Abstract
Aims
Type 2 diabetes (T2D) accelerates progression of chronic liver disease to cirrhosis, yet the effects of most glucose-lowering drugs (GLDs) on cirrhosis risk in T2D are unknown. To address this gap, we compared cirrhosis risk following initiation of newer second-line GLDs vs. thiazolidinediones (TZDs), which improve histology in non-alcoholic fatty liver disease.
Materials and Methods
Using the US Medicare Fee-for-Service database (2007–2015) and an active comparator, new-user design, we estimated crude incidence rates (IRs) and propensity-score adjusted hazard ratios (aHR) for incident cirrhosis, comparing newer GLDs (dipeptidyl peptidase-4 inhibitors (DPP4i), glucagon-like peptide-1 receptor agonists (GLP1RA), and sodium-glucose co-transporter 2 inhibitors (SGLT2i)) vs. TZDs.
Results
Among 239,549 total initiators, we observed 318, 151, and <30 cirrhosis events when comparing DPP4i vs. TZD, GLP1RA vs. TZD, and SGLT2i vs. TZD, respectively. IRs ranged from 1.7 [95% CI, 0.8–3.6] to 3.6 [2.5–5.2] events per 1,000 person-years. Point aHR estimates for cirrhosis were elevated among newer GLD initiators vs. TZD (DPP4i: 1.15 [0.89–1.50]; GLP1RA: 1.34 [0.82–2.20]; SGLT2i: 1.16, [0.44–3.08]), although estimates were imprecise due to short durations of drug exposure.
Conclusions
We observed mildly elevated cirrhosis risk with newer GLDs vs. TZD; however, uncertainty remains due to imprecise and statistically non-significant effect estimates.
INTRODUCTION
Cirrhosis signifies late-stage chronic liver disease (CLD) resulting from many causes, including hepatitis B/C infection, alcohol-related liver disease and non-alcoholic fatty liver disease (NAFLD). Mortality due to cirrhosis has been rising,1 increasing by 72% between 1999 and 2017.2 Type 2 diabetes (T2D) is a potent risk factor for progression of hepatic steatosis to advanced fibrosis in NAFLD 3, and the presence of T2D also heightens risk of cirrhosis from other etiologies 4, including hepatitis B 5, hepatitis C 6, and alcoholic liver disease 7. Despite T2D being a recognized risk factor for progressive liver disease, little is known about how glucose-lowering drugs (GLDs) influence risk of all-cause cirrhosis.
Thiazolidinediones (TZD) are insulin sensitizers with favorable lipid effects, and are known to improve steatosis and fibrosis in non-alcoholic steatohepatitis 8,9. Notably, reduction of hepatic fat by TZDs may also be beneficial in other forms of liver disease, such as with viral hepatitis 10. Newer GLDs, including dipeptidyl peptidase-4 inhibitors (DPP4i), glucagon-like peptide-1 receptor agonists (GLP1RA) and sodium-glucose cotransporter-2 inhibitors (SGLT2i), have only been explored for their benefits in NAFLD.11,12 Both GLP1RA and SGLT2i improve NAFLD by promoting weight loss; weight-independent effects may also exist 12,13. DPP4i do not appear to have a substantial impact on NAFLD, though data are mixed.12–16
Since T2D likely mediates progression of all-cause liver disease by promoting steatohepatitis, and since GLDs have varying liver effects, it is plausible that GLDs differentially modify cirrhosis risk in patients with T2D and CLD, regardless of etiology. Currently, there is insufficient evidence to recommend one GLD over another to reduce risk of cirrhosis in T2D. We aimed to address this gap by estimating the comparative effect of multiple newer second-line GLDs (DPP4i, GLP1RA, SGLT2i) vs. TZD on risk of incident cirrhosis of any etiology in older adult patients with T2D.
MATERIALS AND METHODS
Data Source
We conducted an active comparator, new-user (ACNU) retrospective cohort study 17, using a nationwide 20% random sample of the US Medicare Fee-for-Service (FFS) database. The Medicare FFS database is representative of the US population aged ≥65, with information on Part A (inpatient services), Part B (outpatient services), and Part D (prescription drug) coverage, as well as enrollment and demographic information.
Study Population
The base population consisted of all patients with ≥1 prescription dispensing claim for a SGLT2i, DPP4i, GLP1RA, or TZD between January 1, 2007 and September 30, 2015, identified using National Drug Codes (NDCs). We conducted three pairwise comparisons (Appendix Tables 1, 2) to estimate the comparative risk of cirrhosis with newer GLDs vs. TZD (DPP4i vs. TZD, GLP1RA vs. TZD, and SGLT2i vs. TZD). Pairwise comparisons involving SGLT2i were only conducted using data from 2013–2015, since SGLT2i were not in routine use in the US before 2013. We additionally compared the three newer GLDs (SGLT2i vs. GLP1RA, SGLT2i vs. DPP4i, and DPP4i vs. GLP1RA); however, we observed generally unstable estimates due to limited drug use data in these comparisons, and report these analyses in the Appendix.
Eligible patients were adults aged ≥65 years with ≥12 months of continuous enrollment in Parts A, B, and D prior to first eligible prescription dispensing claim. To ensure new use of the study drugs, we excluded patients who received a prescription for either drug in each pairwise comparison during the 12-month baseline period leading up to drug initiation (washout period). To remove patients with prevalent cirrhosis, we excluded individuals with the following conditions in the 12-month baseline period: 1) previous diagnosis of cirrhosis in either clinic or hospital setting (ICD-9-CM diagnoses 456.0; 456.1; 456.2; 456.21; 567.23; 571.2; 571.5; 572.2; 572.3; 572.4; 789.59, which achieves a sensitivity of 98–99% for exclusion) 18; 2) previous diagnosis of hepatocellular carcinoma or cholangiocarcinoma; or 3) prior hepatectomy or liver transplantation (Appendix Figure 1, 2).
The study was exempted from full Institutional Review Board review by the University of North Carolina at Chapel Hill. The study protocol was registered with the European Network of Centres for Pharmacoepidemiology and Pharmacovigilance, prior to estimating treatment effects on cirrhosis (EU PAS Register Number 31539).
Exposure
Exposure was defined as ≥2 same-drug class prescription dispensing claims of the study drugs in each pairwise comparison. To simulate ongoing use of the initial treatment, we permitted the second prescription to occur within 30 days (grace period) following the first prescription’s days’ supply, to allow for leeway between prescription fills. The second prescription served as the index date for the analysis. We excluded patients who received a prescription for the comparator drug between the first and second prescriptions of the index drug, and vice versa.
Outcome
The primary outcome of interest (Appendix Table 3) was the first diagnosis of cirrhosis, defined by any of the following ICD-9-CM diagnosis codes in the hospital setting: 456.1, 571.2, or 571.5 (cirrhosis definition 1).18 Due to the lack of a standard, validated claims-based definition for cirrhosis, we also assessed the impact of alternative definitions of cirrhosis (Appendix Table 3), including a more sensitive definition that uses the same codes as our primary definition, but in either outpatient or hospital settings (cirrhosis definition 2).18
Follow-up
We conducted the primary analysis using an “as-treated” approach, where follow-up started at the index date (date of the 2nd prescription) and ended at the time an individual experienced either an outcome of interest or censoring event (Appendix Figure 3). Patients were censored for treatment discontinuation, switch or augmentation; disenrollment from Medicare Parts A, B, or D; or at the administrative study end (September 30, 2015), whichever came first.
Patients were considered to have discontinued treatment if they received no new prescription of the cohort drug class within a (prescription days’ supply + pre-defined 30-day grace period) time window after the last prescription of the cohort drug class; censoring occurred at the end of this window. Similarly, patients were considered to have switched or augmented treatment if they filled a prescription for a comparator drug within the same time window after the last prescription of the cohort drug class; censoring occurred at the fill date of the comparator drug class. Patients who switched between, or augmented with, drugs within the same class were not censored.
Confounding Control
We controlled for measured confounding using propensity score weighting, with the following baseline covariates, measured in the 12 months prior to index date, included in the propensity score model: 1) patient demographics (age, sex, race, low income subsidy); 2) diabetes-related comorbidities (retinopathy, nephropathy, neuropathy, peripheral vascular disease); 3) liver-related diseases (viral hepatitis B and C, alcoholic liver disease, non-alcoholic fatty liver disease, drug-induced liver injury, decompensation); 4) cardiovascular comorbidities (retinopathy, nephropathy, neuropathy, cerebrovascular disease, congestive heart failure, acute myocardial infarction, ischemic heart disease, coagulopathy); 5) general health comorbidities (chronic kidney disease, chronic obstructive pulmonary disease, depression, alcoholism, obesity, smoking status, human immunodeficiency virus infection, history of cancer, peritoneal dialysis, peptic ulcer disease, renal disease); 6) diabetic medication use (metformin, sulfonylurea, long-acting insulin, and any TZD, DPP4i, GLP-1RA, SGLT2i not used to define cohorts); 7) other medication use (angiotensin-converting enzyme inhibitor, angiotensin receptor blockers, beta-blockers, calcium channel blockers, statins, aspirin, prescription omega-3 fatty acids, and various diuretics [loop, thiazide, aldosterone antagonists, and other] lactulose, rifaximin); and 8) measures of healthcare utilization (number of hemoglobin A1c tests, hospitalizations, emergency department visits, physician encounters, gastroenterologist encounters, endocrinologist encounters, number of low-density lipoprotein tests).
