https://orcid.org/0000-0003-3612-653X
Abstract
Evidence regarding the risks of serious hypoglycaemia for patients with atrial fibrillation (AF) and diabetes mellitus (DM) taking antidiabetic medications with concurrent non-vitamin K antagonist oral anticoagulants (NOACs) vs. warfarin is limited. This study aimed to investigate this knowledge gap.
This retrospective cohort study used nationwide data from Taiwan's National Health Insurance Research Database and included a total of 56 774 adult patients treated with antidiabetic medications and oral anticoagulants between 1 January 2012 and 31 December 2020. The incidence rate ratios (IRRs) of serious hypoglycaemia were estimated for patients taking antidiabetic drugs with NOACs vs. warfarin. Poisson regression models with generalized estimating equations accounting for intra-individual correlation across follow-up periods were used. Stabilized inverse probability of treatment weighting was used to create treatment groups with balanced characteristics for comparisons. Compared to concurrent use of antidiabetic drugs with warfarin, those with NOACs showed a significantly lower risk of serious hypoglycaemia (IRR = 0.73, 95% CI: 0.63–0.85, P < 0.001). In the analyses of each NOAC, patients taking dabigatran (IRR = 0.76, 95% CI: 0.63–0.91, P = 0.002), rivaroxaban (IRR = 0.72, 95% CI: 0.61–0.86, P < 0.001), and apixaban (IRR = 0.71, 95% CI: 0.57–0.89, P = 0.003) showed a significantly lower risk of serious hypoglycaemia than those taking warfarin.
In patients with AF and DM taking antidiabetic drugs, concurrent use of NOACs was associated with a lower risk of serious hypoglycaemia than concurrent use of warfarin.
Introduction
Both atrial fibrillation (AF) and diabetes mellitus (DM) are major global health issues in the world's aging populations.1–3 These two conditions have a high prevalence and incidence in older adults, and DM is also a major risk factor for AF. Many patients are diagnosed with both conditions. In most diabetes patients with atrial fibrillation, long-term oral anticoagulant treatment is required for stroke prevention. Warfarin, a vitamin K antagonist, is a traditional and cornerstone oral anticoagulant used for decades. Non-vitamin K antagonist oral anticoagulants (NOACs) have been approved as alternatives to warfarin in recent years.4–6
Hypoglycaemia is a dangerous side-effect of diabetes treatment. Hypoglycaemia is linked to an increased risk of cardiovascular events, central nervous system damage, and mortality in patients with DM.7,8 It could also affect cardiac autonomic tone and cause abnormal repolarization. This can further lead to cardiac arrhythmias and sudden cardiac death.9,10 Recent studies have also shown that in diabetes patients with AF, severe hypoglycaemia leads to a worsened prognosis and premature death; thus, it is crucial to find strategies to minimize the risk of hypoglycaemia for this vulnerable population.11
Recent evidence suggests that vitamin K has a beneficial effect on insulin sensitivity and glycaemic control through several mechanisms.12–14 Warfarin, but not NOACs, may interfere with glycaemic control by antagonizing vitamin K function, increasing the risk of DM and its complications, such as hypoglycaemia.15–17 Moreover, previous studies have reported that concurrent use of warfarin compared with non-use may increase the risk of serious hypoglycaemic events in patients treated with specific antidiabetic drugs (e.g. sulfonylureas) owing to the potential drug—drug interactions.18,19 Since DM and AF commonly co-exist in aging populations, many patients need to take antidiabetic drugs and anticoagulants concurrently. Therefore, choosing an optimal oral anticoagulant that reduces the risk of serious hypoglycaemia in patients with diabetes treatment becomes a crucial clinical issue.
However, there is limited evidence comparing the effects of NOACs vs. warfarin on serious hypoglycaemia risk in patients taking antidiabetic drugs. This study used a nationwide cohort to address the knowledge gap on whether patients taking various antidiabetic medications and concurrent NOACs vs. warfarin would have different risks of serious hypoglycaemia.
Methods
Data sources
This nationwide retrospective cohort study was conducted using data from Taiwan's National Health Insurance Research Database (NHIRD). The NHIRD comprises healthcare data from approximately 23.6 million individuals, representing the entire population of Taiwan. A brief description of NHIRD is provided in Supplementary Note 1, and the details of NHIRD have been reported elsewhere.20,21 The Research Ethics Committee of Hualien Tzu Chi Hospital approved this study (REC No: IRB110-170-C); the requirement for informed consent was waived.
Study population, exposure, and follow-up time
The data were initially retrieved from all adult patients aged ≥20 years who had been diagnosed with both DM (ICD-9-CM code 250 or ICD-10-CM codes E08—E13) and AF (ICD-9-CM code 427.31 or ICD-10-CM codes I48.0, I48.1, I48.2, and I48.91) at least once in an inpatient service or twice in outpatient services between 2012 and 2020 in NHIRD. Our study population was limited to patients treated with antidiabetic medications concurrently with any oral anticoagulant.
