OBJECTIVE: Previous studies have developed cardiovascular surgery outcome prediction models using only patient risk factors, but surgery outcomes from the patient's perspective seem to differ between hospitals. We have developed outcome prediction models that incorporate preoperative patient risks, as well as hospital processes and structure. METHODS: Data were collected from the Japan Cardiovascular Database for patients scheduled for cardiovascular surgery between January 2005 and December 2007. We analyzed 33,821 procedures in 102 hospitals. Logistic regression was used to generate risk models, which were then validated through split-sample validation. RESULTS: Odds ratios, 95% confidence intervals, and P values for structures and processes in the mortality prediction model were as follows: "hospital annual adult cardiac surgery volume (continuous; every 1 procedure increase per year)" (odds ratio, 0.998; confidence interval, 0.997-0.999; P < .001); "recommended staffing and equipment" (odds ratio, 0.75; confidence interval, 0.64-0.87; P < .001); "daily conferences with cardiologists" (odds ratio, 0.79; confidence interval, 0.60-1.02; P = .073); "intensivists involved in postsurgical management" (odds ratio, 0.89; confidence interval, 0.77-1.02; P = .90); "public hospitals" (odds ratio, 0.80; confidence interval, 0.70-0.93; P = .003); "surgeons lacking miscellaneous duties" (odds ratio, 0.80; confidence interval, 0.70-0.93; P = .003); and "surgeons who work no more than 32 hours per week" (odds ratio, 0.55; confidence interval, 0.32-0.95; P = .032). The mortality prediction model had a C-index of 0.85 and a Hosmer-Lemeshow P value of .79. CONCLUSIONS: Our models yielded good discrimination and calibration, so they may prove useful for hospital selection based on individual patient risks and circumstances. Improved surgeon work environments were also shown to be important for both surgeons and patients.
OBJECTIVE: Previous studies have developed cardiovascular surgery outcome prediction models using only patient risk factors, but surgery outcomes from the patient's perspective seem to differ between hospitals. We have developed outcome prediction models that incorporate preoperative patient risks, as well as hospital processes and structure. METHODS: Data were collected from the Japan Cardiovascular Database for patients scheduled for cardiovascular surgery between January 2005 and December 2007. We analyzed 33,821 procedures in 102 hospitals. Logistic regression was used to generate risk models, which were then validated through split-sample validation. RESULTS: Odds ratios, 95% confidence intervals, and P values for structures and processes in the mortality prediction model were as follows: "hospital annual adult cardiac surgery volume (continuous; every 1 procedure increase per year)" (odds ratio, 0.998; confidence interval, 0.997-0.999; P < .001); "recommended staffing and equipment" (odds ratio, 0.75; confidence interval, 0.64-0.87; P < .001); "daily conferences with cardiologists" (odds ratio, 0.79; confidence interval, 0.60-1.02; P = .073); "intensivists involved in postsurgical management" (odds ratio, 0.89; confidence interval, 0.77-1.02; P = .90); "public hospitals" (odds ratio, 0.80; confidence interval, 0.70-0.93; P = .003); "surgeons lacking miscellaneous duties" (odds ratio, 0.80; confidence interval, 0.70-0.93; P = .003); and "surgeons who work no more than 32 hours per week" (odds ratio, 0.55; confidence interval, 0.32-0.95; P = .032). The mortality prediction model had a C-index of 0.85 and a Hosmer-Lemeshow P value of .79. CONCLUSIONS: Our models yielded good discrimination and calibration, so they may prove useful for hospital selection based on individual patient risks and circumstances. Improved surgeon work environments were also shown to be important for both surgeons and patients.