Population‐based analysis of health care contacts among suicide decedents: identifying opportunities for more targeted suicide prevention strategies

Abstract

The objective of this study was to detail the nature and correlates of mental health and non‐mental health care contacts prior to suicide death. We conducted a systematic extraction of data from records at the Office of the Chief Coroner of Ontario of each person who died by suicide in the city of Toronto from 1998 to 2011. Data on 2,835 suicide deaths were linked with provincial health administrative data to identify health care contacts during the 12 months prior to suicide. Sub‐populations of suicide decedents based on the presence and type of mental health care contact were described and compared across socio‐demographic, clinical and suicide‐specific variables. Time periods from last mental health contact to date of death were calculated and a Cox proportional hazards model examined covariates. Among suicide decedents, 91.7% had some type of past‐year health care contact prior to death, 66.4% had a mental health care contact, and 25.3% had only non‐mental health contacts. The most common type of mental health contact was an outpatient primary care visit (54.0%), followed by an outpatient psychiatric visit (39.8%), an emergency department visit (31.1%), and a psychiatric hospitalization (21.0%). The median time from last mental health contact to death was 18 days (interquartile range 5‐63). Mental health contact was significantly associated with female gender, age 25‐64, absence of a psychosocial stressor, diagnosis of schizophrenia or bipolar disorder, past suicide attempt, self‐poisoning method and absence of a suicide note. Significant differences between sub‐populations of suicide decedents based on the presence and nature of their health care contacts suggest the need for targeting of community and clinical‐based suicide prevention strategies. The predominance of ambulatory mental health care contacts, often close to the time of death, reinforce the importance of concentrating efforts on embedding risk assessment and care pathways into all routine primary and specialty clinical care, and not only acute care settings.

At least 800,000 people worldwide die from suicide each year, at an estimated rate of 11.4 per 100,000 per year1. Suicide results in devastating personal and societal loss, with immense emotional and economic costs2. There is no singular profile of a person who dies by suicide, and many interconnected factors may lead to this tragic outcome, but mental illness is often at the core, being present among >90% of cases3, 4. The absolute risk of suicide in people with a lifetime contact with specialty mental health services is estimated at 6‐7% among men and 4‐5% among women5.

Within the health care system, there are a variety of different potential points of contact prior to suicide death. Mental health care may be accessed through primary care, ambulatory psychiatric services, emergency departments or inpatient care settings6, 7, 8, 9. Extant data suggest that <50% of people who die by suicide have mental health care contact during the year prior to death10, 11, 12, 13, 14, 15, with specialized mental health services being the most common access point among those with mental health care contact8, 14, 16.

There is a paucity of comprehensive data on factors associated with any mental health contact or with specific mental health contact types. Furthermore, while the most common type of any health care contact during the year prior to suicide death is with primary care physicians16, 17, 18, 19, only a portion of visits involve an assessment of the patients’ safety or have a documented mental health focus16, 20, 21, and there are limited data examining the role of primary care providers in the care of individuals at risk of suicide8, 14. Women and older adults have been reported to be more likely to have had contacts with a primary care physician or mental health services prior to suicide, but most other possible demographic and clinical factors have not received sufficient study8, 14, 22.

There are a large group of people who die by suicide without having any recent mental health care contact. This group is even less well understood, with available data describing and comparing contact and non‐contact groups limited by small sample size and non‐representative samples23, 24. By definition, studies of this group must rely on population‐based datasets for identification of suicide decedents, since health care administrative databases such as those from health maintenance organizations or other similar sources are not designed to sufficiently capture data on people not in treatment.

Population‐based groups that differ based on the presence, type, frequency and recency of health care contacts prior to suicide represent potentially different suicide risk populations that would require different suicide prevention strategies25, 26, 27. A better understanding of the socio‐demographic, clinical and suicide‐related differences between these groups can serve to inform the implementation and evaluation of targeted suicide prevention strategies3, 28, 29.

We examined a large, population‐based sample of suicide decedents in order to detail the nature and correlates of mental health and non‐mental health care contacts prior to suicide death. We sought to address a number of limitations in the literature by utilizing population‐based data sources that combined detailed person‐related information on suicide decedents with health care administrative data that captured all types of mental health and non‐mental health care contacts (and where there is no contact) within primary and specialty care.

