Achieving comparability in glycemic control between antidiabetic treatment strategies in pregnancy when using real world data

Abstract

Purpose: Healthcare utilization databases often lack information on glycemic control, a key confounder when studying the safety of antidiabetic treatments, since patients with worse control are channeled to second-line agents, in particular insulin, versus first-line agents such as metformin. We evaluated whether adjustment for measured characteristics attains balance in glycemic control when comparing antidiabetic treatment strategies in pregnant women with pregestational type 2 diabetes (T2DM).

Methods: In a US insurance claims database, we identified 3360 women with T2DM pregnant between 2004 and 2015, of whom a subset of 996 had data on hemoglobin A_1c (HbA_1c ) levels. We selected insulin only as the comparator group and used propensity score (PS)-matching on comorbidities and proxies of diabetes severity, but not on HbA_1c , to adjust for confounding. We used standardized differences (st.diff) to assess balance in claims-based covariates and mean HbA_1c (% ± SD) in the subset.

Results: There were imbalances in claims-based covariates before PS-matching, with smaller differences when both treatment strategies included insulin. After PS-matching, balance was achieved in most claims-based covariates (st.diff <0.1). Mean HbA_1c was similar before and after PS-matching when both treatments included insulin (e.g., 7.1 ± 1.5 vs. 7.7 ± 1.8 and 7.1 ± 1.5 vs. 7.5 ± 1.7, respectively, for metformin + insulin vs. insulin only). Differences in mean HbA_1c remained after PS-matching when non-insulin treatments were compared to treatments including insulin (e.g., 6.3 ± 1.1 vs. 7.6 ± 1.7 for metformin only vs. insulin only).

Conclusions: Balance in both claims-based characteristics and glycemic control was attained after restricting the population to women with T2DM and comparing treatment strategies indicated for patients with similar diabetes severity. When comparing treatment strategies with versus without insulin, differences in glycemic control persisted after PS-matching even when balance was attained for other measured characteristics.

Keywords: antidiabetic medication; channeling bias; confounding; glycemic control; pregnancy; real world data; type 2 diabetes mellitus.

Figure 1:

Flowchart of study population derivation LMP: estimated start of pregnancy, date of last menstrual period

Figure 2:

Covariate balance before and after propensity score matching for selected baseline claims-based characteristics in the study population of pregnant women with type 2 diabetes comparing first trimester metformin treatment, alone or with insulin, versus insulin only * including nephropathy, retinopathy, neuropathy, diabetic foot, diabetes with other ophthalmic manifestations or retinal edema, chronic kidney disease (Stage III or higher), other renal conditions. ^** including diabetes with other specified or unspecified manifestations Covariate assessment periods: Healthcare utilization: LMP-180 to LMP-1 days; Lab test ordered, comorbidities, medications: LMP-180 to LMP+90 days. Abbreviations: ACE, angiotensin converting enzyme; St.diff, standardized difference; LMP, last menstrual period; PS, propensity score. PS-matching results for the complete list of covariates presented in Supplementary Table S7 and S8.

Figure 3:

Covariate balance before and after propensity score matching for selected baseline claims-based characteristics in the study population of pregnant women with type 2 diabetes with pre-pregnancy metformin only treatment comparing second trimester metformin use, alone or with insulin, versus insulin only. Note^: The post-matching study population contained no individuals with diabetic ketoacidosis in the first comparison, and no individuals with diabetic ketoacidosis or hypoglycemia in the second comparison. * including nephropathy, retinopathy, neuropathy, diabetic foot, diabetes with other ophthalmic manifestations or retinal edema, chronic kidney disease (Stage III or higher), other renal conditions. ^** including diabetes with other specified or unspecified manifestations Covariate assessment periods: Healthcare utilization: LMP-180 to LMP-1 days; Lab test ordered, comorbidities, medications: LMP-180 to LMP+90 days. Abbreviations: ACE, angiotensin converting enzyme; St.diff, standardized difference; LMP, last menstrual period; PS, propensity score. PS-matching results for the complete list of covariates presented in Supplementary Table S13 and S14.