**2. Methods**

#### *2.1. Literature Search Strategy*

All relevant articles published in English between 1 January 1990, and 31 December 2019 were identified by searching PubMed, BioMed Central, Medline, and Google Scholar. Details regarding search terms are provided in supplementary data. Moreover, the reference lists of relevant articles were reviewed to identify eligible studies not derived using these search terms.

#### *2.2. Inclusion and Exclusion of the Literature*

Inclusion criteria were as follows: (1) clear definition of PD diagnosis; (2) clear definition regarding the quantity of caffeine, coffee, or tea consumption; (3) cohort study published as an original article, case series, or letter to the editor; (4) sample size of ≥50 individuals; and (5) published in English. After excluding nonqualified studies, 19 studies were entered the full-article assessment process and another 6 studies were excluded due to the lack of hazard ratio. Finally, 13 studies were included into qualitative synthesis. We further segregated the remaining 13 studies into two categories: the healthy cohort including studies (*n* = 9) that recruited individuals without previous diagnosis of PD, wherein PD diagnosis was performed during follow-up evaluation, and the PD cohort including studies (*n* = 4) on individuals with PD already presenting motor symptoms, wherein PD progression was monitored. The selection process is illustrated in Figure 1.

#### *2.3. Data Extraction*

The following data were extracted: name of the first author; year of publication; country and location; study design; the original cohort or clinical trial; the starting time of cohort; diagnostic criteria for PD; the assessment of caffeine consumption; the amount of coffee or caffeine consumption; mean follow-up period of time; the outcome assessment time; and the outcome of the PD progression. All data were independently reviewed by three investigators (BAI CH, Hong CT, and Chan L), and conflicts were resolved through a consensus. Assessing of quality of all studies were done by three investigators (BAI CH, Hong CT, and Chan L) based on the Newcastle–Ottawa Scale. The study was recommended (>7) by at least 2 investigators into this study as candidate. Data from these 13 candidate studies were independently extracted by two investigators (BAI CH and FAN YC).

**Figure 1.** A schematic representation of the literature search.

#### *2.4. Statistical Analysis*

The hazard ratio (HR) was determined, and 95% CIs were calculated on the basis of a binomial assumption. I<sup>2</sup> was used to assess heterogeneity across studies. All statistical analyses were performed using SAS software (version 9.3; Statistical Analysis System, SAS.com, USA). All reported probability (*p*) values were two-sided, with *p* < 0.05 considered statistically significant.

#### *2.5. Data Availability*

The present study was a meta-analysis and all the studies enrolled into analysis can be found through the provided searching strategy.

#### **3. Results**

Among the nine studies included in the healthy cohort (Table 1) [15–18,21–25], five were conducted in the United States, three in Scandinavia, and one in Singapore. Some studies were large-scale, long-term, population-based epidemiological cohort studies, and the others were specific for individuals with certain characteristics (nurses, healthcare professions, twins, and ancestry of migrants). Caffeine consumption was evaluated using questionnaires, either detailed and comprehensive or simple ones. Five of them investigated overall dietary habits, including coffee, tea, cola, and chocolate consumption, by using the transforming formula. The rest of them only recorded the daily consumption of coffee or tea. PD was diagnosed through either self-report and confirmation of medical records or from the national health care database. Two studies separately reported the results for men and women, and another study reported data only for women.

Most of the included studies categorized caffeine consumption as degree 4–5 based on the amount of caffeine or the number of cups of coffee per day. Only one study simply provided options of "yes" and "no" with regard to regular coffee consumption. Considering the difficulty in transforming the actual caffeine consumption among studies, this study considered results of all individuals consuming coffee at all degrees and considered the no-exposure group as a reference group to determine the HR. Overall, 43 results extracted from nine studies were analyzed herein. Caffeine consumption was significantly associated with a lower risk of developing remarkable symptoms for the diagnosis of PD during the follow-up period of time (HR = 0.797, 95% CI: 0.748–0.849, *p* < 0.001; Figure 2).

This study analyzed the effect of caffeine on patients with PD (Table 2). [26–29] Among the four studies in the PD cohort, three were conducted in European countries and one in the United States. Patients with PD were in the early stage of the disease. Similar to the healthy cohort, levels of caffeine consumption were assessed through either comprehensive questionnaires or simple questions. The four studies set different parameters for PD progression, including the initiation of levodopa, levodopa-induced motor complications, and the transition to Hoehn and Yahr stage III. The average follow-up duration ranged 4 to 10.3 years. Finally, 10 results were extracted from these 4 studies. Caffeine composition among patients at an early stage of PD significantly decelerated PD progression (HR = 0.834, 95% CI = 0.707–0.984, *p* = 0.03; Figure 3).


