Hospital selection patterns are essential for evaluating medical accessibility and optimizing resource management. In the absence of medical records, early studies primarily used accessibility functions to estimate potential selection probabilities (PSPs). With the advent of travel data, data-driven functions have enabled the calculation of observed selection probabilities (OSPs). Comparing PSP and OSP helps to leverage travel data to understand hospital selection preferences and improve medical service evaluation models. This study proposes a selection probability-based accessibility model for calculating PSP and OSP accessibility. A case study in Shenzhen employed nighttime navigation data to reduce interference from different travel modes. The distance decay function was validated, with exponential and Gaussian functions performing best. For hospitals, the PSP distribution closely aligned with OSP, except in areas with high hospital density. This discrepancy may result from the PSP function overestimating the selection probability for nearby hospitals, a limitation that could be addressed by fitting the distance decay function to actual data. PSP-based accessibility and Gini coefficients differ from those of OSP. However, when parameters are fitted to actual data, the PSP- and OSP-based functions produce nearly identical results. Fitting to actual data can notably improve the accuracy of PSP and the corresponding accessibility outcomes. These findings may provide valuable references for medical service evaluation methodologies and offer insights for planning and management.