Abstract
Abstract
Keywords
INTRODUCTION
Parkinson’s disease (PD) is the second most commonly occurring neurodegenerative condition after Alzheimer’s disease [1]. Global estimates of the age-adjusted incidence rate of PD range from 7.9 to 19 per 100,000 person-years and a prevalence of 57 to 230 per 100,000 population [2]. Various risk and protective factors of PD have been extensively investigated over several decades, however the etiology remains largely unclear [3]. There is a great interest in identifying at-risk individuals early, to potentially slow down or prevent neurodegeneration [4]. While genetic and familial environmental exposures are often cited to contribute to PD risk, lifestyle exposures such as smoking, coffee/tea and alcohol consumption have been the focus of research for several decades with varying and often conflicting results.
Interest in determining the association betweenalcohol consumption and risk of PD has to some extent come about from the inverse associations between smoking, coffee consumption and risk of PD. Risk reductions have been observed for caffeine consumers and smokers [5, 6], however causality has yet to be unequivocally established. Early studies that assessed alcohol consumption and risk of PD produced equivocal results [7–9]. However more recently, several meta-analyses have reported an inverse relationship between alcohol intake and PD risk [10–12]. Whether these factors are truly biologically protective is still a matter of debate, and interestingly this inverse relationship is not consistent across studies. The objective of this systematic review was to update earlier reviews [10, 11] using a more rigorous approach and to explore potential sources of variability by critically reviewing the key methodological features of eligible studies, such as study design, control of confounding and measurement of exposure and outcome.
METHODS
Search strategy
The databases PubMed, TRIP [13] and Web of Science Core Collection were searched systematically for relevant literature. The two main constituents of the research question; ‘Parkinson’s disease’ and ‘alcohol’, were used to develop database search terms. These two terms were entered into Roget’s Thesaurus online, to identify as many synonyms as possible [14]. The final search terms used were: (alcohol *
The inclusion and exclusion criteria were developed to screen the search results systematically and ensure that only relevant articles were included in the final review. Primary research articles were included if they were published in English language, in peer reviewed journals between 2000 and May 2014. To be included, studies also had to include a comparison or control group consisting of individuals without PD, report a measure of association between quantity and frequency of alcohol intake and PD risk, and adjust at least for the potential confounding factors of smoking and age. Studies which reported an association between alcoholism and PD risk were also selected for inclusion. Research which measured alcohol exposure as a binary variable only (drinker versus non-drinker), or of cross-sectional design measuring alcohol intake and PD prevalence at the same point in time, wereexcluded.
RESULTS
Sixteen articles (reporting on seventeen studies) were identified (Fig. 1), seven of which utilised a case-control study design to investigate the association between alcohol consumption and PD risk. Four articles used a nested case-control design and the remaining five articles (reporting on six studies) utilised a cohort study design. Studies were mainly conducted in the USA (seven), with two from the United Kingdom and one each from India, Italy, Finland, Japan, Serbia, Singapore and Sweden. The characteristics of the included studies are shown in Tables 1 and 2.
Case ascertainment
Studies used a variety of methods for subject and case ascertainment, however most used standardized and validated scales. Reference to the criteria is found in Tables 1 and 2. The majority of studies indicated that PD diagnosis was accepted if based on a neurologist diagnosis of several cardinal signs corroborated with clinic notes and screening instrument for cognitive impairment such as the Minimal Mental State Examination (MMSE). Research teams indicated reviewing the medical records. PD evaluation scales included one or a combination of the following: the Unified Parkinson’s Disease Rating Scale (UPDRS), UK Parkinson’s Disease Society Brain Bank Clinic Diagnostic Criteria, Queen Square Brain Bank Criteria, International Parkinson and Movement Disorder Society rating scale. Only three studies mentioned the response to anti-parkinsonian drugs eg. L-dopa. Other combinations included International Classification of Diseases codes, DNA collection, questionnaires and hospital discharge registers and death certificates. Assessment grading using the Hoehn and Yahr scale was not cited, however a number of studies screened for cognitive impairment using the Minimal Mental State Examination.