In clinical practice, the use of diabetes medications and anticoagulants often changes over time; it is unreasonable to limit patients to being exposed to the same drug use status during several years of the follow-up. Therefore, we divided each patient's follow-up time into different periods (unit of time: day) according to the combination of specific antidiabetic medication and oral anticoagulant types used concurrently. Each period of concurrent use of the same combination of oral anticoagulants and antidiabetic drugs in each patient was used as the unit of analysis. In other words, each patient's follow-up time was divided into several separate periods for analysis based on each specific combination of the oral anticoagulant and antidiabetic drug; if the oral anticoagulant or antidiabetic drug changed, then the periods before and after the change were considered different samples for analysis. Each period of concurrent use of oral anticoagulants and antidiabetic drugs was then classified into the warfarin or NOAC group based on the type of oral anticoagulant prescribed. This design helped decrease the likelihood of exposure misclassification for patients who had changed their antidiabetic medications and oral anticoagulants during long-term follow-up.
We excluded follow-up periods during which patients did not possess any antidiabetic drugs or oral anticoagulants, as determined by prescription data, from our analysis. Other exclusion criteria included those without any diagnosis of AF before the follow-up periods, as well as those with end-stage renal disease (ESRD), rheumatic heart disease, congenital heart disease, or having valve replacement surgery to avoid potential confounding by indications.22,23
Study outcome
The outcome event was serious hypoglycaemia, ascertained when a patient was admitted to the hospital or treated in the emergency department with a diagnosis of hypoglycaemia (ICD-9-CM codes: 251.0–251.2 or ICD-10-CM codes: E16.0—E16.2, E08.64, E10.64, E11.64, E13.64) accompanied with prescriptions of intravenous high concentration glucose solution (e.g. 50% glucose solution). The incidence rates of serious hypoglycaemia were compared between the periods with NOAC use and those with warfarin use.
Covariates and confounders
Patient demographics, comorbidities, medications, and healthcare utilization were retrieved as covariates for each period. Medical conditions diagnosed at least once in an inpatient or twice in outpatient services within the year before each period were considered pre-existing baseline comorbidities. The Charlson Comorbidity Index and CHA2DS2-VASc score were calculated. Medical utilization (numbers of outpatient visits, emergency room visits, and hospitalizations) in the previous year of each period were retrieved as covariates. Baseline medications were defined as a drug prescribed for at least 30 days within the year prior to the corresponding period. Current diabetes medications indicated the antidiabetic drugs used during the corresponding period; and the number of antidiabetic drugs used was also calculated. Whether there was a history of serious hypoglycaemia within the year before each period was also treated as a covariate. The index year, DM and AF diagnosis duration, monthly income, and hospital level, where oral anticoagulants were prescribed were also retrieved.
Subgroup and stratified analyses
In addition to the main analysis comparing the use of NOAC vs. warfarin for overall patients taking any antidiabetic drugs, also they were compared in the subgroups according to each type of antidiabetic drug, including metformin, sulfonylurea, meglitinide, α-glucosidase inhibitors (AGI), thiazolidinediones (TZD), dipeptidyl peptidase-4 inhibitors (DPP-4i), sodium-glucose cotransporter-2 inhibitors (SGLT-2i), glucagon-like peptide-1 receptor agonist (GLP-1 RA), and insulin. Periods with concurrent prescriptions of a specific antidiabetic drug were included in the subgroup analysis for that antidiabetic drug. We classified the NOACs into four subgroups (dabigatran, rivaroxaban, apixaban, and edoxaban) and compared each NOAC with warfarin regarding their risks of serious hypoglycaemia. Stratified analyses were performed according to age (<65 and ≥65 years), sex, and numbers of antidiabetic drugs. Interaction tests were conducted to determine whether differences in the risk of hypoglycaemia between NOACs and warfarin were moderated by age, sex, and the number of antidiabetic drugs.
Inverse probability of treatment weighting using propensity score
Propensity scores were calculated to estimate the probability of prescribing NOACs for each period using a multivariable logistic regression model, including all covariates in Table 1. We then applied stabilized inverse probability of treatment weighting (IPTW) with truncating the extreme weights to construct a weighted dataset with balanced characteristics between treatment groups. Any weight >5 was considered large, and we reduced any weight >5 down to this threshold.24 The stabilization and truncation of weights helped avoid unstable or extreme weights for subjects with a very low probability of receiving the treatment administered.24–26 Inverse probability of treatment weighting was used to generate the comparison set (weighted dataset) before performing each analysis, including the overall and subgroup/stratified/sensitivity analyses.