Statistical analysis
We estimated propensity scores using multivariable logistic regression and applied them via standardized mortality ratio (SMR) weighting to estimate the average treatment effect in the treated, by reweighting the comparator drug initiators by the propensity score odds (PS/(1-PS)) 19 within each pairwise comparison. This approach seeks to address the question: “what would the observed cirrhosis risk have been if all patients who initiated a newer GLD instead initiated a TZD?” Covariate balance before and after SMR weighting was evaluated using the standardized mean difference (SMD). Asymmetric 1% propensity score trimming was used to remove some patients treated contrary to prediction to reduce the potential for unmeasured confounding.20
To compare incidence of cirrhosis, we estimated crude incidence rates (IR) and 95% confidence intervals (CIs) using Poisson regression, as well as crude and adjusted hazard ratios (aHRs) using SMR-weighted Cox proportional hazards models. Finally, we generated cumulative incidence curves for cirrhosis using weighted Kaplan-Meier methods.
Sensitivity and Subgroup Analysis
We conducted a number of sensitivity and subgroup analyses to assess the impact of various study specifications and definitions. First, we used an initial treatment (IT) approach, ignoring censoring for treatment changes during follow-up (similar to intention-to-treat analyses in randomized clinical trials). Second, we applied 30-, 60-, 90-, 180-, and 270-day induction and latency periods to account for possible diagnostic delay of cirrhosis, where follow-up started e.g., 90 days after index date, and continued for outcomes 90 days after censoring for treatment discontinuation, switch, or augmentation. Third, we re-estimated HRs modeling death as a competing event in the Cox proportional hazards model, using the Aalen-Johansen estimator. To assess the robustness of the SMR weighting approach, we repeated the analysis using a multivariable Cox proportional hazard model. Finally, to assess whether the estimated HRs varied over calendar time, we restricted all analyses to the same 2013–2015 calendar period.
We also repeated the primary analysis within pre-defined, clinically-relevant patient subgroups: 1) patients with/without metformin use during the baseline period; 2) patients with/without baseline liver disease; 3) patients aged 66–75, >75; and 4) men vs. women. In post-hoc analyses, we additionally assessed effect measure modification among 1) patients with/without baseline insulin use; 2) patients with baseline claims for obesity; and 3) patients with self-identified white vs. non-white race.
RESULTS
Study Population
We identified 239,549 patients with initiation of any of the study drugs during the study window. Of those, 103,491 patients were included in the DPP4i (n=69,027) vs. TZD (n=34,464) comparison, 52,473 in the GLP1RA (n=10,728) vs. TZD (n=41,745) comparison, and 18,829 in the SGLT2i (n=7,849) vs. TZD (n=10,980) comparison (Appendix Figure 1, Table 1).
Table 1.
Baseline Characteristics of Eligible Initiators of DPP4i vs. TZD, GLP1RA vs. TZD, and SGLT2i vs. TZD, Before and After Implementation of Standardized Mortality Ratio (SMR) Weightinga (365-Day Washout Period, 1% Asymmetric Propensity Score Trimming)
| DPP4i vs. TZD | GLP1RA vs. TZD | SGLT2i vs. TZD | |||||||
|---|---|---|---|---|---|---|---|---|---|
| Characteristicsb | DPP4i initiators (n=69,027) | TZD initiators (n=34,464) | Weighted TZD initiators (n=68,979) | GLP1RA initiators (n=10,728) | TZD initiators (n=41,745) | Weighted TZD initiators (n=10,689) | SGLT2i initiators (n=7,849) | TZD initiators (n=10,980) | Weighted TZD initiators (n =7,984) |
| Age, mean (std. dev.) | 74.1 (7.1) | 72.9 (6.9) | 74.0 (7.1) | 70.7 (5.3) | 73.1 (6.9) | 70.6 (5.3) | 71.6 (5.6) | 73.0 (6.8) | 71.6 (5.6) |
| Male | 28015 (40.6) | 15281 (44.3) | 27907 (40.5) | 4349 (40.5) | 18512 (44.3) | 4263 (39.9) | 3840 (48.9) | 5371 (48.9) | 3887 (48.7) |
| Race | |||||||||
| White | 52040 (75.4) | 25123 (72.9) | 52178 (75.6) | 9074 (84.6) | 30056 (72.0) | 9059 (84.7) | 6446 (82.1) | 8263 (75.3) | 6622 (82.9) |
| Black | 7499 (10.9) | 4035 (11.7) | 7315 (10.6) | 873 (8.1) | 4765 (11.4) | 883 (8.3) | 565 (7.2) | 1056 (9.6) | 553 (6.9) |
| Other | 9488 (13.7) | 5306 (15.4) | 9486 (13.8) | 781 (7.3) | 6924 (16.6) | 748 (7.0) | 838 (10.7) | 1661 (15.1) | 809 (10.1) |
| Diabetes Comorbiditiesb | |||||||||
| Nephropathy | 5769 (8.4) | 2353 (6.8) | 5791 (8.4) | 1182 (11.0) | 3115 (7.5) | 1176 (11.0) | 634 (8.1) | 1178 (10.7) | 661 (8.3) |
| Neuropathy | 13480 (19.5) | 5465 (15.9) | 13640 (19.8) | 2666 (24.9) | 7096 (17.0) | 2685 (25.1) | 1959 (25.0) | 2169 (19.8) | 2043 (25.6) |
| Retinopathy | 10414 (15.1) | 4893 (14.2) | 10392 (15.1) | 1925 (17.9) | 6248 (15.0) | 1913 (17.9) | 1438 (18.3) | 1609 (14.7) | 1430 (17.9) |
| Peripheral vascular disease | 9521 (13.8) | 3977 (11.5) | 9508 (13.8) | 1208 (11.3) | 5000 (12.0) | 1231 (11.5) | 920 (11.7) | 1213 (11.0) | 964 (12.1) |
| General Health Comorbiditiesb | |||||||||
| Alcoholic liver disease | 31 (0.0) | 16 (0.0) | 30 (0.0) | NTSRc | 17 (0.0) | NTSR | NTSR | NTSR | NTSR |
| Alcohol abuse | 320 (0.5) | 179 (0.5) | 314 (0.5) | 29 (0.3) | 212 (0.5) | 27 (0.3) | 19 (0.2) | 20 (0.2) | 23 (0.3) |
| AMI | 3283 (4.8) | 1176 (3.4) | 3247 (4.7) | 465 (4.3) | 1446 (3.5) | 457 (4.3) | 320 (4.1) | 389 (3.5) | 290 (3.6) |
| Cerebrovascular disease | 13204 (19.1) | 5518 (16.0) | 13124 (19.0) | 1564 (14.6) | 6842 (16.4) | 1596 (14.9) | 1233 (15.7) | 1563 (14.2) | 1287 (16.1) |
| CHF (exclusion) | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| CKD | 12073 (17.5) | 4743 (13.8) | 12080 (17.5) | 1888 (17.6) | 6173 (14.8) | 1894 (17.7) | 939 (12.0) | 2208 (20.1) | 975 (12.2) |
| Coagulopathy | 2377 (3.4) | 935 (2.7) | 2417 (3.5) | 287 (2.7) | 1167 (2.8) | 287 (2.7) | 213 (2.7) | 278 (2.5) | 204 (2.6) |
