Evaluation of skin cancer screening in Germany – a Microsimulation

Summary

Background

Comprehensive skin cancer screening was introduced in Germany in 2008. It is unclear whether subsequently observed changes in the epidemiology of malignant melanoma (MM), squamous cell carcinoma (SCC) and basal cell carcinoma (BCC) are due to the screening. Simulation models are used to compare different screening scenarios with each other and with observed incidence and mortality trends.

Methods

A microsimulation model for MM, SCC, and BCC was programmed and validated separately by sex on skin cancer‐specific mortality data. In addition to the currently practiced screening (biennial offer, annual participation probability 15%), triennial screening, screening with increased participation, e.g., via invitation, and no screening were simulated. Incidence, mortality, costs, and life‐years gained were simulated for 30 years from the start of screening.

Results

Compared with no screening, mortality is reduced by 13.8% in the simulation with current conditions. This effect occurs in the first years after screening starts before reaching a stable level. More screening allows for further increases in incidence, life‐years gained, and costs and decreases in mortality. Comparing simulated and observed effects shows little agreement.

Conclusions

The model is useful for comparing screening scenarios to identify potential optimization opportunities in SCS. Additional, especially risk‐adapted, screening scenarios should be investigated.

INTRODUCTION

Skin cancer is one of the most common types of cancer in Germany. In 2019, about 12,000 women and 13,000 men received a diagnosis of invasive malignant melanoma (MM) of the skin. About 1,250 women and 1,800 men died from this disease. The estimated incidence of non‐melanocytic skin cancer for 2019 was 104,000 in women and 119,000 in men; 445 women and 631 men died from this type of skin cancer. ¹ If skin cancer is detected early, the chance of easy treatment and complete remission is high, and the prognosis after a diagnosis of skin cancer is generally good. ² Relative five‐year survival for MM is 95% (women) and 94% (men), while in late stages (UICC II–IV) it is only between 83% (UICC II; women) and 40% (UICC IV; men). ¹

Skin cancer screening was introduced in Germany in July 2008. All members of statutory health insurance aged 35 and above can receive biennial whole‐body inspection at participating practices. Before introduction of this screening program, a pilot project on skin cancer screening was performed in the federal region of Schleswig‐Holstein between mid‐2003 and mid‐2004. After this pilot project, melanoma mortality dropped by about 50% in both women and men, while no such decrease was observed in the surrounding regions. ³ After the introduction of skin cancer screening throughout Germany, melanoma mortality kept increasing slightly at first. A reversal of this trend in the direction of decreasing mortality in the total population (women and men) was only seen from 2014 onwards. ⁴ , ⁵ We cannot differentiate between a success of the screening and the availability of new treatment options when trying to explain this trend. We still cannot be sure if and by how much skin cancer screening can decrease melanoma mortality, since there is a world‐wide lack of randomized studies on skin cancer screening. Potential options for improvement are still unknown.

One option for investigating the potential of skin cancer screening in routine clinical practice based on the available evidence is to use simulation models which can analyze intervention effects based on theoretical statistics. In the context of the Pertimo project, which is financially supported by the Federal Joint Committee Innovation Fund, a joint microsimulation model for MM, squamous cell carcinoma (SCC), and basal cell carcinoma (BCC) was developed and conducted for four central screening scenarios: currently practiced screening, triennial attendance, high probability of participation, and no screening. Mortality, incidence, costs, and life‐years gained were simulated and compared for the four screening scenarios. In addition, modelled sex‐specific incidence and mortality were compared with the real‐life statistics in Germany. Similar models have been successfully used by the U.S. Preventive Services Task Force (USPSTF) in the last few years to implement and modify their screening programs for colorectal cancer, lung cancer, and other entities. ⁶ , ⁷ , ⁸

METHODS

Microsimulation

The microsimulation model used in this study is an updated and expanded version of an established model for early detection of MM and was implemented in R, version 4.1.1. ⁹ , ¹⁰ Apart from MM, the updated model can also show SCC and BCC including in‐situ forms. SCC and BCC constitute the majority of non‐melanocytic skin cancer forms, but since the two cancer types are very different, ¹¹ , ¹² they are simulated independent of each other. In addition, the simulation model calculates not only incidence and mortality for the years after introduction of skin cancer screening, but also costs and life years gained (LYG).

