Means restriction for the prevention of suicide by jumping

Chukwudi Okolie; Suzanne Wood; Keith Hawton; Udai Kandalama; Alexander C Glendenning; Michael Dennis; Sian F Price; Keith Lloyd; Ann John

doi:10.1002/14651858.CD013543

. 2020 Feb 25;2020(2):CD013543. doi: 10.1002/14651858.CD013543

Means restriction for the prevention of suicide by jumping

Chukwudi Okolie ^1,², Suzanne Wood ³, Keith Hawton ⁴, Udai Kandalama ¹, Alexander C Glendenning ¹, Michael Dennis ¹, Sian F Price ⁵, Keith Lloyd ¹, Ann John ^1,^2,^✉

Editor: Cochrane Common Mental Disorders Group

PMCID: PMC7039710 PMID: 32092795

Abstract

Background

Jumping from a height is an uncommon but lethal means of suicide. Restricting access to means is an important universal or population‐based approach to suicide prevention with clear evidence of its effectiveness. However, the evidence with respect to means restriction for the prevention of suicide by jumping is not well established.

Objectives

To evaluate the effectiveness of interventions to restrict the availability of, or access to, means of suicide by jumping. These include the use of physical barriers, fencing or safety nets at frequently‐used jumping sites, or restriction of access to these sites, such as by way of road closures.

Search methods

We searched the Cochrane Library, Embase, MEDLINE, PsycINFO, and Web of Science to May 2019. We conducted additional searches of the international trial registries including the World Health Organization International Clinical Trials Registry Platform (ICTRP) and ClinicalTrials.gov, to identify relevant unpublished and ongoing studies. We searched the reference lists of all included studies and relevant systematic reviews to identify additional studies and contacted authors and subject experts for information on unpublished or ongoing studies. We applied no restrictions on date, language or publication status to the searches. Two review authors independently assessed all citations from the searches and identified relevant titles and abstracts. Our main outcomes of interest were suicide, attempted suicide or self‐harm, and cost‐effectiveness of interventions.

Selection criteria

Eligible studies were randomised or quasi‐randomised controlled trials, controlled intervention studies without randomisation, before‐and‐after studies, or studies using interrupted time series designs, which evaluated interventions to restrict the availability of, or access to, means of suicide by jumping.

Data collection and analysis

Two review authors independently selected studies for inclusion and three review authors extracted study data. We pooled studies that evaluated similar interventions and outcomes using a random‐effects meta‐analysis, and we synthesised data from other studies in a narrative summary. We summarised the quality of the evidence included in this review using the GRADE approach.

Main results

We included 14 studies in this review. Thirteen were before‐and‐after studies and one was a cost‐effectiveness analysis. Three studies each took place in Switzerland and the USA, while two studies each were from the UK, Canada, New Zealand, and Australia respectively. The majority of studies (10/14) assessed jumping means restriction interventions delivered in isolation, half of which were at bridges. Due to the observational nature of included studies, none compared comparator interventions or control conditions.

During the pre‐ and postintervention period among the 13 before‐and‐after studies, a total of 742.3 suicides (5.5 suicides per year) occurred during the pre‐intervention period (134.5 study years), while 70.6 suicides (0.8 suicides per year) occurred during the postintervention period (92.4 study years) ‐ a 91% reduction in suicides. A meta‐analysis of all studies assessing jumping means restriction interventions (delivered in isolation or in combination with other interventions) showed a directionality of effect in favour of the interventions, as evidenced by a reduction in the number of suicides at intervention sites (12 studies; incidence rate ratio (IRR) = 0.09, 95% confidence interval (CI) 0.03 to 0.27; P < 0.001; I² = 88.40%). Similar findings were demonstrated for studies assessing jumping means restriction interventions delivered in isolation (9 studies; IRR = 0.05, 95% CI 0.01 to 0.16; P < 0.001; I² = 73.67%), studies assessing jumping means restriction interventions delivered in combination with other interventions (3 studies; IRR = 0.54, 95% CI 0.31 to 0.93; P = 0.03; I² = 40.8%), studies assessing the effectiveness of physical barriers (7 studies; IRR = 0.07, 95% CI 0.02 to 0.24; P < 0.001; I² = 84.07%), and studies assessing the effectiveness of safety nets (2 studies; IRR = 0.09, 95% CI 0.01 to 1.30; P = 0.07; I² = 29.3%). Data on suicide attempts were limited and none of the studies used self‐harm as an outcome. There was considerable heterogeneity between studies for the primary outcome (suicide) in the majority of the analyses except those relating to jumping means restriction delivered in combination with other interventions, and safety nets. Nevertheless, every study included in the forest plots showed the same directional effects in favour of jumping means restriction. Due to methodological limitations of the included studies, we rated the quality of the evidence from these studies as low.

A cost‐effectiveness analysis suggested that the construction of a physical barrier on a bridge would be a highly cost‐effective project in the long term as a result of overall reduced suicide mortality.

Authors' conclusions

The findings from this review suggest that jumping means restriction interventions are capable of reducing the frequency of suicides by jumping. However, due to methodological limitations of included studies, this finding is based on low‐quality evidence. Therefore, further well‐designed high‐quality studies are required to further evaluate the effectiveness of these interventions, as well as other measures at jumping sites. In addition, further research is required to investigate the potential for suicide method substitution and displacement effects in populations exposed to interventions to prevent suicide by jumping.

Plain language summary

Means restriction to prevent suicide by jumping

Why is this review important?

Jumping from a height is an uncommon but lethal means of suicide. While there is evidence that restricting access to means of suicide is an effective approach for preventing suicides, the evidence for preventing suicide by jumping is not well established. This review therefore aimed to explore the impact restriction of access would have on suicide by jumping.

Searching for evidence

We searched several databases (the Cochrane Library, Embase, MEDLINE, PsycINFO, and Web of Science) to find studies that assessed the impact of restricting access to means of suicide by jumping. We searched the databases up until May 2019. We included studies that assessed jumping means restriction interventions delivered on their own, such as physical barriers, fencing or safety nets on bridges, or those delivered in combination with other suicide prevention interventions, such as crisis telephones and CCTV cameras. We also searched the reference lists of all included studies and relevant systematic reviews to identify additional studies and contacted authors to obtain missing information. Our main outcomes of interest were suicide, attempted suicide or self‐harm and cost‐effectiveness of interventions.

