Association Between Thiazolidinedione Treatment and Risk of Macular Edema Among Patients with Type 2 Diabetes

Home , Thiazolidinedione

ORIGINAL INVESTIGATION Association Between Thiazolidinedione Treatment and Risk of Macular Edema Among Patients With Type 2 Diabetes

Iskandar Idris, MD, FRCP, FRCP(Edin); Graham Warren, PhD; Richard Donnelly, MD, PhD

Background: Findings of prior studies have been incon- to adjust for missing values, and propensity score analysis clusive about the ocular effects of thiazolidinediones on to exclude for any selection bias, thiazolidinedione use was diabetic macular edema (DME). We evaluated, in pa- associated with an increased risk of DME at 1-year fol- tients with type 2 diabetes (T2D), the short-term and long- low-up (OR, 2.3 [95% CI, 1.5-3.6]) and 10-year fol- term risks of developing DME among users vs nonusers low-up (hazard ratio [HR], 2.3; [95% CI, 1.7-3.0]). The of thiazolidinediones. effect was similar for pioglitazone and rosiglitazone. Com- bination therapy with insulin plus a thiazolidinedione was Methods: A retrospective cohort study of 103 368 pa- associated with a higher risk of DME after propensity score tients with T2D and no DME at baseline using The Health adjustment (HR, 3.0 [95% CI, 1.5-5.9]), while aspirin use Improvement Network (THIN) database. Clinical, bio- (HR, 0.6 [95% CI, 0.4-0.9]) and angiotensin-converting chemical, and demographic information was obtained for enzyme inhibitor use (HR, 0.4 [95% CI, 0.2-0.7]) were as- the period January 1, 2000, through November 30, 2009. sociated with a reduced risk of DME. Results: At 1 year, the incidence of DME was 1.3% (n=41) Conclusion: Among patients with T2D, treatment with and 0.2% (n=227) among thiazolidinedione users (n=3227) and nonusers (n=100 141), respectively (odds a thiazolidinedione was associated with an increased risk ratio [OR], 5.7 [95% CI, 4.1-7.9]). After Cox multiple of DME at 1-year and 10-year follow-up evaluations. regression analysis (adjusted for age; systolic blood pressure; levels of lipids and hemoglobin A1c; and use of aspi- Arch Intern Med. 2012;172(13):1005-1011. rin, fibrates, insulin, oral antidiabetic drugs, or renin- Published online June 11, 2012. angiotensin system blockers), multiple imputation analysis doi:10.1001/archinternmed.2012.1938

HE THIAZOLIDINEDIONES, ure,7 and, more recently, potentially in- pioglitazone and rosiglita- creased risk of bladder cancer.8 zone, are peroxisome pro- The PPAR-␥ receptors are expressed liferator–activated recep- abundantly in endothelial, renal, and reti- tor ␥ (PPAR-␥) agonists nal vascular tissues.9,10 Thiazolidinedione- thatT ameliorate peripheral and hepatic in- induced peripheral and pulmonary edema sulin resistance and are effective glucose- lowering treatments for patients with type 2 diabetes mellitus.1 The American Dia- See Invited Commentary betes Association (ADA) and the Euro- at end of article pean Association for the Study of Diabe- tes (EASD) recommend thiazolidinediones See also page 1014 as second- or third-line therapy in combination with other oral agents or insulin is well recognized, and the underlying Author Affiliations: Sherwood Author Affil to achieve target levels of glycemic con- mechanisms are likely to include vasodi- Forest Hospitals Foundation 2 Forest Hosp Trust, Nottinghamshire, trol. The risk-benefit ratio for thiazoli- lation, renal sodium reabsorption, and di- Trust, Nottin England (Dr Idris); Trent dinedione use has been the subject of in- rect effects on vascular endothelial per- England (Dr Research Design Service for tense discussion following a series of meability.11-13 More recently, however, Research De the East Midlands, Nottingham, metabolic and cardiovascular outcome small clinical studies have suggested an as- the East Mid England (Dr Warren); and studies.3,4 Several analyses have high- sociation between thiazolidinediones and England (Dr School of Graduate Entry 14 School of Gr Medicine and Health, lighted the major adverse effects of thia- diabetic macular edema (DME), one of Medicine an University of Nottingham, zolidinediones, including an increased in- the major sight-threatening complica- University o 5 Nottingham (Drs Idris and cidence of bone fractures, fluid retention tions that affects up to 20% of patients with Nottingham Donnelly). and edema,6 increased risk of heart fail- type 2 diabetes.15 Donnelly).

