Risk of Suicidal Ideation and Behavior in Individuals With Parkinson Disease

This study addresses inconsistent results from studies of suicidal ideation and behavior in patients with Parkinson PD.

Key Points

Question

What is the prevalence and risk of suicidal ideation and behavior in patients with Parkinson disease (PD)?

Findings

This meta-analysis of 28 studies found a high prevalence of suicidal ideation and behavior in patients with PD. Compared with controls, patients with PD experienced 2 times the risk of engaging in suicidal behavior.

Meaning

Patients with PD may be at elevated risks of both suicidal ideation and behavior, and physicians ought to maintain a high index of suspicion for early detection and appropriate management.

Abstract

Importance

Suicide risk may be increased in patients with Parkinson disease (PD), a common neurodegenerative condition. Mood disorders, especially depression, are prevalent in patients with PD who report suicidality.

Objective

To address inconsistent results from studies of suicidal ideation and behavior in patients with PD.

Data Sources

The study team searched MEDLINE and Embase from inception to June 14, 2023, and further screened the bibliographies of relevant studies to ensure a comprehensive search.

Study Selection

Original studies, published in English, discussing either suicidal ideation, behavior, or both in adults with PD were included. Accepted study designs included cross-sectional, case-control, and cohort studies. Studies that only included patients with PD after deep brain stimulation were excluded.

Data Extraction and Synthesis

This meta-analysis was conducted in line with the PRISMA guidelines. Two authors reviewed each study and extracted the data independently, with discrepancies referred to a third independent author.

Main Outcomes and Measures

Outcomes included the prevalence of suicidal ideation and behavior, measured as proportions, and the risk of suicidal behavior in patients with PD relative to controls, measured in both odds ratio (OR) and hazards ratio (HR).

Results

A total of 28 studies comprising 505 950 PD patients were included in the final analysis. The prevalence of suicidal ideation was evaluated in 14 studies (22.2%; 95% CI, 14.6-32.3) and suicidal behavior in 21 studies (1.25%; 95% CI, 0.64-2.41). Excluding 4 outliers, prevalence of suicidal behavior was significantly higher in prospective studies (1.75%; 95% CI, 1.03-2.95) than retrospective studies (0.50%; 95% CI, 0.24-1.01). Excluding 1 outlier, OR of suicidal behavior was pooled across 10 studies and significant (OR, 2.15; 95% CI, 1.22-3.78; P = .01). HR of suicidal behavior was assessed in 9 studies (HR, 1.73; 95% CI, 1.40-2.14; P < .001).

Conclusions and Relevance

This meta-analysis involving more than 500 000 patients with PD found 22.2% and 1.25% of patients with PD to have suicidal ideation and behavior, respectively. Patients with PD had 2 times the risk of suicidal behavior than controls. Early recognition and management of suicidality in PD can help reduce mortality.

Introduction

Late-life suicide is an important global health issue and represents a preventable cause of death. Older adults are at an increased risk owing to a myriad of factors, such as high prevalence of disease, health care barriers, and psychosocial problems.¹ An especially vulnerable group of patients are those with a chronic neurological disorder. Parkinson disease (PD) is a prototypical disease where most patients are not demented and experience the feeling of being trapped in a disabled body.² Beyond motor symptoms, PD is frequently associated with significant nonmotor symptoms, as well as great psychological burden. Depression and anxiety, for example, are commonly present in patients with PD, affecting up to 30 to 50% of patients.³

Suicidality exists as a spectrum, ranging from suicidal ideation to suicidal behavior.⁴ Suicidal ideation could either be passive (desire for death but without planning) or active (with planning for suicide). The risk of suicidal behavior (either an attempt or death by suicide) increases with the frequency, content, such as the severity of depressed mood and hopelessness, and the intent of suicidal ideation, such as the degree to which the suicide is planned.⁵ Unfortunately, patients at risk of suicidal ideation and behavior frequently go undetected.⁶ Ageism and stigma regarding mental health services remain additional barriers to suicide prevention.⁷

