OBJECTIVES: Electronic health records (EHR) have been widely used for comparative effectiveness research. Care-discontinuity (i.e., receiving care outside of an EHR system) was associated with substantial information bias when using EHR as the sole data source. We aimed to develop and validate a prediction score for having high care-continuity to reduce such bias. METHODS: Study cohort comprised all patients ≥ 65 in EHR from two large US provider networks linked with Medicare insurance claims data from 2007/1/1 to 2014/12/31. Based on the linked EHR-claims data, we measured care-continuity by the Mean Proportion of Encounters Captured (MPEC) by the EHR system. With predictors available in EHR, we built a prediction model for MPEC by Lasso regression, using the two EHR systems as training and validation set, respectively. Within deciles of predicted continuity, we quantified misclassification by Mean Standardized Differences between the proportions of 40 key variables based on EHR alone vs. linked claims-EHR data (MSD_{_40_variables}, <0.1 was used to indicate satisfactory variable classification). We compared patient characteristics in those with high vs. low predicted EHR continuity. RESULTS: Based on 104,403 patients in the training and 79,336 in the validation set, we developed a prediction score that was highly correlated with the measured care-continuity in both training and validation sets (Spearman correlation=0.81, 0.83, respectively). In the training set, MSD_{_40_variables} (misclassification based on EHR alone) in the worst predicted continuity decile was 7.8 (95% confidence intervals 6.8-9.1) fold greater than that in the best predicted continuity decile. Those with top 20% predicted continuity were found to have satisfactory variable classification (MSD_{_40_variables}<0.1) and comparable patient characteristics, compared to the rest of population. We found similar results in the validation set. CONCLUSIONS: Restriction to patients with high predicted care-continuity may reduce misclassification of key characteristics and improve validity of drug research relying exclusively on EHR.