Case-control studies
In the case-control studies (Table 1), alcohol intake was summarised in a number of different ways. Studies reported quantity and frequency in drinks, units, decilitres or grams per day/week, duration of drinking in years and participants positive for an alcohol use disorder (AUD). An AUD is a harmful pattern of drinking with serious risks of physical and psychological harm. A consequence of harmful drinking is alcohol dependence, characterised by a strong desire to drink and difficulties controlling drinking according to the Diagnostic and Statistical Manual of Mental Disorders (DSM IV) [15]. Among the studies which measured quantity and frequency, the reference period over which consumption was measured also varied. Studies asked participants to recall typical consumption patterns across their lifetime [16] average alcohol intake over the previous month [17] and drinking habits during the period when their alcohol consumption was the highest [18].
The results were inconsistent in case-control studies that compared different quantities of alcohol intake in drinkers with non-drinkers: two studies showed a moderately decreased risk of PD with greater quantity consumed [17, 19], two showed no evidence of an association [16, 17], and one showed a strong increased risk of PD with higher consumption [20].
Results were also inconsistent for studies in terms of duration of drinking. A study conducted in Italy showed evidence of a decreased risk of PD with ≥46 years of wine drinking (OR 0.45; 95% CI [0.29, 0.68]) [19]. Yet in India, consuming alcohol for >20 years was not associated with PD risk (OR 1.48; 95% CI [0.82, 2.65]), and consuming alcohol for ≤20 years was nearly protective (OR 0.45; 95% CI [0.20, 1.01]) [21]. However, only 48 cases and 59 controls consumed alcohol at all so this finding may be due to a lack of power in this study [21].
Some of the studies also showed a dose-response relationship with level of alcohol exposure (quantity and/or frequency) and risk of PD, albeit in opposing directions [17, 20]. Nicoletti et al. [19], found reduced risks with greater quantity of wine consumed per day, and for units consumed per week. Evans et al. [17], calculated an OR of 0.44; 95% CI [0.26, 0.75] for each category of intake, while Sipetic et al. [20], reported an increased risk for average weekly consumption in decilitres (OR 4.68; 95% CI [2.79, 7.84]).
Prospective studies
Most studies reported alcohol intake in grams per day, [22–25], however drinks per day [26, 27], or month [28], and units per week [29], were also used (Table 2).
In the prospective studies there was no convincing evidence of a decreased likelihood of PD according to different levels of total alcohol consumption [22–30], Most studies observed non-significant associations between alcohol and PD risk. A study conducted in Finland found those who consumed <5 grams of alcohol per day had an increased risk of PD compared to non-drinkers (RR 1.94; 95% CI [1.09, 3.47]) [24]. Another study from the USA found increased risks of PD among men who consumed 10 to 19.9 grams/day (RR 1.48; 95% CI [1.09, 2.01]) and women who consumed 10 to 14.9 grams/day (RR 1.67; 95% CI [1.06, 2.64]) compared to non-drinkers, however there was no clear trend of increasing risk with increasing consumption [23].
The largest study found no association between total alcohol intake and future PD risk, however beverage specific analyses revealed heavy liquor drinking (at least two drinks/day) to be associated with an increased risk (RR 1.35; 95% CI [1.02, 1.80]), and low to moderate beer drinking (less than one drink/day) with a deceased risk (RR 0.79; 95% CI [0.68, 0.92]) [26].
There was some evidence to suggest that smoking modified the association between alcohol consumption and PD risk, with greater risk reductions afforded to ‘never smokers’. Four studies performed stratified analyses to investigate whether the risk of PD due to alcohol consumption, differed between ‘ever smokers’ and ‘never smokers’ [22, 26]. All four observed non-significant associations between total alcohol intake and risk of PD. However, three of these studies [22, 26], found that ‘never smokers’ who consumed alcohol had a greater risk reduction of PD than ‘ever smokers’. Wirdefeldt et al. [25], found a protective effect of alcohol (OR 0.56; 95% CI [0.39, 0.80]) for ‘ever’ versus ‘never’ drinkers when restricting the analysis to ‘never smokers’. Hernán et al. [22], found a marginally reduced risk among male ‘never smokers’ who drank ≥15 grams/day (RR 0.5; 95% CI [0.3, 1.0]), but not for women. Liu et al. [26], observed an inverse association between beer drinking and PD risk among ‘never smokers’, however no such association was found for liquor drinking. Conversely, Palacios et al. did not find any significant differences when conducting stratified analyses between ‘never’ and ‘ever’ smokers [23]. Further investigations would be required to confirm these findings.