Characteristics and comorbidities at baseline among patients with atrial fibrillation and diabetes taking oral anticoagulants and antidiabetic drugs
| Overall population (N = 56 774) | ||
|---|---|---|
| N | % | |
| Age (years)† | 73.5 ± 10.2 | |
| Sex | ||
| Male | 31 482 | 55.5 |
| Female | 25 292 | 44.6 |
| Income level (NTD) | ||
| Financially dependent | 15 912 | 28.0 |
| 15 840–29 999 | 26 222 | 46.2 |
| 30 000–44 999 | 7519 | 13.2 |
| ≥45 000 | 7121 | 12.5 |
| Charlson comorbidity index†,‡ | 3.1 ± 2.1 | |
| CHA2DS2-VASc score†,§ | 4.2 ± 1.7 | |
| Comorbidities | ||
| Hypertension | 43 546 | 76.7 |
| Coronary artery disease | 19 490 | 34.3 |
| Congestive heart failure | 15 381 | 27.1 |
| COPD | 7738 | 13.6 |
| Asthma | 4025 | 7.1 |
| Chronic kidney disease | 8438 | 14.9 |
| Cirrhosis | 2123 | 3.7 |
| Hyperlipidemia | 22 329 | 39.3 |
| Stroke | 17 381 | 30.6 |
| Rheumatoid arthritis | 441 | 0.8 |
| Dementia | 3755 | 6.6 |
| Depression | 2036 | 3.6 |
| Malignancy | 4749 | 8.4 |
| Baseline medication use | ||
| Statins | 25 558 | 45.0 |
| ACEI/ARB | 38 277 | 67.4 |
| β-blocker | 29 583 | 52.1 |
| CCB | 25 738 | 45.3 |
| Diuretics | 20 159 | 35.5 |
| NSAID | 17 024 | 30.0 |
| Corticosteroids | 3694 | 6.5 |
| Antipsychotics | 2999 | 5.3 |
| PPI | 5184 | 9.1 |
| Current diabetes medications | ||
| Metformin | 39 033 | 68.8 |
| Sulfonylurea | 21 210 | 37.4 |
| Meglitinide | 4047 | 7.1 |
| AGI | 5300 | 9.3 |
| TZD | 2972 | 5.2 |
| DPP-4i | 20 713 | 36.5 |
| SGLT-2i | 2538 | 4.5 |
| GLP-1 RA | 183 | 0.3 |
| Insulin | 7720 | 13.6 |
| No. of current diabetes medication types | ||
| 1 | 26 407 | 46.5 |
| 2 | 17 359 | 30.6 |
| ≥3 | 13 008 | 22.9 |
| Duration of diabetes | ||
| <2 years | 12 427 | 21.9 |
| ≥2 years | 44 347 | 78.1 |
| Duration of AF | ||
| <2 years | 37 691 | 66.4 |
| ≥2 years | 19 083 | 33.6 |
| History of serious hypoglycaemia | 819 | 1.4 |
| Index year | ||
| 2012–2014 | 15 727 | 27.7 |
| 2015–2017 | 19 547 | 34.4 |
| 2018–2020 | 21 500 | 37.9 |
| Medical utilization in the previous year | ||
| No. of outpatient visits† | 29.8 ± 19.0 | |
| No. of ER visits† | 1.3 ± 2.0 | |
| No. of hospitalizations† | 1.0 ± 1.4 | |
| Overall population (N = 56 774) | ||
|---|---|---|
| N | % | |
| Age (years)† | 73.5 ± 10.2 | |
| Sex | ||
| Male | 31 482 | 55.5 |
| Female | 25 292 | 44.6 |
| Income level (NTD) | ||
| Financially dependent | 15 912 | 28.0 |
| 15 840–29 999 | 26 222 | 46.2 |
| 30 000–44 999 | 7519 | 13.2 |
| ≥45 000 | 7121 | 12.5 |
| Charlson comorbidity index†,‡ | 3.1 ± 2.1 | |
| CHA2DS2-VASc score†,§ | 4.2 ± 1.7 | |
| Comorbidities | ||
| Hypertension | 43 546 | 76.7 |
| Coronary artery disease | 19 490 | 34.3 |
| Congestive heart failure | 15 381 | 27.1 |
| COPD | 7738 | 13.6 |
| Asthma | 4025 | 7.1 |
| Chronic kidney disease | 8438 | 14.9 |
| Cirrhosis | 2123 | 3.7 |
| Hyperlipidemia | 22 329 | 39.3 |
| Stroke | 17 381 | 30.6 |
| Rheumatoid arthritis | 441 | 0.8 |
| Dementia | 3755 | 6.6 |
| Depression | 2036 | 3.6 |
| Malignancy | 4749 | 8.4 |
| Baseline medication use | ||
| Statins | 25 558 | 45.0 |
| ACEI/ARB | 38 277 | 67.4 |
| β-blocker | 29 583 | 52.1 |
| CCB | 25 738 | 45.3 |
| Diuretics | 20 159 | 35.5 |
| NSAID | 17 024 | 30.0 |
| Corticosteroids | 3694 | 6.5 |
| Antipsychotics | 2999 | 5.3 |
| PPI | 5184 | 9.1 |
| Current diabetes medications | ||
| Metformin | 39 033 | 68.8 |
| Sulfonylurea | 21 210 | 37.4 |
| Meglitinide | 4047 | 7.1 |
| AGI | 5300 | 9.3 |
| TZD | 2972 | 5.2 |
| DPP-4i | 20 713 | 36.5 |
| SGLT-2i | 2538 | 4.5 |
| GLP-1 RA | 183 | 0.3 |
| Insulin | 7720 | 13.6 |
| No. of current diabetes medication types | ||
| 1 | 26 407 | 46.5 |
| 2 | 17 359 | 30.6 |
| ≥3 | 13 008 | 22.9 |
| Duration of diabetes | ||
| <2 years | 12 427 | 21.9 |
| ≥2 years | 44 347 | 78.1 |
| Duration of AF | ||
| <2 years | 37 691 | 66.4 |
| ≥2 years | 19 083 | 33.6 |
| History of serious hypoglycaemia | 819 | 1.4 |
| Index year | ||
| 2012–2014 | 15 727 | 27.7 |
| 2015–2017 | 19 547 | 34.4 |
| 2018–2020 | 21 500 | 37.9 |
| Medical utilization in the previous year | ||
| No. of outpatient visits† | 29.8 ± 19.0 | |
| No. of ER visits† | 1.3 ± 2.0 | |
| No. of hospitalizations† | 1.0 ± 1.4 | |
Data are presented as percentages unless otherwise noted.
