patient-level and physician-level predictors of discharge against medical advice: a multilevel modeling approach

Background: discharge against medical advice (dama) is a complex and multifaceted issue in healthcare, often challenging the continuity of care and affecting patient outcomes. objectives: this study aimed to investigate the predictors of discharge against medical advice (dama) by simultaneously examining patient- and physician-level variables within a unified analytical framework. methods: this cross-sectional study was conducted in 2023 at one of the largest private hospitals in northwest iran. the study included all 16,071 patients admitted in 2022 and 137 attending physicians. a multilevel analysis model was employed to examine the influence of variables related to patients and physicians in predicting dama. results: the study involved patients with a mean age of 45.28 ± 19.59 years, with 67.7% being women and 14.8% not having health insurance. among the physicians studied, the mean age was 56.27 ± 12.29 years, with 67.2% being male and 70.1% being hospital shareholders. patients with dama comprised 6.8% (n = 1094). the null model had a log-likelihood value of -3304.90. when patient-level predictors were added, the value increased to -3041.76, a statistically significant improvement (p < 0.001) based on the chi-square test. subsequently, incorporating physician-level predictors further increased the log-likelihood value to -2996.16, and this increase was also statistically significant compared to the model with only patient-level predictors (p < 0.001). physician-level factors, including specialization, sex, and experience, were associated with dama. significant patient-level variables included age, type of insurance, and type of disease (p < 0.05). conclusions: utilizing multilevel modeling enables the assessment of the significance of both physician-level and patient-level factors. to avoid conflicting results, it is recommended to evaluate hospital performance based on dama by considering both levels.