Key results

We found 14 relevant studies. Three studies each were from Switzerland and the USA, while two studies each were from the UK, Canada, New Zealand, and Australia respectively. The majority of studies had a before‐and‐after study design. Due to the observational nature of our included studies, none compared other interventions or control conditions. Jumping means restriction interventions delivered in isolation or in combination with other interventions were found to reduce the number of suicides by jumping. Data on suicide attempts were limited and no study reported self‐harm. A cost‐effectiveness analysis suggested that the construction of a physical barrier on a bridge would be cost‐effective in the long term. The evidence for these assessments was of low quality because of weaknesses in study design and differences in findings between studies, therefore requiring the need for further high‐quality studies.

Summary of findings

Background

Description of the condition

Suicide rates

Worldwide, suicide is recognised as a serious public health problem. Approximately 800,000 people die from suicide every year; a global mortality rate of 10.5 per 100,000 population (WHO 2019). However, it is suggested that global suicide figures could be underreported due to misclassification of suicides as accidents or other causes of death (WHO 2014). Suicide occurs throughout the lifespan and is the second‐leading cause of death in 15 to 29 year olds worldwide, after road injury (WHO 2018). Globally, the majority of deaths by suicide occur in low‐and‐middle‐income countries (79%), however, high‐income countries have the highest age‐standardised suicide rates (11.5 per 100,000) (WHO 2019). Self‐harm, which includes acts of self‐poisoning or self‐injury carried out by an individual irrespective of motivation (NICE 2011), is much more common than suicide and a significant cause of morbidity and mortality, including by suicide (Sinclair 2010; WHO 2014). In describing self‐harm, we have followed the approach favoured in the UK and some other countries where all intentional self‐harm (suicide attempt) is described/included in a single category, namely self‐harm. (Hawton 2016).

Risk factors/causes

There are a variety of risk factors for suicide. These include mental disorders (particularly depression, anxiety, psychosis disorders and substance misuse), psychological, biological and genetic factors, exposure to role models and early‐life adversity (Hawton 2009; Turecki 2016). A prior suicide attempt is the single most important risk factor for suicide in the general population (WHO 2014). Sex is also a factor, with higher rates of suicide reported in males (13.7 per 100,000) than in females (7.5 per 100,000) (WHO 2019). However, unlike suicide, self‐harm usually occurs more commonly in females (Geulayov 2016; Hawton 2008). Whatever the background factors at the point when a person feels hopeless and suicidal, access to the means of suicide can be decisive (Hawton 2007). Availability of means can increase the likelihood of a suicide attempt, particularly where impulsive behaviour is a factor (Hawton 2007). The nature of the method chosen will influence the outcome (Yip 2012).

Definitions

A range of different terms are used for suicide and suicidal behaviour. In the context of this review, 'suicidal behaviour' refers to any form of intentional self‐injurious or self‐poisoning behaviour with known suicidal intent. 'Suicide' refers to self‐injurious or self‐poisoning behaviour with a fatal outcome and known suicidal intent or where that intent was underdetermined. Self‐harm includes acts of self‐poisoning or self‐injury irrespective of motivation.

Means of suicide

Despite differences between countries, worldwide three principal methods of suicide predominate ‐ poisoning by ingestion of pesticides, hanging, and use of firearms.(WHO 2014). Suicide by jumping and other methods of poisoning (usually poisoning with drugs) are also significant methods. International variations in suicide methods suggest that the observed suicide pattern in many countries depends on the availability of the methods used (Ajdacic‐Gross 2008; Yip 2012). For example, suicide by jumping is the most common method of suicide in Hong Kong and other urban societies with extensive high‐rise housing for both residential and work purposes (Wong 2014).

Jumping from a height, though an uncommon means of suicide, can be lethal (Elnour 2008). Most suicides by jumping occur from high‐rise residential buildings (Beautrais 2007). Other common sites include cliffs, bridges, and terraces – some of which have gained reputations as sites for suicides (Beautrais 2007; Pirkis 2015). Prominent media coverage and style of reporting may contribute to imitation or contagion and thus enhance the status of these sites as places for suicide (Beautrais 2007; John 2017).

Description of the intervention

Restriction of availability or access to lethal methods of suicide (means restriction) is an important universal approach to suicide prevention. Universal prevention strategies are targeted at the general public or entire population groups. These strategies are designed to influence everyone and typically affect people whose suicide risk is otherwise undetected (Yip 2012). Means restriction is underpinned by the concept of intervention being available during acute periods of risk for suicidal behaviour; for example as might occur when a person with depression is exposed to an adverse life event. If access to means is restricted at this point, the chance of survival beyond the stage of acute risk increases. Evidence from research on near lethal suicide attempts supports the idea that, at least for a proportion of people, these acts may be an impulsive response and probably would not have occurred if the means had not been readily available (Hawton 2005). Means restriction is therefore applied to the population as a whole, where it typically affects vulnerable individuals, whose suicide risk may be undetected and who have not sought assistance in the midst of this crisis (Yip 2012).

It has been argued that restricting access to one method will lead to substitution with another. There is evidence, however, that restricting access to means during periods of acute risk can have an impact on an individual’s likelihood of dying from suicide in the longer term. In the UK, an often used example is coal gas. From the late 1950s to the early 1970s, domestic gas supplies changed from toxic coal gas to non‐toxic North Sea natural gas. After the Second World War, suicide rates in the UK had been increasing and carbon monoxide poisoning, using a gas oven, was the most common method. With the change to natural gas, the numbers of these deaths fell. Despite a slight increase in other methods, the net effect was a large reduction in suicide deaths (Kreitman 1976).

