Annals of Oncology Advance Access published online on August 13, 2007
Annals of Oncology, doi:10.1093/annonc/mdm194
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
© 2007 European Society for Medical Oncology
Timely disclosure of progress in childhood cancer survival by ‘period’ analysis in the Automated Childhood Cancer Information System
1 Data Analysis and Interpretation Group, International Agency for Research on Cancer, Lyon, France
2 Division of Clinical Epidemiology and Aging Research, German Cancer Research Center, Heidelberg, Germany
3 Clinical Trial Service Unit and Epidemiological Studies Unit, University of Oxford, Oxford, UK
4 Department of Preventive and Predictive Medicine, Epidemiology Unit, Istituto Nazionale per lo Studio e la Cura dei Tumori, Milano, Italy
* Correspondence to: Dr Eva Steliarova-Foucher, Data Analysis and Interpretation Group, 150 cours Albert Thomas, F-69372 Lyon Cedex 08, France. Tel: +33 (0)4 72 73 84 66; Fax: +33 (0)4 72 73 86 50; E-mail: steliarova{at}iarc.fr
 |
Abstract |
|---|
 Top Abstract introductionpatients and methodsresultsdiscussionAcknowledgementsReferences |
|---|
Background: A few years ago, a new method of survival analysis, denoted ‘period’ analysis, was introduced to provide more up-to-date survival estimates of cancer patients.
Patients and methods: We evaluated the period survival method using the large database of the Automated Childhood Cancer Information System (ACCIS). Our evaluation is based on data from 35 191 children diagnosed with cancer in 13 European countries between 1975 and 1989 and followed for vital status until around 1999.
Results: Using the follow-up data available in 1989, 10-year survival for all children with cancer calculated by the period method for the 1985–89 period was 58%, while it was 43% when calculated by traditional ‘cohort’ life-table analysis (based on children diagnosed in 1975–79). The period method provided a better estimate of the true 10-year survival of 62%, observed 10 years later in the cohort of patients diagnosed in 1985–89. Similar results were observed for each of the common groups of childhood cancer.
Conclusion: Period analysis is especially useful for monitoring childhood cancer survival, because at a given point in time it provides more timely estimates of long-term survival expectations than the cohort life-table method. Using the ACCIS database, up-to-date estimates of period survival for childhood cancer are derived in subsequent papers in this journal.
cancer registries, childhood cancer, population-based, prognosis, survival
|
introduction |
|---|
|
TopAbstractintroductionpatients and methodsresultsdiscussionAcknowledgementsReferences |
|---|
In recent decades, survival of children with cancer has improved dramatically in the populations of the developed world, mainly due to progress in therapy [1–6]. Nevertheless, the full extent of this improvement has only been disclosed with substantial delay to clinicians, the patients, their families and the public, when the calculation was based upon the ‘cohort’-based method, as used traditionally for survival analysis of population-based data.
For example, the EUROCARE project reported in 2001 and 2002 [2, 3] 5-year cohort survival for children diagnosed in various European countries in 1978–89 and 1978–92, respectively. The most recent report on childhood cancer survival from the EUROCARE study, dated 2005, included children diagnosed up to 1994 [5]. With the traditional cohort-based type of analysis used in these studies (as in most other studies of survival), reliable 10-year survival estimates for children diagnosed in those years may only become available another 5 years later. In the meantime, however, there have been further improvements in therapy, and it would be highly desirable to disclose the resulting improvement in survival figures in a timely manner.
Changes in survival of cancer patients can be detected earlier if the ‘period survival method’ is used [7, 8]. This method has been thoroughly evaluated and found useful in the monitoring of long-term survival of adult cancer patients [9–12]. A recent analysis using data from the Surveillance, Epidemiology and End Results (SEER) Program of the United States National Cancer Institute suggested that the method might be particularly appropriate for childhood cancers, where progress in therapy has been generally faster and improvement in prognosis more pronounced than for most cancers of adults [13]. However, due to sample size limitations, the analysis of SEER data was restricted to a few major groups of childhood cancer.
By application of the period method to the world's largest database of childhood cancer collected within the Automated Childhood Cancer Information System (ACCIS) [4], we are able to provide up-to-date estimates of survival for different types of childhood cancer in a series of four subsequent papers in this journal [14–17]. The aim of this paper is to demonstrate the difference between period and cohort methods and to assess, using historical data, how closely the different methods predict the survival actually experienced by the patients.
|
patients and methods |
|---|
|
TopAbstractintroductionpatients and methodsresultsdiscussionAcknowledgementsReferences |
|---|
database
The ACCIS database contains some 160 000 records of childhood and adolescent cancer cases registered over the past 30 years in 78 European population-based cancer registries, covering 1300 million person-years [4]. The data were verified in collaboration with participating registries, analysed centrally and a dataset from each registry was evaluated for comparability [18]. The tumours were classified according to the International Classification of Childhood Cancer (ICCC) [19].