Presented as mean ± standard deviation.
Calculated without including scores for age.
Congestive heart failure, hypertension, age ≥75 years, diabetes mellitus, stroke or transient ischemic attack, vascular disease, age 65–74 years, sex category (CHA2DS2-VASc) score.
Abbreviations: ACEI, angiotensin-converting enzyme inhibitors; AGI, alpha-glucosidase inhibitors; ARB, angiotensin II receptor blockers; AF, atrial fibrillation; CCB, calcium channel blockers; COPD, chronic obstructive pulmonary disease; DPP-4i, dipeptidyl peptidase-4 inhibitors; ER, emergency room; GLP-1 RA, glucagon-like peptide-1 receptor agonist; IPTW, inverse probability of treatment weighting; NOAC, non-vitamin K antagonist oral anticoagulant; NSAID, nonsteroidal anti-inflammatory drugs; NTD, New Taiwan Dollar; OAC, oral anticoagulant; PPI, proton pump inhibitors; SGLT-2i, sodium-glucose cotransporter-2 inhibitors; SMD, standardized mean difference; and TZD, thiazolidinedione.
Characteristics and comorbidities at baseline among patients with atrial fibrillation and diabetes taking oral anticoagulants and antidiabetic drugs
| Overall population (N = 56 774) | ||
|---|---|---|
| N | % | |
| Age (years)† | 73.5 ± 10.2 | |
| Sex | ||
| Male | 31 482 | 55.5 |
| Female | 25 292 | 44.6 |
| Income level (NTD) | ||
| Financially dependent | 15 912 | 28.0 |
| 15 840–29 999 | 26 222 | 46.2 |
| 30 000–44 999 | 7519 | 13.2 |
| ≥45 000 | 7121 | 12.5 |
| Charlson comorbidity index†,‡ | 3.1 ± 2.1 | |
| CHA2DS2-VASc score†,§ | 4.2 ± 1.7 | |
| Comorbidities | ||
| Hypertension | 43 546 | 76.7 |
| Coronary artery disease | 19 490 | 34.3 |
| Congestive heart failure | 15 381 | 27.1 |
| COPD | 7738 | 13.6 |
| Asthma | 4025 | 7.1 |
| Chronic kidney disease | 8438 | 14.9 |
| Cirrhosis | 2123 | 3.7 |
| Hyperlipidemia | 22 329 | 39.3 |
| Stroke | 17 381 | 30.6 |
| Rheumatoid arthritis | 441 | 0.8 |
| Dementia | 3755 | 6.6 |
| Depression | 2036 | 3.6 |
| Malignancy | 4749 | 8.4 |
| Baseline medication use | ||
| Statins | 25 558 | 45.0 |
| ACEI/ARB | 38 277 | 67.4 |
| β-blocker | 29 583 | 52.1 |
| CCB | 25 738 | 45.3 |
| Diuretics | 20 159 | 35.5 |
| NSAID | 17 024 | 30.0 |
| Corticosteroids | 3694 | 6.5 |
| Antipsychotics | 2999 | 5.3 |
| PPI | 5184 | 9.1 |
| Current diabetes medications | ||
| Metformin | 39 033 | 68.8 |
| Sulfonylurea | 21 210 | 37.4 |
| Meglitinide | 4047 | 7.1 |
| AGI | 5300 | 9.3 |
| TZD | 2972 | 5.2 |
| DPP-4i | 20 713 | 36.5 |
| SGLT-2i | 2538 | 4.5 |
| GLP-1 RA | 183 | 0.3 |
| Insulin | 7720 | 13.6 |
| No. of current diabetes medication types | ||
| 1 | 26 407 | 46.5 |
| 2 | 17 359 | 30.6 |
| ≥3 | 13 008 | 22.9 |
| Duration of diabetes | ||
| <2 years | 12 427 | 21.9 |
| ≥2 years | 44 347 | 78.1 |
| Duration of AF | ||
| <2 years | 37 691 | 66.4 |
| ≥2 years | 19 083 | 33.6 |
| History of serious hypoglycaemia | 819 | 1.4 |
| Index year | ||
| 2012–2014 | 15 727 | 27.7 |
| 2015–2017 | 19 547 | 34.4 |
| 2018–2020 | 21 500 | 37.9 |
| Medical utilization in the previous year | ||
| No. of outpatient visits† | 29.8 ± 19.0 | |
| No. of ER visits† | 1.3 ± 2.0 | |
