Study Designs in HIV Research
Colette Smith (based on slides from Fiona Lampe)
UK CAB meeting Thursday 17th August 2017 Main types of research studies
Cross-sectional Ecological Observational Cohort
Case-control Systematic review of observational studies
Experimental Randomised controlled trial (RCT) (Intervention) Non randomised intervention study
Systematic review of experimental studies 2 Research ques on and study design
• A research ques on always relates to a specific popula on • Different study designs are suited to different types of research ques on • O en several study designs can be used to provide evidence to answer a specific ques on
Does male circumcision influence the risk of acquisi on of HIV infec on? 3 Exposures and Outcomes
EXPOSURE EFFECT OUTCOME (also referred to as “risk/ (typically disease) protec ve factor”; or it could be an interven on)
Does male circumcision influence the risk of acquisi on of HIV infec on, among men in countries with high levels of HIV?
What is the exposure, outcome, and popula on of interest, in this research ques on? 4 Session Overview
• Ecological study • Cross sec onal study • Cohort study • Case-control study • Randomized controlled trial • Systema c review of RCTs
5 Ecological study
• Values available are for a popula on, country or group, not an individual • All factors measured on groups rather than individuals (unlike all other major study designs) • Data cannot be broken down to the level of the individual • Also some mes called correla on or geographical studies • Can include comparisons over me ( me series)
6 Ecological study: Male circumcision and HIV in 37 African countries (Bongaart AIDS 1989)
25 Each point on the graph represents a 20 different African country
15
10
5
0
Prevalence (%) of HIV in country Prevalencein HIV (%)of 0 20 40 60 80 100 Prevalence of circumcision (%) among men in each country 7 Features of ecological studies
• Rela vely quick and inexpensive, as generally use exis ng data • O en used for ini al genera on of ideas, before other types of studies are undertaken • Difficult to establish “cause and effect” • Lack of individual level data, so large poten al for confounding • Major problem in interpre ng results of ecological studies is the Ecological fallacy: extrapola ng results from groups to individuals is problema c 8 Session Overview
• Ecological study • Cross sec onal study • Cohort study • Case-control study • Randomized controlled trial • Systema c review of RCTs
9 Cross-sec onal study
• A group of people are studied at a single point in me; it is a “snap shot” of the current state of affairs • Always observa onal • All factors are measured at this me point: there is no follow-up • Used primarily to assess prevalence of disease and compare between different groups • May involve ques onnaire and/or clinical measurements and tests 10 Cross-sec onal study: Factors associated with HIV infec on in Rakai, Uganda (Serwadda AIDS 1992)
• Sample of 1292 adults from 21 communi es in Rakai conducted in 1989 • Interview (to assess factors related to HIV transmission) and blood test for HIV
Number of Number who were Prevalence participants HIV-positive of HIV All participants 1292 255 19.7% Men 594 88 14.8% Women 698 167 23.9% Circumcised men 80 9 11.3% Uncircumcised men 495 79 16.0% 11 Features of cross-sec onal studies
• Can measure disease prevalence (current number of people with the condi on of interest), but not incidence (number of new events that occur over follow-up) • Rela vely quick and inexpensive to conduct • Can be used to see whether condi ons are more frequent in one group than in another, but may be difficult to establish cause and effect (exposure and outcome measured at same me) • As with all observa onal studies, rela onship between exposure and outcome may be distorted by confounding factors
12 Session Overview
• Ecological study • Cross sec onal study • Cohort study • Case-control study • Randomized controlled trial • Systema c review of RCTs
13 Cohort study
Group of people without disease
Study start
Follow-up all participants over time
Time 14 Cohort study Group exposed to the factor
Group of people without disease
Group NOT exposed to the factor
Study start
Follow-up all participants over time
Time Determine exposure status 15 Cohort study Develop disease Group exposed to the factor Do not develop disease Group of people without disease Develop disease Group NOT exposed to the factor Do not develop disease Study start
Follow-up all participants over time
Time Determine Compare occurrence of exposure status disease between exposed 16 and unexposed groups Cohort study Cohort study: Male circumcision and HIV in Rakai district (Gray AIDS 2004) n=18 acquired HIV N=908 were circumcised n=890 did not acquire HIV N=5,516 men without HIV n=154 acquired HIV N=4,608 were uncircumcised Study start n=4,454 did not acquire HIV Determine exposure status Time 2 years of follow-up 17 Cohort study Cohort study: Male circumcision and HIV in Rakai district (Gray AIDS 2004)
