To assess associations between Parkinson disease (PD) and putatively protective factors—smoking, caffeine (coffee, tea, and soft drinks), and nonsteroidal anti-inflammatory drugs (aspirin, ibuprofen, and naproxen).
Family-based case-control study.
Academic medical center clinic.
A total of 356 case subjects and 317 family controls who self-reported environmental exposures.
Main Outcome Measures
Associations between PD and environmental measures (history, status, dosage, duration, and intensity) of smoking, coffee, caffeine, nonsteroidal anti-inflammatory drugs, and nonaspirin nonsteroidal anti-inflammatory drugs were examined using generalized estimating equations with an independent correlation matrix while controlling for age and sex.
Individuals with PD were significantly less likely to report ever smoking (odds ratio = 0.56; 95% confidence interval, 0.41-0.78). Additional measures of smoking revealed significant inverse associations with PD (P<.05) and trends in odds ratios (P<.005). Increasing intensity of coffee drinking was inversely associated with PD (test for trend P = .05). Increasing dosage (trend P = .009) and intensity (trend P = .01) of total caffeine consumption were also inversely associated, with high dosage presenting a significant inverse association for PD (odds ratio = 0.58; 95% confidence interval, 0.34-0.99). There were no significant associations between nonsteroidal anti-inflammatory drugs and PD.
Inverse associations of smoking and caffeine were corroborated using families with PD, thus emphasizing smoking and caffeine as important covariates to consider in genetic studies of PD.
Cigarette smoking, caffeine, and nonsteroidal anti-inflammatory drugs (NSAIDs) are reported protective factors for Parkinson disease (PD), but few family-based studies have examined these associations. Numerous studies have described associations for smoking and caffeine with meta-analysis indicating that smokers and caffeine consumers are significantly less likely to develop PD than those never exposed.1 Initial studies of NSAIDs have proposed that this exposure may also delay or prevent the onset of PD.2,3
Family-based case-control data sets are commonly used for genetic association studies, as this study design is robust to confounding by ethnic background (ie, population stratification). Cases and family controls are generally well matched on unmeasured genetic and environmental factors that may predispose to exposure behaviors and to disease. Examination of environmental associations in family-based samples thus reduces confounding by familial influences on exposure. We analyzed a family-based case-control data set for associations between PD and 3 factors inversely associated with PD in prior studies: smoking, caffeine, and NSAIDs.
Scott et al4 provided a detailed description of our study design to ascertain case subjects with prevalent PD and their family members in our previous examination of smoking and PD. Individuals were recruited through physician- and self-referrals by the Morris K. Udall PD Research Center of Excellence, Duke University Medical Center, Durham, NC, for genetic studies of PD. Environmental exposure data were collected to investigate effects and interactions with candidate genes. Individuals with PD and their siblings, spouses, and parents were recruited. In families with multiple individuals with PD, relatives connecting branches of the family were also recruited. Children were recruited if 1 or more were affected. Study protocols and consent forms were approved by the Duke University Medical Center institutional review board.
In clinical evaluations, individuals with PD had at least 2 cardinal signs of PD (resting tremor, rigidity, and bradykinesia), individuals with unclear clinical status had 1 sign and/or atypical features, and unaffected individuals had no signs of PD. Individuals with PD self-reported the age at which 1 of the cardinal signs began, which was used as the age at onset. Individuals with unclear status were not included in our analyses. The data set was stratified by self-reported race to minimize confounding by ethnicity. Only the white families (375/389 ascertained families) provided sufficient statistical power, so results are presented for this subset only.
Measures of environmental exposures
In a structured telephone questionnaire, participants reported their environmental risk factor history, including dosage (exposures per day) and years of initiating and quitting if applicable. Exposure histories of relatives not enrolled in our study were not known. Those who reported exposure on a weekly basis for 1 month or longer and initiation before the reference age were classified as ever exposed. Otherwise, participants were classified as never exposed. The reference age was defined as the age at onset for cases and as the age at examination (AAE) minus the mean disease duration among cases for controls. Ever-exposed participants were further classified by current or past exposure at the reference age. Dosage and duration (years) were determined, and intensity (pack-years, cup-years, and tablet-years) was calculated as the multiplicative result of dosage and duration. Measures of exposure were also truncated at 10 and 20 years before the reference age (that is, data between the reference age and 10 or 20 years before the reference age were not considered in the analyses) to examine whether the proximity of exposure to disease onset influenced the associations.
