Comparative Effectiveness of Second-Line Agents for the Treatment of Diabetes Type 2 in Preventing Kidney Function Decline

Abstract

Background and objectives 

Diabetes is the leading cause of ESRD. Glucose control improves kidney outcomes. Most patients eventually require treatment intensification with second-line medications; however, the differential effects of those therapies on kidney function are unknown.

Design, setting, participants & measurements 

We studied a retrospective cohort of veterans on metformin monotherapy from 2001 to 2008 who added either insulin or sulfonylurea and were followed through September of 2011. We used propensity score matching 1:4 for those who intensified with insulin versus sulfonylurea, respectively. The primary composite outcome was persistent decline in eGFR≥35% from baseline (GFR event) or a diagnosis of ESRD. The secondary outcome was a GFR event, ESRD, or death. Outcome risks were compared using marginal structural models to account for time-varying covariates. The primary analysis required persistence with the intensified regimen. An effect modification of baseline eGFR and the intervention on both outcomes was evaluated.

Results 

There were 1989 patients on metformin and insulin and 7956 patients on metformin and sulfonylurea. Median patient age was 60 years old (interquartile range, 54–67), median hemoglobin A1c was 8.1% (interquartile range, 7.1%–9.9%), and median creatinine was 1.0 mg/dl (interquartile range, 0.9–1.1). The rate of GFR event or ESRD (primary outcome) was 31 versus 26 per 1000 person-years for those who added insulin versus sulfonylureas, respectively (adjusted hazard ratio, 1.27; 95% confidence interval, 0.99 to 1.63). The rate of GFR event, ESRD, or death was 64 versus 49 per 1000 person-years, respectively (adjusted hazard ratio, 1.33; 95% confidence interval, 1.11 to 1.59). Tests for a therapy by baseline eGFR interaction for both the primary and secondary outcomes were not significant (P=0.39 and P=0.12, respectively).

Conclusions 

Among patients who intensified metformin monotherapy, the addition of insulin compared with a sulfonylurea was not associated with a higher rate of kidney outcomes but was associated with a higher rate of the composite outcome that included death. These risks were not modified by baseline eGFR.

Introduction

One of every three persons with diabetes also has CKD (¹). Diabetes accounts for 45% of ESRD in the United States. The coexistence of CKD and diabetes significantly increases the risk of premature death (²,³). Hence, preventing the development of CKD and progression to ESRD is a high priority to reduce morbidity and mortality in patients with diabetes as well as the burden that diabetes–related ESRD care inflicts on the health care system.

Most randomized, controlled clinical trials of kidney outcomes in diabetes were designed to assess the effects of achieving various glycemic targets on proximal renal outcomes, such as microalbuminuria (^4–10). Recently, the 6-year follow-up to the Action in Diabetes and Vascular Disease: Preterax and Diamicron Modified Release Controlled Evaluation Study showed the benefit of tighter glucose control in type 2 diabetes on distal renal outcomes, such as ESRD (¹¹). Nevertheless, few studies have evaluated the effects of specific antidiabetic drug regimens on renal outcomes (¹²). Our recent comparative effectiveness study demonstrated that metformin had renoprotection when compared to sulfonylurea monotherpy (¹²). The American Diabetes Association recommends that treatment should begin with metformin and lifestyle changes to achieve a hemoglobin A1c (HbA1c) of <7%. However, many patients eventually require a second-line agent to reach or maintain this goal. Currently, a knowledge gap exists with regards to the comparative effectiveness of insulin versus sulfonylureas as second-line therapy (add-on therapy) in kidney function decline (¹³). Adding sulfonylureas and insulin are two of the most commonly used intensification regimens at present (¹⁴).

A recent study by our group reported that intensification of metformin monotherapy with insulin did not increase the risk of nonfatal cardiovascular (CV) events or CV death (¹⁵). However, it increased the risk of all-cause mortality, a finding that requires additional investigation (¹⁵).

In this study, we compared the time to a renal composite outcome of persistent decline in eGFR of 35% from the baseline (GFR event) or reaching ESRD and the time to a composite of GFR event, ESRD, or death in patients who intensified metformin monotherapy by adding either insulin or a sulfonylurea. We also evaluate if the effect of the intensification regimens on both outcomes is modified and varies by baseline eGFR.

