Risk factors for mortality after hospitalization for suicide attempt: results of 11-year follow-up study in Piedmont Region, Italy

Purpose Suicide attempters are at high risk of premature death, both for suicide and for non-suicidal causes. The aim of this study is to investigate risk factors and temporal span for mortality in a cohort of cases admitted to hospital for suicide attempt. Methods The cohort included 1489 patients resident in Piedmont Region, North West of Italy, who had been admitted to hospital or emergency department for suicide attempt between 2010 and 2020. Cox regression models were used to identify risk factors for death. The final multivariate model included gender, age, area deprivation index, family composition, psychiatric disorders, malignant neoplasms, neurological disorders, diabetes mellitus, cardiovascular diseases, chronic obstructive pulmonary disease, and intracranial injury or skull fracture. Results During the observation period, 7.3% of patients died. The highest mortality was observed within the first 12 months after suicide attempt, and remained elevated for many years afterwards. Male gender, older age, high deprivation index of the census area, single-parent family, mood disorders, malignant neoplasms, diabetes mellitus and intracranial injuries or skull fracture were independent predictors of death. Risk factors for natural and unnatural causes of death were also identified. Conclusions The mortality risk of suicide attempters is very high, both in the months immediately following the attempt and afterwards. The identification of high-risk groups can help to plan outpatient care following the hospital discharge. Our findings urge the need to design strategies for the assistance and care of these patients at long term in order to reduce the unfavourable outcomes.

The burden of suicide attempt is difficult to estimate due to underreporting and lack of standard reporting procedures in many countries.The prevalence of lifetime suicide attempt was estimated as 1.8% in 2001-2003 in Europe [22], and 0.3-4.2% in 2001-2007 worldwide [23][24][25].In more recent years, an increasing trends of suicide attempt rates have been reported in some countries [26][27][28][29][30].
The identification of sociodemographic and clinical risk factors predicting premature mortality is of paramount importance to address high-risk subgroups and target them with appropriate health care and prevention interventions.A number of risk factors for mortality have been documented in prospective studies, i.e. male gender, older age, living alone, being single, lower education, income area/ socioeconomic status, unemployment, retirement, violent method of suicide attempt and repetition of attempt [1-7, 9, 11, 13-18, 20, 31-34].Psychiatric disorders and physical illnesses co-occur commonly and represent a noteworthy risk factor for increased mortality among suicide attempters, particularly in the presence of both conditions [2,4,5,21,22,28,31,[35][36][37].Among psychiatric disorders, anxiety, depressive and bipolar disorder, schizophrenia, alcohol and drug abuse, and personality disorder are the most recognized co-existing diagnoses associated with mortality risk, and when co-occurring can significantly reduce the temporal interval from first episode of suicide attempt to death [3,4,6,9,13,20,21,33,37].
Most studies on risk factors for mortality following suicide attempt were conducted in North-Europe where the rates of suicidal behaviours are particularly high.Due to different prevalence rates and context, studies investigating the phenomenon in South-Europe are needed.
The aim of this study is to investigate risk factors and temporal span for mortality in a cohort of cases admitted to hospital or emergency department for suicide attempt between 2010 and 2020 in Piedmont, a North-West Italian Region with high rates of suicidal behaviour [38].

Study sample
The cohort included 1489 patients resident in Piedmont, a region of 4.4 million inhabitants in North West of Italy, who had been admitted to hospital or emergency department (ED) for suicide attempt between 1st January 2010 and 31st December 2020.
All patients who had at least one episode of suicide attempt during the study period were included in the cohort.If there was more than one episode of suicide attempt, the first episode occurred during the study period was used to define the patients' date of entry into the cohort.In order to allow record-linkage of administrative and health data, only subjects who resided in Piedmont at the time of ED or hospital discharge were included in the cohort.Patients aged 12-74 years were included in the study.Patients were followed until death, emigration, or the end of the study.Follow-up started from the date of discharge from hospital or ED.All patients were followed up longitudinally for up to 11 years, i.e. follow-up ended on 31st December 2020.

Data collection
Information on the patients in the study was collected through a record-linkage of the administrative and health data available in the Piedmont Longitudinal Study.Linkage of the data archives was done using an anonymous identification code under the frame of the National Statistical Act that legitimates the use and linkages of data for scientific purposes without the need of an Ethical Committee approval.
The sources of data were: 2011 Population Census of the Piedmont Region and the NHS Regional Population Registry for sociodemographic characteristics; and the hospital discharges and emergency department databases for health-related conditions and diagnoses.Information on all-cause mortality (dead/alive) was extracted from the NHS Regional Population Registry available for the period 2010-2020.Information on cause-specific mortality was obtained from the Mortality Register of the Piedmont Region available for the period 2010-2018.

