Modern approaches to data collections in Neuroepidemiology

Background The use of administrative databases has been implemented in epidemiological fields. Record linkage systems match data across administrative health databases supplying large samples for long-term observational studies, the preferred non-experimental approach to test hypotheses. The present project shows the application of modern methodological approaches to the study of neurological diseases, specifically: I. to investigate the association between exposure to breastfeeding and the occurrence of MS in adulthood using prospectively collected community-based data; II. to assess the effect of antibiotic exposure on the risk of developing MS in the Emilia-Romagna region (RER), Italy; III. to assess the accuracy of death certificates (DC) in the identification of subjects with amyotrophic lateral sclerosis in the Limousin region, France. Materials and Methods Study I, a community-based cohort study, linking prospectively collected information on breastfeeding from the Cohort of Norway community-based surveys on health status with the Norwegian MS Registry for MS clinical information, and the population-based Medical Birth Registry of Norway. Associations between breastfeeding and MS onset were estimated. II: A population based nested case-control study in the Emilia Romagna region. All patients with MS diagnosis seen at MS-centers (2015-2017) aged ≥18 years were eligible. For each patient demographic and clinical information were collected and 5 controls, selected among RER residents, were matched on age, sex, place of residence and index year. Information on antibiotic prescription was obtained through the linkage with the RER drug prescription database (ATC code “J01”). III: The French register of ALS cases in the Limousin region (FRALim) was used as gold standard for the population-based validation study. DC information of patients in the FRALim register and deceased within 2000-2011, including ICD10 codes for specific cause of death, were provided. Sensitivity, positive predictive value (PPV) were calculated. Results I: No association was found between having been breastfed at least 4 months and MS risk, adjusting for various maternal factors (HR = 0.90; 95%CI 0.68-1.19). The estimates did not change for different durations of breastfeeding or adjusting for perinatal factors. II: During the study period 877 subjects with MS and 4,205 controls were included. Exposure to any antibiotic in the three years before index was associated with MS (OR=1.52, 1.29–1.79). An exposure in the 8 and 13 years before showed increased effects size (OR=1.95, 1.44-2.63 and OR=3.04, 1.07-8.68, respectively). Similar results were found for different classes of antibiotics or time lag exposures. No dose-response effect was found. III: DCs were available for 197 ALS patients, of whom 185 (93.9%) were correctly identified by the ALS ICD10 code. The overall sensitivity was 93.9% (95% CI 89.6-96.8) and the PPV was 64.9 (59.1-70.4). Stratification for sex, age and year at death did not show difference in accuracy. Discussion and Conclusion The studies presented showed most of the values and issues in applying epidemiological methods to large collections of data. The different risks of bias in our studies were acknowledged and, whenever possible, overcome. The most frequently reported were information bias, with misclassification of exposure or outcome, and the lack of accuracy of measurement methods. Selection bias was avoided using population-based samples. The lack of covariates in most situations can only be reduced using appropriate surrogate variables. The application of neuroepidemiology to healthcare databases should therefore be an effort to maximize the signal over the noise, since the large amount of data may amplify the risk of bias. The use of this information need to be considered carefully and its validity should be assessed to correctly interpret the findings.