Materials and Methods

Study Design and Data Sources

We assembled a retrospective cohort of patients who initiated metformin monotherapy for diabetes using national Veterans Health Administration (VHA) databases (¹²). Patients initiating metformin monotherapy (new users) between October 1, 2001 and September 30, 2008 were followed through September 30, 2011. We used the VHA databases to track prescriptions dispensed by the VHA (¹⁶), outpatient Internal Classification of Diseases Ninth Revision Clinical Modification codes, procedure codes, laboratory values, and demographics (¹⁷). Supplemental information was obtained from the Centers for Medicare and Medicaid Services, including prescription part D data (¹⁸,¹⁹). We obtained dates of death from the National Death Index (NDI) through September 30, 2009 and VHA vital status files (through 2011) (²⁰). The Institutional Review Boards of Vanderbilt University and the VHA Tennessee Valley Healthcare System approved this study.

Study Population

The study population was veterans ≥18 years old who received regular VHA care (VHA encounter or prescription fill at least once every 180 days) for at least 2 years before enrollment. New users of metformin from October of 2001 to September of 2008 did not use any antidiabetic drugs preceding their first metformin prescription fill (²¹). These metformin initiators became eligible for the treatment intensification cohort on the date that they filled a prescription for insulin or a sulfonylurea (²²). We excluded patients missing baseline serum creatinine measurements and also excluded patients receiving hospice care.

Exposures

The two exposures of interest were metformin and insulin (short/fast, long acting, or mixed insulin) and metformin and sulfonylurea (glyburide, glipizide, or glimepiride). Follow-up began 180 days after intensification to prevent inclusion of patients who were switching rather than intensifying their metformin regimens (²²). Follow-up continued until a study outcome (below), death, or the end of the study (September 30, 2011). Patients could also be censored for loss to follow-up (defined as noncontact with any VHA facility [inpatient, outpatient, or pharmacy use] for 181 days), nonpersistence (defined as >90 days without metformin therapy; i.e., persistent exposure required), or prescription for a third antidiabetic drug.

Outcomes

The primary renal composite outcome encompassed a GFR event and reaching ESRD. A GFR event was defined as a persistent decline in eGFR of ≥35% from baseline (²³). ESRD was defined by a procedure or diagnosis code indicating dialysis, renal transplant, or an eGFR<15 ml/min (Supplemental Table 1). An event required a second confirmatory event, except for renal transplant (i.e., another eGFR measurement indicating a ≥35% decline from baseline or dialysis, a dialysis code, or eGFR<15 ml/min), within the next 3–12 months. The event date was the date of the confirmatory event. This was done to prevent capturing AKI events as CKD progression.

eGFRs were calculated from serum creatinine measurements and demographic characteristics using the Modification of Diet in Renal Disease equation for isotope dilution mass spectrometry–calibrated creatinine (²⁴,²⁵).

The secondary renal composite outcome encompassed a GFR event, reaching ESRD, and all-cause mortality. Date of death was obtained from the Veterans Affairs (VA) Vital Status Master File, which has been shown to be highly accurate compared with the NDI (²⁶).

We also evaluate the interaction of baseline eGFR (kidney function at cohort entry) and the effect of the intensification regimen for the primary and secondary composite outcomes. A stratified analysis was conducted in three eGFR subgroups: eGFR<60, 60–80, and >80 ml/min per 1.73 m²

Covariates

Study covariates were measured during the 730 days before the time of therapy intensification and as time-varying covariates ascertained monthly after that, and they included age; sex; race; fiscal year; physiologic variables (BP, creatinine, HbA1c, LDL levels, and body mass index; baseline value was the measurement closest to the time of intensification); proteinuria categorized on the basis of urine dipstick reading as negative, mild (trace or 1+), or heavy (≥2+) (²⁷); indicators of health care utilization (number of outpatient visits, hospitalization during baseline, months from hospitalization to intensification, and nursing home use); smoking; use of medications known to affect creatinine values (angiotensin–converting enzyme inhibitor, angiotensin receptor blocker, and loop and thiazide diuretics); the presence of other comorbidities; and the location of care (Veterans Integrated Service Network [VISN]) (Supplemental Table 2).

