The Diabetes Health Plan and Healthcare Utilization Among Beneficiaries with Low Incomes

Data Source and Population

The analytic data set is limited to 26 large employer groups that purchased the DHP and standard benefit plans from UHC (2009–2014) that have (1) internal pharmacy contracts, (2) complete pharmacy claims data, (3) sufficient medical claims and lab data to identify employees with type 2 diabetes (T2DM), and (4) fewer than 15% of employees enrolled in high deductible health plans. In addition to the above-mentioned criteria, the DHP employer groups must have at least 1 year of standard benefit plan data, prior to the purchase of the DHP, and comparison employer groups are further limited to those that have overlapping propensity scores with DHP employers after employer-level matching (described further below) and who have at least 2 years of continuous enrollment in the standard benefit plan during the duration of the match to the DHP employer.

A diabetes diagnosis was defined as having any of the following prior to the implementation of the DHP: (1) at least one 250.X ICD-9 diagnosis code from an inpatient, outpatient, or emergency department claim; (2) hemoglobin A1C laboratory value of 6.5% or greater or a 2-h value on an oral glucose tolerance test of greater than 200 mg/dl; or (3) at least one prescription fill for an oral hypoglycemic medication other than metformin or insulin. Estimated household income is obtained from the AmeriLINK data.18 This data source incorporates consumer financial survey responses, publicly available information (public records, census information, and retail transaction records), and zip-code-level information from the Internal Revenue Service to generate individual-level estimates of household income. The sample size flow chart for the unique DHP and comparison beneficiaries are shown in Figures 1 and 2, respectively.

Figure 1figure 1

Provider appointments, ER visits, and inpatient visits—DHP sample.

Figure 2figure 2

Provider appointments, ER visits, and inpatient visits—comparison sample.

Propensity Score Matching

Matching criteria for both the DHP and comparison employers were derived with respect to the 12-month period preceding the date of DHP adoption for the DHP employers or standard insurance plan contract renewal (the index date). The matching criteria included the following as reported by UHC: average employee salary, geographic region, number of employees, % female, % in each racial/ethnic category (White, Black, Asian, Hispanic), health benefit plan generosity, % of employees with a HDHP, and % of beneficiaries with each one of the following claims-based co-morbidities (hypertension, hyperlipidemia, coronary artery disease, anxiety/depression, dementia, osteoarthritis, rheumatoid arthritis, non-skin cancer, chronic obstructive pulmonary disease, congestive heart failure, atrial fibrillation, end-stage renal disease, peripheral vascular disease, stroke, schizophrenia) as well as the index date. A single comparison employer could be matched to more than one DHP employer. Individual-level matching criteria were based on the following pre-index date criteria: race and ethnicity, age, gender, Charlson Comorbidity Index, insulin use status, presence of any diabetes complication (retinopathy, nephropathy, neuropathy, cardio/peripheral vascular disease, history of a diabetes-related hospitalization), and baseline healthcare utilization. Nearest-neighbor matching was conducted with replacement using a caliper equal to 25% of the propensity score standard deviation, in an effort to get 3 comparison matches for each DHP beneficiary.19 The employer and beneficiary matching was done using PROC PSMATCH in SAS version 9.4.

Outcomes

We coded “disease management visits” as a count variable based on the composite number of outpatient visits with providers who may perform diabetes management during the course of a visit (endocrinologist, internal medicine, family practice, urgent care specialist, nurse practitioner, physician assistant). Patients with an unusually high number of disease management appointments in the baseline year were excluded from this analysis prior to matching using the 1.5 interquartile range heuristic for identifying outliers.20 We treated emergency room and hospital utilization as dichotomous variables. Both variables were indicators coded as “1” if the utilization was present during the post-period year and coded as “0” if the utilization was not present. These utilization outcomes were not restricted to those exclusively related to diabetes.

Statistical Analyses

We used a DID study to examine the impact of the DHP on utilization. The key assumption of the DID study is the parallel trends assumption which necessitates that the pre-intervention trends for outcome measures across the treatment and comparison groups are the same.21 If the parallel trend assumption is met, any difference in the pre-post intervention change in slope across treatment and comparison groups is attributed to intervention effects. We use the propensity-matched sample to increase the likelihood that the DHP and comparison groups have a similar trend of utilization during the pre-intervention time period.22 Non-linear statistical models were run for each of the utilization outcomes using the PROC GENMOD procedure in SAS. The model used for disease management visits employed a Poisson distribution with a log link function, and binomial logit models were used to model emergency room and hospital utilization. These models include an indicator for time (post-index vs. pre-index) that was coded as “1” if the observation was from the post-index year and coded as “0” if the observation was from the pre-index (baseline) year, and an indicator for group (DHP group vs. comparison group) that was coded as “1” if the observation was from the DHP group and coded as “0” if the observation was from the comparison group and the interaction between time and group, among our matched samples. Specifically, the between-group differences in the change of the outcome variables, post-index, were estimated by the interaction effects.

Sensitivity Analyses

We conducted an additional test to assess the sensitivity of our results to selection bias by repeating the above-mentioned analyses with DHP employers that use an opt-out enrollment strategy as the sole source of the treatment population. This methodological change should allow for evaluation of the DHP utilization effects among a less motivated subset of beneficiaries than the subset including individuals that proactively enrolled in the DHP.12

留言 (0)

沒有登入
gif