Which Is Responsible for Appendectomy Outcomes: Attending Surgeon, Resident, or Their Communication Lack?
In a study of 54,467 appendectomies using the National Surgical Quality Improvement Program (NSQIP) database, Scarborough and colleagues1 demonstrated that resident participation in surgery was associated with a 30% increase in major complication rates. To estimate an unconfounded effect of resident participation, the authors conducted a multiple logistic regression model in which resident participation was entered as the primary explanatory variable and 31 patient- or surgeon-related factors were entered as the secondary explanatory variables. In results, the adjusted odds ratio of resident participation was 1.27 and its 95% confidence interval was 1.14 to 1.42. Multiple regression analysis is commonly used for stratification for confounding in the cause-effect analysis when direct stratification is difficult to do because of a large sample size and/or a lot of confounders. In constructing a multiple regression model, covariate selection is important because the odds ratio of an explanatory variable will change with addition or elimination of another variable. A method for covariate selection is the use of preliminary bivariate regression analysis. To best predict the primary outcome, then, stepwise regression analysis will be performed because the regression coefficient (odds ratio) of a variable in bivariate regression does not necessarily correlate with that in multiple regressions. When looking at the statistics in the study by Scarborough and colleagues,1 the explanatory variables were prespecified with the assumption that these variables are independently associated with the primary outcome. Although the use of prespecified variables is feasible for evaluating the effects of a set of explanatory variables on the primary outcome (confirmatory data analysis), there is no concern with causality, only with identifying a model in this analysis.2 In explanatory data analysis using bivariate and multivariate models, there is a concern with theory construction to explain a phenomenon when the phenomenon is little studied. When selecting covariates, the odds ratio and 95% confidence interval of each variable are calculated, leading to the ranking of importance of the variables. For example, preoperative functional status was the most important predictor (odds ratio 2.33) and preoperative sepsis was the second important predictor (odds ratio 1.73) of major complications after appendectomy in a study using the NSQIP database.3 Another advantage of explanatory data analysis is obtaining a chance to find modifiable factors that embarrass residents’ performance at operation. In a recent study that estimated the effects of resident involvement on the outcomes of laparoscopic appendectomy for uncomplicated appendicitis using the NSQIP data, resident involvement was associated with a 38% increase in major complication rates.4 We reported that the early outcomes of total gastrectomy for gastric cancer performed by trainees were inferior to those by staff surgeons, but blood loss of more than 500 g was more closely associated with adverse outcomes in multivariate analysis, meaning that the outcomes of total gastrectomy by trainees might be improved by reducing operative blood loss.5 Scarborough and colleagues1 suggest that the decrease in the level of attending supervision during appendectomy is a possible explanation for their findings. Given that among appendectomies registered in the NSQIP a proportion of appendectomy without an attending scrubbed in the operating room was around 1%,6 this suggestion may be justified because the attending surgeon was directly involved in 99% of the appendectomies. A surgical team without an attending scrubbed in the operating room was not associated with increased complication risks in that study,6 suggesting that the quality of appendectomy is not decreased by resident participation itself. Communication lack between attending surgeon and resident may be responsible for the outcomes of appendectomy.