Natural History Model

Our microsimulation comprises a “Natural History” model as well as an Intervention model. In the Natural History model, an individual disease‐specific “natural life” is determined for every person in a simulated population. The system simulates the potential skin cancer risk for every person without participation in a skin cancer screening program. In case of a skin cancer diagnosis, the system simulates when the tumor originates, how large it is at any given point in time, at what age and what tumor stage diagnosis is made, and when death will occur either due to the tumor or due to other reasons. This simulation is based on external data such as velocity of tumor growth, or survival rates.

Time spans between disease‐specific life events and disease‐specific characteristics are based on the following data sources: Life expectancy data derive from mortality tables provided by the Federal Statistical Office of Germany (Statistisches Bundesamt), ¹³ data on tumor growth and other parameters were elicited via literature research, and data on costs and regularity of follow‐up derive from medical billing codes and the relevant guidelines. ¹¹ , ¹² , ¹⁴ , ¹⁵ Treatment costs were extracted from billing data supplied by DAK‐Gesundheit, a health insurance company, while screening and excision costs derive from the German Physicians’ Fee Schedule (EBM, Einheitlicher Bewertungsmaßstab). ¹⁶

Age and tumor stage (T stage) at diagnosis as well as survival times – depending on age, sex, and T stage – were extracted from data provided by the Center for Cancer Registry Data, or they were estimated. ¹⁷ Missing details for MM and SCC, especially regarding T stages, were substituted five times in ten iterations via multiple imputation with the R package MICE. ¹⁸ An overview of the imputation variables used can be found in the online supplement (Table S1). Since too much T stage data (almost 80%) was missing in BCC cases, imputation was not performed for BCC. Other parameters were extracted from the literature ¹⁹ , ²⁰ , ²¹ , ²² , ²³ , ²⁴ , ²⁵ , ²⁶ Details on all input data of the simulation model are listed in the online supplement (Table S2).

There are, of course, no empirical data on the exact percentage of people who develop skin cancer at some point during their lifetime – including all tumors never detected. This value was therefore determined via calibration of the simulated incidence with the observed incidence in Germany before the introduction of skin cancer screening. A more detailed explanation of this process can be found at Eisemann et al. ¹⁰ For MM, age‐standardized 2007 incidence rates were used from the nine federal Länder (states) with the highest mean data completeness between 2005 and 2007, based on estimates by the Robert Koch Institute. These were Bavaria, Bremen, Hamburg, Lower Saxony, Rhineland‐Palatinate, Saarland, Saxony‐Anhalt, Schleswig‐Holstein, and Thuringia). ⁴ , ²⁷ For SCC and BCC, the observed incidence was derived from the cancer registry data for Schleswig‐Holstein and Rhineland‐Palatinate between 2004 and 2007, which are least affected by under‐reporting. ²⁸

The partial models were validated according to cancer registry data, by graphically comparing the simulated incidence and mortality rates without screening with the published (MM) or calculated (SCC) rates from the relevant Länder dating from 2007. ²⁷ While the observed incidence rates are incorporated into the model (albeit indirectly via calibration), simulated mortality derives exclusively from modeling. Thus, congruence between the simulated combined distribution of incidence and mortality with the observed data indicates appropriate modeling. For validation, the same selection of federal Länder was used as had been for calibration of MM. This form of validation could not be used for BCC because mortality for this entity is negligible.

Intervention Model

In the intervention model, the same simulated population as in the Natural History model was offered cancer screening with certain characteristics (such as age groups of the screening participants, probability of participation). This changes a number of the individual simulated lives, since tumors will be detected earlier and potentially at an earlier stage, which in turn may lead to changes in survival time and cause of death (Figure 1).

FIGURE 1.

Schematic representation of two life courses with and without screening.

Screening Scenarios

Four intervention scenarios were analyzed:

• No Screening: No skin cancer screening at all (Natural History model).

• Current: Simulation of the skin cancer screening program currently practiced in Germany, with 30% participation in two years and an interval of at least two years between the examinations. ²⁹ , ³⁰

• Three years: The parameters from the Current scenario remain the same, but the interval between examinations is at least three years.

