706 Care Volume 40, May 2017

Sulfonylureas and the Risks of Laurent Azoulay1,2,3 and Samy Suissa1,2 Cardiovascular Events and Death: A Methodological Meta-Regression Analysis of the Observational Studies Diabetes Care 2017;40:706–714 | DOI: 10.2337/dc16-1943

Recent randomized trials have compared the newer antidiabetic agents to treatments involving sulfonylureas, drugs associated with increased cardiovascular risks and mortality in some observational studies with conflicting results. We reviewed the methodology of these observational studies by searching MEDLINE from inception to December 2015 for all studies of the association between sulfonylureas and cardiovascular events or mortality. Each study was appraised with respect to the comparator, the outcome, and study design–related sources of bias. A meta-regression analysis was used to evaluate heterogeneity. A total of 19 stud- ies were identified, of which six had no major design-related biases. Sulfonylureas

REVIEW were associated with an increased risk of cardiovascular events and mortality in five of these studies (relative risks 1.16–1.55). Overall, the 19 studies resulted in 36 relative risks as some studies assessed multiple outcomes or comparators. Of the 36 analyses, was the comparator in 27 (75%) and death was the outcome in 24 (67%). The relative risk was higher by 13% when the comparator was metformin, by 20% when death was the outcome, and by 7% when the studies had design-related biases. The lowest predicted relative risk was for studies with no major bias, comparator other than metformin, and cardiovascular outcome (1.06 [95% CI 0.92–1.23]), whereas the highest was for studies with bias, metformin comparator, and mortality outcome (1.53 [95% CI 1.43–1.65]). In summary, sulfo- 1Center for Clinical Epidemiology, Lady Davis nylureas were associated with an increased risk of cardiovascular events and mor- Institute, Jewish General Hospital, Montreal, Canada tality in the majority of studies with no major design-related biases. Among studies 2 fi Department of Epidemiology, Biostatistics, with important biases, the association varied signi cantly with respect to the and Occupational Health, McGill University, comparator, the outcome, and the type of bias. With the introduction of new Montreal, Canada antidiabetic drugs, the use of appropriate design and analytical tools will provide 3Gerald Bronfman Department of Oncology, their more accurate cardiovascular safety assessment in the real-world setting. McGill University, Montreal, Canada Corresponding author: Laurent Azoulay, laurent [email protected]. It is well established that is associated with an increased risk of Received 8 September 2016 and accepted 7 cardiovascular morbidity and mortality (1). Although much of this association can be January 2017. attributed to the long-term complications of this disease, there has been growing © 2017 by the American Diabetes Association. fl Readers may use this article as long as the work interest in determining whether certain antidiabetic drugs in uence this risk. In par- is properly cited, the use is educational and not ticular, over the years, there have been concerns regarding the cardiovascular safety of for profit, and the work is not altered. More infor- sulfonylureas, the second most commonly used antidiabetic drugs after metformin. mation is available at http://www.diabetesjournals These safety concerns initiated with the University Group Diabetes Program (UGDP) .org/content/license. conducted in the 1960s, where (a first-generation sulfonylurea) was See accompanying article, p. 629. care.diabetesjournals.org Azoulay and Suissa 707

associated with an increased risk of all- search was limited to studies published risks as a function of the three study de- cause and cardiovascular mortality in English before 31 December 2015. sign factors, namely the presence of ma- compared with placebo (2). Indeed, sulfo- jor bias, the comparator being metformin nylureas have been associated with Inclusion Criteria or other, and the outcome being mortal- weight gain, fluid retention, and hypogly- To be included in this methodological re- ity versus cardiovascular events. The cemia, which are all known cardiovascular view, the observational studies had to have model was weighted by the inverse of risk factors (3). In contrast, meta-analyses 1) compared at least one sulfonylurea the variance of the logarithm of each rel- of sulfonylurea randomized controlled (alone or in combination with other antidi- ative risk. This analysis estimated the ratio fi trials (RCTs) have produced conflicting abetic drugs) with speci c antidiabetic of relative risks (RRR) associated indepen- findings with respect to cardiovascular drugs or no use of sulfonylureas and 2) dently with each of the three factors, events and mortality (4–6). However, reported on at least one of the outcomes along with the 95% CIs. From this model, none of these RCTs were designed or of interest (cardiovascular events, cere- we also estimated the lowest and highest powered to detect cardiovascular events brovascular events, and cardiovascular- predicted relative risks (PRRs) and their fi and the RCTs used different comparators, speci c and all-cause mortality). Studies 95% CIs on the basis of the three study including other oral agents or placebo. comparing different sulfonylureas to each design factors. In contrast to the RCTs, several ob- other or to patients without diabetes were servational studies have associated not included, as were those conducted RESULTS within selected cardiovascular populations, sulfonylureas with an increased risk of As of 31 December 2015, we identified a such as among those previously hospital- cardiovascular events and death (7). How- total of 44 observational studies that ized for myocardial infarction. The latter ever, these observational studies used assessed the cardiovascular safety of represent distinct populations that as- varying approaches to study design and sulfonylureas (8–51). Of those, 15 studies sessed the effects of sulfonylureas on the data analysis that could have introduced assessed outcomes within selected popu- risk of recurrent events. Such studies have several biases. With the introduction of lations with cardiovascular complications their own set of methodological chal- new antidiabetic drugs, such as dipeptidyl (8,15,16,23,24,27,30–33,35,40,41,44,49), lenges, which are outside the scope of peptidase-4 (DPP-4) inhibitors, glucagon- 6 studies performed within-class this review. like peptide 1 (GLP-1) analogs, and comparisons (9,19,28,34,39,47), and sodium–glucose cotransporter 2 (SGLT2) Sources of Bias 4 studies