The late effects study: design and subject representativeness of a Canadian, multi-centre study of late effects of childhood cancer

Headnote

Abstract

The Late Effects Study of the Canadian Childhood Cancer Surveillance and Control Program was designed to assess psychosocial and physical health outcomes among survivors of childhood cancer compared to general population controls. The objectives of this paper are to describe the design and methodology of the multi-centre, retrospective cohort study, present clinical characteristics of the survivor population, and evaluate the representativeness of study controls. Response rates were 63% for survivors (n = 2,152) and 49% for controls (n = 2,432). Survivors with germ cell tumours and carcinomas were slightly under-represented among participants as were those who received more intense or multiple series of therapy. Study controls were similar to Census individuals based on marital and work status but did have a slightly higher level of education and income. Otherwise, no large or systematic differences were found. Thus, these long-term survivors and population controls can be validly studied to evaluate whether and to what extent survivors experience an excess of psychosocial or physical health problems compared to similarly aged Canadians who have never had cancer.

Key words: adolescence; cancer; child; population controls; quality of life; study design; survivors

Introduction

Childhood cancer is the leading cause of death from disease in Canadians less than 20 years of age. Each year, approximately 1,300 children and adolescents are diagnosed with cancer in Canada.1 Despite this, in Canada and other developed countries, survival of childhood cancer patients has risen remarkably in the past 30 years due to the use of aggressive, multi-modal therapies. Today, over 75% of children diagnosed with cancer will survive five or more years post-diagnosis.2·3 Currently, it is estimated that one in every 900 adults in the United States is a survivor of childhood cancer.4 However, the long-term implications of both the disease and its treatment are only beginning to be understood.

In 1992, Health Canada announced the Canadian Childhood Cancer Surveillance and Control Program (CCCSCP), a collaboration involving several stakeholders concerned with childhood cancer including Health Canada, pediatric oncology centres, provincial cancer registries, universities, and voluntary agencies. The program was designed to complement provincial cancer registry and clinical trials data by collecting information on the complete childhood cancer continuum. To achieve its goal, the program included three main study components: Etiology, Treatment and Outcome Surveillance, and Late Effects. The overall aim of the Late Effects Study was to describe the psychosocial and physical health outcomes among survivors of childhood cancer compared to general population controls of the same age and sex with no history of cancer. This is the first in a series of manuscripts to be produced from the Late Effects Study. Subsequent papers will include those reporting on the outcomes of health-related quality of life, academic and social achievements, healthrelated behaviours, health service use, and reproduction.

The objectives of this manuscript are to describe the Late Effects Study design and methodology, to present the clinical characteristics of the survivor population, and to evaluate the representativeness of the study controls.

Methods

Design

The CCCSCP Late Effects study was a multi-centre, retrospective cohort study of childhood cancer survivors and population controls. Thirteen pediatric oncology centres3 with survivors from every province and territory in Canada participated. Data were collected between January 1997 and February 2000. The research ethics boards of the participating pediatric oncology centres, provincial cancer registries, and provincial health insurance agencies approved the study protocol.

Subjects

Survivors of childhood cancer were eligible to participate if they were diagnosed with a first malignancy before 20 years of ageb in 1981 to 1990, survived five years post-diagnosis, were alive at the time of study, and had sufficient language skills in either English or French to complete the questionnaire. Childhood cancer was defined according to the International Classification of Childhood Cancer fJCCC)5 which is based on the ICD-O-2 morphology and topology codes. In eastern Canada (Nova Scotia, Newfoundland, Quebec, and Ontario) only eligible survivors diagnosed and/or treated in the participating pediatric oncology centres were identified for the study. In Western Canada (Manitoba, Saskatchewan, Alberta, and British Columbia), survivors who met the eligibility criteria but were not seen at one of the pediatric oncology centres were identified through the provincial cancer registry.

Study controls were selected to resemble the survivors except for having no cancer history. Thus, population-based controls frequency matched to survivors by age and sex within each province were recruited. Control subjects were eligible to participate if they reported no previous cancer diagnosis and had sufficient language skills in either English or French to complete the questionnaire. A 1:1 survivor to control ratio was planned. In three of the eight provinces (British Columbia, Manitoba, and Quebec), eligible controls were identified through the provincial health insurance agency, which maintains a comprehensive list of residents to provide universal health care coverage. In the other five provinces (Alberta, Saskatchewan, Ontario, Newfoundland and Nova Scotia) random digit dialing (RDD) was used to identify eligible control subjects due to restrictions on access to provincial health insurance data.

