Appendix

Bias analysis

Methods. In addition, we conducted probabilistic bias analysis to formally address the potential for uncontrolled confounding of the PD-smoking association; i.e. we performed “reverse” bias analysis posing the question “what would be characteristics of a confounder (U) that could explain away/nullify the observed smoking/nicotine PD associations?” using previously published bias adjustment formulas (REF: Arah et al. (2008); for examples see Sudan et al Ped Perinat Epidemiol 2013; Parriott & Arah 2013).

We specified varying distributions (prevalence) of a binary risk factor U within categories of smoking and nicotine use and then solved the formulas for the odds ratio of the U-PD association. As depicted in the DAG, the U prevalence was chosen to represent a negative relationship with smoking and nicotine and a positive relationship with PD (see figure e-1). At each given scenario the output is the OR showing how strong the association between U and PD would have to be to nullify the observed smoking–PD associations.

Results. In the supplemental Table e-1 we present results of our bias analysis illustrating the distribution and strength of association (with PD) that would be required for a plausible PD risk factor U to explain the reported smoking-PD associations. E.g. a factor U highly prevalent among never-smokers (70%), less among former smokers (30%), and rare among current smokers (5%) would have to be strongly associated with PD (OR=5.93) to nullify the observed OR of 0.28 among current smokers (see scenario 1). With U less prevalent among never-smokers (e.g. 35%) and rare among former smokers (5%) and current smokers (1%), the U-PD association OR would need to be even larger (9.18) to reduce the OR of 0.28 to 1.Parallel findings apply for the use of nicotine substitutes.

Table e-1: Reverse bias analysis of the smoking-Parkinson’s and nicotine-Parkinson’s associations exploring plausible characteristics of an uncontrolled confounding binary variable or risk factor (U), that, when adjusted for, would explain away the observed association between smoking/nicotine and Parkinson’s disease (assuming true null associations)

Observed OR for the smoking-Parkinson’s association / Assumedtrue (U) adjusted odds ratio for the association between smoking or nicotine and Parkinson’s / Assumed prevalence of U=1 within smoking /nicotine categories / Estimated OR (95% confidence interval) for the association between U and Parkinson’s
Scenario 1
Smoking
-Never smoked / Reference / Reference / 70% / –
-Former smoker / 0.65 (0.56 – 0.76) / 1 / 30% / 3.26 (2.81 – 3.81)
-Current smoker / 0.28 (0.22 – 0.34) / 1 / 5% / 5.93 (4.66 – 7.20)
Nicotine
-Never used nicotine / 1 / Reference / 80% / –
-Used nicotine / 0.54 (0.38 – 0.76) / 1 / 30% / 4.48 (3.17 – 6.34)
Scenario 2
Smoking
-Never smoked / Reference / Reference / 65% / –
-Former smoker / 0.65 (0.56 – 0.76) / 1 / 35% / 5.83 (5.02 – 6.81)
-Current smoker / 0.28 (0.22 – 0.34) / 1 / 5% / 6.45 (5.07 – 7.84)
Nicotine
-Never used nicotine / 1 / Reference / 65% / –
-Used nicotine / 0.54 (0.38 – 0.76) / 1 / 25% / 5.55 (3.93 – 7.86)
Scenario 3
Smoking
-Never smoked / Reference / Reference / 60% / –
-Former smoker / 0.65 (0.56 – 0.76) / 1 / 25% / 3.50 (3.02 – 4.09)
-Current smoker / 0.28 (0.22 – 0.34) / 1 / 3% / 6.22 (4.89 – 7.55)
Nicotine
-Never used nicotine / 1 / Reference / 65% / –
-Used nicotine / 0.54 (0.38 – 0.76) / 1 / 15% / 3.29 (2.33 – 4.65)
Scenario 4
Smoking
-Never smoked / Reference / Reference / 50% / –
-Former smoker / 0.65 (0.56 – 0.76) / 1 / 15% / 3.00 (2.58 – 3.51)
-Current smoker / 0.28 (0.22 – 0.34) / 1 / 2% / 7.00 (5.50 – 8.50)
Nicotine
-Never used nicotine / 1 / Reference / 40% / –
-Used nicotine / 0.54 (0.38 – 0.76) / 1 / 10% / 4.97 (3.51 – 7.02)
Scenario 5
Smoking
-Never smoked / Reference / Reference / 35% / –
-Former smoker / 0.65 (0.56 – 0.76) / 1 / 5% / 2.97 (2.56 – 3.47)
-Current smoker / 0.28 (0.22 – 0.34) / 1 / 1% / 9.18 (7.21 – 9.18)
Nicotine
-Never used nicotine / 1 / Reference / 30% / –
-Used nicotine / 0.54 (0.38 – 0.76) / 1 / 5% / 5.11 (3.61 – 7.22)

Figuree-1: Signed directed acyclic graph showing hypothesized directionality of effects of an unknown powerful risk factor for Parkinson’sDisease, and assuming no causal effects of smoking and nicotine on Parkinson’s

Page 1