• High Participation: This also utilizes the parameters from the Current scenario, though with a participation rate of ten more percentage points in people who participate “sometimes” (see below), inspired by increased participation in colorectal cancer screening due to an invitation system. ³¹

In all scenarios, people between 35 and 85 years of age are eligible for screening. It is assumed that primary screening is performed at the general practitioner's practice with a probability of 58%, and the remainder is performed directly by a dermatologist (estimates from the database at DAK‐Gesundheit). Sensitivity of the screening for detecting all extant skin tumors (MM, SCC, BCC) in dermatological practices is set at 91.7% in our model (otherwise: 85%) according to Gonna et al., and specificity is set at 81.5% in dermatological practices (otherwise: 75%). ³² Based on data from Datzmann et al. as well as the Versorgungs‐Report Früherkennung on participation rates, ³³ , ³⁴ we assume that a certain proportion of the population will seize any opportunity for screening (16%), another part will never participate (38%), and the remainder will attend “sometimes” (46%). The proportion of the population who participates in screening “sometimes” has a participation probability of 15.6% within any year in which they are eligible for screening.

10,000 populations with 100,000 women (or men) each were simulated for each scenario.

All results including costs were age‐standardized for better comparability (old European standard, rate per 100,000).

Non‐standardized results can be found in the online supplement (Tables S3 and S4).

Comparison with observational data

Observed data on incidence and mortality of MM in Germany are derived from the abovementioned selected cancer registries (2006–2016, incidence), and the “Causes of Death” statistics from GBE‐Bund (2007–2017, mortality). ¹⁷ , ³⁵

RESULTS

Model Validation

Model validation indicates good validity of the MM partial model, and moderate validity of the SCC partial model. For MM, the simulated combined distributions of incidence and mortality rates match well with those observed in the federal Länder in both sexes (Figure S1, online supplement). Scattering of the observed mortality rates is, however, somewhat higher than in the simulated population; scattering of the observed incidence rates is somewhat higher for women and lower for men than in the simulated population. In SCC, systematic over‐estimation of simulated mortality was found for both women and men.

Comparison of screening scenarios

After ten years, the Current scenario led to an about 3%–4% higher incidence of all analyzed skin cancer types than No screening. For the Three Years scenario, incidences are also somewhat higher than No screening, but lower than in the Current scenario (Table 1). The difference between screening every two and every three years is higher in men, who generally have higher incidence rates, than in women. Higher participation rates (scenario High Participation) led to a slight increase in incidence rates for MM, SCC, and BCC, but to less clearly increased raw incidence rates for SCC and for BCC in men (Table S3, online supplement).

TABLE 1.

Age‐standardized incidence, mortality, costs, and life years gained in the four screening scenarios, ten years after the introduction of HKS.