Methods

Data from the Office of the Chief Coroner of Ontario

The Office of the Chief Coroner of Ontario (OCC) investigates all suicide deaths in Toronto, Canada. We conducted a systematic extraction of data from records at the OCC of each person who died by suicide in the city of Toronto (approximate population 2.5 million) from 1998 to 2011 (3,091 suicide deaths).

The OCC becomes involved in all sudden or unexpected deaths, and conducts an investigation to determine the cause, which can include suicide, according to a standard of a high degree of probability. Coroner charts compile all relevant information into an investigation report and search for confluence from multiple, independent sources, including a police report, evidence from the scene (e.g., a suicide note), death certificate, post‐mortem examination (pathology report), toxicology report (for self‐poisoning deaths only), collateral information gathered from interviews with family or others, physician/clinical records, and in some cases a full inquest. OCC data are not available for approximately two years after the death, while full investigations are completed.

A standardized data extraction procedure was used, collecting data on: a) socio‐demographics, including age, gender, marital status, and living circumstances; b) recent stressors, including employment/financial, interpersonal conflict, relationship breakup, immigration, medical/health, police/legal, and bereavement; c) clinical variables, including diagnosis of bipolar disorder or schizophrenia, past suicide attempts, and presence of any comorbid medical condition; and d) suicide details, including method, place of death and presence of a suicide note. More than one method of suicide could be recorded, if there was more than one independent cause of death identified (e.g., self‐poisoning and asphyxia). Details regarding intent of prior suicidal behavior were not systematically available in coroner data; therefore we had to rely on the coroner determination of a past suicide attempt without clarity as to whether the behavior should have been better characterized within the notion of deliberate self‐harm, non‐suicidal self‐injury, or other descriptors of suicidal behavior.

Two study investigators (MS and AS) provided onsite training to research staff and were in continuous contact to address any questions and reach consensus regarding coding for more complex cases. Socio‐demographic data and details of the suicide were available in >99% of investigation reports. Information on clinical and stressor variables are primarily collected by the coroner to aid the investigation of the cause and details of the death; as such these variables are subject to missing information which may or may not be associated with prior mental health contact. We included them in the analyses, but interpret the results cautiously. Previous studies using coroner data on bipolar disorder or schizophrenia30, 31 have suggested adequate reliability of diagnosis, but we chose to not include other diagnoses (e.g., depression, anxiety, personality disorders) because of reliability concerns as a result of non‐specific information in the coroner records.

Health administrative data

Provincial health administrative data maintained at the Institute for Clinical Evaluative Sciences (ICES) provide accurate and complete information on residents of Ontario (except the prison population and Aboriginal residents on reserve) and their contact with physicians in the health care system, including outpatient physician visits, emergency department visits, and inpatient hospitalizations.

Datasets that were accessed for this study included: a) registered persons database for basic personal information; b) Ontario Health Insurance Plan for physician visits and billing codes (including specific mental health codes for primary care) and for emergency department visits prior to 2002; c) National Ambulatory Care Reporting System for emergency department visits since 2002; d) Canadian Institute for Health Information – Discharge Abstract Database for hospitalizations with a mental health primary discharge diagnosis; and e) Ontario Mental Health Reporting System for psychiatric hospitalizations subsequent to October 2005. These datasets allowed for a one‐year look back for all years in the study.

A number of mental health and non‐mental health related service contacts were defined a priori. Mental health related primary care physician visits were defined as any Ontario Health Insurance Plan claim with a mental health/addiction diagnostic code, or a fee code for primary mental health care or psychotherapy made by a physician designated as a family physician in the ICES physician database. This definition utilized an existing algorithm validated at ICES32, which results in 96.1% sensitivity and 93.1% specificity when compared with chart abstracted data. All other primary care physician Ontario Health Insurance Plan claims were defined as non‐mental health related. An outpatient psychiatric contact was defined by a standard outpatient Ontario Health Insurance Plan claim made by a psychiatrist (defined by ICES physician database main specialty).