The literature surrounding suicide risk in patients with PD unfortunately remains unclear and there have been discrepancies in findings across various studies. While some studies have demonstrated an increased risk,^8,9,10 there is also evidence that suicide risk is lower in comparison with the general population.^11,12 Given the great psychosocial burden experienced by patients with PD, it is important to examine if, and by how much, the risk of suicide is increased in this patient population. Shepard et al¹³ previously conducted a systematic review in 2019 on the same topic. However, to our knowledge, there has been no meta-analysis providing a quantitative assessment of prevalence and risk of suicidality in patients with PD. To address this gap in knowledge, we conducted a systematic review with meta-analysis of studies that provided the rates and risks of suicide in patients with PD. In the present article, we explored the following outcomes of interest: (1) the prevalence of suicidal ideation, (2) the prevalence of suicidal behavior, (3) and the risks of suicidal behavior when contrasted with the general population.

Methods

Search Strategy

This meta-analysis was registered with PROSPERO at CRD42022346931 and conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guidelines.¹⁴ We searched MEDLINE and Embase using a search strategy comprising keywords and medical participant headings related to “Parkinson’s disease” and “suicide” (eTable in Supplement 1). The search was conducted from inception to June 14, 2023. The bibliographies of related articles were also screened to ensure a comprehensive review. The search results were exported to Zotero for duplicate removal and reference management.

Study Selection

Two authors (A.S.M. and J.H.Y.) reviewed each reference in a blinded manner and any disagreements were resolved through discussion. Any unresolved disagreements were referred to an independent senior author for the final decision (E.K.T.). The review was carried out in 2 stages: first, the titles and abstracts were reviewed and second, the full texts of selected references were retrieved and reviewed. Original studies, published in English, discussing either suicidal ideation or behavior or both in adults with PD were included. Accepted study designs included cross-sectional, case-control, and cohort studies. We excluded review articles (including other systematic reviews and meta-analyses), editorials, commentaries, letters to editor, and conference abstracts. Studies including only patients with PD after deep brain stimulation (DBS) were excluded.

Data Extraction

Data were extracted and confirmed by 4 authors (A.S.M., Y.C., B.X., and Z.Z.) in an independent fashion with any differences resolved by reaching a consensus. If consensus could not be achieved, the decision is referred to an independent senior author (E. K.T.). Outcomes of interest were the prevalence estimates of suicidal ideation and behavior, which included both attempted and completed suicide in patients with PD, as well as the risk of suicidal ideation and behavior in patients with PD compared with control populations. Suicidal ideation includes past and current thoughts about suicide, whether passive or active, and suicidal behavior includes both suicidal attempts and death by suicide. Study-related details (study design, sample size, geographic region of study, disease, and control populations) and demographic information (mean age and proportion of females) were also extracted to identify potential subgroups for analysis.

Continuous variables were extracted as means and SDs, while binary variables were extracted as events and the sample size for each relevant group. Formulae derived by Wan et al¹⁵ were used to convert other measures of continuous data to means and standard deviations when needed.

Statistical Analysis

All analyses were undertaken using RStudio version 4.1.1 using the meta package. Logit-transformed prevalence estimates were pooled using the random-intercept logistic regression model. Odds ratios (ORs) were meta-analyzed using the Mantel-Haenszel method, while the precalculated log-transformed hazards ratios (HRs) were pooled using the inverse variance method. The level of significance is defined as P < .05.

The choice between the fixed-effect and random-effects models was made depending on the I² index and Cochran Q test P value. An I² of less than 25% is indicative of low heterogeneity, 25 to 75% of moderate heterogeneity, and more than 75% of high heterogeneity. In addition, P < .05 for Cochran Q test suggests significant heterogeneity. In cases with minimal heterogeneity, a fixed-effect model was used. Otherwise, a random-effects model was used. All results were presented as their effect sizes with the accompanying 95% CIs, along with the P values where applicable.