used patients without type 2 inhibitors, drugs that will likely undergo One of the major threats to observa- diabetes as comparators (10,18,50,51). the same scrutiny as sulfonylureas, there tional studies is confounding bias, which The remaining 19 studies were con- is a need to understand how bias can be presents important challenges in their ducted within unselected populations introduced (and avoided) in observational conduct. Indeed, any study evaluating (11–14,17,20–22,25,26,29,36–38,42,43, studies assessing the safety of second- to the risk of sulfonylureas will have to 45,46,48). These 19 studies were as- third-line treatments. Moreover, many ensure that the comparator group of pa- sessed independently by the two au- RCTs of the newer drugs have included tients is practically identical to the pa- thors for their study design quality. and will include sulfonylureas in the tients using sulfonylureas. Thus, in The aforementioned studies were clas- comparator group. contrast with RCTs where such compara- sified into one of four categories: expo- The objective of this methodological bility is inherently established by ran- sure misclassification (17,36,42), time-lag review is to detail the most important domization, observational studies must bias (12,13,20,25,26,37,42,45,48), selec- methodological limitations of observa- rely on matching or statistical adjust- tion bias (14,22), and studies with no major tional studies assessing the cardiovascu- ment techniques to minimize the poten- biases (11,21,29,38,43,46). Overall, the lar safety of sulfonylureas and identify the tial for confounding bias and determine relative risks ranged between 1.37 and most robust observational studies of the comparability. Although most observa- 1.70 in studies with exposure misclassi- association between sulfonylureas and tional studies accomplish this quite well fication, between 0.95 and 2.08 in stud- the risk of cardiovascular events. This re- within the limitations of the available ies with time-lag bias, between 1.23 and view will thus provide guidance for future data, there are other sources of bias 1.24 in studies with selection bias, and studies on the use of robust methods to that can arise in their study design, between 0.95 and 1.55 in studies with minimize bias in assessing the cardiovas- which can have a greater impact on their no major biases. These studies are listed cular safety of newer antidiabetic drugs. results. In this article, we focus mainly on in Tables 1 and 2, and their biases are some of these study design limitations described in detail below along with RESEARCH DESIGN AND METHODS that could lead to bias other than from some illustrative examples. Search Strategy confounding, namely exposure misclas- A MEDLINE search was conducted to iden- sification, time-lag bias, and selection Exposure Misclassification fi tify all observational studies assessing the bias. In addition, we discuss the role of Exposure misclassi cation is invariably a effects of sulfonylureas on the incidence the comparator used in these studies. limitation of all observational studies. of cardiovascular and cerebrovascular However, in many studies investigating events, cardiovascular mortality, and all- Data Analysis the association between the use of sul- cause mortality. The search terms in- A meta-regression analysis was used to fonylureas and cardiovascular out- cluded “sulfonylureas,”“cardiovascular,” evaluate the heterogeneity among the comes, this exposure misclassification “myocardial infarction,”“coronary artery relative risks. A multivariate log-linear re- was magnified by using intent-to-treat disease,”“stroke,” and “mortality.” The gression model was used to fit the relative (ITT) analyses (17,36,42). Just as with 708 Sulfonylureas and Cardiovascular Risk Diabetes Care Volume 40, May 2017

Table 1—Methodological limitations of observational studies investigating the effects of sulfonylureas alone or in combination with other antidiabetic drugs on the incidence of cardiovascular events or mortality Author Study design Comparison Outcome(s) Relative riska (95% CI) Exposure misclassification Evans et al. (17) Cohort Sulfonylureas vs. metformin All-cause mortality 1.43 (1.15–1.77) Sulfonylureas vs. metformin Cardiovascular mortality 1.70 (1.18–2.45) Corrao et al. (36) Cohort Sulfonylureas vs. metformin All-cause mortality 1.37 (1.26–1.49) Pantalone et al. (42) Cohort vs. metformin All-cause mortality 1.64 (1.39–1.94) Glyburide vs. metformin All-cause mortality 1.59 (1.35–1.88) vs. metformin All-cause mortality 1.68 (1.37–2.06) Time-lag bias b Mannucci et al. (12) Cohort Sulfonylurea + metformin vs. All-cause mortality 2.08 (1.18–3.67) c other antidiabetic drugs 1.68 (1.01–2.79) Koro et al. (13) Nested case-control Sulfonylureas vs. no treatment CHF 1.19 (1.02–1.39) Kahler et al. (20) Cohort Sulfonylureas vs. metformin All-cause mortality 1.15 (0.91–1.47)d Sulfonylureas vs. TZD All-cause mortality 1.04 (0.75–1.46)d Tzoulaki et al. (25) Cohort Sulfonylureas vs. metformin All-cause mortality 1.37 (1.11–1.71)e Sulfonylureas vs. metformin All-cause mortality 1.24 (1.14–1.35)f Pantalone et al. (26) Cohort Sulfonylureas vs. metformin CHF 1.32 (1.10–1.56)d Sulfonylureas vs. metformin All-cause mortality 1.85 (1.56–2.17)d Sulfonylureas vs. All-cause mortality 1.69 (1.23–2.33)d Horsdal et al. (37) Case-control Sulfonylureas vs. metformin Myocardial infarction 1.16 (1.05–1.28)d Sulfonylureas vs. Myocardial infarction 1.09 (1.01–1.16)d Pantalone et al. (42) Cohort Glipizide vs. metformin All-cause mortality 1.64 (1.39–1.94) Glyburide vs. metformin All-cause mortality 1.59 (1.35–1.88) Glimepiride vs. metformin All-cause mortality 1.68 (1.37–2.06) Currie et al. (45) Cohort Sulfonylureas vs. metformin All-cause mortality 1.75 (1.64–1.86) Ghotbi et al. (48) Cohort Sulfonylureas vs. insulin MACEg 0.95 (0.73–1.22) Sulfonylureas vs. insulin All-cause mortality 1.20 (0.90–1.60) Selection bias Johnson et al. (14) Cohort Sulfonylureas vs. metformin Nonfatal hospitalization, 1.23 (1.03–1.47)d all-cause mortality McAlister et al. (22) Cohort Sulfonylureas vs. metformin CHF 1.24 (1.01–1.54) MACE, major adverse cardiovascular event; TZD, . aRelative risk is used as a generic term for rate ratio, HR, and odds ratio. bEstimated among women. cEstimated among men. dFor consistency with the rest of the table, the relative risks and 95% CIs were inversed when sulfonylureas were the comparator group. eAnalysis based on first-generation sulfonylureas. fAnalysis based on second-generation sulfonylureas. gIncluded nonfatal myocardial infarction, nonfatal stroke, resuscitation after cardiac arrest, or cardiovascular death.

