Rheumatoid arthritis drugs and the risk of Parkinson’s disease – a meta-analysis

Search strategies

To identify relevant observational studies investigating the potential link between the medications used for RA and the development of PD, we searched three medical databases: PubMed, Web of Science, and Scopus. We used combinations of the following terms in our search strategy: rheumatoid arthritis, rheumat*, antirheum*, anti-rheum*, DMARD*, and parkinson’s disease. By adding an * to the end of a term, the search engines pick out other forms of the word. We searched databases from inception until August 2023.

Inclusion and exclusion criteria

Inclusion Criteria

1. Study Design: prospective and retrospective observational studies, including cohort and case series.

2. Population: Adults without a prior PD diagnosis.

3. Exposure: exposure to any of the following drugs: NSAIDs, Glucocorticoids, and Disease-modifying anti-rheumatoid drugs (DMARDs).

4. Outcome: PD occurrence over the course of the follow-up period.

Exclusion Criteria

1. Studies not published in English.

2. Case reports, case series, and cross-sectional studies

3. Non-human studies.

4. Papers that do not provide any original information, such as reviews and meta-analyses.

5. Studies with inaccessible or insufficient data.

Data extraction

We extracted data on patients’ demographics and study characteristics including study design, number of PD cases, effect estimates, and variables adjusted. Two independent reviewers extracted data to a uniform Google sheet. A third independent reviewer further checked the retrieved data to confirm data accuracy. All disagreements were resolved through discussion.

Quality assessment

Two independent groups of authors assessed the quality of the retrieved 26 observational studies according to the Newcastle-Ottawa Scale (NOS). NOS uses a star system (with a maximum of nine stars) to evaluate a study in three domains: (a) selection of participants (4 stars); (b) comparability of study groups (2 stars); (c) the ascertainment of outcomes of interest (3 stars). A study was considered to be at low-risk bias if it received a score of nine stars; subsequently, a study was considered to be at medium and high-risk bias if it received a score of seven to eight and less than seven stars respectively [16].

Statistical analysis

The odds ratios (ORs) and relative risks (RRs) with 95 % confidence interval (CIs) were pooled in meta-analysis models. A RR value >1 indicates an increased risk of PD, and a RR value <1 indicates a decreased risk of PD. The p-value was used to evaluate whether the results were statistically significant. If the p-value was less than 0.05, then it was considered statistically significant.

In cases where multiple effect estimates, such as RRs, were provided due to different methodologies, we prioritized “ever use” over other categories like “past use” or “current use.” When faced with various RRs associated with different durations of drug exposure, we selected the RR corresponding to the longest exposure duration. Furthermore, we favored the effect estimate from the main analyses over sensitivity analyses and preferred the one associated with a lag period over the one that was not.

Dealing with missing data

When a cohort study [17] didn’t provide RR for a specific drug, we calculated RR and its 95 % CI from the provided numbers of exposed and non-exposed cohorts according to Altman [18] using MedCalc statistical software for Windows, version 19.4 (MedCalc software, Ostend, Belgium; https://www.medcalc.org). When another case-control study [19] plotted the RRs and 95 % CI for different drugs on a graph but didn’t report them as numbers, the data were extracted from the graph using the PlotDigitzer software (https://plotdigitizer.com).

Assessment of heterogeneity

We quantified the heterogeneity using the I² statistic, and determined its significance based on the accompanying p-value of the Cochran Q test. An I² value of 0 % indicates no observed heterogeneity, and I² values of 25 %, 50–75 %, and >75 % indicate low, moderate, and high levels of heterogeneity, respectively [20]. In case of significant heterogeneity (Chi-Square p<0.1), a random effects model of meta-analysis was used [21].

Publication bias

To assess publication bias across studies, funnel plots were generated using RevMan software, plotting the pooled effect estimate against its standard error (SE). The presence of publication bias was evaluated based on the symmetry of these funnel plots. Additionally, two statistical tests, namely Egger’s regression test and the Begg and Mazumdar rank correlation test (Kendall’s tau), were employed to further examine the evidence of publication bias. A low p-value (>0.05) suggests evidence against the null hypothesis, indicating the presence of publication bias.

Study selection

Following the screening of 3,255 titles from 3 databases, 448 studies were eliminated as duplicates, and another 2,718 were eliminated through the review of the titles and abstract as they did not seem to be related to the topic. A total of 89 studies were qualified for full-text screening, and 26 studies were eventually included in the analysis. Figure 1 illustrates the flow diagram of the study selection process and a detailed justification for the study’s ineligibility. For example, a Danish cohort study was excluded because it used osteoarthritis diagnosis as a surrogate for NSAID use [22].

Figure 1:

PRISMA flow diagram of the study selection process.

Description of the included studies

Twenty-six studies with a total number of 4,321,104 participants including 150,703 PD cases were included. Table 1 shows the characteristics of the included studies. Fifteen studies that were included in the final analysis were case-control studies [10], 19], [23], [24], [25], [26], [27], [28], [29], [30], [31], [32], [33], [34], [35], 11 were cohort studies [9], 17], [36], [37], [38], [39], [40], [41], [42], [43], [44]. Among the studies included, 11 studies were conducted in the USA [9], 17], 31], 34], [36], [37], [38], [39], [40, 42], 43], 3 in Canada [25], 26], 41], 3 in Taiwan [29], 33], 35], 2 in the Netherlands [23], 44], 2 in Finland [28], 30], 1 in Norway [24], 1 in Denmark [19], 1 in the United Kingdom [27], and 1 in South Korea [10].