For meta-analyses that have high heterogeneity, we performed an influence analysis to determine the contribution of each study to the overall heterogeneity. Based on the resultant Baujat plots and leave-one-out analyses, as well as inspection of the forest plots, we performed a sensitivity analysis in which outliers were excluded. In addition, we conducted subgroup analysis according to whether the study used a prospective or retrospective design.

Quality Assessment

Two blinded reviewers (A.S.M. and J.H.Y.) conducted the assessment and any conflicts were resolved through discussion. We used the Joanna Briggs Institute Critical Appraisal Tools for the quality assessment of included articles. The maximum score attainable (signifying high quality) is 8 points for analytical cross-sectional studies, 10 points for case-control studies, and 11 points for cohort studies.

Results

Summary of Included Articles

A total of 807 studies were found after searching MEDLINE and Embase. Among these, 289 were duplicates and 518 studies remained following duplicate removal. The study team screened the titles and abstracts of these studies and included 35 studies for further review. The study team retrieved the full texts of these 35 studies and 28 studies of 505 950 patients with PD were included in the final analysis (Figure 1).

Figure 1. PRISMA Flow Diagram.

Among these, 15 studies were cross-sectional studies, 8 were retrospective cohort studies, and 3 were case-control studies. Santos-Garcia et al¹⁶ used a prospective cohort design, while Kostic et al¹⁷ used both cross-sectional and cohort study designs. Only 3 of the studies reported a follow-up period: Kostic et al¹⁷ reported that the patients were followed up with for 8 years, Lee et al¹⁸ reported a mean follow-up time of 7.1 (SD, 4.1) years, and Santos-Garcia et al¹⁶ reported that patients were followed up for 2 years.

Most studies were of high quality according to the JBI Critical Appraisal Tools, but 8 studies had 2 points deducted due to the lack of identification and adjustment for confounders. Most studies focused on patients with PD, except for 5 that analyzed a variety of other disease conditions and not specifically PD. A summary of included articles can be found in the Table.

Table. Summary of Included Articles.

