Capture-recapture suggests that the NRCT ascertained nearly all cases of childhood cancer diagnosed during 2003–04 in residents of Britain. Some of the completeness estimates were slightly reduced when it was assumed that all CR notifications received between February 2007 and April 2009 for previously CCLG-only cases were due to feedback through routine data exchange. Even under this assumption, however, the completeness estimate was 99% overall. Stratification by various factors that might affect the probability of ascertainment made almost no difference to the overall estimate.
The capture-recapture estimates must be interpreted with caution, because there was undoubtedly some dependence between the two sources. There may have been direct transfer of data to some individual CRs from their local CCLG clinical centres before the NRCT was notified. More generally, both sources use records from NHS cancer treatment centres, and both might tend to miss patients with non-fatal disease treated by clinicians not specialising in oncology.
Although the HES cross-check did not find convincing evidence of a serious failure of registration by the NRCT, we cannot be certain without further enquiries that no relevant cases were missed. Moreover, HES data are clearly not independent of other hospital-based sources of notification; in particular, the cross-check could not detect any unregistered children who were treated by non-NHS clinicians, or were diagnosed only around the time of death. Recent completeness of ascertainment for early-childhood leukaemia in residents of England may not be typical of other periods, ages and diagnostic groups, or of Britain in general.
It is clear that direct notification to the NRCT from specialist clinicians was a valuable supplement to the general cancer registration system. Of the 2985 patients known to the NRCT who were resident in Britain and diagnosed during 2003–2004, the CRs notified between 92% and 96% overall (depending on the feedback assumption), and no more than 90% in at least one region. Overall, the CCLG notified a similar percentage of registrations (93%).
During 1972–73 in the region of North West England covered by the Manchester Children's Tumour Registry (MCTR), 91% of known childhood cancer cases were ascertained to the MCTR from Hospital Activity Analysis records (a predecessor of HES), 93% from the regional CR, and 93% from clinicians; 98–99% could have been ascertained from any two of these three sources (Leck et al, 1976
). The percentages notified by CR and clinicians seem remarkably similar to those reported here for national registration in Britain 30 years later.
A previous two-source capture-recapture study (Draper et al, 1991
) estimated NRCT completeness to be 99% for childhood leukaemia and non-Hodgkin lymphoma in Britain diagnosed during 1966–83. Although the result resembles the estimates for leukaemia and lymphoma presented here, the assumption of independence was more questionable in the earlier study. In particular, the two sources of ascertainment were defined so that one source included notification from CRs whereas the other included notification from death certificates; during the 1960s and early 1970s, death certificates were an important component of non-CR ascertainment of leukaemia to the NRCT, and death certificates have always been a source of ascertainment to the CRs. Evidence from one region suggested a rather lower level of completeness during 1972–1984: a cross-check with clinical records found that 137 (95%) of 144 childhood leukaemia patients treated at hospitals in the Oxford region had been ascertained by the NRCT (Draper et al, 1989
In New Zealand, for the diagnosis period 1990–93, three different sources of ascertainment notified, respectively, 97%, 98% and 86% of known cases of childhood cancer, with an overall capture-recapture estimate of nearly 100% (Dockerty et al, 1997
). In three cities in Brazil, for childhood acute leukaemia diagnosed in 2001, there were 55, 44 and 28 known cases, respectively; overall, 48% were notified by CRs and 60% by a diagnostic laboratory database (Azevedo-Silva et al, 2009
). Two-source capture-recapture estimates for the true numbers of cases in the three cities were 71, 82 and 89, implying relatively low completeness estimates of 77%, 54% and 31%, consistent with the difficulties of cancer registration in a huge and developing country.
In the present study, as far as possible, univariate stratified analysis was used to assess and allow for potential dependence between sources. This is equivalent to fitting a log-linear capture-recapture model with a single categorical covariate (Tilling et al, 2001
). Analysis was also stratified by SES within diagnostic group and within age group. As there was very little evidence of variation in completeness with any of the factors considered, we chose not to apply the more sophisticated modelling methods that are appropriate for the joint effects of several categorical covariates, or for continuous covariates (Tilling and Sterne, 1999
The proportion of registrations notified by the CCLG was very high (97% or more) for the typical cancers of infancy and early childhood normally diagnosed in specialist paediatric oncology centres (leukaemia, neuroblastoma, retinoblastoma, and renal and hepatic cancers), and rather lower in diagnostic groups that are relatively frequent in older children, teenagers and adults. Accordingly, the proportion notified by the CCLG was slightly lower at ages 10–14 years than in younger age groups. In contrast, the proportion notified by the CRs was lower for infants than for older children under one of the feedback assumptions (and may have been as low as 83% for retinoblastoma).
Both sources notified slightly higher proportions of registrations for patients who died than for other patients. This is consistent with the possibility that registration might be slightly more complete for patients with more aggressive cancers (because they are more likely to be seen by specialist clinicians) and for patients who die (because the death certificate is a source of notification to the CRs).
For leukaemia and lymphoid leukaemia registrations from England and Wales, under one of the two feedback assumptions only, the proportion of CCLG-only notifications increased with deprivation. There was a contrary trend in CR-only notification for CNS tumours, suggesting that in deprived communities a higher proportion of registered cases had been referred to CCLG centres. A slight deficit of CR notifications from deprived communities is a possible explanation for both types of trend.
As described, notification patterns varied slightly according to geographical area, diagnostic group, age at diagnosis, whether the patient had died, and (for CNS tumours and perhaps leukaemia) SES. However, effects on completeness were trivial, because the great majority of registrations were notified by both sources. The capture-recapture estimate was 99–100% in almost every subgroup; it was slightly lower in the Thames region (98–99%), and in two diagnostic groups that are relatively rare in children (germ-cell/gonadal cancer 98–99%, melanoma and non-skin carcinoma 97–98%). There was very little evidence that completeness varied according to age at diagnosis, or whether the patient had died. There was no evidence that ascertainment was appreciably less complete in deprived areas of England and Wales than in affluent areas.