For the current analysis, all childhood cancer cases diagnosed between 1975 and 1989 in children aged 0–14 years in 15 registries in 13 European countries were extracted from the ACCIS database (Table 1). Only those registries with first incidence year 1977 or earlier, and with follow-up for vital status complete until the end of 1997 or later were included in the analyses. Records of patients notified by death certificate only (0.2%) or with missing information on vital status (1.2%) were excluded from the analysis.
|
The registries included in the analysis cover different parts of Europe; eight of them are national. Almost half of all cases were contributed by the childhood cancer registry of England and Wales, UK. Most registries provided records for children diagnosed within the entire time window 1975–89, and had complete follow-up until the end of 1999 (Table 1). Due to non-comparable coding of childhood cancer types of interest in this paper, two registries (national registries from the Czech Republic and Sweden) contributed to the analyses only for all forms of cancer combined. These two datasets represented about 19% of the total dataset.
Statistical methods
Focusing on 10-year survival, we compared three methods of analysis: two traditionally used methods [20], the ‘cohort’ analysis (restricted to cohorts of patients with complete 10-year follow-up), the ‘complete’ analysis (including all patients with complete 10-year follow-up as well as those with censored survival time) and the ‘period’ analysis [7, 8]. The group of patients and the years of follow-up included in each analysis are illustrated in Figure 1. The cohort analysis was applied to the cohort of patients diagnosed between 1975 and 1979 who had a complete 10-year follow-up by the end of 1989. In Figure 1, this corresponds to the upper horizontal black rectangle. The grey trapezium in Figure 1 represents the patients diagnosed in 1975–89, followed for between 0 and 10 years by the end of 1989, who were included in the complete analysis. The third method, period analysis, is represented in Figure 1 by the vertical rectangle with dashed borders, and reflects the survival experience observed in 1985–89 for the patients diagnosed between 1975 and 1989.
|
In all three methods, life-table estimates of cumulative survival are obtained by multiplication of the conditional probabilities of survival of the first 10 yearly follow-up intervals. In contrast to the traditional methods, the period analysis reflects conditional survival probabilities of the children with cancer prevalent in 1985–89 only. Survival probability for the first year after diagnosis is obtained from patients diagnosed between 1984 and 1989, conditional survival probability for the second year is obtained from children diagnosed in 1983–88, and so on, until the survival probability in the tenth year following diagnosis, which is obtained from experience of the children diagnosed in 1975–80. For a detailed description of the period analysis methodology please refer to a recent review article [21].
Although the three methods use different segments of the database, they have a common closing date for survival analysis, the end of 1989. Using the data that were available at this date, we calculated three sets of survival statistics and drew a survival curve for each. Working with historical data permitted us to compare the results of the three methods available at the end of 1989 with the 10-year survival actually observed 10 years later for children diagnosed in 1985–89 (lower horizontal black rectangle in Figure 1). This comparative measure was calculated using the cohort method with the closing date ‘end of 1999’.
For comparison purposes we also report 5-year survival for selected groups of patients. These were obtained using the same methodology and same sets of data.
All analyses were carried out with a recently published SAS macro, which can be used for both traditional survival analysis and for period analysis [22].
|
results |
|---|
|
TopAbstractintroductionpatients and methodsresultsdiscussionAcknowledgementsReferences |
|---|
Overall, 35 191 childhood cancer cases were extracted from the ACCIS database and included in the analyses. The numbers and proportions of patients included in tumour-specific analyses are shown in Table 2. Lymphoid leukaemia was by far the most common diagnosis, accounting for more than one in four childhood cancers. The other 11 forms of cancer assessed in this analysis accounted for between 2.1 and 8.2% of all childhood cancers. The minimum number of cancers per diagnostic group was 597. Almost 80% of all childhood cancers were included in the cancer-specific analyses for 12 selected tumour types.
|
Among children diagnosed with any form of cancer in 1985–89, 65% were still alive after 5 years and 62% after 10 years of follow-up (Table 2, observed survival). However, using the cohort method, the survival estimate available in 1989 was only 47% at 5 years and 43% at 10 years of follow-up. An estimate of survival available at the end of 1989 from complete analysis (pertaining to all children diagnosed in 1975–89) was somewhat higher (55% at 5 years and 50% at 10 years), but still substantially lower than the survival actually experienced by the patients diagnosed in 1985–89. By contrast, 5-year survival estimated by period analysis at the end of 1989 was 63% and 10-year survival 58%, i.e. much closer to the survival estimates eventually observed for children diagnosed in 1985–89.
