A collaborative approach to developing an electronic health record phenotyping algorithm for drug-induced liver injury

Abstract

Objective: To describe a collaborative approach for developing an electronic health record (EHR) phenotyping algorithm for drug-induced liver injury (DILI).

Methods: We analyzed types and causes of differences in DILI case definitions provided by two institutions-Columbia University and Mayo Clinic; harmonized two EHR phenotyping algorithms; and assessed the performance, measured by sensitivity, specificity, positive predictive value, and negative predictive value, of the resulting algorithm at three institutions except that sensitivity was measured only at Columbia University.

Results: Although these sites had the same case definition, their phenotyping methods differed by selection of liver injury diagnoses, inclusion of drugs cited in DILI cases, laboratory tests assessed, laboratory thresholds for liver injury, exclusion criteria, and approaches to validating phenotypes. We reached consensus on a DILI phenotyping algorithm and implemented it at three institutions. The algorithm was adapted locally to account for differences in populations and data access. Implementations collectively yielded 117 algorithm-selected cases and 23 confirmed true positive cases.

Discussion: Phenotyping for rare conditions benefits significantly from pooling data across institutions. Despite the heterogeneity of EHRs and varied algorithm implementations, we demonstrated the portability of this algorithm across three institutions. The performance of this algorithm for identifying DILI was comparable with other computerized approaches to identify adverse drug events.

Conclusions: Phenotyping algorithms developed for rare and complex conditions are likely to require adaptive implementation at multiple institutions. Better approaches are also needed to share algorithms. Early agreement on goals, data sources, and validation methods may improve the portability of the algorithms.

Keywords: Drug-induced liver injury; Electronic health records; Pharmacovigilance; Phenotyping; Rare diseases.

Figure 1

Summary of study methodology. CU, Columbia University; DILI, drug-induced liver injury; Mayo, Mayo Clinic; MSSM, Mount Sinai School of Medicine.

Figure 2

Summary of Columbia University (CU) acute liver injury and drug-induced liver injury (DILI)-related natural language processing (NLP) algorithm results. ICD-9, International Classification of Diseases, revision 9.

Figure 3

Drug-induced liver injury (DILI) case definition and phenotyping algorithm harmonization. (A) Columbia University (CU) phenotyping algorithm: CU's International Serious Events Consortium (iSAEC)-informed algorithm makes a distinction between acute liver injury and chronic liver injury. CU chose to focus on acute liver injury. With respect to drug exposure, CU considered patients with any drug prescribed within 90 days of an acute liver injury diagnosis. Given iSAEC protocol specifications and access to structured data, CU considered iSAEC-specified threshold values for alanine aminotransferase (ALT), intestinal alkaline phosphatase (APh) and intervascular bilirubin. CU excluded 10 diagnoses initially. (B) Mayo Clinic (Mayo) phenotyping algorithm: Mayo's Drug Induced Liver Injury Network (DILIN)-informed algorithm considered any liver injury-related diagnoses. Mayo also considered a subset of drugs of interest to DILIN; and specified the temporal relationship between drug administration, DILI diagnosis and laboratory measures. Given DILIN protocol specifications and access to clinical notes of recruited patients, Mayo used DILIN-specified thresholds and text terms for ALT, aspartate aminotransferase (AST), and international normalized ratio (INR) for use by the cTAKES natural language processing engine. Specific emphasis was laid on investigating the DILI-related complications and medications in various sections of the clinical notes (including chief complaints, impression report plans). Mayo excluded three diagnoses initially. Given Mayo's focus on a smaller number of medications, exclusion criteria were less stringent than CU in order to optimize recall. (C) Harmonized phenotyping algorithm. See supplementary file 1 (available online only) for more detail.

Figure 4

Two implementations of the same electronic health record phenotyping algorithm. Step 1 is implemented the same for both approaches. Step 2 differs by the query anchor and the time frame for which laboratory values are checked to be within normal ranges. The top implementation is anchored on the date of medication administration with laboratory values checked within 30 days before. The bottom implementation is anchored on the date of acute liver injury diagnosis with laboratory values checked between 180 and 90 days before. Step 3 differs by the query anchor and the time frame for which laboratory values are checked for thresholds to qualify as drug-induced liver injury (DILI). The top implementation is anchored on the date of medication administration with laboratory values checked within 180 days following drug administration. The bottom implementation is anchored on the date of acute liver injury diagnosis with laboratory values checked within 90 days before. CU, Columbia University; Mayo, Mayo Clinic; MSSM, Mount Sinai School of Medicine.