Deriving real-world oral antineoplastic treatment endpoints: A comparison of automated data extraction versus manual chart abstraction.

Abstract

e18070 Background: Real world data (RWD) is increasingly being used to generate evidence that informs clinical care. Calculating outcomes measures using RWD, such as time-on-treatment (ToT), requires accurate medication start/end dates. Whereas intravenously administered medication dates are typically available, dates for oral antineoplastics (OANs) are challenging as they are filled by specialty pharmacies and documented separately. To determine the feasibility of automated ToT extraction, we used the Syapse Learning Health Network to compare the utility of automated chart extraction (ACE) and of manual chart abstraction (MCA) from the electronic health record (EHR) in providing high-quality OAN data. Methods: We selected cancer patients (pts) from two histologies for whom OANs were prescribed. For these pts, ACE was compared to MCA. ACE data were derived from EHR structured medication lists through existing interoperability pipelines. MCA was performed by 4 trained data abstractors who reviewed the corresponding progress notes per pt. Expected date of OAN start/end, and actual medication start/end, were recorded as available. Chi-square and descriptive statistics were used for analysis. Results: 61 cancer pts (31 lung, 30 breast; mean age 62y [27-90]; 20% male) who received OAN from 3 multi-hospital medical systems were evaluated. ACE detected a greater number of expected start/end dates as compared to MCA (Table, P < 0.001). In contrast, for actual start/end dates, the converse was true; ACE was inferior to MCA (P < 0.001). Noteworthy, only 26% of pts had actual start/end dates using manual chart abstraction only. There was no concordance between the actual ACE and MCA dates. We evaluated if ACE expected ToT dates could be used as a surrogate for actual MCA ToT dates. Of the 12 pts satisfying this criteria, mean discordance in ToT was 19 days (1-71 days). Conclusions: MCA only captured actual ToT in a small minority of pts. ACE expected dates were a poor surrogate for ToT and highly discordant to over 2 months. Neither MCA nor ACE from EHR data were adequate for the majority of pts. Alternative mechanisms such as integration with additional data sources like specialty pharmacy dispensing records are essential. [Table: see text]