For patients with missing covariates, we conducted a multiple imputation procedure using the Markov Chain Monte Carlo method and a noninformative Jeffrey prior (²⁸). All covariates from the primary analysis, survival time, and a censoring indicator were included in 20 imputation models and used to compute the final estimates.

Statistical Analyses

For the analyses, we first assembled propensity score–matched cohorts. The propensity score modeled the probability of metformin and insulin use given baseline covariates at the time of intensification and the VISN as location of care. Because of differences in the size of the two exposure groups (Supplemental Figure 1), metformin and insulin observations were propensity score matched to metformin and sulfonylurea observations using a 1:4 ratio Greedy Matching Algorithm (²⁹,³⁰).

Marginal structural Cox proportional hazards regression models (MSMs) were used to compare outcomes between the metformin and insulin and the metformin and sulfonylurea matched cohorts (³¹). MSMs estimate the marginal treatment effects on a population through weighting and reweighting observations, while controlling for baseline and time-varying covariates. Potential confounders were controlled through the inverse probability of treatment weights, using a patient’s covariate history to estimate the probability of intensifying therapy with metformin and insulin versus metformin and sulfonylurea (Supplemental Table 3) and the probability of being censored after follow-up begins. To adjust for potential informative censoring, the probability of death was modeled separately from all other censoring criteria (Supplemental Table 4). Because MSM estimates can be unstable in the presence of disproportionately large treatment weights, we used the stabilized inverse probability of treatment weights (³¹) and truncated weights at five, i.e., changed weights >5 to =5.

Sensitivity and Subgroup Analyses

We conducted sensitivity analyses to evaluate the robustness of our main findings to variations in study definitions and analytic assumptions. In an approach similar to the intention to treat analyses from clinical trials, we used the intensification regimen (the addition of either insulin or sulfonylurea to metformin) to define the exposure and ignored subsequent changes in regimens (i.e., persistent exposure not required). Thus, these analyses allow exposure misclassification caused by nonadherence and crossover between exposures. We evaluated the degree of crossover between exposure groups in an attempt to quantify part of the misclassification that occurred (Supplemental Figure 2).

Because death would preclude the occurrence of study outcomes, we performed a separate competing risk analysis to compare the hazard of study outcomes, considering death as a competing event. We also performed our analyses limiting to the first two years of follow-up. This sensitivity analysis was done to test if the observed results were driven by differences in the later years, when the sample size was relatively small. Finally, to assess for potential effect modification, we conducted subgroup analyses with stratification by age (<65 or ≥65 years old), race (black versus nonblack), baseline proteinuria (any versus none), and HbA1c (≤8 or >8) (Supplemental Figure 3).

Analyses were conducted using R, version X64 2.12.1 (R Foundation for Statistical Computing, Vienna, Austria) and SAS software, version 9.2 (SAS Institute Inc., Cary, NC).

Results

Study Population

There were 178,341 patients who initiated metformin monotherapy during the study years. Fifty-two percent (n=92,045) never intensified therapy and had a median follow-up of 50 months (interquartile range [IQR], 19–67); 6% (n=9851) were nonadherent to metformin monotherapy, and 2% (n=3577) had <6 months of follow-up. Among the remaining 72,868 metformin initiators who intensified therapy, 41% (n=29,523) were excluded, because metformin was discontinued or their regimen included nonstudy medications (thiazolidinediones, dipeptidylpeptide-4 inhibitors, or glucagon like peptide-1 agonist analogs). Of the remaining 59% who intensified (n=43,345), we excluded 9% (n=6481) with no creatinine measurement before intensification and 1% (n=407) who had codes indicating hospice care or ESRD before intensification. Participant selection (n=29,523) in a flow chart in Figure 1. Thus, the final cohort before matching included 36,405 individuals, 2393 (7%) who intensified their metformin monotherapy with insulin and 34,012 (93%) who intensified with a sulfonylurea. The cohort characteristics before matching are presented in Supplemental Table 5.