RCTs, the ITT approach assumes patients conducted study, the authors identified misclassification of exposure that results remain on the initial treatment through- a cohort of 70,437 patients who initiated in biasing the risk estimates toward out the follow-up period, regardless of metformin or a sulfonylurea in monother- a null effect. However, there are cer- treatment discontinuation and, in some apy between 2001 and 2003. Patients tain situations, particularly in observa- cases, also regardless of treatment in- were followed from the date of the first tional studies, where the opposite can tensification or switching. Unlike in prescription until the first hospitalization occur. In the present example, the au- RCTs, however, this approach will not for macrovascular disease, death, emigra- thors provided a useful analysis showing ensure that potential confounders tion, or end of the study period (31 July that sulfonylurea users were more likely are well balanced between the exposure 2007). Thus, patients were considered ex- to have switched to insulin during the groups, thus necessitating advanced posed to their initial treatment, either follow-up period than metformin users study design and statistical methods to metformin or sulfonylureas, for up to (HR 1.55 [95% CI 1.43–1.68]). This may control for confounding. The appeal of 7 years after their first prescription, re- be related to the fact that sulfonylureas using this approach within an observa- gardless of treatment termination or have a lower efficacy in lowering HbA1c tional setting is its simplicity of implemen- switch to another antidiabetic drug during than metformin (52). It may also be re- tation, while minimizing potential biases this period. Overall, compared with met- lated to the fact that sulfonylurea users related to censoring patients at the time formin, the use of sulfonylureas was had advanced disease when they initi- of treatment discontinuation, particularly associated with an increased risk of hos- ated treatment and were thus more likely if the latter is related to the outcome of pitalization (hazard ratio [HR] 1.15 [95% to progress to using the last-line treat- interest. CI 1.08–1.21]) and death from any ment (i.e., insulin) than metformin users. An example of this approach is one cause (HR 1.37 [95% CI 1.26–1.49]). As diabetes severity is likely associated study using databases from the Italian re- The ITT approach generally, and partic- with an increased risk of cardiovascular gion of Lombardy (36). In this well- ularly in RCTs, leads to a nondifferential events, the ITT approach in this study care.diabetesjournals.org Azoulay and Suissa 709

Table 2—Observational studies with no major biases assessing the cardiovascular safety of sulfonylureas Author Study design Comparison Outcome(s) Relative riska (95% CI) Cardiovascular events McAfee et al. (21) Cohort Sulfonylureas vs. metformin MI and coronary 1.30 (1.04–1.61) revascularization Schramm et al. (38) Cohort Glimepiride vs. metformin MACE 1.32 (1.24–1.40) Glyburide vs. metformin MACE 1.19 (1.11–1.28) Glipizide vs. metformin MACE 1.27 (1.17–1.38) Tolbutamide vs. metformin MACE 1.28 (1.17–1.39) Roumie et al. (43) Cohort Sulfonylureas vs. metformin MACE 1.16 (1.08–1.25) All-cause mortality Gulliford et al. (11) Cohort Sulfonylureas + metformin vs. metformin All-cause mortality 0.95 (0.64–1.40) Azoulay et al. (29) Nested case-control Sulfonylureas vs. metformin All-cause mortality 1.43 (1.33–1.56) Wheeler et al. (46) Cohort Glyburide vs. metformin All-cause mortality 1.38 (1.27–1.50) Glipizide vs. metformin All-cause mortality 1.55 (1.43–1.67) MACE, major adverse cardiovascular event; MI, myocardial infarction. aRelative risk is used as a generic term for rate ratio, HR, and odds ratio.

likely led to an overestimation of the as- type 2 diabetes, of which 1,301 were di- contribute two time intervals, one corre- sociation. This potential bias could have agnosed with CHF during follow-up (case sponding to metformin and another corre- been minimized by limiting follow-up to a subjects). These case subjects were then sponding to sulfonylurea. As a result, the shorter time period (e.g., 1 year) and/or matched to 7,788 control subjects who 91,521 patients included in the study gen- using an as-treated exposure definition were never diagnosed with CHF during erated 2,843,007 intervals of treatment that would have limited the follow-up to follow-up on age, sex, and calendar year with antidiabetic drugs. Using metformin the time while on treatment, or using a of the CHF diagnosis. Overall, compared monotherapy as the reference, the authors corresponding time-dependent analysis. with sulfonylurea monotherapy, the com- reported that monotherapy with sulfonyl- bination of metformin with sulfonylureas ureas (either first or second generation) Time-Lag Bias wasassociatedwithincreasedrisksofall Time-lag bias arises when comparing was associated with an increased risk of – three outcomes. The methodology used in treatments at different stages of the dis- CHF (odds ratio 1.38 [95% CI 1.13 1.69]). this study may hold well in settings where ease process (53). As duration of diabe- It is important to note that calendar time the time intervals are assumed to be in- tes has been previously associated with was the underlying time axis used to match terchangeable, i.e., where the risk of the cardiovascular outcomes (54), compar- case and control subjects, and not duration outcome of interest is the same regardless ing drugs prescribed at different stages of disease. As such, this may result in a of follow-up or disease duration. In the of the disease can introduce important differential duration of disease between context of the progressive nature of bias (53). For example, in cohort studies, case and control subjects. In this particular type 2 diabetes, however, this assumption time-lag bias can be introduced when example, it is possible that case subjects is unlikely to hold. As a result, comparing comparing a second- to third-line treat- were more likely to be using a combination treatment intervals that are separated by ment (e.g., thiazolinedediones) to a first- therapy if their duration of disease was on several years may introduce time-lag bias, line treatment (e.g., metformin). Such average longer than in their matched con- as it becomes difficult to separate the ef- comparisons introduce confounding by trol subjects. Conversely, because of the fects of the drug from the effects of the disease severity that may be difficult to random selection of control subjects, it is underlying disease progression. While one control for in the analysis (see Fig. 1 for a also possible for control subjects to have a method for minimizing this potential bias graphical representation of this bias). This longer duration of disease compared with can be to adjust the statistical models for bias can also be present in case-control case subjects. As a result, failure to match duration of diabetes (as was done by the studies if the duration of the disease is dif- on disease duration can produce spurious authors), the latter is at best a moderate ferent between case and control subjects. results that can go in either direction (i.e., proxy for diabetes severity. Alternatively, A number of observational studies in- increased risk as observed in this study or matching the treatment intervals on diabe- vestigating the cardiovascular effects of decreased risk). tes duration or defining exposure as a sulfonylureas were likely affected by In another study using the U.K. GPRD, time-varying variable using duration of di- time-lag bias (12,13,20,25,26,37,42,45,48). the authors conducted a large cohort study abetes as the underlying time axis may be a In one study using the U.K. General Practice among 91,521 patients to assess the asso- more effective means of minimizing this Research Database (GPRD) (now known as ciation between different antidiabetic source of confounding. the Clinical Practice Research Datalink drugs and the risk of myocardial infarction, [CPRD]), the authors conducted a nested CHF, and all-cause mortality (25). The unit Choice of the Comparator case-control analysis to assess the associa- of analysis was the time interval on which a The choice of the comparator group is tion between different antidiabetic drugs patient was using a specific antidiabetic another major consideration in obser- and the risk of congestive heart failure drug. For example, a patient starting treat- vational studies assessing the safety (CHF) (13). The underlying cohort included ment with metformin monotherapy and of second- to third-line antidiabetic 21,888 patients newly diagnosed with then switching to a sulfonylurea would drugs, which can affect the risk estimates 710 Sulfonylureas and Cardiovascular Risk Diabetes Care Volume 40, May 2017

Figure 1—Time-lag bias introduced by comparing patients at different stages of the disease progression. A: Treatment trajectory of a hypothetical patient. B: Time-lag bias is introduced when comparing patients at different stages of the disease, in this case, the use of different antidiabetic drugs to metformin.

