Cancer survival analysis in Hamburg 1995–2003: Assessing the data quality within a population-based registry

Abstract

Introduction. Population-based cancer survival analysis constitutes valuable reference material for the clinical field of oncology. The objectives of this study were to assess the quality of the Hamburg Cancer Registry's (HCR) database in this respect, to perform survival analyses by means of selected sites, and to evaluate the results in relation to prevalent opinions and external estimates. Methods. Data quality was assessed by the proportion of cases documented as diagnosed at death, external estimates of completeness, reliability of follow-up, histological verification and information on stage. Included were first primary malignancies of the colon (ICD10 C18, n=4 544), female breast (C50, n=9 259), prostate (C61, n=5 707) and urinary bladder (C67, D09.0, n=3 148), diagnosed in Hamburg residents 1995–2003. Observed and relative survival (OS, RS) were estimated by site, sex, time, age and stage. Results. Regarding female breast cancer in Hamburg, high levels of data quality and completeness exist while the explanatory power concerning malignancies of the colon, prostate and urinary bladder is limited. Age-standardised 5-year relative cancer survival estimates amounted for female breast to 81%, for colon to 49% (male) and 52% (female), for prostate to 81% and for urinary bladder to 71% (male) and 62% (female). Conclusion. The study demonstrates the capacities and limitations of an epidemiological cancer registry to produce convincing survival estimates for clinical use, under the terms of a voluntary case reporting system.

Population-based cancer survival analysis can give useful indications regarding the effectiveness of cancer related activities within a healthcare system. Recently published survival estimates referring to malignancies diagnosed in Germany since 1995 predominantly have been derived from the cancer registries of Saarland and Munich in the South of Germany or from the New Laender in the East of Germany [1–7]. However, a substantial data stock has been collected by the Hamburg Cancer Registry (HCR) in Northern Germany, which started operation in 1929 [8], [9]. Since 1985, when the corresponding federal state law ‘Hamburgisches Krebsregistergesetz’ became effective, epidemiological cancer registration in the 1.7 million metropolitan area of Hamburg has been based on the voluntary right of physicians to notify cases dependent on the patient's informed consent. By 1995 the overall proportion of those reported by death certificates only (DCO-cases) could be reduced to below 20%, and since then varied from 8 to 14% per year with a decreasing trend. Information on actual deaths is continuously derived from matching all death certificates issued in Hamburg with the entire HCR's database. Additionally in 2004 the continuous update of documented personal data concerning changes of name, sex, place of residence and vital status, started by a cooperation with the residential registry office, which was introduced by a total cross-validation of all patients kept as ‘alive’ at that time. Thus for some years before 2004 0.6% registered cancer patients are documented as migrated from Hamburg annually, compared to constantly 4% on average thereafter. Correspondingly an unpublished active follow-up of 2 962 registered colorectal cancer cases in Hamburg residents diagnosed in 1990–1998 proved information concerning vital status to be more than 95% correct (R. Fertmann, unpublished observation, Hamburg 2005). The remaining 5% consisted of <1% unknown deaths, <1% lost to follow-ups and 4% migrants.

Comprehensive cancer survival analysis based on epidemiological registry data from Hamburg has not yet been performed, and comparable results from Northern Germany are scarcely published [6], [7]. Meanwhile several neighbouring European cancer registries have established survival estimates as part of their periodic reports providing a sound basis for quality assessment within clinical oncology [10–14].

The aim of this study is to assess the HCR's data quality concerning survival analysis and to perform analyses within selected malignancies. The results are evaluated in relation to estimates of neighbouring cancer registries. It is intended to appraise the databases clinical usability with respect to cancer survival estimation.

