow-up, and larger sample size, these results provide more assurance of the safety and clinical effectiveness of the initiation of InSTI-based antiretroviral therapy. But they do not suggest that InSTI-based regimens have better clinical outcomes than the EFV-based regimen. Furthermore, our study showed that, in this sample with median age of 40 years, the 6-year risk of clinical outcomes was substantial for both InSTI and EFV groups (14% in intention-to-treat analyses and 12% in per- protocol analyses).

Existing randomized trials focused on short-term biomarkers suggested that patients initiating InSTI-based regimens experienced more rapid and potent antiretroviral activity, and fewer adverse events and drug interactions compared with those initiating other regimens.25–29,31–

33,91 Direct comparisons of efficacy and safety of InSTI-based regimens with EFV-based regimens have been performed in several randomized controlled trials: STARTMRK (RAL vs.

EFV)25,37–40, SINGLE (DTG vs. EFV)28,41, and Study 102 (EVG vs. EFV)31,42,43,91. STARTMRK trial found non-inferiority of the RAL-based regimen to the EFV-based regimen on viral suppression at week 48 and 96, but also suggested significantly fewer drug-related clinical

adverse events in the RAL group.25,37–40 At week 240, the study showed RAL induced significantly better viral suppression.92 In the SINGLE study directly comparing DTG plus abacavir/lamivudine (ABC/3TC) with the co-formulated EFV/FTC/TDF, DTG-based regimen was found to have a more favorable tolerability profile and lower rate of treatment discontinuations. 28,41 The Study 102 trial, which compared co-formulated EVG/COBI/FTC/TDF versus co-formulated EFV/FTC/TDF, showed a significantly better immunological recovery among the EVG group, but comparable rate of viral suppression and similar number of treatment discontinuations. 31,42,43,91 Overall, these trials suggested that InSTIs had a better virological response and tolerability with fewer treatment discontinuations, in contrast with our finding that

InSTIs lead to a higher 6-year risk of treatment changes and had comparable long-term clinical effects with EFV. The discrepancies between clinical trials and real-world observational data may be due in part to the shorter duration of follow-up in clinical trials because some adverse effects may only occur after prolonged treatment exposure. Therefore, further research is warranted to incorporate and investigate the reasons for treatment changes, and provide additional insights on the tolerability of InSTI-based initial antiretroviral therapy in a real-world observational setting.

Our study is subject to limitations. First, one key challenge in the analysis of observational data is that the treatment was not randomly assigned. That means, all baseline confounders should be measured and adjusted for to achieve comparability of treatment groups to emulate a randomized trial. It is possible that residual confounding was present. Second,

ESRD and ESLD outcomes were based on laboratory tests. In addition, acute MI or stroke were based on diagnosis data because adjudicated events in NA-ACCORD were not available for all cohorts throughout the follow-up period. Thus, our outcome data may suffer from measurement

error. Future studies on validated outcomes including validated MI and cancer outcomes are a worthwhile addition. Third, we did not assess additional chronic disease outcomes or their proxies, such as weight gain or metabolic disorders, for which there is some evidence of enhanced risk when comparing InSTI-based to other regimens. Though there are currently both trial and observational data supporting the association of some InSTI-based regimens with more rapid weight gain59,61,95,96, and increases in waist circumference101, these may be intermediates of hard outcomes which we did assess in this analysis. We further felt that quantifying these relationships would be beyond the scope of this analysis and may merit separate study. Last, here we only estimated the overall clinical effects of InSTI class, and did not distinguish the clinical outcomes for each distinct InSTI agent because such detailed analyses would require a larger study.

There are several strengths to our study that are worth noting. We used a large observational pooling project and emulated a randomized trial.94 First, the size, breadth, prolonged follow-up and representativeness of the NA-ACCORD data allowed us to more accurately quantify the clinical effect of InSTI-based initial antiretroviral therapy with adjustment for many potential confounders. Second, we restricted analyses to participants starting antiretroviral therapy so that prior use of ART could not bias the results.72 We also set time zero of our analyses to align with time zero of a randomized trial when eligibility criteria are met and treatment therapies were assigned. Further, we adopted modern causal and statistical approaches to perform intention-to-treat and per-protocol analyses to ameliorate the concern of baseline confounding, differential loss to follow-up, and treatment discontinuations and switches, which are difficult to address using traditional statistical methods, and provide comprehensive clinical evidence about the effectiveness of InSTIs. Our analytical approaches also allowed us to

estimate the more patient-relevant measures of interest, such as absolute risks and risk differences, in addition to hazard ratios.

In conclusion, employing a rigorous methodological approach, our study indicates a similar 6-year risk of composite clinical outcomes for initial InSTI-based regimens and the EFV- based regimen by leveraging a large multisite observational cohort collaboration in the U.S. and

Canada. Given the lack of existing or forthcoming longer-term randomized evidence, prospective cohort studies in real-world settings contribute to the evidence-base and supplement findings from randomized clinical trials. With principled causal and statistical methods, one can leverage observational data to obtain more accurate effect estimates, provide the best available evidence, and address timely and important questions that inform policy decision making and treatment guidelines.

Table 5.1. Incident type-specific AIDS-defining illnesses among 288 adults who initiated an InSTI-based regimen and incurred AIDS and 671 adults who initiated an EFV-based regimen and incurred AIDS as their first composite event between July 2009 and December 2016 in the NA-ACCORD, the United States and Canada InSTI-based EFV-based First diagnosis of AIDS-defining illnesses after regimena regimena ART initiation No. (%) No. (%) Candidiasis of bronchi, trachea, or lungs 0 (0.0) 0 (0.0) Candidiasis esophageal 27 (9.4) 55 (8.2) Candidiasis AIDS-defining unspecified 0 (0.0) 2 (0.3) Cervical cancer, invasive 5 (1.7) 6 (0.9) Coccidioidomycosis, disseminated or 0 (0.0) 5 (0.8) extrapulmonary Cryptococcosis, extrapulmonary 15 (5.2) 26 (3.9) Cryptosporidiosis, chronic intestinal (greater 6 (2.1) 13 (1.9) than one month's duration) Cytomegalovirus disease (other than liver, 2 (0.7) 3 (0.4) spleen or lymph nodes) Cytomegalovirus retinitis (with loss of vision) 15 (5.2) 19 (2.8) Cytomegalovirus, not retinitis AIDS-defining 2 (0.7) 2 (0.3) unspecified Cytomegalovirus, AIDS-defining unspecified 6 (2.8) 13 (1.9) Encephalopathy, HIV related 10 (3.5) 19 (2.8) Herpes simplex: chronic ulcer(s) (more than 1 30 (10.4) 34 (5.1) month in duration); or bronchitis, pneumonitis, or esophagitis Histoplasmosis, disseminated or 5 (1.7) 7 (1.0) extrapulmonary Isosporiasis, chronic intestinal (more than 1 0 (0.0) 0 (0.0) month in duration) Kaposi sarcoma 23 (8.0) 80 (11.9) Non-Hodgkin’s Lymphoma, Burkitt's 9 (3.1) 10 (1.5) Non-Hodgkin’s Lymphoma, immunoblasti 2 (0.7) 0 (0.0) Non-Hodgkin’s Lymphoma, primary, of brain 1 (0.4) 3 (0.5) Non-Hodgkin’s Lymphoma, unspecified 13 (4.5) 35 (5.2) Mycobacterium avium complex or M kansasii, 10 (3.5) 17 (2.5) disseminated or extrapulmonary Mycobacterium tuberculosis, any site 27 (9.4) 79 (11.8) (pulmonary or extrapulmonary) Mycobacterium, other species or unidentified 7 (2.4) 29 (4.3) species, disseminated or extrapulmonary Pneumocystis pneumonia (PCP) 33 (11.5) 106 (15.8) Pneumonia, recurrent 10 (3.5) 17 (2.5) Progressive multifocal leukoencephalopathy 4 (1.4) 5 (0.8) Salmonella septicemia, recurrent 0 (0.0) 0 (0.0)