DISCUSSION
Quality of studies
This review determined several possible methodological weaknesses that could explain the varying and often conflicting results of studies reporting lifestyle exposures such as smoking, coffee/tea and alcohol consumption contributing to PD risk.
Among the case-control studies there was a potential for selection bias due to selection or self-selection of controls [17, 20]. One study found an increased risk of PD due to alcohol consumption, however they used hospital controls receiving treatment for chronic pancreatitis, a condition strongly associated with alcohol consumption [20]. Bias may have been introduced if these controls consumed alcohol at a different rate to the general population.
Few case-control studies used incident cases [16, 20], and this may have increased the potential for inaccurate recollection of exposure. The temporal sequence of events was particularly unclear for one case-control study, where prevalent cases were questioned about their current alcohol intake over the past month [17]. This study found a dose-dependent inverse association between alcohol intake and PD risk. Whereas a prospective cohort study reported PD cases significantly reduced their alcohol intake around the time of their diagnosis and continued to reduce their alcohol intake thereafter [23]. Thus findings from case-control studies that indicate protective effects of alcohol should be interpreted cautiously not only for their limitations and potential for selection bias but also retrospective assessment of alcohol intake, since the recall of past dietary exposures can be affected by current exposures [31].
In further support of this argument, the studies with the most certain temporal sequence of events i.e. those which used cohort and nested case-control study designs [22, 30], tended to find non-significant associations, close to unity, between total alcohol consumption and risk of PD. One nested case-control study found lower risks when alcohol consumption was reported by prevalent cases, compared to incident cases who were identified five years after alcohol exposure information was collected [28].
Two nested case-control studies may have inadvertently included prevalent cases in their baseline sample, which could potentially explain the non-significant inverse associations they observed [25, 27]. One of these studies seemingly made no attempt to exclude the 26 prevalent cases from their analyses [27].
There was no discernible pattern in the results according to the length of follow-up in the prospective studies which varied between 10 years to 20 years (Table 2). For example a nested case-control study followed participants between 1995 and 2001 [29], yet found very similar results to studies that had much longer follow-up periods [22, 26]. Three studies reduced the potential for reverse causality by excluding the first 5 to 10 years of follow-up [23, 26]. Two of these studies observed non-significant associations between total alcohol intake and PD [23, 26], and one study found increasing PD risk with increasing duration between exposure and outcome [24].
The cohort studies were more likely to use validated questionnaires to elicit alcohol exposure information compared to the case-control and nested case-control studies. All cohort studies asked participants about their consumption of specific beverages. These questions have been shown to produce higher estimates and improved recall of alcohol consumption compared to questions on overall consumption alone [32]. For these reasons, the cohort studies may have been more likely to produce valid results.
Overall, most studies used non-drinkers as the reference category in their analyses, which helped when comparing the results, however the definition of a ‘non-drinker’ tended to differ between studies. Apart from the study that measured ‘peak’ alcohol consumption, it was not clear whether any of the studies that measured quantity and frequency, separated lifetime abstainers from former drinkers. This can lead to misclassification if former heavy drinkers are classified as ‘non-drinkers’ [33].
Alcohol use disorders
Two studies Brighina et al. and Hernán et al. classified alcohol exposure according to clinical criteria for drinking that is harmful to health [34, 29]. Brighina et al. a case-control study, screened subjects for an alcohol use disorder with the CAGE questionnaire and also with medical records, using an alcohol-related medical problem as a proxy for alcoholism [34]. The CAGE questionnaire consists of four questions and is used as a screening tool for alcohol abuse and dependence [35]. Hernán et al. a prospective study, classified subjects as clinically-defined alcoholics, before their index date, based on a computerized diagnosis of alcoholism or alcohol-related chronic disease such as alcohol cirrhosis or alcoholic cardiopathy [29].
Brighina et al. found a statistically significant inverse association between subjects with a CAGE score of 2 or more (indicative of an alcohol use disorder) compared with participants with CAGE score <2 and PD risk: adjusted OR 0.63, 95% CI [0.43, 0.93] [34]. The CAGE questionnaire positively identified 7.2% of cases and 11% of controls (based on 779 case-control pairs) for alcohol use disorders reporting a weak but statistically significant trend of decreasing risk of PD with increasing CAGE score [34]. When the analysis was based on participants with a medical history of an alcohol-related health problem (based on 843 case-control pairs) the inverse association was weaker. The report by Hernán et al., indicated that alcoholism was not associated with the risk of PD, however, the smoking-adjusted OR of 1.09, 95% CI [0.67, 1.78] was based on eighteen cases and 174 controls (1.72% ) receiving a diagnosis of alcoholism before the index date [29]. This analysis was based on fewer subjects, and may have lacked power to detect a significant difference. Thus, the methods used to measure alcohol use and the statistical power of the analyses may explain the contrasting results of the studies.
