OBJECTIVES : Measuring adherence to flexible dose therapies such as insulin with secondary databases is challenging. Due to the inaccuracies of recorded days’ supply when using medication possession ratio (MPR), adjusted MPR (aMPR) has been employed in several insulin adherence studies. This study compared adherence among intermediate-acting (NPH), first or second generation long-acting (FGLA, SGLA) insulin using both MPR and aMPR.

METHODS : Texas Medicaid claims from 1/1/2014-6/30/2017 were extracted for type 2 diabetes (T2D) patients 18-63 years, with ≥2 basal insulin claims. The index date was the date of the first basal insulin claim without a claim 6 months prior nor a different basal insulin or premix insulin during 1-year follow-up. aMPR was calculated as MPR multiplied by a product-unique adjustment factor, which was the ratio between average days between refills and average recorded days’ supply (or ratio between two median values in the sensitivity analysis). ANOVA was performed to compare adherence.

RESULTS : Of 5,034 patients included, SGLA users had both the highest MPR and aMPR (MPR: SGLA 0.68±0.25 (N=325) vs. FGLA (N=4,522) 0.59±0.27 vs. NPH (N=187) 0.55±0.27, p<.0001; aMPR: 0.83±0.23 vs. 0.78±0.26 vs. 0.73±0.28; p=.0001, respectively). Adjustment factors ranged from 1.32 (degludec 200 units/mL pens) to 1.64 (glargine 100 units/mL pens), and from 0.97 (degludec 200 units/mL pens) to 1.07 (glargine 100 units/mL vials, pens, and detemir pens) in the sensitivity analysis. The relationship between package size and adjustment factor was not consistent.

CONCLUSIONS : It is unclear whether larger adjustment factors were due to larger package sizes, worse actual adherence behaviors, lower insulin needs, or a combination of factors. Thus, aMPR may introduce bias into adherence comparisons particularly when group characteristics are imbalanced. Selecting means versus medians in calculating adjustment factors may also lead to different results. A simulation study may facilitate identifying scenarios where aMPR is more appropriate.