[Skip to Navigation]
Sign In
Comment & Response
November 3, 2021

Questions About a Risk Prediction Model of Mortality After Esophagectomy for Cancer

Author Affiliations
  • 1Google, Mountain View, California
  • 2Department of Biostatistics, Harvard T. H. Chan School of Public Health, Boston, Massachusetts
JAMA Surg. 2022;157(3):279-280. doi:10.1001/jamasurg.2021.5701

To the Editor D’Journo and colleagues1 used patients’ baseline characteristics to construct a potentially useful prediction procedure for 90-day mortality after esophagectomy. Specifically, a multiple logistic regression model with 10 covariates was fit to data from the development cohort (n = 4172) and then validated in an independent cohort (n = 4231). For the validation process, the model was assessed with respect to discrimination and calibration. For discrimination, the C statistic (area under the receiver operating characteristic) was 0.64. For calibration, the model was examined with a Hosmer-Lemeshow goodness-of-fit test, which was not statistically significant. Instead of using this logistic regression to make predictions directly for the patients, the authors divided the patients into 5 categories based on the prognostic score from the model, ranging from very low risk (score, ≥1; mortality rate, 1.8%) to very high risk (score, ≤−5; mortality rate, 18.2%).

Add or change institution