Source	Design	Patient subgroups	Sample size	Age, y (SD)	Female, No. (%)	Qualitya
Allyson Jones et al,19 2012	Retrospective cohort	IPD	10 910	NA	4963 (45.5)	11/11
Berardelli et al,20 2018	Cross-sectional	IPD	100	65.8 (10.3)	37 (37.0)	6/8b
		Controls (open-angle glaucoma)	91	64.5 (10.6)	39 (42.9)
Chen et al,21 2021	Retrospective cohort	IPD	35 891	72.5 (10.1)	17 482 (48.7)	11/11
		Controls	143 557	72.5 (10.1)	69 928 (48.7)
Chendo et al,22 2023	Cross-sectional	Late-stage IPD	92	75.0 (6.7)	53 (57.6)	8/8
Dissanayaka et al,23 2010	Cross-sectional	IPD	79	67.2 (10.5)	37 (46.8)	8/8
Elfil et al,24 2020	Case control	IPD	186	64.8 (9.2)	94 (50.5)	10/10
		Controls	177	63.6 (5.7)	83 (46.9)
Eliasen et al,25 2018c	Case control	Suicidal behavior (poisoning)	8974	41.5 (15.6)	5626 (62.7)	10/10
		Controls	89 740	41.5 (15.6)	56 260 (62.7)
Erlangsen et al,26 2015c,d	Retrospective cohort	Individuals ≥65 y	1 849 110	NA	1 010 047 (54.6)	11/11
Erlangsen et al,27 2020c	Retrospective cohort	Individuals ≥15 y	7 300 395	NA	3 657 498 (50.1)	11/11
Heronemus et al,28 2022	Retrospective cohort	IPD	80 150	NA	1004 (1.3)	11/11
		Controls	1 416 481	NA	17 737 (1.3)
Hinkle et al,29 2018e	Cross-sectional	IPD	223	NA	NA	8/8
Jeong et al,30 2022	Retrospective cohort	IPD	17 143	71.10 (6.0)	10 682 (62.3)	11/11
		Controls	17 143	71.10 (6.0)	10 682 (62.3)
Kasten et al,31 2012	Cross-sectional	LOPD	90	68.0 (7.0)	31 (34)	6/8b
		EOPD	38	51.0 (7.0)	20 (53)
		GPD (MMC)	23	54.0 (9.0)	10 (44)
		GPD (NMC)	19	48.0 (9.0)	8 (42)
		Controls	127	59.0 (12.0)	61 (48)
Kostic et al,17 2010	Cohort	IPD	102	58.4 (9.9)	47 (46.1)	9/11b
	Cross-sectional	IPD	128	62.7 (9.3)	40 (31.2)	6/8b
Kumar et al,32 2021	Cross-sectional	IPD	239	59.4 (7.6)	96 (40.2)	6/8b
Kummer et al,33 2009	Cross-sectional	IPD with suicidal ideation	13	51.6 (6.7)	7 (53.8)	8/8
		IPD without suicidal ideation	77	57.1 (10.6)	29 (37.7)
Lee et al,18 2016	Case control	IPD with suicidal behavior	29	62.3 (13.7)	10 (34.5)	10/10
		IPD without suicidal behavior	116	63.5 (9.2)	66 (56.9)
Lee et al,34 2018c	Retrospective cohort	Death by suicide	725	NA	206 (28.4)	11/11
		Death by other causes	12 718	NA	4687 (36.9)
		Survived	286 789	NA	139 354 (48.6)
Lee et al,35 2021	Cross-sectional	IPD with depression	50	66.0 (9.0)	21 (42.0)	6/8b
		IPD without depression	125	65.0 (9.4)	42 (33.6)
Merschdorf et al,36 2003	Cross-sectional	IPD	49	NA	NA	6/8b
		MDD	38	NA	NA
Myslobodsky et al37 2001	Cross-sectional	Patients with PD who died	144 364	81.2 (range, 40.0-113.0)	62 077 (43.0)	6/8b
Nazem et al,38 2008	Cross-sectional	IPD	116	64.7 (10.4)	29 (25.0)	8/8
Ou et al,39 2021	Cross-sectional	EOPD	577	43.6 (7.6)	264 (45.8)	8/8
		LOPD	2937	64.4 (8.9)	1378 (46.9)
Ozdilek et al,40 2014	Cross-sectional	IPD with suicidal ideation	14	56.7 (9.9)	4 (28.6)	8/8
		IPD without suicidal ideation	106	62.2 (8.8)	38 (35.8)
Rai et al,41 2015	Cross-sectional	IPD	126	57.9 (10.9)	NA	8/8
Roberts et al,42 2018c	Retrospective cohort	England sample	8 970 811	median, 61.0	4 933 946 (55.0)	11/11
		Wales sample	577 474	median, 62.0	323 385 (56.0)
Santos-Garcia et al,16 2023	Prospective cohort	IPD with suicidal ideation	35	60.7 (10.0)	21 (60.0)	11/11
		IPD without suicidal ideation	658	62.7 (8.9)	262 (39.8)
Willis et al,43 2016f	Cross-sectional	IPD	104 437	NA	NA	8/8
		Controls	3 814 266	NA	NA

Abbreviations: EOPD, early-onset Parkinson disease; GPD, genetic Parkinson disease; IPD, idiopathic Parkinson disease; LOPD, late-onset Parkinson disease; MDD, major depressive disorder; MMC, manifesting mutation carrier; NA, not applicable; NMC, nonmanifesting mutation carrier; PD, Parkinson disease.

^a The quality of each study was assessed using the JBI Critical Appraisal Tool according to their respective study design. The maximum possible score for cross-sectional studies is 8, for cohort studies is 11, and for case-control studies is 10.

