In a case–control design, 750 PD patients (UK Brain Bank Criteria) , with an onset of motor symptoms at age 40 or above, were recruited from the Radboud University Nijmegen Medical Centre as well as from 15 community hospitals in a wide area surrounding Nijmegen. The patients were selected upon review of medical charts, excluding subjects with cognitive decline (MMSE <27) or with physical or psychiatric illnesses that could have influenced their job opportunities. 1300 healthy controls were derived from the Nijmegen Biomedical Study, a large population-based cross-sectional survey in and around Nijmegen . We included men only, to exclude sex-related differences in job opportunities in the elderly population. All participants gave informed consent for data sharing. The study design was approved by the local ethics committee.
Patients and controls completed a structured questionnaire about their lifetime occupational status, years and level of education, and comorbidity. Patients also reported the subjective onset of their motor symptoms as well as their reasons to change or end a job. A research assistant who was blinded to the study aims classified each occupation according to the RIASEC system. This produces a three-letter code for every single occupation, related to the main personality characteristics associated with the occupation, e.g. ERS, RIA, and so on. In this study, we focused on the first letter of this code as it represents the occupation’s most dominant feature.
One hundred seventy-four questionnaires (PD: n = 57/750; controls: n = 117/1300) were not completed adequately and therefore excluded from the analyses. There were no major differences between both databases (PD and controls) in terms of basic characteristics except for a higher median age at the moment of completing the questionnaire in the PD group (69.4 vs. 51.2 years; Table 2). Therefore, we not only adjusted for age in all following calculations, but also performed additional analyses to identify any age effects (see end of this section).
We had two primary hypotheses: (1) a low PD risk for artistic occupations and (2) a high PD risk for conventional occupations. Logistic regression analyses were performed on the very first occupation and on the most recent occupation, which, in case of the patients, was the last occupation prior to the self-reported symptom onset. The primary analysis included the factors ‘artistic occupation (A; yes/no)’, ‘conventional occupation (C; yes/no)’ and ‘age’. Because we evaluated two hypotheses in the primary analysis, we used a Bonferroni correction and considered p < 0.025 statistically significant. In a secondary analysis, we investigated the role of educational level by including this as an additional factor in the logistic regression and by calculating the interaction between education and occupation. This led to p > 0.20 for the interaction term. We carried out separate analyses for subjects with educational levels up to intermediate vocational training, and those above.
We estimated odds ratios (ORs) for the artistic (A) occupations versus each of the remaining five occupational categories (R, I, S, E, C) with respect to their association with PD.
Since previous reports identified farming as a risk factor for PD, the OR of farming versus all other occupations was calculated as a validation of our data set in relationship with those previous reports. We repeated the primary analysis, both with farming as additional factor and without farming. We did not aim to prove or investigate the relationship between PD and farming in the light of toxin exposure.
We repeated the analyses restricted to subjects from the Nijmegen area to verify whether this specific geographical area would yield different results from the group in total.
With respect to the substantial age difference between PD and controls, we carried out the following additional calculations. We challenged the main assumption of linearity on the log odds scale in the logistic regression. When allowing for non-linearity, however, the effects remained almost unchanged. In particular, we performed an analysis with age, [age]2 and [age]3 as covariables, thus allowing for an effect that could change direction even up to two times. We also ran the analysis stratifying for age classes of 5 years. These ORs were then pooled and presented in an overall OR. Such an analysis does not assume any particular shape of the relationship between age, risk of PD or occupation. The mean age difference between PD and controls within the age classes was 0.02 years. We chose not to make the samples age-matched, as such post hoc matching would be rather arbitrary and would have led to a data set far too small for any reliable conclusion. Finally, we explored whether the number of job changes correlated with the risk of PD. We hypothesized that more frequent changes could be associated with a lower PD risk.