FRAMR-EMR: Framework for Prognostic Predictive Model Development Using Electronic Medical Record Data with a Case Study in Osteoarthritis Risk

TL;DR

This paper introduces FRAMR-EMR, a comprehensive framework for developing prognostic predictive models using electronic medical record data, demonstrated through a case study estimating osteoarthritis risk in primary care patients.

Contribution

The paper presents a step-by-step framework, FRAMR-EMR, for constructing prognostic models from EMR data, including addressing common pitfalls and demonstrating with a real case study.

Findings

Developed a 5-year osteoarthritis risk prediction model with good discrimination.

Provided a detailed framework for EMR-based prognostic model development.

Validated the model with moderate calibration and good discrimination.

Abstract

Background-Prognostic predictive models are used in the delivery of primary care to estimate a patients risk of future disease development. Electronic medical record, EMR, data can be used for the construction of these models. Objectives- To provide a framework for those seeking to develop prognostic predictive models using EMR data, and to illustrate these steps using osteoarthritis risk estimation as an example. FRAMR-EMR-The FRAmework for Modelling Risk from EMR data, FRAMR-EMR, was created, which outlines step-by-step guidance for the construction of a prognostic predictive model using EMR data. Throughout these steps, several potential pitfalls specific to using EMR data for predictive purposes are described and methods for addressing them are suggested. Case Study-We used the DELPHI, DELiver Primary Healthcare Information, database to develop our prognostic predictive model for estimation of osteoarthritis risk. We constructed a retrospective cohort of 28447 eligible primary care patients. Patients were included if they had an encounter with their primary care practitioner between 1 January 2008 and 31 December 2009. Patients were excluded if they had a diagnosis of osteoarthritis prior to baseline. Construction of a prognostic predictive model following FRAMR-EMR yielded a predictive model capable of estimating 5-year risk of osteoarthritis diagnosis. Logistic regression was used to predict osteoarthritis based on age, sex, BMI, previous leg injury, and osteoporosis. Internal validation of the models performance demonstrated good discrimination and moderate calibration. Conclusions-This study provides guidance to those interested in developing prognostic predictive models based on EMR data. The production of high quality prognostic predictive models allows for practitioner communication of accurately estimated risks of developing future disease among primary care patients.