Background: Prognostication is an important aspect of medical practice. It relies on statistical modeling testing the correlation of variables with the outcome of interest.
Objective: In contrast with the classic approach of predictive modeling, this study aimed to estimate the unique, individual, and relative contributions. This includes the quantitative contributions of patient-, tumor-, and treatment-related factors to oncologic outcome after rectal cancer resection.
Design: This was a retrospective analysis of prospectively registered data.
Settings: The study included 65 hospitals participating on a voluntary basis in the Project on Cancer of the Rectum, a Belgian multidisciplinary improvement project of rectal cancer care.
Patients: A total of 1470 patients presenting midrectal or low-rectal adenocarcinoma without distant metastasis were included.
Intervention: The study intervention was total mesorectal excision with or without sphincter preservation.
Main outcome measures: The unique, individual, and relative contributions of a set of covariables to the statistical variability of the distant metastasis rate and overall survival have been calculated.
Results: The 5-year distant metastasis rate was 21% and overall survival 76%. A large amount of the variability of the outcomes (ie, 83.6% to 84.2%) could not be predicted by the prognostic factors. Unique contributions of the predictors ranged from 0.1% to 3.1%. The 3 risk factors with the highest unique contribution for distant metastasis were lymph node ratio, pathologic tumor stage, and total mesorectal quality; for overall survival they were age, lymph node ratio, and ASA score.
Limitations: The main weakness of this study was incomplete participation and registration in the Project on Cancer of the Rectum.
Conclusions: Several factors influence oncologic outcomes and are present in prediction models. However, the models predict relatively little of outcome variation.