In the past, population-based cancer registration in West Germany was restricted to a few regions, whereas in East Germany a nationwide comprehensive cancer registry had been in operation. Meanwhile, population-based cancer registries have been built up in all federal states, and, since 2009, cancer diagnoses have been systematically registered in Germany nationwide. Most of the cancer registries reached a level of completeness of case ascertainment and data quality that allows regional comparisons of cancer burden in Germany.
Cancer statistics for Germany on the level of federal states are reported by the German Centre for Cancer Registry Data (ZfKD) and the Association of Population-based Cancer Registries in Germany (GEKID). The ZfKD, which receives the data from the cancer registries of the federal states, analyzes and evaluates the epidemiological cancer data for Germany and publishes the results in ‘Cancer in Germany’ every two years (Robert Koch Institute and the Association of Population-based Cancer Registries in Germany, 2014, 2015). In this report, cancer incidence and mortality by federal state are presented according to cancer site and sex as bar charts in comparison to the nationwide estimates for Germany (Fig. 1a). GEKID provides an interactive online cancer atlas for Germany (Association of Population-based Cancer Registries in Germany GEKID, 2015), visualizing the indicators using bar charts and choropleth maps (Fig. 1b). The GEKID Atlas presents, in addition to incidence and mortality, 5-year cancer survival on the level of federal states.
‘Cancer in Germany’ and in particular the GEKID Atlas make cancer data easily accessible for researchers as well as for the public and allow the comparison of cancer incidence, mortality, and survival between federal states. However, the visualization using bar charts and cancer maps can lead to misinterpretation of the presented figures. Ranked bar charts, as used in ‘Cancer in Germany’ and the GEKID Atlas, are difficult to interpret, as they imply a rank ordering of the compared regions. However, such ranks are very unreliable statistical summaries, and there is great uncertainty with regard to the true ranks (Marshall and Spiegelhalter, 1998). Furthermore, in the presentation of the German regional cancer statistics using bar charts, the variability expected from random variation is not taken into account, which varies with the precision of the estimates, and thus with the size of the population. For small populations, fluctuation is high by chance alone, which results in an overrepresentation at the ends of the rankings. Choropleth maps can also be misleading, as the maps reflect the size of geographic areas rather than the size of the underlying populations, and the appearance of the maps depends on the choice of the categorization used for the color-coding of the displayed figures.
Funnel plots are a visualization tool in which an estimate of an underlying quantity is plotted against a measure of its precision, with ‘control limits’ forming a funnel around a target outcome (Spiegelhalter, 2005). Funnel plots are a standard tool within meta-analysis for detecting publication bias, but were also strongly recommended by Spiegelhalter for comparisons of institutional performance, as they avoid spurious ranking of institutions and emphasize the increased variability expected for smaller institutions (Spiegelhalter, 2005). Funnel plots are very flexible and can be used with a variety of indicators. Recently, funnel plots were used with population-based cancer registry data as well, for example, to assess geographical variation in cancer mortality and survival in the UK (Davies et al., 2008; Rachet et al., 2009; Walters et al., 2011), to assess trends in incidence and relative survival in Japan (Ito et al., 2009; Katanoda et al., 2012), and in a worldwide cancer survival study (CONCORD) (Coleman et al., 2008).
The present study aimed to compare cancer incidence, mortality, and survival in Germany on the level of federal states using funnel plots.
Materials and methods
For the present study, publicly available data were used. Data on sex-specific cancer incidence, mortality, and survival for Germany by federal state and on a national level for the year 2011 were retrieved from the online atlas published by GEKID, the Association of Population-based Cancer Registries in Germany (data status December 2014, retrieval 22 August 2016) (Association of Population-based Cancer Registries in Germany GEKID, 2015). All cancer sites combined (ICD-10 C00-C97 excluding C44+D09.0+D41.4) and the most common cancers were considered for this analysis: cancer of the colorectum (C18–C21), lung (C33–C34), female breast (C50), and prostate (C61). The cancer incidence statistics presented in the GEKID Atlas are based on the new cancer cases registered by the population-based cancer registries of the federal states, which submit their data annually to GEKID. Incidence data for Baden-Württemberg (BW) were not included in the atlas, because the cancer registry of this state was still in the process of being built up. As not all federal cancer registries have reached a complete registration yet, the national incidence for Germany was estimated using a pool of selected regions with high-quality data. Cancer mortality data originate from the German Federal Statistical Office. Data on cancer survival for regions with sufficient data quality were provided by a collaborative project between the population-based German cancer registries and the German Cancer Research Center (DKFZ). Further details with regard to the statistics presented in the GEKID Atlas can be found in the methodological notes of the atlas (Association of Population-based Cancer Registries in Germany GEKID, 2015).