| No. of hospitalizations† | 1.0 ± 1.4 | |
| Overall population (N = 56 774) | ||
|---|---|---|
| N | % | |
| Age (years)† | 73.5 ± 10.2 | |
| Sex | ||
| Male | 31 482 | 55.5 |
| Female | 25 292 | 44.6 |
| Income level (NTD) | ||
| Financially dependent | 15 912 | 28.0 |
| 15 840–29 999 | 26 222 | 46.2 |
| 30 000–44 999 | 7519 | 13.2 |
| ≥45 000 | 7121 | 12.5 |
| Charlson comorbidity index†,‡ | 3.1 ± 2.1 | |
| CHA2DS2-VASc score†,§ | 4.2 ± 1.7 | |
| Comorbidities | ||
| Hypertension | 43 546 | 76.7 |
| Coronary artery disease | 19 490 | 34.3 |
| Congestive heart failure | 15 381 | 27.1 |
| COPD | 7738 | 13.6 |
| Asthma | 4025 | 7.1 |
| Chronic kidney disease | 8438 | 14.9 |
| Cirrhosis | 2123 | 3.7 |
| Hyperlipidemia | 22 329 | 39.3 |
| Stroke | 17 381 | 30.6 |
| Rheumatoid arthritis | 441 | 0.8 |
| Dementia | 3755 | 6.6 |
| Depression | 2036 | 3.6 |
| Malignancy | 4749 | 8.4 |
| Baseline medication use | ||
| Statins | 25 558 | 45.0 |
| ACEI/ARB | 38 277 | 67.4 |
| β-blocker | 29 583 | 52.1 |
| CCB | 25 738 | 45.3 |
| Diuretics | 20 159 | 35.5 |
| NSAID | 17 024 | 30.0 |
| Corticosteroids | 3694 | 6.5 |
| Antipsychotics | 2999 | 5.3 |
| PPI | 5184 | 9.1 |
| Current diabetes medications | ||
| Metformin | 39 033 | 68.8 |
| Sulfonylurea | 21 210 | 37.4 |
| Meglitinide | 4047 | 7.1 |
| AGI | 5300 | 9.3 |
| TZD | 2972 | 5.2 |
| DPP-4i | 20 713 | 36.5 |
| SGLT-2i | 2538 | 4.5 |
| GLP-1 RA | 183 | 0.3 |
| Insulin | 7720 | 13.6 |
| No. of current diabetes medication types | ||
| 1 | 26 407 | 46.5 |
| 2 | 17 359 | 30.6 |
| ≥3 | 13 008 | 22.9 |
| Duration of diabetes | ||
| <2 years | 12 427 | 21.9 |
| ≥2 years | 44 347 | 78.1 |
| Duration of AF | ||
| <2 years | 37 691 | 66.4 |
| ≥2 years | 19 083 | 33.6 |
| History of serious hypoglycaemia | 819 | 1.4 |
| Index year | ||
| 2012–2014 | 15 727 | 27.7 |
| 2015–2017 | 19 547 | 34.4 |
| 2018–2020 | 21 500 | 37.9 |
| Medical utilization in the previous year | ||
| No. of outpatient visits† | 29.8 ± 19.0 | |
| No. of ER visits† | 1.3 ± 2.0 | |
| No. of hospitalizations† | 1.0 ± 1.4 | |
Data are presented as percentages unless otherwise noted.
Presented as mean ± standard deviation.
Calculated without including scores for age.
Congestive heart failure, hypertension, age ≥75 years, diabetes mellitus, stroke or transient ischemic attack, vascular disease, age 65–74 years, sex category (CHA2DS2-VASc) score.
Abbreviations: ACEI, angiotensin-converting enzyme inhibitors; AGI, alpha-glucosidase inhibitors; ARB, angiotensin II receptor blockers; AF, atrial fibrillation; CCB, calcium channel blockers; COPD, chronic obstructive pulmonary disease; DPP-4i, dipeptidyl peptidase-4 inhibitors; ER, emergency room; GLP-1 RA, glucagon-like peptide-1 receptor agonist; IPTW, inverse probability of treatment weighting; NOAC, non-vitamin K antagonist oral anticoagulant; NSAID, nonsteroidal anti-inflammatory drugs; NTD, New Taiwan Dollar; OAC, oral anticoagulant; PPI, proton pump inhibitors; SGLT-2i, sodium-glucose cotransporter-2 inhibitors; SMD, standardized mean difference; and TZD, thiazolidinedione.