Beverage types
Eight studies measured associations between PD risk and the beverage types; beer, wine or liquor, however no specific beverage type was consistently, significantly associated with PD risk. Of these eight studies, three found statistically significant inverse associations. Nicoletti et al. observed a moderate, dose-dependent protective effect of wine drinking, with an OR of 0.45; 95% CI [0.28, 0.74] for ≥3 glasses per day [19]. Hernán et al. and Liu et al. each found a weak, inverse association for beer drinking, however there was little evidence of a dose-dependent trend [22, 26].
Three studies noted that alcohol intake and PD vary according to different kinds of alcohol suggesting increased risks of PD with consumption of specific beverages. A case-control study by Fukushima et al. assessed daily intake separately for each type of alcohol including beer, Japanese sake, Shochu, wine and whisky. There were no significant associations with PD, except moderate to strong, dose-dependent effects of Japanese sake [18]. Additionally, Liu et al. reported an increased PD risk with ≥2 drinks of liquor per day (OR 1.35; 95% CI [1.02, 1.80]) [26]. The remaining study by Sipetic et al. indicated strong associations for brandy and beer, and very strong associations for liquor and wine [20]. However, the findings of this latter study should be interpreted with caution since the authors did not statistically control for any confounding variables when calculating beverage specific associations.
In contrast, Palacios et al. and Brighina et al. two studies which investigated the relationships between beer, wine and liquor drinking and risk of PD, found non-significant results with associations close to unity [23, 34], It would therefore appear that the studies collectively paint an inconsistent picture of the relationship between PD and different beverage types. Variance in typical beverage specific consumption patterns among the studied populations could potentially explain these inconsistencies. Cultural preferences could result in fewer participants selecting certain beverages, and thus lack of significance could reflect a lack of power. However, four studies conducted in the United States of America, observed different results for the same beverage [22, 34].
In studies that evaluated three different alcoholic beverage types, there was no evidence of simultaneously reduced or increased risks for all three beverages beer, wine and liquor [22, 34]. This could indicate that ingredients other than ethanol may be driving the beverage specific associations observed.
Confounding by personality traits
Ten studies analysed the effects of alcohol consumption alongside smoking and caffeine intake as risk factors for PD. Two of these studies found statistically significant inverse associations for all three factors [17, 19]. One of these studies also found a dose-dependent trend for the presence of at least one, two or three of smoking, coffee or wine drinking behaviours, with the greatest risk reduction for all three [19]. The other study suggested that personality traits (such as risky, sensation seeking behaviours) confounded the associations between each of the three factors and PD risk [17]. These findings suggest that a confounding variable common to all three factors may explain or partially explain their hypothetical protective effects. However, against this argument is that the remaining eight studies either did not observe significant associations for all three factors or observed significant effects but in different directions. The majority of studies did not find simultaneously protective effects for smoking, coffee drinking and alcohol consumption, suggesting that the presence of a confounding variable common to all three factors, such as an addiction-avoiding personality trait, is unlikely.
Ethnicity
Most of the selected studies examined PD risk in western countries with predominantly white populations. Two studies recruited participants of East Asian descent. One case-control study recruited participants in Japan [18], and a prospective study recruited Singaporean Chinese participants [30]. Both studies found non-significant associations between total alcohol consumption and PD risk. Fukushima et al. calculated an association close to unity (OR 0.96; 95% CI [0.50, 1.81]) for those who drank alcohol at least six days per week compared to non-drinkers, and Tan et al. found a non-significant decreased risk (RR 0.6, 95% CI [0.31, 1.16]) for at-least-weekly drinkers compared to non- or less-than-weekly drinkers [18, 30]. Fukushima et al collected data on ‘peak’ alcohol drinking rather than average consumption, and statistically controlled for a number of extra potential confounders compared to Tan et al such as body mass index, medication history, cholesterol, vitamin E, vitamin B6, iron and dietary glycemic index, which could potentially explain the differences in results [18, 30].
