Dynamic programming for clinical baseline matching and its application to anti-hepatitis B research

Abstract

Clinical baseline matching is a critical step in the process of transforming real-world clinical records into novel medical knowledge, more precise diagnosis and efficacious treatments. Conventionally, the baseline matching was performed by heuristic methods or even manually. Here, a baseline-matching algorithm called "dynamite" was proposed using the dynamic programming technique. This algorithm minimizes the discrepancy of values of clinical variables or propensity scores between two patient groups in the study, while at the same time includes as many patients as possible so as to maximize statistical power. This algorithm was applied to an anti-hepatitis B research where two classes of approved drugs, peginterferon and nucleos(t)ide reverse transcription inhibitors (NRTI), were compared in terms of their protective effect in patients against hepatocellular carcinoma (HCC). Patients treated by the two classes of medications were retrospectively recruited from clinical records. Initially, age and HBeAg positivity were significantly different between candidate study subjects in the two treatment groups, with 153 and 177 patients respectively. Using the baseline-matching algorithm, the baseline characteristics of the two included patient groups, each comprised 120 patients, were well matched. Longitudinal analysis showed that the peginterferon-treated group achieved better HCC-free survival than the NRTI-treated groups (P = 0.0087293).