A Comparison of a Diabetes Life Expectancy Prediction Model Results from Electronic Health Record Abstraction and Patient Surveys (Preprint)

Abstract

Prediction models are increasingly utilized in clinical practice but the optimal approach to collecting the needed inputs is unknown.

Our objective was to compare a mortality prediction model inputs and scores based on chart abstraction versus patient survey.

Older patients with type 2 diabetes at an urban, Chicago primary care practice were recruited and the Lee Mortality Index was calculated from retrospective chart review and patient surveys.

We saw non-significant differences in mortality scores (p=0.61) and subsequent diabetes in older adult health status class recommendations (p=0.70) comparing chart abstraction to survey data. However, there were large differences in certain domains such as functional status and presence of disease. Differences in chart review and survey resulted in 20% having discordant diabetes recommendations.

Healthcare organizations should work to systematically and routinely collect PROs that are inputs for widely used prediction models.