ARCH INTERN MED/ VOL 172 (NO. 13), JULY 9, 2012 WWW.ARCHINTERNMED.COM 1005

Downloaded From: https://jamanetwork.com/ on 09/24/2021 A cohort study from the Kaiser Permanente database STUDY VARIABLES reported an increased incidence of DME following short- term (1-year) exposure to pioglitazone,16 but no asso- Medical and demographic codes for the parameters of interest ciation between thiazolidinediones and DME was ob- included age; body mass index (BMI; calculated as weight in served in a cross-sectional analysis of baseline data in the kilograms divided by height in meters squared); body weight; 17 levels of hemoglobin A1c (HbA1c) and serum lipids (total cho- ACCORD eye substudy, and neither of these studies in- lesterol, low-density lipoprotein cholesterol [LDL-C], high- vestigated the potential longer-term effects of thiazoli- density lipoprotein cholesterol [HDL-C], and triglycerides); use dinedione use. Accordingly, the aim of this study was to of specific antihypertensive drugs, lipid-lowering therapies, in- evaluate the short- and long-term effects of thiazolidine- sulin, and aspirin; and systolic and diastolic blood pressure (BP). dione treatment (pioglitazone and rosiglitazone) on the risks of developing DME in a large population cohort and DME END POINT to identify any risk factors that may influence visual outcome in patients with type 2 diabetes treated with a thia- The primary outcome measure of the study was a new diagno- zolidinedione. sis of DME as indicated by READ codes (including all codes relating to a diagnosis of DME and/or focal laser photocoagulation for DME) at 1 year and 10 years from the patient’s entry into the study. A diagnosis of DME recorded on the THIN da- METHODS tabase would typically originate from a specialist ophthalmologic assessment of the patient following routine annual eye STUDY DESIGN AND DATA SOURCE screening. Specialist-derived diagnoses of DME on THIN have been validated for accuracy,20 but some degree of measure- ment error is still possible. The study was approved by the National Research Ethics Com- Comprehensive diabetic retinopathy screening in the United mittee East of England–Cambridge South (reference No. 10/ Kingdom is free to patients and is based on a standardized pro- H0305/46). This was a retrospective cohort study using The tocol, independent of drug prescription, comorbidities, and dia- Health Improvement Network (THIN) database. THIN is an betes consultations. National screening services aim to rou- established database that collects anonymized electronic data tinely screen all patients with diabetes yearly for the presence from a volunteer sample of United Kingdom general practices of diabetic retinopathy, including DME. Beginning in 2005, all using the Vision primary care computer system (In Practice Sys- retinopathy screening performed in England and Wales has been tems, London, England). THIN includes medical records on based on digital photographic images, in line with recommen- 9.1 million patients (3.4 million of whom are alive and regis- dations from the National Diabetic Retinopathy Screening pro- tered with a practice currently contributing to THIN at the date gram. Patients with suspected DME at screening are routinely of last collection) attending approximately 479 general prac- referred for specialist ophthalmologic assessment at the Hos- tices in England and Wales. Data recorded in THIN include pital Eye Service. The coded diagnosis of DME, or a coded di- demographic information, medical diagnoses, prescribed medi- agnosis of focal laser photocoagulation (a specific therapy for cations, laboratory results, lifestyle characteristics, and refer- DME), is then recorded in the THIN database. Information on rals to specialists. Past and current medical diagnoses are re- visual acuity is unfortunately not available in THIN. corded using READ codes (a thesaurus of coded medical terms The THIN database was queried to determine the use of pre- maintained and distributed by the United Kingdom Terminol- scribed thiazolidinediones prior to the diagnosis of DME or prior ogy Center). THIN has been shown to be highly representa- to censuring of patients with no diagnosis of DME. Informa- tive of the wider United Kingdom population. It has been well tion on drug prescription was extracted from the database. validated at both the practice and data set level by comparison with other national data sources for demographics and mor- STATISTICAL ANALYSIS bidity, mortality, prevalence, and geographic rates.18,19 Baseline clinical and demographic variables were compared STUDY POPULATION between users and nonusers of thiazolidinediones using the t test or the Wilcoxon rank-sum test for continuous variables, ␹2 Patients selected for analysis had a diagnosis of type 2 diabetes and the test or the 2-tailed Fisher exact test for categorical variables. (defined as diabetes not requiring insulin therapy within 6 A2ϫ2 contingency table was used to calculate the unad- months of diagnosis), were 18 years or older, and were regis- justed odds ratio (OR) of DME incidence at 1 year with rosi- tered with the medical practices for at least 1 year from the in- glitazone and pioglitazone exposure. Multiple logistic regres- dex date of January 2000. A diagnosis of diabetes was based sion analysis was used as the primary analysis to calculate the on READ codes for “diabetes” diagnosis and/or use of glucose- adjusted OR controlling for the following covariates: systolic lowering therapies. The index date was chosen based on the and diastolic BP, HbA1c and lipid levels; body weight; BMI; and licensing of both rosiglitazone and pioglitazone in the United the use of insulin, oral antidiabetic therapies, ACE inhibitors, Kingdom. The last date of follow-up for this study was No- angiotensin II receptor blockers, lipid-lowering therapies, and/or vember 30, 2009. According to figures from the THIN data- aspirin. All values were measured at baseline. The possible con- base, 95% of repeat prescriptions in general practice have a pe- founding variables were prechosen on clinical grounds, includ- riodicity (frequency) of 6 months or less. Patients with less than ing the thiazolidinedione-insulin interaction, and variables to 6 months of exposure to a thiazolidinedione were excluded from be included in the final model were chosen by comparing the the study. Individuals were included in the analysis if they had log-likelihoods of competing models. A diagnostic plot (log- no history of DME before baseline. The patient cohort was iden- [time] vs log[−log(survival function)]) confirmed that the con- tified using relational database tools that involve strategies to stant proportionality assumptions were met. In addition, the design, code, test, and run programs to identify patients’ in- Hosmer and Lemeshow parameter did not show evidence for formation and outcome parameters. lack of fit (P=.44).