			Reference scenario	Difference to scenario
			No screening Median (1st and 3rd quartile)	Current Median (1st and 3rd quartile)	3‐year Median (quartile)	High participation Median (1st and 3rd quartile)
Incidence (ASR per 100,000)	MM	Women	15.61 (14.58; 16.63)	+0.43 (–0.88; 1.7)	+0.29 (–1; 1.55)	+0.5 (–0.81; 1.77)
		Men	14.74 (12.83; 16.71)	+0.52 (–1.85; 2.89)	+0.38 (–1.97; 2.74)	+0.6 (–1.78; 2.96)
		Total	15.18 (13.7; 16.67)	+0.48 (–1.36; 2.3)	+0.33 (–1.48; 2.14)	+0.55 (–1.29; 2.37)
	SCC	Women	25.23 (24.08; 26.39)	+1.02 (–0.39; 2.5)	+0.94 (–0.45; 2.43)	+1.05 (–0.36; 2.53)
		Men	50.05 (48.32; 51.81)	+2.15 (–0.05; 4.4)	+1.97 (–0.22; 4.24)	+2.17 (–0.02; 4.45)
		Total	37.64 (36.2; 39.1)	+1.58 (–0.22; 3.45)	+1.46 (–0.34; 3.34)	+1.61 (–0.19; 3.49)
	BCC	Women	126.19 (119.87; 132.79)	+4.49 (–3.21; 12.54)	+3.93 (–3.76; 11.86)	+4.62 (–3.09; 12.66)
		Men	161.87 (158.25; 165.55)	+6.00 (1.43; 10.94)	+5.25 (0.65; 10.1)	+6.12 (1.56; 11.02)
		Total	144.03 (139.06; 149.17)	+5.24 (–0.89; 11.74)	+4.59 (–1.55; 10.98)	+5.37 (–0.76; 11.84)
Mortality (ASR per 100,000)	MM	Women	1.53 (1.41; 1.65)	−0.2 (–0.35; –0.04)	−0.19 (–0.34; –0.03)	−0.24 (–0.4; –0.09)
		Men	2.51 (2.18; 2.84)	−0.33 (–0.72; 0.05)	−0.31 (–0.71; 0.08)	−0.41 (–0.8; –0.03)
		Total	2.02 (1.79; 2.24)	−0.26 (–0.53; 0.01)	−0.25 (–0.52; 0.03)	−0.32 (–0.6; –0.06)
	SCC	Women	0.52 (0.48; 0.56)	−0.01 (–0.06; 0.05)	−0.01 (–0.06; 0.05)	−0.01 (–0.07; 0.05)
		Men	1.61 (1.52; 1.69)	−0.04 (–0.16; 0.09)	−0.03 (–0.15; 0.09)	−0.05 (–0.17; 0.07)
		Total	1.06 (1; 1.12)	−0.03 (–0.11; 0.07)	−0.02 (–0.1; 0.07)	−0.03 (–0.12; 0.06)
Yearly costs in Euros (ASR per 100,000)	Total	Women	191,500 (185,400; 197,500)	+196,600 (188,400; 204,800)	+148,900 (140,700; 157,200)	+238,800 (230,600; 247,000)
		Men	238,000 (232,300; 243,900)	+197,900 (190,100; 205,700)	+150,800 (142,800; 158,500)	+239,900 (232,200; 247,800)
		Total	214,700 (208,900; 220,700)	+197,200 (189,200; 205,200)	+149,800 (141,800; 157,900)	+239,400 (231,400; 247,400)
Life years gained (ASR per 100,000)	MM	Women	−	9.1 (7.69; 10.57)	8.48 (7.2; 9.89)	10.96 (9.38; 12.6)
		Men	−	12.11 (10.17; 14.2)	11.26 (9.43; 13.27)	14.54 (12.33; 17)
		Total	−	10.61 (8.93; 12.38)	9.87 (8.31; 11.58)	12.75 (10.86; 14.8)
	SCC	Women	−	0.48 (0.26; 0.76)	0.43 (0.22; 0.7)	0.57 (0.32; 0.87)
		Men	−	2.25 (1.69; 2.87)	2.03 (1.51; 2.63)	2.66 (2.03; 3.34)
		Total	−	1.36 (0.98; 1.81)	1.23 (0.86; 1.67)	1.61 (1.17; 2.1)
Costs in Euros per life year gained	Total	Total	−	16,474.52	13,495.5	16,671.31

Abbr.: ASR, age‐standardized rate; BCC, basal cell carcinoma; MM, malignant melanoma; SCC, squamous cell carcinoma

All of the observed differences show a further increase over the next 20 years, except for the circumstance that the incidence rates of SCC and BCC in men will also be higher in the High Participation scenario than in the Current scenario (Table S5).

Mortality due to MM after ten years is about 13% lower for the Current scenario than for the reference scenario No Screening. Higher participation (scenario High Participation) leads to a further decrease in mortality (–3.5 percentage points). Screening offered at larger intervals (scenario Three Years) shows no relevant changes in mortality. Differences in SCC mortality are small in all scenarios.

There is also a small increase in the differences for mortality over the following 20 years (Table S5, online supplement, and Figure 2).

FIGURE 2.

Simulated time courses of melanoma mortality for all four screening scenarios.

Abbr.: ASR, age‐standardized rate

In the Current scenario, age‐standardized costs for skin cancer screening and treatment are almost twice as high as without screening. In the Three Years scenario, cost increases compared to No Screening are about 70% and thus somewhat lower than for the Current scenario. The High Participation scenario leads to the highest cost increase (+110.0%).

Screening according to the Current scenario leads to an annual gain of 12 years of life per 100,000 people after ten years. Triennial screening achieves about three‐quarters of a life year less per 100,000 people, while high participation yields an additional gain of about 2.5 life years.

