Method: The study cohort consisted of patients in Norway diagnosed during 1960-1999 with one of the following six cancer diagnoses with high mortality: Stomach (ICD-10 code: C16), Colon (C18), Rectal (C19-21), Pancreas (C25), Lung (C33-34), or Kidney (C64) (n=149,086). To minimize censoring we required them to be born before 1922, as this allowed potential follow-up until age 90 at the end of follow-up at Dec 31, 2011. Patients were followed until emigration, death, or Jan 1, 2011, whichever came first, with a total follow-up time of 414.2 thousand person-years and an individual median follow-up time of 0.61 years. Overall 0.8% of patients were censored. Median, restricted mean (10 years), and mean survival time were estimated in strata defined by patients' sex, age category at diagnosis (40-49, 50-59, 60-69, 70-79, 80+ years), cancer type, and time period of diagnosis (1960-69, 1970-79, 1980-89, 1990-99). To compute means with censored follow-up time, we used cubic splines to extrapolate the log-hazard function. Exploratory plots were used to identify candidate relationships.
Result: For the six cancer diagnoses there was a linear relationship between the strata specific log-mean survival times and corresponding log-median survival times (Figure), and similarly between the log-ratio of means versus the log-ratio of medians.
Conclusion: For cancer diagnoses with poor prognosis, it is possible to predict mean survival times from corresponding medians from a simple log-linear relationship. This result may be used to validate mean survival times based on extrapolation.