Information on completeness of cancer registration by federal state and cancer site originates from the ZfKD (Robert Koch Institute and the Association of Population-based Cancer Registries in Germany, 2015). Because of different time and conditions of operation, data quality of the German cancer registries is quite heterogeneous. The ZfKD annually checks the completeness of the data received from the population-based cancer registries in Germany. In 2012 cancer registration reached a completeness of case ascertainment of more than 90% in 11 of the 16 federal states.
Table 1 shows the size of the population (data as of 31 December 2011 retrieved from the Federal Statistical Office of Germany, 2016) and the number of registered incident cases and deaths for all cancers combined in 2011 by federal state and for Germany in total (a map of the German federal states is shown in Supplementary Fig. 1, Supplemental digital content 1, http://links.lww.com/EJCP/A196). There is a large variation between cancer registries in the size of the covered region and population, the latter ranging from 0.7 million in the city state of Bremen to 17.5 million in North Rhine-Westphalia.
As described by Spiegelhalter (2005), a funnel plot consists of four components: an indicator Y (e.g. an observed outcome measure), a target value θ for the indicator (specifies the expected value for the outcome), a precision parameter ρ, and control limits yα for given significance levels α (see Supplementary Fig. in Supplemental digital content 2, http://links.lww.com/EJCP/A197, which illustrates the components of a funnel plot). The precision parameter ρ can be taken as a function proportional to the inverse variance and should be selected as a directly interpretable measure. The significance level is usually chosen as α of 5% and α of 0.2% resulting in control limits of 95 and 99.8%, respectively (corresponding ~2 and 3 SDs around the target value). The control limits are independent of the observed data and form a ‘funnel’ around the target value (shown as a horizontal line), the wider part reflecting increased variability from less precise estimates. Funnel plots allow assessing visually which observations are within the range of statistical variation (within the funnel) and which are significantly different from the target value (outside the funnel) and may need further investigation.
In the present study, funnel plots were used to compare cancer incidence, mortality and relative survival in Germany on the level of federal states by sex and cancer site. The cancer statistics were plotted against a measure of their precision, and the target values were taken as the national estimates for Germany. The control limits for the funnel plots were set at 95 and 99.8%. Estimates falling outside the control limits represent the federal states showing wider disparity from the nationwide estimate for Germany than could be expected because of chance alone.
The control limits yα correspond conceptually to the 100 (1−α)% confidence interval. The methods for the calculation of the control limits and the construction of the funnel plots for the regional comparisons of the cancer statistics are described in detail in Supplementary Material (Supplemental digital content 2, http://links.lww.com/EJCP/A197).
Figure 2 shows funnel plots of age-standardized incidence rates for 15 federal states in Germany in 2011 by sex for all cancer sites, cancer of the colorectum, lung, female breast, and prostate in comparison with the national estimates for Germany. In some federal states the registration was not complete for all or certain cancer sites, which means that their incidence rates were underestimated and hence not immediately comparable to the national estimate. A level of completeness of case ascertainment lower than 90% is indicated in the funnel plots by highlighting in light shade.
The overall cancer incidence rate was highest in Hamburg (HH) and North Rhine-Westphalia (NW) among women and in NW among men. In women, the incidence rate was also considerably above the upper 99.8% control limit for Schleswig-Holstein (SH) and Lower Saxony (NI). Among the states with complete registration, incidence was significantly lower than average for Bavaria (BY) in men and women, and for the eastern states Brandenburg (BB), Thuringia (TH), Saxony (SN) in women.
For colorectal cancer, the incidence rates fell within the control limits for the majority of states. Only in NW the incidence rate was substantially higher than the national estimate among women. Only the incidence rates of states with incomplete cancer registration fell below the 99.8% control limit.
Lung cancer incidence showed wide regional variation and a similar pattern for women and men, except for Mecklenburg-West Pomerania (MV) and Saxony-Anhalt (ST), where the incidence among men was higher and among women lower than average. The incidence rates both in women and men were above the 99.8% control limit for the city-states Berlin (BE), Bremen (HB), HH as well as for NW, SH, and Saarland (SL). Furthermore, the incidence rates were above the 99.8% control limit in NI among women and in MV among men. Significantly lower than the nationwide estimate was the incidence rate for BY in women and men, and for TH and SN in women.