Summary
This study has provided a critical summary of the epidemiologic literature related to alcohol consumption and PD risk published since 2000. In general, studies with a prospective design tended to find non-significant associations between total alcohol intake and PD risk, with two studies finding an increased risk with moderate alcohol consumption. The case-control studies were more likely to find protective effects of alcohol on PD risk, however it is unclear to what extent these observations were the result of selection and recall bias.
Among the reviewed studies there was some evidence to suggest that smoking modifies the association between alcohol intake and PD risk. There was also evidence to suggest that beer may be inversely associated with PD risk, however ethanol is unlikely to be the protective ingredient. The evidence does not support the theory that a confounder, such as an addiction-avoiding personality trait, has produced the inverse associations between smoking, coffee and alcohol intake and PD risk. Studies examining the relationship between alcohol use disorders and PD have produced inconsistent results.
We note the lack of uniformity by these primary studies in the definition of drinking, the inconsistency of the units used to report quantity and frequency, and the reference period over which alcohol consumption was measured. We fully appreciate that to obtain valid estimates of alcohol consumption, researchers must utilise methods appropriate for the drinking culture [36]. However, these methodological differences make it difficult to compare results across countries with accuracy, and may explain some of the heterogeneity observed.
Standard measures such as units and drink portions can differ between countries, therefore converting these measures into grams of ethanol for analysis would be beneficial for comparison. Additionally, since the risk of PD may differ by beverage type, researchers should aim to measure beverage specific risks wherever possible. Not only will this help to improve understanding of beverage specific risks but recall may be improved compared to combined alcohol questioning. Whilst volume and duration of alcohol consumption are important to determine cause-effect associations, frequency of heavy drinking episodes may be another important consideration [37], which future research may wish to address.
The subject selection process was determined differently, and did not follow a uniform method across the studies. Assessment grading using the more current Hoehn and Yahr scale was not cited though MMSE was used. This leaves the reader to speculate that PD patients were selected on a scale where cognitive decline did not interfere with the ability to recall events, self-report exposure or complete questionnaires especially in case–control studies, a process that is prone to recall bias and could lead to false-positive associations. Alternatively, PD patients may underreport exposure leading to false-negative associations.
The strengths of our review include the large number of subjects investigated and the comprehensive picture provided. The decision to exclude studies which did not control for the potential confounding effects of smoking was important. Palacios et al. [23], found smoking to be the biggest confounder of the relationship between alcohol and PD risk, and smoking reduced the risk of PD in the majority of the reviewed studies. The conclusions of this study are therefore more likely to reflect the true relationship between alcohol and PD.
A limitation to this review was the introduction of bias. Non-published data and papers published in languages other than English were not included. Studies which measured alcohol exposure as a binary variable only were also excluded, however measuring exposure as ‘ever’ or ‘never’ drinkers would have been inadequate to determine the complexity of alcohol exposure. Despite identifying cohort studies, the review primarily involved case-control studies and the results should be interpreted cautiously due to the recall bias of case-control studies.
Conclusion and implications and directions of future research
This study highlights the need for more prospective studies investigating the relationship between alcohol and PD of adequate sample size. Improvements to reporting of studies by investigators particularly with respect to sample size and power would help others interpret the epidemiological significance of any findings. We note the lack of uniformity in reporting quantity, duration and frequency of alcohol intake. Researchers may wish to consider the comparability of their results when reporting, and the need to be explicit when presenting the data. Finally, as with any observational study, the possibility of residual confounding cannot be ruled out. Although authors controlled for a range of factors, it remains possible that other factors confound the association between alcohol and PD risk.
Interpretation of findings from case-control studies should take into account selection and recall bias. The association between ingredients in beer and PD risk, and effect modification by smoking, may be avenues for future research. The evidence is not strong enough to recommend prioritising beer over other beverages to reduce the risk of PD. From researchers’ responses, most of the studies proved to be preliminary and improving statistical power to detect joint effects was encouraged.
FINANCIAL DISCLOSURE/CONFLICTS OF INTEREST
The authors have no conflict of interest to report.
Footnotes
ACKNOWLEDGMENTS
We wish to thank Dr. Sue Pearson and Dr. Cecile McKeown for their valuable comments on this manuscript.