Automated Detection of Systematic Off-label Drug Use in Free Text of Electronic Medical Records

Abstract

Off-label use of a drug occurs when it is used in a manner that deviates from its FDA label. Studies estimate that 21% of prescriptions are off-label, with only 27% of those uses supported by evidence of safety and efficacy. We have developed methods to detect population level off-label usage using computationally efficient annotation of free text from clinical notes to generate features encoding empirical information about drug-disease mentions. By including additional features encoding prior knowledge about drugs, diseases, and known usage, we trained a highly accurate predictive model that was used to detect novel candidate off-label usages in a very large clinical corpus. We show that the candidate uses are plausible and can be prioritized for further analysis in terms of safety and efficacy.

Figure 1.

Analysis workflow.

Figure 2.

Calculation of drug and disease similarity features. To calculate the similarity of drug m to other drugs used to treat indication n, we find other rows j such that entry (j, n) = 1 (i.e. drug j is known to treat indication n). We calculate the cosine and Jaccard similarities of these rows (indicated by blue arrows) to row m and use the maximum similarity. An analogous calculation is used to calculate indication. similarity.