**Table 1.** List of the included cohort study.

**Figure 2.** Forest plot illustrating the hazard ratio (HR) of Parkinson's

 disease (PD) among healthy individuals

 from cohort studies.


**Figure 3.** Forest plot illustrating the hazard ratio (HR) of progression of Parkinson's disease (PD) among individuals with early-stage

 PD.

#### **4. Discussion**

The results of this study showed that among both healthy individuals and patients with PD, caffeine consumption was significantly associated with a lower HR for the risk or progression of PD, respectively. Considering that steady neurodegeneration in PD precedes the onset of motor symptoms for decades and persists thereafter [30], caffeine was speculated to have disease-modifying potential throughout the course of the disease in this study. Compared with data obtained from case–control studies, data obtained from a combination of multiple cohorts were more likely to demonstrate the beneficial causal relationship between caffeine consumption and the risk of PD.

The potential neuroprotective effect of caffeine consumption against PD was noted on the basis of case–control epidemiological studies. Although another component in coffee, that is, eicosanoyl-5-hydroxytryptamide, is believed to protect against neurodegeneration [31], a similar association found for tea consumption further corroborated the finding that caffeine is the key protective agent in coffee [32]. Instead of psychostimulation, caffeine antagonizes the adenosine A2A receptor. In the central nervous system, the adenosine A2A receptor is exclusively expressed in dopaminergic neurons, and the activation of the adenosine A2A receptor triggers the cAMP-protein kinase A-dependent elevation of intracellular calcium and release of glutamate [33]. Excessive intracellular calcium and glutamate levels are responsible for excitotoxicity in neurodegenerative diseases, including PD [34]. The adenosine A2A receptor is also involved in neuroinflammation-mediated neuronal dysfunction and degeneration [35]. Istradefylline, an FDA-approved adenosine A2A receptor antagonist currently used for treating PD, reduces off time and improves the motor symptoms of patients with PD albeit with complications including the exacerbation of dyskinesia [36]. The neuroprotective effect of istradefylline has been further described in in vivo studies [37–39] but not in clinical.

However, large-scale cohort studies focused on the healthy population, did not consistently demonstrate the risk-reduction effect of caffeine on PD. Studies investigating the effect of caffeine on PD progression among PD patients were also not fruitful. The quantification of daily caffeine consumption is most challenging for studies intending to investigate disease-modifying effects. Except caffeine tablets, the assessment of the daily intake of caffeinated beverages and food products requires a formula for transformation. A structured dietary interview is usually necessary to obtain semi-quantitative data regarding daily caffeine intake. Considering that adults usually adhere to their dietary preferences, the interview would yield reliable data regarding long-term levels of caffeine consumption. However, coffee, tea, cola, and chocolate in different styles, brands, or countries (even areas) have different caffeine contents. Moreover, genetic polymorphism, sex, and heterogeneity in caffeine metabolism also influence the effects of caffeine [20,40,41]. Herein, most studies segregated their participants on the basis of caffeine consumption by relative gradients, thus introducing a slight variation among the enrolled studies. This relative but not absolute grouping deters inter-study comparisons and the obtainment of consistent findings.

Defining PD progression is challenging and problematic. Among healthy individuals, tremors may be visible and can be recognized early. However, the remaining cardinal motor symptoms, including rigidity, bradykinesia, and postural instability, are ambiguous. Most individuals with PD were either misdiagnosed or underwent unnecessary treatment years before reaching a final diagnosis [42]. This delay deters the assessment of disease progression for the cohort study recruiting healthy participants. Regarding disease progression among patients with PD, off-status motor function is the major parameter among clinical trials. However, responses to one-night washout are variable, and the effect of levodopa may last for 2 weeks [43]. Moreover, if the intervention itself causes certain symptomatic effects in conjunction, similar to rasagiline or caffeine, it would be challenging to distinguish between disease modification and symptomatic effect [44]. However, the onset of motor fluctuation and dyskinesia have been considered markers of disease progression among patients with PD. Nevertheless, the degeneration of dopaminergic neurons is not the only underlying factor [45]. The dosage of levodopa, the prescription of dopamine agonists, or amantadine in the early stage of the disease also influences the onset of motor complications [46–48]. One study included herein

considered the initiation of levodopa treatment as a marker for disease progression, which was highly influenced by the subjective and objective conditions of patients with PD [49]. Young or old; employed, self-employed, or retired individuals and the self-expectation had an influence on levodopa initiation. These aforementioned issues deter the accurate assessment of disease progression, thus yielding inconsistent disease-modifying effects of caffeine or any other interventions.