Population-averaged generalized estimating equations (GEE) were implemented using SAS statistical software version 8e (SAS Institute, Cary, NC) to model data from clustered individuals, which involved specification of a within-cluster correlation structure to derive a working correlation matrix. Regression parameters and variances were then derived for each model term while adjusting for correlations. Analysis with GEE using an independent correlation matrix has been shown to be a valid test for environmental association using simulated pedigrees similar to the current data set (D.B.H., unpublished data, July 2006). Even if this matrix does not accurately fit the data, GEE models are quite robust to misspecification of the correlation matrix.5 The GEE method is more powerful than our previously used method, conditional logistic regression, because conditional logistic regression restricts its analyses to within-sibling comparisons, whereas GEE incorporate all relatives regardless of structure and assess within-family and across-family differences (D.B.H., unpublished data, 2006).
The GEE models were constructed using affection status as the outcome and AAE and sex as confounders. Participants never exposed to the relevant factor served as the referent group. Associations between PD and each exposure—smoking, caffeinated coffee, total caffeine (coffee, tea, and soft drinks), total NSAIDs (aspirin, ibuprofen, and naproxen), and nonaspirin NSAIDs—were evaluated. Environmental measures were defined by history (ever and never), status (current, past, and never), dosage, duration, and intensity. For individual sources, continuous variables were categorized into tertiles. For caffeine, NSAIDs, and nonaspirin NSAIDs, individuals were categorized as high, moderate, low, and never as determined by their highest classification from relevant sources. Also, models were constructed using ordinal variables for status (0 = never, 1 = past, 2 = current) as well as dosage, duration, and intensity (0 = never, 1 = low, 2 = moderate, 3 = high) to perform tests for linear trend while controlling for AAE and sex. Because smoking, caffeine, and NSAIDs may confound the associations of each other, the GEE models were reconstructed using AAE, sex, and history of the 2 other exposures as confounders. Pearson correlation coefficients (r) between smoking, caffeine, and NSAIDs history were calculated to ensure that multicollinearity was avoided in these reconstructed models. Adjusted odds ratios (ORs) and 95% confidence intervals or P values were calculated for each model with 2-tailed testing. Results with P<.05 were considered statistically significant. A conservative Bonferroni correction based on 9 tests (3 independent tests [history, dosage, and duration] across 3 environmental factors) was applied to significant results.
Using the simulation of linkage and association program, data sets of 300 siblings with at least 1 affected individual (a simplification of our actual data set) were created.6 A binary risk factor with 10% within-sibling correlation was specified at a frequency of 0.2, 0.4, or 0.8. Data sets at varying relative risks were simulated to determine the minimum detectable relative risk for 80% power at each frequency. Statistical power of the GEE as implemented by SAS statistical software version 8e was based on OR estimates from 1000 replicate data sets. In detecting exposures with a frequency of 0.2 (eg, NSAIDs) at 80%, the minimum detectable relative risk was 0.38. The minimum detectable relative risk was 0.53 when assessing exposures with a frequency of 0.4 (eg, smoking) or exposures with a frequency of 0.8 (eg, caffeine).
Six-hundred seventy-three individuals from 375 families were analyzed. Cases (n = 356) included 338 probands, 8 siblings, 7 extended relatives, 2 children, and 1 parent. Controls (n = 317) included 289 siblings, 10 parents, 10 spouses or other unrelated controls, 6 extended relatives, and 2 children. Two hundred thirty-five case subjects (66.0%) were male, whereas 139 controls (43.9%) were male. The mean ± SD AAE was 66.1 ± 10.7 years in cases and 63.7 ± 12.3 years in controls. Cases had a mean ± SD age at onset of 58.1 ± 11.6 years and a mean disease duration of 8.0 years. Because individuals with PD were more likely than unaffected relatives to be male and older at examination, sex and AAE were considered important confounders in our analyses.
Exposure histories were not obtained for 8.0% of participants for smoking, 5.9% for caffeine, and 5.3% for NSAIDs. Data from individuals with a missing exposure history were excluded from analyses for the relevant factor. Reported exposure histories were correlated as follows: smoking and caffeine, r = 0.19; smoking and NSAIDs, r = 0.05; and caffeine and NSAIDs, r = 0.10. Given these modest correlations, models with multiple exposures were valid.
Results from models examining associations of smoking while controlling for AAE and sex are shown in Table 1. Individuals with PD were 0.56 times as likely to report ever smoking and 0.30 times as likely to report current smoking compared with unaffected relatives. Dosage, duration, and intensity presented ORs indicative of inverse relationships between PD and smoking with significance at most exposure levels. Dose-response associations were detected with increasing status, dosage, duration, and intensity (P<.005). Patterns remained significant after truncating exposure at 10 and 20 years before the reference age (Table 1) and adjusting for caffeine and NSAIDs (data not shown). Even after applying the conservative Bonferroni correction for multiple testing, most associations and trends in ORs for smoking remained significant.
