Temporal patterns of childhood cancer survival 1991 to 2016: A nationwide register‐study based on data from the German Childhood Cancer Registry

Childhood cancer is the leading disease‐related cause of death among under 15 year olds in Europe. Since primary preventive measures are lacking, improving survival probabilities and long‐term well‐being remain primary goals. With this report, we provide the first long‐term assessment and interpretation of patterns in childhood cancer survival in Germany, covering a period of 30 years. Using data from the German Childhood Cancer Registry, we assessed temporal patterns of cancer survival among children (0‐14 years) diagnosed in Germany from 1991 to 2016, by cancer type, age at diagnosis and sex. We calculated overall survival (OS) and average annual percentage changes of the respective 5‐year OS estimates. OS improved across all cancer types, age groups as well as for boys and girls over time. Five‐year OS for all childhood cancers combined increased from 77.8% in 1991‐1995 to 86.5% in 2011‐2016, with stronger improvements during the early 1990s. The most pronounced survival improvement was seen for acute myeloid leukaemia, at 2% annually and 5‐year OS recently reaching 81.5%. Survival improvements for some diagnoses such as neuroblastoma, renal tumours and bone tumours have flattened out. Tremendous enhancements in diagnostics, treatment and supportive care have affected average survival improvements for most cancer types. Recently, survival improvements have decelerated overall and for some cancer types, it plateaued at an unsatisfactory level. As not all children benefited equally from the survival improvements, personal factors (eg, socioeconomic circumstances, health literacy, access to care) likely affect individual prognosis and warrant further investigation.


| INTRODUCTION
Cancer is the leading disease-related cause of death among children in Europe and other high-income countries (HICs). [1][2][3] Even though its annual incidence is comparatively low, with about 170 cases per million children in Germany (ages 0-14 years), 4,5 it is associated with a high burden of disease for patients, their relatives and for the public health. In addition to the risk of death, intensive treatment at a young age and the patients' long remaining life spans put them at risk of somatic late effects as well as adverse psychosocial and socioeconomic consequences later in life. [6][7][8][9] Childhood cancer represents a heterogeneous group of diseases, including malignancies and nonmalignant tumours in the central nervous system (CNS), showing a wide variety of patterns regarding incidence, 5,10,11 aetiology, [12][13][14][15] treatment 8,16,17 and clinical outcomes. 5,[18][19][20][21] In HICs, leukaemias, tumours of the CNS and lymphomas are the most frequent cancers in children. 1,5,11 The aetiology of most childhood cancers remains poorly understood, with established risk factors explaining only about 10% of all incident cases. 13,14 Consequently, approaches for primary preventive measures are lacking, rendering treatment effectiveness to improve survival especially important in reducing the overall burden of disease. Over the past five decades, advances in the understanding of tumour biology, diagnostics, pharmacology, adapted treatment combinations and risk grouping and supportive care have indeed led to remarkable enhancements in the treatment of and survival from childhood cancer, with 5-year survival exceeding 80% in most HICs today. 8,16 Notably, however, not all children benefit equally from these recent diagnostic and therapeutic improvements. Disparities in childhood cancer survival are particularly evident according to cancer type and age at diagnosis. 1,5,8,19,22,23 Moreover, large survival disparities have been observed between countries, with a general pattern of higher survival in countries ranking higher on the Human Development Index. 24 Within Europe, about 10% poorer outcome overall is reported for Eastern nations vs Northern or Western Europe. 8,18,19,22,25,26 Given the heterogeneity across cancer types and the lack of primary preventive measures, further advances in childhood cancer survival remain the primary goal, enabling a higher life expectancy for every affected child.
With this report, we sought to provide the first long-term assessment and interpretation of patterns in childhood cancer survival in Germany using data from the national German Childhood Cancer Registry (GCCR) covering the diagnostic years 1991 to 2016. The nationwide population-based data offer a high granularity for evaluating persisting survival disparities, relating survival patterns to changes in diagnostics and therapeutic regimes, and thus contributing to a better understanding of long-term temporal trends and persistent disparities in childhood cancer survival.