The results of the three methods are compared for the selected diagnostic groups of childhood cancer in Table 2 and Figure 2. Despite major variation of survival between different types of childhood cancer, the survival curves obtained by the period method for 1985–89 (dashed curves in Figure 2) were very close to the true survival curves of patients diagnosed in 1985–89, observed at the end of 1999 (upper black solid curve in Figure 2), for most diagnostic groups. Period analysis almost perfectly predicted survival curves of children with lymphoid leukaemia, Hodgkin lymphoma, primitive neuroectodermal tumours, neuroblastoma, retinoblastoma, osteosarcoma and rhabdomyosarcoma. For the other diagnostic groups, the period estimates were somewhat lower, but much less so than the traditional cohort and complete estimates. The differences between the estimates obtained by the traditional methods and by period analysis were largest for those cancers with the strongest improvement in prognosis over time, i.e. lymphoid leukaemia, acute non-lymphocytic leukaemia, non-Hodgkin lymphoma and osteosarcoma.
|
Standard errors of 10-year survival estimates were similar for cohort and period analysis (ranging from 1.0 to 3.2 % and from 1.0 to 3.6%, respectively, for the different types of cancer) and somewhat lower for complete analysis (range: 0.7 to 2.3%) (Table 2).
|
discussion |
|---|
|
TopAbstractintroductionpatients and methodsresultsdiscussionAcknowledgementsReferences |
|---|
In this article we demonstrate how well the period survival method estimates predicted survival for recently diagnosed patients, using childhood cancer cases from the ACCIS database [4]. In particular, using historical data we showed that the period survival method provides much more up-to-date estimates of eventually observed long-term survival of children with cancer than the traditional cohort and complete life-table analysis. This pattern was consistent for all common forms of childhood cancer included in this analysis. These results are in agreement with and extend the previous findings obtained for cancers of adults and for selected forms of childhood cancer in relatively small samples of children [9, 10, 13, 23].
We have deliberately chosen to report 10-year survival to emphasize the increased mortality affecting 5-year survivors, which may sometimes be forgotten when survival is reported for 5-year follow-up periods.
With very few exceptions [19–25], childhood cancer survival figures available in the literature have been derived using cohort analysis, complete analysis or a mixture of both [1–6, 26]. Although these survival statistics are useful for evaluation of survival of defined cohorts of patients, they are less so when interest lies in obtaining information on survival expectations of recently diagnosed cancer patients. The period survival method is of particular value in this context, and this increases with the length of the follow-up, and with the speed of change in prognosis. In the evaluations presented in this paper, the advantage of period analysis was largest for lymphoid leukaemia, acute non-lymphocytic leukaemia, non-Hodgkin lymphoma and osteosarcoma, the cancers for which prognosis has improved most dramatically in recent decades.
For some diagnostic groups (e.g. acute non-lymphocytic leukaemia) the survival curves obtained by the period survival method tended to be still somewhat pessimistic (albeit much less so than the traditional cohort and complete estimates). This phenomenon is indicative of ongoing further strong improvement of survival of patients diagnosed in 1985–89 during later years of follow-up, i.e. after 1989, which was not reflected in the conditional survival probabilities available at the end of 1989.
In theory, period estimates may also become (transiently) too optimistic, if changes in early detection only increase lead time. A pertinent impact of early detection might be recognized for neuroblastoma, where screening practices or increased awareness have increased early diagnosis or led to overdiagnosis, which artificially prolonged survival time without reducing mortality [27]. However, even for neuroblastoma, the period estimate of 10-year survival for the 1985–89 period was almost identical with the 10-year survival later observed for patients diagnosed in 1985–89. In general, advances in childhood cancer therapy appear to have reduced both early and late cancer deaths [28–30]. This is in agreement with our finding that some underestimation rather than overestimation of long-term survival is the issue to be concerned about in practice, even with period analysis, as long as survival continues to improve. However, such underestimation manifests much less with period analysis than with traditional cohort-based analyses.
When survival remains stable over time, cohort, complete and period estimates are generally very similar. In our analyses, such patterns were seen, for example, for Hodgkin lymphoma and retinoblastoma. However, period estimates provide some additional information even in that situation, by indicating the absence of major recent progress in survival, which could have been missed by the other methods of survival analysis due to their delayed responsiveness. In the case of a deterioration of survival over time, such an alarming development would also be detected more timely by period analysis.
