Regional Variation in Relative Survival—Quantifying the Effects of the Competing Risks of Death by Using a Cure Fraction Model with Random Effects

Abstract

We extend a mixture cure fraction model with random effects to allow estimation of relative survival of cancer patients by region in a country with a parsimonious number of parameters. The heterogeneity in the expected survival was taken into account such that the expected mortality rate was considered as a random quantity varying across regions. Two sets of random effects were used to describe regional variation, both in the cure fraction and in the relative survival of the non-cured patients. This hierarchical model was fitted in a Bayesian framework to the follow-up data of Finnish colon cancer patients diagnosed in 1975-2004. The public health importance of the regional differences was described by the extra survival time and avoidable deaths that could be gained, if the regional variation were eliminated and all the hospital districts in Finland achieved the most favourable level of relative survival. The differences in relative survival between the districts were quite small during the three decades. If all the districts had achieved the most favourable level of relative survival, the mean survival times of the female patients diagnosed in 1975-1984, 1985-1994 and 1995-2004 would have increased by 5%, 4% and 2% respectively.

Estimating multilevel regional variation in excess mortality of cancer patients using integrated nested Laplace approximation

“…Considering hospital district‐specific population mortality rates to be fixed without random variation was shown to be a reasonable assumption for the estimation of excess mortality . However, variation in excess mortality within hospital districts includes variation due to potential differences in population mortality between municipalities within each hospital district.…”

“…However, variation in excess mortality within hospital districts includes variation due to potential differences in population mortality between municipalities within each hospital district. In municipality‐specific population mortality rates, random variation should be taken into account by using an extended model that includes the likelihood for population mortality …”

“…The decrease in the standard deviation of the net survival in breast cancer does not indicate that the public health importance of regional variation has decreased because breast cancer incidence increased simultaneously. When assessing the public health importance, the changes in incidence can be taken into account, for example, by estimating potential numbers of person years or deaths that would be saved, if regional variation were eliminated …”

Modeling clustered long‐term survivors using marginal mixture cure model

“…The likelihood-based estimation methods for random effects models also makes missing data handling easier. Random effects based cure models for clustered data have been well studied in the literature (Chen, Ibrahim & Sinha, 2002b;Herring & Ibrahim, 2002;Lai & Yau, 2008;Seppä, Hakulinen, Kim & Läärä, 2010;Seppä, Hakulinen & Läärä, 2014;Xiang, Ma & Yau, 2011;Yau & Ng, 2001;Yin, 2008). However, the estimation methods for random effects models tend to be more computationally intensive.…”

Do cancer survival statistics for every hospital make sense?

“…10 If we analyse patient data that include hospital characteristics (e.g. the presence of a multidisciplinary team, or small caseload), we can ascertain whether a certain characteristic is associated with lower survival.…”