above and beyond the biases described patients using antidiabetic drugs in which corresponds by numerical inver- above. For instance, using as compara- monotherapy, combination users, and sion to an increased risk with sulfonyl- tor , which have been insulin users. As such, this comparator ureas (HR 1.23 [95% CI 1.03–1.47]), associated with an increased risk of car- group includes patients with different compared with metformin. Although diovascular events (55), could lead to an disease severities, rendering the find- this study had several strengths, a num- underestimation of the association. In ings difficult to interpret. ber of exclusions may have affected the contrast, another approach has been findings. Specifically, out of the 12,188 to compare the use of sulfonylureas Selection Bias patients prescribed oral antidiabetic with “no use,” a comparator group con- Selection bias is an important threat to drugs during the study period (1991– sisting of patients not currently using the validity of observational studies. Se- 1999), 6,486 (53.2%) patients were ex- any antidiabetic drug (13). As this com- lection bias is a systematic error that cluded on the basis of future events oc- parator group may include patients with occurs when the selection of patients curring during the follow-up period less severe disease not requiring phar- into a study is influenced by their expo- (such as a dispensation of insulin [n = macological treatment, comparing an sure and disease status. Some of the 1,443], ,1 year of oral antidiabetic antidiabetic drug to this comparator observational studies investigating the drug use [n = 2,009], and receiving less mayleadtoanoverestimationofthe association between sulfonylureas and than the recommended average daily association. Indeed, in one study assess- cardiovascular outcomes likely suffered dose during one or more 6-month inter- ing the association between different from selection bias (14,22). For exam- vals [n = 3,034]). These exclusion criteria antidiabetic drugs and the risk of CHF, ple, in one study using the Saskatche- all involve immortal time, which implies sulfonylurea monotherapy was associ- wan Health administrative databases, that the patient necessarily had to be ated with a 19% increased risk (HR the authors conducted a cohort study alive during part of the follow-up period 1.19 [95% CI 1.02–1.39]) when com- among 5,702 patients to assess whether to satisfy one of the criteria. For exam- pared with no use (13). Interestingly, a metformin monotherapy is associated ple, the 1,443 patients excluded be- similar HR was observed with metformin with a lower risk of a composite end cause they received insulin some time monotherapy (HR 1.20 [95% CI 0.97– point of first nonfatal hospitalization during follow-up had to be alive at the 1.48]), a drug thought to have neutral for any cause or death, compared with time that they received the insulin; such effects on cardiovascular risk (13). sulfonylurea monotherapy (14). Overall, exclusions will introduce immortal time Finally, a common approach has been metformin monotherapy, compared bias (56). The magnitude of this bias will to compare a specific antidiabetic drug with sulfonylureas, was associated be particularly important if the time to with “any use” of other antidiabetic with a decreased risk of the composite insulin is longer in one group than the drugs. The latter may include a mix of end point (HR 0.81 [95% CI 0.68–0.97]), other, which we suspect to be the case care.diabetesjournals.org Azoulay and Suissa 711

with metformin (longer time to starting Table 3—Crude and adjusted RRRs of adverse event associated with sulfonylurea insulin) and sulfonylureas (shorter use according to study design parameters from the meta-regression analysis of time). Consequently, selection bias was 36 estimates of relative risk from the observational studies listed in Tables 1 and 2 likely introduced in an effort to create Number of relative Mean Crude Adjusted RRR* mutually exclusive groups of sulfonyl- risk estimates relative risk RRR (95% CI) urea and metformin monotherapy users Comparator during the entire follow-up period. Metformin 27 1.43 1.08 1.13 (1.01–1.27) Other (reference) 9 1.32 1.00 1.00 (reference) Studies With No Major Biases Outcome All observational studies are prone to some Death 24 1.50 1.24 1.20 (1.07–1.34) degree of bias. However, certain design Cardiovascular (reference) 12 1.21 1.00 1.00 (reference) and analytical choices could minimize Major bias these biases. To date, six observational Yes 26 1.44 1.13 1.07 (0.95–1.20) studies did not suffer from the major biases No (reference) 10 1.28 1.00 1.00 (reference) discussed above (11,21,29,38,43,46). For *Adjusted for one another and weighted by the inverse of the variance. example, in the study by Roumie et al. (43), the authors used the Veterans Health Administration databases to assess highest PRR was for studies with major to important selection bias. Instead, whether sulfonylurea monotherapy (n = bias, where the comparator was metfor- such patients should be allowed to be in- 98,665) was associated with an increased min and where the outcome was mortal- cluded in the cohort, and other mecha- risk of a composite of acute myocardial in- ity (highest PRR 1.53 [95% CI 1.43–1.65]). nisms such as censoring could be used to farction, stroke, or death, compared with mitigate this issue (assuming the censor- metformin monotherapy (n = 155,025). In CONCLUSIONS ing is noninformative). Finally, rigorous addition to its large sample size, this study Assessing the cardiovascular safety of sec- adjustment of potential confounders is restricted the cohort to patients newly ond- to third-line antidiabetic drugs can necessary; this can be achieved with the treated with these drugs (defined as no be challenging. As discussed