Christian Stock1, Ute Mons2, Hermann Brenner3. 1. Division of Clinical Epidemiology and Aging Research, German Cancer Research Center (DKFZ), Im Neuenheimer Feld 581, 69120 Heidelberg, Germany. Electronic address: c.stock@dkfz.de. 2. Division of Clinical Epidemiology and Aging Research, German Cancer Research Center (DKFZ), Im Neuenheimer Feld 581, 69120 Heidelberg, Germany; Cancer Prevention Unit, German Cancer Research Center (DKFZ), Im Neuenheimer Feld 280, 69120 Heidelberg, Germany. 3. Division of Clinical Epidemiology and Aging Research, German Cancer Research Center (DKFZ), Im Neuenheimer Feld 581, 69120 Heidelberg, Germany; Division of Preventive Oncology, German Cancer Research Center (DKFZ) and National Center for Tumor Diseases (NCT), Im Neuenheimer Feld 460, 69120 Heidelberg, Germany; German Cancer Consortium (DKTK), German Cancer Research Center (DKFZ), Im Neuenheimer Feld 280, 69120 Heidelberg, Germany.
Abstract
BACKGROUND: Cancer incidence projections are of major interest for resource allocation in healthcare and medical research. Previous reports of cancer incidence projections have often been deterministic, i.e. lacking quantification of uncertainty. We project cancer incidence in Germany by applying an approach that allows for probabilistic interpretation of outcomes. MATERIAL AND METHODS: German cancer registry data from 1999 to 2013 are used to predict cancer incidence for 27 sites until the year 2030. We apply Bayesian Poisson and negative binomial models to obtain probabilistic estimates of future site-, year-, sex- and age-specific cancer incidence rates. Results from cancer incidence models are combined with probabilistic population projections to estimate numbers of incident cancer cases. Comparisons of overall and stratum-specific cancer incidence rates and case numbers are made between the years 2015 and 2030 by estimating absolute and relative change along with uncertainty intervals. RESULTS: The overall standardized incidence rate is expected to increase by 5% (95%-credible interval: 0%, 13%) until 2030. Incident case numbers are expected to increase by 23% (95%-credible interval: 17%, 29%) which is mostly driven by demographic change. The probability (expressed as %) that the change will be >10%, >20% or >30% was calculated to be >99%, 66% and 7%, respectively. CONCLUSIONS: The analysis provides evidence on the future cancer burden in Germany by applying a fully Bayesian approach that offers advantages in terms of flexibility, probabilistic interpretability, and transparency. It may especially be an alternative when long-term cancer incidence data are missing.
BACKGROUND:Cancer incidence projections are of major interest for resource allocation in healthcare and medical research. Previous reports of cancer incidence projections have often been deterministic, i.e. lacking quantification of uncertainty. We project cancer incidence in Germany by applying an approach that allows for probabilistic interpretation of outcomes. MATERIAL AND METHODS: German cancer registry data from 1999 to 2013 are used to predict cancer incidence for 27 sites until the year 2030. We apply Bayesian Poisson and negative binomial models to obtain probabilistic estimates of future site-, year-, sex- and age-specific cancer incidence rates. Results from cancer incidence models are combined with probabilistic population projections to estimate numbers of incident cancer cases. Comparisons of overall and stratum-specific cancer incidence rates and case numbers are made between the years 2015 and 2030 by estimating absolute and relative change along with uncertainty intervals. RESULTS: The overall standardized incidence rate is expected to increase by 5% (95%-credible interval: 0%, 13%) until 2030. Incident case numbers are expected to increase by 23% (95%-credible interval: 17%, 29%) which is mostly driven by demographic change. The probability (expressed as %) that the change will be >10%, >20% or >30% was calculated to be >99%, 66% and 7%, respectively. CONCLUSIONS: The analysis provides evidence on the future cancer burden in Germany by applying a fully Bayesian approach that offers advantages in terms of flexibility, probabilistic interpretability, and transparency. It may especially be an alternative when long-term cancer incidence data are missing.
Authors: Maximilian Knoll; Jennifer Furkel; Jürgen Debus; Amir Abdollahi; André Karch; Christian Stock Journal: BMC Med Res Methodol Date: 2020-10-15 Impact factor: 4.615