Toxoplasmosis of brain 6 (2.1) 13 (1.9) Wasting syndrome due to HIV 19 (6.6) 67 (10.0) Non-Hodgkin’s lymphoma, large cell 1 (0.4) 6 (0.9) Total 288 (100.0) 671 (100.0) Abbreviations: InSTI, integrase strand transfer inhibitor; EFV, efavirenz; NA-ACCORD, the North American AIDS Cohort Collaboration on Research and Design a Both regimens included the same backbone of tenofovir disoproxil fumarate [or tenofovir alafenamide], and emtricitabine

Table 5.2. Missing baseline covariates among 15,993 HIV-infected adults initiating an InSTI- based regimen or an EFV-based regimen between July 2009 and December 2016 in the NA- ACCORD, the United States and Canada No. (%) of missingness InSTI-based EFV-based Characteristic regimena regimena Overall (n=5,824) (n=10,169) (n=15,993) Age 0 (0.0) 0 (0.0) 0 (0.0) Female 0 (0.0) 0 (0.0) 0 (0.0) Black race 135 (2.3) 317 (3.1) 452 (2.8) Hispanic ethnicity 135 (2.3) 317 (3.1) 452 (2.8) Body mass index 143 (2.5) 273 (2.7) 416 (2.6) Injection drug use 0 (0.0) 0 (0.0) 0 (0.0) Male-to-male sexual contact 0 (0.0) 0 (0.0) 0 (0.0) Heterosexual behavior 0 (0.0) 0 (0.0) 0 (0.0) Previous AIDS diagnosis 0 (0.0) 0 (0.0) 0 (0.0) Hepatitis B 28 (0.5) 30 (0.3) 58 (0.4) Hepatitis C 0 (0.0) 0 (0.0) 0 (0.0) Previous depression diagnosis 30 (0.5) 31 (0.3) 61 (0.4) Previous anxiety diagnosis 52 (0.9) 78 (0.8) 130 (0.8) Diabetes mellitus 2 (0.0) 1 (0.0) 3 (0.0) Hypertension 25 (0.4) 42 (0.4) 67 (0.4) Statin prescription 25 (0.4) 27 (0.3) 52 (0.3) Elevated total cholesterol 28 (0.5) 30 (0.3) 58 (0.4) Baseline CD4 countb 1,034 (17.8) 2,425 (23.8) 3,459 (21.6) Baseline viral loadc 1,167 (20.0) 2,769 (27.2) 3,936 (24.6) Calendar year at initiation 0 (0.0) 0 (0.0) 0 (0.0) Abbreviations: HIV, human immunodeficiency virus; InSTI, integrase strand transfer inhibitor; EFV, efavirenz; AIDS, acquired immunodeficiency syndrome a Both regimens included the same backbone of tenofovir disoproxil fumarate [or tenofovir alafenamide], and emtricitabine b Baseline CD4 count was captured at the closest date to ART initiation within the window from 90 days before to 7 days after ART initiation c HIV viral load was captured at the closest date to ART initiation within the window from 90 days before to 7 days after ART initiation

Table 5.3. Characteristics at ART initiation of 15,993 HIV-infected adults initiating an InSTI- based regimen or an EFV-based regimen between July 2009 and December 2016 in the NA- ACCORD, overall and by treatment regimens, the United States and Canada No. (%) of Participants InSTI-based EFV-based Characteristic regimena regimena Overall (n=5,824) (n=10,169) (n=15,993) 40.0 (30.0- Age, median (IQR), yrs 37.0 (28.0-48.0) 41.0 (31.0-50.0) 50.0) Female 894 (15.3) 1,101 (10.8) 1,995 (12.5) Black race 2,351 (40.4) 4,611 (45.3) 6,962 (43.5) Hispanic ethnicity 722 (12.4) 1,350 (13.3) 2,072 (13.0) 25.1 (22.3- Body mass index, median (IQR) 25.1 (22.3-28.7) 25.1 (22.3-28.6) 28.6) Injection drug use 566 (9.7) 1,033 (10.2) 1,599 (10.0) Male-to-male sexual contact 3,209 (55.1) 4,523 (44.5) 7,732 (48.4) Heterosexual behavior 1,349 (23.2) 2,030 (20.0) 3,379 (21.1) Previous AIDS diagnosis 480 (8.2) 735 (7.2) 1,215 (7.6) Hepatitis B 214 (3.7) 426 (4.2) 640 (4.0) Hepatitis C 564 (9.7) 1,123 (11.0) 1,687 (10.6) Previous depression diagnosis 859 (14.8) 1,038 (10.2) 1,897 (11.9) Previous anxiety diagnosis 692 (11.9) 727 (7.2) 1,419 (8.9) Diabetes mellitus 284 (4.9) 545 (5.4) 829 (5.2) Hypertension 817 (14.0) 1,792 (17.6) 2,609 (16.3) Statin prescription 340 (5.8) 806 (7.9) 1,146 (7.2) Elevated total cholesterol 224 (3.9) 508 (5.0) 732 (4.6) Baseline CD4 count, median 349.0 (173.0- 323.0 (178.0- 332.0 (177.0- (IQR), cells/μL 524.0) 461.0) 485.0) 38440.0 Baseline viral load, median 40432.0 (8829.0- 36754.5 (7482.5- (8017.0- (IQR), copies/mL 139000.0) 123182.0) 128000.0) Calendar year at initiation, 2011 (2010- 2012 (2010- median (IQR) 2014 (2013-2015) 2012) 2014) Abbreviations: ART, antiretroviral therapy; HIV, human immunodeficiency virus; InSTI, integrase strand transfer inhibitor; EFV, efavirenz; NA-ACCORD, the North American AIDS Cohort Collaboration on Research and Design; IQR, interquartile range; AIDS, acquired immunodeficiency syndrome a Both regimens included the same backbone of tenofovir disoproxil fumarate [or tenofovir alafenamide], and emtricitabine