A mental health related emergency department visit was defined in one of three ways: a) prior to 2002, as an Ontario Health Insurance Plan claim with a mental health/addiction diagnostic code or mental health service code billed in the emergency department setting; b) since 2002, within National Ambulatory Care Reporting System, as an emergency department visit with a mental health/addiction diagnostic code (ICD‐9 codes 290‐319 or ICD‐10 codes F00‐F99); c) since 2002, as an emergency department visit that involved suicide‐related behavior based on National Ambulatory Care Reporting System coding of self‐inflicted poisoning or injury (ICD‐9 codes E950‐9 or ICD‐10 codes X60‐X84)33, 34. All other emergency department visits that did not meet the above criteria were defined as non‐mental health related.

Mental health related hospitalization was defined as any Canadian Institute for Health Information – Discharge Abstract Database record with a mental health/addiction diagnostic code (ICD‐9 codes 290‐319 or ICD‐10 codes F00‐F99), or any Ontario Mental Health Reporting System discharge record after October 2005. All other hospitalizations were defined as non‐mental health related.

Linking data from the OCC and ICES

These datasets were linked using unique, encoded identifiers and analyzed at ICES. For each person who died by suicide, we attempted to link via probabilistic matching of name, gender, date of birth, date of death, and home postal code (first three characters) since the OCC charts do not contain health card numbers that are the basis for identification within health administrative data. Unsuccessful linkage occurs when there is insufficient information on the key linkage variables from either dataset to establish a definitive match.

Successful linkage was obtained for 94.7% of cases. The linkage rate showed a weak increasing trend over years (R²=0.335, p=0.031), with a range from 88.7% to 97.7%. A further 3.0% of cases were excluded from analysis after quality check because of discrepancies between ICES and coroner data on age (by >3 years), gender, or date of death (by >2 days), or because the date of death was prior to a recorded health care contact. This resulted in an analyzable sample of 2,835 suicide deaths (91.7% of total suicide deaths).

A comparison of the analysis and excluded groups was conducted. Decedents under age 25 years (p=0.031), and those with an immigration stressor (p≤0.001), a bipolar disorder diagnosis (p=0.046), a past suicide attempt (p=0.002) and a medical diagnosis (p≤0.001) were significantly more likely to be excluded.

Mental health and non‐mental health care contacts

A mental health care contact in the 12 months prior to the date of suicide was defined as one or more of the following: a) a mental health related primary care outpatient physician contact; b) an outpatient psychiatric contact; c) a mental health related emergency department visit; or d) a mental health related hospitalization. Each of these mental health contacts was also examined separately.

A non‐mental health contact was defined as one or more of the following: a) a non‐mental health related primary outpatient physician contact; b) a contact with an ICES physician database‐defined specialty other than primary care or psychiatry; c) a non‐mental health related emergency department visit; or d) a non‐mental health related hospitalization. Emergency department visits or hospitalizations that included a component of both non‐mental health and mental health care were considered to be mental health related.

Statistical analysis

Among the analyzable sample of 2,835 suicide deaths, the proportion of subjects with a mental health care contact, only a non‐mental health care contact, or neither type of contact within the 12 months prior to suicide was described. Bivariate analyses compared socio‐demographics, clinical variables, recent stressors and suicide details between subjects with any mental health contact, only non‐mental health contact and no contact.

Multivariate logistic regression for any past‐year mental health contact was then conducted using generalized estimating equation models to test associations of any mental health contact in the year prior to suicide. Variables tested included age, gender, marital status, living circumstances, recent and past suicide attempt, diagnosis of bipolar disorder, diagnosis of schizophrenia, recent medical diagnosis, method of death, place of death, and presence of a suicide note. Models were run with and without year of death to test for a secular trend in the results, and crude and adjusted odds ratios were obtained for the independent variables.

The proportion of subjects with each specific type of mental health care contact within the past 12 months prior to suicide was also described. Mental health contacts were then categorized as either acute (emergency department visit or hospitalization) or ambulatory (outpatient psychiatric or primary care visits), and bivariate analyses compared subjects with any acute mental health care versus those with only ambulatory mental health care. Further secondary analyses compared four subgroups of subjects who had: a) a mental health inpatient or emergency department visit; b) mental health emergency department visit(s), but no hospitalization; c) outpatient psychiatric or mental health primary care visits, but no acute care; or d) only outpatient mental health primary care visits.