| COPD | 15371 (22.3) | 6704 (19.5) | 15265 (22.1) | 2431 (22.7) | 8182 (19.6) | 2387 (22.3) | 1585 (20.2) | 2064 (18.8) | 1678 (21.0) |
| Decompensation | 3532 (5.1) | 1182 (3.4) | 3525 (5.1) | 391 (3.6) | 1505 (3.6) | 378 (3.5) | 195 (2.5) | 446 (4.1) | 199 (2.5) |
| Depression | 10174 (14.7) | 4163 (12.1) | 10116 (14.7) | 1717 (16.0) | 5087 (12.2) | 1685 (15.8) | 1088 (13.9) | 1388 (12.6) | 1132 (14.2) |
| Diabetes | 66541 (96.4) | 31795 (92.3) | 66456 (96.3) | 10289 (95.9) | 38887 (93.2) | 10252 (95.9) | 7750 (98.7) | 10290 (93.7) | 7885 (98.8) |
| Drug induced liver | 513 (0.7) | 281 (0.8) | 537 (0.8) | 66 (0.6) | 371 (0.9) | 67 (0.6) | 44 (0.6) | 53 (0.5) | 50 (0.6) |
| Dyslipidemia | 57787 (83.7) | 26207 (76.0) | 57698 (83.6) | 9130 (85.1) | 32418 (77.7) | 9069 (84.8) | 7053 (89.9) | 8936 (81.4) | 7186 (90.0) |
| HIV | 125 (0.2) | 72 (0.2) | 133 (0.2) | 13 (0.1) | 93 (0.2) | 15 (0.1) | NTSR | 19 (0.2) | NTSR |
| Ischemic heart disease | 22087 (32.0) | 9052 (26.3) | 21991 (31.9) | 3295 (30.7) | 11359 (27.2) | 3324 (31.1) | 2409 (30.7) | 2827 (25.7) | 2451 (30.7) |
| Liver disease | 155 (0.2) | 65 (0.2) | 153 (0.2) | 13 (0.1) | 90 (0.2) | 13 (0.1) | 23 (0.3) | 19 (0.2) | 17 (0.2) |
| Non-alcoholic liver disease | 1840 (2.7) | 680 (2.0) | 1874 (2.7) | 400 (3.7) | 885 (2.1) | 416 (3.9) | 332 (4.2) | 323 (2.9) | 356 (4.5) |
| Obesity | 9142 (13.2) | 3344 (9.7) | 9097 (13.2) | 2900 (27.0) | 4119 (9.9) | 2860 (26.8) | 1916 (24.4) | 1599 (14.6) | 1965 (24.6) |
| Peptic ulcer disease | 1358 (2.0) | 639 (1.9) | 1368 (2.0) | 136 (1.3) | 799 (1.9) | 135 (1.3) | 91 (1.2) | 176 (1.6) | 85 (1.1) |
| Renal disease | 13316 (19.3) | 5325 (15.5) | 13311 (19.3) | 2099 (19.6) | 6917 (16.6) | 2104 (19.7) | 1068 (13.6) | 2383 (21.7) | 1098 (13.7) |
| Smoking | 6542 (9.5) | 2410 (7.0) | 6534 (9.5) | 1085 (10.1) | 2979 (7.1) | 1074 (10.0) | 887 (11.3) | 1181 (10.8) | 952 (11.9) |
| Viral hepatitis | 504 (0.7) | 218 (0.6) | 528 (0.8) | 52 (0.5) | 281 (0.7) | 47 (0.4) | 36 (0.5) | 58 (0.5) | 34 (0.4) |
| History of cancer | 11034 (16.0) | 4599 (13.3) | 11068 (16.0) | 1564 (14.6) | 5699 (13.7) | 1559 (14.6) | 1240 (15.8) | 1576 (14.4) | 1279 (16.0) |
| Peritoneal dialysis | 33 (0.0) | 27 (0.1) | 36 (0.1) | NTSR | 32 (0.1) | NTSR | NTSR | NTSR | NTSR |
| Prior Medication Useb | |||||||||
| ACEI | 32481 (47.1) | 17161 (49.8) | 32399 (47.0) | 4995 (46.6) | 20722 (49.6) | 5004 (46.8) | 3534 (45.0) | 5175 (47.1) | 3657 (45.8) |
| ARB | 21700 (31.4) | 8771 (25.4) | 21739 (31.5) | 3637 (33.9) | 11293 (27.1) | 3587 (33.6) | 2962 (37.7) | 3267 (29.8) | 3021 (37.8) |
| Aspirin | 2570 (3.7) | 1236 (3.6) | 2570 (3.7) | 568 (5.3) | 1485 (3.6) | 559 (5.2) | 378 (4.8) | 361 (3.3) | 391 (4.9) |
| Beta blockers | 34125 (49.4) | 14687 (42.6) | 34161 (49.5) | 5116 (47.7) | 18203 (43.6) | 5086 (47.6) | 3817 (48.6) | 5020 (45.7) | 3920 (49.1) |
| Calcium channel blockers | 25268 (36.6) | 11387 (33.0) | 25178 (36.5) | 3587 (33.4) | 14127 (33.8) | 3535 (33.1) | 2629 (33.5) | 3834 (34.9) | 2713 (34.0) |
| Lactulose | 896 (1.3) | 376 (1.1) | 882 (1.3) | 90 (0.8) | 477 (1.1) | 89 (0.8) | 50 (0.6) | 102 (0.9) | 62 (0.8) |
| Omega-3 | 2068 (3.0) | 712 (2.1) | 2135 (3.1) | 389 (3.6) | 1038 (2.5) | 389 (3.6) | 341 (4.3) | 318 (2.9) | 358 (4.5) |
| Rifaximin | 58 (0.1) | 19 (0.1) | 60 (0.1) | 11 (0.1) | 25 (0.1) | NTSR | NTSR | NTSR | NTSR |
| Statins | 49068 (71.1) | 22641 (65.7) | 48928 (70.9) | 7841 (73.1) | 28040 (67.2) | 7720 (72.2) | 5980 (76.2) | 8011 (73.0) | 6061 (75.9) |
| Loop diuretics | 11631 (16.8) | 4933 (14.3) | 11642 (16.9) | 2196 (20.5) | 5926 (14.2) | 2172 (20.3) | 1095 (14.0) | 1403 (12.8) | 1069 (13.4) |
| Other diuretics | 16826 (24.4) | 8056 (23.4) | 16910 (24.5) | 2766 (25.8) | 9954 (23.8) | 2808 (26.3) | 1944 (24.8) | 2544 (23.2) | 2031 (25.4) |
| Thiazide diuretics | 10457 (15.1) | 5183 (15.0) | 10334 (15.0) | 1671 (15.6) | 6157 (14.7) | 1634 (15.3) | 1116 (14.2) | 1562 (14.2) | 1140 (14.3) |
| Aldosterone antagonists | 1529 (2.2) | 593 (1.7) | 1491 (2.2) | 349 (3.3) | 712 (1.7) | 332 (3.1) | 197 (2.5) | 235 (2.1) | 225 (2.8) |
| Metformin | 49045 (71.1) | 22373 (64.9) | 49086 (71.2) | 7144 (66.6) | 27765 (66.5) | 7127 (66.7) | 6051 (77.1) | 7409 (67.5) | 6182 (77.4) |
| Sulfonylureas | 35193 (51.0) | 18047 (52.4) | 35504 (51.5) | 5094 (47.5) | 22742 (54.5) | 5128 (48.0) | 4008 (51.1) | 5910 (53.8) | 4102 (51.4) |
| Thiazolidinediones | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| DPP4i | --- | --- | --- | 3254 (30.3) | 7456 (17.9) | 3430 (32.1) | 3369 (42.9) | 3331 (30.3) | 3592 (45.0) |
| GLP1RA | 1469 (2.1) | 815 (2.4) | 1596 (2.3) | --- | --- | --- | 1198 (15.3) | 501 (4.6) | 1317 (16.5) |
| SGLT2i | 331 (0.5) | 100 (0.3) | 351 (0.5) | 247 (2.3) | 174 (0.4) | 239 (2.2) | --- | --- | --- |
| Long-acting insulin | 10803 (15.7) | 5058 (14.7) | 10852 (15.7) | 3919 (36.5) | 5896 (14.1) | 3902 (36.5) | 2458 (31.3) | 1754 (16.0) | 2629 (32.9) |
| Measures of Healthcare Utilization in Year Prior to Index Dateb | |||||||||
| Low-income subsidy (LIS) status | |||||||||
| 0 – no subsidy | 38026 (55.1) | 17261 (50.1) | 38242 (55.4) | 7000 (65.2) | 20815 (49.9) | 6735 (63.0) | 5656 (72.1) | 6889 (62.7) | 5719 (71.6) |
| 1 – 100 premium subsidy | 27451 (39.8) | 15013 (43.6) | 27120 (39.3) | 3172 (29.6) | 18398 (44.1) | 3575 (33.4) | 1914 (24.4) | 3626 (33.0) | 2034 (25.5) |
| 2 – partial (25–75) premium subsidy | 3550 (5.1) | 2190 (6.4) | 3617 (5.2) | 556 (5.2) | 2532 (6.1) | 380 (3.6) | 279 (3.6) | 465 (4.2) | 232 (2.9) |
| N HbA1c tests in past year | |||||||||
| 0 | 7352 (10.7) | 6513 (18.9) | 7658 (11.1) | 1102 (10.3) | 7250 (17.4) | 1153 (10.8) | 353 (4.5) | 1507 (13.7) | 410 (5.1) |
| 1 | 11907 (17.2) | 6649 (19.3) | 11533 (16.7) | 1474 (13.7) | 7574 (18.1) | 1426 (13.3) | 900 (11.5) | 1684 (15.3) | 826 (10.3) |