Statistical analysis
The difference in baseline characteristics between the treatment groups was evaluated using standardized mean difference, with a value of <0.1 indicating a negligible difference. Because the outcome event, serious hypoglycaemia, can occur more than once during each of the follow-up periods within a patient, we used Poisson regression models to estimate the incidence rate ratios (IRRs) of serious hypoglycaemia for patients taking antidiabetics with NOACs vs. warfarin.27 The regression models were weighted by IPTW to control potential confounders; generalized estimating equations were applied to account for intra-individual correlation, because the follow-up periods from the same patient were not entirely independent.28 The risk of serious hypoglycaemia during periods prescribed with warfarin was used as the reference category. The details and statistical codes for how we performed Poisson regression models weighted by IPTW and applied generalized estimating equations accounting for intra-individual correlation are described in Supplementary Materials (Statistical Code section). A two-tailed P value < 0.05 was considered statistically significant. Data management and statistical analyses were performed using R software version 4.1.3 (R Foundation for Statistical Computing, Vienna, Austria) and SAS software version 9.4 (SAS Institute, Inc., Cary, NC, USA).
Sensitivity analyses
We did a sensitivity analysis by excluding patients with any diagnosis of chronic kidney disease (CKD) since the renal function may influence the choice of oral anticoagulants, but we could not obtain the renal function data in the claims-based dataset. In addition, three negative control outcomes (occurrence of influenza, urinary tract infection, and suicide death) were evaluated, for which no causal effect was expected, to indirectly investigate whether potential unmeasured/unknown confounders existed.29
Results
Patient characteristics
A total of 56 774 patients with AF and DM taking antidiabetic drugs with concurrent use of NOACs or warfarin were included in the study population. Their characteristics at baseline (according to their first observed follow-up periods) are shown in Table 1. The mean age was 73.5 years, and females accounted for 44.6% of the study population. We identified 800 691 and 286 384 periods attributable to concurrent use of antidiabetic agents with NOACs and warfarin, respectively. After IPTW, all characteristics were appropriately balanced between the NOAC and warfarin groups, with a standardized mean difference of <0.1 for all variables (see Supplementary material online, Table S1). The flow diagram of patient selection is shown in Supplementary Figure S1.
Risk of serious hypoglycaemia in those with concurrent use of NOACs vs. warfarin
Overall, concurrent use of NOACs with antidiabetic drugs had a significantly lower risk of serious hypoglycaemia than that of warfarin with antidiabetic drugs (IRR = 0.73, 95% CI: 0.63–0.85, P < 0.001) (Figure 1). The detailed results (e.g. the follow-up time, event numbers, and incidence rates) are provided in Supplementary Table S2.
Risk of serious hypoglycaemia in patients taking antidiabetic drugs with concurrent use of NOACs vs. warfarin. Abbreviations: CI, confidence interval; IRR, incidence rate ratio; and NOAC, non-vitamin K antagonist oral anticoagulant.
Stratified analyses revealed that, in both male and female subgroups, concurrent use of NOACs was associated with a significantly lower risk of serious hypoglycaemia compared to warfarin use. The significantly lower risk of serious hypoglycaemia linked to concurrent NOAC use was observed only in older patients aged ≥65 years and those taking two or more antidiabetic drugs (see Figure 1 and Supplementary material online, Table S2). However, none of the interaction tests reached statistical significance (P for interaction with sex = 0.661; P for interaction with age = 0.977; and P for interaction with antidiabetic drug numbers = 0.219).
In the subgroup analyses of each individual NOAC, a significantly lower risk of serious hypoglycaemia was found in dabigatran (IRR = 0.76, 95% CI: 0.63–0.91, P = 0.002), rivaroxaban (IRR = 0.72, 95% CI: 0.61–0.86, P < 0.001), and apixaban (IRR = 0.71, 95% CI: 0.57–0.89, P = 0.003) users. Although the analysis for edoxaban did not reach statistical significance, a similar trend of lower hypoglycaemia risk in edoxaban users was observed (IRR = 0.77, 95% CI: 0.57–1.03, P = 0.078) (see Figure 2 and Supplementary material online, Table S3).
Risk of serious hypoglycaemia in patients taking antidiabetic drugs with concurrent use of each NOAC vs. warfarin. Abbreviations: CI, confidence interval; IRR, incidence rate ratio; and NOAC, non-vitamin K antagonist oral anticoagulant.
Risk of serious hypoglycaemia according to each antidiabetic drug with concurrent use of NOACs vs. warfarin
The significantly lower risk of serious hypoglycaemia associated with concurrent NOAC use was found in patients taking metformin (IRR = 0.73, 95% CI: 0.59–0.89, P = 0.002), DPP-4i (IRR = 0.73, 95% CI: 0.59–0.92, P = 0.007), or sulfonylurea (IRR = 0.68, 95% CI: 0.56–0.81, P < 0.001). The difference in serious hypoglycaemia risks between NOAC and warfarin was not statistically significant in patients taking other antidiabetic drugs (see Figure 3 and Supplementary material online, Table S4).
Risk of serious hypoglycaemia in patients taking various antidiabetic drugs with concurrent use of NOACs vs. warfarin. *Not applicable because of the limited sample size. Abbreviations: AGI, alpha-glucosidase inhibitors; CI, confidence interval; DPP-4i, dipeptidyl peptidase-4 inhibitors; GLP-1 RA, glucagon-like peptide-1 receptor agonist; IRR, incidence rate ratio; NA, not applicable; NOAC, non-vitamin K antagonist oral anticoagulant; SGLT-2i, sodium-glucose cotransporter-2 inhibitors; and TZD, thiazolidinedione.
