Quality control of automatically defined cancer cases by the automated registration system of the Venetian Tumour Registry. Quality control of cancer cases automatically registered

Quality control on digital cancer registration

Population-based cancer registration methods are subject to internationally-established rules. To ensure efficient and effective case recording, population-based cancer registries widely adopt digital processing (DP) methods. At the Veneto Tumor Registry (RTV), about 50% of all digitally-identified (putative) cases of cancer are further profiled by means of registrars' assessments (RAs). Taking these RAs for reference, the present study examines how well the registry's DP performs. A series of 1,801 (putative) incident and prevalent cancers identified using DP methods were randomly assigned to two experienced registrars (blinded to the DP output), who independently re-assessed every case. This study focuses on the concordance between the DP output and the RAs as concerns cancer status (incident versus prevalent), topography, and morphology. The RAs confirmed the cancer status emerging from DP for 1,266/1,317 incident cancers (positive predictive value [PPV] = 96.1%) and 460/472 prevalent cancers (PPV = 97.5%). This level of concordance ranks as "optimal", with a Cohen's K value of 0.91. The overall prevalence of false-positive cancer cases identified by DP was 2.9%, and was affected by the number of digital variables available. DP and the RAs were consistent in identifying cancer topography in 88.7% of cases; differences concerned different sites within the same anatomo-functional district (according to the International Agency for Research on Cancer [IARC]) in 9.6% of cases. In short, using DP for cancer case registration suffers from only trivial inconsistencies. The efficiency and reliability of digital cancer registration is influenced by the availability of good-quality clinical information, and the regular interdisciplinary monitoring of a registry's DP performance.

Indicators of guideline-concordant care in lung cancer defined with a modified Delphi method and piloted in a cohort of over 5,800 cases

BACKGROUND

To identify indicators of guideline-concordant care in lung cancer, to implement such indicators with cancer registry data linked to health databases, and to pilot them in a cohort of patients from the cancer registry of the Milan Province.

METHODS

Thirty-four indicators were selected by revision of main guidelines by cancer epidemiologists, and then evaluated by a multidisciplinary panel of clinicians involved in lung cancer care and working on the pathway of lung cancer diagnosis and treatment in the Lombardy region, Italy. With a modified Delphi method, they assessed for each indicator the content validity as a quality measure of the care pathway, the degree of modifiability from the health professional, and the relevance to the health professional. Feasibility was assessed using the cancer registry and the routine health records of the Lombardy region. Feasible indicators were then calculated in the cohort of lung cancer patients diagnosed in 2007-2012 derived from the cancer registry of the Milan Province. Criterion validity was assessed reviewing clinical records of a random sample of 114 patients (threshold for acceptable discordance ≤20%). Finally, reliability was evaluated at the provider level.

RESULTS

Initially, 34 indicators were proposed for evaluation in the first Delphi round. Of the finally 22 selected indicators, 3 were not feasible because the required information was actually not available. The remaining 19 were calculated on the pilot cohort. After assessment of criterion validity (3 eliminated), 16 indicators were retained in the final set and evaluated for reliability.

CONCLUSION

The developed and piloted set of indicators is now available to implement and monitor, over time, quality initiatives for lung cancer care in the studied health system.

Evolution of Cancer Registration Combining Online Reporting with Follow-up in the Community: Practices in Guangzhou, China

Background:

An efficient registration system with accurate and timely information on cancer incidence and mortality is key to development of policies to prevent and control cancer. A traditional registration system usually needs 3-4 years to collect data and publish a cancer report. However, researchers, policymakers and healthcare professionals need to know the latest cancer registration data quickly.

Methods:

A computer system has been operating with cases reported online by hospitals and followed up in communities at the Cancer Registry of Guangzhou (CRG) since 2008. The comparability, completeness, accuracy and timeliness of collected data were here evaluated.

Results:

From 2010 to 2014, 181,194 cancer cases from 1,916,253 medical records of cancer were reported to the CRG online. 53,473 cases were deleted as duplicates (47,906), wrong diagnoses (410) or residents of other places (5,157) during the follow up. Successful final follow-up rates were over 90% for both newly and previously diagnosed cases by general practitioners in community clinics. The CRG coding and classification system follows international standards. The annual trends for all sites by sex were stable from 2010 to 2014. All age-specific incidence rates for childhood cancers were within the limits of the respective international references. The overall M: I ratio for all sites but C44 was 56.7%., ratios for most sites in Guangzhou being between Hong Kong and Shanghai. A total of 75.7% of the cancer cases reported in 2010–2012 were morphologically verified. Ninety five percent of new cases completed registration within 29.0 months in 2010, reducing to 8.0 months in 2014.

Conclusion:

The online report system with community follow up at the CRG yields reasonably accurate and close-to-complete data. It takes less time to confirm diagnosis and other information so that reporting annual incidence one year after the close of registration becomes possible.

