Effect of Financial Incentives for Process, Outcomes, or Both on Cholesterol Level Change

This randomized clinical trial evaluates the use of financial incentives to improve adherence to statin therapy and achieve a decrease in low-density lipoprotein cholesterol levels in individuals at risk for atherosclerotic cardiovascular disease.


B) Eligibility related to statin use
All participants attested to an active prescription to any statin medication. Patients also had to demonstrate evidence of nonadherence to the statin, which was ascertained using the 8-item Morisky Medication Adherence Questionnaire (MMAS-8). The MMAS includes 8 items that ask about medication adherence in different ways, such as asking about forgetting, skipping, and problems with adherence during travelling. Any response indicating adherence adds a point to the score. The score ranges from 0 to 8. Imperfect adherence was defined as a score less than 8 on the MMAS assessed during enrollment. A patient had to indicate with ≥1 response that they had less than perfect adherence to their statin.

eMethods 4: Smart pill bottles used in the trial
Prior to randomization, every trial participant received a smart bill bottle, which used a cellular network to transmit a record of the bottle opening to the Way to Health platform. Over the course of the trial, we sequentially provided three different bottles from different manufacturers to participants. Model 1 was removed from service with the phasing out of a cellphone network standard; the bottle was incompatible with the new standard. Model 2 was initially chosen to replace Model 1, but the manufacturer was unable to supply a sufficient quantity of devices in a timely fashion and feedback about the devices indicated some dissatisfaction among participants.

eMethods 6: Methods related to secondary and sensitivity analyses
The linear model for the primary outcome was refit using the same covariates and outcome of Δ from baseline to 18 months, 6 -months after the conclusion of the interventions, to assess durability of response.
We assessed differences in mean Δ from baseline across the four visits of the intervention period using a mixed effects model. We contrasted the control arm with each of the intervention arms using a linear combination of the 3, 6, 9 and 12-month visits. We repeated these analyses by pre-specified subgroups; no test of an interaction was used as the sample lacked statistical power to detect meaningful differences.
Using a simple linear model, we reported measured adherence for both the 12-month intervention period, and differences between any intervention arms and control.
Sensitivity analyses included (1) a complete case analysis and (2) a model adjusted for gender, education, income, and race.
As a post-hoc analysis, we assessed differences in mean Δ from 3-months to 6, 9, and 12-months. Models were adjusted for LDL-c at baseline and stratified by site, using multiply imputed data. Values are number of participants (%) or means and standard deviations (SD). a Analysis of Variance (ANOVA) for continuous data and Pearson Chi-square for categorical data b Values may not sum to the total number of subjects due to missing data c Subjects were asked to give their annual pre-tax household income in $10,000 intervals. Those who did not wish to answer the question were asked if they were willing to share whether their income was greater or less than $50,000.

eFigure 2A-F: Subgroup analyses of the primary outcome of change in LDL-C (mg/dL) from baseline to 12 months using a linear mixed-effects model
Methods: Mean change in LDL-c was estimated by a linear mixed effects model within each subgroup using arm, visit, interaction of arm and visit, and baseline LDL-c as covariates with a random intercept term. We tested the null hypothesis that the linear combination of LDL-c change from baseline across the 3, 6, 9 and 12-month visits was identical for the control and each intervention group. Complete data were used in the analysis. Results are considered exploratory and no adjustment for multiple comparisons was made.
Forest Plots: Squares are the mean reduction in LDL-c from baseline at each timepoint, and after adjustment for the mean baseline value across all arms. Error bars are 95% confidence intervals (CI). Negative values indicate reduction in LDL-c versus baseline. Dashed line is the overall mean for the LDL-c reduction for control in the subgroup. eFigure 2E: Income. Income category based on a baseline survey question asking pre-tax household income levels in $10,000 intervals with a follow-up question for subjects who did not wish to answer of: "Are you willing to share if your income is above or below $50,000?". The 121 participants who did not wish to disclose their income in response to either question are excluded from this analysis. For those with income>US $50,000 (n=324), the results of the hypothesis test of each intervention arm versus control is Process (p=.367), Outcomes (p=.869), Process plus Outcomes (p=.941). For those of income <US $50,000 (n=286) the results of the hypothesis test of each intervention arm versus control is Process (p=.097), Outcomes (p=.047), Process plus Outcomes (p=.006).

eFigure 3: Secondary analysis of data from all visits during the intervention (3, 6, 9 and 12 months)
Methods: Mean change in LDL-c was estimated by a linear mixed effects model using arm, visit, interaction of arm and visit, and baseline LDL-c as covariates, and a random intercept term. We tested the null hypothesis that the linear combination of LDL-c change from baseline across the 3, 6, 9 and 12 month visits was identical for the control and each intervention group. Complete data were used in the analysis (n=731). No adjustment for multiple comparisons was made.    eFigure 5: Spillover analysis flowchart of Penn Medicine patients with hypertension and diabetes plus usual care data for hemoglobin A1c and systolic blood pressure, respectively, in the pre-randomization period (up to 6 months prior) and later period (9 -15 months postrandomization)