Caffeinated coffee associations with PD relative to never consuming it were assessed while controlling for AAE and sex. When truncating exposure at the reference age, a significant dose response for intensity (trend P = .05) was observed. When truncating exposure at 10 years before the reference age, high dosage (>2.0 cups/d) was significantly inversely associated with PD (OR = 0.64; 95% confidence interval, 0.42-0.99), and increasing dosage showed a significant trend in ORs (P = .05). No significant associations were observed after truncating exposure at 20 years before the reference age or adjusting for smoking and NSAIDs (data not shown).
Overall caffeine associations with PD compared with never consuming it were also assessed while controlling for AAE and sex. These results are presented in Table 2. Case subjects were neither more nor less likely than controls to report ever consuming caffeine. However, a significant inverse association with PD at high dosage and significant inverse gradients for dosage and intensity were detected. The significant inverse gradient for dosage persisted after truncating exposure at 10 and 20 years before the reference age. The trends in ORs for dosage and intensity were nearly significant after adjusting for AAE, sex, smoking, and NSAIDs (trend P = .06 for both), but the significant trends shown in Table 2 did not withstand the conservative Bonferroni correction (trend P = .08 for dosage and trend P = .09 for intensity after Bonferroni correction).
Most users of NSAIDs reported a current exposure of relatively short duration, so assessment of status and exposure truncation before the reference age was not possible. No significant associations between NSAIDs and PD were detected (Table 3). There were also no significant associations between nonaspirin NSAIDs and PD (data not shown). Adjustment for smoking and caffeine did not change these results (data not shown).
Our findings support smoking and caffeine as being inversely associated with PD. The inverse association of smoking with PD was corroborated in twins and siblings (a subset of the current data set), but to our knowledge no other family-based studies have examined the association between PD and smoking.4,7 This work extends our previous findings by using a more powerful method in detecting inverse associations between PD and smoking as well as caffeine in an expanded data set (375 vs 140 families) with all of the sampled relatives (not just siblings) included in analyses.4 Meta-analysis suggests that the inverse association of PD with coffee drinking may be slightly weaker than the association with smoking.1 In our data, increasing dosage of caffeinated coffee was not significantly associated with a lower PD risk, whereas increasing intensity was marginally significantly associated with a decreasing risk of PD. These inverse trends became significant at P<.01 when considering total caffeine consumption. This result suggests that multiple caffeine sources should be considered when assessing associations with PD.
To our knowledge, our findings represent the first examination of the association between PD and caffeine in a case–family control setting, where cases are well matched to family controls on many unmeasured genetic and environmental factors. Given this matching and our controlling for confounding by sex, age, and race, our sample may not be representative of all PD cases and our results may not be generalizable at the population level. Nonetheless, persistence of significant inverse associations for smoking and caffeine in families with PD suggests that these associations are likely not confounded by familial influences on exposure. Further, residual confounding between the environmental factors themselves may lead to spurious associations, particularly with appreciable recall error, but inverse trends for caffeine and smoking remained significant (or nearly significant) after adjustment for one another. This indicates that these associations are not completely confounded by each other.
Recall bias may exist if the accuracy of reported exposures differed between cases and controls, but this bias generally leads to positive associations rather than inverse associations, as case subjects might be more aware of possible risk factors for disease. We also cannot eliminate the possibility of selection bias in our data owing to differential survival between prevalent and incident cases or differential mortality between exposed and unexposed individuals. Nonetheless, consistency of our results with previous studies, including prospective cohort studies not subject to recall and selection biases, suggests that our findings are not strongly influenced by such biases.1
No associations between NSAIDs and PD were observed in our moderate-sized sample, as ORs for all of the measures centered on the null value (OR = 1) with no significant trends. Two cohort studies2,3 provided initial evidence that NSAIDs may protect against PD. Subsequently, a large population-based case-control study found NSAIDs to be significantly associated with a lower PD risk but only in men,8 whereas 2 population-based case-control studies9,10 of moderate size failed to replicate the inverse association. Our family-based case-control sample had sufficient power to detect major effects, especially for more common exposures. However, for the less common NSAIDs exposure, 80% power was expected only for effects with an OR less than or equal to 0.39. Thus, small- to moderate-sized case-control samples may be underpowered to detect an NSAIDs effect.