| 2 | 16796 (24.3) | 7635 (22.2) | 16000 (23.2) | 2282 (21.3) | 9236 (22.1) | 2236 (20.9) | 1825 (23.3) | 2442 (22.2) | 1700 (21.3) |
| ≥3 | 32972 (47.8) | 13667 (39.7) | 33788 (49.0) | 5870 (54.7) | 17685 (42.4) | 5874 (55.0) | 4771 (60.8) | 5347 (48.7) | 5048 (63.2) |
| N LDL tests in past year | |||||||||
| 0 | 11682 (16.9) | 8413 (24.4) | 12065 (17.5) | 1758 (16.4) | 9506 (22.8) | 1798 (16.8) | 801 (10.2) | 2119 (19.3) | 849 (10.6) |
| 1 | 19513 (28.3) | 9703 (28.2) | 19084 (27.7) | 2901 (27.0) | 11506 (27.6) | 2802 (26.2) | 2114 (26.9) | 3124 (28.5) | 2086 (26.1) |
| 2 | 18132 (26.3) | 7979 (23.2) | 17383 (25.2) | 2715 (25.3) | 9937 (23.8) | 2699 (25.3) | 2173 (27.7) | 2745 (25.0) | 2110 (26.4) |
| ≥3 | 19700 (28.5) | 8369 (24.3) | 20447 (29.6) | 3354 (31.3) | 10796 (25.9) | 3391 (31.7) | 2761 (35.2) | 2992 (27.2) | 2938 (36.8) |
| Flu shot received in past year | 37731 (54.7) | 16254 (47.2) | 37780 (54.8) | 6089 (56.8) | 20200 (48.4) | 6013 (56.3) | 4883 (62.2) | 5907 (53.8) | 4968 (62.2) |
| N hospitalizations | |||||||||
| 0 | 56180 (81.4) | 28873 (83.8) | 56191 (81.5) | 9311 (86.8) | 34944 (83.7) | 9228 (86.3) | 7073 (90.1) | 9536 (86.8) | 7161 (89.7) |
| 1 | 7204 (10.4) | 3313 (9.6) | 7266 (10.5) | 939 (8.8) | 4009 (9.6) | 970 (9.1) | 553 (7.0) | 904 (8.2) | 598 (7.5) |
| 2 | 3321 (4.8) | 1333 (3.9) | 3184 (4.6) | 319 (3.0) | 1627 (3.9) | 303 (2.8) | 168 (2.1) | 341 (3.1) | 141 (1.8) |
| ≥3 | 2322 (3.4) | 945 (2.7) | 2337 (3.4) | 159 (1.5) | 1165 (2.8) | 188 (1.8) | 55 (0.7) | 199 (1.8) | 84 (1.0) |
| N days in hospital | |||||||||
| 0 | 56180 (81.4) | 28873 (83.8) | 56191 (81.5) | 9311 (86.8) | 34944 (83.7) | 9228 (86.3) | 7073 (90.1) | 9536 (86.8) | 7161 (89.7) |
| 1–2 | 3133 (4.5) | 1490 (4.3) | 3145 (4.6) | 452 (4.2) | 1825 (4.4) | 473 (4.4) | 275 (3.5) | 414 (3.8) | 302 (3.8) |
| 3–5 | 3289 (4.8) | 1435 (4.2) | 3305 (4.8) | 433 (4.0) | 1739 (4.2) | 430 (4.0) | 244 (3.1) | 429 (3.9) | 244 (3.1) |
| 5–10 | 1732 (2.5) | 780 (2.3) | 1705 (2.5) | 168 (1.6) | 916 (2.2) | 173 (1.6) | 101 (1.3) | 156 (1.4) | 105 (1.3) |
| >10 | 4693 (6.8) | 1886 (5.5) | 4632 (6.7) | 364 (3.4) | 2321 (5.6) | 385 (3.6) | 156 (2.0) | 445 (4.1) | 172 (2.2) |
| N emergency department visits | |||||||||
| 0 | 48019 (69.6) | 25361 (73.6) | 47994 (69.6) | 8080 (75.3) | 30621 (73.4) | 8013 (75.0) | 6028 (76.8) | 8211 (74.8) | 6099 (76.4) |
| 1 | 13052 (18.9) | 5816 (16.9) | 13114 (19.0) | 1769 (16.5) | 7090 (17.0) | 1807 (16.9) | 1247 (15.9) | 1833 (16.7) | 1255 (15.7) |
| ≥2 | 7956 (11.5) | 3287 (9.5) | 7871 (11.4) | 879 (8.2) | 4034 (9.7) | 870 (8.1) | 574 (7.3) | 936 (8.5) | 630 (7.9) |
| N physician encounters | |||||||||
| 0 | 2055 (3.0) | 2317 (6.7) | 2054 (3.0) | 391 (3.6) | 2547 (6.1) | 406 (3.8) | 76 (1.0) | 580 (5.3) | 76 (1.0) |
| 1–3 | 4183 (6.1) | 3641 (10.6) | 4177 (6.1) | 689 (6.4) | 4030 (9.7) | 704 (6.6) | 298 (3.8) | 994 (9.1) | 293 (3.7) |
| 4–6 | 5713 (8.3) | 3454 (10.0) | 5728 (8.3) | 757 (7.1) | 4030 (9.7) | 755 (7.1) | 592 (7.5) | 1045 (9.5) | 575 (7.2) |
| ≥7 | 57076 (82.7) | 25052 (72.7) | 57021 (82.7) | 8891 (82.9) | 31138 (74.6) | 8824 (82.6) | 6883 (87.7) | 8361 (76.1) | 7040 (88.2) |
| N gastroenterologist visits | |||||||||
| 0 | 59121 (85.6) | 30507 (88.5) | 59234 (85.9) | 9140 (85.2) | 36733 (88.0) | 9166 (85.7) | 6673 (85.0) | 9679 (88.2) | 6763 (84.7) |
| 1 | 4077 (5.9) | 1653 (4.8) | 3862 (5.6) | 689 (6.4) | 2074 (5.0) | 653 (6.1) | 528 (6.7) | 588 (5.4) | 549 (6.9) |
| 2 | 2813 (4.1) | 1094 (3.2) | 2686 (3.9) | 491 (4.6) | 1393 (3.3) | 416 (3.9) | 336 (4.3) | 359 (3.3) | 351 (4.4) |
| ≥3 | 3016 (4.4) | 1210 (3.5) | 3197 (4.6) | 408 (3.8) | 1545 (3.7) | 455 (4.3) | 312 (4.0) | 354 (3.2) | 322 (4.0) |
| N endocrinologist visits | |||||||||
| 0 | 60351 (87.4) | 32025 (92.9) | 60986 (88.4) | 8029 (74.8) | 38296 (91.7) | 8233 (77.0) | 6078 (77.4) | 9750 (88.8) | 6145 (77.0) |
| 1 | 2947 (4.3) | 818 (2.4) | 2080 (3.0) | 819 (7.6) | 1114 (2.7) | 502 (4.7) | 372 (4.7) | 357 (3.3) | 382 (4.8) |
| 2 | 1944 (2.8) | 504 (1.5) | 1619 (2.3) | 504 (4.7) | 706 (1.7) | 455 (4.3) | 323 (4.1) | 242 (2.2) | 301 (3.8) |
| ≥3 | 3785 (5.5) | 1117 (3.2) | 4293 (6.2) | 1376 (12.8) | 1629 (3.9) | 1499 (14.0) | 1076 (13.7) | 631 (5.7) | 1156 (14.5) |
| Year of initiation | |||||||||
| 2008 | 4123 (6.0) | 7099 (20.6) | 14059 (20.4) | 427 (4.0) | 7696 (18.4) | 1594 (14.9) | --- | --- | --- |
| 2009 | 5085 (7.4) | 7892 (22.9) | 15362 (22.3) | 466 (4.3) | 8874 (21.3) | 1852 (17.3) | --- | --- | --- |
| 2010 | 6187 (9.0) | 6298 (18.3) | 12420 (18.0) | 578 (5.4) | 7373 (17.7) | 1715 (16.0) | --- | --- | --- |
| 2011 | 9232 (13.4) | 3829 (11.1) | 7319 (10.6) | 934 (8.7) | 4765 (11.4) | 1129 (10.6) | --- | --- | --- |
| 2012 | 10825 (15.7) | 2101 (6.1) | 4025 (5.8) | 1484 (13.8) | 2740 (6.6) | 721 (6.7) | 0 (0) | 281 (2.6) | 189 (2.4) |
| 2013 | 11457 (16.6) | 2345 (6.8) | 4881 (7.1) | 2148 (20.0) | 3247 (7.8) | 1026 (9.6) | 550 (7.0) | 3408 (31.0) | 2359 (29.5) |
| 2014 | 12749 (18.5) | 2794 (8.1) | 6202 (9.0) | 2473 (23.1) | 4085 (9.8) | 1467 (13.7) | 3279 (41.8) | 4257 (38.8) | 3193 (40.0) |
| 2015 | 9369 (13.6) | 2106 (6.1) | 4711 (6.8) | 2218 (20.7) | 2965 (7.1) | 1185 (11.1) | 4020 (51.2) | 3034 (27.6) | 2242 (28.1) |
Abbreviations: SGLT2i, sodium-glucose cotransporter-2 inhibitors; DPP4i, dipeptidyl peptidase-4 inhibitors; GLP1RA, glucagon-like peptide-1 receptor agonist; TZD, thiazolidinediones; AMI, acute myocardial infarction; CHF, congestive heart failure; CKD, chronic kidney disease; COPD, chronic obstructive pulmonary disease; ACEI, angiotensin-converting-enzyme inhibitor; ARB, angiotensin receptor blockers; HbA1c, hemoglobin A1c; LDL, low-density lipoprotein; NTSR, number too small to report c
Weighted by standardizing the comparator drug initiators to the population of index treatment initiators, using the propensity score odds (PS/(1-PS)), to estimate the treatment effect in the treated (ATT).
