Abstract
The aim of this study was to investigate the association between thiazolidinediones (TZDs) vs. other antidiabetic drugs and risk of atrial fibrillation (AF) in diabetic patients.
Diabetes mellitus (diabetes) increases the risk of AF by approximately 34%. TZD is an insulin sensitizer that also has anti-inflammatory effects, which might decrease the risk of AF compared with other antidiabetic drugs. We used data from the Danish nationwide registries to study 108 624 patients with diabetes and without prior AF who were treated with metformin or sulfonylurea as first-line drugs. The incidence of AF was significantly lower with TZD as the second-line antidiabetic treatment compared with other second-line antidiabetic drugs (P < 0.001). The 10 year cumulative incidence [95% confidence interval (95% CI)] of AF was 6.2% (3.1–9.3%) with TZD vs. 10.2% (9.8–10.6%) with other antidiabetic drugs. The decreased risk of AF remained significant after adjusting for age, sex, and comorbidities with a hazard ratio (95% CI) of 0.76 (0.57–1.00), P = 0.047 associated with TZD treatment compared with other antidiabetic drugs.
Use of a TZD to treat diabetes was associated with reduced risk of developing AF compared with other antidiabetic drugs as second-line treatment.
Introduction
Patients with type 2 diabetes mellitus (diabetes) have an increased risk of atrial fibrillation (AF) and furthermore diabetes is a strong risk factor for developing a stroke in patients with AF.1–4 The potential mechanisms underlying the predisposition to AF in patients with diabetes are not entirely understood; however, these may include inflammatory interstitial fibrosis and atrial remodelling.5–8 Interventions that alter these processes might reduce the risk of AF.
Thiazolidinediones (TZDs) belong to a class of peroxisome proliferator-activated receptor gamma activators. Binding to this transcription factor alters glucose and lipid homeostasis, increases sensitivity to insulin, and decreases the inflammatory-response.9–11
Despite their benefits in reducing glucose levels in diabetes, TZDs have been associated with heart failure leading FDA to place a black box warning on TZDs. Furthermore, troglitazones have been withdrawn from the market, leaving rosiglitazone and pioglitazone the only TZDs still available in the USA.12 TZDs are not recommended as first-line antidiabetic drug in international guidelines, but remains option as second-line treatment for diabetes.13 TZDs may also have potential benefits for the AF prevention in both development of new-onset AF in patients with diabetes and in post-ablation patients.14–17
We sought to determine whether treatment with TZDs was associated with decreased risk of AF in patients with diabetes by analysing data from the extensive Danish nationwide cohorts.
Method
All residents of Denmark are, at birth or immigration, issued a permanent unique civil registration number that enables individual-level linkage between administrative registries. The Danish National Patient Register holds information on all hospital visits of both inpatient admissions and outpatient visits.9 Each hospitalization discharge is coded with one primary and, if appropriate, one or more secondary diagnosis codes according to the International Classification of Diseases (ICD). Data on pharmacy prescriptions were identified from the Danish Registry of Medicinal Product Statistics, which has record of all drug prescriptions dispensed by Danish pharmacies since 1995. Each drug dispensing is coded according to the Anatomical Therapeutic Chemical system, including the date of dispensing, quantity dispensed, strength, formulation, and affiliation of the physician issuing the prescription. The reimbursement of drug expenses by the Danish health care system requires all pharmacies to register each drug dispensing in the National Prescription Registry.