ARCH INTERN MED/ VOL 172 (NO. 13), JULY 9, 2012 WWW.ARCHINTERNMED.COM 1006

Downloaded From: https://jamanetwork.com/ on 09/24/2021 Kaplan-Meier curves were constructed using 10-year outcome data to illustrate graphically the difference in DME inci- Table 1. Baseline Patient Demographic dence according to exposure to a thiazolidinedione. A log- and Clinical Characteristicsa rank test was used to check for statistically significant differences between users and nonusers of thiazolidinediones for the 10- Nonusers of Users of year follow-up and subgroup analysis for 6-month follow-up. Thiazolidinediones Thiazolidinediones Cox regression analysis was used to calculate DME inci- Characteristic (n = 100 141) (n = 3227) dence hazard ratios (HRs) adjusted for confounding variables. Age, y 56.4 (15.7) 58.2 (13.4) All possible confounders, including the thiazolidinedione- Sex, No. (%) insulin interaction and aspirin and/or ACE-inhibitor use, were Female 52 323 (52.2) 1869 (57.9) preplanned, and the model was selected by comparing model Male 47 818 (47.8) 1358 (42.1) log-likelihoods. Threshold significance was given at the con- BMI 30.7 (6.3) 32.4 (7.2) ventional level of P=.05, and 95% CIs are shown for indi- Weight 86.6 (19.9) 92.6 (23.1) vidual HRs. Systolic BP, mm Hg 138.7 (19.0) 134.5 (16.6) To address missing values in the Cox regression analysis, Diastolic BP, mm Hg 81.0 (10.4) 77.7 (9.8) multiple imputation was performed using a linear regression HbA1c, % 7.6 (1.7) 8.1 (1.7) Total cholesterol, 5.2 (1.2) 4.6 (1.1) method to impute 10 HbA1c values per missing case. Multiple imputation was not used in the final logistic regression model mmol/L because HbA level, the variable with a significant number of HDL cholesterol, 1.4 (1.0) 1.3 (0.3) 1c mmol/L missing values, was not included in the final model. LDL cholesterol, 3.4 (1.5) 2.9 (1.7) In the United Kingdom Prospective Diabetes Study, HbA1c mmol/L level was shown to increase with diabetes duration, regardless Prescriptions, No. (%) 21 of intensive or conventional glucose control. We therefore per- Aspirin 8935 (8.9) 35 (1.1) formed a sensitivity analysis for both the multiple logistic and Fibrates 552 (0.6) 0 Cox regression results by performing propensity score analy- Insulin 5467 (5.5) 323 (10.0) 22 ses, as determined by HbA1c level and other covariates to ad- RAS blockers 10 208 (10.2) 54 (1.7) just for treatment selection bias and to ensure that both co- ACE inhibitors 8752 (8.7) 46 (1.4) horts were comparable. A nonparsimonious logistic regression Statins 7781 (7.8) 49 (1.5) analysis was used to generate a propensity score, which is a pre- dictor for thiazolidinedione prescription based on the ob- Abbreviations: ACE, angiotensin-converting enzyme; BMI, body mass served covariates (sex, Townsend index, age, BMI, systolic and index (calculated as weight in kilograms divided by height in meters diastolic BP, HbA level, and aspirin and/or insulin use). Pre- squared); BP, blood pressure; HbA1c, hemoglobin A1c concentration; 1c HDL, high-density lipoprotein; LDL, low-density lipoprotein; vious studies have shown that stratification by propensity score RAS, renin-angiotensin system. into 5 strata is sufficient to remove 90% to 95% of the selec- SI conversion factors: To convert all cholesterol measurements tion bias.23 An alternative method of kernel matching in which to conventional units, divide by 0.0259. the outcome for each exposed individual was compared with a a Unless otherwise indicated, data are reported as mean (SD) values. weighted mean of the outcomes of unexposed individuals was used to check for the sensitivity of the method of using the pro- sis to correct for missing values, we found that exposure pensity scores to reduce bias.24 STATA 11 software (StataCorp LP) was used for the statis- to a thiazolidinedione was associated with a signifi- tical analysis and IBM Modeler 13 for the database work. Re- cantly increased risk of DME at 1 year (OR, 3.3 [95% CI, sults are expressed as ORs for logistic regression and HRs for 2.2-5.0]). The analysis for pioglitazone and rosiglita- Cox regression analysis. zone individually showed that each drug was separately associated with a significantly increased risk of DME at 1 RESULTS year, and there was no significant difference between the 2 thiazolidinediones (pioglitazone OR, 3.6 [95% CI, 2.0- 6.6]; rosiglitazone OR, 3.1 [95% CI, 1.9-5.1]). The analysis identified 109 295 patients in 464 prac- The increased risk of DME at 1 year was still present tices who fulfilled the initial selection criteria. Of these, when propensity score methods used to remove selec- 5927 patients were not included in the analysis because tion bias (stratification OR, 2.3 [95% CI, 1.5-3.6]; kernel- the baseline date for these records was earlier than the matching OR, 2.7 [95% CI, 1.6-4.4]). The potential con- acceptable mortality reporting date for the practice, which founders that did not change the magnitude of associations made them potentially less reliable. Thus, the cohort for of interest and were therefore excluded were age, HbA1c, further analysis included 103 368 patients with type 2 dia- BMI, lipid measurements, and use of aspirin, fibrates or betes. The baseline characteristics for users and nonus- angiotensin receptor blockers (Table 3). ers of thiazolidinediones are listed in Table 1. OUTCOMES AT 10 YEARS OUTCOMES AT 1 YEAR Ten years of follow-up data for the patient cohort The incidence of DME at 1 year of follow-up according (n = 103 359) were used to construct Kaplan-Meier curves to thiazolidinedione use is listed in Table 2. Patients for time to DME incidence and were further analyzed by exposed to a thiazolidinedione had a significantly higher Cox regression analysis. The log-rank test yielded an ␹2 risk of developing DME (OR, 5.7 [95% CI, 4.1-7.9]) statistic of 373 (P Ͻ .001) and showed a clear differ- (Fisher exact test P Ͻ .001). ence in DME incidence between users and nonusers of Following multiple logistic regression to correct thiazolidinediones (HR, 5.2 [95% CI, 4.3-6.3]) (P Ͻ .001) for confounding factors and multiple imputation analy- (Figure). A log-rank test was also performed for time

