Summary of findings
Among primary care patients in CPRD, we found a 20-fold increase in risk of suicide in those diagnosed with PD compared to those with no mental disorder, and an over four-fold increase in risk when compared with patients with other mental disorder diagnoses. Females with PD had a significantly greater elevation in suicide risk than male patients with PD, compared to their gender and age-matched peers without any mental illness diagnoses. Significant risks associated with PD diagnosis were identified across all age ranges, although the greatest elevations were in the younger age ranges, 16–39 years. A diagnosis of borderline PD doubled the risk of suicide when compared to patients diagnosed with other types of PD. Clinically significant comorbid alcohol misuse significantly increased the risk of suicide in those with a diagnosis of PD.
Comparison with previous literature
To our knowledge this is the first epidemiological study to examine suicide risk in a representative primary care sample of UK patients with PD, making it difficult to compare with other studies. However, our main findings are generally in line with published findings, except for the much larger effect sizes we found. For example, Schneider et al. [
5] reported that PD was associated with a seven-fold increase in suicide risk compared with those in the general population with no mental illnesses [
5]. Risks were particularly raised amongst those with a diagnosis of Cluster B personality disorders (which includes borderline type), with an odds ratio of 7.5 compared with 4.5 for Cluster A and 4.1 for Cluster C. In our study, a diagnosis of PD was associated with a 20-fold elevated risk compared with those with no mental disorders while borderline PD was linked with double the risk versus other types of PD. Schneider et al. [
5] also found that being diagnosed with comorbid PD and alcohol use disorders elevated the risk 15-fold compared with those with no psychiatric diagnosis, while we found a 45-fold increase. This finding is particularly alarming, suggesting that comorbid PD and alcohol misuse represents a potent determinant of elevated suicide risk in community settings.
In a national Danish cohort study of psychiatric inpatients, Hiroeh et al. [
7] reported that PD was associated with a 16-fold increase in suicide risk in women (standardised mortality ratio, SMR = 1568) and 12-fold in men (SMR = 1198) compared with people who had not been admitted for psychiatric treatment [
7]. Similar patterns were found in our primary care-based study, although the effect sizes were larger still (ORs = 38 for women and 16 for men). The greater relative risk found for women may be strongly linked with the high self-harm rate in women with PD [
28].
Strengths and limitations
The CPRD uniquely provided a large, detailed, and nationally representative computerised cohort of primary care-treated patients, including those with mental illness and comorbid alcohol misuse, with complete linkage and follow-up of cause-specific mortality. Using a nested case-control design to capture all cases in the cohort at risk, and with random sampling of living matched controls from the cohort, examining risk factors for rare adverse events like suicide was possible. Biases that commonly flaw epidemiological studies can also be minimised. For example, information bias was precluded because the data were collected prospectively without knowledge of the subsequent outcome. Selection into primary care was an unlikely source of major bias because almost all UK residents are registered with a GP soon after they are born, and when people move address and register with a new practice their complete primary health care record is transferred automatically. Thus, the CPRD yields a representative community-based epidemiological sample.
The major limitation of our study was the potential for incomplete ascertainment in the CPRD of all patients diagnosed with PD, all other types of mental illness, and comorbid alcohol misuse. This is a widely recognised issue for all CPRD-based studies [
17]. A second limitation concerns diagnostic validity and reliability in the Read codes that were recorded in the CPRD. All information recorded in the Datalink can be linked to a health professional within the general practice, but the link does not necessarily reflect the true diagnostic path, especially regarding information from secondary care (for example, hospital letters inputted by administrative staff). However, our study benefited from the lead author (MD) and two of the co-authors (JS and LA) having extensive clinical experience in the assessment and management of PD, thus minimising the risk of misclassification. Although we did not know how or by whom PD was diagnosed, we applied a stringent set of diagnostic codes to identify people with PD to ensure we had a clinically relevant sample. For PD to be recorded in primary care notes, it may have been diagnosed in specialist care - it is not a diagnosis commonly made by GPs and attributed to patients; thus, symptoms may be severe. Therefore, the high risk identified in the study reflects clinically evident PD rather than all patients with PD traits. Identification of ‘cases’ of clinically significant alcohol misuse was also challenging, because it is not consistently and systematically recorded in the CPRD. However, as described in the Methods section, we used a robust approach to identify alcohol misuse deemed to be at a clinically significant level. Misclassification that occurs when using the CPRD for epidemiological purposes may have attenuated the observed odds ratios conservatively towards unity because they were distributed non-differentially between suicide cases and controls [
29]. A third important limitation was that we could not estimate incidence rates or absolute risk of suicide from the nested case-control design. This is true of all case-control studies, although the ‘nested’ variant with random sampling of controls from the cohort at risk is a robust statistically efficient design for examining rare adverse events such as suicide. Two final limitations of this CPRD-based study include inability to stratify PD and other mental disorders according to their degree of severity, and to examine suicide risk in patients with more than one diagnosed PD.