All baseline characteristics measured in the one year (365 days) prior to date of cohort drug initiation
Number too small to report (<11), per Medicare Data Use Agreement
Initiators of TZDs exhibited higher proportions of male and non-white patients compared to initiators of the three newer second-line GLDs. Comorbidities were generally more prevalent among initiators of newer GLDs; we observed the largest crude differences between baseline covariate distributions in the SGLT2i vs. TZD comparison (Appendix Figures 4, 5), relative to other comparisons. Prevalence of codes for baseline obesity, as well as the proportion of patients with DPP4i and insulin use during the baseline period, were notably elevated in GLP1RA and SGLT2i users. TZD initiators, on the other hand, had higher prevalence of renal disease compared to SGLT2i initiators, and generally more prior sulfonylurea and ACEI use. Covariate balance was improved by SMR weighting (Appendix Figure 5).
Incidence of Cirrhosis
In primary as-treated analysis, we observed 318, 151, and <30 cirrhosis events when comparing DPP4i vs. TZD, GLP1RA vs. TZD, and SGLT2i vs. TZD, respectively. Crude incidence rates of cirrhosis ranged from 1.7 [95% CI, 0.8–3.6] events per 1,000 person-years in the SGLT2i cohort, to 3.6 [2.5–5.2] events per 1,000 person-years in the GLP1RA cohort, and were generally higher among patients treated with DPP4i and GLP1RA, vs. TZD (Table 2). After weighting, we observed elevated aHR point estimates for all three newer second-line GLDs (Table 2). Compared to TZD initiators, the estimated hazard for cirrhosis was most elevated for GLP1RA initiators (aHR 1.34, 95% CI 0.82–2.20), followed by SGLT2i initiators (aHR 1.16, 95% CI 0.44–3.08) and DPP4i initiators (aHR 1.15, 95% CI 0.89–1.50). The elevated aHR estimates in the GLP1RA comparison were supported by separation of the cumulative incidence curves (Figure 1), although curves crossed after 420 days of follow-up, likely due to low numbers of patients at-risk in the GLP1RA group. Alternative cirrhosis definitions yielded further-increased aHR estimates among newer GLD users for all TZD comparisons (Table 3, Figure 1).
Table 2.
Crude and Adjusted Hazard Ratio (HR) Estimates for Incident Liver Cirrhosis with Second-Line Glucose-Lowering Drug Initiation (365-Day Washout Period, As-Treated Analysis, 1% Asymmetric Propensity Score Trimming)
| Comparison | Cohort | Number of Patients | Median (IQR) Follow-up Time, Yearsa | Total Person-Years | Number of Eventsb | Crude Incidence Rate, per 1,000 Person-Years | Crude HR (95% CI) | PS-weighted HR (95% CI) |
|---|---|---|---|---|---|---|---|---|
| DPP4i vs. TZD | ||||||||
| AT Analysis | DPP4i | 69027 | 0.66 (0.33–1.51) | 78778 | 220 | 2.8 (2.4–3.2) | 1.09 (0.86–1.39) | 1.15 (0.89–1.50) |
| TZD | 34464 | 0.64 (0.33–1.45) | 38393 | 98 | 2.6 (2.1–3.1) | |||
| IT Analysis | DPP4i | 69027 | 2.08 (0.91–3.00) | 131309 | 392 | 3.0 (2.7–3.3) | 1.03 (0.88–1.21) | 1.06 (0.89–1.26) |
| TZD | 34464 | 3.00 (1.52–3.00) | 79036 | 232 | 2.9 (2.6–3.3) | |||
| GLP1RA vs. TZD | ||||||||
| AT Analysis | GLP1RA | 10728 | 0.45 (0.25–0.96) | 8264 | 30 | 3.6 (2.5–5.2) | 1.41 (0.95–2.11) | 1.34 (0.82–2.20) |
| TZD | 41745 | 0.64 (0.33–1.42) | 45741 | 121 | 2.6 (2.2–3.2) | |||
| IT Analysis | GLP1RA | 10728 | 1.66 (0.67–3.00) | 18355 | 57 | 3.1 (2.4–4.0) | 1.06 (0.80–1.41) | 0.95 (0.67–1.34) |
| TZD | 41745 | 3.00 (1.40–3.00) | 93420 | 281 | 3.0 (2.7–3.4) | |||
| SGLT2i vs. TZD | ||||||||
| AT Analysis | SGLT2i | 7849 | 0.40 (0.21–0.70) | NTSRb | NTSR | 1.7 (0.8–3.6) | 0.91 (0.38–2.21) | 1.16 (0.44–3.08) |
| TZD | 10980 | 0.58 (0.30–1.11) | 8507 | 17 | 2.0 (1.2–3.2) | |||
| IT Analysis | SGLT2i | 7849 | 0.60 (0.31–1.03) | NTSR | NTSR | 1.6 (0.8–3.1) | 0.62 (0.30–1.29) | 0.80 (0.38–1.67) |
| TZD | 10980 | 1.10 (0.51–1.75) | 12953 | 35 | 2.7 (1.9–3.8) | |||
Abbreviations: SGLT2i, sodium-glucose cotransporter-2 inhibitors; DPP4i, dipeptidyl peptidase-4 inhibitors; GLP1RA, glucagon-like peptide-1 receptor agonists; TZD, thiazolidinediones; IQR, interquartile range; PS, propensity score; AT, as-treated; IT, initial treatment (similar to intention-to-treat approach in randomized controlled trials); NTSR, number too small to report b
Follow-up in IT analyses were capped at 3 years following index date.
Number too small to report (<11), per Medicare Data Use Agreement
Figure 1. SMR-Weighteda Cumulative Incidence (Kaplan-Meier) Curves for Incident Liver Cirrhosis following Second-Line Glucose-Lowering Drug Initiation (365-Day Washout Period, As-Treated Analysis, 1% Asymmetric Propensity Score Trimming).
Abbreviations: SMR, standardized mortality ratio; SGLT2i, sodium-glucose cotransporter-2 inhibitors; DPP4i, dipeptidyl peptidase-4 inhibitors; GLP1RA, glucagon-like peptide-1 receptor agonists; TZD, thiazolidinediones; HR, hazard ratio
a. Weighted by standardizing the comparator drug initiators to the population of index drug initiators, using the propensity score odds (PS/(1-PS)), to estimate the treatment effect in the treated (ATT).
Table 3.
Crude and Adjusted Hazard Ratio (HR) Estimates for Incident Liver Cirrhosis under Alternative Outcome Definitions (365-Day Washout Period, As-Treated Analysis, 1% Asymmetric Propensity Score Trimming)
| High Specificity Definitionsa | High Sensitivity Definitionsa | |||||||
|---|---|---|---|---|---|---|---|---|
| Comparison | Cohort | Number of Patients | Crude Incidence Rate, per 1,000 Person-Years | Crude HR (95% CI) | PS-weighted HR (95% CI) | Crude Incidence Rate, per 1,000 Person-Years | Crude HR (95% CI) | PS-weighted HR (95% CI) |
| DPP4i vs. TZD | ||||||||
| Lapointe et al, 2018 | DPP4i | 69027 | 2.8 (2.4–3.2) | 1.09 (0.86–1.39) | 1.15 (0.89–1.50) | 6.2 (5.7–6.8) | 1.34 (1.13–1.59) | 1.38 (1.14–1.66) |
| Definitions 1 & 2 | TZD | 34464 | 2.6 (2.1–3.1) | 4.6 (4.0–5.4) | ||||
| Nehra et al, 2013 | DPP4i | 69027 | 1.0 (0.8–1.2) | 1.03 (0.69–1.52) | 1.03 (0.68–1.56) | 6.9 (6.3–7.5) | 1.20 (1.02–1.40) | 1.25 (1.05–1.48) |
| Definitions 3 & 4 | TZD | 34464 | 1.0 (0.7–1.3) | 5.8 (5.1–6.6) | ||||
| GLP1RA vs. TZD | ||||||||
| Lapointe et al, 2018 | GLP1RA | 10728 | 3.6 (2.5–5.2) | 1.41 (0.95–2.11) | 1.34 (0.82–2.20) | 7.0 (5.4–9.1) | 1.37 (1.03–1.83) | 1.21 (0.85–1.70) |
| Definitions 1 & 2 | TZD | 41745 | 2.6 (2.2–3.2) | 4.9 (4.3–5.6) | ||||
| Nehra et al, 2013 | GLP1RA | 10728 | 1.9 (1.2–3.2) | 1.93 (1.09–3.41) | 1.77 (0.79–3.97) | 8.3 (6.5–10.5) | 1.32 (1.02–1.73) | 1.19 (0.86–1.65) |
| Definitions 3 & 4 | TZD | 41745 | 1.0 (0.7–1.3) | 5.9 (5.3–6.7) | ||||
| SGLT2i vs. TZD | ||||||||
| Lapointe et al, 2018 | SGLT2i | 7849 | 1.7 (0.8–3.6) | 0.91 (0.38–2.21) | 1.16 (0.44–3.08) | 7.8 (5.5–11.1) | 1.48 (0.92–2.39) | 1.64 (0.94–2.86) |
| Definitions 1 & 2 | TZD | 10980 | 2.0 (1.2–3.2) | 4.8 (3.6–6.6) | ||||
| Nehra et al, 2013 | SGLT2i | 7849 | 0.7 (0.2–2.3) | 1.04 (0.26–4.17) | 1.99 (0.49–8.02) | 7.8 (5.5–11.1) | 1.28 (0.81–2.04) | 1.38 (0.80–2.38) |
| Definitions 3 & 4 | TZD | 10980 | 0.8 (0.4–1.7) | 5.5 (4.2–7.4) | ||||
Abbreviations: SGLT2i, sodium-glucose cotransporter-2 inhibitors; DPP4i, dipeptidyl peptidase-4 inhibitors; GLP1RA, glucagon-like peptide-1 receptor agonists; TZD, thiazolidinediones; IQR, interquartile range; PS, propensity score; AT, as-treated; IT, initial treatment (similar to intention-to-treat approach in randomized controlled trials)
Definition 1: 456.1; 571.2; 571.5; INPATIENT ONLY (specificity 91–96%; sensitivity 57–77%)
Definition 2: 456.1; 571.2; 571.5; INPATIENT (INPT) & OUTPATIENT (OUTPT) (specificity 61–77%; sensitivity 98–99%)
Definition 3: 456.0; 456.2; 456.21; 572.4; INPT/OUTPT (specificity 98.3%; sensitivity 11.3%)
Definition 4: 456.0; 456.1; 456.2; 456.21; 571.2; 571.5; 572.2; 572.3; 572.4; 567.23; INPT/OUTPT (specificity 43%; sensitivity 97.7%)
Subgroup and Sensitivity Analyses
Results remained generally consistent across a number of sensitivity analyses (Figure 2). In all TZD comparisons, aHR estimates were elevated among patients who were younger (age ≤75), female, and who had a history of liver disease. Hazards also appeared to be increased among patients who self-identified as non-white in the GLP1RA and SGLT2i comparisons, while the reverse was observed in the DPP4i comparison. Subgroup aHR estimates were generally imprecise in all comparisons involving SGLT2i due to few observed events.
