Fast and Efficient Feature Engineering for Multi-Cohort Analysis of EHR Data.

We present a framework for feature engineering, tailored for longitudinal structured data, such as electronic health records (EHRs). To fast-track feature engineering and extraction, the framework combines general-use plug-in extractors, a multi-cohort management mechanism, and modular memoization. Using this framework, we rapidly extracted thousands of features from diverse and large healthcare data sources in multiple projects.

A Novel Computational Tool for Mining Real-Life Data: Application in the Metastatic Colorectal Cancer Care Setting.

BACKGROUND: Randomized clinical trials constitute the gold-standard for evaluating new anti-cancer therapies; however, real-life data are key in complementing clinically useful information. We developed a computational tool for real-life data analysis and applied it to the metastatic colorectal cancer (mCRC) setting. This tool addressed the impact of oncology/non-oncology parameters on treatment patterns and clinical outcomes. METHODS: The developed tool enables extraction of any computerized information including comorbidities and use of drugs (oncological/non-oncological) per individual HMO member. The study in which we evaluated this tool was a retrospective cohort study that included Maccabi Healthcare Services members with mCRC receiving bevacizumab with fluoropyrimidines (FP), FP plus oxaliplatin (FP-O), or FP plus irinotecan (FP-I) in the first-line between 9/2006 and 12/2013. RESULTS: The analysis included 753 patients of whom 15.4% underwent subsequent metastasectomy (the Surgery group). For the entire cohort, median overall survival (OS) was 20.5 months; in the Surgery group, median duration of bevacizumab-containing therapy (DOT) pre-surgery was 6.1 months; median OS was not reached. In the Non-surgery group, median OS and DOT were 18.7 and 11.4 months, respectively; no significant OS differences were noted between FP-O and FP-I, whereas FP use was associated with shorter OS (12.3 month; p <0.002; notably, these patients were older). Patients who received both FP-O- and FP-I-based regimens achieved numerically longer OS vs. those who received only one of these regimens (22.1 [19.9-24.0] vs. 18.9 [15.5-21.9] months). Among patients assessed for wild-type KRAS and treated with subsequent anti-EGFR agent, OS was 25.4 months and 18.7 months for 124 treated vs. 37 non-treated patients (non-significant). Cox analysis (controlling for age and gender) identified several non-oncology parameters associated with poorer clinical outcomes including concurrent use of diuretics and proton-pump inhibitors. CONCLUSIONS: Our tool provided insights that confirmed/complemented information gained from randomized-clinical trials. Prospective tool implementation is warranted.

Identifying and Investigating Unexpected Response to Treatment: A Diabetes Case Study.

The availability of electronic health records creates fertile ground for developing computational models of various medical conditions. We present a new approach for detecting and analyzing patients with unexpected responses to treatment, building on machine learning and statistical methodology. Given a specific patient, we compute a statistical score for the deviation of the patient's response from responses observed in other patients having similar characteristics and medication regimens. These scores are used to define cohorts of patients showing deviant responses. Statistical tests are then applied to identify clinical features that correlate with these cohorts. We implement this methodology in a tool that is designed to assist researchers in the pharmaceutical field to uncover new features associated with reduced response to a treatment. It can also aid physicians by flagging patients who are not responding to treatment as expected and hence deserve more attention. The tool provides comprehensive visualizations of the analysis results and the supporting data, both at the cohort level and at the level of individual patients. We demonstrate the utility of our methodology and tool in a population of type II diabetic patients, treated with antidiabetic drugs, and monitored by the HbA1C test.