Circumcised Uncircumcised Number of men at start of study 908 4608 Number who developed HIV 18 154 during follow-up
(Incidence) Risk of HIV infec on 18 / 908 = 2.0% 154 / 4608 =3.3% over the 2 year period
18 Features of cohort studies
• Par cipants classified on the basis of their exposure (risk factor) status, and followed-up (“longitudinal”) to determine who develops the outcome • Can calculate and compare incidence of outcome in the exposed and unexposed groups • Can make some a empts at determining cause-and-effect/ temporality: direc on of associa on between exposure and disease is known (exposure precedes disease onset) • Large numbers of par cipants needed, especially if disease is rela vely rare; me-consuming and expensive • As with all observa onal studies, associa ons may be distorted by confounding factors 19 Session Overview
• Ecological study • Cohort study • Case-control study • Systema c review of observa onal studies • Randomized controlled trial • Cross sec onal study
20 Case-control study
Group of people with disease: CASES
Group of people WITHOUT disease: CONTROLS Study start
Look back to determine exposure status IN THE PAST
Time Select cases and controls on the basis of disease 21 status Case-control study Group exposed to the factor Group of people with disease: Group NOT exposed to CASES the factor
Group exposed to the factor Group of people WITHOUT disease: Group NOT exposed to CONTROLS the factor Study start
Look back to determine exposure status IN THE PAST
Compare frequency of Time Select cases and controls exposure between on the basis of disease 22 cases and controls status Case-control study: Male circumcision and HIV in Abidjan, Ivory Coast (Sassan-Morokro JAIDS 1996)
415 / 490 Circumcised (n=415) CASES = 85% (HIV-posi ve men): NOT circumcised n=490 (n=75)
221 / 239 Circumcised (n=221) = 93% CONTROLS (HIV-nega ve men): NOT circumcised n=239 (n=18) Study start: Compare frequency of Look back to determine Select cases and controls exposure between on the basis of disease exposure 23 cases and controls status Features of case-control studies
• Par cipants selected on the basis of their outcome (“cases” if they have the condi on; “control” if they have not) and compared with respect to previous exposure. • Well suited to studying rare diseases (select cases first) • Usually less expensive and quicker than cohort study (fewer par cipants needed and no follow-up involved) • Poten al for recall error / bias in determining exposures • As with all observa onal studies, there may be confounding 24 Confounding in observa onal studies
• Confounding is a major problem in observa onal studies • A confounder is a factor that is associated with the exposure and with the outcome • Confounders can distort the rela onship between exposure and outcome
Exposure Outcome
Confounder
25 Confounding in observa onal studies
• Confounding is a major problem in observa onal studies • A confounder is a factor that is associated with the exposure and with the outcome • Confounders can distort the rela onship between exposure and outcome
Absence of male HIV infection circumcision
26 Ecological study: Male circumcision and HIV in 37 African countries (Bongaart AIDS 1989)
25 Rwanda 20 Uganda Zambia Eastern Africa 15 Burundi
Northern Africa 10 Malawi (Egypt , Morocco, Guinea Bissau Ivory Coast Tunisia, Sudan (N)) 5 Ghana Tanzania 0
Prevalence (%) of HIV in country Prevalencein HIV (%)of 0 20 40 60 80 100 Prevalence of circumcision (%) among men in each country 27 Confounding in observa onal studies
• Confounding is a major problem in observa onal studies • A confounder is a factor that is associated with the exposure and with the outcome • Confounders can distort the rela onship between exposure and outcome
Absence of male HIV infection circumcision
Religion
28 Confounding in observa onal studies
• Confounders in associa on between circumcision and HIV include religion, ethnicity, age • In addi on to simple ‘unadjusted’ analyses, adjusted analysis can be performed to control for confounding factors, using a sta s cal model • Associa on between lack of circumcision and HIV in observa onal studies was generally found to be at least as strong in adjusted analyses • But difficult to measure and adjust for all possible confounding factors
29 Systema c review of observa onal studies: Male circumcision and risk of HIV among men in sub- Saharan Africa (Weiss AIDS 2000)
‖ “Randomised controlled trials of male circumcision are needed (in areas where it would be acceptable to the local community). ‖ Such trials would overcome the limita ons of observa onal studies, and provide reliable evidence on the overall impact of introduc on of male circumcision on HIV incidence”.