Figure 2. Adjusted Hazard Ratio (HR) Estimates for Incident Liver Cirrhosis with Second-Line Glucose-Lowering Drug, Sensitivity and Subgroup Analyses, TZD Comparisonsb.
Abbreviations: SGLT2i, sodium-glucose cotransporter-2 inhibitors; DPP4i, dipeptidyl peptidase-4 inhibitors; SU, sulfonylureas; IQR, interquartile range; SMR, standardized mortality ratio; HR, hazard ratio; AT, as-treated; ITT, intention-to-treat; Hx, history
a. Weighted by standardizing the comparator drug initiators to the population of SGLT2i initiators, using the propensity score odds (PS/(1-PS)), to estimate the treatment effect in the treated (ATT).
b. Primary analyses performed with both induction and latency periods = 0
c. Propensity scores were re-estimated within each subgroup
DISCUSSION
This is the first study examining the comparative effect of second-line glucose-lowering drugs (GLDs) on risk of incident cirrhosis in patients with T2D. In general, point estimates suggested a possible trend towards lowest cirrhosis risk with TZD compared to newer second-line GLDs, which was largely consistent across a number of sensitivity analyses. However, short exposure durations to the study drugs yielded wide 95% CIs and limited our ability to draw strong conclusions from these data.
We chose to compare newer GLDs to TZD since less is known about the liver effects of these newer agents, whereas TZD have well-established benefits in NAFLD; including resolution of non-alcoholic steatohepatitis and improvement in fibrosis.8,21,22 GLP1RA and SGLT2i are associated with weight loss and preliminary evidence also suggests benefit in NAFLD,23–26 so we included these agents as comparisons. Additionally, we examined DPP4i since they are the most commonly prescribed branded second-line GLD,27 and there is conflicting evidence regarding their impact on NAFLD.13–16 Despite sulfonylureas being the most commonly prescribed second-line agent in T2D,27 they are substantially cheaper than newer agents, so were excluded given potential for socioeconomic confounding.
Impact of glucose-lowering drugs on cirrhosis appears to mirror NAFLD literature
Our findings on the impact of GLDs and cirrhosis risk follow similar patterns to those seen in NAFLD. For instance, TZD have the strongest evidence for benefit in NAFLD, and we likewise observed HR estimates consistently above 1 for newer GLD vs. TZD comparisons in as-treated, propensity score-weighted analyses. Both GLP1RA and SGLT2i have been shown to improve liver enzymes and reduce hepatic steatosis in patients with T2D and NAFLD.23,25,26,28 Few randomized controlled trials (RCTs) have compared these agents to TZD, and they position GLP1RA and SGLT2i as either similar or better than TZD at improving liver enzymes and histology.29–32 In our study, we observed slightly higher estimated hazards of incident cirrhosis among both GLP1RA and SGLT2i initiators compared to TZD initiators, although aHR estimates were reduced in the GLP1RA analysis when including outpatient and expanded cirrhosis diagnosis codes. Precision was also limited for our SGLT2i comparisons, with fewer years of SGLT2i data (2013–2015) likely contributing to this uncertainty.
DPP4i use was associated with higher estimated hazard of incident cirrhosis compared to TZD, particularly when using cirrhosis definitions with greater sensitivity (i.e., definitions 2 and 4). This trend may reflect DPP4i’s more neutral effect observed in NAFLD.12 Interestingly, serum dipeptidyl peptidase 4 (DPP4) levels are elevated in NASH, and are positively associated with histopathological grade and fibrosis. 33,34 A recent randomized study by Yan et al comparing the efficacy of liraglutide, sitagliptin or insulin glargine (on background metformin) for NAFLD found that add-on liraglutide and sitagliptin led to comparable reductions in liver fat, whereas insulin glargine did not.35 However, a dedicated RCT by Cui et al examining the effects of sitagliptin (vs placebo) on NAFLD in patients with T2D did not demonstrate improvement in hepatic fat content. 14
Gender and other subgroup differences
In subgroup analyses, we observed higher estimated aHRs for cirrhosis among women, in particular for the DPP4i vs. TZD comparison (Figure 2, primary definition, women, aHR 1.50 [95% CI 1.04, 2.16]; men; 0.86 [95% CI 0.59–1.25]). We observed similar trends for GLP1RA and SGLT2i, vs. TZD, though estimates were more imprecise in these comparisons. The reasons for this gender difference are unclear, although data on NAFLD suggest an important role of sex hormones, as men are at higher risk during reproductive years, while women are at higher risk after menopause.36 It is therefore feasible that the metabolically distinct livers of men and women interact differently with medications, including GLDs. Some studies have reported TZD to be more effective in women than in men, though data are mixed.37,38 Evidence does not suggest meaningful differences in efficacy of GLP1RA and SGLT2i by gender, 37–39 though women are more likely to experience side effects of these medications (e.g., nausea and vomiting associated with GLP1RA).37,38 It is plausible that more frequent examinations and testing in response to medication-related symptoms could lead to earlier discovery of cirrhosis (e.g., via abdominal imaging or endoscopy) in women compared to men. It is notable that sex differences also exist in other forms of CLD40 – how this is influenced by GLDs is largely unknown, and an interesting area for future research.
We additionally observed a higher estimated hazard of cirrhosis among initiators of GLP1RA vs. TZDs (aHR 1.82 [95% CI 1.09, 3.06]) in patients not on baseline insulin, whereas this was not the case for those on insulin. One potential explanation is that GLP1RA are often initiated in patients whose glycemic control has worsened such that they necessitate insulin, but who decline or prefer to delay recommended insulin therapy. Also, patients prescribed GLP1RA in the absence of insulin may have worse obesity. In both cases, poor metabolic health in this subgroup may contribute to more rapid progression of liver disease. Conversely, patients with baseline insulin use, which is a surrogate measure of increased diabetes duration or severity, may have experienced higher baseline cirrhosis risk in both drug groups, which may have contributed to the attenuated aHR estimates in that subgroup. We did not observe these trends in the SGLT2i vs. TZD comparison, although estimates in that analysis were generally difficult to interpret due to low precision.
Implication of findings
Our findings suggest a possible trend towards greater benefit of TZD over newer GLDs in terms of cirrhosis risk. Since we included an older population of patients with T2D, it is likely that NAFLD accounted for a large proportion of CLD in this study. Thus, while our results reflect the impact of GLDs on all-cause cirrhosis, this may be largely driven by known effects in NAFLD, rather than impact on other forms of CLD (i.e. viral, alcoholic). One small study examining hepatic steatosis by magnetic resonance spectroscopy found that pioglitazone was able to reduce hepatic steatosis in individuals with human immunodeficiency virus and hepatitis C coinfection. 10 However, data examining use of pioglitazone and other GLDs in non-NAFLD CLD are limited, and further studies would be needed before TZD could be recommended above other GLDs for the purpose of reducing risk of progression to cirrhosis.