Study cohort and follow-up
We included all individuals with diabetes who were treated with two different types of antidiabetic drugs between 2000 and 2012. Claimed prescriptions of the specific types of antidiabetic drugs were identified with first-line treatment as either metformin or sulfonylurea, and the second-line treatment as either metformin, sulfonylurea, insulin, TZD, carbamoyl methyl benzoic acid derivative (CBD), dipeptidyl peptidase-4 inhibitor (DPP) or glucagon-like peptide-1 (GLP-1). We excluded all patients below 18 and above 100 years of age and all patients with prior AF. Prior AF was defined as a primary or secondary diagnosis of AF in both in- and outpatients. Patients who received TZDs as second-line antidiabetic drug entered the ‘TZD group’, and patients who received other antidiabetic drugs than TZDs were assigned to the ‘other group’. In the study design information on follow-up was limited to 31 December 2012; hence this was end of study. Patients entered the study on first day they initiated the second-line antidiabetic drug, and they were followed until end of study (31 December 2012), last date with second-line antidiabetic drug, introduction of a third-line antidiabetic drug, emigration from Denmark, death, or AF development, whichever came first.
Comorbidities
Comorbidities were identified using ICD-8 and ICD-10 codes for stroke, heart failure, all cancers, ischaemic heart disease, chronic obstructive pulmonary disease, chronic kidney disease, liver disease, and vascular disease. Claimed drug prescriptions were used to identify hyperthyroidism, hypothyroidism, hypertension, and statin use. Nordic procedure codes were used to identify prior coronary artery bypass graft (CABG) and percutaneous coronary intervention (PCI). Information on death came from National Causes of Death Register (see Supplementary material online, Table S1).
Study outcome
The study outcome was first time diagnosis of AF. This was identified using ICD-10 codes (I48) in the Danish National Patient Register, in either in- or outpatients and either as a primary or secondary diagnosis code.
Statistical methods
Categorical data were presented as counts with percentages, and statistical differences were tested using χ2 test and Fisher’s exact test where appropriate. Continuous variables were presented as means with standard deviations for normal distributed data, and as medians with interquartile range for non-normal distributed data. Statistical differences were tested using Student’s t-test and Wilcoxon rank-sum test where appropriate. Cumulative incidence of AF with 95% confidence interval (95% CI) was calculated using the Aalen–Johansen estimator accounting for death as competing risk. Statistical difference between the curves was tested using Fine and Gray test. Relative risks were presented as hazard ratios with 95% CI calculated in three Cox regression models. Model 1 was univariate, Model 2 was adjusted for age and sex, and Model 3 was adjusted for age, sex and comorbidities. A P-value < 0.05 was considered as statistically significant.
A propensity score was calculated using information on prior stroke, heart failure, all cancers, hyperthyroidism, ischaemic heart disease, chronic obstructive pulmonary disease, chronic kidney disease, liver disease, vascular disease, hypertension, follow-up time, prior CABG, prior PCI, and statin use. Patients in second-line treatment with TZD were matched 1:5 with patients in second-line treatment with other antidiabetic drugs with ‘exact matching’ on sex and age and ‘nearest neighbour matching’ (a greedy match) on the remaining variables. Both Aalen–Johansen cumulative incidence rates and Cox regression analyses were performed on the matched cohort.
Data management and statistical analyses were conducted using R statistics [R Core Team (2015). R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing. URL: http://www.R-project.org/]. Following CRAN packages were used ‘tableone’, ‘riskRegression’, and ‘MatchIt’.
Ethics approval
In Denmark, retrospective register studies do not require approval from the ethics committees. The Danish Data Protection Agency approved this study (ref. no.: 2007-58-0015/GEH-2014-016 I-Suite no.: 02734) and data were available in an anonymized format such that specific individuals could not be identified.
Results
From the 311 631 patients treated with an antidiabetic drug between 2000 and 2012, we excluded 203 007 patients (first antidiabetic drug that was not metformin or sulfonylurea, antidiabetic mono-therapy, below 18 or above 100 years of age, prior AF, or in meglitinides treatment). This left us with a study cohort of 108 624 patients prescribed a first-line antidiabetic drug of either metformin or sulfonylurea, and an additional second-line antidiabetic drug of either metformin, sulfonylurea, insulin, DPP, GLP-1, CBD, or TZD. A minority of 2658 patients entered the ‘TZD group’ and 105 966 entered the ‘other group’. The selection of the study cohort is depicted in Figure1.