Results of sensitivity analyses
The sensitivity analysis excluding patients with CKD demonstrated findings consistent with the primary analysis (see Supplementary material online, Figure S2); these results suggest that this study's findings were not confounded by CKD status. Regarding the analysis of negative control outcomes (occurrence of influenza, urinary tract infection, and suicide death), none of the outcomes demonstrated a significant association with concurrent use of NOACs compared with warfarin (see Supplementary material online, Figure S3). These sensitivity analyses supported the robustness of the study' findings.
Discussion
This nationwide retrospective cohort study demonstrated that patients with DM and AF taking antidiabetic drugs with concurrent use of NOACs had a significantly lower risk of serious hypoglycaemia than those with concurrent use of warfarin. Each NOAC (dabigatran, rivaroxaban, apixaban, and edoxaban) had a similar trend of lower hypoglycaemia risk compared with warfarin.
Two previous observational studies have evaluated the risk of serious hypoglycaemia associated with warfarin use vs. non-use in patients taking antidiabetic medications and suggested an increased risk of serious hypoglycaemia in patients taking sulfonylureas with concurrent use of warfarin.18,19 However, the comparison in those studies was made only between warfarin use and non-use, without an active comparator. Those findings are more prone to confounding by indication bias as patients/periods with or without anticoagulant treatment may represent different populations or patient conditions (e.g. different disease severity or comorbidity status). Furthermore, if patients are indicated for an oral anticoagulant treatment to prevent or treat stroke/thromboembolism, such treatment would be prescribed regardless of the potential risk of hypoglycaemia in clinical practice. Therefore, selecting an appropriate oral anticoagulant with lower hypoglycaemia risk is a more clinically important issue for patients with diabetes requiring anticoagulant treatment. As NOACs have been introduced as alternatives to warfarin, evidence comparing hypoglycaemia risk between NOACs and warfarin in this specific population is needed. However, limited evidence regarding this issue has been reported to date. A previous study in 2020 was the first to suggest that NOACs may lead to a lower risk of hypoglycaemia than warfarin in AF patients with diabetes in its sensitivity analysis.17 However, the study made no distinction between various concurrent antidiabetic medications,17 which is crucial as the underlying mechanisms may be related to potential drug—drug interactions.18 Furthermore, the study design did not consider the potential time-varying exposures and confounders, which is a critical issue as the use of medications, especially oral anticoagulants and antidiabetic drugs, can change during several years of follow-up.17 In addition, a recent observational study focusing exclusively on patients taking sulfonylureas reported an increased risk of severe hypoglycaemia in those taking both sulfonylureas and warfarin compared to sulfonylurea alone,30 consistent with prior literature.18,19 However, the authors proposed residual confounding as a potential issue, as no significant difference in hypoglycaemia risk between warfarin and NOACs was observed in their supplementary analyses.30 Nonetheless, only a limited number of hypoglycaemia events in the NOAC group (43 events) were observed. A larger study with a greater number of events is therefore needed. Our present study focused on patients with AF and DM and used an active-comparator design to compare the risk of serious hypoglycaemia between different oral anticoagulants (NOACs vs. warfarin) in patients taking different antidiabetic drugs. Here, the design of partitioning each patient's follow-up time into different periods according to the antidiabetic medication and oral anticoagulant use and retrieving information on exposures and covariates according to each period considered the time-varying exposures and confounders. Additionally, more comprehensive analyses were performed (including various subgroup, stratified, and sensitivity analyses) than those in previous studies.17–19 Therefore, this study addressed the previous knowledge gaps and provided more robust findings on this crucial clinical issue.
Patients with certain diseases such as severe valvular heart disease and end-stage renal disease tend to be prescribed warfarin during our study period. To avoid potential confounding by indication, our study population has excluded these patients. Also, we have made efforts to control for various potential confounders, such as the index year, age, sex, income level, Charlson Comorbidity Index, CHA2DS2-VASc score, various comorbidities, baseline medications, current diabetes medications, medical utilization, duration of diabetes and atrial fibrillation, and hypoglycaemia history. In our analysis, all these potential confounding factors were well-balanced after IPTW was undertaken. Thus, the difference in the probabilities of being prescribed to patients with clinically complex situations has been minimized. Indeed, our subgroup analyses revealed that the hypoglycaemia risk was lower in NOAC users compared to warfarin users particularly in elderly patients and those taking multiple antidiabetic drugs who are more likely to have a clinically complex situation. These observations suggest that clinical complexity of patients is unlikely to be a confounding factor for our findings.
While the precise mechanisms remain undetermined, some hypotheses have been proposed. First, warfarin is a vitamin K antagonist, whereas NOACs have no such effect. Accumulated evidence has suggested the beneficial effects of vitamin K on insulin sensitivity and diabetes control.12–14 In fact, recent clinical studies suggested that NOACs may offer better glycemic control and lower diabetes complication risks compared to warfarin.16,17,31 Patients with poorly controlled diabetes have heightened risk of hypoglycaemia events.8,32 Thus, the difference in the drug effects on vitamin K may account for the findings observed in this study. Second, warfarin and certain antidiabetic medications (e.g. sulfonylureas) are both primarily metabolized through the cytochrome P450 2C9 hepatic pathway.18,19,33 The shared pathway and a potential drug—drug interaction may hinder sulfonylurea metabolism and increase hypoglycaemia risk. Our study detected the lowest IRR of serious hypoglycaemia within the sulfonylurea subgroup (Figure 3), aligning with the hypothesis that interactions between warfarin and sulfonylureas may precipitate hypoglycaemia events. However, the exact biological mechanisms involved in our study findings remain uncertain, necessitating further investigation in future research.
