Methodological aspects on measurements of temporal trends in melanoma survival

We compared measurements of temporal trends in cancer survival either based on observed survival or relative survival. Survival rates for cutaneous malignant melanoma diagnosed in Sweden 1960-1982 were used as an example. In a proportional hazards model based on observed survival, the hazard ratio for patients diagnosed during the period 1975-1979 was 0.56 (95% confidence limits 0.49-0.64) in men and 0.45 (0.38-0.53) in women when the period 1960-1964 was the reference. When relative survival was the outcome measure, during the same period, the hazard ratio decreased to 0.47 (0.39-0.56) in men and 0.36 (0.29-0.44) in women. Thus, the effect of period of diagnosis was larger when the analysis was based on relative survival rates. The reasons for the differences between these two alternative measurements are explained in the article and advantages/disadvantages of the two approaches are discussed.