Telemedicine use among physicians in the German outpatient sector: A secondary analysis of a cardiologist-dominated web-based survey with Bayesian model averaging and exploratory machine learning.

内容摘要

Cardiovascular diseases remain a major health burden in Germany, and telemedicine (TM) offers promising solutions for outpatient care, yet barriers limit uptake. While prior studies relied on qualitative or conventional statistical methods, they often struggled with model uncertainty and complex relationships. Building on a national survey, this study applies Bayesian Model Averaging (BMA) and extreme gradient boosting (XGBoost). This study aimed to explore candidate associations and patterns related to TM use among physicians in the German outpatient sector. We conducted a secondary analysis of a web-based survey carried out between 2023 and 2024. BMA was applied to identify explanatory associations with TM use, explicitly accounting for model uncertainty. XGBoost with SHAP values was used to explore classification patterns in a hypothesis-generating framework using repeated nested cross-validation. Of the 165 respondents, 95 (58%) reported using TM. BMA revealed a limited number of variables with moderate to high posterior inclusion probabilities (PIP), alongside substantial overall model uncertainty, with TM use associated with receiving information from professional associations or insurers and perceived TM benefits (e.g., improving patients' everyday quality of life, improving doctor-patient relationship). Practicing in Lower Saxony was associated with lower TM use. XGBoost demonstrated limited discriminative ability, with performance statistically indistinguishable from chance. SHAP-based analyses therefore identified exploratory patterns, including features such as information status, workplace, perceptions of TM's benefits for patients (e.g., health literacy, compliance and adherence) and TM's barriers (e.g., data protection, implementation incompatibility), as well as employment status. TM adoption in Germany's outpatient sector appears associated with structural-economic factors and physicians' perceptions of patient-related benefits. However, given the substantial model uncertainty in BMA and the limited predictive performance of the machine learning model, all findings should be interpreted cautiously. The machine learning component should be considered exploratory and hypothesis-generating.