Strengths and Limitations
Ours is the first study to examine the impact of second-line GLDs on risk of incident cirrhosis in a large, nationally-representative population of older adult patients. The active comparator, new user design provides implicit control for confounding by indication among patients receiving similar-line GLDs, and the restriction to patients with ≥2 prescriptions of a study drug increases confidence that patients are continuously taking those drugs. Finally, as shown in standardized difference plots, propensity score weighting methods were successful in controlling for measured confounders.
There were important limitations to this study. First, longer drug exposure times may be necessary to detect significant differences between GLDs, since it can take many years to develop and diagnose cirrhosis. In our study, the relatively short (0.40–0.66 years) on-treatment times observed in the Medicare FFS population precluded our ability to establish a robust causal relationship between second-line GLD use and incidence of cirrhosis. Second, our study was not linked to the electronic health record, so we were unable to adjust for several relevant confounding factors, such as blood pressure, HbA1c, lipids, and importantly, body mass index. Notably, obesity is known to be sub-optimally captured in administrative claims data, with high specificity but poor sensitivity. We observed higher prevalence of obesity codes among new users of GLP1RA and SGLT2i than in TZD users; while reasonable covariate balance was achieved after propensity score weighting, we acknowledge the possibility for residual confounding. We conducted a sensitivity analysis to examine the impact of baseline obesity and found no meaningful difference in results, though this should be interpreted with caution since coding for obesity may be incomplete in clinical practice. Future head-to-head pragmatic randomized trials comparing liver effects of TZDs to other second-line GLDs may help to address issues of baseline unmeasured confounding.
Finally, when interpreting results of our study, it is important to consider the natural history of cirrhosis and outcome definitions used. Cirrhosis can take years to develop, and diagnoses often occur when patients transition from compensated to decompensated disease; the latter of which may necessitate hospitalization.41 Since we anticipated low counts for cirrhosis over our 8-year study period, we examined multiple validated coding definitions (Appendix Table 3). Our primary coding definition required ICD-9 codes to be placed in the hospital setting only, while our secondary definition included the same codes but allowed them to be placed in either inpatient or outpatient settings. Since most cases of compensated cirrhosis receive care in the primary care setting,42 our secondary definition had greater sensitivity for detecting cirrhosis and likely captured more individuals with both stable and decompensated disease, vs. only hospitalized patients with more severe disease or clinical decompensation (i.e., primary definition). Event rates for cirrhosis were expectedly higher with our secondary definition, and comparisons were better powered. Larger studies with higher event rates would be needed to provide additional insight into the impact of GLDs on cirrhosis risk when very specific (yet less sensitive) outcomes definitions are used.
Conclusion
In conclusion, we observed a possible trend towards lower risk of incident cirrhosis with TZD use vs. newer second-line GLDs, although some uncertainty remains due to imprecise estimates resulting from short durations of on-treatment follow-up in this population. Our results mirror preliminary data on the impact of GLDs on NAFLD, and these findings merit further study to better understand which GLDs should be prioritized for reducing cirrhosis risk among patients with type 2 diabetes.
Supplementary Material
Highlights.
Type 2 diabetes (T2D) accelerates progression of chronic liver disease to cirrhosis, yet the effects of most glucose-lowering drugs (GLDs) on cirrhosis risk in T2D are unknown.
After propensity score weighting, adjusted hazard ratio estimates for incident cirrhosis were mildly elevated among newer second-line GLD initiators vs. thiazolidinediones (TZD) (DPP4i vs. TZD: 1.15 [0.89–1.50]; GLP1RA vs. TZD: 1.34 [0.82–2.20]; SGLT2i vs. TZD: 1.16, [0.44–3.08]).
This is the first study examining the comparative effect of second-line glucose-lowering drugs on risk of incident cirrhosis in patients with T2D.
ACKNOWLEDGEMENTS
Pate V: VP receives salary support from R01 AG056479 (National Institute on Aging), R01 HL118255 (National Institutes of Health, NIH), and the National Center for Advancing Translational Sciences (NCATS, UL1TR002489), NIH.
Barritt AS: Dr. Barritt reports consulting fees from Target Pharmasolutions, Intercept, and Genfit, Inc.
Crowley MJ: Dr. Crowley receives support from the Durham Center of Innovation to Accelerate Discovery and Practice Transformation (ADAPT) (Health Services Research and Development grant CIN 13–410) at the Durham Veterans Affairs Health Care System.
Buse JB: Outside of the submitted work, Dr. Buse reports contracted consulting fees and associated travel support paid to his employer from Adocia, AstraZeneca, Dance Biopharm, Dexcom, Eli Lilly, Fractyl, GI Dynamics, Intarcia Therapeutics, Lexicon, MannKind, Metavention, NovaTarg, Novo Nordisk, Orexigen, PhaseBio, Sanofi, Senseonics, vTv Therapeutics, and Zafgen; grants and related travel support from AstraZeneca, Eli Lilly, Intarcia Therapeutics, Johnson & Johnson, Lexicon, Medtronic, Novo Nordisk, Sanofi, Theracos and vTv Therapeutics; personal fees from Cirius Therapeutics Inc, CSL Behring; stock or stock options from Mellitus Health, PhaseBio, Stability Health, and Pendulum Health; grants from the US National Institutes of Health (UL1TR002489, U01DK098246, UC4DK108612, U54DK118612, P30DK124723), PCORI and American Diabetes Association.
Stürmer T: TS receives investigator-initiated research funding and support as Principal Investigator (R01 AG056479) from the National Institute on Aging (NIA), and as Co-Investigator (R01 CA174453; R01 HL118255, R21-HD080214), National Institutes of Health (NIH). He also receives salary support as Director of Comparative Effectiveness Research (CER), NC TraCS Institute, UNC Clinical and Translational Science Award (UL1TR002489), the Center for Pharmacoepidemiology (current members: GlaxoSmithKline, UCB BioSciences, Merck, Takeda), from pharmaceutical companies (Amgen, AstraZeneca, Novo Nordisk), and from a generous contribution from Dr. Nancy A. Dreyer to the Department of Epidemiology, University of North Carolina at Chapel Hill. Dr. Stürmer does not accept personal compensation of any kind from any pharmaceutical company. He owns stock in Novartis, Roche, BASF, AstraZeneca, and Novo Nordisk.
The database infrastructure used for this project was funded by the Department of Epidemiology, UNC Gillings School of Global Public Health; the Cecil G. Sheps Center for Health Services Research, UNC; the CER Strategic Initiative of UNC’s Clinical Translational Science Award (UL1TR002489); and the UNC School of Medicine.
Guarantors’ statement: JY Yang and AS Alexopoulos served as the guarantors for the study.
Grant Support:
This research was supported, in part, by grants from the National Institutes of Health R01 AG056479 (TS, VP), T32 DK007634 (JYY, AMM, HK), T32DK007012 (ASA), UL1TR002489 (JB, TS, VP), and by the University of North Carolina Royster Society of Fellows (JYY).