The incidence rates of breast cancer were significantly higher than the nationwide estimate for HH, NI, NW, and SH and lower in the eastern federal states BB, MV, SN, TH, and ST, the latter with incomplete registration. For prostate cancer, outliers with higher incidence than the estimate for Germany were NI, NW, and SH. A significantly lower incidence was found in BY. In the other federal states with incidence estimates below the 99.8% control limit, the registration was incomplete.
The funnel plots in Fig. 3 display age-standardized cancer mortality for all German federal states by sex for all cancer sites, cancer of the colorectum, lung, female breast, and prostate in comparison with the average mortality for Germany.
The overall cancer mortality was substantially above the upper 99.8% control limit among women in BE, HH, NW, SL and among men in MV, NW, SL, and ST. Significantly lower mortality compared with Germany was found for BW and BY both in women and men as well as for SN in women and for Hesse in men.
The mortality of colorectal cancer showed little variation with only few outliers. Among women, the mortality rate was above the upper 99.8% control limit in NW and below the lower 99.8% control limit in BW and BY. Among men, only for SH a significantly lower than average mortality rate was found.
For lung cancer, the regional variation in mortality showed a similar pattern to incidence for most of the states. Exceptions were for women in NI and BB, where mortality was not significantly different from average and below the 99.8% control limit, respectively. Among men, the mortality of lung cancer in HB and SH did not exceed significantly the German estimate, and, in ST, it was significantly higher than average, in contrast to incidence.
The highest mortality rates of female breast cancer were observed for HB, HH, and SL and were slightly above the upper 99.8% control limit. Lower outliers with regard to breast cancer mortality were SN and TH. For prostate cancer, the mortality fell within the 99.8% control limits for all states but NW and ST, where the mortality was significantly higher than the average for Germany.
The results of the regional comparisons of cancer survival are shown in Fig. 4. The funnel plots display age-standardized 5-year relative survival by sex and federal state for cancer of the colorectum, lung, female breast, and prostate in relation to the survival estimate for Germany. For all investigated cancer sites, the survival estimates fell within the 99.8% control limits for almost all federal states, both among men and women. Only lung cancer survival in NI was below the 99.8% control limit among women.
The regional comparisons of cancer estimates in Germany using funnel plots showed large variation in incidence and mortality between federal states, with many outliers in relation to the national estimates, whereas for cancer survival, essentially no excess variation could be observed. One outlier, a federal state for which all survival estimates were suspiciously low, could be traced back to erroneous data, which subsequently were corrected in the GEKID Atlas.
Regional differences in cancer incidence and mortality varied by sex and cancer site. The incidence rates of states with incomplete registration were in general lower than average and often below the 99.8% control limit. The largest variation in incidence and mortality could be observed for all cancers combined and lung cancer in both sexes and for breast cancer in female individuals. Incidence and mortality in NW were consistently and, in most cases, substantially higher than average for the investigated cancer sites in both sexes. In contrast, the rates for BW and BY were most often significantly lower than the national estimates. The eastern federal states with the exception of BE showed lower incidence and mortality than average for lung and breast cancer among female individuals.
Underlying causes of regional disparities in cancer incidence and mortality are complex and interrelated. Cancer disparities are linked to differences in distribution of environmental and lifestyle risk factors for cancer as well as access to healthcare (screening, treatment). As socioeconomic status is associated with behavioral risk factors for cancer and screening attendance, variation in social structure is responsible to a large extent for regional cancer disparities. The high lung cancer rates in the city-states BE, HH, and HB as well as in NW and SL, traditional mining and heavy industry states, may be related to smoking behavior and air pollution, both risk factors being increased in large cities and industrial regions. The lower lung cancer rates in the southern federal states can be explained by the lower smoking prevalence in these regions. The difference in female breast cancer rates between eastern and western federal states can be attributed to different reproductive behaviors (age at first birth, parity) and hormone replacement therapy use in the two former German countries. Variation in prostate cancer rates may reflect regional differences in prostate-specific antigen testing.
The results of the regional comparisons should be interpreted in the light of limitations of cancer incidence and survival data from the German cancer registries. Variation in the quality of registration data (completeness of registration, proportion of cases notified by death certificate only, follow-up of vital status) could contribute to the regional differences identified in cancer incidence and survival. Further research should aim to clarify potential causes for regional cancer health disparities.