^b These studies did not score the maximum score possible for their respective JBI Critical Appraisal tool due to both a lack of identification and adjustment for potential confounders.

^c This study analyzed the risk of suicide in other disease conditions as well; information regarding individuals with PD not complete. The data were reported for the subgroups defined by the study.

^d Analyzed the risk of suicide in various disease conditions and presented the results separately for men and women. An overall analysis of both sexes was not provided.

^e Used the data set from Methods of Optimal Depression Detection in Parkinson disease study. Mean age and proportion of females were not reported for the overall sample.

^f Used the National Inpatient Sample and investigated hospitalizations for mental health or substance abuse conditions. The sample size for demographic variables were varied due to differences in the extent of missing data.

Suicidal Ideation

Fourteen studies involving 3323 patients with PD and 637 events of suicidal ideation were pooled using the random-effects model and found the prevalence to be 22.2% (95% CI, 14.6-32.3). The I² index was 94.8% and the Cochran Q test was significant at P < .001.

Influence analysis revealed 3 outliers: Merschdorf et al,³⁶ Santos-Garcia et al,¹⁶ and Dissanayaka et al.²³ A sensitivity analysis excluding them was conducted and found the prevalence to be 24.0% (95% CI, 19.1-29.7). The I² index was 89.0% and the Cochran Q test was significant at P < .001.

Suicidal Behavior

Prevalence

Twenty-one studies involving 423 148 participants (inclusive of non-PD controls) and 2700 events of suicidal behavior were pooled and the prevalence was found to be 1.25% (95% CI, 0.64-2.41). The I² index was 99.6% and the Cochran Q test was significant at P < .0001.

Influence analysis revealed 4 outliers: Jones et al,¹⁹ Myslobodsky et al,¹¹ Eliasen et al,⁹ and Kasten et al.³¹ A sensitivity analysis exclusive of these studies was undertaken and the prevalence estimate was 0.89% (95% CI, 0.52-1.52). The I² index was 94.3% and the Cochran Q test was significant at P < .001.

In addition, the study team conducted a subgroup analysis according to whether the study used a prospective or retrospective design. Prior to outlier removal, the prospective studies (k = 10) demonstrated a prevalence of 2.34% (95% CI, 1.27-4.27), while the retrospective studies (k = 11) demonstrated a prevalence of 0.78% (95% CI, 0.28-2.18). The subgroup differences trended toward but did not achieve statistical significance.

However, following outlier removal, the prospective studies demonstrated a significantly higher prevalence compared with the retrospective studies (P = .01). Respectively, the prevalence estimates were 1.75% (95% CI, 1.03-2.95) and 0.50% (95% CI, 0.24-1.01). The forest plot is presented in Figure 2.

Figure 2. Forest Plot of Subgroup Analysis of Prevalence of Suicidal Behavior Comparing Consecutive Recruitment vs Retrospective Database Review, Following Outlier Removal.

OR

Eleven studies involving 19 371 103 participants (inclusive of non-PD controls) and 103 435 events were pooled to find a nonsignificant OR of 1.57 (95% CI, 0.69-3.60; P = .28). The I² index was 99.6% and the Cochran Q test was significant at P < .001.

Influence analysis revealed 1 outlier: Willis et al.¹² A sensitivity analysis excluding this study was conducted and the OR became statistically significant at 2.15 (95% CI, 1.22-3.78; P = .01). The forest plot is presented in Figure 3. Irrespective of outlier removal, subgroup analysis by study design (prospective vs retrospective) did not find any statistical significance (before outlier removal, P = .84; after outlier removal, P = .61).

Figure 3. Forest Plot of Odds Ratio (OR) of Suicidal Behavior in Parkinson Disease vs Controls Following Outlier Removal.

