Health-Related Quality of Life in European Childhood Cancer Survivors: Protocol for a Study Within PanCareLIFE

Introduction

At present, more than 80% of children diagnosed as having cancer in Europe survive [1]. Almost half a million childhood cancer survivors are estimated to live in Europe in 2020 [2]. However, cancer itself, or its treatment, causes chronic health conditions, including a broad spectrum of somatic [3] and mental sequelae [4]. In the St. Jude Lifetime Cohort Study, nearly all childhood cancer survivors had at least one chronic health condition by the age of 50 years and twice the burden of disease than the general population [5]. Chronic health conditions such as heart failure, second neoplasms, or pulmonary dysfunction can be life threatening; other health conditions such as fertility and hearing impairment can affect survivors’ life planning and daily life, which may reduce their health-related quality of life (HRQoL) [6-8].

HRQoL is a multidimensional concept that includes elements of physical, functional, social, and psychological health as well as perceived health status and well-being [9]. Many studies assessing HRQoL in childhood cancer survivors used different questionnaires or different reference groups or varied in characteristics of the study population (eg, treatment era, age of survivors, cancer diagnostic groups), making it difficult to compare results between studies [10]. Results from large childhood cancer survivor studies in the United Kingdom (British Childhood Cancer Survivor Study, BCCSS) [11], the United States (Childhood Cancer Survivor Study) [12], and Switzerland (Switzerland Childhood Cancer Survivor Study, SCCSS) [6] suggest that, on average, childhood cancer survivors have similar HRQoL compared with the general population. However, there were significant differences in HRQoL between subgroups of survivors. Women, survivors with low educational background, survivors of brain tumors, and survivors who had undergone radiotherapy had the lowest HRQoL [6,11,12]. It is still unclear which other factors influence HRQoL and whether HRQoL in childhood cancer survivors differs among European countries.

Within the PanCareLIFE project, funded by the European 7th Framework Program (FP7), we aim to study HRQoL in a large database of childhood cancer survivors from 5 European countries using a homogeneous approach to assess and analyze HRQoL. In particular, we aimed to compare HRQoL in European childhood cancer survivors with normative data and between European countries to determine predictors of HRQoL and describe the effect of hearing and fertility impairment on HRQoL [13]. This study provides an overview of the design, data origin, and data collection strategies and summarizes the characteristics of survivors who participated in this study.

Methods

The PanCareLIFE Research Framework

The European FP7 project PanCareLIFE (grant agreement no. 602030) started in 2013 [14]. Institutions from 10 countries provided data on more than 15,000 childhood, adolescent, or young adult cancer survivors. Within the PanCareLIFE framework, this study focused on long-term HRQoL in childhood cancer survivors [13]. It was based in the University Hospitals of Münster (2013-2015) and Bonn (2016-2018). The Institute of Social and Preventive Medicine at the University of Bern provided methodological support and conducted the analyses. Each institution obtained ethical approval according to their local and/or national authority regulations before collecting the data. For all survivors, either written informed consent was obtained or the ethics committee agreed that an individual’s consent was not required for this questionnaire study.

Origin of Data and Inclusion Criteria

The PanCareLIFE HRQoL study population was composed of 6 national or regional cohorts, which had slightly different inclusion criteria. Table 1 provides an overview of the countries and institutions that provided data for the PanCareLIFE HRQoL study.

We combined data from Brno and Prague to 1 Czech cohort. The identification of eligible survivors depended on the respective cohort. Switzerland and Germany prospectively identified the survivors from the national childhood cancer registry, the Czech Republic and the Netherlands retrospectively identified the survivors and France implemented both ways of identification (Table 1) [15,17-19]. The identification of survivors was population-based in Switzerland, the Rhone-Alpes region (France), and Germany. In the Netherlands and the Czech Republic, it was hospital-based, with an estimated coverage of 95% of the national childhood cancer population. Switzerland, Germany, and the Netherlands enrolled their cohorts before and France and the Czech Republic after the launch of PanCareLIFE. Details on the period of data collection and sample characteristics of each cohort are listed in Table 2.

Survivors were eligible for the study if they were aged ≤18 years at the time of diagnosis, had been residents of the respective country at the time of diagnosis, had a cancer diagnosis according to the International Classification of Childhood Cancer, 3rd edition (ICCC-3) [21] or had been diagnosed as having Langerhans cell histiocytosis, had survived ≥5 years after cancer diagnosis, were aged ≥18 years at the time of the questionnaire survey, were not undergoing treatment for cancer at the time of the study, and did not refuse to registration in the national or local childhood cancer cohort. The last criterion did not apply to Czech survivors because patients were identified in a clinical routine. We could not calculate the exact age at the time of survey and the survival time, as we did not receive information on the day of birth, day of diagnosis, and day of survey from the institutions because of PanCareLIFE data protection rules. Therefore, we allowed for a tolerance of 2 months and included survivors who were 2 months older than 18 years at the time of cancer diagnosis, those who survived 2 months less than 5 years, and those who were 2 months younger than 18 years at the time of the survey.