Table 5.4. Estimated 6-year risk differences and hazard ratios of the composite outcome comparing adults initiating an InSTI-based regimen with those initiating an EFV-based regimen between July 2009 and December 2016 in the NA-ACCORD, the United States and Canada Analysis No. of Person- No. of 6-year 6-year Risk Hazard Ratio Participants years Outcomesa Risk, % Difference (95% CI) (95% CI), % Crudeb EFV 10,169 38029.3 1,097 14.2 0 1 InSTI 5,824 13240.8 440 13.1 -1.1 (-4.1 to 0.99 (0.88 to 1.9) 1.10) Intention-to-treatc EFV 10,169 38029.3 1,097 14.3 0 1 InSTI 5,824 13240.8 440 14.6 0.3 (-2.7 to 1.08 (0.97 to 3.3) 1.19) Per-protocold EFV 10,169 25100.8 672 11.9 0 1 InSTI 5,824 9313.0 325 12.2 0.3 (-3.0 to 1.09 (0.96 to 3.7) 1.25) Abbreviations: InSTI, integrase strand transfer inhibitor; EFV, efavirenz; NA-ACCORD, the North American AIDS Cohort Collaboration on Research and Design; CI, confidence interval a Composite outcome included AIDS-defining illnesses, acute myocardial infarction or stroke, end-stage renal diseases, end-stage liver diseases, or death b Crude analysis did not account for baseline confounding, different loss to follow-up and treatment changes (uncensored). c Intention-to-treat analyses accounted for baseline confounding and differential loss to follow-up d Per-protocol analyses accounted for baseline confounding, differential loss to follow-up, and treatment changes (i.e., treatment discontinuations or switches)

Table 5.5. Secondary analyses for the effect on composite outcome comparing adults initiating an InSTI-based regimen with those initiating an EFV-based regimen between July 2009 and December 2016 in the NA-ACCORD, the United States and Canada 6-year Risk Difference (95% Hazard Ratio (95% CI) CI), % Secondary analysis 1 a AIDS and death Intention-to-treat b -0.4 (-3.8 to 3.0) 1.07 (0.96 to 1.20) Per-protocol c 0.9 (-2.7 to 4.6) 1.12 (0.98 to 1.29) Serious non-AIDS events Intention-to-treat 0.9 (-5.1 to 6.9) 1.16 (0.92 to 1.47) Per-protocol 0.2 (-7.3 to 8.0) 1.29 (0.96 to 1.73) Secondary analysis 2 d Intention-to-treat 0.2 (-2.5 to 3.0) 1.06 (0.96 to 1.18) Per-protocol 1.1 (-2.3 to 4.6) 1.12 (0.98 to 1.27) Secondary analysis 3 CD4 cell count ≤200 cells/μL Intention-to-treat 2.1 (-3.4 to 7.6) 0.96 (0.81 to 1.13) Per-protocol 0.4 (-6.0 to 6.9) 0.99 (0.81 to 1.22) CD4 cell count >200 cells/μL Intention-to-treat 1.1 (-2.2 to 4.4) 1.22 (0.99 to 1.50) Per-protocol 0.0 (-3.8 to 3.8) 1.20 (0.96 to 1.51) Abbreviations: InSTI, integrase strand transfer inhibitor; EFV, efavirenz; NA-ACCORD, the North American AIDS Cohort Collaboration on Research and Design; CI, confidence interval. a In secondary analysis 1, we restricted the composite outcome only to AIDS or death b Intention-to-treat analyses accounted for baseline confounding and differential loss-to-follow- up. We compared the InSTI group to the EFV group for the effect estimate c Per-protocol analyses accounted for baseline confounding, differential loss-to-follow-up, and treatment changes (i.e., treatment discontinuations or switches). We compared the InSTI group to the EFV group for the effect estimate d In secondary analysis 2, we additionally adjusted for individual cohorts as potential confounding

Figure 5.1. Secular trend in initiation of InSTI- versus EFV-based regimens between July 2009 and December 2016 in the NA-ACCORD, the United States and Canada. Abbreviations: InSTI, integrase strand transfer inhibitor; EFV, efavirenz; NA-ACCORD, the North American AIDS Cohort Collaboration on Research and Design; ART, antiretroviral therapy.

Figure 5.2. Crude risks of loss-to-follow-up and treatment changes during 6-year follow-up period among 15,993 HIV-infected adults initiating an InSTI-based regimen or an EFV-based regimen between July 2009 and December 2016 in the NA-ACCORD. Composite events were censored. Treatment changes that were considered allowable exceptions and thus were not included in this analysis were: 1) a change from one InSTI-based regimen to another InSTI- based regimen (for instance, from a RAL-based regimen to a DTG-based regimen); 2) a change from the EFV-based regimen to a non-nucleoside reverse transcriptase inhibitor (NNRTI)-based regimen of rilpivirine, TDF and FTC; and 3) switch between TDF and TAF.Abbreviations: InSTI, integrase strand transfer inhibitor; EFV, efavirenz; NA-ACCORD, the North American AIDS Cohort Collaboration on Research and Design; ART, antiretroviral therapy.

Figure 5.3. Risk of the composite outcome (AIDS-defining illnesses, acute myocardial infarction or stroke, end-stage renal diseases, end-stage liver diseases, or death) among 15,993 HIV- infected adults initiating an InSTI-based regimen or an EFV-based regimen between July 2009 and December 2016 in the NA-ACCORD. The left panel represent intention-to-treat analyses that accounted for baseline confounding and differential loss-to-follow-up, and the right panel represent per-protocol analyses that accounted for baseline confounding, differential loss to follow-up, and treatment changes.

CHAPTER 6: CONCLUSION

6.1 Overview

New InSTI-based regimens have been recommended as first-line antiretroviral therapy

(ART) for adults with human immunodeficiency virus. But evidence on long-term clinical effectiveness of InSTI-based regimens remains limited. In this work, we explored the long-term effects of InSTI on virologic and clinical outcomes using the large pooled observational data. We used data from the prospective cohort HIV collaboration to emulate a randomized trial through a sophisticated study design, and estimated observational analogs of both the intention-to-treat and per-protocol effects to add needed information to the evidence base about the long-term comparative effectiveness of InSTI-based regimens.

In Aim 1, we estimated the consecutive proportions virally suppressed among those who initiated the InSTI-based regimens and those who initiated the EFV-based regimens across 7- year follow-up period. These estimates were aimed to first replicate and then extend the results of randomized trials, and examine if InSTI-based regimens have a better viral response compared with EFV-based regimens. The results may be of particular interest to clinical trialists and clinicians, and give them a sense how observational data can be used to replicate and extend randomized evidence.

The results from Aim 2 are particularly important from the patient-centered perspective.

The HIV randomized trials are primarily focused on the surrogate biomarkers including CD4 cell counts and HIV viral loads. However, the clinicians and the patients are more concerned about

the clinical effects of these ART drugs. Although InSTI-based regimens have been recommended as first-line ART treatment, this knowledge gap about the clinical effectiveness of

InSTI has not been adequately addressed in existing research. We used pooled data from HIV cohort studies in the United States and Canada, and employed new epidemiology study designs, coupled with causal and statistical inference methods to assess the impact of InSTI-based regimens on clinical outcomes including AIDS-defining illnesses, serious non-AIDS events and all-cause mortality over prolonged follow-up.

6.2 Study findings

In Chapter 4, we estimated the effects of InSTI-based regimens on viral suppression over

7-years of follow-up. We found that initiating InSTI-based regimens led to more rapid decline in

HIV viral load than initiating EFV-based regimens among adults living with HIV, especially during the first quarter (0.25 years) since ART initiation, which aligns with the results of randomized trials that compared the InSTI-class drug (i.e., RAL, EVG, DTG) with EFV.

