The activity of cancer registries represents a multistep process that starts by gathering information from a variety of sources. Such information is checked, linked, enriched and handled to produce high-quality original data capable of being informative enough to prove useful in answering specific epidemiological and clinical questions.
This thesis is part of a PhD by Prior Publication grounded in six published papers. These papers deal with different steps in the production of cancer registry data, enhancing the contribution of registries to cancer epidemiology. Skin melanoma has been used as an example, but all the presented methods and concepts apply to any cancer type.
Materials and methods
1. The first paper (related to cancer registry data quality) tests the hypothesis whether the distribution of the first digit (from one to nine) of crude incidence rates obeys Benford law. Pearson’s coefficient of correlation and different distance measures were applied to compare the theoretical distribution to the observed one in a sample of 43 population-based cancer registry populations randomly drawn from the volume X of Cancer Incidence in 5 Continents.
2. In the second paper, an innovative index for measuring the amount of internal variability among the sub-areas underlying an overall incidence rate is presented. The measure is a ratio, where the numerator is the difference between the highest and the lowest age-adjusted standardised rate in sub-areas. The denominator is the overall area age-adjusted standardised rate. Such measure was applied to age-standardised incidence rates for ‘all cancer sites excluding non-melanoma skin cancer’, for men, in 2014, for Nordic countries as a whole, for each country (Denmark, Faroe Islands, Finland, Greenland, Iceland, Sweden and Norway) and their regions.
3. In paper three, to make cancer registry data useful in the clinical setting, melanoma incidence during 1985–2004 in the Tuscan cancer registry (Italy) was analysed including both standard (site, morphology, sex, age, calendar period) and clinically relevant variables, as in situ melanoma and Breslow’s thickness. For the time trend analysis, the annual percent change (APC) of the rates was computed.
4. Paper 4 presents the results of an age-period-cohort model applied to 1977 skin melanomas, incident in the Tuscan cancer registry. Such a method allows us to understand the time trend better and to forecast future change. Moreover, a non-linear regression model was applied to estimate the expected number of new cases in a more recent period.
5. Paper 5 shows a skin melanoma survival analysis based on 1403 patients from two Italian registries (Tuscan and Reggio-Emilia). The focus was on two different approaches: the multivariate Cox proportional hazard model and the Classification And Regression Trees analysis. The latter is an automatic method that splits data through a recursive process creating a ‘tree’ of groups with different profiles of risk of death. Both ways were applied to the following variables: age, sex, Breslow thickness, Clark level, Registry, sub-site and morphologic type.
6. In Paper 6, the quality of melanoma diagnosis and care in the Tuscan region is measured based on 13 newly realised process indicators, which encompassed early diagnosis, pathology reporting and surgical treatment. We evaluated the clinical adherence to these indicators in two years: 2004 and 2008, using a population-based series of incident skin melanomas, measuring the possible changes in the indexes following the implementation of specific regional recommendations.
1. The distribution of the first significant digits of cancer incidence rates was shown to belong to numbers that abide by Benford law, in the whole dataset (146,590 rates) by sex and cancer registries. The correlation coefficient between observed and expected distributions was extremely high (0.999), and the distance measures very small.
2. The index for internal variability highlighted a quite relevant heterogeneity among Nordic countries (index 57.1% = the difference between the Nordic country with the highest and the one with the lowest rate is 57.1% of the Nordic overall age-adjusted rate). Within countries, the variability was negligible in Iceland (9.6%), and high in Sweden (37.1%).
3. During the four analysed periods standardized melanoma incidence rose significantly, for both invasive (APC = + 5.1%) and in situ lesions (APC = + 11.1). Over time, the median value of thickness decreased from 1.68 to 0.8 mm (P < 0.001), but only for <=1.00 mm melanomas. Although the rates of thin melanoma have increased, rates for thick ones did not decrease.
4. The model that best fitted the data included age and ‘drift’. The linear effect (‘drift’) showed, in each age group, an increase of the risk of malignant melanoma diagnosis of about 36.6% every five years of period or cohort. For the period 2002–2006, 1112 new cases were predicted with a standardised rate (age 15–84 years) of 19.2 × 100.000. In the Tuscany Cancer Registry area, no clues for malignant melanoma incidence rates levelling off were documented. Growing rates and numbers of malignant melanoma are expected soon.
5. The Cox proportional hazard model found sex, age, Breslow thickness, Clark and morphologic type to have a significant independent prognostic value. The Classification And Regression Trees analysis identified six groups of different risks based on Breslow thickness, age and sex. The best prognostic group (5-year observed survival, 98.1%) included those subjects with Breslow less than 0.94 mm and age 19–44 years. The same thickness but an older age (50–69 years) was associated with a statistically significant different prognosis (5-year observed survival, 92.8%).
6. As regards the quality of care, there were statistically significant increases in the percentage of thin (<= 1 mm) melanomas from 2004 to 2008 (from 50.7 to 61.3%) and in the number of pathology reports that mentioned ulceration (from 61.4 to 84.6%) and margin statuses (from 76.8 to 84.3%). The percentage of patients staged by sentinel lymph node biopsy was stable (63%) and was higher for patients younger than 75 years of age (74%). The number of lymph nodes almost invariably exceeded the proposed site-specific cut-off reference, and, in 2008, the number of nodes removed was always reported for lymphadenectomy. From 2004 to 2008, surgical and pathological waiting times increased.
The six presented papers cope cohesively with consecutive steps in the procedure of cancer registration.
1. The check for Benford law abidance may become a preliminary test in the process of data quality.
2. The heterogeneity index offers a new, simple (to be produced and to be understood) and noteworthy information.
3. The analysis of some clinically crucial variables raises the interest of clinicians and makes cancer registries closer to the real world.
4. The use of methods with higher statistical involvement, e.g., the age-period-cohort model, provided further information on melanoma time trends in the area. Moreover, estimates were projected to a more recent period bridging the Registry’s timeliness gap.
5. Prognosis is a piece of vital information for both patients and clinicians. Hazard ratios and patients grouping showed almost the same risk patterns but conveyed by a different message (relative vs absolute), with a different understandability.
6. Population-based quality indexes allow to check the practical application of guidelines and recommendations, highlighting critical situations to be improved.
Cancer registration is a unique process made by different but connected steps. The improvement of each of them positively affects others. It is a sort of virtuous circle in which new methods, new uses and new users are all involved in a common aim: exploiting cancer registries’ activity.
Registries can have a real informative power only if the whole process, from the collection of raw data to the provision of relevant information for various stakeholders, is accomplished.