Figure 1.: Study flow chart.

After propensity score matching, our study cohort included 1989 metformin and insulin users and 7956 metformin and sulfonylurea users. Baseline characteristics were similar, and standardized differences were small both before (Supplemental Table 5) and after propensity score matching (Table 1). Overall, patients were 94% men and 71% white, median age was 60 years old (IQR, 54–67), creatinine was 1.0 mg/dl (IQR, 0.9–1.11), and HbA1c 8.1% (IQR, 7.1%–9.9%). Seventy-four percent (n=7372) had a urine dipstick measurement available, and 29% of these had documented proteinuria. The median follow-up after intensification was 1.1 years (IQR, 0.5–2.3) in the metformin and insulin group and 1.1 years (IQR, 0.5–2.2) in the metformin and sulfonylurea group. Patient characteristics at 12 months after intensification were not statistically different (Supplemental Table 6).

Table 1. - Characteristics of patients at the time of add-on therapy

	Propensity-Matched Cohort		Standardized Difference
	Metformin and Sulfonylurea, n=7956	Metformin and Insulin, n=1989
Characteristics
Age, yr, median (IQR)	60 (54–67)	60 (54–67)	0.01
Men, %	94	95	0.04
Race, %
White	71	70	0.01
Black	16	17	0.02
Other/unknown	4	5	0.02
Clinical measures
eGFR, ml/min per 1.73 m2, median (IQR)	83 (69–99)	82 (68–98)	0.00
Urine protein measurement available, no. (%)	5897 (74)	1475 (74)	0.00
Missing measurements, no. (%)	2059 (26)	514 (26)	0.00
Negative	4181 (71)	1049 (74)
Mild to moderate	1420 (24)	358 (24)
Moderate to severe	296 (5)	68 (5)
Creatinine, mg/dl, median (IQR)	1.0 (0.9–1.1)	1.0 (0.9–1.17)	0.02
Systolic BP, mmHg, median (IQR)	131 (120–142)	131 (120–142)	0.00
Diastolic BP, mmHg, median (IQR)	76 (70–83)	76 (69–84)	0.00
Missing measurements, no. (%)	167 (2)	41 (2)
HbA1c, %, median (IQR)	8.1 (7.1–9.8)	8.2 (6.9–10.0)	0.00
Missing measurements, no.	633	152
Body mass index, kg/m2, median (IQR)	33 (29–37)	33 (29–37)	0.02
Missing measurements, no. (%)	258 (3)	62 (3)
Baseline use of medications
ACEI or ARBs, %	71	72
Loop diuretics, %	17	16	0.03
Statins, %	77	78	0.02
Antihypertensive drugs, %	73	73	0.00
Months to additional therapy, median (IQR)	18 (8–34)	17 (7–34)	0.02
Baseline comorbidities, %
Cardiovascular disease	34	35	0.00
Malignancy	10	10	0.01
Liver or respiratory failure	6	5	0.02
Congestive heart failure	9	9	0.00
HIV	1	1	0.03
Arrhythmias	11	11	0.01
Smoking	23	23	0.00
COPD/asthma	21	21	0.00
Indicators of health utilization
Hospitalized in the prior year, %	29	28	0.03
Outpatients visit in the last year, % median (IQR)	7 (4–12)	7 (4–12)	0.02
Nursing home utilization in the prior year, %	0.2	0.2	0.00
Calendar year, %			0.03
2002	0	0
2003	3	3
2004	6	6
2005	9	10
2006	14	14
2007	16	16
2008	19	19
2009	17	16
2010	13	12
2011	3	3

IQR, interquartile range; HbA1c, hemoglobin A1c; ACEI, angiotensin–converting enzyme inhibitor; ARB, angiotensin receptor blocker; COPD, chronic obstructive pulmonary disease.

