OBJECTIVES: Patients who self-discharge against medical advice (AMA) may be at higher risk for a hospital readmission if the unauthorized discharge was premature. The study objective is to examine the relationship between discharges AMA and cardiovascular disease (CVD) hospital readmissions while addressing bias due to potential confounding, selection, and hospital-level clustering. METHODS: This cross-sectional study uses confidential inpatient hospital discharge data covering the years 2000 to 2005. The outcome variables captured readmissions for a CVD-related condition following an index CVD-related admission. The covariate of interest was an indicator variable for a discharge AMA in the index hospitalization. The relationship between discharges AMA and 7-day, 31-day, 180-day, and 365-day hospital readmissions was examined using generalized linear models with adjustments for clustering and selection bias.  RESULTS: The sample included 443,088 patients, of which 22,757 (5.1%) were readmitted to the same hospital. Approximately 1% of the patients who were readmitted in the hospital during the study period left AMA on the index admission while 0.89% of those who were not readmitted left AMA (p=0.047). The odds of a CVD-related readmission within 7 days, 31 days, 180 days and 365 days were 145% (p<0.001), 55% (p<0.001), 26% (p<0.001) and 15% (p=0.0011) higher for patients discharged AMA on index admission compared to those who were discharged formally. Results are robust to corrections for observable selection bias (via propensity score analysis) and hospital-level clustering.  CONCLUSIONS: A self-discharge AMA among patients admitted for CVD is predictive of CVD-related readmissions and the strength of association increases as the time between admissions decreases.