above, cer- use of methods such as propensity scores prescription in the year before), thereby tain design and analytical decisions can or marginal structural models (57). minimizing time-lag bias (53). It also used help circumvent some of these challenges In addition to the methodological is- an as-treated exposure definition that fol- and the biases that ensue. First, given the sues described above, it is important to lowedpatientsuntilaswitchoranaddi- progressive nature of type 2 diabetes, the recognize that there are differences in the tion of another antidiabetic drug, an primary exposure definition should be pharmacodynamic and pharmacokinetic outcome, or one of the study’scensoring based on an as-treated approach or mod- properties of the different sulfonylureas; events, an approach that minimized expo- eling exposure as a time-varying variable. these may have an impact on their asso- sure misclassification. In addition, the An ITT exposure could also be used to ciationwithcardiovascular outcomes study minimized the potential for con- complement the aforementioned expo- (58). Indeed, some sulfonylureas are not founding bias, with the use of propensity sure definitions but will need to be lim- selective for pancreatic b-cells and thus score matching of the patients on sulfo- ited to a relatively short follow-up period may increase the risk of cardiovascular nylurea monotherapy with those on met- (e.g., 1 year) to avoid the exposure mis- outcomes by binding to receptors in other formin monotherapy. Overall, the use of classification issues described above. Sec- tissues, such as cardiomyocytes and vas- sulfonylureas was associated with a mod- ond, it is important to compare drugs cular smooth muscle cells (59). As such, it est increased risk of the composite end used at a similar stage of the disease to will be necessary for future studies to as- point (HR 1.16 [95% CI 1.08–1.25]). avoid time-lag bias (53). For example, sess the effects of sulfonylureas as a class, Meta-Regression Analysis comparing a combination of metformin as well as the effects of each individual The results of the meta-regression log-lin- and sulfonylurea (a second- to third-line sulfonylurea. ear model analysis are summarized in treatment strategy) to metformin mono- Understanding the cardiovascular ef- Table 3. Overall, the relative risk of an therapy (a first-line treatment strategy) fects of sulfonylureas is also important adverse event is higher by 13% when the will introduce important confounding by because many RCTs of the newer antidi- comparator is metformin with an RRR of indication that may be difficult to adjust abetic drugs use sulfonylureas as the 1.13 (95% CI 1.01–1.27). Moreover, the for in the statistical models. The alterna- comparator drug or include them as relative risk is higher by 20% when death tive would be to compare a combination part of the comparator regimen. Thus, is the outcome (RRR 1.20 [95% CI 1.07– therapy with another combination used the safety assessment of these newer 1.34]). Finally, for the studies with major at a similar stage of the disease, or alter- drugs can inherently possibly conceal bias as defined above, the relative risk was natively match the combination users an increased cardiovascular risk if the higher by 7% (RRR 1.07 [95% CI 0.95– with monotherapy users on diabetes du- comparator group includes a treatment 1.20]). As a result of this model, the lowest ration. Third, inclusion of patients in a that increases this risk. Likewise, it is PRR was for studies with no major bias, study should not be based on future possible that the benefits on cardiovas- where the comparator was other than events occurring during the follow-up pe- cular and mortality outcomes observed metformin and where the outcome riod (such as excluding patients eventu- in some of the recent trials may be due, was a nonfatal cardiovascular event allyexposedtoaspecificantidiabetic at least in part, to the fact that a higher (lowest PRR 1.06 [95% CI 0.92–1.23]). The drug); such inclusion criteria could lead proportion of patients in the placebo 712 Sulfonylureas and Cardiovascular Risk Diabetes Care Volume 40, May 2017

arms initiated sulfonylureas during the need to use appropriate methodologi- increase early mortality in patients with diabe- follow-up period (60–62). Thus, our cal approaches to minimize bias when tes mellitus after direct angioplasty for acute present meta-regression analysis of assessing the safety of second- to third- myocardial infarction. J Am Coll Cardiol 1999; 33:119–124 the observational studies to data sug- line antidiabetic drugs. This is highly rel- 9. Olsson J, Lindberg G, Gottsater¨ M, et al. In- gests that some caution be used in in- evant in this era of new antidiabetic creased mortality in type II diabetic patients terpreting the data from such recent drugs (such as DPP-4 inhibitors, GLP-1 an- using sulphonylurea and metformin in combina- trials when sulfonylureas are involved alogs, and SGLT2 inhibitors), where there tion: a population-based observational study. Di- – in the comparator. will be a need to assess their cardiovas- abetologia 2000;43:558 560 10. Klamann A, Sarfert P, Launhardt V, Schulte The cardiovascular safety of some of cular safety in the real-world setting. G, Schmiegel WH, Nauck MA. Myocardial infarc- the newer antidiabetic agents, such as tion in diabetic vs non-diabetic subjects. Sur- DPP-4 inhibitors, has also recently been vival and infarct size following therapy with the subject of some concern. In the Sax- Funding. This research was funded in part by sulfonylureas (). Eur Heart J – agliptin Assessment of Vascular Out- grants from the Canadian Institutes of Health 2000;21:220 229 Research and the Canadian Foundation for In- 11. Gulliford M, Latinovic R. Mortality in type 2 comes Recorded in Patients with Diabetes diabetic subjects prescribed metformin and sul- – novation. L.A. is the recipient of a Chercheur- Mellitus Thrombolysis in Myocardial In- Boursier award from the Fonds de recherche du phonylurea drugs in combination: cohort study. – farction (SAVOR-TIMI) 53 trial, patients Quebec´ - Sante.