Building a model for disease classification integration in oncology, an approach based on the national cancer institute thesaurus

BACKGROUND

Identifying incident cancer cases within a population remains essential for scientific research in oncology. Data produced within electronic health records can be useful for this purpose. Due to the multiplicity of providers, heterogeneous terminologies such as ICD-10 and ICD-O-3 are used for oncology diagnosis recording purpose. To enable disease identification based on these diagnoses, there is a need for integrating disease classifications in oncology. Our aim was to build a model integrating concepts involved in two disease classifications, namely ICD-10 (diagnosis) and ICD-O-3 (topography and morphology), despite their structural heterogeneity. Based on the NCIt, a "derivative" model for linking diagnosis and topography-morphology combinations was defined and built. ICD-O-3 and ICD-10 codes were then used to instantiate classes of the "derivative" model. Links between terminologies obtained through the model were then compared to mappings provided by the Surveillance, Epidemiology, and End Results (SEER) program.

RESULTS

The model integrated 42% of neoplasm ICD-10 codes (excluding metastasis), 98% of ICD-O-3 morphology codes (excluding metastasis) and 68% of ICD-O-3 topography codes. For every codes instantiating at least a class in the "derivative" model, comparison with SEER mappings reveals that all mappings were actually available in the model as a link between the corresponding codes.

CONCLUSIONS

We have proposed a method to automatically build a model for integrating ICD-10 and ICD-O-3 based on the NCIt. The resulting "derivative" model is a machine understandable resource that enables an integrated view of these heterogeneous terminologies. The NCIt structure and the available relationships can help to bridge disease classifications taking into account their structural and granular heterogeneities. However, (i) inconsistencies exist within the NCIt leading to misclassifications in the "derivative" model, (ii) the "derivative" model only integrates a part of ICD-10 and ICD-O-3. The NCIt is not sufficient for integration purpose and further work based on other termino-ontological resources is needed in order to enrich the model and avoid identified inconsistencies.

Automated selection of relevant information for notification of incident cancer cases within a multisource cancer registry

OBJECTIVE

The aim of this study was to develop and evaluate a selection algorithm of relevant records for the notification of incident cases of cancer on the basis of the individual data available in a multi-source information system.

METHODS

This work was conducted on data for the year 2008 in the general cancer registry of Poitou-Charentes region (France). The selection algorithm hierarchizes information according to its level of relevance for tumoral topography and tumoral morphology independently. The selected data are combined to form composite records. These records are then grouped in respect with the notification rules of the International Agency for Research on Cancer for multiple primary cancers. The evaluation, based on recall, precision and F-measure confronted cases validated manually by the registry's physicians with tumours notified with and without records selection.

RESULTS

The analysis involved 12,346 tumours validated among 11,971 individuals. The data used were hospital discharge data (104,474 records), pathology data (21,851 records), healthcare insurance data (7508 records) and cancer care centre's data (686 records). The selection algorithm permitted performances improvement for notification of tumour topography (F-measure 0.926 with vs. 0.857 without selection) and tumour morphology (F-measure 0.805 with vs. 0.750 without selection).

CONCLUSION

These results show that selection of information according to its origin is efficient in reducing noise generated by imprecise coding. Further research is needed for solving the semantic problems relating to the integration of heterogeneous data and the use of non-structured information.

Probabilistic classifiers and automated cancer registration: an exploratory application

A test of the performance of two probabilistic classifiers (random forests and multinomial logit models) in automatically defining cancer cases has been carried out on 5608 subjects, registered by the Venetian Tumour Registry (RTV) during the years 1987-1996 and manually checked for possible second cancers that occurred during the 1997-1999 period. An eightfold cross-validation was performed to estimate the classification error; 63 predictive variables were entered into the model fitting. The random forest allows to automatically classify 45% of subjects with a classification error lower than 5%, while the corresponding error is 31% for the multilogit model. The performance of the former classifier is appealing, indicating a potential drop of manually checked cases from 1750 to 960 per incidence year with a moderate error rate. This result suggests to refine the approach and extend it to other categories of manually treated cases.

Consistency and accuracy of diagnostic cancer codes generated by automated registration: comparison with manual registration

Background

Automated procedures are increasingly used in cancer registration, and it is important that the data produced are systematically checked for consistency and accuracy. We evaluated an automated procedure for cancer registration adopted by the Lombardy Cancer Registry in 1997, comparing automatically-generated diagnostic codes with those produced manually over one year (1997).

Methods

The automatically generated cancer cases were produced by Open Registry algorithms. For manual registration, trained staff consulted clinical records, pathology reports and death certificates. The social security code, present and checked in both databases in all cases, was used to match the files in the automatic and manual databases. The cancer cases generated by the two methods were compared by manual revision.

Results

The automated procedure generated 5027 cases: 2959 (59%) were accepted automatically and 2068 (41%) were flagged for manual checking. Among the cases accepted automatically, discrepancies in data items (surname, first name, sex and date of birth) constituted 8.5% of cases, and discrepancies in the first three digits of the ICD-9 code constituted 1.6%. Among flagged cases, cancers of female genital tract, hematopoietic system, metastatic and ill-defined sites, and oropharynx predominated. The usual reasons were use of specific vs. generic codes, presence of multiple primaries, and use of extranodal vs. nodal codes for lymphomas. The percentage of automatically accepted cases ranged from 83% for breast and thyroid cancers to 13% for metastatic and ill-defined cancer sites.

Conclusion

Since 59% of cases were accepted automatically and contained relatively few, mostly trivial discrepancies, the automatic procedure is efficient for routine case generation effectively cutting the workload required for routine case checking by this amount. Among cases not accepted automatically, discrepancies were mainly due to variations in coding practice.