Material and methods

The study was limited to first primary malignancies diagnosed in Hamburg residents. With regard to an exemplary and explorative analysis, four cancer sites and an appropriate period of diagnosis were selected according to the following criteria: Entities were considered among the ten most frequent malignancies in Hamburg, with at least 100 notified cases per sex and year, representing gender-specific occurrence vs. similar manifestation in both sexes and diverse degrees of completeness. A maximum allowance of 20% cases annually documented as diagnosed at death (DCO-notifications and clinical reports omitting date of diagnosis) was set to restrict the possible bias of survival estimates to a tolerable level [15]. Estimates of completeness by the Central Cancer Surveillance Programme at the Robert Koch-Institute (RKI) were consulted as an external rating of data quality, although they refer to notified cases regardless whether first and second primaries. This indicator is derived by an indirect method which estimates the disease's frequency in the catchment area concerned, based on the data of a cancer registry where case registration is complete. Under the assumption that no major differences exist with regard to the case fatality of cancers among German states and that different regional cancer risks are reflected by differences in mortality, the age-specific proportions of incidence to mortality are used to estimate the frequency of cancer indirectly. An iterative procedure is used to model the quotients in a log-linear approach with polynomial time trends. Initially, the reference registry has been represented by the Saarland Cancer Registry [16], [17].

The selection resulted in malignancies of the colon (ICD10 C18), female breast (ICD10 C50), prostate (ICD10 61) and urinary bladder (ICD10 C67 and D09.0) of which were diagnosed in 1995–2003 (Table I). Excluded in each site were second primary tumours, DCO-cases and other cases with date of diagnosis documented as equal to death. Sporadically dubious case documentation was clarified individually by comparing originally reported data and processed best-of information. Urinary bladder invasive and in situ neoplasms (ICD10 C67 and D09.0) were grouped together because of changes in the coding of non-invasive papillary transitional cell carcinoma in 2000 [18], [19]. The selection of 15–99-year-old patients resulted in an actual age range of 18–99 years at diagnosis in the present study. Ninety-one percent of the included cases were histologically verified. For each site the trend of case numbers, indicators of completeness and the proportion of cases documented with stage (local, regional, distant metastases) and T-category according to the TNM scheme was identified. The mortality-to-incidence (M/I) ratio trends are commented in respect of each site. The overall values are listed in Table I.

Table I.  Selection of database for exemplary survival analysis from Hamburg Cancer Registry (HCR)

		Colon		Breast	Prostate	Urinary bladder
		Men	Women	Women	Men	Men	Women
(A)	Incidence /100,000 1995–2003	31.8	25.2	105.1	85.3	27.5^3	6.7^3
	in Hamburg (European standard)
(B)	Proportion of ICD10 C00-C96	7%	10%	29%	20%	6%^3	3%^3
	without C44 (1995-2003) in %
(C.)	Mortality Incidence ratio M/I	0.58	0.64	0.38	0.36	0.37	0.52
(D)	Estimate of completeness^1
	in % of expected cases	77%	72%	93%	77%	64%^3	63%^3
(E)	First primary malignancies in
	Hamburg as in HCR 1995-2003	2322	3216	9986	6646	2480	1017
(F)	DCO and other excluded	362	632	727	939	200	149
	cases^2 (% of (E))	(16%)	(20%)	(7%)	(14%)	(8%)	(15%)
(G)	Included cases	1960	2584	9259	5707	2280	868
	(% of (E))	(84%)	(80%)	(93%)	(86%)	(92%)	(85%)

¹Estimates of completeness referring to all primary malignancies without DCO, as calculated by the Robert Koch Institute [2], here: average estimates concerning HCR-data of 1995-2003, anent urinary bladder ICD10 C67 only (Dr B. Görsch, RKI)

²DCO-cases and others with documented date of diagnosis equal to date of death

³Referring to C67 only

Information on survival was derived from routine matching with the Hamburg death certificates and with the Hamburg residential registration office's data up to the end of 2004. Survival time was calculated using the difference in quarters between the date of diagnosis and date of moving outside of Hamburg, death or December 31, 2004, whichever was earliest. Measures of observed and relative survival (OS, RS) were calculated for multiple-year cohorts. The observed survival, defined as the proportion of patients surviving a specific amount of time after cancer diagnosis, was computed according to the life-table method [20–22]. Relative survival, defined as the ratio between observed and expected survival in a comparable group of the regional population, was estimated according to Hakulinen [22–24]. Reference data for expected survival estimates originated from sex- and calendar year-specific life tables for Hamburg by the statistical state office. The life tables’ ranges of age (0–90 years) were extended up to the age of 99, according to the Gompertz distribution by means of logistic regression [25]. Standard errors (SE) of the estimated relative survival were obtained using Greenwood's formula; the 95%-confidence interval (95%CI) was approximated by addition and subtraction of twice the standard error [22], [26]. Crude relative survival ratios were estimated by site, sex, age-group and consecutive triennial calendar periods of diagnosis. In addition, for the purpose of comparison, age standardized 3- and 5-year relative survival ratios were calculated by the direct method using European standard population for cancer-survival analysis [27].