Measures
The following sociodemographic characteristics were extracted: gender, date of birth, age at the index episode of suicide attempt, deprivation index of the census area, marital status, education, occupation and family composition.Area deprivation index was based on five conditions describing social and material deprivation, measured at the census section level: % of population with low education, % of unemployed, % of population living in rented houses, in crowded households, in single-parent families.The area deprivation index was then categorized in quintiles (1st = lowest to 5th = highest deprivation) [39].

Statistical analysis
The primary outcome under study was all-cause death (yes/ no).
Secondary analyses were conducted on cause-specific mortality (natural and unnatural).
Descriptive statistics were used to describe the study population by gender.Kaplan-Meier survival analysis was used to plot temporal patterns of death after ED or hospital discharge for suicide attempt.The curves of the subgroups were compared through log-rank test.For each patient in the cohort, person-years of survival were calculated starting from the date of discharge from ED or hospital of the index episode to the date of death or the end of follow-up.
Gender, age, area deprivation index, marital status, education, family composition, psychiatric disorders and physical illnesses were evaluated as risk factors for death in Cox regression models.
Unadjusted Cox regression models were performed to identify risk factor independently from other covariates.Collinearity among variables was checked.The non-collinear and statistically significant variables were then selected to build the multivariate model.Due to correlation between marital status and family composition (r = 0.7), only family composition was included in the final model.However, sensitivity analysis was performed with marital status included in the model instead of family composition.Some variables had several missing data, therefore, to minimize the reduction of cases in the regression models, the additional level "Missing" was created.Analyses of Schoenfeld residuals and Kaplan-Meier were performed to test proportional hazard assumptions, before building the final multivariate Cox regression model.The proportional hazard assumption was satisfied for all variables except for psychiatric disorders and chronic obstructive pulmonary diseases in the model focused on unnatural deaths.Since the Kaplan-Meier survival curves showed violation only in the first months of the follow-up, both factors were added in the Cox regression model.However, results of this model should, therefore, be considered in light of this limitation.
The final Cox model included: gender, age, area deprivation index, family composition, psychiatric disorders, malignant neoplasms, neurological disorders, diabetes mellitus, cardiovascular disorders, chronic obstructive pulmonary disease and intracranial injury or skull fracture.As regards psychiatric disorders, the diagnoses included in the category of "other disorders" (personality disorder, drug and alcohol dependence, anxiety disorder, adjustment disorder, dementia and other disorders due to organic conditions) showed protective effect compared to no disorder at all in the preliminary analysis.Therefore, in the multivariate analysis, these two groups were merged as reference level, and psychiatric disorders were re-grouped into three categories: none/other disorders; schizophrenia; and mood disorders (bipolar and depressive disorders).The multivariate Cox regression models assessing risk factors for natural and unnatural causes of death were run on the restricted sample of patients entering the cohort between 2010 and 2018.
All the analyses were performed by using STATA 16 statistical software [40].

Descriptive statistics
Sociodemographic and clinical characteristics of the study sample are shown in Table 1.

Survival analysis
The Kaplan-Meier survival curves by gender and psychiatric disorders are shown in Figs. 1 and 2.
Mortality was highest during the first 12 months after the index suicide attempt: 45.4% of deaths occurred in the first year, 14.8% in the second year, and the others were sparse over the following 7 years.No deaths occurred in the last 2 years of the observation period.
Through the entire follow-up, the survival rate was significantly lower for males than females (log rank p < 0.001); 33 males and 16 females died in the first 12 months (Fig. 1).
The survival rate was significantly lower for patients diagnosed with schizophrenia and mood disorders than those with no disorders or other disorders (log rank p = 0.003) (Fig. 2).This was observed also when looking at natural causes of death (data not shown).

Cox regression models for all-cause mortality
In unadjusted Cox regression models, the risk of death was significantly higher for males, progressively increased with age, and was significantly higher for widowed, couples without children, single-parent family, and single.The risk of death was significantly associated with schizophrenia, mood disorders, malignant neoplasms, neurological disorders, diabetes mellitus, cardiovascular diseases and chronic obstructive pulmonary diseases.The association of high area deprivation index and intracranial injury or skull fracture with the risk of death was marginally significant (Table 2).
In the multiple Cox regression model, male suicide attempters had 86% higher risk of premature death than females (HR 1.86, 95%CI 1.24-2.78).The risk of death was associated with age, with 6% increased risk for each year of increase in age (HR 1.06, 95%CI 1.04-1.08).High deprivation index of the residence area (4th quintile) was associated with 71% increased risk of mortality compared to low deprivation (1st-3rd quintile) (p = 0.037).Single-parent family was a significant predictor of death as compared to family defined as couples with children, and the risk was doubled (HR 2.48, 95%CI 1.36-4.53).In sensitivity analysis including marital status in the model instead of family composition, widowed patients were associated with twofold higher risk of death (HR 2.44, 95%CI 1.22-4.90)(data not shown).Mood disorders predicted 72% increased risk of death as compared to no or other psychiatric diagnosis (p = 0.037).The diagnosis of intracranial injury or skull fracture and malignant neoplasms were associated with 1.67-2.71times greater mortality (p = 0.050, p < 0.001, respectively), whilst the other diseases did not reach significance level (Table 3).