How the intervention might work

Suicides by jumping normally occur from man‐made and natural points of elevation, such as high‐rise buildings, bridges, cliffs and terraces. Means restriction interventions enacted at these sites work by either providing a physical obstruction to prevent jumping from these sites (e.g. physical barriers, fences, guard rails) or by restricting access to these sites (e.g., road closure, prevention of vehicular access). A range of studies have demonstrated that the construction of barriers and fencing at high‐risk jumping sites, especially bridges, can lead to a reduction in the number of suicides by this method (Beautrais 2001; Bennewith 2007; Pelletier 2007; Pirkis 2013; Pirkis 2015). These interventions are oftentimes used in combination with other suicide prevention measures such as interventions aimed at increasing opportunities for help‐seeking (e.g. crisis telephones), or those aimed at increasing the opportunity and capacity for human intervention (e.g. CCTV camera surveillance). Other measures such as responsible media reporting ‐ including not reporting on method and context of a suicide (e.g. details of bridge), are also important for suicide prevention when used in combination with means restriction interventions.

Why it is important to do this review

Although there is clear evidence of the effectiveness of restricting access to lethal means as a suicide prevention measure, the evidence with respect to means restriction for the prevention of suicide by jumping is not well established. Previous reviews have focused on either suicide prevention at frequently‐used locations (Cox 2013; Pirkis 2015), or general prevention of suicide by jumping (Beautrais 2007). A review published in 2013 assessed the effectiveness of structural interventions at frequently‐used locations (Pirkis 2013). This review did not follow strict Preferred Reporting Items for Systematic Reviews and Meta‐Analyses (PRISMA) guidelines (Moher 2010), did not assess risk of bias of included studies, and had study searches conducted in July 2012. An updated systematic review specifically focused on means restriction interventions to prevent jumping suicides, using PRISMA guidelines, is therefore timely and necessary.

Objectives

Methods

Criteria for considering studies for this review

Types of studies

We included primary research studies with the following study designs in this review: randomised controlled trials (RCTs), cluster‐RCTs, cross‐over RCTs, and quasi‐RCTs (trials in which allocation of participants to study arms is not truly random). Other study types eligible for inclusion included before‐and‐after studies, and studies using interrupted time series design. Due to the nature of this area of study, we anticipated that randomised trials would be unlikely, in which case we considered the best available evidence. We considered both published and unpublished studies. However, we excluded publications with no effectiveness data, such as editorials and case reports.

Types of participants

Adults or children of all ethnicities were eligible for inclusion in the review. Participants included individuals exhibiting suicidal behaviour. We also included studies with participants diagnosed with a mental disorder, as well as those in which a diagnosis had not been made prior to suicide or attempted suicide.

Types of interventions

Experimental intervention

We considered studies assessing the effectiveness of interventions to restrict the availability of, or access to, means of suicide by jumping. These include the use of physical barriers, fencing or safety nets at frequently‐used jumping sites, or restriction of access to these sites, such as by way of road closures.

Some or all of these interventions could operate at multiple levels within the universal, selective, and indicated hierarchy as described below.

Universal interventions: targeted at the general public or whole populations.
Selective interventions: targeted at individuals or groups within a population at increased risk of suicidal behaviours
Indicated interventions: targeted at individuals with known suicidal behaviours.

The majority of jumping means restriction interventions operate at the universal level, however, where placements of barriers, nets or fences are in proximity to schools, psychiatric hospitals, and prisons, these interventions could operate at the indicated/selective levels.

We excluded studies assessing the effectiveness of:

interventions aimed at educating professionals or the public about means of suicide;
interventions to restrict cognitive availability of means of suicide, for example the impact of media portrayals;
interventions aimed at improving recognition, screening for risk, treatment or the understanding of causes and risk factors of suicidal behaviour (including mental illness);
interventions solely aimed at increasing opportunities for help‐seeking or third‐party involvement (e.g. studies only assessing the effects of crisis telephones or CCTV cameras on bridges).

However, studies assessing means restriction interventions in combination with other suicide prevention measures such as interventions designed to increase help‐seeking or third‐party involvement or responsible media reporting, were included. We also excluded interventions on trains and railway networks as they are beyond the scope of this review, but will be included in a separate review, as part of our portfolio on means restriction (John 2018).

Comparator intervention

Comparator interventions or control conditions included any other intervention delivered in isolation e.g. crisis phones or CCTV on bridges, or no intervention. We also included studies with head‐ to‐head comparator interventions, e.g. signage on bridges versus structural changes.

Types of outcome measures

Primary outcomes

Suicide
Attempted suicide or self‐harm
Study withdrawal

Secondary outcomes

Change in hospital admission rates for attempted suicide or self‐harm
Cost‐effectiveness of interventions

Timing of outcome assessment

The effects of the interventions are expected to be immediate; however studies have shown that a short duration of follow‐up (less than 2 years) following the introduction of a means restriction intervention may not be adequate to detect a change in the rate of suicides (Hawkins 2007; Hawton 2007). We therefore considered that for a study to be eligible for inclusion in a meta‐analysis, it had to have a postintervention evaluation of at least two years follow‐up for outcome assessment.

Search methods for identification of studies

Electronic searches

We searched the following electronic databases on 31 May 2019 (from the dates of inception).

The Cochrane Library (Issue 5 of 12, May 2019)
Ovid Embase (1974 onwards);
Ovid MEDLINE (1946 onwards);
Ovid PsycINFO (1806 onwards);
Web of Science (Science Citation Index (SCI) (1900 onwards) and Social Science Citation Index (SSCI) (1956 onwards).

The search strategy can be found in Appendix 1, Appendix 2, and Appendix 3.

We did not apply date, language, or publication status restrictions to these searches. In addition, we searched international trial registries via the WHO International Clinical Trials Registry Platform (ICTRP) (www.who.int/ictrp) and ClinicalTrials.gov (clinicaltrials.gov) to identify unpublished and ongoing studies.

Searching other resources

Reference searching

We searched the reference lists of all included studies and relevant papers known to our team, as well as the reference lists of relevant systematic reviews focusing on interventions to reduce suicide at jumping sites (Beautrais 2007; Cox 2013; Pirkis 2013; Pirkis 2015).

Correspondence

We contacted corresponding authors, chief investigators and subject experts for information on unpublished or ongoing studies or to request additional study data.