The acceptance of calculated values into graphs was limited to the condition that at the beginning of the respective interval at least 30 patients were alive and SE ≤25%. Descriptive data analysis was conducted by means of SPSS version 12.0 and Microsoft Excel 2003. Survival analysis was performed by using SURV3, a shareware package from the Finnish Cancer Registry [28], [29].

Results

Colon

The ratio of female (f) to male (m) colon cancer patients was 1.4:1. From 1995 to 2003 the annual numbers of included cases slowly dropped from 263 to 203 (m) and from 320 to 247 (f), i.e. by approximately a fifth in both sexes without obvious changes in the distribution of age and stage. Sixty-nine point eight years represented the median age at diagnosis in men compared to 75.6 years in women. Sixty-eight percent of the included cases were staged, 59% were provided with T-categories. Sixteen percent (m) and 20% (f) of exclusions, fluctuating M/I ratios between 0.47–0.62 (m) and 0.57–0.78 (f), and an estimated completeness of 77% (m) and 72% (f) indicate a certain amount of underregistration (Table I).

The survival analysis shows almost equal crude estimates for both sexes (5 years RS 50% (m) and, 51% (f)) except minor differences in the ratios by age-group (Table II). An obvious increase in long-term relative survival curves is seen between the 1995–1997- and the 1998–2000-cohort, yet no further improvement for those diagnosed in 2001–2003 (Figure 1a, 1b).

Figure 1.  Crude relative survival (RS) stratified by tumour site, sex and period of diagnosis

Table II.  One- and five-year observed and relative survival (OS and RS in %) for selected tumour sites diagnosed in 1995–2003, by sex and age-group.

				Three-year survival (%)			Five-year survival (%)
Tumour Site		Age-group	Number of Patients	OS	RS	(95% CI)	OS	RS	(95% CI)
Colon	Men	Total crude	1960	50	58	(55–61)	38	50	(47–53)
		Total age-standardized		51	58	(52–64)	40	49	(42–56)
		15–44	58	71	71	(59–84)	59	60	(45–74)
		45–54	127	53	54	(45–63)	40	41	(31–51)
		55–64	467	63	66	(61–70)	50	54	(49–60)
		65–74	641	53	59	(55–64)	42	51	(46–56)
		75 +	667	35	49	(43–54)	25	43	(37–50)
Colon	Women	Total crude	2584	48	56	(54–59)	39	51	(48–53)
		Total age-standardized		54	59	(54–65)	45	52	(46–58)
		15–44	58	63	63	(50–77)	57	57	(43–71)
		45–54	124	68	68	(60–77)	55	57	(46–67)
		55–64	348	63	64	(58–70)	52	54	(48–60)
		65–74	663	55	58	(54–62)	46	52	(47–56)
		75 +	1391	40	51	(48–55)	30	47	(43–52)
Breast	Women	Total crude	9259	84	89	(88–90)	75	83	(82–84)
		Total age-standardized		80	87	(85–89)	71	81	(79–84)
		15–44	970	88	88	(86–90)	80	81	(78–84)
		45–54	1779	92	93	(91–94)	86	87	(85–89)
		55–64	2872	89	91	(90–92)	82	85	(83–87)
		65–74	1986	84	88	(87–90)	75	83	(80–85)
		75 +	1652	64	79	(76–82)	51	75	(71–79)
Prostate	Men	Total crude	5707	79	90	(89–91)	68	86	(84–88)
		Total age-standardized		78	87	(63–81)	67	81	(77–86)
		15–54	235	88	90	(85–95)	81	84	(78–90)
		55–64	1764	92	96	(95–98)	84	92	(89–94)
		65–74	2325	86	95	(93–96)	76	92	(89–94)
		75–84	1058	59	76	(72–80)	42	66	(61–72)
		85 +	325	27	47	(39–56)	15	39	(28–51)
Bladder	Men	Total crude	2280	65	76	(74–78)	54	70	(67–73)
		Total age-standardized		68	76	(72–81)	58	71	(65–77)
		15–44	59	86	86	(77–96)	82	84	(72–95)
		45–54	157	82	84	(77–90)	76	79	(72–87)
		55–64	501	77	80	(76–84)	69	75	(70–80)
		65–74	756	73	81	(77–85)	58	71	(66–76)
		75 +	807	47	64	(59–69)	35	61	(54–67)
Bladder	Women	Total crude	868	55	63	(59–67)	47	56	(56–65)
		Total age-standardized		60	65	(56–74)	53	62	(52–72)
		15–44	19	63	63	(41–86)	53	53	(30–76)
		45–54	45	67	68	(53–83)	67	69	(54–83)
		55–64	115	68	70	(61–79)	64	67	(57–77)
		65–74	272	67	71	(65–78)	61	68	(61–75)
		75 +	417	42	54	(48–61)	32	50	(43–58)