The study flowchart. AF, atrial fibrillation; TZD, thiazolidinedione; Met, metformin; SU, sulfonylurea; Insu, insulin; DPP, dipeptidyl peptidase-4 inhibitor; GLP-1, glucagon-like peptide-1; CBD, carbamoyl methyl benzoic acid derivative.
Patients in the ‘TZD group’ were generally younger with a median age of 59.6 years vs. 62.4 years in the ‘other group’ (P < 0.001) with similar proportion of men and women (P = 0.14). The patients in the ‘TZD group’ had less cardiovascular comorbidities at baseline: 2.3% vs. 4.9% (P < 0.001) had heart failure, 9.9% vs. 12.9% (P < 0.001) had ischaemic heart disease, and 4.3% vs. 5.6% (P < 0.005) had vascular disease. The baseline characteristics of the study cohort are presented in Table1.
Baseline characteristics of the ‘TZD group’ and ‘Other group’
| TZD | Other | P-value | |
|---|---|---|---|
| N | 2658 | 105 966 | |
| First line metformin | 2163 (81.4) | 51 914 (49.0) | <0.001 |
| First line sulfonylurea | 495 (18.6) | 54 052 (51.0) | <0.001 |
| Men, n (%) | 1507 (56.7) | 61 610 (58.1) | 0.141 |
| Age, median (IQR) | 59.59 (50.6–67.5) | 62.40 (53.6–71.2) | <0.001 |
| Age categories | <0.001 | ||
| <40 | 202 (7.6) | 5146 (4.9) | |
| 40–64 | 1628 (61.2) | 56 414 (53.2) | |
| 65–74 | 555 (20.9) | 26 671 (25.2) | |
| >74 | 273 (10.3) | 17 735 (16.7) | |
| Stroke, n (%) | 123 (4.6) | 7657 (7.2) | <0.001 |
| Heart failure, n (%) | 62 (2.3) | 5236 (4.9) | <0.001 |
| Cancer, n (%) | 155 (5.8) | 9260 (8.7) | <0.001 |
| Hyperthyroidism, n (%) | 47 (1.8) | 1890 (1.8) | 1.000 |
| Ischaemic heart disease, n (%) | 263 (9.9) | 13 627 (12.9) | <0.001 |
| Chronic obstructive pulmonary disease, n (%) | 98 (3.7) | 6145 (5.8) | <0.001 |
| Chronic kidney disease, n (%) | 20 (0.8) | 1522 (1.4) | 0.002 |
| Liver disease, n (%) | 32 (1.2) | 2583 (2.4) | <0.001 |
| Vascular disease, n (%) | 115 (4.3) | 5955 (5.6) | 0.005 |
| Hypertension, n (%) | 1334 (50.2) | 51 335 (48.4) | 0.079 |
| Antiadrenergic drug, n (%) | 121 (4.6) | 3953 (3.7) | 0.031 |
| Diuretics, n (%) | 1166 (43.9) | 45 193 (42.6) | 0.217 |
| RAS inhibitors, n (%) | 1564 (58.8) | 59 218 (55.9) | 0.003 |
| Loop diuretics, n (%) | 532 (20.0) | 24 059 (22.7) | 0.001 |