Data collection and analysis

Selection of studies

A two‐stage screening process was undertaken by two review authors (CO and UK). Firstly, both review authors independently assessed all citations from the searches and identified relevant titles and abstracts. In the second stage, both reviewer authors independently assessed full texts of potentially eligible studies to identify studies to be included in the review. Disagreements at full text stage were resolved through discussion. Where disagreements could not be resolved, a third review author (AJ) was consulted.

Data extraction and management

Three review authors (CO, UK, SW) extracted data from all included studies using a modified version of the Cochrane Public Health Group Data Extraction and Assessment Template. This extraction tool was piloted on a random selection of four included studies prior to its use in the main review. Disagreements were resolved in the first instance through discussion, and where this failed, by referral to a fourth review author (AJ). Where necessary, authors of studies were contacted for clarification and for obtaining missing information or further unpublished data.

Assessment of risk of bias in included studies

Three review authors (CO, UK, SW) independently assessed risk of bias for each included study. We hypothesised that our included studies could consist of both randomised and non‐randomised studies, therefore the assessment criteria were based on Cochrane's 'Risk of bias' assessment tool (Higgins 2011), and the Cochrane ROBINS‐I tool for non‐randomised studies (Sterne 2016a). The ROBINS‐I tool is currently available for cohort study designs but adaptations are being developed for other non‐randomised study types (Sterne 2016b).

For randomised trials, we made a judgement of 'low risk', 'unclear risk', or 'high risk' of bias for the following domains.

Random sequence generation.
Allocation concealment.
Blinding of participants, personnel and outcome assessors.
Incomplete outcome data.
Selective outcome reporting.
Other sources of bias.

For non‐randomised studies, we made a judgement of 'low risk', 'moderate risk', 'serious risk', or 'critical risk' of bias for the following domains.

Confounding.
Selection bias.
Bias in classification of interventions.
Bias due to deviations from intended interventions.
Bias due to missing data.
Bias in measurement of outcomes.
Bias in selection of the reported result.

We critically appraised economic evaluations using the Cochrane recommended British Medical Journal Checklist for authors and peer reviewers of economic submissions (also known as the Drummond checklist; Drummond 1996). This checklist is grouped in 10 sections under three headings: study design; data collection; and analysis and interpretation of results. We made a judgement of 'yes', 'no', or 'not clear' for the following domains.

Study question.
Selection of alternatives.
Form of evaluation.
Effectiveness data.
Benefit measurement and valuation.
Costing.
Modelling.
Adjustments for timing of costs and benefits.
Allowance for uncertainty.
Presentation of results.

Disagreements were resolved in the first instance by discussion and where this failed, by referral to a fourth review author (AJ).

Measures of treatment effect

Continuous data

Similar to previous reviews (Pirkis 2013; Pirkis 2015), we reported continuous data by calculating the pooled incidence rate ratio (IRR) using a random‐effects Poisson regression analysis. Poisson regression was used as the outcome measure, IRR, follows a Poisson distribution, that is, the distribution describing independent, random events in a fixed time interval. In addition, this analysis estimates the change in suicide incidence from pre‐ to postintervention, while also assessing both baseline and between‐study variability in the intervention effect (Spittal 2015). To measure the effect of the interventions, we conducted six meta‐analyses, including: (1) all studies assessing jumping means restriction interventions (delivered in isolation or in combination with other suicide prevention interventions); (2) studies assessing jumping means restriction interventions delivered in isolation; (3) studies assessing jumping means restriction interventions delivered in combination with other suicide prevention interventions; (4) studies assessing the effectiveness of barriers; (5) studies assessing the effectiveness of road closures that limited access to jumping sites; (6) studies assessing the effectiveness of safety nets.

Unit of analysis issues

We did not identify any cluster‐RCTs, cross‐over trials, or studies with multiple intervention/comparator groups in this review. Given the nature of the area of study, it is unlikely there will be any studies with multiple comparators or studies that are not observational in design.

Dealing with missing data

We did not conduct imputation of missing data because of the potential for introduction of bias. Where important data or information about the study design were missing, we contacted investigators or study sponsors in order to verify key study characteristics and obtain missing numerical outcome data, where possible.

Assessment of heterogeneity

We assessed heterogeneity between studies using the I² statistic (Higgins 2011). The I² statistic indicates the percentage of between‐study variation due to chance and can take any value from 0% to 100% (Higgins 2011). Thresholds for interpreting I² are as follows: 0% to 40% ‐ might not be important; 30% to 60% ‐ may represent moderate heterogeneity; 50% to 90% ‐ may represent substantial heterogeneity; 75% to 100% ‐ considerable heterogeneity. The random‐effects Poisson regression analysis employed for this review included an assessment of between‐study heterogeneity (Spittal 2015). Where substantial heterogeneity was detected, possible causes were explored. Where not appropriate, studies were summarised in tables and narrative synthesis conducted.

Assessment of reporting biases

Reporting bias occurs when the decision to publish a research finding is influenced by the direction and significance of its results (Egger 1997). We planned to assess small study effects and potential publication bias using a funnel plot if a meta‐analysis included results of at least 10 studies.

Data synthesis

We performed statistical analyses using statistical software R (version 3.4.3) (R Development Core Team 2017). Where more than one study examined the same intervention, and we judged the study populations and methods as being sufficiently similar, we conducted a meta‐analysis using a random‐effects Poisson regression model to provide an overall estimate of treatment effect. We reported continuous data by calculating the pooled IRR. Where we deemed meta‐analysis inappropriate due to significant heterogeneity, we provided a narrative synthesis of results.

Subgroup analysis and investigation of heterogeneity

Planned subgroup analyses included:

history of self‐harm versus no known history of self‐harm;
diagnosis of mental disorder versus no known history of mental disorder.

Sensitivity analysis

For this review, we planned on analysing the effects of excluding studies that were methodologically different from other studies, studies judged to be at high risk of bias across one or more domain(s), and studies that contributed substantial levels of statistical heterogeneity. If the exclusion of these studies did not substantially alter the direction of effect or the precision of the effect estimates, then we included data from these studies in the analysis.