Two private firms that maintain banks of operating 10-digit telephone numbers conducted the random digit dialling. The first three digits identify the province of residence. Individual banks of operating numbers were randomly selected and the final two digits were removed and regenerated using a computer program that randomly chooses a number from 00 to 99. This method was repeated until the full sample was drawn. RDD thus allows the identification of potential control subjects with both listed and unlisted numbers. Each telephone number was called at least five different times during the week before being removed from the sample as an unscreened number. Eligible controls who agreed to participate during the initial call were placed in the correct age/sex strata. Once the stratum quotas were filled, the information was forwarded to the pediatric oncology centre responsible for recruitment and follow-up.

Data collection

Essentially similar procedures were used at each participating centre. The required numbers of survivors and controls from the pediatric oncology centres, the provincial cancer registries, and the provincial health insurance agencies were randomly selected from the sampling frames provided. All controls identified by RDD were included for recruitment. A clinical research assistant from each centre was assigned to recruit the selected subjects. Telephone and address information for study subjects and/or their parents were provided with the lists of eligible subjects. Information on current family physician was also obtained, if available, as an additional resource to locate subjects. Contact information was often out of date, in which case electronic telephone directories were used to trace subjects and/or their parent(s).

Initial contact was generally made by mail using a personalized letter explaining the study, its objectives, and the criteria for participation. However, due to varying requirements of the research ethics committees, initial contact was made by telephone in some centres using prepared scripts. After consent was obtained by mail or telephone, the questionnaires were sent by mail and up to two follow-up calls were made to remind subjects to return their questionnaires. Reminder calls were made three weeks after initial mailing and two weeks after the first follow-up call. Up to 12 attempts with a minimum of five attempts were made at different times during the day and days of the week to reach respondents who had not returned their questionnaires. Centres that contacted their subjects by telephone first, rather than by letter, had the highest response rates. Response rates were also slightly higher in centres that mailed a reminder letter signed by the pediatric oncologist after the follow-up telephone calls.

Information on psychosocial and physical health outcomes was gathered using two age-appropriate questionnaires, one for respondents who were under 16 years of age at study and one for those who were 16 years and older. Where possible the questionnaires included previously validated measures with known psychometric characteristics (Table 1). Subjects 16 years and older at the time of study completed a self-report questionnaire while a parent or guardian responded for subjects less than 16 years old. Both questionnaires were estimated to take between 45 minutes and one hour to complete, based on the results of the feasibility study undertaken prior to the main study.

For survivors, clinical information was extracted from the treatment medical record at the oncology centre. Information obtained included cancer type, site, and morphology; stage at diagnosis; treatments undergone including chemotherapy, surgery, and radiotherapy; and outcomes, such as relapse and death, which occurred at the treating institution.

Sample size

A target sample size of 2,000 completed questionnaires from both survivors and controls was set to provide 80 % power to detect a difference in mean scores of 5% or more for psychosocial outcomes or a relative risk of 1.5 for common and 2.5 for rare physical outcomes, when stratified by major diagnostic category and age at study (< 16 years, > 16 years). The number of subjects targeted by each centre was proportional to the size of its patient population. An over-selection of subjects ensured that the targeted sample size was achieved.

Analysis

Response rates were calculated by dividing the number of subjects who returned completed questionnaires by the total number of eligible subjects, including those who were lost to follow-up or refused to participate. Subjects considered lost to follow-up included passive refusals (i.e., subjects who agreed to participate but never returned their questionnaires) as well as subjects who were never located. Some of the survivors who were never located may have been deceased.

For control subjects identified by RDD, the denominator of the response rate included a proportion of unscreened telephone numbers. Unscreened numbers consisted of unanswered numbers as well as answered numbers where the interviewer was not able to complete the screening process. The number added to the denominator of eligible RDD subjects was estimated by multiplying the number of unscreened telephone numbers by an estimated probability that each would have resulted in the identification of an eligible subject. The estimated probability was calculated by dividing the ? of successfully screened telephone numbers resulting in the identification of eligible subjects by the ? of successfully screened telephone numbers.

Among survivors, clinical characteristics for participants were compared to those who refused to participate and those lost to followup. Characteristics assessed included major diagnostic group, age at diagnosis, year of diagnosis, types of treatments received, and number of treatment series completed during therapy. A treatment series included chemotherapy, radiation therapy and/or surgery and usually followed a predetermined protocol. More than one treatment series occurred when the initial planned treatment was extended due to incomplete or lack of response or after a relapse.