| Beta-blockers, n (%) | 836 (31.5) | 33 365 (31.5) | 0.987 |
| Statin, n (%) | 1541 (58.0) | 56 176 (53.0) | <0.001 |
| Coronary artery bypass grafting, n (%) | 45 (1.7) | 2518 (2.4) | 0.026 |
| Percutaneous coronary intervention, n (%) | 80 (3.0) | 4403 (4.2) | 0.004 |
| TZD | Other | P-value | |
|---|---|---|---|
| N | 2658 | 105 966 | |
| First line metformin | 2163 (81.4) | 51 914 (49.0) | <0.001 |
| First line sulfonylurea | 495 (18.6) | 54 052 (51.0) | <0.001 |
| Men, n (%) | 1507 (56.7) | 61 610 (58.1) | 0.141 |
| Age, median (IQR) | 59.59 (50.6–67.5) | 62.40 (53.6–71.2) | <0.001 |
| Age categories | <0.001 | ||
| <40 | 202 (7.6) | 5146 (4.9) | |
| 40–64 | 1628 (61.2) | 56 414 (53.2) | |
| 65–74 | 555 (20.9) | 26 671 (25.2) | |
| >74 | 273 (10.3) | 17 735 (16.7) | |
| Stroke, n (%) | 123 (4.6) | 7657 (7.2) | <0.001 |
| Heart failure, n (%) | 62 (2.3) | 5236 (4.9) | <0.001 |
| Cancer, n (%) | 155 (5.8) | 9260 (8.7) | <0.001 |
| Hyperthyroidism, n (%) | 47 (1.8) | 1890 (1.8) | 1.000 |
| Ischaemic heart disease, n (%) | 263 (9.9) | 13 627 (12.9) | <0.001 |
| Chronic obstructive pulmonary disease, n (%) | 98 (3.7) | 6145 (5.8) | <0.001 |
| Chronic kidney disease, n (%) | 20 (0.8) | 1522 (1.4) | 0.002 |
| Liver disease, n (%) | 32 (1.2) | 2583 (2.4) | <0.001 |
| Vascular disease, n (%) | 115 (4.3) | 5955 (5.6) | 0.005 |
| Hypertension, n (%) | 1334 (50.2) | 51 335 (48.4) | 0.079 |
| Antiadrenergic drug, n (%) | 121 (4.6) | 3953 (3.7) | 0.031 |
| Diuretics, n (%) | 1166 (43.9) | 45 193 (42.6) | 0.217 |
| RAS inhibitors, n (%) | 1564 (58.8) | 59 218 (55.9) | 0.003 |
| Loop diuretics, n (%) | 532 (20.0) | 24 059 (22.7) | 0.001 |
| Beta-blockers, n (%) | 836 (31.5) | 33 365 (31.5) | 0.987 |
| Statin, n (%) | 1541 (58.0) | 56 176 (53.0) | <0.001 |
| Coronary artery bypass grafting, n (%) | 45 (1.7) | 2518 (2.4) | 0.026 |
| Percutaneous coronary intervention, n (%) | 80 (3.0) | 4403 (4.2) | 0.004 |
‘TZD group’ is patients with thiazolidinedione as second-line antidiabetic drug. ‘Other group’ is patients with metformin, sulfonylurea, insulin, DPP, GLP-1, or CBD as second-line antidiabetic drug.