Risk for the Primary and Secondary Outcomes for the Overall Cohort and Stratified by eGFR Group at Cohort Entry

For the primary renal composite outcome, there were 98 events in patients who added insulin and 328 events in patients who added a sulfonylurea (Table 2). Almost all of the renal composite outcomes were caused by a clinically significant decline in eGFR of 35% (97 of 98 in insulin intensifiers and 324 of 328 in sulfonylurea intensifiers). There were 31 events per 1000 person-years of metformin and insulin use versus 26 events per 1000 person-years of metformin and sulfonylurea use (adjusted hazard ratio [aHR], 1.27; 95% confidence interval [95% CI], 0.99 to 1.63). Unadjusted cumulative incidence curves for the outcome in the propensity score–matched cohorts are shown in Figure 2. Figure 3 depicts the eGFR values over time for both treatment regimens after intensification at the median values for the 75th and 25th percentiles. The eGFR values overlap for both groups until year 4, which supports the lack of significance for the hazard ratio for the primary analysis. During later years, the number of remaining participants was small, and the estimates were less precise.

Table 2. - Event rates and adjusted hazard ratios for the primary and secondary outcomes among those who intensify with insulin versus those who intensify with sulfonylurea

	Metformin and Sulfonylurea, n=7956	Metformin and Insulin, n=1989
Persistent exposure required a
Person-yr	12,478	3176
Primary renal composite b	328	98
eGFR events: ≥35% decline	324	97
ESRD	4	1
Unadjusted rate per 1000 person-yr	26	31
Adjusted hazard ratio c (95% CI)	Reference	1.27 (0.99 to 1.63)
Secondary renal composite: GFR, ESRD, or death events	611	203
Death events	283	105
Unadjusted rate per 1000 person-yr	49	64
Adjusted hazard ratio c (95% CI)	Reference	1.33 (1.11 to 1.59)
Persistent exposure not required d
Person-yr	22,792	5577
Primary renal composite b	586	178
eGFR events: ≥35% decline	577	173
ESRD	9	5
Unadjusted rate per 1000 person-yr	26	32
Adjusted hazard ratio c (95% CI)	Reference	1.22 (1.02 to 1.46)
Secondary renal composite: GFR, ESRD, or death events	1246	403
Death events	660	225
Unadjusted rate per 1000 person-yr	55	72
Adjusted hazard ratio c (95% CI)	Reference	1.38 (1.17 to 1.63)
Competing risk analysis
Adjusted hazard ratio (95% CI)	Reference	1.20 (1.01 to 1.43)

95% CI, 95% confidence interval.

^aPrimary analysis requires persistence on metformin; patients are censored after 90 days without either medication.

^bRenal event: eGFR decline of 35% or ESRD confirmed with a second event in 3–12 months.

^cAdjusted hazard ratio is derived from the marginal structural model for the time to outcome truncated weights at five, i.e., changed weights >5 to =5. Refer to Supplemental Table 3 for the probability of intensifying with metformin and insulin and Supplemental Table 4 for the probability of censoring or death during the follow-up.

^dPatients remain in their exposure group, regardless of persistence on drug therapy, until outcome or end of the study.

Figure 2.: Crude cumulative incidence of GFR events or ESRD by drug combination groups. (A) Propensity score–matched cohort with persistent exposure required in the intensified regimen. (B) Propensity score matched with persistent exposure not required: patients remain in their exposure group, regardless of persistence with the intensified regimen. met-ins, Metformin and insulin; met-sul, metformin and sulfonylurea.

Figure 3.: Plots of eGFR over time for both treatment regimens after intensification. Met+Ins, metformin and insulin; Met+Sul, metformin and sulfonylurea.

For the secondary composite outcome of renal events or death, there were 203 events (105 deaths) in insulin intensifiers and 611 events (283 deaths) in sulfonylurea intensifiers (Table 2). There were 64 events per 1000 person-years of metformin and insulin use versus 49 events per 1000 person-years of metformin and sulfonylurea use (aHR, 1.33; 95% CI, 1.11 to 1.59).

We performed a stratified analysis on the basis of baseline eGFR at cohort entry to evaluate the interaction of baseline eGFR and the association of intensification regimen with outcomes for both the primary and secondary outcomes. The eGFR subgroups were eGFR<60, 60–80, and >80 ml/min (Figure 4). We did not observe significant differences in the association of the intensification regimen with outcomes by specific eGFR subgroups. Although the eGFR<60 subgroup’s hazard ratios suggested metformin and insulin as protective for both the primary and secondary outcomes (aHR, 0.49; 95% CI, 0.16 to 1.48 and aHR, 0.85; 95% CI, 0.48 to 1.53, respectively), the 95% CIs were wide, and the effects were not statistically significant. A test for a therapy by eGFR group interaction was also not significant for the primary outcome (P=0.39) or the secondary outcome (P=0.12).