ARCH INTERN MED/ VOL 172 (NO. 13), JULY 9, 2012 WWW.ARCHINTERNMED.COM 1007

Downloaded From: https://jamanetwork.com/ on 09/24/2021 Table 2. Unadjusted Incidence of DME After 1 Year According to Thiazolidinedione Usea

DME at 1-Year Follow-up, No. (%) of Patients [95% CI] Thiazolidinedione Status Yes No Total No. TZD users 41 (1.3) [0.9-1.7] 3186 (98.7) [98.3-99.1] 3227 Non TZD users 227 (0.2) [0.2-0.3] 99 914 (99.8) [99.7-99.8] 100 141 Total NR NR 103 368

Abbreviations: DME, diabetic macular edema; NR, not reported. a The odds ratio for development of DME at 1 year among thiazolidinedione users vs nonusers was 5.7 (95% CI, 4.1-7.9) (Fisher exact test P Ͻ .001).

Table 3. ORs for Incidence of DME After 1 Year Among Thiazolidinedione Users After Adjustment for Significant Covariatesa

Multivariable Model Propensity Score Stratification Clinical Characteristic OR (95% CI) P Value OR (95% CI) P Value Glitazone, N 1 [Reference] NA 1 [Reference] NA Insulin, Nb Glitazone, Y 3.33 (2.23-4.96) Ͻ.001 2.31 (1.49-3.58) Ͻ.001 Insulin, N Glitazone, Y 8.73 (4.3-17.46) Ͻ.001 5.94 (2.72-12.98) Ͻ.001 Insulin, Y Systolic BP per 10–mm Hg 1.07 (1.00-1.14) .045 1.08 (1.00-1.17) .06 increase ACE inhibitor use 0.39 (0.18-0.82) .01 0.28 (0.11-0.69) .005 Age, per 10-y increase 1.08 (1.00-1.17) .06 1.00 (0.91-1.10) .97

Abbreviations: ACE, angiotensin-converting enzyme; BP, blood pressure; DME, diabetic macular edema; N, no; NA, not applicable; OR, odds ratio; Y, yes. a Main logistic regression with multiple imputation analysis analyzed and shown with and without propensity score stratification to exclude indication bias. Nonprescription of drugs is the reference in all cases, so an OR greater than 1 indicates that the use of that drug increases the risk of DME. The confounders adjusted for in this model but not included in the table because not significant (P Ͼ .05) and not improving the model significantly (comparing model

log-likelihoods) were hemoglobin A1c concentration, BMI, total cholesterol level, serum triglycerides levels, and aspirin, fibrates, and angiotensin receptor blocker use. The unadjusted OR for development of DME at 1 year among thiazolidinedione users vs nonusers was 5.66 (95% CI, 4.05-7.91). Following propensity score analysis, the unadjusted OR was 2.28 (95% CI, 1.56-3.34). b Reference category incidence was 0.2%.

tients exposed to a thiazolidinedione (HR, 2.8 [95% CI, 12.5 2.1-3.5]) (P Ͻ .001); after adjusting for selection bias TZD + insulin TZD using propensity score stratification, we found a similar 10.0 No TZD result (HR, 2.3 [95% CI, 1.7-3.0]). In this multiple regression model, HbA1c, systolic BP, aspirin use, ACE- 7.5 inhibitor use, and serum triglyceride levels were variables that produced substantial changes in the magnitudes 5.0 of the associations of interest (Table 4). A plot to test for the proportional hazard assumption showed that the 2.5