For those with any prior mental health care contact, mean, median and categorical time periods from last mental health contact to date of death were calculated for any and each type of mental health contact. A time‐to‐event curve was generated for any and each type of mental health contact. We structured this similarly to data by Ahmedani et al8 to facilitate comparison across datasets.

We then examined time from last mental health contact (any type) to suicide death among subjects with at least one contact. The relationship between baseline covariates and the time from last mental health contact of any type until suicide was modeled using a Cox proportional hazards model to obtain both crude and adjusted hazard ratios. The proportional hazards assumption was tested, and time dependent variables were added to the basic model for medical diagnosis, diagnosis of schizophrenia, living circumstances and method of death. Both adjusted and unadjusted hazard ratios were obtained for the baseline covariates. This model was also run with year of death as a covariate.

For those with prior contact, the frequency of each type of mental health care contact was also reported as mean, median and range.

Ethical approval and privacy

The OCC granted approval to this study and provided full access to their records for the purposes of completing this study. The study was approved by the Research Ethics Board at Sunnybrook Health Sciences Centre, Toronto, Canada. Strict privacy procedures utilized by the OCC and ICES were fully adhered to.

RESULTS

Of the 2,835 people who died by suicide, a total of 91.7% had some type of health care contact during the year prior to suicide death. The majority (66.4%) had a mental health contact, with 25.3% having only non‐mental health related contacts.

Table 1 summarizes the bivariate analyses across three groups based on contact during the year prior to suicide death: a) subjects with a mental health contact; b) subjects with only a non‐mental health contact; and c) subjects with no mental health or non‐mental health contact.

Table 1.

Comparison of people who died by suicide by type of health care contact in prior year in Toronto, Canada, 1998‐2011

	Any mental health contact (N=1883)	Only non‐mental health contact (N=716)	No contact (N=236)	Test value (F/χ2)	df	p
Socio‐demographics
Age (mean, years± SD)	47.0 ± 16.5	50.4 ± 20.5	41.0 ± 14.7	27.19	2	<0.0001
Age (%)
≤ 24 years	7.4	12.4	14.8	109.23	4	<0.0001
25‐64 years	77.4	60.6	80.5
≥ 65 years	15.2	27.0	4.7
Gender (% male)	64.9	79.6	84.7	79.12	2	<0.0001
Marital status (%)
Single/no status available	54.6	49.6	65.3	25.46	4	<0.0001
Divorced, separated or widowed	20.8	19.3	17.4
Married, including common law	24.5	31.1	17.4
Living circumstances (%)
Alone	44.2	37.8	48.3	22.54	4	0.0002
Family/friends	49.1	58.1	44.9
Other	6.6	4.1	6.8
Recent stressors
Bereavement (%)	5.6	5.3	6.8	0.74	2	0.6915
Employment/financial (%)	15.7	21.4	31.4	39.51	2	<0.0001
Relationship (%)	8.2	8.9	9.7	0.9	2	0.6382
Interpersonal conflict (%)	16.1	18.6	17.8	2.33	2	0.3116
Medical health (%)	10.4	20.7	4.7	64.39	2	<0.0001
Police/legal (%)	6.8	6.6	7.6	0.34	2	0.8454
Immigration (%)	1.0	0.8	x	3.12	2	0.2097
Any stressor present (%)	46.5	61.9	59.3	55.3	2	<0.0001
Clinical variables
Bipolar disorder diagnosis (%)	8.6	1.4	x	55.28	2	<0.0001
Schizophrenia diagnosis (%)	9.7	1.1	x	68.28	2	<0.0001
Past suicide attempt (%)	36.6	12.2	8.9	201.03	2	<0.0001
Medical diagnosis (%)	34.1	41.9	9.3	83.96	2	<0.0001
Suicide details
Method of death (%)
Hanging	26.7	32.8	37.3	133.67	12	<0.0001
Self‐poisoning	24.4	13.7	8.5
Fall/jump from height	23.5	22.8	25.4
Subway/train/car collision	8.8	4.1	5.1
Other asphyxia	6.9	8.5	11.9
Shooting	3.1	9.6	4.2
Other	6.6	8.5	7.6
Place of death (%)
Own home	61.7	68.4	64.8	34.81	8	<0.0001
Other residence	2.7	2.4	5.1
Outdoors	12.0	11.0	16.1
Subway/railway	7.8	3.5	5.1
Other	15.9	14.7	8.9
Suicide note (% yes)	29.4	35.8	33.1	10.21	2	0.0061

x‐data with N≤5 that have been suppressed due to privacy limits; significant differences are highlighted in bold prints