However, both the intention-to-treat and per-protocol analyses showed that InSTI-based and

EFV-based regimens showed similar longer-term (7-year) effects on viral suppression. To our knowledge, this is the first study to have examined the virologic effect of InSTI with such a long follow-up duration. The most relevant study has shown similar results over a 2.5-year follow- up.102

In Chapter 5, we estimated the effects of InSTI-based regimens on clinical outcomes including AIDS, death, and serious non-AIDS events over 6-years of follow-up. Similar to Aim

1, we used observational data to mimic a randomized trial, and estimated the observational analogs of intention-to-treat and per-protocol effects. To our knowledge, this is the first study

that examined the effects of InSTI on a broad spectrum of clinical outcomes over such prolonged follow-up. As a result, we found that InSTI-based and EFV-based initial ART regimens had similar 6-year effects on the composite clinical outcome including AIDS-defining illnesses, serious non-AIDS events, and death. These results contradicted our prior expectation or hypothesis that the newer class InSTI should have a more favorable clinical profile compared with EFV. As randomized trials are mainly focused on short-term surrogate biomarkers, our results raise the question whether current HIV randomized trials are sufficient to provide a strong evidence base on the effectiveness of HIV drugs, and whether and how policy makers might use both randomized and observational evidence for improved decision making. With advances in statistical methods and increased availability of observational data, more research should be devoted to using observational data to estimate the clinical effectiveness of HIV ART drugs under the real-world settings.

In summary, our studies suggest although InSTI presented a more rapid viral potency compared with EFV, the long-term effects of InSTI-based initial regimens on virologic and clinical outcomes were similar to EFV-based regimens. In addition, our data showed the increased popularity in initiating InSTI-based regimens compared with EFV-based regimens since 2009, as expected. However, what is unexpected is that both our studies have shown that patients that initiated InSTI-based regimens had a larger proportion of loss to follow-up and treatment changes (i.e., treatment switches and discontinuations) compared with those initiated

EFV-based regimens, which contradicted the randomized evidence that showed InSTI-based regimens had fewer treatment discontinuations. This discrepancy may be because: 1) clinical trials have relatively shorter follow-up duration but some adverse effects may only occur after prolonged treatment exposure which could only be observed with observational data over

prolong follow-up; 2) the differences in study populations between the clinical trials and our observational study, as clinical trials often excluded a portion of sample due to strict eligibility criteria and monitored patients under more controlled settings, and our pooled observational data presented a broad, more heterogeneous population; 3) there were some prognostic factors for treatment discontinuations that we could not adjust for in our observational data; and 4) potential measurement error during the electronic medical recording for treatment switches and discontinuations in our observational studies. Overall, the main findings from our studies are that

InSTI-based and EFV-based regimens have a similar long-term effects on virologic and clinical outcomes, which should be of strong interest to a broad party of stakeholders, including health providers, policy makers, patients and pharmaceutical companies, especially in the current context that the InSTI-class drug dolutegravir are being rolled out globally and extensive research is being conducted over InSTI for maintenance therapy and pre-exposure prophylaxis.

6.3 Strengths

Our work resulted in a valuable addition to current literature. This work is the first and longest study that examined the long-term comparative effectiveness of InSTI-based regimens on comprehensive outcomes including viral suppression and a wide variety of clinical outcomes.

Our study used the large pooled HIV cohort collaboration (NA-ACCORD) to address our study aims. The NA-ACCORD presents a heterogeneous and representative population living with HIV in the United States and Canada, which enables us to generate directly applicable and policy-relevant effect estimates of InSTI among patients living with HIV under routine care beyond traditional randomized trials. The rich and broad sample allows us to more precisely estimate the impact of InSTI-based regimens compared with studies of single cohorts. In

addition, measurement of a multitude of covariates give us a unique opportunity to adjust for many important confounders and thereby estimate more accurate effects.

The new user design72 and sophisticated analytic approach for this work also add to its strength. We restricted analyses to participants initiating antiretroviral therapy so that prior use of

ART could not bias the results, and this approach also allows us to mimic a randomized trial that compares InSTI-based regimens and EFV-based regimens. In addition, we identified the baseline covariate measurement at ART initiation, defined the eligibility criteria, and set time zero as the

ART initiation, which align with the settings of randomized trials. Further, we adopted appropriate statistical approaches to perform the observational analogs of intention-to-treat and per-protocol analyses by adjusting for a multitude of baseline and time-varying covariates to mitigate the concern of baseline confounding, differential loss to follow-up, and treatment discontinuations and switches that could occur in the observational settings. These concerns could not be addressed using traditional statistical methods (e.g., straightforward Cox proportional hazard models). The observational analogs of intention-to-treat and per-protocol effects provide comprehensive clinical evidence about the effectiveness of InSTIs. Last, we used multiple imputation to impute the baseline missing covariates. Although missing, or “dark” data are always a concern, recently epidemiologists and statisticians have been concentrating attention of this issue.103 However, so-called complete-case analysis remains the main approach in epidemiologic studies, which could bias the results if the missingness is not completely at random. Using multiple imputation to impute the missing data enables us to generate more efficient and valid effect estimates, by relaxing our assumption to the data being missing at random, given the multitude of measured covariates.

6.4 Limitations

Using observational data is well-suited and inescapable for our aims, in the absence of randomized evidence existing or forthcoming. However, our study is also subject to several limitations. First, a key limitation in the analysis of observational data is that the treatment was not randomly assigned. To achieve comparability of treatment groups all confounders must be measured and appropriate adjustments made. However, such assumptions are unverifiable using observational data. Although we adjusted for a large number of likely confounders, it is still possible that uncontrolled confounding remains. For example, education level is not recorded in the NA-ACCORD, and it may be a potential confounder for the effects of InSTI on clinical outcomes. Further, varying prescribing behaviors of ART drugs by different clinicians are not possible to be measured, but could potentially impact the effects of InSTI.

Second, our study may suffer from potential measurement error. In Aim 1, we chose to use the cutoff of HIV RNA <200 copies/mL instead of <50 copies/mL that was used in most published randomized trials, because in our study there were a portion of HIV viral load measurement with lower limit of detection greater than 50 copies/mL. Our choice may lead to potential misclassification and result in bias in our effects estimates. In Aim 2, some clinical outcomes including acute MI or stroke, ESRD, and ESLD are only based on laboratory tests and diagnosis, and are not fully adjudicated, which could result in possible outcome misclassification.

Third, apart from the comparability of treatment groups (known as exchangeability90) and no measurement error, identification of causal intention-to-treat and per-protocol effects also can be identified with following assumptions of treatment version irrelevance (namely, any versions of treatment are irrelevant)104–106, positivity (namely, there is a nonzero probability of being

observed for every combination of values of treatment and covariate histories for treatment models, and for every combination of values of censoring and covariate histories for censoring models)107, no interference between patients included in the study108, no measurement error and correct specification of statistical models (e.g., treatment, censoring). Most of these assumptions are untestable. Any violations of these assumptions could lead to biased estimates of our causal effects of InSTI. For example, as we included many baseline covariates and time-varying covariates for our treatment and censoring models, it is possible that positivity assumption is not met. Fortunately, the large sample size, and the somewhat haphazard nature of treatment and censoring, combine to mitigates this concern.

Model specification may also be an issue, as we used parametric models to predict the probability of treatment and censoring. The parametric model assumptions are probably violated.

However, by using flexible model techniques, such as quadratic splines terms for continuous covariates, we hope to well-approximate the underlying relationships. We also could not test the assumption of treatment version irrelevance. In our work, we only have the prescribing record of

ART drugs from electronic medical records, and however, we do not have the available data on the true adherence from each patient. We are bolstered by the fact that both the InSTI- and EFV- based regimens come in standard fixed-dose preparations. It is possible that we treat all patients on InSTI as simply taking InSTI, but different patients may have varying adherence patterns, which are unobserved but can lead to different levels of virologic and clinical severity. Future work is needed to explore this possibility.