A potential disadvantage of period estimates compared with complete estimates is their somewhat higher standard error. However, in our analyses the differences between the point estimates were typically much greater than the standard errors of either estimate. An increase in precision is of much less consequence than the greater delay by which progress in prognosis is captured. The issue of precision could become more important, however, in analyses of smaller datasets, or when improvement in survival has levelled off.
Fortunately, improvement in survival seems to be ongoing, and for most childhood cancers the pace of change is dramatic; long-term period survival estimates at a given time are much closer to long-term survival experienced by concurrently diagnosed patients than traditional survival estimates in this very situation. We therefore recommend that period analysis be more widely used for timely monitoring of changes in prognosis of the cancers of childhood at the population level.
|
Acknowledgements |
|---|
|
TopAbstractintroductionpatients and methodsresultsdiscussionAcknowledgementsReferences |
|---|
This project was funded by the European Commission from the Europe Against Cancer programme (contract SPC.2002303), jointly with the International Agency for Research on Cancer. E.S.-F. was partly supported by the Cancéropôle Lyon Auvergne Rhône Alpes and the Federal Ministry of Health of the Federal German Government. The authors are grateful to the following registries participating in ACCIS for their contribution to this study. Czech Republic: Czech Cancer Registry; Estonia: Estonian Cancer Registry; Finland: Finnish Cancer Registry; France: Bas-Rhin Cancer Registry; Hungary: Hungarian Paediatric Cancer Registry; Iceland: Icelandic Cancer Registry; Italy: Childhood Cancer Registry of Piedmont and the Lombardy Cancer Registry; Norway: Cancer Registry of Norway; Slovenia: Cancer Registry of Slovenia; Spain: Navarra Cancer Registry; Sweden: Swedish Cancer Registry; Switzerland: Geneva Cancer Registry; United Kingdom: Childhood Cancer Registry for England and Wales and the Scottish Cancer Registry.
Received for publication November 17, 2006. Revision received April 3, 2007. Accepted for publication April 11, 2007.
|
References |
|---|
|
TopAbstractintroductionpatients and methodsresultsdiscussionAcknowledgementsReferences |
|---|
1. Ries LAG, Smith MA, Gurney JG, et al, eds. Cancer incidence and survival among children and adolescents: United States. SEER Program 1975–1995. NIH Pub No. 99-4649 (1999) Bethesda, MD: National Cancer Institute.
2. Capocaccia R, Gatta G, Magnani C, et al. Childhood cancer survival in Europe 1978–1992: the EUROCARE study. Eur J Cancer (2001) 37:671–816.[CrossRef][Web of Science][Medline]
3. Gatta G, Capocaccia R, Coleman MP, et al. Childhood cancer survival in Europe and in the United States. Cancer (2002) 95:1767–1772.[CrossRef][Web of Science][Medline]
4. Steliarova-Foucher E, Stiller CA, Kaatsch P, et al. Geographical patterns and time trends of cancer incidence and survival among children and adolescents in Europe since the 1970s (the ACCIS project): an epidemiological study. Lancet (2004) 364:2097–2105.[CrossRef][Web of Science][Medline]
5. Gatta G, Capocaccia R, Stiller C, et al. Childhood cancer survival trends in Europe: A EUROCARE Working Group Study. J Clin Oncol (2005) 23:3742–3751.
6. Steliarova-Foucher E, Coebergh JW, Kaatsch P, et al. Cancer in children and adolescents in Europe. Eur J Cancer (2006) 42:1913–2183.[CrossRef][Web of Science]
7. Brenner H, Gefeller O. An alternative approach to monitoring cancer patient survival. Cancer (1996) 78:2004–2010.[CrossRef][Web of Science][Medline]
8. Brenner H, Gefeller O. Deriving more up-to-date estimates of long term patient survival. J Clin Epidemiol (1997) 50:211–216.[CrossRef][Web of Science][Medline]
9. Brenner H, Hakulinen T. Up-to-date survival curves of patients with cancer by period analysis. J Clin Oncol (2002) 20:826–832.
10. Brenner H, Söderman B, Hakulinen T. Use of period analysis for providing more up-to-date estimates of long-term survival rates: empirical evaluation among 370 000 cancer patients in Finland. Int J Epidemiol (2002) 31:456–462.
11. Brenner H, Hakulinen T. Advanced detection of time trends in long-term cancer patient survival: experience from 50 years of cancer registration in Finland. Am J Epidemiol (2002) 156:566–577.
