[Skip to Content]
Access to paid content on this site is currently suspended due to excessive activity being detected from your IP address Please contact the publisher to request reinstatement.
[Skip to Content Landing]
Special Communication
September 26, 2001

Comparison of 2 Methods for Calculating Adjusted Survival Curves From Proportional Hazards Models

Author Affiliations

Author Affiliations: Departments of Medicine (Drs Ghali, Knudtson and Ms Galbraith) and Community Health Sciences (Drs Ghali, Quan, Brant, and Faris), University of Calgary, Calgary, Alberta; Department of Public Health Sciences, University of Alberta, Edmonton, Alberta (Ms Norris); and the Institute for Social and Preventive Medicine, University of Lausanne, Lausanne, Switzerland (Dr van Melle).

JAMA. 2001;286(12):1494-1497. doi:10.1001/jama.286.12.1494

Context Adjusted survival curves are often presented in medical research articles. The most commonly used method for calculating such curves is the mean of covariates method, in which average values of covariates are entered into a proportional hazards regression equation. Use of this method is widespread despite published concerns regarding the validity of resulting curves.

Objective To compare the mean of covariates method to the less widely used corrected group prognosis method in an analysis evaluating survival in patients with and without diabetes. In the latter method, a survival curve is calculated for each level of covariates, after which an average survival curve is calculated as a weighted average of the survival curves for each level of covariates.

Design, Setting, and Patients Analysis of cohort study data from 11 468 Alberta residents undergoing cardiac catheterization between January 1, 1995, and December 31, 1996.

Main Outcome Measures Crude and risk-adjusted survival for up to 3 years after cardiac catheterization in patients with vs without diabetes, analyzed by the mean of covariates method vs the corrected group prognosis method.

Results According to the mean of covariates method, adjusted survival at 1044 days was 94.1% and 94.9% for patients with and without diabetes, respectively, with misleading adjusted survival curves that fell above the unadjusted curves. With the corrected group prognosis method, the corresponding survival values were 91.3% and 92.4%, with curves that fell more appropriately between the unadjusted curves.

Conclusions Misleading adjusted survival curves resulted from using the mean of covariates method of analysis for our data. We recommend using the corrected group prognosis method for calculating risk-adjusted curves.