´ S.S. is the recipient of the James Diabetes Metab Res Rev 2004;20:239 245 randomized to had a higher McGill Professorship award. 12. Mannucci E, Monami M, Masotti G, risk of hospitalization for CHF compared Duality of Interest. This research was funded in Marchionni N. All-cause mortality in diabetic patients treated with combinations of sulfonyl- with placebo (HR 1.27 [95% CI 1.07–1.51]) part by a grant from Boehringer Ingelheim. No other potential conflicts of interest relevant to ureas and . Diabetes Metab Res Rev (63). In contrast, this association was not this article were reported. 2004;20:44–47 observed in the vast majority of the ob- Author Contributions. L.A. drafted the man- 13. Koro CE, Bowlin SJ, Weiss SR. Antidiabetic servational studies conducted to date uscript. L.A. and S.S. conceived and designed the therapy and the risk of heart failure in type 2 (64–72), with the exception of two stud- study, performed the statistical analyses, inter- diabetic patients: an independent effect or con- founding by indication. Pharmacoepidemiol ies (73,74). Although the absence of an preted the data, and critically revised the man- uscript for important intellectual content. Drug Saf 2005;14:697–703 association in most of these studies pro- 14. Johnson JA, Simpson SH, Toth EL, Majumdar vides some reassurance with respect to SR. Reduced cardiovascular morbidity and mortal- References CHF, the assessment of other cardiovas- ity associated with metformin use in subjects with 1. Huxley R, Barzi F, Woodward M. Excess risk – cular outcomes will require careful atten- type 2 diabetes. Diabet Med 2005;22:497 502 of fatal coronary heart disease associated with 15. Danchin N, Charpentier G, Ledru F, et al. tion to design-related decisions. As the diabetes in men and women: meta-analysis of Role of previous treatment with sulfonylureas newer antidiabetic drugs are intended 37 prospective cohort studies. BMJ 2006;332: in diabetic patients with acute myocardial in- to be used as second- to third-line treat- 73–78 farction: results from a nationwide French reg- ments, it will be imperative that they are 2. Meinert CL, Knatterud GL, Prout TE, Klimt CR. istry. Diabetes Metab Res Rev 2005;21:143–149 A study of the effects of hypoglycemic agents on compared with drugs used at a similar 16. Eurich DT, Majumdar SR, McAlister FA, vascular complications in patients with adult- Tsuyuki RT, Johnson JA. Improved clinical out- stage of the disease, and using the appro- onset diabetes. II. Mortality results. Diabetes comes associated with metformin in patients priate exposure definitions. Failure to 1970;19(Suppl.):789–830 with diabetes and heart failure. Diabetes Care consider these important design-related 3. Tahrani AA, Barnett AH, Bailey CJ. Pharma- 2005;28:2345–2351 decisions will likely introduce bias and, in cology and therapeutic implications of current 17. Evans JM, Ogston SA, Emslie-Smith A, drugs for type 2 diabetes mellitus. Nat Rev En- the process, generate more uncertainty Morris AD. Risk of mortality and adverse cardio- docrinol 2016;12:566–592 vascular outcomes in type 2 diabetes: a compar- on the safety of these drugs. 4. Selvin E, Bolen S, Yeh HC, et al. Cardiovascu- ison of patients treated with sulfonylureas and In summary, the majority of the studies lar outcomes in trials of oral diabetes medica- metformin. Diabetologia 2006;49:930–936 reporting on the association between sul- tions: a systematic review. Arch Intern Med 18. Johnsen SP, Monster TB, Olsen ML, et al. – fonylureas and cardiovascular risk had 2008;168:2070 2080 Risk and short-term prognosis of myocardial in- 5. Varvaki Rados D, Catani Pinto L, Reck farction among users of antidiabetic drugs. Am J design-related biases, such as exposure Remonti L, Bauermann Leitão C, Gross JL. – fi Ther 2006;13:134 140 misclassi cation, time-lag bias, and selec- The association between sulfonylurea use and 19. Simpson SH, Majumdar SR, Tsuyuki RT, tion bias. However, the majority of studies all-cause and cardiovascular mortality: a meta- Eurich DT, Johnson JA. Dose-response relation with no major designed-related biases analysis with trial sequential analysis of ran- between sulfonylurea drugs and mortality in reported increased risks of cardiovascular domized clinical trials. PLoS Med 2016;13: type 2 diabetes mellitus: a population-based e1001992 – events and mortality with sulfonylureas. cohort study. CMAJ 2006;174:169 174 6. Bain S, Druyts E, Balijepalli C, et al. Cardio- 20. Kahler KH, Rajan M, Rhoads GG, et al. Im- Overall, the role of the comparator drug vascular events and all-cause mortality associ- pact of oral antihyperglycemic therapy on all- used had an important bearing on the risk ated with sulfonylureas compared with other cause mortality among patients with diabetes estimates. Indeed, the lowest PRR was antihyperglycaemic drugs: a Bayesian meta- in the Veterans Health Administration. Diabetes for studies with none of the major analysis of survival data. Diabetes Obes Metab. Care 2007;30:1689–1693 14 November 2016 [Epub ahead of print]. 21. McAfee AT, Koro C, Landon J, Ziyadeh N, biases described above, where the com- DOI: 10.1111/dom.12821 Walker AM. Coronary heart disease outcomes parator was not metformin, and where 7. Pladevall M, Riera-Guardia N, Margulis AV, in patients receiving antidiabetic agents. Phar- the event definition was based on a car- Varas-Lorenzo C, Calingaert B, Perez-Gutthann macoepidemiol Drug Saf 2007;16:711–725 diovascular outcome. In contrast, the S. Cardiovascular risk associated with the use of 22. McAlister FA, Eurich DT, Majumdar SR, highest PRR was for studies with major glitazones, metformin and sufonylureas: meta- Johnson JA. The risk of heart failure in patients analysis of published observational studies. with type 2 diabetes treated with oral agent biases, where metformin was the compar- BMC Cardiovasc Disord 2016;16:14 monotherapy. Eur J Heart Fail 2008;10:703–708 ator, and where the outcome was mor- 8. Garratt KN, Brady PA, Hassinger NL, Grill DE, 23. Horsdal HT, Johnsen SP, Søndergaard F, tality. This heterogeneity highlights the Terzic A, Holmes DR Jr. Sulfonylurea drugs Rungby J. Type of preadmission glucose- care.diabetesjournals.org Azoulay and Suissa 713