Interpretation of the findings and their implications
The findings from this study highlight the importance of considering age, gender and concurrent alcohol misuse when assessing suicide risk among primary care patients diagnosed with PD. With increasing age, certain risky behaviours associated with PD and suicide, such as impulsivity and aggression, may become less common, contributing to the lower elevated suicide risks amongst those from the older age groups [
30]. However, comorbid psychiatric disorders may enhance the expression of other dysfunctional traits such as dementia and social withdrawal in older people making it more complex to diagnose PD in this group [
30,
31]. The diagnostic criteria for PD and in particular borderline PD - including repeated non-fatal self-harm, impulsivity, affective instability, and mood disorder - have consistently been found to be strongly linked to suicidal behaviour [
2,
4,
32,
33]. Self-harming behaviour among people with a diagnosis of PD, but especially borderline PD, tends to be ambivalent in intent where emotional instability and impulsivity are very common. This is complicated further as often the level of intent of self-harm can fluctuate rapidly and trying to accurately predict a rare event such as suicide is seldom possible. Therefore, enhancing the skills of GPs, through training opportunities, to identify and assess PD and personality traits factors could help in reducing the risk of suicide; this would need to take account of the high false positive rate amongst those who repeatedly self-harm. A standardised assessment tools such as the Standardised Assessment of Personality [
34] may be helpful as a first-stage screen as part of a two-stage procedure for case identification [
35], which could be used regardless of the age or gender of patients and could be integrated into a primary care consultation.
Similar to mental health services, it is possible that the elevated risk of suicide in primary care is related to a paucity of effective interventions to manage the risk. There is a lack of consensus about the extent to which people with PD are amenable to therapeutic interventions [
36] and primary health care staff may view people with PD as ‘untreatable’. A sense of futility combined with a lack of therapeutic skills could be exacerbating an already elevated risk among these patients. Closely linked to perceived ‘untreatabilty’ is the tendency for patients with PD to be viewed negatively by treating psychiatrists and other clinicians due to their antisocial, disruptive and treatment resistant symptoms [
37], especially if they self-harm [
38]. There is also some evidence that those with a diagnosis of PD are less likely than those without to be referred by GPs to specialist psychiatric services [
14] and GPs have been shown to view those with a PD as ‘less compliant’, ‘less likeable’ and ‘more stressful’ to deal with [
14]. This may be further exacerbated by time-limited consultations.
Nevertheless, GPs remain in a unique position to assess the risk of suicide in people with PD and to intervene. This would mean working with specialist care, agreed clinical pathways and availability of services for comorbidities such as alcohol misuse, as well as opportunities for GPs to develop specific clinical skills. Presence of PD traits, alcohol misuse and suicidal behaviour and frequent attendance at clinic could be the basis for a ‘flag’ alert in primary care records for treatment and/or referral [
16]. Given that a diagnosis of alcohol use disorders is associated with a much heightened risk of suicide in patients with PD, detection and treatment for alcohol misuse should be a priority [
39]. In addition, since impulsivity is a trait that is common among people diagnosed with PD, substance use and suicidal behaviour [
32,
40] treatments should also target difficulties in impulse control.
Given the strong link found in this study between PD and suicide in primary care, and the paucity of robust evidence in this setting, there is a clear need for further research. Attempts should be made to try and replicate the findings from this study in different primary care samples from other countries. From a clinical perspective, the way that GPs assess, support and manage people with PD should be further explored, both quantitatively and qualitatively. A review of the links between primary and specialist mental health services would be useful to establish the mutual expectations of the referrer and referee and to evaluate treatment options available [
41]. The role of specialist mental health services and the interface with primary care services in managing the needs and risks among people with PD also requires evaluation.