HR

Nine studies were meta-analyzed to find a statistically significant HR of 1.73 (95% CI, 1.40-2.14; P < .001). The I² index was 32.8% and the Cochran Q test was not statistically significant with P = .16. Visual inspection of the forest plot (Figure 4) and influence analysis did not find any outliers. There were also no statistically significant subgroup differences between studies that used a prospective vs retrospective design.

Figure 4. Forest Plot of Hazards Ratio (HR) of Suicidal Behavior in Parkinson Disease vs Controls.

Biological Sex and Suicidal Behavior

Four studies analyzed suicide risk in men and women separately and a random-effects meta-analysis was conducted. The pooled HR is 1.83 (95% CI, 1.54-2.18; P < .001) for men and 2.30 (95% CI, 1.85-2.86; P < .001) for women. The heterogeneity is very low 0%, Cochran Q test P = .71, and P = .82 for men and women, respectively. The forest plot for men in eFigure 1 in Supplement 1 and for women is presented in eFigure 2 in Supplement 1.

Discussion

This meta-analysis of 28 studies demonstrated a high prevalence of suicidal ideation, affecting over a quarter of patients with PD. There is also a substantial prevalence of suicidal behavior (at about 1%), though studies with a consecutive recruitment process detected a significantly higher prevalence compared with studies using retrospective data. We also found patients with PD to be at a greater risk of suicidal behavior compared with other nonneurologic diseases or the general population. While there is significant heterogeneity observed in the meta-analysis of prevalence estimates, there was minimal heterogeneity detected when meta-analyzing estimates of OR and HR. The quality of included studies was generally high, though 8 of 28 studies did not explicitly identify and adjust for confounders.

It is important to recognize suicidality exists as a spectrum, starting from suicidal ideation and progressing to suicidal behavior in those who are most vulnerable. Patients with PD possess multiple risk factors for suicidality, such as advanced age, living with a chronic condition, as well as limitations in physical mobility and functional ability. The association between physical illness and suicidality had been consistently demonstrated with multiple illnesses relating strongly to a higher risk of suicide.⁴⁴ Functional disability was also shown to be another independent risk factor of both suicidal ideation and behavior, possibly due to diminished social connectivity and feelings of isolation or loneliness.⁴⁵ The association between the 2 is likely to be mediated through mental health factors, with prior studies demonstrating patients with PD to be at greater risk of developing mood disorders, such as depression and anxiety. Importantly, the current literature suggests depression to be among the greatest risk factors for suicidal ideation and behavior in late life, though it is unfortunately underdiagnosed and undertreated.^44,45,46

Patients with PD often experience great psychiatric comorbidity, of which the most prominent is depression. Depressive mood disorders are the greatest risk factors for both suicidal ideation and suicidal behavior and are present in almost half of patients with PD.¹⁷ Additional risk factors include sleep disorders and feelings of hopelessness. First, sleep disorders are among the most frequent nonmotor symptoms in PD and include insomnia, rapid eye-movement sleep behavior disorder, and excessive daytime sleepiness. The presence and severity of sleep disorders are associated with depression severity,^47,48 which is in turn related to suicidality. Second, feelings of hopelessness (also called demoralization) are also common in patients with PD and are increasingly recognized as a significant risk factor for suicide risk.⁴⁹ Lastly, a diagnosis of PD is often a stressful life event for patients, which can acutely exacerbate suicidality in those who are vulnerable.⁵⁰ Efforts directed at identifying and addressing these risk factors, such as improving the quality and quantity of sleep through medications, could be helpful for these patients.