30 Session Overview
• Ecological study • Cross sec onal study • Cohort study • Case-control study • Randomized controlled trial • Systema c review of RCTs
31 Randomized controlled trial
Group of people without disease
Study start
Follow-up all par cipants over me
Time 32 Randomized controlled trial
INTERVENTION group Group of people without disease CONTROL group (does NOT receive interven on)
Study start
Follow-up all par cipants over me
Time RANDOMLY ALLOCATE each par cipant to interven on group or control group 33 Randomized controlled trial
Develop disease INTERVENTION group Do not develop disease Group of people without disease Develop disease CONTROL group (does NOT receive interven on) Do not develop disease Study start
Follow-up all par cipants over me
RANDOMLY ALLOCATE each Compare occurrence of Time disease between par cipant to interven on group 34 or control group exposed and unexposed groups RCT of male circumcision for HIV preven on, in Rakai district, Uganda (Gray et al, Lancet 2007)
N=22 N=2,387 acquired HIV immediate circumcision N=2,365 did not N=4,817 acquire HIV uncircumcised men without HIV N=45 N=2,430 acquired HIV delayed circumcision Study start N=2,385 did not RANDOMLY ALLOCATE acquire HIV each par cipant to interven on group or control group Time Follow-up all par cipants over 2 years 35 Cohort study RCT of male circumcision for HIV preven on, in Rakai district, Uganda (Gray et al, Lancet 2007)
INTERVENTION: CONTROL: circumcision No circumcision Number of men at start of study 2387 2430 Number who developed HIV 22 45 during follow-up
(Incidence) Risk of HIV infec on 22 / 2387 = 0.92% 45 / 2430 = 1.9%
36 Features of randomized controlled trials
• Par cipants randomly allocated to either the interven on group or the control group: should ensure two groups have similar characteris cs at start of trial (i.e. no confounding) • Presence of a control group (usually receive the current “gold standard treatment) to act as a comparison • Individuals are followed up to determine who develops disease (i.e. longitudinal study), so RCTs usually inves gate incidence • Aim is to treat both groups iden cally, so that any difference in outcome between interven on and control groups must be due to the interven on, rather than some other factor • Feasibility and ethical issues may prevent RCT being done 37 Session Overview
• Ecological study • Cross sec onal study • Cohort study • Case-control study • Randomized controlled trial • Systema c review of RCTs
38 Systema c review of RCTs: Male circumcision and risk of heterosexual HIV acquisi on among men (Siegfried Cochrane Database of Systema c Reviews 2009) • Studies published up to June 2007 • 3 large RCTs were found including around 11,000 par cipants in total • They concluded “There is strong evidence that medical male circumcision reduces the acquisi on of HIV by heterosexual men by between 38% and 66% over 24 months.”
39 Type of study and strength of evidence Type of study and strength of evidence Strength of evidence for causal associa on between an exposure and a disease outcome
Systema c review of RCTs Strongest evidence
RCT Systema c review of observa onal studies Non randomized interven on study Cohort study
Case-control study
Cross sec onal survey
Ecological study Weakest evidence 40 Accumula on of evidence Studies inves ga ng the rela onship between male circumcision and the risk of HIV acquisi on
Ecological study in 37 African Case-control study in Cohort study of in Systema c countries Abidjan, Ivory Coast Rakai district review of RCTs (Bongaart et al) (Sassan-Morokro et al) (Gray et al) (Siegfried et al) Cross-sec onal study of factors Systema c review of associated with HIV observa onal studies infec on, in Rakai in sub-Saharan Africa RCT in Rakai (Serwadda et al) (Weiss et al) (Gray et al)
Year of publica on 41 Accumula on of evidence
1994: “There may be an associa on between male circumcision and risk for HIV infec on, but considerable scep cism remains” (Moses et al, AIDS)
2000: “There is considerable evidence suppor ng a protec ve effect of male circumcision on HIV-infec on in men” (van Dam et al, Horizons Project)
2007: “Circumcision reduces risk of HIV infec on and can be recommended for HIV preven on in men” (Gray et al, Lancet)
42 On-going issues and ques ons
43 Which study design?
• For individuals currently with a CD4 count>500 cells/mm3, would it be be er to start HIV-treatment immediately, or wait un l the CD4 count reaches 350 cells/mm3? • Do previously an retroviral-naïve individuals star ng an an retroviral regimen containing a new protease inhibitor have a be er response than those receiving a standard of care regimen? • Does alcohol consump on have an impact on future development of depression in HIV-posi ve individuals? • Do individuals co-infected with TB and HIV have a similar first- line response to an retroviral therapy compared to individuals who are HIV-posi ve but do not have TB? 44