In addition to the general eligibility criteria, there were country-specific reasons for the exclusion of eligible survivors and not all reasons applied to all cohorts. The Czech Republic, France, and the Netherlands excluded survivors with severe mental sequelae. In Switzerland and the Czech Republic, survivors were excluded from invitation because of physicians’ decisions mainly because of palliative care or relapsed disease. France, the Czech Republic, and Switzerland did not exclude survivors because of competing surveys. In the Netherlands and Germany, 2.21% (552/24,993) of survivors could not be included because they were already enrolled in competing surveys (Table 3). The French cohort did not include leukemia survivors.

Variables

Exposure variables and confounders covered the following topics: demographic characteristics, socioeconomic measures, lifestyle, cancer diagnosis, cancer treatment, hearing impairment, and female fertility impairment.

The outcome variables were the 36 HRQoL questions from the Short Form 36 (SF-36, version 1 or 2). The Netherlands, Germany, and the Czech Republic used version 1 of the SF-36, and Switzerland and France used version 2 [22,23]. Both versions differ only slightly in a few items, whereas the psychometric properties are comparable [24]. With the exception of variables concerning demographic characteristics, cancer diagnosis, and cancer treatment, all variables were self-reported.

Self-reported confounder variables included living with a partner (yes or no), education, occupational status, migration background, alcohol consumption, smoking, and body mass index. Another substudy of the PanCareLIFE research project investigated female fertility impairment and coded the data using 8 different criteria into a binary variable (fertility impairment: yes or no). The detailed procedure is outlined by van den Berg et al [25]. Fertility impairment data on male survivors were unfortunately not available in most of the cohorts; therefore, we will analyze female survivors only.

Hearing impairment data (yes or no) were collected together with information on HRQoL via questionnaires. Questions on hearing impairment differed slightly between data providers. Data on hearing impairment were not available for the Dutch cohort. All relevant variables (outcomes, exposures, and confounders) are listed in Multimedia Appendix 1.

The central PanCareLIFE data center defined a baseline variable list, including sex, cancer, and treatment-related variables and a minimal set of information provided for nonparticipants. The data center collected and harmonized the variables from each institution. We also recoded a few variables used specifically for the HRQoL study (eg, on hearing impairment) in cooperation with the institutions. All partners transferred the data according to the PanCareLIFE data protection standards.

The sources of cancer-related and treatment information, sources of self-reported data, and the methods of written questionnaire assessments differed slightly among the 5 cohorts (Table 4).

Data on HRQoL and HRQoL-specific exposure variables and self-reported confounders were assessed using the same questionnaire. An exception were the Dutch survivors, who answered the questionnaire on HRQoL on average 4 years later than the questions on sociodemographic characteristics and lifestyle for the purpose of their national Dutch Childhood Oncology Group Survivor (DCOG LATER) study.

SF-36 Data From the General Population

Country-specific reference data from the general population did not exist for Switzerland, France, and the Czech Republic at the time of the study. We used the German SF-36 version 1 normative data from 1998 for norm-based scoring of data from all participating countries [26]. This norm-based scoring allowed us to compare the HRQoL between countries [24]. We also scored HRQoL raw data with more recent SF-36 version 2 normative data from Germany (2008-2011) [24] and from Switzerland (2015-2016) [27].

Statistical Analyses

We compared survivors who participated in the survey with those who did not participate using chi-square tests. We used the software package Stata (version 14, StataCorp) for all analyses.

For all future analyses, we documented the planned statistical approaches for all HRQoL-related research questions of PanCareLIFE in a statistical analysis plan before we received the data from the central data center. In brief, SF-36 consists of 36 items (questions), which use a 2- to 6-point Likert scale, depending on the item. Between 2 and 10 items can be summarized into 1 of the 8 scales (physical functioning, role physical, bodily pain, general health, vitality, social functioning, role emotional, and mental health) and further into 2 summary scores (Physical Component Summary and Mental Component Summary). Because we will perform comparisons with both version 1 and version 2 normative data, we will transform all scales into the respective version with an algorithm provided by the developer of SF-36 [23,24,26]. We will convert raw scores into T-scores (mean 50, SD 10), and depending on the normative data used, we will stratify for subgroups, for example, age and sex. Higher scores indicate better HRQoL. Survivors with >50% missing data in any of the SF-36 scales will be excluded from the analyses. We will implement appropriate measures to deal with missing values, including imputation procedures. We will use regression analyses with HRQoL as an outcome variable to address all research questions to (1) compare HRQoL in European childhood cancer survivors with normative data and between European countries and to determine predictors of HRQoL, (2) describe the effect of hearing impairment on HRQoL, and (3) describe the effect of fertility impairment on HRQoL. We will include country, gender, and age in all multivariable models. We will also include the most important risk factors as defined by a selection criterion of P<.05 in the univariable regression. Multimedia Appendix 1 gives an overview of the variables that we plan to include in the regression analyses. If HRQoL scores have a skewed distribution, we will run logistic regressions (poor HRQoL vs normal or high HRQoL) and define poor HRQoL and high HRQoL, respectively, as scores below [11] and above the 10th percentile of the control population.