Study controls were compared to Canadian Census data to assess the extent to which controls resembled the Canadian population.6 Parental socio-demographic information was used for the comparisons of controls < 16 years of age at study. Key variables common to our study and the census included: achieved education, marital status, work status, and personal income. Census data from 1996 were used as this was the census year closest to the year the majority of subjects were recruited. Comparison with Census data was done by province, weighted by sample size, sex, and age. Because the Census data are reported in five-year age groups the study controls compared were limited those who were 20 to 44 years of age at study. This group represented 74% of all study controls.

Chi-square tests were used to compare the socio-demographics of study survivors and controls, and to compare the clinical characteristics of participant survivors to survivors who refused to participate and to survivors who were lost to follow-up. Cells that contributed to the significant difference between columns were highlighted. Due to the large number of comparisons made only alpha values below 0.01 were reported.

Results

Three thousand four hundred and seventy three eligible survivors and 5,063 eligible controls were selected for study. Questionnaires were completed and returned for a total of 2,216 survivors and 2,507 controls (Figures 1 & 2). After removing questionnaires completed by subjects who were later determined to be ineligible (e.g., out of age range, diagnosis not included in the ICCC) there were 2,152 survivors and 2,432 controls available for outcome analysis. Response rates were 63.1% (2,152/3,409) for survivors and 48.8% (2,432/4,988) for controls.

Response rates for survivors by ICCC major diagnostic category were over 60% with the exception of survivors of bone tumours, germ cell tumours, and carcinomas that are more commonly found in older children. The response rates for these cancers were 57.4%, 50.2% and 43.9%, respectively. Response rates were highest for survivors of neuroblastoma (82.1%) and hepatic carcinomas (72.7%), while survivors of the remaining diagnostic groups had response rates between 60% and 70%.

Socio-demographic characteristics of study survivors and controls are shown in Table 2. Approximately 50% of survivors and controls were male and the average age at study was 19 years. The large majority of subjects reported Caucasian ethnicity, followed by Asian (3.5%) and multiple ethnicities (3.0%). Similar and substantial proportions from each group were residing in Central or Western Canada.

The distribution of clinical characteristics for participant and non-participant (refusals and lost to follow-up) survivors is shown in Table 3. Over half of the survivors were diagnosed with leukemia, lymphoma or central nervous system tumours, regardless of their participation status. Compared to the participant survivors, those lost to follow-up were significantly more likely to be germ cell tumour or carcinoma survivors and less likely to have been diagnosed with leukemia. Accordingly, the average age at diagnosis was significantly lower among participating survivors (seven years) compared to survivors who were lost to follow-up (nine years), and participants were more likely to have been diagnosed in the latter part of the decade (1987 to 1990). Survivors who participated were also more likely to have received all three forms of therapy (chemotherapy, radiation therapy, and surgery) compared to survivors who were lost to follow-up. Finally, survivors who were lost to follow-up were less likely to have had more than one treatment series compared to participants.

Sociodemographic characteristics of study controls were generally similar to those of the population as reported in the 1996 census (Table 4). Approximately 60% of the controls and census individuals were married or living in common-law relationships and nearly three quarters of subjects in both cohorts were employed either full or part-time. Some differences were noted in education and income levels: the control subjects reported a higher level of education and personal income compared to the census individuals.

Discussion

The national, multi-centre design employed in the Late Effects Study resulted in the identification of a cohort of children, adolescents, and young adults who survived childhood cancer and a population-based comparison group of similar age and sex with no cancer history. Although study participation was not as high as anticipated, we believe these two groups can be validly compared to evaluate whether and to what extent survivors experience an excess of psychosocial or physical health problems.

First, the long-term survivors who participated in the study had clinical characteristics that were similar to those of survivors who either refused to participate or were lost to follow-up. This suggests they were representative of all children and young adults diagnosed with cancer in Canada before the age of 20 in 1981 to 1990 and were alive during the study period. With the exception of cancers more commonly diagnosed in older children (i.e., germ cell tumours and carcinomas) response rates did not vary by cancer type. Specifically, the lower participation among survivors of germ cell tumours and carcinomas likely reflects the difficulty in recruiting young adults rather than the loss of eligible subjects with a particular treatment or outcome. This is further reflected in the older average age at diagnosis and earlier period of diagnosis among survivors who were lost to follow-up. As well, carcinoma and germ cell tumour survivors were less likely to receive multi-modal care, possibly due to their earlier period of diagnosis as well as their cancer type, with more than two-thirds receiving surgery only or surgery with chemotherapy. Finally, over 90% of survivors with carcinoma and germ cell tumours reported no relapses during therapy.

Second, by design, the population controls recruited for study have similar age and sex distributions to the survivors. They are also similar in terms of ethnic origin and area of residence in Canada. With considerable information on potential confounding factors available for both survivors and controls, any differences that might affect the outcomes of interest can be statistically controlled during analyses. Thus, we are confident that any differences observed when comparing the physical and psychosocial health of survivors and controls can be attributed to having survived childhood cancer.