Methodological quality of the included studies

After applying NOS to the 26 included studies, we categorized the studies into high, medium, or low quality, with scores of 9, 7–8, and <7, respectively. 2 studies were found to be of high quality, 11 studies were found to be of medium quality, and 13 studies were found to be of low quality. The NOS assessment and score of each study is detailed in Table 2.

Description of the included drugs

The included studies reported an association between the PD risk and many classes of drugs. These included NSAIDs (18 studies), acetaminophen (6 studies), corticosteroids (5 studies), and DMARDs (6 studies). Individual NSAIDs included aspirin (9 studies), ibuprofen (6 studies), diclofenac (2 studies), and naproxen (2 studies). Individual corticosteroids included prednisolone (2 studies), dexamethasone (2 studies), hydrocortisone (2 studies), prednisone (1 study), methylprednisolone (1 study). Individual DMARDs included methotrexate (4 studies), anti-tumor necrosis factor (anti-TNF) agents (2 studies), chloroquine/hydroxychloroquine (2 studies), sulfasalazine (2 studies), and other DMARDs (azathioprine, leflunomide, mycophenolate, mesalamine; 1 study). Table 3 shows the pooled RRs and 95 % CIs for each of these drugs.

NSAIDs and risk of PD

Main analysis

A total of 18 studies were included in the analysis to assess the overall effect of NSAIDs on the risk of PD (Figure 2). The RR with a 95 % CI was 0.96 (0.83–1.11), with a p-value of 0.57. This suggests that there is no statistically significant association between the use of NSAIDs and the risk of PD.

Figure 2:

Forest plot for the association between overall NSAID use and the risk of Parkinson’s disease.

Subgroup analysis by individual NSAIDs

We examined individual NSAIDs, including aspirin, ibuprofen, diclofenac, and naproxen, to explore their specific associations with the risk of PD (Figure 3).

Figure 3:

Forest plots for the association between individual NSAIDs use and the risk of Parkinson’s disease.

Aspirin: Nine studies were included in this analysis, and the pooled RR with a 95 % CI was 1.06 (0.91–1.25) with a p-value of 0.45. These results suggest that there is no statistically significant association between Aspirin use and the risk of PD.

Ibuprofen: A total of six studies were included, and the pooled RR was 1.00 (0.78–1.28) with a p-value of 1.00. This indicates no significant association between Ibuprofen and PD.

Diclofenac: The analysis of two studies yielded a pooled RR of 1.11 (0.82–1.49) with a p-value of 0.51. These findings do not suggest a significant association between Diclofenac and PD.

Naproxen: Two studies were included in this analysis, and the pooled RR was 0.96 (0.83–1.12) with a p-value of 0.64. This implies that there is no statistically significant association between Naproxen and PD.

Acetaminophen and risk of PD

A total of six studies were included in this analysis [23], 28], 30], 37], 38], 40]. The pooled RR with a 95 % CI was 1.09 (1.01–1.18), and the p-value was 0.02 (Figure 4). These results suggest that there is a statistically significant association between acetaminophen use and an increased risk of PD.

Figure 4:

Forest plot for the association between acetaminophen use and the risk of Parkinson’s disease.

Corticosteroids and risk of PD

We began by examining the overall effect of corticosteroids on the risk of PD using data from five studies [10], 19], 29], 32], 42]. The pooled RR with a 95 % CI was 0.80 (0.77–0.84), and the associated p-value was <0.00001 (Figure 5). These results indicate a statistically significant association between corticosteroid use and a reduced risk of PD.

Figure 5:

Forest plot for the association between corticosteroids use and the risk of Parkinson’s disease.

To provide a more comprehensive understanding, we conducted subgroup analyses for individual corticosteroids (Figure 6), including:

Figure 6:

Forest plots for the association between individual corticosteroids use and the risk of Parkinson’s disease.

Prednisolone: Data from two studies showed a pooled RR of 0.87 (0.24–3.21), with a p-value of 0.84. While the point estimate suggests a reduced risk, the wide confidence interval indicates uncertainty regarding the association.

Dexamethasone: Analysis of two studies resulted in a pooled RR of 0.69 (0.60–0.79), with a p-value of <0.00001. This suggests a statistically significant association between dexamethasone use and a reduced risk of PD.

Hydrocortisone: Our analysis of two studies yielded a pooled RR of 0.89 (0.69–1.15), with a p-value of 0.38. These results do not indicate a statistically significant association between hydrocortisone and PD.

DMARDs and risk of PD

We started by assessing the overall effect of DMARDs on the risk of PD using data from six studies [9], 10], 32], 33], 43]. The pooled RR with a 95 % CI was 0.92 (0.78–1.08), and the associated p-value was 0.29 (Figure 7). These results indicate that there is no statistically significant association between the use of DMARDs and the risk of PD. To provide a comprehensive analysis, we further explored the effects of specific DMARDs (Figure 8), including:

Figure 7:

Forest plot for the association between DMARDs use and the risk of Parkinson’s disease.

Figure 8:

Forest plot for the association between individual DMARDs use and the risk of Parkinson’s disease.

Publication bias assessment

The symmetry of the funnel plot (Figure 9) indicates the absence of publication bias in the studies reporting the association between NSAIDs use and the risk of PD. Moreover, Egger’s test and Kindalls Tau test p-values (0.274 and 0.410 respectively) confirm the absence of publication bias. We did not perform the publication bias assessment for the studies on acetaminophen, corticosteroids, and DMARDs because of the small number of studies on these drugs.

Figure 9:

Funnel plot for assessment of the publication bias for the association between NSAIDs use and risk of Parkinson’s disease.