Comparison of observed and simulated incidence and mortality rates for MM

The increase in incidences after introduction of skin cancer screening as predicted by the simulation model can also be found in similar form in the observed incidence data (Figure 3). As opposed to the predicted decrease of incidences in the following years, however, the observed incidences remain more or less constantly on the higher level, without any substantial changes.

FIGURE 3.

Melanoma incidence from 2006 to 2016 and melanoma mortality from 2007 to 2017 from the simulation scenario Current and from observations of the regions used for calibration; HKS 2008 was introduced both in the observed data and in the simulation.

Abbr.: ASR, age‐standardized rate

The simulation model predicts a continuous decrease in mortality, starting about two years after initiation of the screening program and stabilizing after about 25 years at a new level of 1.53 per 100,000 in women and 2.51 in men. Observed mortality shows a trend in the opposite direction which only reaches the original level again after about ten years.

DISCUSSION

Model validation

The MM model shows good accordance with the cancer registry validation data for both women and men. The SCC model, however, systematically overestimates mortality. This might be caused by underreporting of non‐lethal SCC cases. Thus, our SCC model remains generally suitable for comparisons between various simulated screening scenarios, but it should not be used as a reference for expected mortality rates. No validation from cancer registry data was possible for BCC; BCC incidence is not comprehensively documented due to the frequency of this diagnosis.

Comparison of the screening scenarios

As expected, the Current scenario leads to a higher skin cancer incidence, a lower skin cancer mortality (mainly caused by a decrease in MM mortality), higher costs, and life years gained compared with the No Screening scenario. In the Three Years scenario, all of these effects were somewhat lessened (the reduction in MM mortality, however, less than expected), while the effects increased for the High Participation scenario.

Unexpectedly, raw incidence rates for SCC and BCC after ten years of screening with higher participation are lower than in the Current scenario. We suspect that this is caused by the age structure of skin cancer patients with different entities as well as by the design of our model which determines that people can only participate in the screening program until they have received a skin cancer diagnosis. However, the effect is very small and disappears over the course of the following years.

If you calculate the additional costs per life year gained via Current screening programs as compared with No Screening, every life year gained costs about € 16,500. If screening was offered only on a triennial basis, each life year gained would cost € 13,500. High participation would result in costs of about € 16,100 per life year gained. Cost effectiveness is therefore comparable to mammography or cervical cancer screening. In 1994, costs of € 9,500 to € 12,900 (converted from German marks, the national currency in Germany until 2001) were estimated for every life year gained via mammography screening in Germany, ³⁶ and for biennial mammography screening of women aged 50–69 in high‐income countries, estimates vary from € 3,350 to € 19,000. ³⁷ Biennial PAP smears in women aged 20–29 and biennial HPV screening in women above 30 leads to estimated costs of € 28,400 per life year gained. ³⁸

The choice of scenarios presented here is currently still limited. In the future, we should investigate if risk‐adapted screening approaches such as earlier and/or more frequent screenings for people with increased skin cancer risk may gain more life years while still conserving financial resources.

If the aim is to utilize comparisons between scenarios to derive recommendations for improving skin cancer screening programs, it is necessary to weigh the various parameters: the desired goal (life years gained), the resources available (financial and human), and cost effectiveness (costs per life year gained).

Models can only be transferred into reality if the simulated interventions are comparable with those in real life. If skin cancer screening in real life is less sensitive than assumed, for example because investigations are frequently incomplete or are not performed with the necessary diligence, the promising results found in the models might conceivably not be achieved.

Comparison of observed and simulated incidence and mortality rates for MM

As opposed to the expectations from the simulation model, neither incidence (after the initial increase) nor mortality of MM showed a corresponding decrease after the introduction of skin cancer screening. While the simulation assumes a constant risk of disease and mortality, hitherto observed time courses indicate a background trend of increasing risks of morbidity and mortality; specifically, increases in morbidity risks have been observed in all of Europe. ⁴ , ³⁹ , ⁴⁰ , ⁴¹ As Hübner et al. have discussed for Germany, ⁴ there are considerable uncertainties about the strength and continuity of this background trend. In addition, we cannot simply assume that a situation “without screening” will immediately switch to a situation “with screening” once the screening program has been implemented. Early detection activities were indeed performed in the years before implementation of the screening program, although their extent is mostly unknown. Thus, we cannot draw any definite conclusions on model quality from divergences in observed and simulated incidence rates, respectively mortality rates. The primary goal of our simulation model was not to depict reality but to investigate differences in the results from varying screening scenarios. Our model is only suitable for a direct comparison with observed numbers inasmuch as a horizon of expectations for the results of skin cancer screening is determined and hypotheses on unknown factors in the real course of skin cancer can be generated.