Cox regression models for naturaland unnatural-cause mortality
Age was a significant predictor of natural-cause mortality, with 8% increased risk for each year of increase in age (HR 1.08, 95%CI 1.05-1.12).Living in couples without children and single-parent family was associated with threefold (HR 3.05, 95%CI 1.09-8.49)and fourfold (HR 4.68, 95%CI 1.49-14.71)increased risk of mortality compared to couples with children, respectively.As regards chronic diseases, malignant neoplasms (HR 5.40, 95%CI 2.68-10.87)and diabetes mellitus (HR 3.14, 95%CI 1.31-7.54)were significant predictors of natural death (Table 3).When looking at unnatural-cause mortality, males had twofold higher risk of death compared to females (HR 2.51, 95%CI 1. 21-5.24).Age was associated with 3% increased risk of death for each year of increase in age (HR 1.03, 95%CI 1.00-1.06).All other variables did not reach statistical significance.In sensitivity analysis including marital status instead of family composition in the model, widowed patients had fivefold greater risk of death (HR 5.00, 95%CI 1.24-20.11)(data not shown).

Discussion
We conducted a cohort study by linking health and administrative data of the Piedmont Region following prospectively 1489 patients who were admitted to hospital or ED for suicide attempt between 2010 and 2020.To our knowledge, this is the first study to investigate risk factors for mortality in a cohort of hospitalized suicide attempters over 11-year span in Italy.Our study extends prior findings by adding new evidence from the South-European context.In this study, male gender, older age, high area deprivation index, single-parent family, comorbidity with mood disorders, malignant neoplasms and intracranial injury or skull fracture were predictors of all-cause death following the first hospital admission for attempted suicide.Older age, living in couples without children and single-parent family, being affected by malignant neoplasms and diabetes mellitus predicted natural-cause death, whereas gender and older age increased the risk of unnatural-cause death.The risk of death was highest during the first 12 months after suicide attempt, and remained elevated for many years afterwards.
During the observation period, 7.3% of subjects died, similar to previously observed 6.4% in Canadian and 8.7% in USA cohorts [9,10], but lower than 12.1% reported in the Italian cohort by Pavarin and colleagues [16].Although the risk for suicide and non-suicidal mortality persists for many years following the first episode, it is particularly pronounced shortly after the attempt episode [1-3, 6, 7, 10-13, 16, 17, 20, 21, 50].In our study, 45% of deaths occurred in the first 12 months: this result points out that the period following the attempt episode is the most critical, and suggests that integrated and intense prevention and care interventions should be implemented in this period.
In the adjusted regression model, several characteristics emerged as significant predictors of mortality following suicide attempt.It is already well established that male gender is a significant predictor of mortality, particularly for suicide, among suicide attempters [1, 4-7, 9, 16, 18, 20, 32, Statistically significant results are marked in bold PY person-years, HR hazard ratios, CI confidence interval, P p-value a Other: families with two or more nucleus or other relatives b Other disorders: personality disorder, drug and alcohol dependence, anxiety disorder, adjustment disorder, dementia and other mental disorders due to organic condition c Mood disorders: bipolar disorder, depressive disorder .Consistently, also in our study, the survival time was significantly lower for males than females [1,2,4,5,33,50].This may be attributable to a higher tendency of males to use violent methods and their higher intention to die, a relevant risk factors for suicide particularly within the first months after the attempt [3, 5-7, 13-15, 18, 31, 34, 37, 42, 51].Moreover, males in our cohort had more alcohol and drug related problems than female suicide attempters, a behaviour that may contribute to violent causes of death [3,4,6,7,20,42,51].This result is similar to that observed in the Finnish cohort by Suominen and colleagues [4].Moreover, in our study, a greater proportion of males was diagnosed with intracranial injury or skull fracture, conditions that may act as a proxy of violent suicide methods and therefore be linked to a higher risk of death for unnatural causes.
In agreement with earlier studies, the risk of death increased with the increase of age at the time of suicide attempt [1, 5-7, 9, 16, 18, 20, 31-33, 49], possibly due to higher prevalence of health problems in older patients [31,35].Indeed, in our study, the prevalence of physical and mental illnesses increased proportionally with increased age (data not shown).Moreover, with advanced age the proportion of widowed patients increases, which may result in loneliness and living alone, recognized risk factors for death in general population and in suicide attempters [6,11,20].Indeed, the association of widowed condition with the risk of all-cause and unnatural-cause deaths was also observed in our study (data not shown).
Suicide attempters living in highly deprived areas were at 71% greater risk of mortality as compared to those living in less deprived areas, congruently with what observed elsewhere using income and education level as indicators of socioeconomic status [14,32].Patients of high deprived areas may be at higher risk of psychiatric disorders or physical illnesses, have limited access to social and medical resources, poor social support, bad diet habits, substance use related problems and poor skills to cope with adverse life situations compared to those of more advantaged areas.These characteristics may therefore exacerbate the negative outcomes of a frail condition as suicide attempt.Particular attention should be paid to patients of low socioeconomic level after suicide attempt.