| Study population and data source
The study population comprised all incident cancer diagnoses reported to the GCCR from 1991 to 2016 (including subsequent primary neoplasms [SPNs]) among children aged 0 to 14 years residing in Germany at time of diagnosis. SPNs were defined as any new primary tumours diagnosed in a child (0-14 years of age) with a previous cancer diagnosis, regardless of the time between the two diagnoses. 5 The GCCR was established in 1980 and has been collecting and monitoring data on incident diagnoses of cancers in children aged 0 to 14 years (since 2009 expanded to 0-17 years) with a completeness of incident cases exceeding 95%. 5 The cancer diagnoses are classified according to the International Classification of Childhood Cancer-third edition (ICCC-3). 27

| Statistical analysis
We defined 5-year overall survival (OS) as our main outcome of interest and censored follow-up at 5 years after diagnosis. Patients were followed from date of cancer diagnosis until death from any cause, emigration, end of the 5-year follow-up, or March 15, 2022, whichever came first.
Using the Kaplan-Meier method, we calculated 1-, 3-and 5-year OS estimates with corresponding 95% confidence intervals (95% CIs) for all cancers combined, as well as stratified by diagnostic group, sex and age group at diagnosis per diagnostic period. We also calculated the sex ratio of the 5-year OS estimates (male OS/female OS) and corresponding pointwise 95% CI, based on Katz' method and adapted by using the Kaplan-Meier survival estimates and Greenwood's variance estimate. 28,29 Since the age distribution of Germany's child population has changed over the past decades, all OS estimates were adjusted for age using the method established in the EUROCARE-6 project. 23 We thus calculated age-adjusted OS by applying weights according to the age distribution of the respective diagnostic group in Germany during 1991 to 2016 (Table S4). In additional analyses, we calculated ageadjusted OS using the weights from the EUROCARE-6 project (considering only tumours with malignant behaviour, since EUROCARE included solely malignant tumours) to enable a direct comparison to the European survival estimates of the EUROCARE-6 study. 23 To assess the temporal changes, Joinpoint regression models 30 were applied using the German age-weighted 5-year OS estimates and corresponding standard errors to derive average annual percentage changes (AAPC) and corresponding 95% CI for 5-year OS estimates.
We allowed up to five joinpoints.
Statistical analyses were performed using SAS Statistical Software 9.4 31

| Overall survival by cancer type
Five-year OS for all childhood cancers combined improved from 77.8% in 1991-1995 to 86.5% in 2011-2016 (Table 2). This pattern was evident across all diagnoses and all age groups. In the most recent diagnostic period studied, the highest survival probabilities were T A B L E 1 Characteristics and quality criteria of the study population comprising children diagnosed with cancer below the age of 15 years in Germany by diagnostic group, sex, age at diagnosis and diagnostic period (German Childhood Cancer Registry, 1991 Survival estimates were similar to those of the main analysis after applying the EUROCARE-6 study age weights.

| Overall survival by age and sex
Across age groups, 5-year OS has increased similarly since 1991, by about 10 percentage points. During the entire study period, survival probabilities were higher in children aged 1 to 4 years at diagnosis than in older or younger children ( Table 2, Figure 3A), primarily driven by the superior prognosis of the large group of 1-to 4-year olds with LL ( Figure 3B,C). Children aged <1 year at diagnosis showed lowest survival for LL and CNS tumours, yet highest survival for neuroblastoma ( Figure 3B-F). We also observed sex-specific survival differences over time for some diagnostic groups. Survival for LL and malignant CNS tumours appeared to have been slightly higher in girls during the 1990s and rather similar to boys in more recent years. Sex ratios for survival from hepatic tumours indicated somewhat better outcomes for boys compared to girls during the 1990s. While recent sex ratios for germ cell tumour survival indicate superior survival for girls, we observed indications of superior survival in boys with AML and non-Hodgkin lymphoma (NHL) ( Table 2). although the average estimated slope decreased from 1.6% to 0.4%

