Concordance of National Cancer Registration with self-reported breast, bowel and lung cancer in England and Wales: a prospective cohort study within the UK Collaborative Trial of Ovarian Cancer Screening.

BACKGROUND: It has been suggested that lower UK cancer survival may be due to incomplete case ascertainment by cancer registries.
METHODS: We assessed concordance between self-reported breast, bowel and lung cancer and cancer registration (CR) for 1995-2007 in England and Wales in the UK Collaborative Trial of Ovarian Cancer Screening.
RESULTS: Concordance of breast cancer CR was higher (94.7%:95% CI: 94.1-95.3%) than for bowel (85.1%:95% CI: 82.1-87.8%) and lung (85.4%:95% CI: 76.3-92.0%). CR concordance was lower in breast cancer (94.5% vs 98.8%) survivors compared with deceased but the difference was small. No difference was found for bowel (85.3% vs 94.6%) or lung (87.1% vs 90.5%) cancer.
CONCLUSION: Concordance of CR and self-reported cancer is high. Incomplete registration is unlikely to be a major cause of lower UK survival rates.

International comparisons of cancer incidence and mortality are dependent on completeness of cancer registration (CR). Recently, lower survival rates have been reported for those diagnosed with primary bowel, lung, breast and ovarian cancer in the UK (1995–2007) compared with countries with similar national cancer plans (Coleman ). Concern was raised that incomplete case ascertainment by cancer registries particularly of survivors may have contributed to this finding (Robinson ). We report using independent case ascertainment (Parkin ; self-reporting in the course of a clinical trial) on concordance between CR in England and Wales for 1995–2007 and self-reported breast, bowel or lung cancer in women participating in the UK Collaborative Trial of Ovarian Cancer Screening (UKCTOCS).

Materials and Methods

This prospective study was nested within the UKCTOCS cohort of 189 063 women, aged 50–74 years recruited from England and Wales (Menon , 2009). Women recruited from Northern Ireland were excluded. At recruitment, women reported on breast, bowel, lung or other cancer diagnosed before enrollment and date of diagnosis (Menon ). At postal follow-up 3–5 years later, women were asked about ovarian, breast, bowel, lung or other cancers diagnosed after joining UKCTOCS, year of diagnosis and where treatment took place. All women provided written consent.

Women were ‘flagged' through Health and Social Care Information Centre (HSCIC). All deaths and cancer diagnoses were notified using the International Classification of Disease Codes (ICD-9 and ICD-10). No data were provided on whether CR was death certificate-initiated or death certificate-only. Up-to-date CR was received on 11 October 2012.

Women self-reporting a single breast, bowel or lung cancer diagnosis at recruitment or follow-up with diagnosis dates between 1 January 1995 and 31 December 2007 were included. CR was defined as concordant if the woman had a relevant CR irrespective of number/year of CRs. Sensitivity analysis included women who self-reported more than one of the above cancers. Subgroup analysis was undertaken by regional cancer registry with Gateshead/Middlesborough assigned to Northern and Yorkshire, Liverpool/Manchester to North Western, Nottingham/Derby to Trent, Portsmouth/Bristol to South West, Royal Free/Barts to Thames, Cardiff/North Wales to Wales Cancer Registries.

Timely reporting of death is a legal requirement in the UK and the mention of cancer on the death certificate triggers collation of additional medical information. We therefore assessed concordance of CR in survivors vs deceased women with the relevant cancer as primary or contributory cause of death. Women who had died with no mention of the relevant cancer on the death certificate were excluded from this analysis. In those residing in England who self-reported relevant cancer but had missing CR, in-patient and out-patient Hospital Episodes Statistics (HES) records for 2001–2010 were searched for relevant ICD codes to explore whether inclusion would improve CR concordance.

Results

Of 189 055 (99.9%) who gave consent for follow-up, 10 390(5.5%) self-reported breast, bowel or lung cancer. In all, 1313 were excluded as they had not reported a diagnosis date (463) or had reported diagnosis after 2007 (256), two separate diagnosis dates (229) and multiple primary cancers (365). Of the remaining 9067 women, 5917 (5210 breast, 618 bowel and 89 lung) self-reported cancer diagnosis between 1995 and 2007 (Table 1).