We will analyze subcohorts depending on the research question (Multimedia Appendix 2). For the analyses of differences in HRQoL between participating countries and predictors of HRQoL and for the analyses of the effect of hearing on HRQoL, we will use 2 main subsets for the analysis. In the first, we will include all survivors aged ≥25 years at the time of the study from all participating countries. The second subset will include survivors aged <25 years from all countries except Germany because Germany did not collect data in this age group. For the analyses on the effect of fertility impairment on HRQoL, we will investigate female survivors aged ≥25 years only. We will describe the detailed procedure in the respective publication of each research question.

Results

We identified 24,993 eligible 5-year childhood cancer survivors (Table 3). Of those, 5725 had died before the questionnaire survey, had severe mental problems, were living abroad at the time of the study, had no available contact data (lost to follow-up), were enrolled in competing surveys, their physicians decided not to invite them, or for other reasons. Of the 19,268 survivors who received an invitation to the HRQoL survey, 9192 did not respond or refused to participate and 206 had incomplete SF-36 information, preventing full scoring. In total, 9871 survivors (4725 men and 5146 women) participated in the HRQoL analyses, thus resulting in response rates of 39.50% (9871/24,993) of eligible survivors and of 51.23% (9871/19,268) of invited survivors. Of the 9871 participating survivors, almost half were from Germany 4726 (47.88%), 2146 (21.74%) from the Netherlands, 1585 (16.06%) from Switzerland, 1029 (10.42%) from the Czech Republic, and 385 (3.90%) from France (Table 5).

Most survivors were diagnosed as having cancer between the ages of 10 and 14 years (3488/9871, 34.93%) or <5 years (3201/9871, 32.43%). Almost half (4680/9871, 47.41%) were diagnosed as having cancer between 1985 and 1994, most were survivors of leukemia (3157/9871, 31.97%), lymphoma (2075/9871, 21.02%), or CNS tumors (1356/9871, 13.74%), had no history of a subsequent tumor (9225/9871, 93.46%), and received chemotherapy with or without surgery and/or radiotherapy (7645/9871, 77.45%). More than half (5460/9871, 55.31%) of the participants were aged 25 to 34 years at the time of the HRQoL study.

Response rates differed between countries. The proportions of eligible survivors who were invited to participate were 83.13% (2513/3023) in Switzerland and 82.99% (10,078/12,144) in Germany, 80.28% (861/1060) in France, 75.37% (4250/5639) in the Netherlands, and 50.40% (1576/3127) in the Czech Republic (Table 3). The proportion of responders was highest in the Czech Republic, with 65.3% (1029/3127) and lowest in France (385/1060, 45.2%).

Participants were more often women, slightly younger at the time of the survey, and slightly older at cancer diagnosis; their cancer diagnosis tended to be in more recent years, and they were more often survivors of leukemia or lymphoma and less often survivors of CNS tumors than nonparticipants (Table 6). The Czech participants had the highest proportion of women (682/3127, 66.3%) because women were prioritized during recruitment as they were also part of an associated PanCareLIFE study on female fertility [25]. In other countries, the proportion of women ranged between 48.0% (761/3023, Switzerland) and 56.6% (218/1060, France).

Discussion

Central Message

PanCareLIFE has successfully constructed a large population of 9871 childhood cancer survivors with comprehensive data to assess HRQoL and its predictors. This rich data set of survivor cohorts from 5 different countries will allow in-depth investigation of the differences in HRQoL between countries, including the effect of female fertility and hearing impairment on HRQoL. The sample size and the wide range of treatment era, type of diagnosis, and age at diagnosis provide an excellent basis for risk stratification. It will thereby provide new scientific information on risk factors for impaired HRQoL after childhood cancer. This study will be the first to estimate the association between well-defined female fertility information and HRQoL with ample statistical power. Our results will help caregivers to identify survivors at risk for decreased HRQoL and will contribute to the development of evidence-based interventions toward better HRQoL of future childhood cancer survivors. Our data allow us to investigate similar predictors of HRQoL as in previous studies [6,11] (except for annual income) and, in addition, the role of female fertility impairment.