Fourth, in both Aim 1 and Aim 2, we only estimated the overall effects of InSTI class, and did not estimate single components of each InSTI-class drug (i.e., RAL, EVG, DTG). We recognize that there are potential differences in advantages and disadvantages with regards to

these 3 distinct InSTI-class drug. However, such detailed individual analyses would require a larger study, and the sets of covariates required to make the individual drugs exchangeable would be difficult to identify.

Fifth, in our per-protocol analyses, we censored all the treatment switches and discontinuations, as our observational data did not record the reasons for stopping or discontinuing the ART regimens. However, the per-protocol analyses usually do not censor when patients stop treatment due to clinically indicated toxicity because these treatment changes due to toxicity is often defined in the protocol specified prior to the study. Thus, our censoring approach may have some impact on the validity of the per-protocol effects. This assumption is testable only if we had additional data on reason for stopping or switching treatments. In addition, when dealing with per-protocol analyses in Aim 2, we censored patients at the minimum of loss to follow-up or treatment changes. Creating weights based on a combination of loss to follow-up and treatment changes assumes a common set of variables and association between loss to follow-up, treatment changes and the composite clinical outcome of interest. this assumption is verifiable, but only given enough data to estimate the parameters separately

Last, in our study, we only explored the effectiveness of InSTI-based regimens. We did not take into account the other features including the drug interactions, convenience, cost- effectiveness and resistance mutations of the ART regimens. To better inform the HIV treatment guidelines and policy, our findings could be combined with other existing and forthcoming work on these features to provide a more comprehensive evidence base.

Even with these limitations, our work provides valuable evidence that adds to the existing literature on the effectiveness of InSTI-based regimens, and contributes to HIV treatment

decision and policy making. The novel study design along with sophisticated statistical methods employed make this work a powerful and important contributions to the HIV treatment literature.

6.5 Future directions

We have examined the long-term effects of InSTI-based initial regimens on virologic and clinical outcomes. However, as demonstrated in the limitations, in Aim 2, some components of the composite outcome (e.g., acute MI or stroke) were not adjudicated, and thus may suffer from measurement error. In addition, cancer outcomes, as an important endpoint, were not included because they were not fully validated during the analysis phase. Thus, future work could be aimed to examine the effect of InSTI-based regimens on validated outcomes including MI and cancer, which is a worthwhile addition to our work.

Second, as follow-up duration is prolonged and sample size on those initiating InSTI- based regimens increase, future studies are warranted to explore the clinical effects of individual

InSTI-based regimen. Each individual InSTI drugs may have different advantages and disadvantages, in terms of either the clinical effectiveness or other aspects such as cost, convenience, and resistance mutations. Thus, results from such individual detailed analyses can provide more informative perspectives for patients, clinicians and policy makers.

Third, cost-effectiveness research of the InSTI-based regimen is necessary to better inform HIV treatment guidelines and policy-making, especially in the resource limited setting.

Incorporating cost-effectiveness into our effect estimates will aid policy making by 1) providing estimates relevant in a resource-constrained setting, especially in the developing world and allowing policies to balance the benefit and cost of the InSTI-based regimens over other

regimens, and 2) providing evidence that could be used in negotiations with pharmaceutical manufacturers.

Fourth, our work shows that patients that initiated the InSTI-based regimens had higher proportion of loss to follow-up and treatment changes compared with those that initiated the

EFV-based regimen, which contradicted the findings from randomized trials. However, the reasons that caused such discrepancy have not been fully explored. Future work should be devoted to incorporating and investigating the reasons for treatment changes, and providing deeper insights on the tolerability of InSTI-based initial antiretroviral therapy in a real-world observational setting.

Last, future studies should also focus on other newly-developed InSTI-class drugs including bictegravir and cabotegravir, which have not been examined in the current work. These newly developed InSTI-class drugs have received increasing attention due to their potency and convenience. Assessment of long-term clinical effectiveness of these promising new drugs will also be necessary.

6.6 Public health significance and conclusion of our work

From the substantive perspective on HIV treatment, our studies showed that although

InSTI-based initial ART regimens had a more rapid viral response compared with the EFV-based initial ART regimen, the long-term effects on virologic and clinical outcomes were similar.

Along with the increasing studies showing the potential adverse effects associated with InSTI including neuropsychiatric toxicity64–66, neural tube defects36, weight gain59,61,95,96, our results on a relatively similar long-term effects bring into question whether InSTI-based regimens should

be favored over other regimens, especially in the current context that dolutegravir-based regimens are being rolled-out globally.

From the methodologic perspective, our work demonstrates that, in the absence existing or forthcoming longer-term randomized evidence, observational data in real-world settings can be used to replicate, extend, and complement findings from randomized clinical trials, and contribute to the evidence base. Using principled methods for causal and statistical inference, one can leverage observational data to obtain more accurate effect estimates, providing the best available evidence, and directly address important questions that inform treatment guidelines, policies, and decision making.

REFERENCES

1. Maartens G, Celum C, Lewin SR. HIV infection: Epidemiology, pathogenesis, treatment, and prevention. Lancet. 2014;384(9939):258-271.

2. Sierra S, Kupfer B, Kaiser R. Basics of the virology of HIV-1 and its replication. J Clin Virol. 2005;34(4):233-244.

3. Kuller LH, Tracy R, Belloso W, et al. Inflammatory and coagulation biomarkers and mortality in patients with HIV infection. PLoS Med. 2008;5(10):1496-1508.

4. Hsue PY, Scherzer R, Hunt PW, et al. Carotid Intima‐Media Thickness Progression in HIV‐Infected Adults Occurs Preferentially at the Carotid Bifurcation and Is Predicted by Inflammation. J Am Heart Assoc. 2012;1.

5. Marks MA, Rabkin CS, Engels EA, et al. Markers of microbial translocation and risk of AIDS-related lymphoma. AIDS. 2013;27:469-474.

6. Ancuta P, Kamat A, Kunstman KJ, et al. Microbial translocation is associated with increased monocyte activation and dementia in AIDS patients. PLoS One. 2008;3:e2516.

7. Andrade BB, Hullsiek KH, Boulware DR, et al. Biomarkers of inflammation and coagulation are associated with mortality and hepatitis flares in persons coinfected with HIV and hepatitis viruses. J Infect Dis. 2013;207:1379-1388.

8. Lu H, Surkan PJ, Irwin MR, et al. Inflammation and Risk of Depression in HIV: Prospective Findings From the Multicenter AIDS Cohort Study. Am J Epidemiol. 2019;188(11):1994-2003.

9. Lucas S, Nelson AM. HIV and the spectrum of human disease. J Pathol. 2015;235(2):229-241.

10. WHO. Global Health Observatory (GHO) data: HIV/AIDS. who.int.

11. Deeks SG, Lewin SR, Havlir D V. The end of AIDS: HIV infection as a chronic disease. Lancet. 2013;382:1525-1533.

12. World Health Organization. Consolidated guidelines on the use of antiretroviral drugs for treating and preventing HIV infection: recommendations for a public health approach. 2016;(June).

