Objective: To describe a data artefact in degree-of-spread at first presentation in the New South Wales Central Cancer Registry (NSW CCR), the only Australian cancer registry that records degree-of-spread data for all solid tumours.
Method: Trends in the proportions of cancer cases diagnosed annually over 1972-2004 by degree-of-spread categories of localised, regional, distant and unknown were calculated for each major cancer type.
Results: Excepting breast cancer and melanoma, the proportion of localised cancer cases reported from 1993-1998 was approximately 5% lower than expected, and was mirrored by an artefactual increase in unknown degree-of-spread cases.
Conclusion: This artefact was caused by the introduction of the Electronic Notification System and cannot easily be remedied retrospectively. However, regional and distant categories of degree-of-spread in the NSW CCR data are reliably recorded for the 1972-2004 period.
Implications: It is important that past and present cancer data users are notified and understand the quality issues with NSW CCR degree-of-spread data, and use it as recommended to avoid anomalous results or conclusions.
The New South Wales Central Cancer Registry (NSW CCR) is the only Australian Cancer Registry that records degree-of-spread at first presentation for all malignant solid tumours. Degree-of-spread data have been collected since the registry began in 1972. These data are frequently requested and incorporated into analyses by clinicians, researchers, health planners and others.
Two major uses of the degree-of-spread data are to evaluate the effectiveness of screening programs1–3 and to analyse the survival of patients by extent of disease at first presentation.4,5
Degree-of-spread represents the most aggressive extent of the disease based on diagnostic and therapeutic evidence. The NSW CCR defines degree-of-spread, within four months of diagnosis as one of four summary stage categories: localised, regional, distant or unknown within four months of diagnosis, in accordance with international guidelines6 which are widely used by other cancer registries worldwide (e.g. The Surveillance, Epidemiology and End Results (SEER) Program in the US,7 The Denmark Cancer Registry).8 Localised spread is invasive cancer confined to the tissue of origin; regional is when the tumour has invaded adjacent tissues or organs; distant is when the tumour has extended beyond the primary site either outside the regional boundary or by metastasis to distant lymph nodes or distant sites; and unknown is where there is insufficient information to assign a stage. Summary stage is less detailed than other staging methods such as the Tumour, Nodes and Metastases (TNM) Staging System. Nonetheless, as a summary staging method, degree-of-spread effectively delineates differences in survival by each category4,7 In addition, the proportion of people presenting with localised disease is often used as a clinical indicator. This is to explore the differentials in access to healthcare services prior to diagnosis – where diagnosis with more advanced disease is seen as a sign of disadvantage,8 and to evaluate the effectiveness of screening programs – where a shift to higher proportions of more localised degree-of-spread in screened populations compared to unscreened populations, is a key indicator of the benefit of screening.9–11
This paper describes an artefact in the NSW CCR degree-of-spread data that was detected in all solid-tumour cancers excepting breast and melanoma, occurring from 1993 to 1998.
The population-based NSW CCR has recorded malignant cancer diagnoses in NSW residents since 1972. Since 1986 it has been a statutory requirement for all public and private hospitals, pathology laboratories, outpatient and radiotherapy departments, nursing homes and day-procedure centres to notify all invasive cancer cases to the NSW CCR.
The proportion of localised, regional, distant and unknown cases for each maj or invasive solid tumour cancer type diagnosed annually from 1972 to 2004 was calculated from the NSW CCR incidence data and plotted as line graphs, with delineation of the six-year period in which the degree-of-spread reporting artefact occurred (1993-98). Expected numbers of cases by degree-of-spread during the artefact period were estimated using linear interpolation based on actual cases diagnosed three years before and after the period.
The degree-of-spread trend for all cancers combined shows a decline in localised cancers (5.2% below expected) mirrored as an increase (7.5% above expected) in the unknown degree-of-spread category for the period 1993-1998 (Figure 1a). Excepting breast cancer and melanoma, this characteristic trend pattern was present for all maj or cancers, but trends in the regional and distant degree-of-spread categories were not affected (see Figures 1b-d, also supplementary figures available from author on request).