Table 1

(A) Number of women from England and Wales reporting breast, bowel or lung cancer diagnosis between 1995 and 2007 who had relevant cancer; (B) in women who self-reported more than one of these cancers

(A)
ICD-9 and ICD-10 codes	No. of women with CR	% Concordance	95% CI
Self-reported cancer history
Breast cancer (n=5210)a
C50/174 CR	4358	83.6	82.6–84.6%
D05/233 CR	578	11.1	10.3–12.0%
Total (invasive Br Ca+in situ carcinoma)	4936	94.7	94.1–95.3%
Bowel cancer (n=618)
C18/C19/C20/C21/C17/153/154 CRb (includes D01 in situ carcinoma)	526	85.1	82.1–87.8%
Lung cancer (n=89)
C33/C34/162 CR	76	85.4	76.3–92.0%
(B)
ICD-9 and ICD-10 codes	No. of women with CR	% Agreement	95% CI
Breast cancer (n=5274)a
C50/174 CR	4389	84.2	82.6–84.6%
D05/233 CR	582	11.2	10.3–12.0%
Total (invasive Br Ca+in situ carcinoma)	4971	94.3	94.1–95.3%
Bowel cancer (n=669)
C18/C19/C20/C21/C17/153/154 CRb (includes D01 in situ carcinoma)	559	83.6	81.7–87.5%
Lung cancer (n=125)
C33/C34/162 CR	81	64.8	76.3–92.0%

Abbreviations: Br Ca= Breast cancer; CI=confidence interval; CR=cancer registry; DCIS=ductal carcinoma in situ; ICD-9; ICD-10=international causes of diseases and health-related problems (9th and 10th revision).

One woman did not give permission to ONS flagging.

Includes two women with C17..

The numbers/percentages in bold represent the results per breast (overall), bowel and lung cancer separately.

For breast cancer, the concordance of CR was 94.7% (95% CI: 94.1–95.3%) with 11.1% (95% CI: 10.3–12.0%) of women having a ductal carcinoma in situ only registration (Table 1A). For bowel (including four carcinoma in situ) and lung cancer, the concordance of CR was 85.1% (95%CI: 82.1–87.8%) and 85.4% (95% CI: 76.3–92.0%), respectively. The difference in concordance of CR between breast, bowel and lung cancer was significant (P<0.0001). Concordance improved with increasing year of diagnosis for self-reported breast cancers but not for bowel or lung cancer (data not shown). Sensitivity analysis including the 75 women who self-reported multiple of these cancers demonstrated similar concordance for breast (94.3%) and bowel (83.6%) but lower (64.8%) for lung (Table 1B). Age of diagnosis (self-reported) differed in those with CR (mean 58.5, s.d. 6.8) compared with those without (mean 57.3, s.d. 7.2; P=0.001). Agreement between self-reported year of diagnosis and first relevant CR diagnosis date (defined as ±1 year) was 95.0% (5260 out of 5538) and was similar across the cancers.

CR concordance was lower in breast cancer (94.5% vs 98.8%, P=0.001) survivors compared with deceased but the difference was small. No difference was found for bowel (85.3% vs 94.6%, P=0.106) or lung (87.1% vs 90.5%, P=0.681) cancer (Table 2).

Table 2

Comparison of concordance of CR and self-reporting of breast, bowel or lung cancer between 1995 and 2007 in survivors vs deceased

		CR for relevant cancer not available	CR
	No. of women	No. of women	No. of women	%	95% CI	P-value
Self-reported breast cancer
Breast cancer primary or contributory cause of deatha	333	4	329	98.8	96.9–99.7%	0.001
Survivors in cohort	4704	260	4444	94.5	93.8–95.1%
Self-reported bowel cancer
Bowel cancer primary or contributing cause of deathb,c	37	2	35	94.6	81.8–99.3%	0.106
Survivors in cohort	545	80	465	85.3	82.1–88.2%
Self-reported lung cancer
Lung cancer primary or contributing cause of death	21	2	19	90.5	69.6–98.2%	0.681
Survivors in cohort	62	8	54	87.1	76.1–94.3%

Abbreviations: CR=cancer registration, includes in situ carcinomas of the breast and bowel; CI=confidence interval.