All significant variables were entered in a multivariate logistic regression for any past‐year mental health contact. Table 2 displays the significant findings from this regression. Past‐year mental health contact was significantly associated with female gender, age 25‐64, absence of an identified psychosocial stressor, diagnosis of schizophrenia, diagnosis of bipolar disorder, past suicide attempt, self‐poisoning method of suicide, and the absence of a suicide note. There were no major secular trends associated with any mental health contact, except for an effect for year 2001 vs. 1998 (adjusted OR=1.86, 95% CI: 1.14‐3.03, p=0.014), which was likely accounted for by a change in data source for emergency department visits in year 2000.

Table 2.

Multivariate logistic regression for any past‐year mental health contact in persons who died by suicide

	Odds ratio, adjusted	Lower confidence limit	Upper confidence limit	p
Gender: male vs. female	0.535	0.43285	0.6613	<0.0001
Age at death: 10 to 24 vs. 25 to 64	0.5639	0.41137	0.7729	0.0004
Age at death: 65 and over vs. 25 to 64	0.7422	0.5683	0.9694	0.0287
Any stressor: yes vs. no	0.6638	0.47935	0.9193	0.0136
Bipolar disorder diagnosis: yes vs. no	5.0475	2.85056	8.9378	<0.0001
Schizophrenia diagnosis: yes vs. no	6.6147	3.6734	11.9112	<0.0001
Past suicide attempt: yes vs. no	3.6598	2.89332	4.6293	<0.0001
Method of death: self‐poisoning vs. hanging	1.575	1.1871	2.0898	0.0016
Note left: yes vs. no	0.8214	0.67917	0.9935	0.0427

Significant differences are highlighted in bold prints

Within the group that had a mental health contact, the most common type of contact was a mental health outpatient primary care visit (54.0%), followed by an outpatient psychiatric visit (39.8%), a mental health emergency department visit (31.1%), and a mental health hospitalization (21.0%).

Mental health contacts were also divided into acute care (emergency department visits and hospitalizations) and ambulatory care (outpatient psychiatric or primary care visits). Table 3 shows the comparison between subgroups who accessed any acute mental health care (N=882), those who accessed only ambulatory mental health care (N=1001), and those with no mental health contact of any type (N=952). The acute care subgroup was younger, less likely to be married or to have an identified psychosocial stressor, more likely to have a major mental illness or past suicide attempt, and less likely to die at home or produce a suicide note.

Table 3.

Comparison of people who died by suicide by type of mental health contact (acute vs. ambulatory care) in prior year

	Accessed acute mental health care (N=882)	Accessed only ambulatory mental health care (N=1001)	No contact (N=952)	Test value (F/χ2)	df	p
Socio‐demographics
Age (mean, years±SD)	44.5 ± 15.6	49.2 ± 16.9	48.1 ± 19.6	18.38	2	<0.0001
Age (%)
≤ 24 years	9.8	5.4	13.0	75.75	4	<0.0001
25‐64 years	79.4	75.6	65.5
≥ 65 years	10.9	19.0	21.4
Gender (% male)	65.9	64.1	80.9	77.56	2	<0.0001
Marital status (%)
Single/no status available	59.2	50.6	53.5	19.61	4	0.0006
Divorced, separated or widowed	20.0	21.6	18.8
Married, including common law	20.9	27.8	27.7
Living circumstances (%)
Alone	44.1	44.3	40.4	25.52	4	<0.0001
Family/friends	46.9	51.0	54.8
Other	8.8	4.6	4.7
Recent stressors
Bereavement (%)	5.1	6.0	5.7	0.72	2	0.6992
Employment/financial (%)	14.1	17.2	23.8	30.79	2	<0.0001
Relationship (%)	8.3	8.1	9.1	0.77	2	0.6804
Interpersonal conflict (%)	14.7	17.4	18.4	4.59	2	0.1008
Medical health (%)	7.3	13.2	16.7	37.91	2	<0.0001
Police/legal (%)	6.8	6.8	6.8	0	2	0.9993
Immigration (%)	1.4	0.6	1.2	2.93	2	0.2309
Any stressor present (%)	41.4	51.0	61.2	72.37	2	<0.0001
Clinical variables
Bipolar disorder diagnosis (%)	10.8	6.7	1.5	68.64	2	<0.0001
Schizophrenia diagnosis (%)	13.4	6.4	1.4	103.72	2	<0.0001
Past suicide attempt (%)	49.5	25.3	11.3	336.68	2	<0.0001
Medical diagnosis (%)	29.9	37.8	33.8	12.83	2	0.0016
Suicide details
Method of death (%)
Hanging	25.1	28.1	33.9	144.21	12	<0.0001
Self‐poisoning	24.1	24.7	12.4
Fall/jump from height	25.7	21.6	23.4
Subway/train/car collision	11.8	6.1	4.3
Other asphyxia	4.9	8.6	9.3
Cutting/stabbing	3.1	3.1	3.9
Other	5.3	7.9	12.7
Place of death (%)
Own home	52.7	69.6	67.5	87.73	8	<0.0001
Other residence	2.8	2.5	3.0
Outdoors	13.6	10.6	12.3
Subway/railway	10.5	5.3	3.9
Other	20.3	12.0	13.2
Suicide note (% yes)	24.6	33.6	35.1	27.13	2	<0.0001