12. Brenner H, Hakulinen T. Very long-term survival rates of patients with cancer. J Clin Oncol (2002) 20:4405–4409.
13. Brenner H. Up-to-date survival curves of children with cancer. Br J Cancer (2003) 88:1693–1697.[CrossRef][Web of Science][Medline]
14. Brenner H, Steliarova-Foucher E, Arndt V. Up-to-date monitoring of childhood cancer long-term survival in Europe: methodology and application to all forms of cancer combined. Ann Oncol (2007) 18. in press.
15. Brenner H, Coebergh JW, Parkin DM, et al. Up-to-date monitoring of childhood cancer long-term survival in Europe: leukaemias and lymphomas. Ann Oncol (2007) 18. in press.
16. Arndt V, Kaatsch P, Steliarova-Foucher E, et al. Up-to-date monitoring of childhood cancer long-term survival in Europe: central nervous system tumours. Ann Oncol (2007) 18. in press.
17. Arndt V, Lacour B, Steliarova-Foucher E, et al. Up-to-date monitoring of childhood cancer long-term survival in Europe: Tumours of the sympathetic nervous system, retinoblastoma, renal and bone tumours and soft sarcomas. Ann Oncol (2007) 18. in press.
18. Steliarova-Foucher E, Kaatsch P, Lacour B, et al. Quality, comparability and methods of analysis of data on childhood cancer in Europe(1978–1997): Report from the ACCIS project. Eur J Cancer (2006) 42:1915–1951.[CrossRef][Web of Science][Medline]
19. Kramárová E, Stiller CA, Ferlay J, et al. International classification of childhood cancer. IARC Technical International Agency for Research on Cancer. (1996) Lyon. Report No. 29.
20. Parkin DM, Hakulinen T. Analysis of survival. In: Cancer Registration. Principles and Methods. IARC Scientific Publications No. 95—Jensen OM, Parkin DM, Maclennan R, et al, eds. (1991) Lyon: International Agency for Research on Cancer. 159–176.
21. Brenner H, Gefeller O, Hakulinen T. Period analysis for up-to-date cancer survival data. Theory, empirical evaluation, computational realisation and applications. Eur J Cancer (2004) 40:326–335.[CrossRef][Web of Science][Medline]
22. Brenner H, Gefeller O, Hakulinen T. A computer program for period analysis of survival. Eur J Cancer (2002) 38:690–695.[CrossRef][Web of Science][Medline]
23. Burkhardt-Hammer T, Spix C, Brenner H, et al. Long-term survival of children with neuroblastoma prior to the neuroblastoma screening project in Germany. Med Pediatr Oncol (2002) 39:156–162.[CrossRef][Web of Science][Medline]
24. Brenner H, Kaatsch P, Burkhardt-Hammer T, et al. Long-term survival of children with leukemia achieved by the end of the second millenium. Cancer (2001) 92:1977–1983.[CrossRef][Web of Science][Medline]
25. German Childhood Cancer Registry. Annual Report 2002 (2002) Mainz: German Childhood Cancer Registry.
26. Stiller CA. Population-based survival rates for childhood cancer in Britain, 1980–1991. BMJ (1994) 309:1612–1616.
27. Spix C, Pastore G, Sankila R, et al. Neuroblastoma incidence and survival in European children (1978–1997): Report from the Automated Childhood Cancer Information System project. Eur J Cancer (2006) 42:2081–2091.[CrossRef][Web of Science][Medline]
28. Robertson CM, Hawkins MM, Kingston JE. Late deaths and survival after childhood cancer: implications for cure. BMJ (1994) 309:162–166.
29. Hudson MM, Jones D, Boyett J, et al. Late mortality of long-term survivors of childhood cancer. J Clin Oncol (1997) 15:2205–2213.
30. Möller TR, Garwicz S, Barlow L, et al. Decreasing late mortality among five-year survivors of cancer in childhood and adolescence: a population-based study in the Nordic countries. J Clin Oncol (2001) 19:3173–3181.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  H. Brenner, E. Steliarova-Foucher, and V. Arndt Up-to-date monitoring of childhood cancer long-term survival in Europe: methodology and application to all forms of cancer combined Ann. Onc., September 1, 2007; 18(9): 1561 - 1568. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
H Brenner, J. Coebergh, D. Parkin, I Izarzugaza, J Clavel, V Arndt, and E Steliarova-Foucher Up-to-date monitoring of childhood cancer long-term survival in Europe: leukaemias and lymphomas Ann. Onc., September 1, 2007; 18(9): 1569 - 1577. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||