Funnel plots proved to be a suitable tool for regional comparisons of cancer incidence, mortality, and survival between federal states in Germany. Funnel plots allow a fair comparison of the estimates for the federal states against the national estimates by taking into account increased variability in small populations and avoiding spurious ranking. The advantage of using funnel plots for regional comparisons is illustrated by the example in Fig. 5, which shows prostate cancer mortality by federal state in Germany in 2010 and 2011 presented as ranked bar charts and funnel plots, respectively. For a state with a small population such as HB there can be a substantial change in rank between different years (Fig. 5a). The visual representation using funnel plots, however, shows that the mortality rates for HB fall inside of the control limits for both years, indicating expected variability (Fig. 5b).
Accounting for statistical variation, funnel plots allow identifying outliers which may require further investigations and are thus an appropriate aid to health service decision-making (Marshall et al., 2004). Funnel plots can be used for routine regional comparisons and therefore help to improve population-based health management and eliminate cancer health disparities. Furthermore, funnel plots are also well suited for the presentation of comparative cancer data to a general audience, as they can be readily understood by lay people (Rakow et al., 2015).
The authors thank our colleague Dorothee Twardella for her helpful and critical comments on the manuscript.
Conflicts of interest
There are no conflicts of interest.
Association of Population-based Cancer Registries in Germany
(GEKID) (2015). Atlas of cancer incidence
and mortality in Germany
(GEKID Atlas). Data submission December 2014. Lübeck. Available at: http://www.gekid.de/
. [Accessed 22 August 2016].
Coleman MP, Quaresma M, Berrino F, Lutz JM, De Angelis R, Capocaccia R, et al (2008). Cancer survival
in five continents: a worldwide population-based study (CONCORD). Lancet Oncol 9:730–756.
Corazziari I, Quinn M, Capocaccia R (2004). Standard cancer patient population for age standardising survival ratios. Eur J Cancer 40:2307–2316.
Davies E, Mak V, Ferguson J, Conaty S, Moller H (2008). Using funnel plots to explore variation in cancer mortality
across primary care trusts in South-East England. J Public Health (Oxf) 30:305–312.
Federal Statistical Office of Germany
(2016), Population data. Available at: http://www.destatis.de/
. [Accessed 2 September 2016].
Ito Y, Ioka A, Tsukuma H, Ajiki W, Sugimoto T, Rachet B, et al (2009). Regional differences in population-based cancer survival
between six prefectures in Japan: application of relative survival models with funnel plots. Cancer Sci 100:1306–1311.
Katanoda K, Ajiki W, Matsuda T, Nishino Y, Shibata A, Fujita M, et al (2012). Trend analysis of cancer incidence
in Japan using data from selected population-based cancer registries. Cancer Sci 103:360–368.
Marshall EC, Spiegelhalter DJ (1998). Reliability of league tables of in vitro fertilisation clinics: retrospective analysis of live birth rates. BMJ 316:1701–1704; (discussion 1705).
Marshall T, Mohammed MA, Rouse A (2004). A randomized controlled trial of league tables and control charts as aids to health service decision-making. Int J Qual Healthcare 16:309–315.
Rachet B, Maringe C, Nur U, Quaresma M, Shah A, Woods LM, et al (2009). Population-based cancer survival
trends in England and Wales up to 2007: an assessment of the NHS cancer plan for England. Lancet Oncol 10:351–369.
Rakow T, Wright RJ, Spiegelhalter DJ, Bull C (2015). The pros and cons of funnel plots as an aid to risk communication and patient decision making. Br J Psychol 106:327–348.
Robert Koch Institute and the Association of Population-based Cancer Registries in Germany
(2014). Cancer in Germany
2009/2010, 9th ed. Berlin: Robert Koch Institute.
Robert Koch Institute and the Association of Population-based Cancer Registries in Germany
(2015). Cancer in Germany
2011/2012, 10th ed. Berlin: Robert Koch Institute.
Spiegelhalter DJ (2005). Funnel plots for comparing institutional performance. Stat Med 24:1185–1202.
Walters S, Quaresma M, Coleman MP, Gordon E, Forman D, Rachet B (2011). Geographical variation in cancer survival
in England, 1991–2006: an analysis by Cancer Network. J Epidemiol Community Health 65:1044–1052.