13. Gulick RM, Ribaudo HJ, Shikuma CM, et al. Three- vs Four-Drug Antiretroviral Regimens for the Initial Treatment of HIV-1 Infection. J Am Med Assoc. 2006;296(7):769-781.

14. Cassetti I, Madruga JVR, Suleiman JMAH, et al. The Safety and Efficacy of Tenofovir DF in Combination with Lamivudine and Efavirenz Through 6 Years in Antiretroviral- Naïve HIV-1—Infected Patients. HIV Clin Trials. 2007;8(3):164-172.

15. Staszewski S, Moreales-Ramirex J, Tashima KT, Rachils A. Efavirenz plus Zidovudine and Lamivudine, Efavirenz plus Indinavir, and Indinavir plus Zidovudine and Lamivudine in the Treatment of HIV-1 Infection in Adults. N Engl J Med. 1999;341(25):1865-1873.

16. Riddler SA, Haubrich R, DiRienzo AG, et al. Class-Sparing Regimens for Initial Treatment of HIV-1 Infection. N Engl J Med. 2008;358(20):2095-2106.

17. Robbins GK, Gruttola V De, Shafer RW, et al. Comparison of Sequential Three-Drug Regimens as Initial Therapy for HIV-1 Infection. N Engl J Med. 2003;349(24):2293-2303.

18. Gulick RM, Ribaudo HJ, Shikuma CM, et al. Triple-nucleoside regimens versus efavirenz-containing regimens for the initial treatment of HIV-1 infection. N Engl J Med. 2004;350(18):1850-1861.

19. Clifford DB, Evans S, Yang Y, et al. Long Term Impact of Efavirenz on Neuropsychological Performance and Symptoms in HIV Infected Individual (ACTG 5097s). HIV Clin Trials. 2009;10(6):343-355.

20. Clifford DB, Evans S, Yang Y, et al. Impact of efavirenz on neuropsychological performance and symptoms in HIV-infected individuals. Ann Intern Med. 2005;143(10).

21. Muñoz-Moreno JA, Fumaz CR, Ferrer MJ, et al. Neuropsychiatric symptoms associated with efavirenz: Prevalence, correlates and management. A neurobehavioral review. AIDS Rev. 2009;11(2):103-109.

22. Rihs TA, Begley K, Smith DE, et al. Efavirenz and chronic neuropsychiatric symptoms: A cross-sectional case control study. HIV Med. 2006;7(8):544-548.

23. Mollan KR, Smurzynski M, Eron JJ, et al. Association between Efavirenz as Initial Therapy for HIV-1 Infection and Increased Risk of Suicidal Ideation, Attempted, or Completed Suicide. Ann Intern Med. 2014;161(1):1-10.

24. Ciccarelli N, Fabbiani M, Di Giambenedetto S, et al. Efavirenz associated with cognitive disorders in otherwise asymptomatic HIV-infected patients. Neurology. 2011;76:1403- 1409.

25. Lennox JL, DeJesus E, Lazzarin A, et al. Safety and efficacy of raltegravir-based versus efavirenz-based combination therapy in treatment-naive patients with HIV-1 infection: a multicentre, double-blind randomised controlled trial. Lancet. 2009;374(9692):796-806.

26. Lennox JL, Dejesus E, Berger DS, et al. Raltegravir versus Efavirenz regimens in treatment-naive HIV-1-infected patients: 96-week efficacy, durability, subgroup, safety, and metabolic analyses. J Acquir Immune Defic Syndr. 2010;55(1):39-48.

27. Clotet B, Feinberg J, van Lunzen J, et al. Once-daily dolutegravir versus darunavir plus ritonavir for antiretroviral-naive adults with HIV-1 infection (FLAMINGO): 48 week results from the randomised open-label phase 3b study. Lancet HIV. 2014;383:2222-2231.

28. Walmsley SL, Antela A, Clumeck N, et al. Dolutegravir plus Abacavir–Lamivudine for the Treatment of HIV-1 Infection. N Engl J Med. 2013;369(19):1807-1818.

29. Eron JJ, Cooper DA, Steigbigel RT, et al. Efficacy and safety of raltegravir for treatment of HIV for 5 years in the BENCHMRK studies: Final results of two randomised, placebo- controlled trials. Lancet Infect Dis. 2013;13(7):587-596.

30. Sax PE, DeJesus E, Mills A, et al. Co-formulated elvitegravir, cobicistat, emtricitabine, and tenofovir versus co-formulated efavirenz, emtricitabine, and tenofovir for initial treatment of HIV-1 infection: A randomised, double-blind, phase 3 trial, analysis of results after 48 weeks. Lancet. 2012;379(9835):2439-2448.

31. Zolopa A, Sax PE, DeJesus E, et al. A randomized double-blind comparison of coformulated elvitegravir/cobicistat/emtricitabine/tenofovir disoproxil fumarate versus efavirenz/emtricitabine/tenofovir disoproxil fumarate for initial treatment of HIV-1 infection: analysis of week 96 results. J Acquir Immune Defic Syndr. 2013;63(1):96-100.

32. Sax PE, Pozniak A, Montes ML, et al. Coformulated bictegravir, emtricitabine, and tenofovir alafenamide versus dolutegravir with emtricitabine and tenofovir alafenamide, for initial treatment of HIV-1 infection (GS-US-380–1490): a randomised, double-blind,

multicentre, phase 3, non-inferiori. Lancet. 2017;390(10107):2073-2082.

33. Gallant J, Lazzarin A, Mills A, et al. Bictegravir, emtricitabine, and tenofovir alafenamide versus dolutegravir, abacavir, and lamivudine for initial treatment of HIV-1 infection (GS- US-380-1489): a double-blind, multicentre, phase 3, randomised controlled non-inferiority trial. Lancet. 2017;390(10107):2063-2072.

34. Günthard HF, Saag MS, Benson CA, et al. Antiretroviral Drugs for Treatment and Prevention of HIV Infection in Adults: 2016 Recommendations of the International Antiviral Society-USA Panel. JAMA. 2016;316(2):191-210.

35. Panel on Antiretroviral Guidelines for Adults and Adolescents DHHS. Guidelines for the Use of Antiretroviral Agents in Adults and Adolescents Living with HIV. 2018, March.

36. Zash R, Makhema J, Shapiro RL. Neural-Tube Defects with Dolutegravir Treatment from the Time of Conception. N Engl J Med. 2018:NEJMc1807653.

37. Rockstroh JK, DeJesus E, Lennox JL, et al. Durable Efficacy and Safety of Raltegravir Versus Efavirenz When Combined With Tenofovir/Emtricitabine in Treatment-Naive HIV-1–Infected Patients. JAIDS J Acquir Immune Defic Syndr. 2013;63(1):77-85.

38. Rockstroh JK, Lennox JL, Dejesus E, et al. Long-term treatment with raltegravir or efavirenz combined with tenofovir/emtricitabine for treatment-naive human immunodeficiency virus-1-infected patients: 156-week results from STARTMRK. Clin Infect Dis. 2011;53(8):807-816.

39. Lennox JL, DeJesus E, Berger DS, et al. Raltegravir versus efavirenz regimens in treatment-naive HIV-1-infected patients: 96-week efficacy, durability, subgroup, safety, and metabolic analyses. J Acquir Immune Defic Syndr. 2010;55(1):39-48.

