Study populations for period analyses of cancer survival.

ABSTRACT

BACKGROUND: Period analysis is increasingly used to compute long-term cancer survival, as it provides better prediction of survival of newly diagnosed patients than traditional cohort analysis. However, the patient population to which period survival estimates best pertain to and which should be described in a study is less obvious. METHODS: Using Finnish Cancer Registry data on 23 common cancer sites, age-standardized period estimates of 5-, 10-, 15-, and 20-year relative survival were computed for each 2-, 5-, and 10-year calendar period in 1954-2003 and compared with survival estimates for two cohorts by means of mean, mean absolute and mean squared differences a full cohort of all patients potentially contributing some data to the survival analysis and a restricted cohort of patients diagnosed in the period of interest. RESULTS: In most computations, survival estimates for the full cohorts were on average closer to the period estimates for the majority of cancer sites. For 10-year survival, results were less obvious with respect to the mean difference. However, mean squared and mean absolute differences were smaller for the majority of cancers when using the full cohort. CONCLUSION: Our results suggest that the full cohort should be described in reports of period survival analysis.