Statistical analysis
Univariate differences in cost were compared using nonparametric tests (Wilcoxon rank-sum test or Kruskal– Wallis test). To assess the multivariate association between cost and postoperative complications, loss of independence and hospital mortality, we performed a multiple linear regression, controlling for age, ASA, OS and FI-CGA. These factors have been associated with health care costs in previous research.4,20–24 Even after truncating the data, the data continued to be right-skewed. The cost data were logarithmically transformed to adjust for the skewness. Log-transforming the data improved the fit diagnostics of the models, demonstrated by more homogeneous, random residuals plots. Additionally, the R2 value of the cost model increased from 0.3554 to 0.5735 with log-transformation of cost data, indicating better fit to the regression line. We calculated adjusted median costs by exponentiating the least squares means of the logtransformed total costs using general linear models. The increase in cost associated with each level of an ordinal factor was calculated by exponentiating the β coefficient (e.g., x unit increase in complication severity results in exp [coefficient] increase in cost).19,25 We calculated the costs attributable to adverse events (most severe postoperative complication, hospital mortality and loss of independence) using a regression-based approach.26
Statistical analysisUnivariate differences in cost were compared using nonparametric tests (Wilcoxon rank-sum test or Kruskal– Wallis test). To assess the multivariate association between cost and postoperative complications, loss of independence and hospital mortality, we performed a multiple linear regression, controlling for age, ASA, OS and FI-CGA. These factors have been associated with health care costs in previous research.4,20–24 Even after truncating the data, the data continued to be right-skewed. The cost data were logarithmically transformed to adjust for the skewness. Log-transforming the data improved the fit diagnostics of the models, demonstrated by more homogeneous, random residuals plots. Additionally, the R2 value of the cost model increased from 0.3554 to 0.5735 with log-transformation of cost data, indicating better fit to the regression line. We calculated adjusted median costs by exponentiating the least squares means of the logtransformed total costs using general linear models. The increase in cost associated with each level of an ordinal factor was calculated by exponentiating the β coefficient (e.g., x unit increase in complication severity results in exp [coefficient] increase in cost).19,25 We calculated the costs attributable to adverse events (most severe postoperative complication, hospital mortality and loss of independence) using a regression-based approach.26
การแปล กรุณารอสักครู่..
