Myopic and Forward Looking Behavior in Branded Oral Anti-Diabetic Medication Consumption: An Example from Part D

Context: The non-linear Medicare Part D benefit contains incentives for consumption decisions based on end-of-year (future) out-of-pocket prices, with beneficiaries facing different cost sharing prices in deductible (100%), initial (25%), gap (100%), and catastrophic (5-7%) phases. Many analyses have focused on medication consumption in the coverage gap, when out-of-pocket prices are highest, but not considered whether beneficiaries exhibited forward looking behavior by reducing medication consumption before reaching these end-of-year prices. To address this question, we evaluate medication consumption changes during the coverage year, taking into account generic prescribing increases that could affect consumption decisions.

Methods: Data Source: IMS Health Longitudinal Database, with de-identified patient-level data for over 60% of the prescriptions filled in the U.S. Study Sample: Elderly Part D diabetes patients who used medications in four oral anti-diabetic drug classes, two dominated by higher priced branded (mean prescription price $200) and two by inexpensive generic products ($13-$15 mean price) in 2009, the last year Part D coverage gap fully in effect. Empirical strategy: Compare Part D standard (non-LIS) to low-income-subsidy (LIS) patients, who face no non-linear budget constraints. Primary outcome variable: Adherence by medication class and month, with patients considered adherent when ratio of days supplied to calendar days at least 0.8. Generalized Estimating Equations estimate adherence odds by month and coverage type (non-LIS vs. LIS), relative to LIS January referent; models run separately for each medication class.

Results: Among 38,054 propensity matched patients, 85% (N=32,520) used primarily generic and fewer (N=8,657; 23%) branded-only class oral anti-diabetics. In adjusted analyses, non-LIS had 0.74 (Branded1; 95% CI: 0.60-0.92) and 0.66 (Branded2: 95% CI: 0.60-0.74) times the odds of branded adherence in January, relative to LIS referent. These odds declined to 0.54 in June (Branded1; 95% CI: 0.34-0.67 vs. LIS: 0.89; 0.74-1.1) and 0.55 in July (Branded2: 95% CI: 0.49-0.61; vs. LIS: 0.8; 0.73-0.87). By December, non-LIS branded adherence odds were 0.35 and 0.28 lower than LIS (Branded1: OR: 0.39 (95% CI: 0.31-0.47 vs. 0.75; 95% CI: 0.63-0.91; Branded2: 0.35; 95% CI: 0.32-0.39 vs. 0.63 (95% CI: 0.57-0.7). Non-LIS and LIS primarily generic class adherence odds remained similar through the coverage year.

Discussion: We observed consumption responses to the non-linear Part D benefit that were neither fully forward-looking nor fully myopic. Branded oral anti-diabetic adherence odds decreased rapidly through the first 7 months of the coverage year, before most individuals would have reached a $2,700 gap spending threshold, suggesting that beneficiaries reduced more costly medication use as gap grew closer. These findings are consistent with both forward looking behavior and discounting of future costs. Significantly lower non-LIS branded adherence odds in January, however, when beneficiaries were furthest from out-of-pocket coverage gap prices, could reflect myopic responses to high branded cost-sharing deductible ($200) and initial coverage phase ($50) out-of-pocket prices. When the gap is fully phased out, some beneficiaries may increase more costly branded medication consumption before their expenditures reach the former gap spending threshold; others may reduce branded consumption in response to deductible and initial coverage phase out-of-pocket prices.