Cumulative DME Incidence, % observed and predicted curves agreed and only deviated when the number of DME events in the thiazolidinedi- 0.0 0 20 40 60 80 100 120 one group small at longer-term follow-up. Time to DME Diagnosis, mo Further Cox regression and multiple imputation analysis using an interaction model showed that combina- Patients at risk 103 359 92 451 81 139 70 885 58 178 43 449 25 817 tion therapy with insulin and a thiazolidinedione increased the HR of DME even further (HR, 4.4 [95% CI, Figure. Kaplan-Meier time to diabetic macular edema (DME) curves according 2.5-7.8]; or HR with propensity score adjustment, 3.0 to thiazolidinedione use with or without insulin. The log-rank test gives an ␹2 statistic of 373 (P Ͻ .001), which shows a clear difference in DME incidence [95% CI, 1.5-5.9]). Conversely, concurrent use of aspi- according to thiazolidinedione use. In a comparison of thiazolidinedione use rin (HR, 0.7 [95% CI, 0.5-0.9]) and an ACE inhibitor (HR, with nonuse, the hazard ratio was 5.19 (95% CI, 4.31-6.25). 0.5 [95% CI, 0.3-0.9]) was associated with a reduction in the HR of developing DME independent of other co- to DME up to 6 months with P Ͻ .001, which shows more variates and after propensity score stratification: HR, 0.6 clearly that the increased risk was also present at short- (95% CI, 0.4-0.9) and HR, 0.4 (95% CI, 0.2-0.7), respec- term follow-up. tively (Table 4). Following adjustment for confounding factors using To further strengthen the analysis, propensity scores Cox regression and multiple imputation analysis to ad- were derived to exclude indication bias and produce com- just for missing values, we found that there was a sig- parable subgroups for a stratified analysis. The poten- nificantly increased risk of developing DME among pa- tial confounding variables for users and nonusers of thia-

ARCH INTERN MED/ VOL 172 (NO. 13), JULY 9, 2012 WWW.ARCHINTERNMED.COM 1008

Downloaded From: https://jamanetwork.com/ on 09/24/2021 Table 4. Hazard Ratios for Incidence of DME After 10 Years Among Thiazolidinedione Users After Adjustment for Significant Covariatesa

Multivariable Model Propensity Score Stratification Clinical Characteristic HR (95% CI) P Value HR (95% CI) P Value Glitazone, N and insulin, Nb 1 [Reference] NA 1 [Reference] NA Glitazone, Y and insulin, N 2.75 (2.13-3.54) Ͻ.001 2.29 (1.74-3.02) Ͻ.001 Glitazone, Y and insulin, Y 4.39 (2.46-7.84) Ͻ.001 3.00 (1.54-5.88) .001 Systolic BP per 10–mm Hg increase 1.09 (1.04-1.13) Ͻ.001 1.07 (1.02-1.12) .004 Triglycerides per 1–mmol/L increase 0.94 (0.90-0.99) .02 0.93 (0.88-0.99) .02 ACE inhibitor 0.53 (0.30-0.92) .03 0.37 (0.20-0.69) .002 Aspirin 0.67 (0.50-0.89) .01 0.58 (0.39-0.86) .02

HbA1c, Ն9% vs Ͻ9% 1.38 (1.15-1.66) .001 1.70 (1.30-2.10) Ͻ.001

Abbreviations: ACE, angiotensin-converting enzyme; BP, blood pressure; DME, diabetic macular edema; HbA1c, hemoglobin A1c; N, no; NA, not applicable; OR, odds ratio; Y, yes. a This analysis uses all 10 years of follow-up for the patients in a Cox regression analysis with multiple imputation, results shown with and without propensity score stratification. The confounders adjusted for in this model but not included in the table because not significant and not improving the model significantly are age, BMI, total cholesterol level, and angiotensin receptor blocker use. The unadjusted hazard ratio (95% CI) comparing use with nonuse of thiazolidinedione for the risk of DME was 5.19 (4.31-6.25). Following propensity score analysis, the unadjusted hazard ratio was 2.21 (1.79-2.73). b Reference category incidence was 0.2%.

Table 5. Distributions of Key Variables by Thiazolidinedione Use and Propensity Score Quintilea

Age, y HbA1c, % Systolic BP, mm Hg Insulin Use, % Quintileb No TZD TZD No TZD TZD No TZD TZD No TZD TZD 1 55.7 (13.6) 55.5 (12.2) 8.4 (2.1) 8.5 (2.0) 129 (15.2) 129 (14.8) 5.6 4.3 2 58.6 (13.7) 58.3 (12.9) 7.6 (1.7) 7.8 (1.6) 133 (15.0) 133 (13.9) 7.7 6.8 3 59.4 (13.8) 58.9 (13.1) 7.3 (1.5) 7.6 (1.3) 137 (15.1) 137 (15.3) 10.1 8.9 4 60.4 (14.4) 59.0 (12.7) 7.3 (1.4) 7.7 (1.4) 143 (16.5) 142 (16.1) 16.7 18.5 5 64.0 (13.9) 59.2 (14.3) 7.3 (1.4) 8.2 (1.6) 148 (20.9) 147 (20.4) 23.5 41.6

Abbreviations: BP, blood pressure; DME, diabetic macular edema; HbA1c, hemoglobin A1c level; TZD, thiazolidinedione. a Unless otherwise indicated, data are reported as mean (SD) values. b Comparison of age, HbA1c, systolic BP, and proportion using insulin between TZD users and nonusers by propensity score quintile. Patients in quintile 1 were more likely to be users of a thiazolidinedione.