Excluded 173 women who died from causes unrelated to breast cancer.

Excluded 26 women who died from causes unrelated to bowel cancer.

Includes C17 and C21.

Concordance for all three cancers combined by Cancer Registry ranged from 96.0% (Wales) to 86.4% (Thames), (Figure 1). In those diagnosed between 2001 and 2007 with missing CR, HES records contained additional information on 31 (1.3%) women with breast, 9 (2.7%) with bowel and 3 (5.2%) with lung cancer. These records have the potential to contribute to cancer registry data in the future. Inclusion increased concordance of CR to 96.7% for breast, 90.1% for bowel and 89.7% for lung cancer (data not shown).

Figure 1

Concordance of self-reporting and CR for relevant cancers diagnosed between 1995 and 2007 in England and Wales by regional Cancer Registry.        .

Discussion

Our study of 6000 women nested in a cohort of 190 000 shows 95% concordance of national CR for self-reported breast cancer in England and Wales between 1995 and 2007, 85% for bowel and 85% for lung cancers. This compares well with reports (91–99%) from other population-based registries (Jensen , Lang , Gathani ). The concordance of CR was significantly lower in breast cancer survivors compared with women with a death certificate where it was a primary or contributing cause of death but the difference was only 5%. Modelling suggests that disparity between concordance of CR for deceased patients and survivors needs to be around 40% to account for half of the survival differences between UK and other countries (Woods ). Our data refute the suggestions (Beral and Peto, 2010) that the lower UK breast cancer survival is influenced mainly by lower CR in survivors. Recent reports suggests that more likely contributors to lower UK cancer survival are low cancer awareness, increased perceived barriers to symptomatic presentation, concern about wasting the doctor's time (Forbes ), suboptimal treatment of advanced ovarian (Maringe ) and bowel cancer (Dejardin ).

Highest concordance of CR for self-reported breast followed by bowel and lung cancer was in keeping with previous findings (Parikh-Patel , Dominguez ). The majority of women in these studies were aged >50 years and similar to those in our cohort. Concordance of CR is linked to the accuracy of self-reporting. Women have been found to be more accurate in reporting cancer history than men (Desai ). Presence of established screening programmes also increases concordance. Self-reporting is also likely to be more accurate for carcinomas such as breast and bowel, with a definite histological diagnosis compared with lung cancer where presence of secondary lesions can lead to misinterpretation (Colditz ). This is supported by the lower concordance with CR of self-reported lung cancer.

We further supplemented CR with HES data as Cancer Registries are using the hospital in-patient episode data as a registration source since 2009 (UKACR, 2011). This increased overall concordance by 1.6% with the highest increase (5.2%) in lung cancer. Similar results have been reported by Moller ) and are in keeping with Swedish findings that a proportion of missing CR relates to cancer cases with no pathology reports where the diagnosis is recorded on the clinical letters, operation notes and imaging reports (Barlow ).

The concordance varied by Cancer Registry from 86% (Thames Cancer Registry) to 96% (Wales Cancer Registry), possibly reflecting increased use of private health care in London. It has been previously reported that patients receiving private cancer care have missing registrations (Stotter ) as CR in the UK (NCIN, 2012), unlike in Sweden, is not mandatory (Barlow ). The mobile nature of the London population may have also contributed.

Strengths include the size (6000 women nested within a prospective cohort of 190 000), complete ‘flagging' through HSCIC because of availability of NHS number, use of additional information from HES and death certificates and follow-up for 5-years post censorship. We focussed on breast, bowel and lung cancer as participants were specifically asked about these cancers on two occasions (recruitment and follow-up). We chose cancer self-reporting as an independent measure against which to compare CR. The former is not a gold-standard and has limitations—some women may misreport a cancer or self-report in the absence of cancer (Fourkala ). Another potential bias is that those who died shortly after their diagnosis may not have had the chance to complete the follow-up questionnaire. Small numbers make it difficult to draw definitive conclusions on lung cancers.

Knowledge of whether CR is complete and/or representative of all patients diagnosed with cancer is crucial to researchers as well as policymakers and clinicians who interpret and act on the research findings. Our study suggests high overall concordance of national CR with self-reported breast, bowel and lung cancer. Ongoing evaluation is important.