Significant differences are highlighted in bold prints

The number of mental health contacts and the categorized time from last mental health contact (any, and by type) to suicide death are shown in Table 4. The cumulative weekly percentages of subjects receiving mental health care (any, and by type) in the year prior to suicide death are shown in Figure 1.

Table 4.

Number of mental health contacts and timing of last mental health contact in persons who died by suicide

	Outpatient primary care visit (N=1531)	Outpatient psychiatric visit (N=1127)	Emergency department visit (N=690)	Inpatient visita (N=596)	Any mental health contact (N=1883)b
Number of mental health care contacts
Mean±SD	6.5 ± 9.9	11.4 ± 15.1	2.4 ± 2.8	1.9 ± 1.4
Median (range)	3 (1‐153)	6 (1‐134)	1 (1‐24)	1 (1‐12)
Time from last mental health contact to death
Mean time, days (SD)	87.3 (94.8)	66.5 (86.4)	87.4 (99.1)	99.9 (98.7)	52.6 (77.4)
Median time, days (IQR)	47 (14‐134)	26 (8‐86)	42 (9‐138)	62.5 (16.5‐168.5)	18 (5‐63)
0 to 24 hrs, N (%)	12 (0.8)	14 (1.2)	26 (3.8)	31 (5.2)	65 (3.5)
1 to 7 days, N (%)	219 (14.3)	246 (21.8)	127 (18.4)	76 (12.8)	541 (28.7)
8 to 30 days, N (%)	390 (25.5)	337 (29.9)	138 (20.0)	103 (17.3)	561 (29.8)
31 to 90 days, N (%)	383 (25.0)	256 (22.7)	156 (22.6)	144 (24.2)	357 (18.9)
>90 days, N (%)	527 (34.4)	274 (24.3)	243 (35.2)	242 (40.6)	359 (19.1)

^aHospitalizations that immediately followed an emergency department visit were excluded if the main diagnosis was non‐mental health related and/or if it appeared to be directly related to the suicide event

^bUses the type of visit that has the shortest period prior to death

Figure 1.

Cumulative weakly percentage of subjects receiving health service in the year prior to suicide death, by visit type

Cox proportional hazards model found that time from last contact to suicide death was significantly longer among males (adjusted hazard ratio, HR=0.785, 95% CI: 0.708‐0.871, p≤0.0001), people aged 10‐24 years (adjusted HR = 1.426, 95% CI: 1.183‐1.720, p=0.0002), and suicide decedents without an identified psychosocial stressor (adjusted HR = 0.759, 95% CI: 0.634‐0.908, p=0.003).

A shorter time from last contact to suicide death was identified for those with a bipolar disorder diagnosis (adjusted HR=1.935, 95% CI: 1.634‐2.291, p≤0.0001), a schizophrenia diagnosis (adjusted HR=1.531, 95% CI: 1.270‐1.846, p≤0.0001), a past suicide attempt (adjusted HR=1.768, 95% CI: 1.596‐1.958, p≤0.0001), and those who died in hospital (adjusted HR = 1.891, 95% CI: 1.168‐3.060, p=0.0095).