The strength of this study is the delineation of the disease-modifying effect of caffeine on PD. The inclusion of exclusive cohort studies was superior to case–control studies owing to the potential temporal association between caffeine and PD, and the prevention of recall bias on the dietary habit. Furthermore, this study pooled all the HR of PD from moderate-to-high levels of caffeine consumption together and determined the lower limit as reference, thus eliminating the ambiguous cut-off level of daily caffeine intake in numerous studies. This blurred "beneficial dosage of caffeine" varied among studies and confounded clinicians and the population. Moreover, no remarkable J-shaped curve was previously obtained for the risk of PD and caffeine consumption, thus yielding an upper limit of permissible caffeine consumption. The present results indicate that caffeine consumption potentially alters the PD risk and progression among both healthy individuals and those with PD, and this concept is easier to pick up by general population and health professions.

This study has some limitations. First, considering the diagnosis of PD heavily relying on the development of the motor symptoms, which may be delayed for decades after the beginning of the neurodegeneration, the utilization of the diagnosis of PD as outcome assessment in the healthy cohorts may bias by either under or overvaluation. Second, variations in the levels of caffeine consumption among studies undoubtedly introduced heterogeneity among studies. Certain studies focused on caffeinated food products, and another study focused only on coffee or tea. The effect of some promising ingredients in coffee, such as eicosanoyl-5-hydroxytryptamide [31] or methylxanthine [50] had not been investigated in the present study due to the lack of standard assessment, such as the dietary questionnaires for the amount of daily intake. None of the studies included in this research focused on the effect of pure caffeine tablets, which would directly demonstrate the effects of caffeine rather than the mixed effects of caffeinated food products. Based on this information provided from the dietary questionnaires, it was not possible to define the optimal daily dosage of caffeine and the food source of caffeine. Third, instead of coffee or caffeine, some factors are known to affect the risk of PD, such as diabetes, pesticide exposure and the well-water drinking [51]. Female is well-known for the lower risk of PD, and the protective role of sex hormone is speculated [52]. However, in the present study, two cohorts separated the results between men and women, which revealed no significant heterogenicity. Meanwhile, one study also sub-grouped female participants based on the hormone replacement therapy, and there was no remarkable heterogenicity either. Moreover, several genetic and environmental factors interact with caffeine, and life style, socioeconomic status, exercise, high-fat diet and alcohol consumption may also be associated with the habitual coffee drinking. There was no clear information about those environmental factors from the included studies for the authors to adjust those possible confounding factors. Lastly, genetic polymorphism may affect the metabolism of caffeine, but to the best of our knowledge, only case-control studies were found to investigate the gene–caffeine interaction in PD [53–55], which did not fit into the inclusion criteria of the present meta-analysis.

In conclusion, this meta-analysis shows that caffeine is associated with a low risk of developing PD in healthy individuals and the deceleration in the progression of motor symptoms in patients with PD. Additional studies are required to investigate not only the optimal daily dosage and food source of caffeine for PD, but also the possible mechanisms underlying the bioprotective effects of caffeine on PD. Among individuals with PD, caffeine intake should be encouraged if adverse effects are tolerable.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2072-6643/12/6/1860/s1.

**Author Contributions:** C.T.H.: study design, article review, quality assessment, draft written. L.C.: study design, article review, quality assessment, manuscript revise. C.-H.B.: study design, article review, quality assessment, data extraction, statistical analysis, manuscript revise. Financial Disclosures of all authors: none to disclose. All authors have read and agreed to the published version of the manuscript.

**Funding:** The author(s) received no financial support for the research, authorship, and/or publication of this article.

**Acknowledgments:** We appreciated YC Fan for the help in data extraction and Shennie Tan and Yanie Tan for the proofreading and English editing.

**Conflicts of Interest:** The authors declare no conflict of interest.
