Longitudinal surveys after medication dispensing: real-world data from an at-home smart hub (Preprint)

BACKGROUND Real-World Data (RWD) has gained strategic, financial, and regulatory importance in the pharmaceutical industry during both development and commercialization. However, the actual RWD submitted to regulatory bodies often fails to be considered primary evidence. Fit for purpose longitudinal survey methods allow researchers to be targeted in their data acquisition but face the challenges of panel recruitment, panel attrition, and low data quality from an incentivized population. An untapped platform for administering longitudinal surveys to patients is the digital medication dispensing device. One such system is an in-home smart hub called "spencer." OBJECTIVE We evaluated whether the spencer stand-alone medication dispensing smart hub could be a primary generator of longitudinal patient reported outcomes. Using the spencer platform, we measured platform persistency, survey compliance, and the psychometric properties of the survey responses. METHODS We analyzed 4,138 patients on the spencer smart hub platform during their first two years of tenure. A discrete survival framework was used to estimate platform persistency. Survey opt-out and response rates were computed by patient tenure. For accessing reliability and validity, we examined a spencer question on reported falls called "Q_FALL." For reliability, we looked at the within-patient correlation between tenure years one and two mean Q_FALL and computed transition rates between intervals, assuming persistence of vigor among healthy patients. For evidence of validity, we measured the degree of association between Q_FALL and known factors influencing fall risk, including age, biological sex, quality of life, physical and emotional health, and use of SSRIs or SNRIs. RESULTS For the spencer platform, 51.0% of patients were retained beyond two years (24 thirty-day periods). For 1,832 patients retained past one year, 82.0% kept their surveys enabled through the 12th month of tenure with an aggregate response rate of 84.1% in the period. Two thirds of patients had near-perfect response rates for the entire year while one third answered less frequently over time. For 234 patients with >= 5 Q_FALL responses in the first two years of tenure, the within patient Pearson correlation was 0.723. In the rare falls category, 84% in year one remained in the same category in year two. Q_FALL did not show the expected relationship with sex (P=.66) or age (P=.76) but had significant positive relationships (all having P<0.001) with SSRI / SNRI usage, quality of life, depressive symptoms, physical health, disability, and trips to the emergency room. CONCLUSIONS Longitudinal surveys administered on a medication dispensing smart hub can generate robust, fit for purpose RWD. Patients were persistent to the spencer platform for years and continued to answer survey questions at high rates. Response patterns showed evidence that reliable and valid measures of important health constructs are possible.