Breast

The included cases of female breast cancer ranged on a fluctuating level of approximately 1 000–1 200 per year. The median age at diagnosis gradually rose from 60 (1995) to 62 years (2003). The low proportion of exclusions (7%), a steadily from 0.40 to 0.34 declining M/I ratio, and an estimated completeness of 95%, indicate a high degree of coverage (Table I). A good data quality is confirmed by 94% histologically verified diagnoses and 83% specified stages. T-categories were available for 67% of included cases.

The relative survival curves for female breast cancer are almost lineally and slowly declining on a high level (5 years RS 83%). Regarding age-specific survival only patients above 74 years have distinctly inferior values (5 years RS 75%) compared to the other age-groups (Figure 1d, Table II). Crude relative survival curves relating to three consecutive triennials of diagnosis show constant increases beyond the first year after diagnosis. Stratification by stage applied to 5-years RS of moving diagnosis periods from 1995–1997 to 1998–2000 supports this finding for local and regional stages only (Figure 2).

Figure 2.  Trends of stage-specific five-year relative survival in breast cancer patients (moving three-year cohorts). Loc: Local stage, Reg: Regional spread, Met: Distant Metastases

Prostate

Prostate cancer case numbers increased constantly from less than 500 cases per year (1995) to more than 900 (2003), simultaneously the median age at diagnosis dropped from 70 to 67 years. 14% of excluded cases correspond reasonably to overall 77% estimated completeness. The annual M/I ratio declined from 0.39 to 0.23 (Table I). Of the included, 61% are staged and 52% provided with T-categories.

Long-term observed and relative survival diverges more distinctly compared to other sites investigated in this study (Table II). The descent of the relative survival curve is constantly slow for the whole cohort (5 years RS 86%). The 55–74-year-old patients present the best estimates (5 years RS 92%) in contrast to significantly subjacent rates of those aged 75 years and more (5 years RS 66% for 75–84 and 39% for 85 + ) (Table II). Three successive periods of diagnosis show clear increases for crude relative survival curves of prostate cancer (Figure 1c).

Urinary bladder

The cohort shows rather constant annual case numbers of approximately 250 (m) and 100 (f), men being affected 2.4 times as often and at a younger age than women (median age at diagnosis 70.7 (m) and 73.9 (f) years). Sixty-eight percent of the cases got staged, 64% were equipped with T-categories. Concerning indicators of completeness, comparatively minor proportions of DCO and equivalent cases (8% (m) and 15% (f)) diverge from the RKI-estimates (64% (m) and 63% (f)). M/I ratios fluctuated particularly in respect of women in a range of 0.43–0.61, compared to 0.30–0.41 in men (Table I).

Female urinary bladder cancer patients exhibit significantly worse observed and relative survival than male patients (5 years RS 56% (f) vs. 70% (m)) (Table II). Relative survival of the age-groups 75 years and more is inferior to that of younger patients with a major difference in women. Crude relative survival curves of consecutive triennials of diagnosis show scant alterations in men. In women, an apparent increase between the 1995–1997 cohort (5 years RS 56%) and the 1998–2000 cohort (5 years RS 64%) is followed by a decrease (Figure 1e, f). Female urinary bladder cancer patients diagnosed 2001–2003 appear to have survived similarly to those from 1995–1997 during the first two years, but then show a moderately improved relative survival remaining below that of the 1998–2000 cohort.