Baseline characteristics of the ‘TZD group’ and ‘Other group’
| TZD | Other | P-value | |
|---|---|---|---|
| N | 2658 | 105 966 | |
| First line metformin | 2163 (81.4) | 51 914 (49.0) | <0.001 |
| First line sulfonylurea | 495 (18.6) | 54 052 (51.0) | <0.001 |
| Men, n (%) | 1507 (56.7) | 61 610 (58.1) | 0.141 |
| Age, median (IQR) | 59.59 (50.6–67.5) | 62.40 (53.6–71.2) | <0.001 |
| Age categories | <0.001 | ||
| <40 | 202 (7.6) | 5146 (4.9) | |
| 40–64 | 1628 (61.2) | 56 414 (53.2) | |
| 65–74 | 555 (20.9) | 26 671 (25.2) | |
| >74 | 273 (10.3) | 17 735 (16.7) | |
| Stroke, n (%) | 123 (4.6) | 7657 (7.2) | <0.001 |
| Heart failure, n (%) | 62 (2.3) | 5236 (4.9) | <0.001 |
| Cancer, n (%) | 155 (5.8) | 9260 (8.7) | <0.001 |
| Hyperthyroidism, n (%) | 47 (1.8) | 1890 (1.8) | 1.000 |
| Ischaemic heart disease, n (%) | 263 (9.9) | 13 627 (12.9) | <0.001 |
| Chronic obstructive pulmonary disease, n (%) | 98 (3.7) | 6145 (5.8) | <0.001 |
| Chronic kidney disease, n (%) | 20 (0.8) | 1522 (1.4) | 0.002 |
| Liver disease, n (%) | 32 (1.2) | 2583 (2.4) | <0.001 |
| Vascular disease, n (%) | 115 (4.3) | 5955 (5.6) | 0.005 |
| Hypertension, n (%) | 1334 (50.2) | 51 335 (48.4) | 0.079 |
| Antiadrenergic drug, n (%) | 121 (4.6) | 3953 (3.7) | 0.031 |
| Diuretics, n (%) | 1166 (43.9) | 45 193 (42.6) | 0.217 |
| RAS inhibitors, n (%) | 1564 (58.8) | 59 218 (55.9) | 0.003 |
| Loop diuretics, n (%) | 532 (20.0) | 24 059 (22.7) | 0.001 |
| Beta-blockers, n (%) | 836 (31.5) | 33 365 (31.5) | 0.987 |
| Statin, n (%) | 1541 (58.0) | 56 176 (53.0) | <0.001 |
| Coronary artery bypass grafting, n (%) | 45 (1.7) | 2518 (2.4) | 0.026 |
| Percutaneous coronary intervention, n (%) | 80 (3.0) | 4403 (4.2) | 0.004 |
| TZD | Other | P-value | |
|---|---|---|---|
| N | 2658 | 105 966 | |
| First line metformin | 2163 (81.4) | 51 914 (49.0) | <0.001 |
| First line sulfonylurea | 495 (18.6) | 54 052 (51.0) | <0.001 |
| Men, n (%) | 1507 (56.7) | 61 610 (58.1) | 0.141 |
| Age, median (IQR) | 59.59 (50.6–67.5) | 62.40 (53.6–71.2) | <0.001 |
| Age categories | <0.001 | ||
| <40 | 202 (7.6) | 5146 (4.9) | |
| 40–64 | 1628 (61.2) | 56 414 (53.2) | |
| 65–74 | 555 (20.9) | 26 671 (25.2) | |
| >74 | 273 (10.3) | 17 735 (16.7) | |
| Stroke, n (%) | 123 (4.6) | 7657 (7.2) | <0.001 |
| Heart failure, n (%) | 62 (2.3) | 5236 (4.9) | <0.001 |
| Cancer, n (%) | 155 (5.8) | 9260 (8.7) | <0.001 |
| Hyperthyroidism, n (%) | 47 (1.8) | 1890 (1.8) | 1.000 |
| Ischaemic heart disease, n (%) | 263 (9.9) | 13 627 (12.9) | <0.001 |
| Chronic obstructive pulmonary disease, n (%) | 98 (3.7) | 6145 (5.8) | <0.001 |
| Chronic kidney disease, n (%) | 20 (0.8) | 1522 (1.4) | 0.002 |
| Liver disease, n (%) | 32 (1.2) | 2583 (2.4) | <0.001 |
| Vascular disease, n (%) | 115 (4.3) | 5955 (5.6) | 0.005 |
| Hypertension, n (%) | 1334 (50.2) | 51 335 (48.4) | 0.079 |
| Antiadrenergic drug, n (%) | 121 (4.6) | 3953 (3.7) | 0.031 |
| Diuretics, n (%) | 1166 (43.9) | 45 193 (42.6) | 0.217 |
| RAS inhibitors, n (%) | 1564 (58.8) | 59 218 (55.9) | 0.003 |
| Loop diuretics, n (%) | 532 (20.0) | 24 059 (22.7) | 0.001 |
| Beta-blockers, n (%) | 836 (31.5) | 33 365 (31.5) | 0.987 |
| Statin, n (%) | 1541 (58.0) | 56 176 (53.0) | <0.001 |
| Coronary artery bypass grafting, n (%) | 45 (1.7) | 2518 (2.4) | 0.026 |
| Percutaneous coronary intervention, n (%) | 80 (3.0) | 4403 (4.2) | 0.004 |
‘TZD group’ is patients with thiazolidinedione as second-line antidiabetic drug. ‘Other group’ is patients with metformin, sulfonylurea, insulin, DPP, GLP-1, or CBD as second-line antidiabetic drug.