Strengths and limitations

As compared with other simulation models, our microsimulation imitates many details from real‐life screening and skin cancer experiences. In spite of its high degree of complexity, however, our model also contains a number of necessary simplifications, specifically due to the fact that there are no or insufficient data available in some areas.

Firstly, participation patterns in real life vary greatly between individuals and over time, but our model uses three simple and distinct groups: One group participates in every possible screening opportunity, another does not make use of screening programs at all, and the third group attends “sometimes” with a pre‐determined probability. Group distribution and probability of participation of the “sometimes” group were determined so that observed participation rates were approximated after two and seven years. However, other combinations would also conform with the observed data. Moreover, the model does not incorporate any self‐selection effects. One might speculate that in real life, screening is primarily utilized by persons who have noticed suspicious alterations on their skin and who therefore do not experience much (additional) benefit from a physician's examination. In such a case, the simulation – just as an RCT – would overestimate the real effect at population level.

Secondly, in our microsimulation every person can get any of the three skin cancer entities only once. This has no influence on incidence and mortality, but it does have an influence on the simulated costs. In the simulation, treatment costs only accrue in the year of diagnosis since information is only available for this year. The model will therefore tend to underestimate costs.

Thirdly, the model underestimates overdiagnosis. The observational data on diagnoses at early stages show numbers far above the simulated estimates. We must assume that in our model, too few skin cancers occur in older age groups, and too few tumors with extremely slow growth are included. However, we are not aware of any existing meaningful data on this topic.

Overall, the quality of the model results depends on the validity and precision of the parameter values used. Our validation of the MM model shows that the interaction of the various parameters leads to reliable prediction of MM mortality. Validation of the SCC model, however, shows that the SCC data available in Germany have only limited informative value. Most cancer registries do not collect systematic data on this type of tumors, ⁴² and in particular we must assume that mainly tumors with poor prognosis are registered. This results in substantial overestimation of mortality. This kind of underreporting is even more pronounced for BCC, which is more common as well. There are also fewer published data on tumor biology of keratinocytic skin cancer entities, such as on tumor growth, which has an influence on the time frame allowing for early diagnosis within a screening program. This increases the risk of unsuitable input parameters for the model.

Conclusion

Our microsimulation model on skin cancer screening generated valid simulation results for MM. For SCC, we must assume a bias in that mortality is likely too high. The BCC model could not be validated.

In spite of its limitations, the microsimulation model can be used effectively to compare various screening scenarios regarding incidence, mortality, costs, and life years gained, so possible optimization potential can be identified.

The scenarios show that skin cancer screening, compared with no screening, has the potential to decrease skin cancer mortality substantially and to gain additional years of life. Higher participation could potentially increase this effect, while triennial screening would be only slightly less effective. Screening increases skin cancer incidence as well as costs. In the Current screening scenario, for example, every life year gained costs about € 15,800. The differences between the simulated data and the observed data may be caused by a lower than assumed quality of skin cancer screening in real life. Simulations such as these should in future be extended to include risk‐adapted screening approaches.

FUNDING

This study has received funding from the innovation fund of the Federal Joint Committee (Gemeinsamer Bundesausschuss G‐BA; reference 01VSF18002) according to §§ 92a and 92b SGB V.

CONFLICT OF INTEREST STATEMENT

M.A. has received consultation fees and/or has given lectures and/or has been remunerated for travel expenses and/or has taken part in clinical studies supported by the following companies: Almirall, DAK, Leo.

The other authors declare no conflicts of interest.

Supporting information

Supporting Information

Baltus H, Hübner J, Garbe C, et al. Evaluation of skin cancer screening in Germany – a Microsimulation. JDDG: Journal der Deutschen Dermatologischen Gesellschaft. 2025;23:19–27. 10.1111/ddg.15539