A subsequent investigation found that the introduction of the hospital Electronic Notification (EN) System in 1994 caused the artefact. With the changeover to EN, degree-of-spread was not included as a field in the original EN system but was available in the upgraded system in 1999. Regional and distant categories were still accurately recorded because these could be derived from the site-of-metastasis field on the EN System. Thus, the unknown category from 1993-98 consists of a proportion of localised cases not recorded as localised by the EN System, and true unknown cases.
Public hospitals started to use the EN System after it was first implemented in 1994, but private and smaller public institutions continued to submit paper notifications. For cases diagnosed in 1994, 39.4% of inpatient notifications were received electronically, growing to 52.2% by 1998. Hence, there remains a proportion of cancer cases that still could be assigned accurately as localised or unknown degree-of-spread during 1993-98. For the cases submitted by hospital EN, degree-of-spread data had to be extracted from pathology reports which were, and remain to this day, notified on paper. Consequently, if the information provided on the pathology report was such that degree-of-spread could not be derived, then these cases would have been assigned an unknown degree-of-spread value. Thus, the proportion of unknown degree-of-spread cases for most cancers increased artefactually during 1993-98. When the EN System was upgraded and the degree-of-spread field was added in 1999, localised cases were again identifiable directly.
Although the implementation of the EN System in 1994 caused the artefact, cases diagnosed in 1993 were still affected, and when the EN System was upgraded in 1999 cases diagnosed in 1998 were the last year to be affected by the artefact. This is because notifications are usually submitted to the NSW CCR in the course of the year following diagnosis. Hence, the artefact can be detected in cases diagnosed from 1993 to 1998.
A true case of unknown degree-of-spread (at first presentation) occurs when the information notified to the NSW CCR is insufficient to assign a summary stage category. The NSW CCR has rigorous criteria for determining which pathology samples can be used for reporting degree-of-spread. For example, excision biopsies, where the tumour has been completely excised with a narrow margin of normal tissue, are valid samples to derive degree-of-spread from, but incision and punch biopsies, tissue fragments or fine needle aspirations are not. Accordingly, the proportion of true unknown category can vary across cancer sites, and can depend on the treatment practice for a given cancer and consequently the specimens typically taken. In particular, for some sites such as melanoma and breast, the pathology reports contain well-documented information (e.g. Clark's level) enabling coding staff to assign degree-of-spread without the additional clinical information being received from the hospital. For other cancers, where clinical information is required to assign the degree of spread, the artefact is most noticeable.
It is recommended that users of the degree-of-spread data for 1993-98 combine the unknown and localised categories if using aggregated data. For assessing survival by degree-of-spread, survival rates for 1993-98 for localised and unknown categories could be estimated from interpolating the survival rate trends calculated before and after this period. Screening program effectiveness can still be evaluated during the artefact period by determining whether the proportion of regional and distant cases decreased, as these categories are unaffected by the data artefact, rather than if the proportion of localised cases has increased.
Unfortunately, it is not possible to correct retrospectively the degree-of-spread artefact. The hospital records in question would almost certainly have been archived or destroyed, and the resources required to retrieve and re-code this information preclude the possibility of rectifying the artefact.
As the NSW CCR degree-of-spread data has been widely requested, we anticipate that many studies that used the artefactual data may have anomalous conclusions. To exemplify, from a brief literature search we suspect that the results of Yu et al., 2006 may be significantly affected by the artefactual degree-of-spread data5 and other publications perhaps less so.3,12
Accordingly, we strongly recommend that researchers who may have used degree-of-spread information provided by the NSW CCR in the past review carefully the implications of this report in relation to any results that they may have reported, or with regard to any policies that may have emanated from such results.
We would particularly like to thank staff of the Central Cancer Registry for their work in processing and coding the data. We would also like to thank Claire Cooke-Yarborough, NSW CCR Medical Advisor, for helpful discussions concerning the degree-of-spread coding and Adrian Grundy for formatting the figures. We appreciate the co-operation of statutory notifiers in the supply of invaluable notifications and the assistance of medical records personnel, clinicians and pathologists in meeting requests for supplementary information. The NSW Central Cancer Registry is managed and funded by the Cancer Institute NSW under a Memorandum of Understanding with the NSW Health Department.