The incidence of AF was significantly lower in the ‘TZD group’ (P < 0.001) with a 10 year cumulative incidence of AF of 6.2% (95% CI 3.1–9.3%) compared with 10.2% (95% CI 9.8–10.6%) in the ‘other group’ (Figure2). In the Cox proportional hazard models patients treated with TZD had significantly lower risk of AF with an HR of 0.58 (95% CI 0.44–0.77), P < 0.001 in the univariate model. Adjustment attenuated this association, but it remained significant at 0.72 (95% CI 0.54–0.94), P = 0.018 after adjusting for age and sex and after full adjustment it was still significant at 0.76 (95% CI 0.57–1.00), P = 0.047 (Figure3). The lower risk of AF in patients treated with TZD was present irrespective of the specific second-line antidiabetic drugs with all point estimates being below 1.0 indicating decreased risk of AF with TZD (Figure4). Significant associations were only found with insulin and sulfonylurea.
Cumulative incidence of atrial fibrillation with TZD and other antidiabetic treatment. Aalen–Johansen cumulative incidence of AF in the ‘TZD group’ and the ‘other group’. The model takes into account competing risk of death. P-value from Fine and Gray competing risks regression model.
Cox hazard ratio of atrial fibrillation in the ‘TZD group’ with ‘other group’ as reference. The ‘fully adjusted’ model is adjusted for age, sex, stroke, heart failure, all cancer, hyperthyroidism, ischaemic heart disease, chronic obstructive pulmonary disease, chronic kidney disease, liver disease, vascular disease, hypertension, statin use, prior CABG, and prior PCI.
Cox hazard ratio of risk of atrial fibrillation for the individual antidiabetic drugs with the individual antidiabetic drugs as reference. The model is adjusted for age, sex, stroke, heart failure, all cancer, hyperthyroidism, hypothyroidism, ischaemic heart disease, chronic obstructive pulmonary disease, chronic kidney disease, liver disease, vascular disease, hypertension, statin use, prior CABG, and prior PCI. TZD, thiazolidinedione; met, metformin; SU, sulfonylurea; insu, insulin; DPP, dipeptidyl peptidase-4 inhibitor; GLP-1, glucagon-like peptide-1; CBD, carbamoyl methyl benzoic acid derivative.
In a sensitivity analysis, we identified a subgroup of patients receiving exclusively metformin as first-line antidiabetic drug and either sulfonylurea, insulin, DPP, GLP-1, CBD or TZD as additional second-line additional antidiabetic drug (n = 54077). In this subgroup, the cumulative incidence of AF was also significantly lower (P = 0.004) in the ‘TZD group’ with 4.4% compared with the ‘other group’ with 9.8% after 10 years (Figure5). This decreased risk was also found in the Cox regression analysis after adjusting age, sex and comorbidities with an HR of 0.69 (95% CI 0.50–0.96), P = 0.026 in the ‘TZD group’ with the ‘other group’ as reference.
Aalen–Johansen cumulative incidence of AF in the ‘TZD group’ and the ‘other group’ in patients exclusively treated with metformin as first-line antidiabetic drug. The model takes into account competing risk of death. P-value from Fine and Gray competing risks regression model.