Discussion

The HCR's data stock is heterogeneous in respect of periods and sites. Generally data quality concerning follow-up information is maintained on a high level. According to an active trace back study in 2005, the vital status has been documented accurately in more than 95% of the cases, and the current migration ascertainment of registered patients will improve this aspect even more. With regard to the influence of vital status misclassification on survival estimates, a magnitude of 5% at the most might result in a deviation of approximately 1% in 5-year relative survival ratios [15]. A major proportion of histologically verified diagnoses and declining trends of DCO cases since the mid-nineties indicate a substantial and increasing degree of registration. In the future further improvement will be achieved by the currently established pathologist's legal obligation to report cancer cases to the HCR, which is in force since 2007 April 24. Yet the proportion of excluded cases in this study, which are documented as diagnosed at death either as DCO, autopsy or incomplete case report, still exceeds that of many publications on population-based cancer survival analysis [30–33]. In addition the ascertained two-thirds amount of staged cases is in need of enhancement, not to mention merely 52–67% of T-categories available at present.

Population based survival analysis referring to malignancies of the colon, female breast, prostate and urinary bladder in Hamburg however revealed a differentiated picture. In the following common statements and currently available data on screening activities, measures of early detection and therapeutical developments are taken into account. A comparison with five-year relative survival ratios, issued in the context of the EUROCARE-4 study and by neighbouring cancer registries for population-based patient cohorts diagnosed during similar periods, supplements the site-specific reflection [33]. Though age-standardized values are included as far as possible, the comparison between regions has to be regarded with utmost caution given the differences concerning populations, conditions of registration, inclusion criteria and methodological details (Figure 3 a), 3 b), detailed references there).

Figure 3.  Population-based 5-year relative survival estimates for colon, breast and prostate cancer patient cohorts in neighbouring regions, diagnosed in the stated period and age-standardised if labelled ‘a.s’(references in legend)

Colon malignancies in Hamburg residents were registered as having been diagnosed prior to death at a rate of 70–80% according to the surrogate markers of completeness. Fluctuating M/I-ratios with an average exceeding most of those from neighbouring cancer registries (as published for the years 1998–2002 in Cancer in Five Continents IX) add to the suspicion of underregistration [34]. Furthermore the almost 20% drop of case numbers over time is inconsistent with the constant trend estimated for the whole of Germany [2]. There is no lucid explanation for this decline apart from growing obligatory medical documentation which may effect voluntary requirements like cancer reports to the HCR [35].

The similarity of relative survival among male and female patients, a decline in higher age-groups and the improvement during the nineties correspond to observations within other population-based survival studies [2], [6], [30], [32]. Still the absence of further increase for the most recent patient cohorts and an altogether 5-year relative survival ratio of 50%, which ranks among the lower estimates of comparative registries, remain difficult to be interpreted (Figure 3a). The assumption of up to a fourth of non-reported cases and missing information on stages for roughly half of all presumably incident colon malignancies in Hamburg during the study period impedes further findings.

The collection of data on survival relevant therapeutical care is beyond the scope of this study. Concerning early detection colorectal cancer screening activities during the observation period were limited to optional faecal occult blood test and digital rectal examination from the age of 45, with an estimated attendance of less than half of the entitled women and below seventh of entitled men in Germany [36], [37]. In October 2002 screening colonoscopy was introduced into the National Cancer Prevention Programme in Germany. Less than 3% of the eligible population aged 55 and above responded to this subsequently [36], [38], [39].

Female breast cancer in Hamburg has been documented at a 93% level of completeness, and approximately 80% of staged cases permit differentiated survival-analysis. The relative survival curves show the characteristic shape lacking a plateau within the first decade after diagnosis [40], [41]. A continuous increase during the period of observation corresponds to a slow decline of M/I ratios and to the presumption of earlier diagnoses due to opportunistic screening in Germany [42], [43]. According to available data in 1996 5.2 mill mammographies have been performed, averaging out to one in 6.5 female Germans above the age of 20 [42]. A similar calculation for Hamburg in 2000 results in a numerical proportion of 1:4.7, suggesting an increasing use of mammograms during the period of observation (Hamburg Association of Statutory Health Insurance Physicians (KVH), unpublished communication, Hamburg 2005).

