2.6. Statistical analysis

First, we calculated the number of hospitals and the mean annual surgical volume by the DCCH type and described the distribution of the basic characteristics of patients per DCCH type. Using a multivariable Cox proportional hazard model, HRs for each hospital type (non‐DCCHs constituted the reference group) were estimated after adjustment of potential confounders. In the model, 95% CIs of the HR were adjusted using robust estimators of variance because patients who underwent surgery in the same hospital would have cluster correlations. Based on the postestimation of the Cox proportional hazard model, adjusted survival probabilities were estimated.

Subgroup analyses were undertaken for specific cancer sites. We selected six major cancers, stomach, colorectal, lung, breast, uterus, and prostate, because of their high incidence, and each site accounted for 4.0% or more of all cancer cases (Table S1). Patients with the remaining cancer sites were combined as “others” for the analysis. P < .05 indicated statistical significance. The Stata 15.1 statistical software package (Stata Corp, College Station, Texas, USA) was used for all analyses in this study.