Third, the population controls successfully recruited for our study closely resembled individuals of the same age and sex from the general Canadian population. Proportions of study controls and those identified in the census who were married or living common law or employed full- or part-time during the study were virtually identical. However, our controls had higher educational and income levels compared to census individuals. Observed differences in personal income may be explained in part because census income was adjusted to 1995 dollars whereas our participants were recruited from 1997 to 2001.

It should be noted that we used stringent standards to define the number of eligible individuals to be included in the denominator of the response rate. In our calculation of response rates we used the total number of eligible subjects as the denominator, not the number of subjects who agreed to participate. Furthermore, some of the survivors considered lost to follow-up may have, in fact, died in the time between surviving five years and being recruited for study, artificially inflating the number of eligible survivors. As well, we included an estimated proportion of unscreened calls recruited by RDD in the total number of eligible control subjects, which may have also inflated the denominator and lowered the estimated response rate. Still, our response rates of 63 % for survivors and 49% for controls are similar to other large case-control studies of cancer.7-9 For example, using the same method of calculating participation, the U.S. Childhood Cancer Survivor Study achieved response rates of 69% for survivors (14,054 out of 20,275) and 61 % for sibling controls (3,585 out of 5,857) .10 Modest response rates are worrisome if it is assumed that those who are harder to reach differ in substantial ways from their easier-to-recruit counterparts. If they did, participants could differ from the target population.11"18 In the data presented here we have demonstrated the absence of any large or systematic differences among the participating survivors compared to non-participants and study controls compared to census individuals. Moreover, many authors suggest that slight differences between participants and non-participants should not greatly affect estimates of risk.11-15·16·19·20

The most formidable challenge in conducting this multi-centre, retrospective study was locating and recruiting study subjects. This challenge was confounded by having to solicit the participation of a cohort of young adults, a group known to be highly mobile, more difficult to reach, and least likely to participate in such studies.21 Although we used a number of techniques known to help recruitment (personalized letters, follow-up contacts, stamped return envelopes, contacting participants prior to sending out the questionnaire) we can also speculate on additional ways participation could be improved in future studies.22·23 Contacting subjects by telephone first, rather than by letter, seems to help. More follow-up attempts with subjects who have been contacted but have not returned their questionnaires would likely improve participation. A final, practical suggestion for other researchers in such circumstances is to consider the use of small symbolic monetary incentives to improve participation among those contacted. It has been demonstrated that such incentives can be effective in increasing participation in population-based cohort studies of long-term cancer survivors.24

In summary, the results of this first report of the CCCSCP Late Effects Study indicate the long-term childhood cancer survivors and population controls can be validly compared to evaluate whether and to what extent survivors experience an excess of psychosocial or physical health problems. Papers currently in preparation by our group comparing both adult and child survivors to their respective controls can thus be interpreted as representing the nature and extent of any late effects of cancer diagnosed during childhood in Canada.

Acknowledgements

The authors gratefully acknowledge the contributions of study participants, participating pediatric oncology centres, and members of the CCCSCP Management Committee. The CCCSCP is fully funded by Health Canada. E. Marniseli was the recipient of a National Health Research and Development Program (NHRDP), National Health Research Scholar Award at the time this work was started, and currently holds an Investigator Award from the Canadian Institutes of Health Research (CIHR).

Footnote

a Participating centres included: British Columbia Children's Hospital, Alberta Children's Hospital, Cross Cancer Institute, Allan Blair Cancer Centre, Saskatoon Cancer Centre, Manitoba Cancer Treatment and Research Foundation, Chedoke-McMaster Hospital, Hospital for Sick Children, Hôpital Ste- Justine, Montreal Children's Hospital, Centre Hospitalier de l'Université Laval, Isaac Walton Killam Children's Hospital, Janeway Child Health Centre.

b In British Columbia, Alberta, Saskatchewan, and Manitoba subjects 0 to 19 years old at diagnosis were included. In Ontario, Quebec, Nova Scotia, and Newfoundland subjects 0 to 17 years old at diagnosis were included.

References

References

1 . National Cancer Institute of Canada. Canadian Cancer Statistics 2002. Toronto, Canada.

2. Ries LAG, Eisner MP, Kosaiy CL, Hankey BF, Miller BA, Clegg L et aL SEER Cancer Statistics Review, 1973-1999. National Cancer Institute. Bethesda, MD. 2002.

3. Eiser C. Practitioner review: Long-term consequences of childhood cancer. J Child Psychol Psychiatry 1998;39(5):621-33.