In the interaction analyses, sex, age, hypertension status, statin use, and ischaemic heart disease were investigated and no interactions/effect modification were found (Figure6). In the matched analysis, the propensity scores were matched 99.99%, and age and sex were matched 100% making the mean propensity scores practically identical in the two groups. Subsequently testing between baseline characteristics showed no significant difference between the ‘TZD group’ and the 1:5 matched ‘other group’ (see Supplementary material online, Table S2). The cumulative incidence of AF was significantly lower (P < 0.001) in the ‘TZD group’ with 6.2% compared with the ‘other group’ with 10.4% after 10 years. This decreased risk was also found in the Cox regression analysis with an HR of 0.60 (95% CI 0.45–0.80), P ≤ 0.001 in the ‘TZD group’ with the ‘other group’ as reference.
Interaction analysis for sex, age, hypertension, statin, and ischaemic heart disease. Presented are both events/patients; incidence rates per 100 person years, Cox hazard ratio in the ‘TZD group’ with ‘other group’ as reference. The model is adjusted for age, sex, stroke, heart failure, all cancer, hyperthyroidism, hypothyroidism, ischaemic heart disease, chronic obstructive pulmonary disease, chronic kidney disease, liver disease, vascular disease, hypertension; statin use, prior CABG, and prior PCI. In the last column is the P-value for interaction.
Discussion
This nationwide study found that additional second-line treatment with TZD was associated with 24% lower risk of developing AF compared with other second-line antidiabetic drugs.
This study is the largest study to investigate the association between TZD and the risk of developing AF with TZD as second-line antidiabetic drug. This is an important point because guidelines do not recommend TZD as first-line treatment but as additional second-line antidiabetic drug, and this makes our study clinically more relevant compared with studies investigating TZD as first-line drug. International guidelines on diabetes patients recommend metformin as the preferred first-line antidiabetic drug,13 and this was investigated in a subgroup analysis with patients exclusively in metformin treatment as first-line drug. In this subgroup analysis, TZD was associated with a decreased risk of AF of 31% compared with other antidiabetic drugs. This association with TZD and decreased risk of AF is even stronger than found in the main analysis with a 24% decreased risk.
Our study is also the first to investigate TZD vs. individual antidiabetic drugs one by one. In these analyses all HRs were <1, indicating a decreased risk of AF with TZD compared with all of the other individual antidiabetic drugs. Notably, significant results were only found with sulfonylurea and insulin as reference. The patients treated with TZD in our study suffered from less heart failure, prior stroke, ischaemic heart disease, and vascular disease than patients in both the sulfonylurea and insulin subgroup. These comorbidities are all strong predictors of AF, and although our statistical models were fully adjusted for these risk factors, it can still be argued that the decreased risk of AF could to some extend be attributed to the younger and healthier TZD group. As TZD has a black box warning from FDA, because TZDs were found associated with worsening in heart failure among diabetes patients, clinicians might be more reluctant with prescribing TZD in patients with heart failure.
The evidence regarding TZD and risk of developing AF has only been investigated in a few other studies. A recent nationwide study by Chao et al.,16 who investigated TZD treatment associated with risk of AF in 12 065 diabetic patients, found a decreased risk of developing AF with TZD compared with no-TZD with an HR of 0.69 (95% CI 0.49–0.91), P = 0.028. The direction and magnitude of the HR found compliments in our study of 0.76 (95% 0.57–1.00), P = 0.047. The PROactive study (PROspective pioglitAzone Clinical Trial In macroVascular Events) was a randomized controlled trial investigating the risk of cardiovascular events in patients with diabetes treated with pioglitazone vs. placebo.18 The study showed a non-significant (P = 0.374) decreased risk of developing AF between the two groups. The reason for this lack of significance could be low incidence of AF (2%) or that AF was not a predefined endpoint in the study.