The importance and mechanisms of family composition that may impact the effect of attempted suicide on mortality have been understudied.In our study, living in single-parent family was associated with an increased risk of all-cause and natural-cause mortality compared to living in couples with children.This is expected since, single parents may experience stressful life events, economic disadvantage, and social and psychological pressure to a greater extent than two-parent families.However, in a previous Finnish study, an increased risk of natural-and violent-cause mortality was observed among single-parent families of general population, but not among suicide attempters [11].
Accounting for all other variables included in the model, being affected by mood disorders was an independent risk factor for all-cause mortality, similarly to what observed in other studies [3,7,9,13,21,33,37].However, although the direction of effect was consistent, the statistical significance was not reached when the impact on natural and unnatural deaths was tested separately, probably due to limited statistical power of the restricted sample.Moreover, due to violation of proportional hazard assumption, the results of the model assessing unnatural causes of death should be taken with caution.Nevertheless, these findings point out the need to apply particular attention in diagnosis of concurrent psychiatric disorders, and in prescribing effective treatment in order to prevent mortality risk.
As regards physical comorbidities, an increased risk of all-cause and natural-cause deaths was observed for patients suffering from malignant neoplasms and diabetes mellitus.Physical illnesses are significant predictors of death among suicide attempters [2,5], especially in case of chronic and severe conditions.Moreover, it is recognised that suicidal patients with chronic conditions are more depressed [35], which may be associated with severe health outcomes.Finally, intracranial injury or skull fracture was also a significant predictor of all-cause mortality, a diagnosis that may be related to suicide attempt method itself, e.g.violent attempts could lead to serious traumatic injuries, and in these cases the risk of death can be higher.
This study has a number of strengths.The sample included all episodes of suicide attempts referred from hospitals and ED in the region, irrespective of the method used.Several information on the study subjects were collected from administrative and health registers ensuring richness of the data and investigation of both sociodemographic factors and comorbidities.Completeness of the mortality data was high.However, the study results should be considered also in a light of some limitations.The suicide attempts are potentially underestimated due to misclassified and underreported cases, i.e. those not accessing ED or hospital, and those registered with other diagnoses, e.g.accidents.So, the sample may not be fully representative of all suicide attempts in the regional catchment area but biased toward the most seriously affected cases.We did not have the information on cause-specific mortality for the whole timeline of the cohort, but only until 2018.Therefore, Cox regression models assessing natural and unnatural causes of death were run on restricted samples, possibly resulting in a loss of statistical power for some factors.Moreover, the comparison of results across studies may be challenging due to the ambiguous and inconsistent definition of suicide attempt sometimes including self-injurious acts with and without suicidal intent.Finally, the study would deserve a longer follow-up.
In conclusion, our results showed a risk of mortality particularly high within the first year after admission to hospital or ED for attempted suicide.Male gender, increased age, high area deprivation index, single-parent family, mood disorders, malignant neoplasm, diabetes mellitus and intracranial injury or skull fracture were important risk factors for mortality after suicide attempt.Focusing on these predictors can help to identify high-risk groups, inform clinicians on medical and psychiatric needs of suicidal patients, and plan outpatient care following hospital discharge.Our findings urge the need to design strategies for the assistance and care

SD
standard deviation, P p-value a Other: student, housewife, other conditions b Other: families with two or more nucleus or other relatives c Other: dementia and other mental disorders due to organic condition

Fig.
Fig. 1 Kaplan-Meier survival curve for patients admitted to hospital or ED for suicide attempt between 2010 and 2020, in Piedmont Region, Italy, by gender

Table 1
Characteristics of patients admitted to hospital or ED for suicide attempt between 2010 and 2020, in Piedmont Region, Italy, by gender

Table 3
Risk factors for all-cause, natural-cause and unnatural-cause death: multivariate Cox regression models Statistically significant results are marked in bold Adj HR adjusted hazard ratios, CI confidence interval, P p-value, n/N number of deaths/Number of patients in the group *Statistical test showed violation of proportional hazard for these factors.These results should be taken with caution a Other: families with two or more nucleus or other relatives