Background Previous studies have demonstrated negative associations between cigarette smoking, alcohol consumption and caffeine intake and Parkinson's disease (PD), but most data come from unrepresentative cohorts (1). One previous study suggested different associations with these exposures in different PD phenotypes (2). There are no data describing change in these exposures after diagnosis: a differential change between cases and controls would introduce bias to prevalence-based case-control studies.

Methods We recruited a community-based, inception cohort of PD and matched controls and gathered data about previous duration and intensity of cigarette smoking and alcohol and caffeine intake with face-to-face questionnaires. We calculated odds ratios and adjusted odds ratios using logistic regression analysis to examine the associations between these exposures and PD and the tremor-dominant and postural instability and gait difficulty (PIGD) subtypes. We also assessed the change in exposures after recruitment in patients and controls with a linear mixed model.

Results 201 patients with PD and 249 controls were analysed. Moderate and high cumulative levels of smoking, OR 0.42 (0.24–0.74) and 0.48 (0.28–0.82) respectively; high alcohol intake, OR 0.27 (0.13–0.56); and moderate and high caffeine consumption, ORs 0.48 (0.28–0.82) and 0.39 (0.22–0.68) respectively; were statistically significant negative associations with PD. There were statistically significant inverse trends between rising levels of consumption for each exposure and PD (P<0.001 for smoking, P=0.003 for alcohol and P<0.001 for caffeine). A similar pattern of associations with these exposures was seen in the PIGD and tremor-dominant patients as in the whole PD group. Smoking and alcohol consumption declined after diagnosis, but not at different rates between patients and controls.

Conclusions Our study therefore provides further evidence for dose-dependent negative associations between these exposures and PD in a truly-representative inception cohort. We found no evidence of different associations with the tremor-dominant and PIGD PD subtypes. The lack of a differential reduction in exposure is important for interpretation of prevalent studies.