Thiazolidinedione and risk of developing AF have also been investigated in patients with high risk of AF after cardiothoracic surgery and after ablation.14,15 In the study investigating patients after cardiothoracic surgery, 40 diabetic patients in treated with TZD patients vs. 144 diabetic patients not treated with TZD were enrolled, and a non-significant decreased risk of developing AF was found with an adjusted odds ratio of 0.80 (95% CI 0.32–1.99), P = 0.63. Although the odds ratio was non-significant, it still suggests that TZD could have a protective effect towards AF, and the direction and magnitude of the odds ratio is similar to the HR found in our study. In the ablation study, the risk of recurrent atrial tachycardia was investigated in diabetic patients randomized to pioglitazone vs. not in treatment with pioglitazone. With an odds ratio of 0.32 (95% CI 0.12–0.86), P = 0.024 pioglitazone was found to both decrease the risk of atrial tachycardia recurrence and as a significant predictor of absence of atrial tachycardia after ablation.
Our study supports the existing evidence of a decreased risk of AF in diabetes patients treated with TZD, and it contributes with the largest cohort study investigating this to date. Furthermore, our study offers novel information on the clinically relevant question regarding the risk of AF with TZD as additional antidiabetic drug to metformin. This is a very important notion as this is the most clinically relevant use of TZD. Our finding implies increased use of TZD, and prescription with TZD as additional second-line antidiabetic drug to metformin is according with current guidelines. Notably, our study is registry based and therefore all findings are associations. Knowledge regarding efficacy and safety of TZD treatment and AF risk is needed, and future randomized controlled trails could be considered in order to investigate this.
Limitations
The main limitation of this study is inherent in our observational design of the study and lack of clinical information and randomization to treatment groups. Albeit the models are adjusted for known confounders, there is still a challenge with residual confounding in the observational design that can affect the results. To compensate this, we performed propensity score-matched analyses. In the matched analyses, we found an even greater association between TZD and decreased risk of AF with an HR of 0.60 (95% CI 0.45–0.80), P < 0.001 than we found in our main analyses with an HR of 0.76 (95% CI 0.57–1.00), P = 0.047.
The follow-up period was 12 years, and it was a limitation that patients included near the study end (31 December 2012) had decreased risk-time to develop AF, which could have underestimated the cumulative incidences of developing AF depicted in Figures2and5. The method of identifying AF with ICD codes does not identify patients with unrecognized AF, or patients attending their general practitioner; hence there is a risk of misclassification. On the other hand, the risk of including false-positive patients is very limited, as the method of identifying AF in our study is validated with a positive predictive value of 92.6%.19 Using anti-arrhythmic drugs as proxies for AF is a method that has not yet been validated in the Danish registers. Although this method could decrease the proportion of false-negative AF events in our study, the tradeoff could be introduction of false-positive AF events.
One of the strengths of our study is that the National Prescription Register is linked to the partial reimbursement policy for drug expenses by the national health security system and has been shown to be accurate and reduce the risk of surveillance bias.20,21 In Denmark healthcare is tax financed and is available for every citizen without charge, and by including the entire Danish population, we avoided selection bias. The utility of TZDs in treatment of patients with diabetes has declined in the past decade largely due to adverse effects of fluid retention. Although there has been a decline, TZD may still be the right choice of antidiabetic drug for some patients where good glucose control is hard to achieve with other antidiabetic drugs.
In conclusion, treatment with TZD as second-line antidiabetic drug in patients with diabetes is associated with decreased risk of AF compared with treatment with other antidiabetic drugs.
Supplementary material
Supplementary material is available at European Heart Journal – Cardiovascular Pharmacotherapy online.
Funding
J.L.P. and L.S. have received research funding from Boehringer-Ingelheim. G.H.G. is supported by an unrestricted clinical research scholarship from the Novo Nordisk Foundation. He has ownership of stocks in Novo Nordisk Pharmaceuticals, which produce glucose-lowering drugs.
Conflict of interest: none declared.