zolidinediones were generally balanced across the with those patients who develop thiazolidinedione- propensity score quintiles (Table 5). induced pedal edema,14,27 and in large clinical trials the highest incidence of fluid retention and peripheral edema COMMENT (16%) was observed in the subgroup of patients undergoing combination therapy with a thiazolidinedione and 28 This large retrospective cohort study analyzed the pri- insulin. mary care electronic medical records of more than 100 000 Although glycemic control did not modulate the short- patients with type 2 diabetes and showed that, even af- term (1-year) risk of DME among users of thiazolidine- ter adjustment for various confounding factors known diones, HbA1c was significant in the longer-term analy- to influence diabetic retinopathy, exposure to a thiazol- sis. In addition, the use of aspirin and/or ACE inhibitor idinedione is associated with an increased risk of devel- drugs was associated with a modest reduction in the HR oping DME. The association was evident with both pio- of DME, which is consistent with previous clinical and 25 glitazone and rosiglitazone. A previous short-term study experimental studies. from Kaiser Permanente reported a similar increase in Diabetic macular edema is the leading cause of blind- risk of DME after 1 year (OR, 2.6),16 but the analysis did ness among patients with type 2 diabetes.15,29 The abso- not adjust for several key variables, in particular the use lute rate of new-onset DME was much lower in the pres- of BP-lowering drugs such as angiotensin receptor block- ent study (about 1%) than has been reported in previous ers and ACE inhibitors, which may affect the progres- clinical trials of diabetic retinopathy.29 This probably re- sion of diabetic retinopathy.25,26 flects the low detection rates in routine clinical practice Diabetic macular edema is a sight-threatening chronic and highlights the importance of a national systematic condition,14,15 so an important new observation in the pres- screening program for diabetic retinopathy. The cause ent study is that the increased risk of DME continued to of DME is complex, but disruption to the blood-retinal accrue during the 10-year of follow-up. The patients at barrier (BRB), mediated in part by local release of vas- greatest risk of developing DME were those taking thia- cular endothelial growth factor (VEGF),30 results in zolidinediones in combination with insulin. Diabetic chronic leakage of fluid and reduced visual acuity.15 Im- macular edema has been associated in clinical practice portant regulators of VEGF include hypoxia, hypergly-

ARCH INTERN MED/ VOL 172 (NO. 13), JULY 9, 2012 WWW.ARCHINTERNMED.COM 1009

Downloaded From: https://jamanetwork.com/ on 09/24/2021 cemia, and angiotensin II, but PPAR-␥ agonists also in- PPAR-␥ (or dual ␣,␥) agonists should prospectively crease VEGF expression,31 and our research group12 has evaluate the ocular safety of these drugs. Clinicians previously shown that thiazolidinediones increase vas- should be vigilant in the clinical screening for DME cular endothelial permeability in vitro. A number of sys- among those patients taking thiazolidinediones. temic changes associated with thiazolidinediones may also contribute (eg, sodium and fluid retention, vasodila- 32 Accepted for Publication: March 13, 2012. tion, changes in BP, microvascular perfusion). Thus, Published Online: June 11, 2012. doi:10.1001 there are plausible mechanisms, both local and sys- /archinternmed.2012.1938 temic, by which thiazolidinediones might adversely affect Correspondence: Iskandar Idris, MD, FRCP, FRCP the risk of DME, particularly in insulin-treated patients. (Edin), Department of Diabetes and Endocrinology, Kings The strengths of this cohort analysis include a very Treatment Centre, Clinic 5, Sherwood Forest Hospitals large sample size and a long duration of follow-up. To Foundation Trust, Nottinghamshire, England (iidris@aol increase the reliability of the study, we included only pa- .com). tients who were apparently compliant with thiazolidine- Ͼ Author Contributions: Dr Idris had full access to all the dione therapy ( 6 months and in receipt of repeat pre- data in the study and takes responsibility for the integ- scriptions), and we measured new-onset DME during a rity of the data and the accuracy of the data analysis. Study 10-year observation period. Diabetic macular edema is concept and design: Idris and Donnelly. Acquisition of data: a specialist diagnosis that would normally be made only Idris. Analysis and interpretation of data: Idris and War- by an ophthalmologist when the patient is referred fol- ren. Drafting of the manuscript: Idris and Donnelly. Criti- lowing routine annual eye screening. cal revision of the manuscript for important intellectual con- It is an important limitation of this study that informa- tent: Idris, Warren, and Donnelly. Statistical analysis: Idris tion about duration of diabetes and duration of indi- and Warren. Obtained funding: Idris. Administrative, tech- vidual patient exposure to a thiazolidinedione was not avail- nical, and material support: Donnelly. Study supervision: able. However, the baseline characteristics of patients in Donnelly. the 2 cohorts were comparable in terms of age, sex, and Financial Disclosure: None reported. BMI. The percentage of patients receiving insulin and mean Funding/Support: Sherwood Forest Hospitals Chari- HbA1c levels were slightly higher in the thiazolidinedione table Trust Fund (no external funding). user group, but this was balanced by slightly higher BP Role of the Sponsor: The sponsor’s role was only to pro- and more patients receiving aspirin, ACE inhibitors, and vide indemnity arrangement. statins in the nonuser group. Furthermore, the propen- Additional Contributions: We thank Mary Thomson, sity score analysis mitigates against indication bias and MD, research manager at THIN, for providing access to strengthens the interpretation of a true association be- the database. tween DME and use of a thiazolidinedione. During the study period (2000-2009), thiazolidinediones were routinely and widely used in clinical prac- REFERENCES tice in patients with diabetes of varying duration (eg, as dual or initial combination therapy with metformin in 1. Stumvoll M, Ha¨ring HU. Glitazones: clinical effects and molecular mechanisms. Ann Med. 2002;34(3):217-224. those with short-duration disease and as third-line agents 2. Nathan DM, Buse JB, Davidson MB, et al; American Diabetes Association; Eu- added to metformin/sulfonylurea combination therapy). ropean Association for Study of Diabetes. Medical management of hyperglyce- Despite propensity score analysis and the similar base- mia in type 2 diabetes: a consensus algorithm for the initiation and adjustment line profiles, it is possible, but in our view unlikely, that of therapy: a consensus statement of the American Diabetes Association and the European Association for the Study of Diabetes. Diabetes Care. 2009;32(1): the thiazolidinedione users were significantly different 193-203. than the nonusers either in diabetes duration or in their 3. Kahn SE, Haffner SM, Heise MA, et al; ADOPT Study Group. Glycemic durability cardiovascular risk profile. The present study confirms of rosiglitazone, metformin, or glyburide monotherapy. N Engl J Med. 2006; and extends other observations implicating thiazolidine- 355(23):2427-2443. dione use with an increased risk of DME, but the data 4. Home PD, Pocock SJ, Beck-Nielsen H, et al; RECORD Study Team. Rosiglita- zone evaluated for cardiovascular outcomes in oral agent combination therapy require cautious interpretation. for type 2 diabetes (RECORD): a multicentre, randomised, open-label trial. Lancet. In conclusion, this large-population study has shown 2009;373(9681):2125-2135. that, even after adjustment for a range of confounding 5. Bodmer M, Meier C, Kraenzlin ME, Meier CR. Risk of fractures with glitazones: a variables, thiazolidinediones are associated with an critical review of the evidence to date. Drug Saf. 2009;32(7):539-547. 6. Thomas ML, Lloyd SJ. Pulmonary edema associated with rosiglitazone and increased risk of DME in patients with type 2 diabetes, troglitazone. Ann Pharmacother. 2001;35(1):123-124. especially those undergoing insulin therapy. The risk 7. Singh S, Loke YK, Furberg CD. Thiazolidinediones and heart failure: a teleo-analysis. increased continuously over the 10-year duration of this Diabetes Care. 2007;30(8):2148-2153. study. The explanation for this association is unclear, 8. Lewis JD, Ferrara A, Peng T, et al. Risk of bladder cancer among diabetic pa- but plausible mechanisms described previously exist by tients treated with pioglitazone: interim report of a longitudinal cohort study. Dia- betes Care. 2011;34(4):916-922. which these drugs might adversely influence this 9. Marx N, Duez H, Fruchart JC, Staels B. Peroxisome proliferator-activated recep- important ocular complication of type 2 diabetes. A tors and atherogenesis: regulators of gene expression in vascular cells. Circ Res. larger and more detailed meta-analysis of randomized 2004;94(9):1168-1178. controlled trials (ideally in high-risk patients) will be 10. Tawfik A, Sanders T, Kahook K, Akeel S, Elmarakby A, Al-Shabrawey M. Sup- pression of retinal peroxisome proliferator-activated receptor gamma in experi- needed to clearly establish the risk-benefit profile of mental diabetes and oxygen-induced retinopathy: role of NADPH oxidase. In- thiazolidinediones in patients with, or at risk of, DME. vest Ophthalmol Vis Sci. 2009;50(2):878-884. In addition, future clinical trials to evaluate new 11. Mudaliar S, Chang AR, Henry RR. Thiazolidinediones, peripheral edema, and type

