Cohort description

Full details of the UKCCS case-control study (www.UKCCS.org), which provides the foundation of the matched cohort described here, have been published.8 10 18 Briefly, the sampling frame for cases and controls comprised all children registered with the National Health Service (NHS) in England, Scotland or Wales at the time the study was conducted (>98% of the childhood population in these countries); unfortunately, for logistical reasons Northern Ireland (~3.5% of the total UK population aged 0–14) was not included.

Overseen by a management committee that included epidemiologists, statisticians, paediatric oncologists/haematologists and expert scientists that was chaired by Professor Sir Richard Doll, 10 UKCCS administrative areas covering the whole of Britain were created for the purposes of data collection, with each being the responsibility of a well-established epidemiological research centre8 (online supplemental figure 1).

The study began on 1 January 1991 in Scotland and on either 1 April 1992 or 1 September 1992 in the nine UKCCS administrative areas of England and Wales. In Scotland, case accrual ended on 31 December 1994, and in England and Wales, it was restricted to children diagnosed with leukaemia or non-Hodgkin's lymphoma throughout 1995 and leukaemia alone throughout 1996. In order to ensure ascertainment completeness, proactive cancer notification systems were established in all diagnostic/treatment centres across the UK, the majority of cases being notified at the point of diagnosis directly to the relevant UKCCS administrative centre by paediatric oncologists or haematologists practising in the study area. Subsequent crosschecks were made against regional childhood cancer registries (where they existed); the population-based National Registry of Childhood Tumours (NRCT), which covered England, Scotland and Wales8 19; and clinical trial datasets (acute lymphoblastic leukaemia).

With the permission of the child’s clinical team, the parents of children with cancer were subsequently contacted. Each child whose parents agreed to be interviewed (3835/4430; 87%) was individually matched on sex, date of birth (month and year) and UKCCS region of residence to 10 controls who were randomly selected from the same population-based health service authority list as the case. With their general practitioner’s (GP’s) permission, the parents of the first two controls on the list (‘first-choice’ controls) were contacted and asked to participate in the study; but if the GP declined, or the parents did not wish to be interviewed, the next child was selected, and this procedure was repeated until two control families agreed to take part. By the end of the data collection phase, 96.7% (3786/3835) cases had two participating controls and 1.3% (49/3835) had one.

In order to monitor the characteristics of respondents and non-respondents across the study, and examine the potential impact of selection/participation bias, the registration details of all cases and controls were retained, regardless of whether or not their families were interviewed. Using standard methods, address postcodes were used to allocate deprivation scores for the home at diagnosis (all registered cases) or pseudo-diagnosis (all registered controls; date coinciding with the exact age that the corresponding case was diagnosed) and also at the time they were born (using mother’s address on the child’s birth certificate). Full details of the methods used have been previously published.8 18 Briefly, deprivation categories were then derived by dividing the continuous score for the national 1991 census enumeration areas into five equally sized groups, with group 1 representing the most affluent and group 5 the most deprived.

Illustrating the UKCCS’s potential to examine issues relating to selection bias, the first two columns of table 1 distribute all registered subjects in the case-control study according to the area-based deprivation score of their home at diagnosis (cases) or pseudo-diagnosis (controls; date coinciding with the exact age that the corresponding case was diagnosed). With roughly 20% of the population in each quintile, the deprivation distribution of the 5-year survivor cohorts (last two columns of table 1, right side) mirrors that of the case-control study (table 1, left side).

As expected, the deprivation distribution of those whose parents participated in the case-control study differed from those who did not (table 2). With 32.7% of non-participating controls residing in the most deprived quintile, the effect was more pronounced than in the case population (28.4%). Exacerbated by the fact that participation among case families was higher (3835/4430=86.6%) than among first-choice control families (5526/7658=72.2%), the deprivation distributions of interviewed cases (n=3835, column 1) and first-choice controls (n=5526, column 3) differ significantly from each other (χ²=15.1, p=0.004). As expected, distributing individuals in the 5-year survivors cohort by the same categories produced similar differences (table 2, right side). Such categorisation will allow us not only to evaluate the potential impact of deprivation at birth and diagnosis on hospital activity among 5-year survivors but also to examine selection bias.

Follow-up

The fact that basic details (matching variables, postcode, NHS number) of all cases and controls were registered and retained, regardless of the families’ participation, enabled the conversion of the study into a matched cohort that could track all registered cases (n=4430) and their corresponding first-choice controls (n=7658) who were targeted for interview (table 1; first two columns). In addition to the targeted case and control populations, replacement controls are being tracked, enabling the study to also make comparisons between the 3835 cases (table 2, column 1) and 7621 controls (5526 first-choice (table 2, column 3) and 2095 replacement controls) whose parents participated.

Tracking all of the registered cases and controls, the UKCCS now operates on a legal basis that permits follow-up information to be obtained from NHS administrative healthcare records without explicit consent, enabling all cases and their corresponding controls to be ‘flagged’ at the national level and tracked via linkage to nationwide information on deaths, cancer registrations and hospital episode data (inpatient day and overnight admissions; outpatient appointments and visits; accident and emergency presentations and maternity admissions). Importantly, in line with current ethical and governance requirements, the matched cohorts are now pseudonymised; NHS identifiers are no longer held by the study but by the national administrative bodies supplying linked data (NHS England, Public Health Scotland, NHS Central Register Scotland, and the Secured Anonymised Information Linkage Databank).