Treatment-related factors can also increase the risk of suicidality in patients with PD. The perceived treatment status of fluctuations in patients with PD is associated with depression and suicidality. Patients experiencing untreated motor fluctuations are more likely to have depressive disorders and report suicidality.⁵¹ There is evidence pointing toward an association between the use of dopaminergic agents and suicide,⁸ though some authors argue that such suicide cases are due to preexisting depression or other disease-related factors.⁵² Another contentious association is reported between DBS and increased suicide risk. A recent meta-analysis of patients with PD after DBS showed similar rates of suicidal behavior (1%) as found in our study,⁵³ while another study demonstrated an increased rate of completed suicide in patients with PD post-DBS compared with general population.⁵⁴ However, there has been no convincing evidence that DBS increases the risk of suicide compared with pharmacotherapy alone.⁵⁴ In either case, physicians must appreciate the increased suicide risk in all patients with PD patients and care should be taken when considering patients for DBS or other treatment options. Monitoring and managing suicidality should be a priority, especially for those with psychiatric risk factors, such as preexisting depression or other mood disorders.

Sex-related differences in PD have generated significant interest as such differences in symptomatology could have a significant impact on diagnostic accuracy and management strategies. The relationship between the female sex and numerous nonmotor symptoms has been shown in current literature, such as increased risk of profound depression, persistent anxiety, and greater pain severity.⁵⁵ Nonetheless, our study found no evidence that there are sex-related differences in suicidal behavior, albeit our analysis was limited by the paucity of data. Future research should investigate if there is a so-called “gender paradox” in patients with PD with suicidality, where women experience a greater risk of suicide attempts and men have a greater risk of complete suicides.⁵⁶

Suicidality can be difficult to detect in patients with PD and is frequently missed even in those on long-term follow up. When recognized, suicidality tends to be severe with either active ideation or even suicide attempts.^3,57 Numerous barriers prevent patients from reaching out and getting the help they need, such as stigma toward receiving mental health care, lack of knowledge, limited accessibility, and financial or logistical issues.^58,59 As such, a treatment gap exists that requires much effort to reduce. Interestingly, we found that studies with a retrospective design detected a significantly lower prevalence of suicidal behavior compared with those with a prospective design. This suggests that prevalence estimates derived from retrospective studies may be lower than the true prevalence. A potential explanation is that only completed suicides and suicidal attempts requiring medical attention are detected and documented in medical databases. Prospective studies with patient interviews may hence be more sensitive to suicidal ideation and behavior and provide more reliable data.

Suicide remains a major health problem, especially in developed countries. Early detection and management of suicidality is an area where further efforts are needed. Implementing pilot programs could help in understanding viable solutions to mitigate the barriers to health care. Promoting awareness regarding mental health and help-seeking behavior through educational campaigns is another potential solution. Management of both medical, such as sleep disorders, and psychosocial risk factors, such as feelings of loneliness, hopelessness, and depressed mood, could be useful in lowering suicide risk in patients with PD.

Limitations

This study has some limitations. First, there is heterogeneity across studies due to a variety of factors, such as study design, geographic variation, and differences in patient demographics. To mitigate this, we have performed extensive sensitivity analyses such as influence analyses and exclusion of outliers. We have also performed subgroup analyses according to study design. Second, suicide research has evolved over the past decades with changes in terminologies. Thus, there is unavoidable variability in the definitions of suicidal ideation and behavior used in the included studies. Third, many of the included studies used a prospective cross-sectional design and thus are unable to account for completed suicides and potentially underestimating the prevalence of suicidal behavior. To assess the effect of this bias, we undertook a subgroup analysis comparing prospective studies and retrospective ones. Even so, we found a statistically significantly higher prevalence estimate of suicidal behavior in prospective studies.

Conclusions

This meta-analysis involving more than 500 000 patients with PD demonstrated a 22.2% prevalence of suicidal ideation and 1.25% suicidal behavior in PD. In this study, patients had 2 times the risk of suicidal behavior than controls. Early recognition and management of suicidality in PD can help reduce mortality and improve quality of life.

Supplement 1.

eTable. Search strategies for Medline and Embase

eFigure 1. Forest plot of hazards ratio of suicidal behavior in men with Parkinson disease vs controls

eFigure 2. Forest plot of hazards ratio of suicidal behavior in women with Parkinson disease vs controls

Supplement 2.

Data sharing statement