40. DeJesus E, Rockstroh JK, Lennox JL, et al. Efficacy of Raltegravir Versus Efavirenz When Combined With Tenofovir/ Emtricitabine in Treatment-Naïve HIV-1–Infected Patients: Week-192 Overall and Subgroup Analyses From STARTMRK. HIV Clin Trials. 2012;13(4):228-232.

41. Walmsley S, Baumgarten A, Berenguer J, et al. Dolutegravir Plus Abacavir/Lamivudine for the Treatment of HIV-1 Infection in Antiretroviral Therapy-Naive Patients: Week 96 and Week 144 Results from the SINGLE Randomized Clinical Trial. J Acquir Immune Defic Syndr. 2015;70(5):515-519.

42. Zolopa AR, Berger DS, Lampiris H, et al. Activity of Elvitegravir, a Once‐Daily Integrase Inhibitor, against Resistant HIV Type 1: Results of a Phase 2, Randomized, Controlled, Dose‐Ranging Clinical Trial. J Infect Dis. 2010;201(6):814-822.

43. Wohl DA, Cohen C, Gallant JE, et al. A Randomized, Double-Blind Comparison of Single-Tablet Regimen Elvitegravir/Cobicistat/Emtricitabine/Tenofovir DF Versus Single-Tablet Regimen Efavirenz/Emtricitabine/Tenofovir DF for Initial Treatment of HIV-1 Infection: Analysis of Week 144 Results. J Acquir Immune Defic Syndr. 2014.

44. Van Lunzen J, Maggiolo F, Arribas JR, et al. Once daily dolutegravir (S/GSK1349572) in combination therapy in antiretroviral-naive adults with HIV: Planned interim 48 week results from SPRING-1, a dose-ranging, randomised, phase 2b trial. Lancet Infect Dis. 2012;12(2):111-118.

45. Stellbrink HJ, Reynes J, Lazzarin A, et al. Dolutegravir in antiretroviral-naive adults with HIV-1: 96-week results from a randomized dose-ranging study. Aids. 2013;27(11):1771- 1778.

46. Pozniak A. Surrogacy in HIV-1 clinical trials. Lancet. 1998;351(9102):536-537.

47. An evaluation of HIV RNA and CD4 cell count as surrogates for clinical outcome. Delta Coordinating Committee and Virology Group. AIDS. 1999;13(5):565-573.

48. Lin DY, Fischl MA, Schoenfeld DA. Evaluating the role of CD4-lymphocyte counts as surrogate endpoints in human immunodeficiency virus clinical trials. Stat Med. 1993;12(9):835-842.

49. Choi S, Lagakos SW, Schooley RT, Volberding PA. CD4+ lymphocytes are an incomplete surrogate marker for clinical progression in persons with asymptomatic HIV infection taking zidovudine. Ann Intern Med. 1993;118(9):674-680.

50. Hughes MD, Daniels MJ, Fischl MA, Kim S, Schooley RT. CD4 cell count as a surrogate endpoint in HIV clinical trials: a meta-analysis of studies of the AIDS Clinical Trials Group. AIDS. 1998;12(14):1823-1832.

51. Prentice RL. Surrogate endpoints in clinical trials: Definition and operational criteria. Stat Med. 1989;8(4):431-440.

52. Fleming TR, DeMets DL. Surrogate End Points in Clinical Trials: Are We Being Misled?

Ann Intern Med. 1996;125(7):605-613.

53. Fleming TR, Powers JH. Biomarkers and surrogate endpoints in clinical trials. Stat Med. 2012;31(25):2973-2984.

54. Cole SR, Edwards JK, Hall HI, et al. Incident AIDS or death after initiation of human immunodeficiency virus treatment regimens including raltegravir or efavirenz among adults in the United States. Clin Infect Dis. 2017;64(11):1591-1596.

55. Palella Jr. FJ, Baker RK, Moorman AC, et al. Mortality in the highly active antiretroviral therapy era: changing causes of death and disease in the HIV outpatient study. J Acquir Immune Defic Syndr. 2006;43(1):27-34.

56. Lau B, Gange SJ, Moore RD. Risk of non-AIDS-related mortality may exceed risk of AIDS-related mortality among individuals enrolling into care with CD4+ counts greater than 200 cells/mm3. J Acquir Immune Defic Syndr. 2007;44(2):179-187.

57. Neuhaus J, Angus B, Kowalska JD, et al. Risk of all-cause mortality associated with nonfatal AIDS and serious non-AIDS events among adults infected with HIV. AIDS. 2010;24(5):697-706. doi:10.1097/QAD.0b013e3283365356

58. Lee FJ, Carr A. Tolerability of HIV integrase inhibitors. Curr Opin HIV AIDS. 2012;7(5):422-428.

59. Norwood J, Turner M, Bofill C, et al. Weight gain in persons with HIV switched from efavirenz-based to integrase strand transfer inhibitor-based regimens. J Acquir Immune Defic Syndr. 2017;76(5):527-531.

60. Taramasso L, Ricci E, Menzaghi B, et al. Weight Gain: A Possible Side Effect of All Antiretrovirals. Open Forum Infect Dis. 2017;4(4):3-5.

61. Menard A, Meddeb L, Tissot-Dupont H, et al. Dolutegravir and weight gain: An unexpected bothering side effect? Aids. 2017;31(10):1499-1500.

62. Peñafiel J, De Lazzari E, Padilla M, et al. Tolerability of integrase inhibitors in a real-life setting. J Antimicrob Chemother. 2017;72(6):1752-1759.

63. Monteiro P, Perez I, Pich J, Gatell JM, Martínez E. Creatine kinase elevation in HIV-1-

infected patients receiving raltegravir-containing antiretroviral therapy: A cohort study. J Antimicrob Chemother. 2013;68(2):404-408.

64. De Boer MGJ, Van Den Berk GEL, Van Holten N, et al. Intolerance of dolutegravir- containing combination antiretroviral therapy regimens in real-life clinical practice. In: AIDS. Vol 30. ; 2016:2831-2834.

65. Hoffmann C, Welz T, Sabranski M, et al. Higher rates of neuropsychiatric adverse events leading to dolutegravir discontinuation in women and older patients. HIV Med. 2017;18(1):56-63.

66. Kheloufi F, Boucherie Q, Blin O, Micallef J. Neuropsychiatric events and dolutegravir in HIV patients: A worldwide issue involving a class effect. AIDS. 2017;31(12):1775-1777.

67. Dutertre M, Cuzin L, Demonchy E, et al. Initiation of antiretroviral therapy containing integrase inhibitors increases the risk of IRIS requiring hospitalization. J Acquir Immune Defic Syndr. 2017;76(1):e23-e26.

68. Psichogiou M, Basoulis D, Tsikala-Vafea M, Vlachos S, Kapelios CJ, Daikos GL. Integrase Strand Transfer Inhibitors and the Emergence of Immune Reconstitution Inflammatory Syndrome (IRIS). Curr HIV Res. 2017;15:405-410.

69. Horberg MA, Oakes AH, Hurley LB, Towner WJ, Chao CR, Silverberg MJ. Association of raltegravir use with long-term health outcomes in HIV-infected patients: an observational post-licensure safety study in a large integrated healthcare system. HIV Clin Trials. 2018.

70. Edwards JK, Cole SR, Hall HI, et al. Virologic suppression and CD4 + cell count recovery after initiation of raltegravir or efavirenz-containing HIV treatment regimens. AIDS. 2018;32(2):261-266.