'Summary of findings' tables

We prepared 'Summary of findings' tables for the main outcome measure, suicide or attempted suicide. We used the GRADE system to assess the quality of evidence for each outcome (Schünemann 2011). Using the four standard GRADE levels of evidence (high, moderate, low and very low) we assigned evidence from RCTs an initial quality rating of high and evidence from observational studies an initial quality rating of low (Balshem 2011). We upgraded or downgraded these levels based on our judgements regarding risk of bias, precision, consistency, indirectness of study results, and publication bias. We interpreted the overall quality rating across studies in line with recommendations by Balshem and colleagues (Balshem 2011).

Results

Description of studies

Results of the search

We identified a total of 5409 citations from database searches. We identified 42 additional citations from other sources. After deduplication, the initial number of citations decreased to 3476. Of these, we excluded 3422 citations after screening of titles and abstracts. We assessed 54 full‐text articles for eligibility. We found 18 articles (representing 14 studies) to be eligible for inclusion in the review. Articles which relied on the same core data were grouped together and not discussed as individual studies. We included 12 studies in the meta‐analysis overall.

We excluded 36 articles (see Characteristics of excluded studies table). The results of all searches are presented in a PRISMA flow diagram (Figure 1; Moher 2010).

Included studies

Fourteen studies met all the inclusion criteria and were included in the review. The details of the individual studies are in the Characteristics of included studies table.

Population

Eight studies reported demographic data on the individuals who died by suicide by jumping from a height. All eight studies reported a male predominance of jumping suicides. The highest number of deaths was seen in men between the ages of 20 and 49 years. In three studies from four articles (Beautrais 2009; Isaac 2005; Skegg 2009), the majority of those who died by jumping had a comorbid psychiatric disorder or were under the care of mental health services at the time of the suicide.

Design

Of the 14 included studies, 13 were before‐and‐after studies. One study from two articles employed an A‐B‐A (reversal) design (Beautrais 2009). The only study not utilising a before‐and‐after study design was a cost‐effectiveness analysis (Whitmer 2013). Due to the nature of this topic, none of the included studies employed a randomised controlled design as these would be difficult to implement.

Setting

Three studies each took place in Switzerland and the USA, while two studies each were from the UK, Canada, New Zealand, and Australia respectively. The majority of included studies evaluated interventions at bridges (n = 7), while the remaining studies evaluated interventions at headlands/ocean cliffs (n = 3), viaducts (n = 1), terraces (n = 1) and high‐rise buildings (n = 1). Seven studies included details of the jump site such as the height of bridge/cliff drop (n = 7), or the type of surface underneath the bridge (n = 4). Hemmer 2017 compared the effectiveness of a range of suicide prevention interventions at 15 jump sites (13 bridges, 1 terrace, and 1 multistorey car park).

Interventions

Ten studies assessed jumping means restriction interventions delivered in isolation. Six of these studies assessed the effectiveness of barriers installed on either bridges (n = 5) or viaducts (n = 1); two studies assessed the effectiveness of restricting road access to headlands/ocean cliffs; one study assessed the effectiveness of installing a safety net at a high terrace; and one study assessed the effectiveness of a guard rail installed at a high‐rise hospital window. In both cases of road closure, the intervention was not primarily aimed at suicide prevention, but rather a coincidence of the need to restrict accessibility to the sites due to construction work and a Foot and Mouth outbreak respectively. Two studies from three articles assessed jumping means restriction interventions delivered in combination with other suicide prevention measures. In Lockley 2014, three types of interventions at an ocean cliff were studied collectively: barriers (means restriction), crisis telephones and signs (measures to increase opportunities for help‐seeking), and CCTV cameras (measures to increase the opportunity for human intervention). Bennewith 2007 examined the effectiveness of barriers (means restriction) in combination with CCTV cameras and regular staff patrols (measures to increase the opportunity for human intervention) on a bridge. No study examined the effectiveness of media reporting in combination with jumping means restriction interventions. Hemmer 2017 compared the effectiveness of different suicide prevention interventions at 15 jumping sites. These interventions included barriers (n = 11) and safety nets (n = 4). In this study, six of the areas were secured by barriers and three of the areas secured by safety nets were additionally equipped with Help signs. This study also included an assessment of the extent of structural interventions, i.e. the effectiveness of structural interventions that secure an entire jump area (complete safety measures) or those that secure only a part of the jump area (incomplete safety measures). Three studies included details on the height of barriers/fences. Whitmer 2013 analysed the cost‐effectiveness of a suicide barrier on a bridge.

Observation periods

All of the included observational studies reported interventions that were assessed at two time points ‐ before and after installation of the respective interventions. Pre‐intervention periods ranged from four years to 22 years (median 10 years), while the postintervention periods ranged from 0.4 years to 22 years (median 5 years). The total pre‐ and postintervention periods were 134.5 years and 92.4 years, respectively (Table 6). In line with empirical evidence on the duration of time required for means restriction interventions to demonstrate impact on suicide rates (Hawkins 2007; Hawton 2007), we considered a postintervention evaluation period of at least two years for study eligibility in a meta‐analysis. Only one study had a postintervention period of less than two years (Isaac 2005; Table 6). This study assessed the effectiveness of restricting road access to a jump site. We excluded data from this study from the meta‐analysis but we have provided a narrative description. Hemmer 2017 compared the effectiveness of different suicide prevention interventions at 15 jump sites and provided individual pre‐ and postintervention observation period data for each location. The mean pre‐ and postintervention observation periods were therefore calculated for this study (14.9 years and 6.1 years respectively). The cost‐effectiveness study did not report pre‐ and postintervention periods (Whitmer 2013), but instead reviewed suicides that occurred over a 70‐year period at one site and calculated what might be the cost per life saved over a 20‐year period by the construction of a proposed bridge barrier.

