gms | German Medical Science

Background: In 2008, Germany introduced the first nationwide skin cancer screening. Statutory insured citizens aged 35 years and older have an opportunity to be screened every two years by a qualified primary care physician or dermatologist. While there has been a promising one-year pilot screening in 2003/2004 in the federal state of Schleswig-Holstein, the evidence for the effectiveness of skin cancer screening is limited. We expand an existing microsimulation for melanoma screening to apply also to non-melanoma skin cancer, to take into account risk factors for skin cancer, and to provide estimations of health care costs. We will present and discuss challenges we have encountered in this project.

Methods: A microsimulation is similar to a simulated randomized controlled trial. A population of persons with individual personal and disease-related characteristics is simulated. Each simulated person remains either healthy throughout their whole life-time or develops a tumor at a certain age, followed by disease progression, (regular) diagnosis, a survival time depending on sex, age and tumor size, and finally death – either from the tumor or another cause – at an individual age. Knowing all individual natural disease histories, this population is 1) not being offered screening, and 2) being offered screening. The results of these two groups with identical baseline characteristics are compared (incidence, mortality, costs). Screening-related variables like the interval of screening can be further modified to compare the effects of different screening scenarios.

We consider melanoma and non-melanoma skin cancer in the form of squamous cell carcinoma (SCC) and basal cell carcinoma (BCC) as well as their in situ variants. Each simulation model requires many different kinds of input data, for example from the German Centre for Cancer Registry Data (ZfKD) to get population-based information on age and – after multiple imputation – T-stage at diagnosis in addition to information on survival time. Information on tumor growth, sensitivity and specifity of screening were extracted from scientific publications, while costs were derived from scientific publications, insurance data, and the uniform assessment standard (“Einheitlicher Bewertungsstandard”).

Results: During the still ongoing development of the simulation model, we encountered several challenges. First, the choice of which endpoints to implement, for example age-adjusted mortality and life-years gained. Second, the choice of which risk factors and screening conditions to implement, for example skin type, screening interval or risk-adaptive screening. Third, the scarcity of data needed for some parts of the model. Fourth, the complexity of programming a microsimulation with as many components as planned here. Fifth, although the microsimulation can be run on parallel cores, the number of cores and their processing power are a restrictive factor on getting results within an adequate time frame.

Discussion: While developing a complex microsimulation in R is challenging, it's possible. We recommend to others to plan with enough time and to start with a basic model with the possibility to add on more details. To get results in an adequate time frame, one should invest in enough processing power and collaborate closely with experts to get the best input data.

Existing microsimulation model [1]

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.