br Cancer type Hospital group Average annual
Cancer type Hospital group Average annual volume Number of patients (%) Disease-specific 5-year mortality All-cause 5-year mortality
Hazard ratio (95% confidence interval)a
Crude Adjusted Crude Adjusted
a Adjusted for age, sex, comorbidity, calendar year of surgery and tumor stage.
Please cite this 1346242-81-6 article as: Gottlieb-Vedi E et al., Annual hospital volume of surgery for gastrointestinal cancer in relation to prognosis, European Journal of Surgical Oncology, https://doi.org/10.1016/j.ejso.2019.03.016
E. Gottlieb-Vedi et al. / European Journal of Surgical Oncology xxx (xxxx) xxx 5
specific 5-year mortality, the adjusted HR comparing the 4th and 1st quartiles of hospital volume was 0.87 (95% CI 0.56e1.36), and the effect modification analyses revealed no statistically significant associations (Table 5).
Higher hospital volume of colon cancer surgery was associated with a decreased risk of 5-year mortality outcomes (Table 4). Comparing the 4th and 1st quartile volume hospitals showed an adjusted HR of 0.89 (95% CI 0.84e0.96) for disease-specific 5-year mortality. This association was more pronounced in patients aged 74 years, with more comorbidity, and with more advanced tumor stage (Table 5).
Higher hospital volume of rectal cancer surgery did not decrease the 5-year mortality in most analyses (Table 4). The HR for disease-specific 5-year mortality was not statistically significantly decreased when comparing the 4th and 1st quartile of volume hospitals (adjusted HR 0.92, 95% CI 0.82e1.04). However, there was a decreased risk of all-cause 5-year mortality in the 3rd quartile volume hospitals compared to the 1st (HR 0.90, 95% CI 0.81e0.99), and the effect modification analyses indicated a decreased disease-specific 5-year mortality at higher volume hospitals compared to the 1st quartile in female and older patients (Table 5).
This study indicates a long-term prognostic benefit of higher hospital volume in colon cancer, especially in patients of older age, more comorbidity, and more advanced tumor stage. A survival benefit of higher hospital volume was also indicated in subgroups of patients with cancer of the esophagus, pancreas, and rectum. Hospital volume did not influence the survival in cancer of the stomach, liver, bile ducts, or small bowel.
Strengths of this study include the population-based cohort design and the long and complete follow-up. Additionally, the data on cancer diagnoses, hospital volume, mortality, and confounders were complete and accurate. The statistical power for assessing the survival in colon cancer was high. Among limitations is the fact that Sweden has a relatively low incidence of some of the studied tu-mours, and surgery of some tumours is still conducted at many hospitals. Thus, the hospital volume of some tumours was lower than in many other countries. There is also uncertainty of how hospital volume should ideally be defined to reflect reality. The moving average for assessing hospital volume took into account recent experiences regarding the specific type of surgery under investigation, which would seem justified. The use of quartiles counteracted arbitrary cut-offs, but it is possible that the highest quartile did not represent surgery of high enough volume to identify associations. Experience of other similar cancer procedures than the one under study might also influence the performance, which is difficult to assess. Another limitation is possible residual confounding, e.g. from tumor stage because of missing data. This problem was especially large for small bowel tumours, where 38% of tumour stage data was missing. However, the complete case and imputation analyses showed similar results, indicating that missing tumor stage data may not have influenced the results. One tumor stage classification was used for all cancers with rough categori-zation which might have introduced misclassification and residual confounding. Although the study was nationwide, there was limited precision for some tumours, particularly in the smallest patient groups, i.e. those with cancer of the liver, bile ducts, and small bowel. Thus, some true associations might not have been identified because of type II errors. In other words, lack of
associations may to some degree be due to limited statistical power, rather than lack of true associations. The patients operated during the latest date of the study period could have maximum 3 years of follow up for disease-specific mortality and approximately 4,5 years for all-cause mortality. However, this was only the case for a minority of all patients in the whole cohort, and should not explain the risk estimates, but only reduce the statistical power. Also, although the evaluation of several tumors in one study makes it possible to compare results, it also introduces a risk of chance findings due to multiple testing. Therefore, the positive findings from the subgroup analyses should be interpreted with caution.