Strengths and Limitations

This study is one of the largest studies worldwide to examine HRQoL and its predictors in childhood cancer survivors. It is, in terms of size, comparable with large studies from the United States (CCSS) [12] and the United Kingdom (BCCSS) [11]. This study combines a series of population-based or regionally well-defined cohorts from different European countries or regions. This will allow comparisons between national cohorts and representative in-depth analyses of HRQoL in survivors and their influencing factors. The central data center processed the raw data from all institutions to minimize coding and data cleaning errors before data pooling.

We faced some challenges when assembling the data from 6 institutions from 5 countries, as recruitment and study design differed between countries. Some countries had specific exclusion criteria before they invited the survivors to the survey, which may have led to selection biases. The Netherlands, the Czech Republic, and France excluded survivors with severe mental sequelae and other severe impairments. However, these were only a few patients (only 0.1% of eligible patients, n=28), and therefore, a potential bias toward an overestimation of HRQoL in these countries is negligible. Germany did not contact survivors <25 years on the date of the survey, France did not include survivors of childhood leukemia, and the Netherlands did not send data on hearing impairment. Depending on the research question, we will stratify the overall data set by age at survey (18-24 years and ≥25 years) and/or by country (including or excluding Germany and/or France and/or the Netherlands). The time elapsed between diagnosis and survey differed between countries. In Switzerland and the Czech Republic, survivors received a questionnaire already ≥5 years after diagnosis, whereas in Germany, France, and the Netherlands, it was much longer (≥10 years). However, in the SCCSS, the time since diagnosis did not influence HRQoL after adjusting for age at the time of the study [6]. We will, therefore, include age at the time of diagnosis and age at the time of the survey in all multivariable analyses. The Motol Teaching Hospital in Prague preferred female survivors to participate in the study. A high proportion of women may lead to the underestimation of HRQoL because women reported lower HRQoL than men in both CCSS [12] and BCCSS [11]. We will adjust for sex in all regression models and stratify the results by sex.

Of all countries, France had the highest percentage of survivors who were lost to follow-up and the Netherlands had the lowest. In the Netherlands, the proportion of 5-year survivors who died before the mailing of the questionnaire was highest, most probably because the Dutch cohort had the longest follow-up. Age at the time of study was a predictor for some domains of HRQoL in the SCCSS [6] and in a study from the European Organization for Research and Treatment of Cancer [28] pooling data from the general population. In future analyses, we will adjust for age at the time of the survey in multivariable regression models.

Different assessment logistics among the countries may have influenced response rates. In the Czech cohort, where most survivors were asked to participate during clinical follow-up visits, the participation rate was higher than in Switzerland, the Netherlands, or Germany, who invited all eligible participants via letters. Survivors who have a high risk for late effects are followed up with clinical visits more often than those with lower risk [29]. In the SCCSS, survivors with late effects were at risk for low HRQoL [6]. Therefore, the different sampling procedures in the Czech cohort could have resulted in a lower HRQoL score compared with the other countries. Differences in mentality and general willingness to take part in surveys may also explain the differences in response rates.

The participating survivors differed only slightly from those who did not respond, suggesting that bias from selective response may not be large. The higher proportion of participating women than men reflects a commonly observed self-selection bias in questionnaire health surveys [30]. The frequencies of observed diagnostic groups are similar to those reported from population-based cancer registries, with the following exceptions: more survivors of lymphoma and less survivors of brain tumors participated, reflecting a prognosis-based disease bias.

The countries contributing to this study assessed treatment data differently: most countries collected as-treated data retrospectively from medical records. Switzerland complemented intention-to-treat (ITT) data, with retrospective data from medical records. Germany used solely ITT data from a study protocol database, which did not have treatment data information for 23% of the survivors.

Overall, the limitations refer to different aspects of the data and are mostly specific for each country; therefore, we will provide overall and country-specific results and perform sensitivity analyses excluding countries with specific properties. We will also include a careful evaluation of the differences in data ascertainment and data quality between countries when interpreting the results for specific research questions.

Conclusions and Prospects for Further Research

With careful interpretation, the large data set of this PanCareLIFE study provides a unique opportunity to study long-term HRQoL among childhood cancer survivors across Europe. It will contribute to the knowledge on HRQoL after childhood cancer while acknowledging the differences between countries in treatment traditions and long-term care. It will also allow the investigation of the role of female fertility and hearing impairment on HRQoL. The results may uncover unknown risk factors for reduced HRQoL and will help inform clinicians about certain groups of survivors who have greater needs for counseling and psychological support to obtain the best possible HRQoL.