lowering treatment and prognosis among pa- with the use of metformin and sulphonylureas in monotherapy and matched, non-diabetic con- tients hospitalised with myocardial infarction: type 2 diabetes: a population-based cohort study trols. Diabetes Obes Metab 2014;16:1165–1173 a nationwide follow-up study. Diabetologia in Italy. Eur J Clin Pharmacol 2011;67:289–299 51. Claesen M, Gillard P, De Smet F, Callens M, 2008;51:567–574 37. Horsdal HT, Søndergaard F, Johnsen SP, De Moor B, Mathieu C. Mortality in individuals 24. Sadikot SM, Mogensen CE. Risk of coronary Rungby J. Antidiabetic treatments and risk of treated with glucose-lowering agents: a large, artery disease associated with initial sulphony- hospitalisation with myocardial infarction: a na- controlled cohort study. J Clin Endocrinol Metab lurea treatment of patients with type 2 diabe- tionwide case-control study. Pharmacoepide- 2016;101:461–469 tes: a matched case-control study. Diabetes Res miol Drug Saf 2011;20:331–337 52. Genuth S. Should sulfonylureas remain an Clin Pract 2008;82:391–395 38. Schramm TK, Gislason GH, Vaag A, et al. acceptable first-line add-on to metformin therapy 25. Tzoulaki I, Molokhia M, Curcin V, et al. Risk Mortality and cardiovascular risk associated in patients with type 2 diabetes? No, it’stimeto of cardiovascular disease and all cause mortality with different insulin secretagogues compared move on! Diabetes Care 2015;38:170–175 among patients with type 2 diabetes prescribed withmetforminintype2diabetes,withor 53. Suissa S, Azoulay L. Metformin and the risk oral antidiabetes drugs: retrospective cohort without a previous myocardial infarction: a na- of cancer: time-related biases in observational study using UK General Practice Research Data- tionwide study. Eur Heart J 2011;32:1900–1908 studies. Diabetes Care 2012;35:2665–2673 base. BMJ 2009;339:b4731 39. Pantalone KM, Kattan MW, Yu C, et al. The 54. Huang ES, Laiteerapong N, Liu JY, John PM, 26. Pantalone KM, Kattan MW, Yu C, et al. The risk risk of overall mortality in patients with type 2 Moffet HH, Karter AJ. Rates of complications of developing coronary artery disease or conges- diabetes receiving different combinations of and mortality in older patients with diabetes tive heart failure, and overall mortality, in type 2 sulfonylureas and metformin: a retrospective mellitus: the diabetes and aging study. JAMA diabetic patients receiving , piogli- analysis. Diabet Med 2012;29:1029–1035 Intern Med 2014;174:251–258 tazone, metformin, or sulfonylureas: a retrospec- 40. Jørgensen CH, Gislason GH, Bretler D, et al. 55. Nissen SE, Wolski K. Effect of rosiglitazone tive analysis. Acta Diabetol 2009;46:145–154 Glyburide increases risk in patients with diabe- on the risk of myocardial infarction and death 27. Horsdal HT, Johnsen SP, Søndergaard F, tes mellitus after emergent percutaneous inter- from cardiovascular causes. N Engl J Med 2007; et al. Sulfonylureas and prognosis after myocar- vention for myocardial infarction–a nationwide 356:2457–2471 dial infarction in patients with diabetes: a pop- study. Int J Cardiol 2011;152:327–331 56. Suissa S. Immortal time bias in pharmaco- ulation-based follow-up study. Diabetes Metab 41. Andersson C, Gislason GH, Jørgensen CH, epidemiology. Am J Epidemiol 2008;167:492–499 Res Rev 2009;25:515–522 et al. Comparable long-term mortality risk asso- 57. Cole SR, Hernan´ MA. Constructing inverse 28. Khalangot M, Tronko M, Kravchenko V, ciated with individual sulfonylureas in diabetes probability weights for marginal structural mod- Kovtun V. Glibenclamide-related excess in total patients with heart failure. Diabetes Res Clin els. Am J Epidemiol 2008;168:656–664 and cardiovascular mortality risks: data from Pract 2011;94:119–125 58. Krentz AJ, Bailey CJ. Oral antidiabetic large Ukrainian observational cohort study. 42. Pantalone KM, Kattan MW, Yu C, et al. agents: current role in type 2 diabetes mellitus. Diabetes Res Clin Pract 2009;86:247–253 Increase in overall mortality risk in patients Drugs 2005;65:385–411 29. Azoulay L, Schneider-Lindner V, Dell’aniello with type 2 diabetes receiving glipizide, glybur- 59. Abdelmoneim AS, Hasenbank SE, Seubert S, Schiffrin A, Suissa S. Combination therapy ide or glimepiride monotherapy versus metfor- JM, Brocks DR, Light PE, Simpson SH. Variations with sulfonylureas and metformin and the pre- min: a retrospective analysis. Diabetes Obes in tissue selectivity amongst insulin secreta- vention of death in type 2 diabetes: a nested Metab 2012;14:803–809 gogues: a systematic review. Diabetes Obes case-control study. Pharmacoepidemiol Drug 43. Roumie CL, Hung AM, Greevy RA, et al. Metab 2012;14:130–138 Saf 2010;19:335–342 Comparative effectiveness of sulfonylurea and 60. Zinman B, Wanner C, Lachin JM, et al.; EMPA- 30. Andersson C, Olesen JB, Hansen PR, et al. metformin monotherapy on cardiovascular REG OUTCOME Investigators. Empagliflozin, Metformin treatment is associated with a low events in type 2 diabetes mellitus: a cohort cardiovascular outcomes, and mortality in risk of mortality in diabetic patients with heart study. Ann Intern Med 2012;157:601–610 type 2 diabetes. N Engl J Med 2015;373: failure: a retrospective nationwide cohort 44. Juurlink DN, Gomes T, Shah BR, Mamdani 2117–2128 study. Diabetologia 2010;53:2546–2553 MM. Adverse cardiovascular events during 61. Marso SP, Daniels GH, Brown-Frandsen K, 31. Evans JM, Doney AS, AlZadjali MA, et al. treatment with glyburide (glibenclamide) or gli- et al.; LEADER Steering Committee; LEADER Trial Effect of metformin on mortality in patients clazide in a high-risk population. Diabet Med Investigators. and cardiovascular with heart failure and type 2 diabetes mellitus. 