In clinical practice, it is essential to achieve an optimal time in the therapeutic range (TTR) for patients using warfarin. A previous study from our hospital in Taiwan reported the mean and distribution of TTR among patients receiving warfarin as follows: mean TTR = 0.70; TTR < 0.6: 29.6%; TTR 0.6–0.9: 36.1%; and TTR > 0.9: 34.2%.34 An increase in dose of warfarin may be needed in some warfarin users who do not achieve the optimal TTR; in these cases, the potential negative impact of warfarin on hypoglycaemia risk may be enhanced.
The present study has important clinical implications. Hypoglycaemia, a dangerous but common side effect of diabetes treatment, can cause neurologic damage, cardiovascular events, and death. Prevention of serious hypoglycaemia is crucial for reducing the personal, healthcare, and socioeconomic burden.7,35 This study provides direct, real-world evidence that NOACs are associated with a lower risk of serious hypoglycaemia than warfarin in patients taking antidiabetic agents. Such findings suggest that clinicians should select appropriate medications to prevent serious hypoglycaemia in patients with AF and DM indicated for both anticoagulant and antidiabetic treatment. In patients treated with antidiabetic drugs, NOACs may be used instead of warfarin to decrease the risk of serious hypoglycaemia.
The present study was strengthened by its large-scale, nationwide design using real-world data. The design using periods as the unit of analysis and collecting patient characteristics based on each period helps control potential time-sensitive factors. Nevertheless, several limitations should be addressed. First, the detailed data on patient lifestyle, substance use, and laboratory examination results could not be retrieved from the database. The database also does not contain information regarding patients' medical compliance, disease severity, clinical complexity, and social and domestic factors. While IPTW was applied to balance various patient characteristics between the study groups, it is still possible that there was imbalance in unmeasured confounder between these two groups that may account for our observed results. Second, our study focuses on serious hypoglycaemic events with the need of administration of high-concentration intravenous glucose in the emergency department or during hospitalization. However, some patients, especially those with mild symptoms, might not seek medical assistance and the present study did not include these mild events. As such, our findings cannot be generalized to patients with mild hypoglycaemia. Third, only a few patients were prescribed SGLT-2i and GLP-1 RA in the database because these medications were introduced and covered by Taiwan's National Health Insurance relatively late. The small sample sizes with few event numbers in the subgroup analyses for SGLT-2i and GLP-1 RA resulted in a wide confidence interval or the unavailability to estimate the study outcome. Further studies are required to address this issue. Fourth, given the nature of an observational study, we could not reliably determine whether the modest treatment effect found in our study was causal. Further research is required to confirm our findings. Lastly, our findings were based on the analyses of data in Taiwan, and certain patient groups (e.g. those with ESRD or undergoing valve replacement surgery) were not included. As such, our results may not be generalized to different populations or other areas/countries.
In conclusion, this real-world nationwide cohort study demonstrated that patients with AF and DM taking antidiabetic drugs with concurrent use of NOACs had a significantly lower risk of serious hypoglycaemia than those with concurrent use of warfarin. These findings are based on more evidence from the patients using relatively old antidiabetic agents, but less from the users with modern antidiabetic agents. It is still necessary to further evaluate whether the findings are consistent across patients using different antidiabetic drugs, especially for the most modern antidiabetic agents. Further studies, particularly randomized controlled trials, are needed to validate our findings.
Acknowledgement
The authors thank the Health and Welfare Data Science Center, Ministry of Health and Welfare, Taiwan for approving our access to the database, and the Health and Welfare Data Science Center of Tzu Chi University for facilitating data extraction. The authors thank Editage for English language editing.
Funding
This work was supported by grants from Hualien Tzu Chi Hospital (TCRD111-050 and TCRD111-051). The funder had no role in study design, data collection, data analysis, data interpretation, writing of the report, decision to submit for publication, or approval of the manuscript for publication.
Conflict of interest: None declared.
Ethics approval and consent to participate
The Research Ethics Committee of Hualien Tzu Chi Hospital approved our study (REC No: IRB110-170-C); the requirement for informed consent was waived.
Consent for publication
Not applicable.
Data availability statement
The dataset used in this study is managed by the Taiwan Ministry of Health and Welfare and thus cannot be made available publicly. Researchers interested in accessing this dataset can submit a formal application to the Ministry of Health and Welfare to request access (Taiwan Ministry of Health and Welfare, No. 488, Section 6, Zhongxiao E Rd, Nangang District, Taipei 115, Taiwan; website: https://dep.mohw.gov.tw/DOS/cp-2516-59203-113.html).