Despite epidemiologic support for these environmental factors in neuroprotection, biological mechanisms of protection, if existent, have eluded investigators. As an alternative explanation for the noted inverse associations of smoking and caffeine, some investigators propose a preclinical aversion hypothesis in which the apparent protective effects of smoking and caffeine are merely reflective of the 7- to 10-year preclinical state of PD when smoking and consuming caffeine may be less rewarding.11,12 We detected significant associations for smoking and caffeine when truncating exposures at the reference age, at 10 years before the reference age, and at 20 years before the reference age, suggesting that the preclinical aversion hypothesis does not adequately explain these data.
Given the complexity of PD, these environmental factors likely do not exert their effects in isolation, thus highlighting the importance of gene-environment interactions in determining PD susceptibility. For instance, nitric oxide synthase genes are associated with PD but each potentially interacts with smoking, caffeine, and/or NSAIDs, thereby necessitating analysis of effect modification.13-16 Our evaluation of potentially inversely associated risk factors for PD implicates smoking and caffeine as important environmental exposures even in families with PD, where confounding by unmeasured influences on exposure is reduced. Smoking and caffeine possibly modify genetic effects in families with PD and should be considered as effect modifiers in candidate gene studies for PD.
Correspondence: William K. Scott, PhD, Miami Institute for Human Genomics, University of Miami, PO Box 019132 (M-860), Miami, FL 33101 (firstname.lastname@example.org).
Accepted for Publication: November 2, 2006.
Author Contributions: Ms Hancock and Dr W. K. Scott had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Martin, Vance, and W. K. Scott. Acquisition of data: Hancock, Stajich, Jewett, Stacy, B. L. Scott, Vance, and W. K. Scott. Analysis and interpretation of data: Hancock, Martin, and W. K. Scott. Drafting of the manuscript: Hancock. Critical revision of the manuscript for important intellectual content: Hancock, Martin, Stajich, Jewett, Stacy, B. L. Scott, Vance, and W. K. Scott. Statistical analysis: Martin and W. K. Scott. Obtained funding: Hancock, Martin, Vance, and W. K. Scott. Administrative, technical, and material support: Hancock, Stacy, Vance, and W. K. Scott. Study supervision: Martin, Vance, and W. K. Scott.
Financial Disclosure: None reported.
Funding/Support: This work was supported by grants NS39764 and NS056632 from the National Institute of Neurological Disorders and Stroke, National Institutes of Health.
Acknowledgment: We thank the patients with Parkinson disease and their families for their participation as well as the personnel at the Duke Center for Human Genetics for their contributions.
JJ A meta-analysis of coffee drinking, cigarette smoking, and the risk of Parkinson's disease. Ann Neurol
2002;52276- 284PubMedGoogle ScholarCrossref
et al. Nonsteroidal antiinflammatory drug use and the risk of Parkinson's disease. Ann Neurol
2005;58963- 967PubMedGoogle ScholarCrossref
et al. Nonsteroidal anti-inflammatory drugs and the risk of Parkinson disease. Arch Neurol
2003;601059- 1064PubMedGoogle ScholarCrossref
JM Family-based case-control study of cigarette smoking and Parkinson disease. Neurology
2005;64442- 447PubMedGoogle ScholarCrossref
ER Pedigree generation for analysis of genetic linkage and association. Pac Symp Biocomput
2004;993- 103PubMedGoogle Scholar
LA Nonsteroidal anti-inflammatory drugs and the incidence of Parkinson disease. Neurology
2006;661097- 1099PubMedGoogle ScholarCrossref
et al. Nonsteroidal anti-inflammatory drugs and risk of Parkinson's disease. Mov Disord
2006;21964- 969PubMedGoogle ScholarCrossref
WA Immunologic diseases, anti-inflammatory drugs, and Parkinson disease: a case-control study. Neurology
2006;67494- 496PubMedGoogle ScholarCrossref
R Parkinsonian signs in older people: prevalence and associations with smoking and coffee. Neurology
2003;6124- 28PubMedGoogle ScholarCrossref
DJ Measuring the rate of progression and estimating the preclinical period of Parkinson's disease with [18F]dopa PET. J Neurol Neurosurg Psychiatry
1998;64314- 319PubMedGoogle ScholarCrossref
N Neuroprotective effects of non-steroidal anti-inflammatory drugs by direct scavenging of nitric oxide radicals. J Neurochem
2001;761895- 1904PubMedGoogle ScholarCrossref
KFA Inhibitory effects of cigarette smoke on glial inducible nitric oxide synthase and lack of protective properties against oxidative neurotoxins in vitro. Neurotoxicology
2005;2649- 62PubMedGoogle ScholarCrossref
et al. NOS2A and the modulating effect of cigarette smoking in Parkinson's disease. Ann Neurol
2006;60366- 373PubMedGoogle ScholarCrossref