DISCUSSION

To our knowledge, this is the largest study to date that comprehensively reports on the nature and correlates of mental health and non‐mental health care contacts prior to suicide death in a population‐based sample. A key finding is that, among the 2,835 suicide decedents, 66.4% had a mental health care contact during the year prior to death. A mental health focus within a primary care physician contact was identified as the most common specific type of contact (54.0%), followed in descending order by an outpatient psychiatric visit (39.8%), a mental health emergency department visit (31.1%), and a mental health hospitalization (21.0%). There were a number of socio‐demographic, clinical and suicide‐specific differences associated with the presence, type and timing of health care contacts prior to suicide.

The likelihood of any mental health care contact was considerably higher in this study than previous estimates10, 11, 12, 13, 14, 15. This may be best accounted for by our inclusion of a validated method for identifying a mental health care focus within a primary care physician contact, especially as this type of contact was the most common. As well, our sample derived exclusively from an urban setting with limited barriers within a universal health care model that promotes the centrality of primary health care delivery. Any mental health contact was significantly and positively correlated with female gender, age at death of 25‐64 years, absence of an identified recent psychosocial stressor, bipolar disorder diagnosis, schizophrenia diagnosis, past suicide attempt, method of suicide being self‐poisoning, and the absence of a suicide note being left. The age, gender, and schizophrenia effects replicate prior findings11, 12, 13, 14, 35, and overall the results expand our understanding of factors that are associated with treatment contact prior to suicide.

These findings have a number of implications. First, the fact that a substantial majority of people received some form of more broadly‐defined mental health care argues for the great opportunity inherent in clinically‐based suicide prevention interventions. Second, the better characterization of the sub‐population that received mental health care allows for potentially better targeting of clinically‐based suicide prevention interventions36, 37, 38. Mental health treatment has been clearly shown to lower suicide risk4, 27, 39, 40, 41, 42, 43; however, a stronger basis for targeted clinical interventions can be provided by better characterization of groups distinguished by type, frequency and recency of mental health care contacts prior to suicide44, 45. Third, our finding that the most common types of contact occur in ambulatory care strongly reinforces the importance of designing and integrating suicide prevention strategies into routine clinical care rather than viewing suicide prevention strategies as only being relevant in high‐risk, acute care environments. Such strategies should include evidence‐based guidance on most appropriate screening for suicide risk in ambulatory settings and care pathways for different levels of risk1, 46.

Our results indirectly support Finnish data indicating that the prominence of outpatient psychiatric services is a key mental health system variable associated with lower suicide rates39. The challenge, however, is the perception and reality that suicide remains a rare outcome, so that, while many people who die by suicide sought care in ambulatory settings, the majority of people receiving ambulatory care are not going to die by suicide. We found the group of suicide decedents who accessed only ambulatory mental health care to be older, more likely to be single and to have a medical health stressor or comorbid medical diagnosis, which may drive a more medical focus to outpatient psychiatric service visits.

There is broad consensus that comprehensive suicide prevention efforts benefit from both community‐based and clinically‐based interventions36, 47. For the 33.6% of people that had no past‐year mental health treatment contact, community‐based measures such as public education, anti‐stigma campaigns, online self‐help, gatekeeper training, crisis lines, and broad‐based means restriction are paramount40, 47. If specific groups such as men, youth, and older adults are significantly less likely to access mental health treatment prior to suicide, then community‐based interventions should consider these specific demographic groups as critically important target populations.

We also found that approximately one quarter of all suicide decedents only had a non‐mental health physician contact, which could include a primary care visit without a mental health focus, or contact with other medical specialties. People who only had a non‐mental health contact were significantly older (27% above age 65 years), and more likely to be male, married, living with others, and to have a recent medical stressor or any type of identified stressor. The presence of such a large group that was only seen in the context of non‐mental health care highlights the importance of gatekeeper training and general medical education on identification of suicide risk through simple screening measures1, 46. The finding that not only medical stressors, but also psychosocial stressors were associated with non‐mental health contact highlights the powerful link between stress and physical symptomatology that is directing people towards physicians, but not necessarily with their mental health needs as a stated or identified priority.

