The role of patient, clinician, and neighborhood characteristics in predicting telemedicine engagement and modality in primary care: a cross-sectional analysis

The ability to use telemedicine may differ by social and clinician characteristics, however, studies examining telemedicine use have predominantly focused on associations with patient characteristics. This study assessed how patient, clinician, and neighborhood characteristics contributed to the prediction of telemedicine use and modality in primary care. Cross-sectional study using 2022 electronic health record data from adult primary care patients within a large Midwestern integrated healthcare system. We identified primary care visit mode (any telemedicine visits vs. in-person only) and telemedicine modality (any video vs. audio-only). Multivariable logistic regression models were constructed using distinct and combined sets of patient, clinician, and neighborhood characteristics. Model performance was assessed by area under the curve (AUC), Akaike Information Criterion, and Bayesian Information Criterion, and nested models were compared using likelihood ratio (LR) tests. Of 994,133 patients, 5.9% had telemedicine visits, with 14.5% of those being phone-only. The predictive power of clinician characteristics alone and neighborhood characteristics alone were more modest than patient characteristics alone, yet models incorporating clinician and neighborhood characteristics significantly improved prediction and fit for both outcomes (p < 0.001 for all LR tests). The comprehensive model, including all three covariate sets, consistently showed the highest discriminatory ability and goodness-of-fit (i.e., AUC = 0.7431 for any telemedicine and AUC = 0.8225 for audio-only). Addition of clinician and neighborhood characteristics to models meaningfully modified the associations between patient characteristics and both telemedicine outcomes. Incorporating clinician and neighborhood characteristics improved model fit and modified association with patient characteristics when analyzing telemedicine patterns. Better prediction of telemedicine use will help healthcare systems determine strategies for optimizing healthcare delivery modes (including telemedicine) and ensuring access to optimize health.