1. Pre‐ and postintervention suicides and annual average suicides per year (observational studies).

Author and year	Observation period (years)		Total suicides		Suicides per year
	Pre‐intervention	Post‐intervention	Pre‐intervention	Post‐intervention	Pre‐intervention	Post‐intervention
Beautrais 2001/Beautrais 2009	6	4	19	0	3.2	0
Bennewith 2007/Bennewith 2011	5	5	41	20	8.2	4
Hemmer 2017	14.9	6.1	21.8	2.5	1.5	0.4
Isaac 2005	15	0.4	230	0	15.3	0
Law 2014	4	19	22	16	5.5	0.8
Lester 1993/O'Carroll 1994	7	5	25	1	3.6	0.2
Lockley 2014	10.6	2.4	79	16	7.5	6.7
Mohl 2012	9.5	6.5	10	1	1.1	0.2
Pelletier 2007	22	22	14	0	0.6	0
Perron 2013	15.5	5	155	13	10	2.6
Reisch 2005	4	4	8	0	2	0
Sinyor 2010/Sinyor 2017	11	11	104.5	1.1	9.5	0.1
Skegg 2009	10	2	13	0	1.3	0

Total	134.5	92.4	742.3	70.6	5.5	0.8

Open in a new tab

Outcomes

Primary outcome assessment

Primary outcomes initially planned for evaluation in this review were suicide, attempted suicide, self‐harm and study withdrawal rate.

Suicide

Thirteen of the included studies reported number of suicides per year in both the pre‐ and postintervention periods.

Attempted suicide or self‐harm

Only two studies reported data on attempted suicide (Bennewith 2007; Skegg 2009). Whilst Skegg 2009 reported individual data on suicides and suicide attempts, data on suicides were not disaggregated from data on attempted suicide in Bennewith 2007. We therefore contacted the study authors for the individual data on suicide attempts. None of the studies used self‐harm as an outcome.

Study withdrawal

Given the study design of all the included studies, study withdrawal rates were not reported in any of them. This outcome was therefore not assessed in the current review.

Secondary outcome assessment

Secondary outcome measures planned for evaluation in this review included change in hospital admission rates for attempted suicide or self‐harm and cost‐effectiveness of interventions.

Change in hospital admission rates for attempted suicide or self‐harm

None of the included studies reported data on change in hospital admission rates for attempted suicide or self‐harm. Suicide attempts were deemed more relevant to the current review.

Cost‐effectiveness of interventions

Only one study reported cost‐effectiveness data. This study evaluated the cost‐effectiveness of a proposed barrier on a bridge (Whitmer 2013).

Excluded studies

We excluded 36 articles after reviewing full‐texts (see Characteristics of excluded studies). Among these, 35 articles did not report interventions to restrict access to means of suicide by jumping, while one study reported data which was not interpretable as it did not report clearly defined observational periods.

Ongoing studies

We are not aware of any ongoing studies.

Studies awaiting classification

There were no studies awaiting classification.

Risk of bias in included studies

Since none of our included studies employed a randomised controlled design, we used the Cochrane ROBINS‐I tool to assess risk of bias in the 13 included observational studies (Table 7; Table 8; Table 9; Table 10; Table 11; Table 12; Table 13; Table 14; Table 15; Table 16; Table 17; Table 18; Table 19). We assessed the methodological quality of the single cost‐effectiveness analysis study using the 'Drummond checklist' for critically appraising economic evaluations (Table 20).

2. Risk of bias ‐ Beautrais 2001/Beautrais 2009.

Domain	Authors' judgement	Support for judgement
Bias due to confounding	Moderate risk	The study did not take into account the possibility of potential confounding effects as a result of media publicity which could have promoted the jump site
Bias in selection of participants/data	Low risk	The study assessed interventions that were aimed at the general population, therefore all individuals who would have been eligible to take part in the study were included
Bias in classification of interventions	Low risk	Intervention status is well defined
Bias due to deviations from intended interventions	Low risk	There were no deviations from the intended interventions that were likely to impact on the outcome
Bias due to missing data	Moderate risk	The analysis is unlikely to have removed the risk of bias arising from the missing data. Data were unavailable for suicides after 1998
Bias in measurement of outcomes	Low risk	The methods of outcome assessment were comparable across intervention periods
Bias in selection of reported result	Unclear risk	No access to study protocols for this study, therefore it is difficult to assess this domain

Open in a new tab

Low risk of bias: the study is comparable to a well‐performed randomised trial with regard to this domain.  Moderate risk of bias: the study is sound for a non‐randomised study with regard to this domain but cannot be considered comparable to a well‐performed randomised trial.  Unclear risk of bias: too few details are available to make a judgement of ‘critical’, 'serious', 'moderate', or ‘low’ risk

3. Risk of bias ‐ Bennewith 2007/Bennewith 2011.

Domain	Authors' judgement	Support for judgement
Bias due to confounding	Moderate risk	The study did not take into account the possibility of potential confounding effects as a result of media publicity which could have promoted the jump site
Bias in selection of participants/data	Low risk	The study assessed interventions that were aimed at the general population, therefore all individuals who would have been eligible to take part in the study were included
Bias in classification of interventions	Low risk	Intervention status is well defined
Bias due to deviations from intended interventions	Low risk	There were no deviations from the intended interventions that were likely to impact on the outcome
Bias due to missing data	Moderate risk	The analysis is unlikely to have removed the risk of bias arising from the missing data
Bias in measurement of outcomes	Low risk	The methods of outcome assessment were comparable across intervention periods
Bias in selection of reported result	Unclear risk	No access to study protocols for this study therefore it is difficult to assess this domain

Open in a new tab

4. Risk of bias ‐ Hemmer 2017.

Domain	Authors' judgement	Support for judgement
Bias due to confounding	Moderate risk	The study did not take into account the possibility of potential confounding effects as a result of media publicity which could have promoted the jump site
Bias in selection of participants/data	Low risk	The study assessed interventions that were aimed at the general population, therefore all individuals who would have been eligible to take part in the study were included
Bias in classification of interventions	Low risk	Intervention status is well defined
Bias due to deviations from intended interventions	Low risk	There were no deviations from the intended interventions that were likely to impact on the outcome
Bias due to missing data	Low risk	Data were reasonably complete
Bias in measurement of outcomes	Low risk	The methods of outcome assessment were comparable across intervention periods
Bias in selection of reported result	Unclear risk	No access to study protocols for this study therefore it is difficult to assess this domain

