Abstract 1202: Electronic Health Record To Predict CHF In Primary Care
Background: Congestive Heart Failure (CHF) is often diagnosed long after symptoms first emerge. We used longitudinal electronic health record (EHR) data at the Geisinger Clinic(GC) (580000 patients, 41 community practice sites) on primary care patients to develop a model for early detection of CHF diagnosis.
Methods: The prediction model used a nested case-control study design. Inclusion criteria were CHF due to systolic dysfunction, GC primary care physician for atleast 12 months & atleast one matched control. Combination of risk factors and sensitivity/specificity trade-offs (ROC analysis) at different time points were analyzed. GEE logistic regression models were fitted and used to account for clustering (matching, repeated measures). Bootstrap re-sampling was used on candidate models to make sure that associations held up for a variety of samples.
Results: The prevalence of CHF in this population was validated against the NHANES data. 420 cases and 3963 controls were included in the final analysis. The AUC for the prediction model (table⇓) was 0.74 for 6-month prediction date i.e. probability that a randomly selected case will have a higher risk score than a randomly selected control. The AUC is 0.69 and 0.65 for the 12 and 18 month prediction dates, respectively.
Conclusion: This robust predictive model could help identify patients at high risk of future incidence of CHF, allow risk stratification in real time and thereby allow more targeted and aggressive treatment of established risk factors. Further, this model for detecting CHF in the pre-diagnostic period might allow opportunities to explore the predictive power of newer biomarkers for early diagnosis of CHF.