In the presence of distant malignancies the survival trend of breast cancer patients did not improve, thus consistent with recently published results (Figure 2) [5], [44]. Whether the positive tendency found both for local and regional stages could be explained by stage migration cannot convincingly be elucidated by the available database: Crude stage-specific incidence for the included metastasized breast cancer cases remained stable on a level of 7/100 000 over the period of observation while that for local tumours slightly fell from 50 to 46/100 000, and that for regional spreading rose from 24 to 32/100 000. An overall age-standardised 5-year relative survival estimate for female breast cancer of 81% ranges near the upper equivalent figures issued by neighbouring European regions (Figure 3a), all of which have been performing mammography screening schemes during the period of observation [45].

The rising prostate cancer incidence figures have to be cautiously interpreted in the presence of presumably spreading analyses of prostate-specific antigen (PSA) [46]. A considerable increase of PSA-testing may be assumed according to accessible information: In 1995 1.6 mill PSA-analyses have been billed within the German statutory health insurance system, i.e. approximately one test in 25 members of the male population [47]. In 2000 the same ratio in Hamburg amounted to 1:6 (KVH, unpublished communication, Hamburg 2005). Further the proportion of under-70-year-old patients increased from 46% in 1995 to 65% in 2003, and within the T-staged fraction T1-tumors advanced from 18% to 28%. These observations support an incidence elevating effect of the suspected opportunistic PSA-screening. Regarding data quality rising estimates of up to 96% completeness in 2003 and simultaneously growing proportions of staged cases (above 70% in 2003) hint at an improving, yet not throughout satisfactory condition.

Survival analysis resulted in a putatively pronounced increase of prognosis since the mid-nineties. This observation must be interpreted in connection with spreading PSA-testing and its effects of advancing lead-time and overdetection of prostate cancer [48]. As in other studies, the youngest patients do not have the best outcome [49]. The overall 5-year relative survival of prostate cancer affected men in Hamburg ranges among the upper half of comparative external estimates (Figure 3a).

Regarding urinary bladder cancer in this study, the degree of coverage remains difficult to assess because of contrasting indicators of completeness. This may be interpreted partially as a consequence of classification changes during the relevant period and their heterogeneous realization among German cancer registries. The applicable RKI estimates of completeness depend on the definition of ICD10 C67 as including non-invasive papillary tumours or not by the contributing German registries during the relevant period.

Concerning survival estimates for Hamburg, male bladder cancer patients have significantly better probabilities than female ones, a finding that had been confirmed by others before lacking conclusive explanation [50]. The time trends of relative survival curves do show almost unchanged shapes in men while female bladder cancer patients seem to improve until 2000 and worsen thereafter. The latter observation has to be relativized regarding the largely overlapping confidence intervals of second and third diagnosis periods’ 5-year relative survival: 62–75% for 1998–2000 vs. 57–73% for 2001–2003.

As survival for bladder cancer patient cohorts has not been estimated within the EUROCARE-4 study, alternative publications of neighbouring cancer registries are supplied and shown in a separate diagram (Figure 3 b) [2], [7], [10], [11], [13], [14], [51], [52]. The more or less comparable European 5-year relative survival estimates exhibit a considerable range of variation of up to 22%, partially due to lack of stratification by sex and too diverse classification of non-invasive malignancies (Figure 3). The overall rate for male bladder cancer patients from Hamburg ranges near a mean estimate of neighbouring regions, while the female counterpart marks the lower end.

In conclusion the present study demonstrates the capacities and limitations of an epidemiological cancer registry to produce convincing survival estimates for clinical use, under the terms of a voluntary case reporting system. The situation is determined by a high standard of data maintenance on the part of the HCR, by the capability of improvement on the part of those generating notifications, and by an obvious necessity of numerical data on oncological care. Nevertheless, the basis of clinical usability is formed in terms of comparing survival estimates of selected oncological patients with those of Hamburg. Evaluations like the present one, with a timely feedback to the clinical field, reveal the assessment of oncological performance quality, and have the potential to set off a productive dialogue among those who are engaged in prevention and therapy of cancer.

Acknowledgements

We are grateful to Ms. Nolene Sheppard for the textual revision of this paper.