Footnotes
Disclosures:
Yang JY: No conflicts to disclose
Alexopoulos AS: No conflicts to disclose
Moon A: No conflicts to disclose
Kim H: No conflicts to disclose
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
REFERENCES
- 1.Beste LA, Leipertz SL, Green PK, Dominitz JA, Ross D, Ioannou GN. Trends in burden of cirrhosis and hepatocellular carcinoma by underlying liver disease in US veterans, 2001–2013. Gastroenterology. 2015;149(6):1471–1482 e1475; quiz e1417–1478. [DOI] [PubMed] [Google Scholar]
- 2.Moon AM, Singal AG, Tapper EB. Contemporary Epidemiology of Chronic Liver Disease and Cirrhosis. Clin Gastroenterol Hepatol. 2019. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Bril F, Cusi K. Nonalcoholic Fatty Liver Disease: The New Complication of Type 2 Diabetes Mellitus. Endocrinol Metab Clin North Am. 2016;45(4):765–781. [DOI] [PubMed] [Google Scholar]
- 4.El-Serag HB, Tran T, Everhart JE. Diabetes increases the risk of chronic liver disease and hepatocellular carcinoma. Gastroenterology. 2004;126(2):460–468. [DOI] [PubMed] [Google Scholar]
- 5.Huo TI, Wu JC, Lee PC, Tsay SH, Chang FY, Lee SD. Diabetes mellitus as a risk factor of liver cirrhosis in patients with chronic hepatitis B virus infection. J Clin Gastroenterol. 2000;30(3):250–254. [DOI] [PubMed] [Google Scholar]
- 6.Li X, Gao Y, Xu H, Hou J, Gao P. Diabetes mellitus is a significant risk factor for the development of liver cirrhosis in chronic hepatitis C patients. Sci Rep. 2017;7(1):9087. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Raff EJ, Kakati D, Bloomer JR, Shoreibah M, Rasheed K, Singal AK. Diabetes Mellitus Predicts Occurrence of Cirrhosis and Hepatocellular Cancer in Alcoholic Liver and Non-alcoholic Fatty Liver Diseases. J Clin Transl Hepatol. 2015;3(1):9–16. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Musso G, Cassader M, Paschetta E, Gambino R. Thiazolidinediones and Advanced Liver Fibrosis in Nonalcoholic Steatohepatitis: A Meta-analysis. JAMA Intern Med. 2017;177(5):633–640. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Khan R, Bril F, Cusi K, Newsome PN. Modulation of Insulin Resistance in NAFLD. Hepatology. 2018. [DOI] [PubMed] [Google Scholar]
- 10.Matthews L, Kleiner DE, Chairez C, et al. Pioglitazone for Hepatic Steatosis in HIV/Hepatitis C Virus Coinfection. AIDS Res Hum Retroviruses. 2015;31(10):961–966. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Kahl S, Gancheva S, Strassburger K, et al. Empagliflozin Effectively Lowers Liver Fat Content in Well-Controlled Type 2 Diabetes: A Randomized, Double-Blind, Phase 4, Placebo-Controlled Trial. Diabetes Care. 2020;43(2):298–305. [DOI] [PubMed] [Google Scholar]
- 12.Ranjbar G, Mikhailidis DP, Sahebkar A. Effects of newer antidiabetic drugs on nonalcoholic fatty liver and steatohepatitis: Think out of the box! Metabolism. 2019;101:154001. [DOI] [PubMed] [Google Scholar]
- 13.Khan RS, Bril F, Cusi K, Newsome PN. Modulation of Insulin Resistance in Nonalcoholic Fatty Liver Disease. Hepatology. 2019;70(2):711–724. [DOI] [PubMed] [Google Scholar]
- 14.Cui J, Philo L, Nguyen P, et al. Sitagliptin vs. placebo for non-alcoholic fatty liver disease: A randomized controlled trial. J Hepatol. 2016;65(2):369–376. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Sumida Y, Seko Y, Yoneda M, Japan Study Group of N. Novel antidiabetic medications for non-alcoholic fatty liver disease with type 2 diabetes mellitus. Hepatol Res. 2017;47(4):266–280. [DOI] [PubMed] [Google Scholar]
- 16.Joy TR, McKenzie CA, Tirona RG, et al. Sitagliptin in patients with non-alcoholic steatohepatitis: A randomized, placebo-controlled trial. World J Gastroenterol. 2017;23(1):141–150. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Lund JL, Richardson DB, Sturmer T. The active comparator, new user study design in pharmacoepidemiology: historical foundations and contemporary application. Curr Epidemiol Rep. 2015;2(4):221–228. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Lapointe-Shaw L, Georgie F, Carlone D, et al. Identifying cirrhosis, decompensated cirrhosis and hepatocellular carcinoma in health administrative data: A validation study. PLoS One. 2018;13(8):e0201120. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Sato T, Matsuyama Y. Marginal structural models as a tool for standardization. Epidemiology. 2003;14(6):680–686. [DOI] [PubMed] [Google Scholar]
- 20.Sturmer T, Rothman KJ, Avorn J, Glynn RJ. Treatment effects in the presence of unmeasured confounding: dealing with observations in the tails of the propensity score distribution--a simulation study. Am J Epidemiol. 2010;172(7):843–854. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Chalasani N, Younossi Z, Lavine JE, et al. The diagnosis and management of nonalcoholic fatty liver disease: Practice guidance from the American Association for the Study of Liver Diseases. Hepatology. 2018;67(1):328–357. [DOI] [PubMed] [Google Scholar]
- 22.Cusi K, Orsak B, Bril F, et al. Long-Term Pioglitazone Treatment for Patients With Nonalcoholic Steatohepatitis and Prediabetes or Type 2 Diabetes Mellitus: A Randomized Trial. Ann Intern Med. 2016;165(5):305–315. [DOI] [PubMed] [Google Scholar]
- 23.Dong Y, Lv Q, Li S, et al. Efficacy and safety of glucagon-like peptide-1 receptor agonists in non-alcoholic fatty liver disease: A systematic review and meta-analysis. Clin Res Hepatol Gastroenterol. 2017;41(3):284–295. [DOI] [PubMed] [Google Scholar]
- 24.Bolinder J, Ljunggren O, Kullberg J, et al. Effects of dapagliflozin on body weight, total fat mass, and regional adipose tissue distribution in patients with type 2 diabetes mellitus with inadequate glycemic control on metformin. J Clin Endocrinol Metab. 2012;97(3):1020–1031. [DOI] [PubMed] [Google Scholar]
- 25.Li B, Wang Y, Ye Z, et al. Effects of Canagliflozin on Fatty Liver Indexes in Patients with Type 2 Diabetes: A Meta-analysis of Randomized Controlled Trials. J Pharm Pharm Sci. 2018;21(1):222–235. [DOI] [PubMed] [Google Scholar]
- 26.Cusi K Time to Include Nonalcoholic Steatohepatitis in the Management of Patients With Type 2 Diabetes. Diabetes Care. 2020;43(2):275–279. [DOI] [PubMed] [Google Scholar]
- 27.Montvida O, Shaw J, Atherton JJ, Stringer F, Paul SK. Long-term Trends in Antidiabetes Drug Usage in the U.S.: Real-world Evidence in Patients Newly Diagnosed With Type 2 Diabetes. Diabetes Care. 2018;41(1):69–78. [DOI] [PubMed] [Google Scholar]
- 28.Scheen AJ. Beneficial effects of SGLT2 inhibitors on fatty liver in type 2 diabetes: A common comorbidity associated with severe complications. Diabetes Metab. 2019;45(3):213–223. [DOI] [PubMed] [Google Scholar]
- 29.Bi Y, Zhang B, Xu W, et al. Effects of exenatide, insulin, and pioglitazone on liver fat content and body fat distributions in drug-naive subjects with type 2 diabetes. Acta Diabetol. 2014;51(5):865–873. [DOI] [PubMed] [Google Scholar]
- 30.Garcia Diaz E, Guagnozzi D, Gutierrez V, et al. Effect of incretin therapies compared to pioglitazone and gliclazide in non-alcoholic fatty liver disease in diabetic patients not controlled on metformin alone: An observational, pilot study. Endocrinol Nutr. 2016;63(5):194–201. [DOI] [PubMed] [Google Scholar]
- 31.Ohki T, Isogawa A, Iwamoto M, et al. The effectiveness of liraglutide in nonalcoholic fatty liver disease patients with type 2 diabetes mellitus compared to sitagliptin and pioglitazone. ScientificWorldJournal. 2012;2012:496453. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Ito D, Shimizu S, Inoue K, et al. Comparison of Ipragliflozin and Pioglitazone Effects on Nonalcoholic Fatty Liver Disease in Patients With Type 2 Diabetes: A Randomized, 24-Week, Open-Label, Active-Controlled Trial. Diabetes Care. 2017;40(10):1364–1372. [DOI] [PubMed] [Google Scholar]
- 33.Balaban YH, Korkusuz P, Simsek H, et al. Dipeptidyl peptidase IV (DDP IV) in NASH patients. Ann Hepatol. 2007;6(4):242–250. [PubMed] [Google Scholar]
- 34.Wang X, Hausding M, Weng SY, et al. Gliptins Suppress Inflammatory Macrophage Activation to Mitigate Inflammation, Fibrosis, Oxidative Stress, and Vascular Dysfunction in Models of Nonalcoholic Steatohepatitis and Liver Fibrosis. Antioxid Redox Signal. 2018;28(2):87–109. [DOI] [PubMed] [Google Scholar]
- 35.Yan J, Yao B, Kuang H, et al. Liraglutide, Sitagliptin, and Insulin Glargine Added to Metformin: The Effect on Body Weight and Intrahepatic Lipid in Patients With Type 2 Diabetes Mellitus and Nonalcoholic Fatty Liver Disease. Hepatology. 2019;69(6):2414–2426. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Lonardo A, Nascimbeni F, Ballestri S, et al. Sex Differences in Nonalcoholic Fatty Liver Disease: State of the Art and Identification of Research Gaps. Hepatology. 2019;70(4):1457–1469. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Arnetz L, Ekberg NR, Alvarsson M. Sex differences in type 2 diabetes: focus on disease course and outcomes. Diabetes Metab Syndr Obes. 2014;7:409–420. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Kautzky-Willer A, Harreiter J. Sex and gender differences in therapy of type 2 diabetes. Diabetes Res Clin Pract. 2017;131:230–241. [DOI] [PubMed] [Google Scholar]
- 39.Radholm K, Zhou Z, Clemens K, Neal B, Woodward M. Effects of sodium-glucose co-transporter-2 inhibitors in type 2 diabetes in women versus men. Diabetes Obes Metab. 2020;22(2):263–266. [DOI] [PubMed] [Google Scholar]
- 40.Guy J, Peters MG. Liver disease in women: the influence of gender on epidemiology, natural history, and patient outcomes. Gastroenterol Hepatol (N Y). 2013;9(10):633–639. [PMC free article] [PubMed] [Google Scholar]
- 41.Muir AJ. Understanding the Complexities of Cirrhosis. Clin Ther. 2015;37(8):1822–1836. [DOI] [PubMed] [Google Scholar]
- 42.Liu TL, Barritt AI, Weinberger M, Paul JE, Fried B, Trogdon JG. Who Treats Patients with Diabetes and Compensated Cirrhosis. PLoS One. 2016;11(10):e0165574. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.