BACKGROUND: It is vital that unbiased estimates of relative survival are estimated and reported by cancer registries. A single figure of relative survival is often required to make reporting simpler. This can be obtained by pooling all ages or, more commonly, by using age-standardisation. The various methods for providing a single figure estimate of relative survival can give very different estimates. METHODS: The problem is illustrated through an example using Finnish thyroid cancer data. The differences are further explored through a simulation study that investigates the effect of age on the estimates of relative survival. RESULTS: The example highlights that in practice the all-age estimates from the various methods can be substantially different (up to 6 percentage units at 15 years of follow-up). The simulation study confirms the finding that differing estimates for the all-age estimates of relative survival are obtained. Performing age-standardisation makes the methods more comparable and results in better estimation of the true net survival. CONCLUSIONS: The all-age estimates of relative survival rarely give an appropriate estimate of net survival. We feel that modelling or stratifying by age when calculating relative survival is vitally important as the lack of homogeneity in the cohort of patients leads to potentially biased estimates. We feel that the methods using modelling provide a greater flexibility than life-table based approaches. The flexible parametric approach does not require an arbitrary splitting of the time-scale, which makes it more computationally efficient. It also has the advantage of easily being extended to incorporate time-dependent effects.
BACKGROUND: It is vital that unbiased estimates of relative survival are estimated and reported by cancer registries. A single figure of relative survival is often required to make reporting simpler. This can be obtained by pooling all ages or, more commonly, by using age-standardisation. The various methods for providing a single figure estimate of relative survival can give very different estimates. METHODS: The problem is illustrated through an example using Finnish thyroid cancer data. The differences are further explored through a simulation study that investigates the effect of age on the estimates of relative survival. RESULTS: The example highlights that in practice the all-age estimates from the various methods can be substantially different (up to 6 percentage units at 15 years of follow-up). The simulation study confirms the finding that differing estimates for the all-age estimates of relative survival are obtained. Performing age-standardisation makes the methods more comparable and results in better estimation of the true net survival. CONCLUSIONS: The all-age estimates of relative survival rarely give an appropriate estimate of net survival. We feel that modelling or stratifying by age when calculating relative survival is vitally important as the lack of homogeneity in the cohort of patients leads to potentially biased estimates. We feel that the methods using modelling provide a greater flexibility than life-table based approaches. The flexible parametric approach does not require an arbitrary splitting of the time-scale, which makes it more computationally efficient. It also has the advantage of easily being extended to incorporate time-dependent effects.
Authors: M D Chirlaque; D Salmerón; J Galceran; A Ameijide; A Mateos; A Torrella; R Jiménez; N Larrañaga; R Marcos-Gragera; E Ardanaz; M Sant; P Minicozzi; C Navarro; M J Sánchez Journal: Clin Transl Oncol Date: 2017-07-17 Impact factor: 3.405
Authors: Lasse H Jakobsen; Therese M-L Andersson; Jorne L Biccler; Tarec C El-Galaly; Martin Bøgsted Journal: BMC Med Res Methodol Date: 2019-01-28 Impact factor: 4.615
Authors: William J Hulme; Matthew Sperrin; Glen Philip Martin; Nick Curzen; Peter Ludman; Evangelos Kontopantelis; Mamas A Mamas Journal: BMJ Open Date: 2019-02-19 Impact factor: 2.692