Beginning with the numbers registered in the original case-control study, the flow diagram presented in figure 1 gives the number of cases and controls who survived 5 years or more, followed by the number of cases who were successfully linked to national administrative data sources. Cases enter the survivors’ cohort exactly 5 years after their cancer was first diagnosed and controls 5 years after their corresponding case’s cancer was first diagnosed (pseudo-diagnosis date). Controls who emigrated, died or were lost to follow-up before 5 years had elapsed are not included in the survivor cohort, nor are controls who were matched to cases who died within 5 years of their diagnosis, since, where applicable, their follow-up was truncated when their matched case died (figure 1; bottom row). All those who were targeted for interview in the original case-control study are shown on the left side of figure 1 and all those whose parents were interviewed on the right side; the total number of individuals that are being followed is shown in the yellow boxes in the centre of the figure.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

UKCCS subjects distributed by parental interviewed status (total targeted and total interviewed) and follow-up status; the number of 5-year survivors with linked health data who are being followed is presented in the bottom row.

With respect to the available data, the time periods currently covered by the various linked healthcare datasets, which are updated on an annual basis, are shown in online supplemental figure 2; and the key data domains and fields are listed in table 3.

Supplemental material

[bmjopen-2023-073712supp002.pdf]

View this table:

• View inline
• View popup

Table 3

National administrative data available for comparative analysis: United Kingdom Childhood Cancer Study (UKCCS)

Findings to date

The UKCCS provides a valuable resource within which to examine both the potential causes and long-term health consequences of childhood cancer. Thus far, over 90 peer-reviewed articles containing UKCCS data have been published, and a full list is available on the study website (www.UKCCS.org). Assembling data from multiple sources, key reports include ALL genome-wide association studies (GWAS),14 20 and those examining the potential aetiological roles of a wide range of biological (eg, infections/markers of infectious exposure),21–23 physical (eg, ionising/non-ionising radiation)12 13 24 25 and chemical (eg, prescription drugs/smoking) agents acting on parents before the child’s conception and the child during pregnancy/early life,10 26 27 as well as relationships with other illnesses (eg, allergy) and birth characteristics (eg, birth weight).21 28 29 Importantly, the UKCCS accessed medical records to help answer key questions relating to prescription drugs, illness histories and comorbidities; such contemporaneously recorded information, which by its nature is free from differential recall bias, often producing very different results from that obtained by self-report.11 22 30 31

With respect to the longer term health and healthcare needs of childhood cancer survivors, the conversion of the UKCCS into a matched cohort is now yielding important results about the morbidity and mortality of childhood cancer survivors. Early findings based on the case population alone highlighted survival variations with age and sex across a range of subtypes,15 and for acute lymphoblastic leukaemia, survival associations with area-based measures of deprivation at the time of diagnosis were also noted.16 With additional years of follow-up, we are now able to report a 25-year survival of 65.3% for all cancers combined (95% CI 63.8–66.7). More recently, we have begun examining secondary care hospital activity patterns, comparing findings in the case population to those in the control population over the first 25 years. Adding to excess risks of death and cancer, our first report demonstrated that survivors of childhood acute lymphoblastic leukaemia (ALL) experienced excess outpatient and inpatient activity across their teenage and young adult years, which was not related to routine follow-up monitoring. Indeed, with hospital activity being higher than expected in specialties covering most organ and tissue systems, survivors were more than twice as likely to fall under the care of endocrinology, cardiology, respiratory medicine, ophthalmology, neurology and/or gastroenterology specialists.17 Furthermore, these differences showed no signs of diminishing in the first 25 years of follow-up, underscoring the need to take prior cancer drug and/or radiation treatment into account when interpreting seemingly unrelated symptoms in later life.

Strengths and limitations

Covering England, Scotland and Wales, findings from this population-based age- and sex-matched cohort study are based on routinely compiled linked health data, not self-report. With minimal selection bias and loss to follow-up, the matched-cohort design allows more granular analyses to be performed than is possible in case cohorts that rely solely on general population rates, as has been done in other UK-based studies.32–35 For example, at the point of cohort entry, the individually age- and sex-matched control population has a similar deprivation distribution to the case population, providing a robust baseline against which to evaluate the potential impact of deprivation across the life-course. In addition, because the UKCCS cohorts can be divided into those whose families participated in the original case-control study and those who did not, the study is well-placed to investigate the impact of selection bias: an analysis that most cohorts cannot undertake, largely because they were not designed to do so, most being either record-based (as on the left side of figure 1)36–39 or interview-based (as on the right side of figure 1).40 41 Thus far, with follow-up extending over 25 years from diagnosis, we have examined health through the teenage and early adult years, and individuals in the matched cohorts will continue to be tracked as they age. With respect to limitations, lack of reliable details about ethnicity as well as information from primary care and psychosocial services are obvious deficiencies that currently affect all UK record-linkage studies of the type described here.42–44 Analyses are also constrained by the fact that national administrative healthcare data are primarily collected for administrative and clinical purposes and not for research.

Collaboration

To protect privacy and confidentiality, approval for the linkage of the UKCCS to health data is provided under strict conditions for the storage, retention and use of the data. The current approvals permit storage of the data at one site (University of York) for use by individually named researchers. We encourage interested parties to contact us to discuss potential analyses. Reasonable requests for data access can be submitted to the study Principal Investigator, ER (eve.roman@york.ac.uk), for review by the UKCCS investigator team. The PI will contact the relevant agencies to explore the feasibility of data sharing subject to ethical and data access agreements; should data sharing be agreeable with the agencies, approval for data access and specific analyses will be required from the appropriate ethics committees and data sharing agreements put in place.