An examination of time from last mental health contact until date of death found a fairly even proportion of persons having their last contact during the week prior to death, the month prior to death, and one to twelve months prior to death. The median time from any type of last mental health contact until death was 18 days. Therefore, while the suicide rate is clearly highest during the period soon after a treatment contact, and is significantly correlated with a number of clinical factors such as diagnosis, past suicide attempt and age, the time to event analyses reveal a sizeable number that have a clear gap between date of last contact and their death. Among the specific types of mental health care contacts, the longest median time from last contact to death was for inpatient hospitalizations (62.5 days), with 64.8% of deaths occurring >30 days after the last inpatient contact, similar to other studies10. Inpatient hospitalizations are the least frequent type of contact, and frequency was inversely associated with time from last contact until death, but nonetheless these results highlight the importance of not exclusively focusing on the relevance of very recent hospitalizations, which identify high risk periods but are least common.

The results of this study should be interpreted in the context of some limitations. First, the precise nature of the clinical care delivered during various types of treatment contacts was not known, and therefore we do not know if patients had been identified as being at higher risk of suicide and if any interventions were utilized. Similarly, we do not know if the absence of care was as a result of system issues related to lack of access, or whether care was not sought by the individual7. Our study should, therefore, be understood as a descriptive analysis that must be followed up by studies that enhance the delivery and content of care.

Second, coroner data were used to identify suicides. While this provides detailed information on suicide, determination of suicide as the cause of death is inherently complex and may be influenced by the presence of mental health contacts, with a small proportion of deaths likely to be misclassified in each direction. Previous coroner data validation studies have identified that under‐reporting of suicide is greater than over‐reporting, and that the overall rates are quite low, suggesting that the results are likely highly valid but may not be representative of 100% of suicides48, 49.

Similarly, health administrative data maintained at ICES have been extensively utilized for many mental health studies, with the main limitation being successful linkage with external data sources. Our analysis cohort was comprised of 91.7% of all suicide deaths, which is well within the acceptable range, but does indicate that numerous suicide deaths were not included in our analysis. Decedents under age 25 years and those with an immigration stressor were significantly more likely to be excluded from our analyses, and comprise small but important subgroups that are less well represented in our data.

Third, ICES data only captured physician‐based clinical services, and thus we have no information on other important sources of mental health care provided by psychologists, social workers, the educational system, community agencies, and others. While this is clearly a gap, the nature of the Canadian health care system is that physician‐based services are covered through universal health care, while the other sources must be paid for out of pocket, through employer‐based insurance plans, or by institutions such as schools or universities. The typical pathway of care delivery would be for any person identified with significant mental health care needs or any indication of suicide risk to be referred to some form of physician‐based services.

Finally, the study only examined suicides among people living in the city of Toronto, a large urban centre with ample mental health resources. It is unknown whether the results would be applicable in other non‐Canadian or rural settings.

In conclusion, two thirds of people who died from suicide had mental health care contacts during the year prior to death, most commonly primary and specialty outpatient care. Our data suggest that clinically‐based suicide prevention strategies should adjust to the predominance of opportunities within ambulatory care. For the one third of decedents who do not have any mental health care contact, and who are more likely to be male and youth or older adults, community‐based suicide prevention opportunities are a key source of suicide prevention and should be designed and delivered with those at highest need in mind. Overall, understanding the factors that influence the likelihood and nature of mental health care provided prior to suicide can aid in developing evidence‐based care delivery and suicide prevention interventions.

ACKNOWLEDGEMENTS

The authors thank J. Edwards, J. Lindsell, A. Stephen and the staff at the Office of the Chief Coroner of Ontario for making this research possible. Funding for this project was provided by the Ontario Mental Health Foundation in the form of an operating grant. The study was also supported by the Institute for Clinical Evaluative Sciences (ICES), which is funded by an annual grant from the Ontario Ministry of Health and Long‐Term Care (MOHLTC). The opinions, results and conclusions reported in this paper are those of the authors and are independent from the funding sources. No endorsement by ICES or the Ontario MOHLTC is intended or should be inferred. Parts of this paper are based on data and information compiled and provided by the Canadian Institute of Health Information (CIHI). However, the analyses, conclusions, opinions and statements expressed herein are those of the authors, and not necessarily those of CIHI.