br Annual hospital volume exposure br
Annual hospital volume exposure
The exposure was annual hospital volume for planned
resectional surgery for each of the eight gastrointestinal cancer locations under study. Hospital volume was categorized into four about equal-sized groups (quartiles) separately for each cancer type. Hospitals with the 25% lowest annual surgery volume constituted the 1st quartile, and hospitals with the highest 25% annual surgery volume were in the 4th quartile, whereas quartiles 2 and 3 were inbetween. This pre-defined categorization was used to avoid arbitrary cut-offs. To account for temporary fluctuations in annual hospital volume between calendar years and to incorporate the experience achieved during the last few years prior to each operation, hospital volume was calculated using a 4-year moving average number of operations for each cancer procedure at each hospital, starting in 2002 (3 years before the first study year) and ending in 2013 (last study year). Thus, each patient was allocated the moving average of hospital volume for the year of their oper-ation plus 3 years earlier. For example, for an operation conducted in 2005, the exposure was the average hospital volume of that Thapsigargin procedure at that hospital during the 4 years 2002, 2003, 2004 and 2005.
The main outcome was disease-specific mortality within 5 years of surgery, i.e. with the cancer location under study being recorded as the main or contributing cause of death. The secondary outcome was all-cause mortality within 5 years of surgery, i.e. deaths occurring independent of the cause of death.
Hazard ratios (HRs) with 95% confidence intervals (CIs) were calculated using multivariable Cox proportional hazards models. The HRs were adjusted for five well-established prognostic factors shared by gastrointestinal cancers (with categorizations in brackets): age (continuous), sex (male or female), comorbidity (Charlson comorbidity score 0, 1 or 2, excluding any cancer) , calendar year of surgery (continuous), and pathological tumor stage (0-II or III-IV).
Information was complete for all variables, except for tumor stage. To manage missing data for tumor stage, both complete case analysis and multiple imputation analysis were conducted. The number of imputed datasets was 20 and the monotone logistic method in PROC MI was used based on the assumption that missing occurred randomly (MAR) . The seven variables included in the imputation analysis were age (continuous), sex (male and female), comorbidity (Charlson comorbidity score 0, 1, or 2), calendar year of surgery (continuous), follow-up time (until death or maximum 5 years), disease-specific 5-year mortality, and hospital volume (quartiles). Furthermore, PROC MIANALYZE was used to combine the results from the analyses of the 20 datasets.
A sensitivity analysis was conducted in which patients with any cancer diagnosis prior to the gastrointestinal cancer under study were excluded. To evaluate effect modification, an interaction term was included for each of the five above mentioned prognostic factors and hospital volume one by one. Thereafter HRs with 95% CIs were calculated for the association for each category of the prognostic factor. Each prognostic factor was categorized as above, except for age and calendar year of surgery which were categorized as below and above the median.
An experienced biostatistician (FM) conducted all data man-agement and statistical analyses according to a pre-defined study protocol. All analyses were conducted using SAS Statistical Package (version 9.4, SAS Institute Inc., Gary, NC).
Please cite prostaglandins 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 3
Characteristics of patients who have undergone resectional surgery for gastrointestinal cancers in Sweden in 2005e2013.
Esophagus Stomach Liver Pancreas Bile ducts Small bowel Colon Rectum
Charlson comorbidity score
The study included 45,908 patients who had undergone resec-tional surgery for any of the eight studied gastrointestinal cancer locations. Patients’ characteristics by cancer location are presented in Table 1. Most patients had colon cancer (n ¼ 26,688) and the fewest had bile duct cancer (n ¼ 363). The median age was between 65 and 74 years in the various cancer groups. Male sex was most prevalent in esophageal cancer (80.0%) and least prevalent in bile duct cancer (35.8%). A Charlson comorbidity score 1 was most common in patients with liver cancer (64.8%) and least common in bile duct cancer patients (33.3%). The highest proportion with advanced tumor stage (III-IV) was in patients with colon cancer (49.2%) and the lowest proportion in liver cancer patients (9.9%). Tumor stage was missing for 4.1e11.9% for all cancer locations, except for a higher rate in small bowel cancer (38.5%). The disease-specific 5-year mortality was >50% in cancer of the esophagus, stomach, pancreas, or bile ducts (Table 1).