Open in a new tab

5. Risk of bias ‐ Isaac 2005.

Domain	Authors' judgement	Support for judgement
Bias due to confounding	Moderate risk	The study did not take into account the possibility of potential confounding effects as a result of media publicity which could have promoted the jump site
Bias in selection of participants/data	Low risk	The study assessed interventions that were aimed at the general population, therefore all individuals who would have been eligible to take part in the study were included
Bias in classification of interventions	Low risk	Intervention status is well defined
Bias due to deviations from intended interventions	Low risk	There were no deviations from the intended interventions that were likely to impact on the outcome
Bias due to missing data	Low risk	Data were reasonably complete
Bias in measurement of outcomes	Low risk	The methods of outcome assessment were comparable across intervention periods
Bias in selection of reported result	Unclear risk	No access to study protocols for this study therefore it is difficult to assess this domain

Open in a new tab

6. Risk of bias ‐ Law 2014.

Domain	Authors' judgement	Support for judgement
Bias due to confounding	Moderate risk	The study did not take into account the possibility of potential confounding effects as a result of media publicity which could have promoted the jump site
Bias in selection of participants/data	Low risk	The study assessed interventions that were aimed at the general population, therefore all individuals who would have been eligible to take part in the study were included
Bias in classification of interventions	Low risk	Intervention status is well defined
Bias due to deviations from intended interventions	Low risk	There were no deviations from the intended interventions that were likely to impact on the outcome
Bias due to missing data	Moderate risk	Data prior to 1990 were not available
Bias in measurement of outcomes	Low risk	The methods of outcome assessment were comparable across intervention periods
Bias in selection of reported result	Unclear risk	No access to study protocols for this study therefore it is difficult to assess this domain

Open in a new tab

7. Risk of bias ‐ Lester 1993/O'Carroll 1994.

Domain	Authors' judgement	Support for judgement
Bias due to confounding	Moderate risk	The study did not take into account the possibility of potential confounding effects as a result of media publicity which could have promoted the jump site
Bias in selection of participants/data	Low risk	The study assessed interventions that were aimed at the general population, therefore all individuals who would have been eligible to take part in the study were included
Bias in classification of interventions	Low risk	Intervention status is well defined
Bias due to deviations from intended interventions	Low risk	There were no deviations from the intended interventions that were likely to impact on the outcome
Bias due to missing data	Low risk	Data were reasonably complete
Bias in measurement of outcomes	Low risk	The methods of outcome assessment were comparable across intervention periods
Bias in selection of reported result	Unclear risk	No access to study protocols for this study therefore it is difficult to assess this domain

Open in a new tab

8. Risk of bias ‐ Lockley 2014.

Domain	Authors' judgement	Support for judgement
Bias due to confounding	Moderate risk	The study did not take into account the possibility of potential confounding effects as a result of media publicity which could have promoted the jump site
Bias in selection of participants/data	Low risk	The study assessed interventions that were aimed at the general population, therefore all individuals who would have been eligible to take part in the study were included
Bias in classification of interventions	Low risk	Intervention status is well defined
Bias due to deviations from intended interventions	Low risk	There were no deviations from the intended interventions that were likely to impact on the outcome
Bias due to missing data	Moderate risk	Some of the quantitative data sets were limited
Bias in measurement of outcomes	Low risk	The methods of outcome assessment were comparable across intervention periods
Bias in selection of reported result	Unclear risk	No access to study protocols for this study therefore it is difficult to assess this domain

Open in a new tab

9. Risk of bias ‐ Mohl 2012.

Domain	Authors' judgement	Support for judgement
Bias due to confounding	Moderate risk	The study did not take into account the possibility of potential confounding effects as a result of media publicity which could have promoted the jump site
Bias in selection of participants/data	Low risk	The study assessed interventions that were aimed at the general population, therefore all individuals who would have been eligible to take part in the study were included
Bias in classification of interventions	Low risk	Intervention status is well defined
Bias due to deviations from intended interventions	Low risk	There were no deviations from the intended interventions that were likely to impact on the outcome
Bias due to missing data	Low risk	Data were reasonably complete
Bias in measurement of outcomes	Low risk	The methods of outcome assessment were comparable across intervention periods
Bias in selection of reported result	Unclear risk	No access to study protocols for this study therefore it is difficult to assess this domain

Open in a new tab

10. Risk of bias ‐ Pelletier 2007.

Domain	Authors' judgement	Support for judgement
Bias due to confounding	Moderate risk	The study did not take into account the possibility of potential confounding effects as a result of media publicity which could have promoted the jump site
Bias in selection of participants/data	Low risk	The study assessed interventions that were aimed at the general population, therefore all individuals who would have been eligible to take part in the study were included
Bias in classification of interventions	Moderate risk	The authors stated that electronic death records between 1960 and 1974 did not meet current data quality standards, leading to a potential underestimation of suicides during this period of time. We judged that if issues with data acquisition led to a potential underestimation of suicides during the time period, then it would most likely have been problems with classification or suicide ascertainment
Bias due to deviations from intended interventions	Low risk	There were no deviations from the intended interventions that were likely to impact on the outcome
Bias due to missing data	Moderate risk	Data from 1968 were not available
Bias in measurement of outcomes	Low risk	The methods of outcome assessment were comparable across intervention periods
Bias in selection of reported result	Unclear risk	No access to study protocols for this study therefore it is difficult to assess this domain

Open in a new tab

11. Risk of bias ‐ Perron 2013.

Domain	Authors' judgement	Support for judgement
Bias due to confounding	Moderate risk	The study did not take into account the possibility of potential confounding effects as a result of media publicity which could have promoted the jump site
Bias in selection of participants/data	Low risk	The study assessed interventions that were aimed at the general population, therefore all individuals who would have been eligible to take part in the study were included
Bias in classification of interventions	Low risk	Intervention status is well defined
Bias due to deviations from intended interventions	Low risk	There were no deviations from the intended interventions that were likely to impact on the outcome
Bias due to missing data	Moderate risk	Suicides occurring between July 1 2004 and December 31 2004 were excluded from analysis
Bias in measurement of outcomes	Low risk	The methods of outcome assessment were comparable across intervention periods
Bias in selection of reported result	Unclear risk	No access to study protocols for this study therefore it is difficult to assess this domain