Figure 4.: Adjusted hazard ratios for the primary composite outcome of GFR event or ESRD and the secondary outcome of GFR event, ESRD, or death among patients according to their eGFR baseline strata. Hazard ratios greater than one show an increased risk for composite outcome with the combination metformin plus insulin (Met+Ins) compared with the combination metformin plus sulfonylurea (Met+Sul) in the 1:4 match cohort. Adjusted hazards were derived using Cox proportional hazard marginal structural model for time to the composite renal outcome, truncated weights at five, i.e., changed weights >5 to =5. Models were adjusted for age; sex; race; fiscal year; physiologic variables (BP, creatinine, hemoglobin A1c, LDL levels, and body mass index; baseline value was the measurement closest to the time of intensification); proteinuria categorized on the basis of urine dipstick reading as negative, mild (trace or 1+), or heavy (≥2+) (²⁷); indicators of health care utilization (number of outpatient visits, hospitalization during baseline, months from hospitalization to intensification, and nursing home use); smoking; use of medications known to affect creatinine values (angiotensin–converting enzyme inhibitors, angiotensin receptor blockers, and loop and thiazide diuretics); the presence of other comorbidities; and the location of care (Veterans Integrated Service Network). All continuous variables were modeled as third degree polynomials.

Sensitivity Analyses and Other Subgroup Analyses

Our first sensitivity analysis allowed each individual to remain in his/her original exposure group throughout follow-up, regardless of regimen adherence. For the primary composite outcome of a GFR event (35% decline in eGFR) or ESRD, there were 178 and 586 renal composite events in patients who added on insulin versus a sulfonylurea, respectively, yielding event rates of 32 per 1000 person-years of insulin use versus 26 per 1000 person-years of sulfonylurea use (aHR, 1.22; 95% CI, 1.02 to 1.46) (Table 2). For the secondary composite outcome of a GFR event (35% decline in eGFR), ESRD, or death, there were 403 and 1246 composite events in patients who added on insulin versus a sulfonylurea, respectively, yielding event rates of 72 per 1000 person-years of metformin and insulin use versus 55 per 1000 person-years of metformin and sulfonylurea use (aHR, 1.38; 95% CI, 1.17 to 1.63).

We quantified exposure misclassification caused by crossover in this sensitivity analysis by measuring the percentage of person-time that did not correspond to the original study regimen. During study follow-up, 15% in the metformin and sulfonylurea group received insulin at any time, whereas only 7% in the metformin and insulin group received sulfonylurea at any time (Supplemental Figure 2).

Our sensitivity analyses included a competing risk analysis for GFR event or ESRD with death as the competing event (aHR, 1.20; 95% CI, 1.01 to 1.43). We also conducted an analysis limited to the first 2 years of follow-up, which yielded results consistent with the main analyses for both the primary and secondary outcomes (aHR, 1.25; 95% CI, 0.97 to 1.62; P=0.09 and aHR, 1.43; 95% CI, 1.19 to 1.73; P<0.001, respectively). This analysis supports the robustness of the main analysis performed over all of the years of follow-up. Finally, stratified analyses by subgroups defined a priori did not identify significant interactions by age, race, HbA1c, or the presence of proteinuria. Supplemental Figure 3 shows the forest plot with the hazard ratios for the subgroup analyses.

Discussion

We compared the risk of renal composite outcomes between patients intensifying metformin monotherapy with insulin and patients intensifying metformin monotherapy with sulfonylurea. Intensifying with sulfonylurea and intensifying with insulin are the two most common patterns of intensification in the VA system (¹⁴). The risk of renal and death events was higher among patients who added insulin compared with those who added sulfonylurea. The difference in renal events was not statistically significant in the main analysis, which required persistence on the intensified drug regimen. However, there was an increased risk for the composite outcome that included death as well as renal events.