ARCH INTERN MED/ VOL 172 (NO. 13), JULY 9, 2012 WWW.ARCHINTERNMED.COM 1010

Downloaded From: https://jamanetwork.com/ on 09/24/2021 2 diabetes: incidence, pathophysiology, and clinical implications. Endocr Pract. 22. D’Agostino RB Jr. Propensity score methods for bias reduction in the compari- 2003;9(5):406-416. son of a treatment to a non-randomized control group. Stat Med. 1998;17(19): 12. Idris I, Gray S, Donnelly R. Rosiglitazone and pulmonary oedema: an acute dose- 2265-2281. dependent effect on human endothelial cell permeability. Diabetologia. 2003; 23. Rosenbaum PR, Rubin DB. Reducing bias in observational studies using sub- 46(2):288-290. classification on the propensity score. J Am Stat Assoc. 1984;79(3):516-524. 13. Sotiropoulos KB, Clermont A, Yasuda Y, et al. Adipose-specific effect of rosi- 24. Heckman J, Ichimura H, Todd P. Matching as an econometric evaluation estimator. glitazone on vascular permeability and protein kinase C activation: novel mecha- Rev Econ Stud. 1998;65(2):261-294. nism for PPARgamma agonist’s effects on edema and weight gain. FASEB J. 2006; 25. Kim JH, Kim JH, Yu YS, Cho CS, Kim KW. Blockade of angiotensin II attenuates 20(8):1203-1205. VEGF-mediated blood-retinal barrier breakdown in diabetic retinopathy. J Cereb 14. Ryan EH Jr, Han DP, Ramsay RC, et al. Diabetic macular edema associated with Blood Flow Metab. 2009;29(3):621-628. glitazone use. Retina. 2006;26(5):562-570. 26. Gilbert RE, Kelly DJ, Cox AJ, et al. Angiotensin converting enzyme inhibition 15. Ciulla TA, Amador AG, Zinman B. Diabetic retinopathy and diabetic macular edema: reduces retinal overexpression of vascular endothelial growth factor and pathophysiology, screening, and novel therapies. Diabetes Care. 2003;26(9): hyperpermeability in experimental diabetes. Diabetologia. 2000;43(11):1360- 2653-2664. 1367. 16. Fong DS, Contreras R. Glitazone use associated with diabetic macular edema. 27. Colucciello M. Vision loss due to macular edema induced by rosiglitazone treat- Am J Ophthalmol. 2009;147(4):583-586, e1. ment of diabetes mellitus. Arch Ophthalmol. 2005;123(9):1273-1275. 17. Ambrosius WT, Danis RP, Goff DC Jr, et al; ACCORD Study Group. Lack of as- 28. Raskin P, Rendell M, Riddle MC, Dole JF, Freed MI, Rosenstock J; Rosiglitazone sociation between thiazolidinediones and macular edema in type 2 diabetes: the Clinical Trials Study Group. A randomized trial of rosiglitazone therapy in patients ACCORD eye substudy. Arch Ophthalmol. 2010;128(3):312-318. with inadequately controlled insulin-treated type 2 diabetes. Diabetes Care. 2001; 18. Bourke A, Dattani H, Robinson M. Feasibility study and methodology to create a 24(7):1226-1232. quality-evaluated database of primary care data. Inform Prim Care. 2004;12(3): 29. Klein R, Klein BEK, Moss SE, Cruickshanks KJ. The Wisconsin Epidemiologic 171-177. Study of Diabetic Retinopathy. XV. The long-term incidence of macular edema. 19. Lewis JD, Schinnar R, Bilker WB, Wang X, Strom BL. Validation studies of the Ophthalmology. 1995;102(1):7-16. health improvement network (THIN) database for pharmacoepidemiology research. 30. Aiello LP, Avery RL, Arrigg PG, et al. Vascular endothelial growth factor in ocu- Pharmacoepidemiol Drug Saf. 2007;16(4):393-401. lar fluid of patients with diabetic retinopathy and other retinal disorders. N Engl 20. Martin-Merino E, Fortuny J, Plana E, Rivero E, Garcia-Rodriguez L. Incidence J Med. 1994;331(22):1480-1487. of Diabetic Retinopathy and Diabetic Macular Oedema in newly diagnosed 31. Emoto M, Anno T, Sato Y, et al. Troglitazone treatment increases plasma vas- type 1 and type 2 diabetic patients. Pharmacoepidemiol Drug Saf. 2010;19: cular endothelial growth factor in diabetic patients and its mRNA in 3T3-L1 S1-S347. adipocytes. Diabetes. 2001;50(5):1166-1170. 21. UK Prospective Diabetes Study (UKPDS) Group. Intensive blood-glucose control 32. Nesto RW, Bell D, Bonow RO, et al; American Heart Association; American Diabe- with sulphonylureas or insulin compared with conventional treatment and risk of tes Association. Thiazolidinedione use, fluid retention, and congestive heart fail- complications in patients with type 2 diabetes (UKPDS 33). Lancet. 1998;352 ure: a consensus statement from the American Heart Association and American (9131):837-853. Diabetes Association. October 7, 2003. Circulation. 2003;108(23):2941-2948.

INVITED COMMENTARY Thiazolidinediones and Macular Edema

he thiazolidinediones, rosiglitazone and pioglit- results of a retrospective, population-based study of azone, are peroxisome proliferator–activated re- adult patients with type 2 diabetes in the United King- T ceptor ␥ (PPAR-␥) agonists. They effectively re- dom. The investigators used data from The Health duce glycated hemoglobin among patients with type 2 Improvement Network (THIN), which allowed for long diabetes mellitus by approximately 1 to 1.5 percentage observation of the exposed patients and adequate adjust- points compared with placebo and are used as second- ment for important confounders, including body weight, line treatment agents. However, rosiglitazone and pio- glycated hemoglobin levels, and concomitant medica- glitazone have been associated with peripheral edema, tions. After propensity score adjustment, the authors congestive heart failure, and bone fractures.1,2 An in- reported a significantly increased odds of DME in users creased risk of myocardial ischemia has been attributed of thiazolidinediones compared with nonusers (odds to rosiglitazone.2 An increased risk of bladder cancer as- ratio [OR], 2.3 [95% CI, 1.5-3.6]) at 1 year. Both pioglit- sociated with pioglitazone is noted in the current label. azone and rosiglitazone were associated with a similar Spontaneous reports of macular edema with the use of increased risk. This increased risk persisted at 10 years thiazolidinediones has resulted in regulatory warnings of follow-up. on this potential association. However, the causality re- Importantly, the authors appropriately used mul- mains unclear. tiple imputation methods to handle missing data in this The prevalence of diabetic macular edema (DME) in cohort and used propensity score methods to address con- one or both eyes among patients with diabetes is esti- founding by indication, which is a prominent risk with mated to be 7.1%.3 Approximately one-third of these observational data; however, several limitations pre- affected eyes lose vision. Since PPAR-␥ receptors are clude definitive conclusions. First, the authors did not present in the retinal vasculature, PPAR-␥–mediated have information on the duration of thiazolidinedione fluid retention has been postulated to contribute to exposure or duration of diabetes in the THIN data; that DME. In this issue of the Archives, Idris et al4 report the is, they compared prevalent users of thiazolidinediones—

ARCH INTERN MED/ VOL 172 (NO. 13), JULY 9, 2012 WWW.ARCHINTERNMED.COM 1011

Downloaded From: https://jamanetwork.com/ on 09/24/2021