Open in a new tab

12. Risk of bias ‐ Reisch 2005.

Domain	Authors' judgement	Support for judgement
Bias due to confounding	Moderate risk	The study did not take into account the possibility of potential confounding effects as a result of media publicity which could have promoted the jump site
Bias in selection of participants/data	Low risk	The study assessed interventions that were aimed at the general population, therefore all individuals who would have been eligible to take part in the study were included
Bias in classification of interventions	Low risk	Intervention status is well defined
Bias due to deviations from intended interventions	Low risk	There were no deviations from the intended interventions that were likely to impact on the outcome
Bias due to missing data	Low risk	Data were reasonably complete
Bias in measurement of outcomes	Low risk	The methods of outcome assessment were comparable across intervention periods
Bias in selection of reported result	Unclear risk	No access to study protocols for this study therefore it is difficult to assess this domain

Open in a new tab

13. Risk of bias ‐ Sinyor 2010/Sinyor 2017.

Domain	Authors' judgement	Support for judgement
Bias due to confounding	Low risk	The study took into account all potential confounding effects
Bias in selection of participants/data	Low risk	The study assessed interventions that were aimed at the general population, therefore all individuals who would have been eligible to take part in the study were included
Bias in classification of interventions	Low risk	Intervention status is well defined
Bias due to deviations from intended interventions	Low risk	There were no deviations from the intended interventions that were likely to impact on the outcome
Bias due to missing data	Low risk	Data were reasonably complete
Bias in measurement of outcomes	Low risk	The methods of outcome assessment were comparable across intervention periods
Bias in selection of reported result	Unclear risk	No access to study protocols for this study therefore it is difficult to assess this domain

Open in a new tab

14. Risk of bias ‐ Skegg 2009.

Domain	Authors' judgement	Support for judgement
Bias due to confounding	Moderate risk	The study did not take into account the possibility of potential confounding effects as a result of media publicity which could have promoted the jump site
Bias in selection of participants/data	Low risk	The study assessed interventions that were aimed at the general population, therefore all individuals who would have been eligible to take part in the study were included
Bias in classification of interventions	Low risk	Intervention status is well defined
Bias due to deviations from intended interventions	Low risk	There were no deviations from the intended interventions that were likely to impact on the outcome
Bias due to missing data	Low risk	Data were reasonably complete
Bias in measurement of outcomes	Low risk	The methods of outcome assessment were comparable across intervention periods
Bias in selection of reported result	Unclear risk	No access to study protocols for this study therefore it is difficult to assess this domain

Open in a new tab

15. Quality assessment ‐ Whitmer 2013.

Domain	Authors' judgement
Study design	Yes
Selection of alternatives	No
Form of evaluation	Yes
Effectiveness data	Not clear
Benefit measurement and valuation	Yes
Costing	Yes
Modelling	Not clear
Adjustments for timing of costs and benefits	Yes
Allowance for uncertainty	Not clear
Presentation of results	Yes

Open in a new tab

Observational studies

Confounding

Baseline confounding occurs when one or more prognostic variables also predicts the intervention received at baseline (Sterne 2016a). We only judged one study (from 2 articles) to be at low risk of bias due to confounding (Sinyor 2010). We judged the remaining included observational studies (n = 12) to be at moderate risk of bias due to confounding. In these studies, publicity (of the intervention/jump site) by media reports could have had a potentially confounding effect. This was not adjusted for in the analyses.

Selection bias

Selection bias occurs when exclusion of some eligible participants, or the initial follow‐up time of some participants, or some outcome events, is related to both intervention and outcome (Sterne 2016a). Our included studies assessed interventions that were aimed at the general population; therefore all individuals who would have been eligible to take part in the study were included. We judged all 13 included observational studies to be at low risk of selection bias.

Bias in classification of interventions

Bias in classification of interventions occurs when there is either differential or non‐differential misclassification of intervention status (Sterne 2016a). We judged 12 of the 13 included observational studies to be at low risk of bias in classification of interventions. We judged the remaining study to be at moderate risk of bias because electronic death records used in the study between 1960 and 1974 did not meet current data quality standards (Pelletier 2007), thereby leading to a potential underestimation of suicides during this time period.

Bias due to deviations from intended interventions

Bias due to deviations from intended interventions arises when there are systematic differences between experimental intervention and comparator groups in the care provided (Sterne 2016a). Since our included studies are observational in design – with no comparator groups, assessment of this domain was based on the effect of starting and adhering to the intervention. We judged all 13 included observational studies to be at low risk of bias in this domain as there were no deviations from the intended interventions.

Bias due to missing data

Bias due to missing data arises when follow‐up data are missing for individuals initially included in a study or when individuals with missing information about intervention status or other variables, are excluded (Sterne 2016a). We judged 7 of the 13 included observational studies to be at low risk of bias due to missing data. We judged the remaining six studies to be at moderate risk of bias, mainly due to the unavailability of suicide data at certain time periods.

Bias in measurement of outcomes

Bias in measurement of outcomes occurs when there is either differential or non‐differential errors in measurement of outcome data (Sterne 2016a). We judged all 13 included observational studies to be at low risk of bias in this domain because the same methods were used to assess outcomes.

Bias in selection of reported results

Selective reporting arises when the reporting of results depends on the findings and prevents the estimate from being included in a meta‐analysis (or other synthesis) (Sterne 2016a). We did not have access to study protocols for the studies included in this review, and as such it is difficult to assess this domain. We therefore gave a rating of ‘unclear risk’ for this domain to the 13 included observational studies.

Cost‐effectiveness analysis study

For the evaluation of the cost‐effectiveness analysis study (Table 20), we made a judgement of 'yes' for the following domains: study design; form of evaluation; benefit measurement and valuation; costing; adjustments for timing of costs and benefits; and presentation of results. We made a judgment of 'not clear' for effectiveness data, modelling, and allowance for uncertainty. The only domain with a judgement of 'no' was selection of alternatives, as this was not described.

Effects of interventions

See: Table 1; Table 2; Table 3; Table 4; Table 5