| Temporal trend analysis
per year, the latter point estimate and its CI do represent an ongoing improvement of survival probabilities. This trend was mostly driven by girls (Table 3). Accordingly, the difference between 1-and 5-year OS decreased over time from around 13 to 8 percentage points ( Figure 1A). OS survival improved across all age groups. The Joinpoint regression revealed one joinpoint for the age group 5 to 9 years at diagnosis indicating a stronger increase in OS from 1991 to 2003 followed by a period of overall unchanged survival probabilities (Tables 2 and 3).
Analyses by cancer type demonstrated some differences in the temporal survival patterns. On average, 5-year OS for leukaemia improved by 0.6% (95% CI: 0.5-0.7) annually. The increase was largely due to the survival improvements in AML patients with an increase of 30 percentage points from 51.3% to 81.5% (AAPC = 2.0%, 95% CI: 1.5-2.5), which was the strongest enhancement across all cancer types (Tables 2 and 3). Notably, this trend was least pronounced in infants ( Figure 3C). The gap between LL and AML 5-year OS narrowed over time from almost 37 in 1991 to 1995 to 13 percentage points in 2011 to 2016 (Tables 2 and 3). Overall, the difference between 1-and 5-year OS in leukaemia narrowed considerably during the study period. This pattern was evident for both LL and AML ( Figure 1B,C).  Figure 1D). Survival for nonmalignant CNS tumours has improved slightly and levelled off at around 98% in the most recent years studied ( Figure 1E). Five-year OS of neuroblastoma only increased during the early 1990s (AAPC = 6.8%, 95% CI: 1.2-12.8 in 1991-1996) and remained largely unchanged at around 80% after the mid-1990s, the AAPC for bone tumour survival indicated plateauing survival probabilities (Tables 2 and 3, Figure 1F,G). For soft tissue sarcomas, we observed a steady 0.6% annual improvement in 5-year OS over the entire study period (Table 3).

F I G U R E 1 (Continued)
Due to low incidence and mortality, the Joinpoint regression method was not suitable for generating AAPCs for Hodgkin lymphoma, retinoblastoma, germ cell tumours, epithelial tumours and melanomas and for the group of "other malignant neoplasms." However, the period-specific 5-year OS estimates also indicated survival improvements for these cancer types (Table 2). The Joinpoint regression analysis indicated the strongest improvements were in AML survival. Indeed, the treatment protocols for AML have been constantly revised, including tailoring risk stratification and improving second-line treatment. 16,39,40 During the 1990s, AML relapse was a highly fatal event and, as such, a major obstacle for improving the overall prognosis of patients affected by AML. Since 1997, AML relapse treatment has been increasingly standardised, including safer conditioning regimes for stem cell transplantation (SCT). 41 Finally, the implementation of the AML SCT-BFM 2007 trial was likely the main contributor to the tremendous survival improvements for AML and offered a realistic chance for cure for more than half of the children with AML relapse. 41,42 Adapted from BFM studies, NHL treatments showed only moderate efficacy during the 1990s. 16 Treatment of mature B-ALL or B-NHL according to the BFM trail group was and still is very toxic.   Germany during the 1990s compared to Western Germany and was followed by strong improvements. 64 An overview of the international comparison of patterns in childhood cancer survival is given in Table S5.

| Strengths and limitations
First, this present study is strengthened by the high-quality data from the GCCR, with a high degree of completeness of cases (>95%) 5 and follow-up information (98.9%). Germany's sizable population permitted stratified analyses with high statistical power. This resource offers a solid basis for future investigations on selected disease groups, allowing researchers to zoom in on characteristics specific to certain childhood cancer subtypes and other clinical details. Moreover, the age-adjusted survival estimates following the methodology of the EUROCARE-6 project and OS estimates based on EUROCARE-6 age weights enabled international comparisons across Europe for the first time.
As the data availability of EUROCARE-6 depended on many European cancer registries, the data presentation differed to some extent from our study. Although we applied the same methodology for age standardisation, the comparison of survival estimates should be interpreted with caution, as the age distribution used for the survival analysis was data driven in both studies. For future international investigations on childhood cancer survival, we would recommend conducting survival analyses while applying an external standard population, as has been recently described by Miranda-Filho et al. 69 Since other international investigations on childhood cancer may have been influenced by respective population size, organisation of healthcare, cancer registration and possible differences in data quality, some points for discussion are speculative and comparisons should also be made with caution.

| CONCLUSIONS
The tremendous improvements observed in childhood cancer survival are likely due to advances in diagnostics and risk-adapted treatment allocation. Nevertheless, not all affected children benefited equally.
Some childhood cancer types continue to be associated with poor outcomes or survival recently plateauing at an unsatisfactory level.
Social and socioeconomic inequalities seem to influence survival prob-