Multiple clinical trials have shown that tight glycemic control improves microvascular outcomes (⁴,⁸,⁹). For diabetic nephropathy, this beneficial effect has been shown for both proximal outcomes, such as incidence and worsening of proteinuria (⁴,⁸,⁹), and more recently, distal outcomes, such as incident CKD in type 1 diabetes (⁵) and progression to ESRD in type 2 diabetes (¹¹,³²). Although these studies evaluated the effect of glycemic control on renal outcomes, none compared specific drug regimens (⁶,⁷,¹⁰). In contrast to the question posed by these trials of whether different glycemic targets were preferred, our study addressed outcome differences between specific drug regimens. In our study cohort, glycemic control was similar between groups at 12 (Supplemental Table 6) and 36 months (Supplemental Table 7).

In our sensitivity analysis, in which persistent exposure was not required, patients were analyzed as remaining in their original regimen at cohort entry. Although this sensitivity analysis added considerably more follow-up time and renal events, it has inherent exposure misclassification by design, because it tolerates nonadherence and crossover (Supplemental Figure 2). Nevertheless, the results were similar with those from the primary analysis, which required stricter adherence to the study regimens.

Our group recently reported that intensification of metformin monotherapy with insulin compared with sulfonylurea did not improve nonfatal CV events or CV death rates; however, it was associated with a 44% increase in the risk of all-cause mortality (¹⁵). Our study suggests a higher risk of mortality in patients on metformin and insulin that is not related to CKD events. These observations are highly relevant given that the four most common intensification patterns within the VA system are the addition of sulfonylurea (79%), the addition of insulin (8%), the addition of thiazolidinedione (6%), or a switch to insulin monotherapy (2%) (¹⁴).

Our comparative effectiveness study has noteworthy strengths. We used propensity score matching to balance differences in baseline characteristics and comorbidities between exposure groups. We applied a new user design for the assessment of intensification regimens (³³). We applied strict criteria to minimize misclassification of exposures, outcomes, and covariates. Our analyses accounted for available laboratory and physiologic measurements that complemented administrative data, reducing concerns about residual confounding. Additionally, we used MSMs to account for baseline and time-varying confounders. This approach also controlled for differential censoring (e.g., change to a new regimen) caused by potential differences in glycemic control between the study regimens.

Our study has limitations. First, our cohort consisted mainly of veterans who were men, and findings should be generalized to other populations with caution. Second, although the VHA is a closed system, there is the potential for veterans to seek care outside the system. To minimize this concern, Medicare, Medicaid, and NDI data were merged with the VHA files, and we required patients to have evidence of continuing care in the VA system during follow-up. Third, refill data were used as a proxy for medication taking, and this may result in misclassification of exposures. However, this approach is virtually free of recall issues and has good concordance with self-reported medication use (¹²,³⁴). Although we used a rigorous design and MSM analyses to minimize potential differences between exposure groups, this observational study is still vulnerable to potential residual confounding. However, planned sensitivity and subgroup analyses yielded similar results to our primary analyses. Finally, our study applied a new user design for a second-line agent. Although this approach reduces confounders and minimizes bias, increasing the study’s internal validity, its application led to a relatively small sample size and reduced precision, which can be appreciated in the width of the 95% CIs of our estimates. A sensitivity analysis limiting the analysis to the first 2 years of the study produced similar results as the primary analysis conducted over all of the years of follow-up for both the primary and secondary outcomes, supporting the robustness of the study findings.

In summary, our study shows that, in patients on metformin monotherapy, the addition of insulin was not associated with a significantly higher rate of renal events compared with the addition of a sulfonylurea. However, we did find a higher rate of the composite outcome that included death as well as renal events. The association of the intensification regimen with both the primary and secondary outcomes was not modified by the baseline eGFR at the time of cohort entry. This is concordant with results of our previous study, in which we reported an increased risk of all-cause mortality when intensifying with insulin (¹⁵). These studies suggest that early intensification of metformin monotherapy with insulin should not be a preferred strategy. Larger studies, ideally pragmatic trials, conducted in the general population would be useful to confirm these observations.

Disclosures

None.