Automation in cancer registration

Background Algorithms aimed at replacing the manual decision-making process, usually carried out by registry personnel on ad hoc registry forms, were first introduced in the early 1970s by the Ontario Cancer Registry (OCR). The forms containing the information on cancer patients were coded and computerized at the OCR and subsequently treated by software developed by the OCR for this purpose. All the diagnoses of cancer were assigned by the program using the data available electronically (Clarke et al., 1991). In the early 1990s, the project of the Venetian Tumour Registry (RTV) explored for the first time the possibility of using, as primary sources of data for the registration process, electronic routine data from the hospitals, pathology departments, and other health institutions in the north-east region of Italy. The data were coded according to ICD-9 (hospital discharges and death certificates) or SNOMED (pathology records). By applying a rather simple algorithm (Table 1) to the electronic data, it was possible to assign a diagnosis to the majority of the incident cases (Simonato et al., 1996). The methodology was subsequently adopted by the Northern Ireland Cancer Registry (NICR), and partially by the Thames Cancer Registry, with similar results. More recently, a network of registries in the northeast of Italy, the North-East of Italy Cancer Surveillance Network (NEICSN) adopted and further developed the registration system. Basically the method consists of a binary decisional system of concordance/ discordance, through which a potential incident case is accepted with a consolidated diagnosis of cancer, or rejected. Cases rejected by the program are resolved manually by the registry personnel. Figure 1 illustrates the standard data flow, while an example of its application by NEICSN is shown in Figure 2. The example reported in Figure 2 shows how incidence was obtained by NEICSN for the period 1999–2000. All the electronic data available from the three sources, consisting of 4,401,914 hospital discharges, 197,859 death certificates and 2,516,832 pathology records are used in the process, in which the first phase consists of record linkage with the various sources, in order to eliminate cancer cases with diagnosis before 1 January 1999. Of these, 1,103,147 (15.4%) records with a diagnosis of neoplasia are selected and summarized into 305,369 subjects with at least one record of cancer (average number of records per subject 3.6). Out of these, 246,655 (80.8%) were prevalent cases, and 8,869 (2.9%) turned out to be non-residents at the moment of diagnosis. This leaves 49,845 subjects potentially affected by at least one incident cancer. In the following step, the cases are consolidated by the program, and 39,148 cases are entered in the registry database. These constitute 78.5% of all cancer cases. The remaining 10,697 cases are revised and 21.5% of the total are entered by the registry personnel. A large and increasing number of cases accepted belong to the categories of benign tumours, in situ tumours, and tumours of uncertain nature. This would allow follow-up studies of non-malignant tumours.