71. Hernán MA, Alonso A, Logan R, et al. Observationals studies analyzed like randomized experiments: an application to postmenopausal hormone therapy and coronary heart disease. Epidemiology. 2013;19(6):766-779.

72. Brookhart MA. Counterpoint: The Treatment Decision Design. Am J Epidemiol. 2015;182(10):840-845.

73. Hernán MA, Robins JM. Per-protocol analyses of pragmatic trials. N Engl J Med.

2017;377(14):1391-1398.

74. INSIGHT START Study Group. Initiation of Antiretroviral Therapy in Early Asymptomatic HIV Infection. N Engl J Med. 2015;373(9):795-807.

75. CDC. 1993 Revised classification system for HIV infection and expanded surveillance case definition for AIDS among adolescentes and adults. MWR Morb Mortal Wkly Re. 1993;41:1-20.

76. Sun B, Perkins NJ, Cole SR, et al. Inverse-Probability-Weighted Estimation for Monotone and Nonmonotone Missing Data. Am J Epidemiol. 2018;187(3):585-591.

77. Robins JM, Hernán MA, Brumback B. Marginal structural models and causal inference in epidemiology. Epidemiology. 2000;11(5):550-560.

78. Robins JM, Finkelstein DM. Correcting for noncompliance and dependent censoring in an AIDS Clinical Trial with inverse probability of censoring weighted (IPCW) log-rank tests. Biometrics. 2000;56:779-788.

79. Cain LE, Cole SR. Inverse probability-of-censoring weights for the correction of time- varying noncompliance in the effect of randomized highly active antiretroviral therapy on incident AIDS or death. Stat Med. 2009;28(12):1725-1738.

80. Lau B, Cole SR, Gange SJ. Competing risk regression models for epidemiologic data. Am J Epidemiol. 2009;170(2):244-256.

81. Cole SR, Lau B, Eron JJ, et al. Estimation of the standardized risk difference and ratio in a competing risks framework: Application to injection drug use and progression to AIDS after initiation of antiretroviral therapy. Am J Epidemiol. 2015;181(4):238-245.

82. Bernard R. Fundamentals of Biostatistics, (5th Edition).; 2000.

83. Gange SJ, Kitahata MM, Saag MS, et al. Cohort profile: The North American AIDS Cohort Collaboration on Research and Design (NA-ACCORD). Int J Epidemiol. 2007.

84. Howe CJ, Cole SR, Westreich DJ, Greenland S, Napravnik S, Eron JJ. Splines for trend analysis and continuous confounder control. Epidemiology. 2011;22(6):874-875.

85. White IR, Royston P, Wood AM. Multiple imputation using chained equations: Issues and guidance for practice. Stat Med. 2011;30(4):377-399.

86. White IR, Royston P. Imputing missing covariate values for the Cox model. Stat Med. 2009;28(15):1982-1998.

87. Huque MH, Carlin JB, Simpson JA, Lee KJ. A comparison of multiple imputation methods for missing data in longitudinal studies 01 Mathematical Sciences. BMC Med Res Methodol. 2018;18(1):1-16.

88. Hubbard AE, Ahern J, Fleischer NL, et al. To GEE or Not to GEE: Comparing Population Average and Mixed Models for Estimating the Associations Between Neighborhood Risk Factors and Health. Source Epidemiol. 2016;21(4):467-474.

89. Lu H, Cole SR, Hall HI, et al. Generalizing the per-protocol treatment effect: The case of ACTG A5095. Clin Trials. 2019;16(1):52-62.

90. Cole SR, Hernán MA. Constructing inverse probability weights for marginal structural models. Am J Epidemiol. 2008;168(6):656-664. =

91. Sax PE, DeJesus E, Mills A, et al. Co-formulated elvitegravir, cobicistat, emtricitabine, and tenofovir versus co-formulated efavirenz, emtricitabine, and tenofovir for initial treatment of HIV-1 infection: A randomised, double-blind, phase 3 trial, analysis of results after 48 weeks. Lancet. 2012;379(9835):2439-2448. =

92. Rockstroh JK, DeJesus E, Lennox JL, et al. Durable efficacy and safety of raltegravir versus efavirenz when combined with tenofovir/emtricitabine in treatment-naive HIV-1- infected patients: final 5-year results from STARTMRK. J Acquir Immune Defic Syndr. 2013;63(1):77-85. =

93. Jacobson K, Ogbuagu O. Integrase inhibitor-based regimens result in more rapid virologic suppression rates among treatment-naïve human immunodeficiency virus-infected patients compared to non-nucleoside and protease inhibitor-based regimens in a real-world clinical setting: A r. Med (United States). 2018;97(43). ==

94. Hernán MA, Robins JM. Using Big Data to Emulate a Target Trial When a Randomized Trial Is Not Available. Am J Epidemiol. 2016;183(8):758-764. =

95. Bourgi K, Rebeiro PF, Turner M, et al. Greater Weight Gain in Treatment-naive Persons

Starting Dolutegravir-based Antiretroviral Therapy. Clin Infect Dis. 2019.

96. Rizzardo S, Lanzafame M, Lattuada E, et al. Dolutegravir monotherapy and body weight gain in antiretroviral naïve patients. Aids. 2019;33(10):1673-1674.

97. Althoff KN, Chandran A, Zhang J, et al. Life Expectancy Disparities Among Adults With HIV in the United States and Canada: the Impact of a Reduction in Drug- and Alcohol- Related Deaths Using the Lives Saved Simulation Model. Am J Epidemiol. 2019;188(6):2097-2109.

98. Saag MS, Benson CA, Gandhi RT, et al. Antiretroviral drugs for treatment and prevention of HIV infection in adults: 2018 recommendations of the international antiviral society- USA panel. JAMA - J Am Med Assoc. 2018;320(4):379-396.

99. Kaplan EL, Meier P. Nonparametric Estimation from Incomplete Observations. J Am Stat Assoc. 1958;53(282):457-481.

100. Cole SR, Hernán MA. Adjusted survival curves with inverse probability weights. Comput Methods Programs Biomed. 2004;75(1):45-49.

101. Bhagwat P, Ofotokun I, McComsey GA, et al. Changes in waist circumference in HIV- infected individuals initiating a raltegravir or protease inhibitor regimen: Effects of sex and race. Open Forum Infect Dis. 2018;5(11):1-9.

102. Edwards JK, Cole SR, Hall HI, et al. Virologic suppression and CD4 + cell count recovery after initiation of raltegravir or efavirenz-containing HIV treatment regimens. Aids. 2018;32(2):261-266.

103. Perkins NJ, Cole SR, Harel O, et al. Principled Approaches to Missing Data in Epidemiologic Studies. Am J Epidemiol. 2018;187(3):568-575.

104. Cole SR, Frangakis CE. The consistency statement in causal inference: A definition or an assumption? Epidemiology. 2009;20(1):3-5.

105. Vander Weele TJ. Concerning the consistency assumption in causal inference. Epidemiology. 2009;20(6):880-883. 8

106. Pearl J. On the Consistency Rule in Causal Inference. Epidemiology. 2010;21(6):872-875.

107. Westreich D, Cole SR. Invited commentary: Positivity in practice. Am J Epidemiol. 2010;171(6):674-677.

108. Hudgens MG, Halloran ME. Toward causal inference with interference. J Am Stat Assoc. 2008;103(482):832-842.