2012;29:1524–1528 outcomes in type 2 diabetes. N Engl J Med Am J Cardiol 2010;106:1006–1010 45.CurrieCJ,PooleCD,EvansM,PetersJR, 2016;375:311–322 32. Jørgensen CH, Gislason GH, Andersson C, Morgan CL. Mortality and other important 62. Marso SP, Bain SC, Consoli A, et al.; et al. Effects of oral glucose-lowering drugs on diabetes-related outcomes with insulin vs other SUSTAIN-6 Investigators. and car- long term outcomes in patients with diabetes antihyperglycemic therapies in type 2 diabetes. diovascular outcomes in patients with type 2 mellitus following myocardial infarction not J Clin Endocrinol Metab 2013;98:668–677 diabetes. N Engl J Med 2016;375:1834–1844 treated with emergent percutaneous coronary 46. Wheeler S, Moore K, Forsberg CW, et al. Mor- 63. Scirica BM, Bhatt DL, Braunwald E, et al.; intervention–a retrospective nationwide cohort tality among veterans with type 2 diabetes initi- SAVOR-TIMI 53 Steering Committee and Inves- study. Cardiovasc Diabetol 2010;9:54 ating metformin, sulfonylurea or rosiglitazone tigators. Saxagliptin and cardiovascular out- 33. MacDonald MR, Eurich DT, Majumdar SR, monotherapy. Diabetologia 2013;56:1934–1943 comes in patients with type 2 diabetes mellitus. et al. Treatment of type 2 diabetes and outcomes 47. Bo S, Castiglione A, Ghigo E, et al. Mortality N Engl J Med 2013;369:1317–1326 in patients with heart failure: a nested case-control outcomes of different sulphonylurea drugs: the 64. Yu OH, Filion KB, Azoulay L, Patenaude V, study from the U.K. General Practice Research results of a 14-year cohort study of type 2 diabetic Majdan A, Suissa S. Incretin-based drugs and the Database. Diabetes Care 2010;33:1213–1218 patients. Eur J Endocrinol 2013;169:117–126 risk of congestive heart failure. Diabetes Care 34. Pantalone KM, Kattan MW, Yu C, et al. The 48. Ghotbi AA, Køber L, Finer N, et al. Association 2015;38:277–284 risk of overall mortality in patients with type 2 of hypoglycemic treatment regimens with cardio- 65. Chen DY, Wang SH, Mao CT, et al. diabetes receiving glipizide, glyburide, or glime- vascular outcomes in overweight and obese sub- and cardiovascular outcomes in diabetic pa- piride monotherapy: a retrospective analysis. jects with type 2 diabetes: a substudy of the tients with chronic kidney disease and acute Diabetes Care 2010;33:1224–1229 SCOUT trial. Diabetes Care 2013;36:3746–3753 myocardial infarction: a nationwide cohort 35. Zeller M, Danchin N, Simon D, et al.; French 49. Nagendran J, Oudit GY, Bakal JA, Light PE, study. Int J Cardiol 2015;181:200–206 Registry of Acute ST-Elevation and Non-ST- Dyck JR, McAlister FA. Are users of sulphonylur- 66. Ou SM, Shih CJ, Chao PW, et al. Effects on Elevation Myocardial Infarction investigators. eas at the time of an acute coronary syndrome clinical outcomes of adding dipeptidyl peptidase- Impact of type of preadmission sulfonylureas on at risk of poorer outcomes? Diabetes Obes 4 inhibitors versus sulfonylureas to metformin mortality and cardiovascular outcomes in diabetic Metab 2013;15:1022–1028 therapy in patients with type 2 diabetes mellitus. patients with acute myocardial infarction. J Clin 50. Bannister CA, Holden SE, Jenkins-Jones S, Ann Intern Med 2015;163:663–672 Endocrinol Metab 2010;95:4993–5002 et al. Can people with type 2 diabetes live longer 67. Wang SH, Chen DY, Lin YS, et al. Cardiovas- 36. Corrao G, Romio SA, Zambon A, Merlino L, than those without? A comparison of mortality in cular outcomes of sitagliptin in type 2 diabetic Bosi E, Scavini M. Multiple outcomes associated people initiated with metformin or sulphonylurea patients with acute myocardial infarction, a 714 Sulfonylureas and Cardiovascular Risk Diabetes Care Volume 40, May 2017

population-based cohort study in Taiwan. patients with type 2 diabetes: an observational 72. Filion KB, Azoulay L, Platt RW, et al.; CNO- PLoS One 2015;10:e0131122 study. Diabetes Care 2016;39:726–734 DES Investigators. A multicenter observational 68. Giorda CB, Picariello R, Tartaglino B, et al. 70. Chang CH, Chang YC, Lin JW, et al. No in- study of incretin-based drugs and heart failure. Hospitalisation for heart failure and mortality creased risk of hospitalization for heart failure N Engl J Med 2016;374:1145–1154 associated with dipeptidyl peptidase 4 (DPP- for patients treated with dipeptidyl peptidase- 73.WeirDL,McAlisterFA,SenthilselvanA, 4) inhibitor use in an unselected population 4 inhibitors in Taiwan. Int J Cardiol 2016;220: Minhas-Sandhu JK, Eurich DT. Sitagliptin use in of subjects with type 2 diabetes: a nested 14–20 patients with diabetes and heart failure: a pop- case-control study. BMJ Open 2015;5: 71. Toh S, Hampp C, Reichman ME, et al. Risk ulation-based retrospective cohort study. JACC e007959 for hospitalized heart failure among new users Heart Fail 2014;2:573–582 69. Fu AZ, Johnston SS, Ghannam A, et al. As- of saxagliptin, sitagliptin, and other antihyper- 74. Wang KL, Liu CJ, Chao TF, et al. Sitagliptin and sociation between hospitalization for heart fail- glycemic drugs: a retrospective cohort study. the risk of hospitalization for heart failure: a pop- ure and dipeptidyl peptidase 4 inhibitors in Ann Intern Med 2016;164:705–714 ulation-based study. Int J Cardiol 2014;177:86–90