4. Oeffinger KC, Eshelman DA, Tomlinson GE, Buchanan GR. Programs for adult survivors of childhood cancer. J Oin Oncol 1998;16(8): 2864-7.

5. Kramarova E, Stiller CA, Ferlay J, Parkin DM, Draper GJ, Michaelis J et al. International Cíasáfication of Childhood Cancer 1996. Lyon, France, International Agency for Research on Cancer. 1996.

6. Statistics Canada. 1996 Census data: The nation series - complete edition. Ottawa, Canada. 1998.

7. Sharp L, Chilvers CE, Cheng KK, McKinney PA, Logan RF, Cook-Mozaffari ? et al. Risk factors for squamous cell carcinoma of the oesophagus in women: A case-control study. Br J Cancer 2001;85(1 1):1667-70.

8. Moorman PG, Newman B, Millikan RC, Tse CK, Sandler DP. Participation rates in a casecontrol study: The impact of age, race, and race of interviewer. Ann Epidemiol 1999;9(3): 188-95.

9. Radman I, Basic N, Labar B, Kovacevic J, Aurer I, Bogdanic V et al. Long-term results of conventional-dose salvage chemotherapy in patients with refractory and relapsed Hodgkin's disease (Croatian experience). Ann Oncol 2002;13(10):1650-5.

10. Robison LL, Mertens AC, Boice JD, Breslow NE, Donaldson SS, Green DM et al. Study design and cohort characteristics of the Childhood Cancer Survivor Study: A multiinstitutional collaborative project. Med Pediatr Oncol 2002;38(4):229-39.

11. O'Toole BI, Marshall RP, Grayson DA, Schureck RJ, Dobson M, French M et al. The Australian Vietnam Veterans Health Study: 1. Study design and response bias, hit J Epidemiol 1996;5(2):307-18.

12. Brussaard JH, Brants HA, Bouman M, Lowik MR. The study population: General characteristics and potential confounding factors. Eur J Oin Nutr 1997;51 Suppl 3:S19-S24.

13. Etter JF, Perneger TV. Analysis of nonreponse bias in a mailed health survey. J Clin Epidemiol 1997;50(10):1 123-8.

14. Larroque B, Kaminski M, Bouvier-Colle ??, Hollebecque V. Participation in a mail survey: Role of repeated mailings and characteristics of nonrespondents among recent mothers. Paediatr Perinat Epidemiol 1999; 13(2):218-33.

15. Klesges RC, Williamson JE, Somes GW, Talcott GW, Lando HA, Haddock CK. A population comparison of participants and nonparticipants in a health survey. Am J Public Health 1999;89(8):1228-31.

16. Madigan MP, Troisi R, Potischman N, Brogan D, Gammon MD, Malone KE et al. Characteristics of respondents and non-respondents from a case-control study of breast cancer in younger women Int J Epidemiol 2000;29(5):793-8.

17. Hatch EE, Kleinerman RA, Linet MS, Tarone RE, Kaune WT, Auvinen A, et al. Do confounding or selection factors of residential wiring codes and magnetic fields distort findings of electromagnetic fields studies? Epidemiology 2000;ll(2):189-98.

18. Rupp I, Triemstra M, Boshuizen HC, Jacobi CE, Dinant HJ, van den Bos GA. Selection bias due to non-response in a health survey among patients with rheumatoid arthritis. Eur J Public Health 2002;12(2):131-5.

19. Kreiger N, Nishri ED. The effect of nonresponse on estimation of relative risk in a case-control study. Ann Epidemiol 1997; 7(3):194-9.

20. Brogger J, Bakke P, Eide GE, Gulsvik A. Contribution of follow-up of nonresponders to prevalence and risk estimates: A Norwegian respiratory health survey. Am J Epidemiol 2003;157(6):558-66.

21. McCormick LK, Crawford M, Anderson RH, Gittelsohn J, Kingsley B, Upson D. Recmiting adolescents into qualitative tobacco research studies: experiences and lessons learned. J Sell Health 1999;69(3):95-9.

22. Edwards P, Roberts I, Clarke M, DiGuiseppi C, Pratap S, Wentz R et al. Increasing response rates to postal questionnaires: Systematic review. Br Med J 2002;324(7347): 1183.

23. Dillman D. Mail and Internet Surveys. 2nd ed. New York: John Wiley and Sons, Ine, 2000.

24. Maunsell E, Labbe J, Brisson C, Brisson J, Masse B. Monetary incentives and participation in a population-based study of 3-year breast cancer survivors and controls: randomized trial. Ref type: Unpublished work. 2004.