Suicide attempt risk factors in individuals with psychiatric disorders and multi-affected families: a network analysis Authors: Gabriele Torino, Giuseppe Delvecchio, Alessandro Pigoni, Domenico Madonna, Silvana Zito, Paolo Brambilla Published: 2026 ABSTRACT Psychiatric disorders are characterized by familiarity, which could also be implicated in the process leading to suicide attempts. This study aimed to evaluate the associations between sociodemographic, clinical, and psychopharmacological factors and previous suicide attempts in individuals with psychiatric disorders and multi-affected families. Separate networks for the sociodemographic, clinical, and psychopharmacological domains were estimated using Gaussian graphical models. The most influential variables were then combined to estimate an overall network. Results indicated that previous suicide attempts were primarily associated with alcohol abuse, female sex, and a younger age of onset of psychiatric disorders. The sociodemographic network revealed that age, sex, and marital status were directly related to a previous suicide attempt. In the clinical network, alcohol abuse and peripartum complications were central, while mood stabilizer therapy duration emerged as a significant factor within the psychopharmacological network. These findings underscore the importance of a multidimensional approach in suicide risk assessment to address the complex determinants of suicidal behavior effectively. Introduction Psychiatric disorders, particularly those on the affective and psychotic spectrum, are distinguished by a notable degree of familiarity, signified by the recurrence of symptoms within the same household1–3. This phenomenon is frequently exacerbated by genetic, environmental, and psychosocial predisposing factors that heighten vulnerability to these disorders, and that also contribute to reinforcing public stigma, lowering self-esteem, and increasing social distancing4,5. Similarly, suicidal behaviors appear to show a significant degree of familial transmission, independent of the familial transmission of psychiatric disorders6. However, further studies are needed to investigate this relationship and confirm the hypothesis of suicidal behavior transmission and any underlying mechanisms. As suicide is a multifaceted behavior triggered by several biopsychosocial causes, instead of an epiphenomenon of a psychiatric disorder, it is particularly difficult to identify predisposing factors for increased suicide risk. In accordance with the suicide familiarity hypothesis, a substantial body of research has identified specific hereditary personality traits, such as impulsivity and perfectionism, as major predisposing factors for suicidal behaviors7,8. This link is particularly pronounced in cases where the individual is also suffering from psychiatric disorders, which have a considerable degree of heritability as well9. Therefore, there is a sixteen-fold increased risk of suicide in individuals with mental disorders compared to those without them10, with an estimated lifetime prevalence of suicide attempts (SAs) above 30% among those with affective disorders11,12. When considering sociodemographic risk factors for suicidal behaviors, age and biological sex require important consideration13. Indeed, even though almost three out of four deaths by suicide worldwide are men, evidence indicated that SAs are 78% more frequent among females14. Concerning age groups, while the majority of suicidal behaviors are enacted by older adults, suicide represents the second leading cause of death among young adults aged between 15-2913. Furthermore, the estimated ratio of attempted to completed suicides among adolescents ranges from 50:1 to 100:115. Additionally, being divorced or unmarried, unemployed, or having a lower level of education also increases the risk of attempting suicide16–18. In the clinical setting, evidence has revealed that the major factors consider when assessing suicide risk include previous attempts and hospitalizations, as well as substance abuse, age of onset of psychiatric disorders, and a family history of mental disorder and suicide19–22. Conversely, in the domain of psychopharmacological treatment, the extant research indicates contradictory findings, as the efficacy of long-term psychotropic medications in reducing the lifetime risk of suicidal behaviors remains uncertain23,24. Consequently, when conducting a suicide risk assessment, it is crucial to consider all aspects of an individual’s life instead of examining certain domains separately. However, there is often a predominant focus merely on the clinical aspects related to psychiatric symptoms, which limits a holistic understanding of the other factors influencing the psychology of suicidal behavior25. Nonetheless, given the multifaceted nature of suicide, its prediction is a highly complex undertaking, even in individuals deemed to be at elevated risk. Consequently, conventional clinical practices and traditional statistical methodologies have frequently proven ineffective and inaccurate in predicting suicidal behavior26–28. To address this gap, the application of sophisticated statistical approaches such as Gaussian graphical models29,30 (GGMs), which can offer more information than traditional multiple regression models, is becoming more widespread in suicidology31,32. A clear advantage of GGMs is their ability to simultaneously identify a set of several variables of interest (“nodes”) and observe their direct and indirect relationships (“edges”), ultimately facilitating the identification of proximal, distal, and mediator factors. The present study is the first to apply GGMs in a sample of individuals receiving psychiatric care and belonging to multi-affected families, defined as households with at least one first-degree relative with a documented history of psychiatric or neurodegenerative disorders, or a history of suicide. Specifically, this study aims to estimate networks that highlight associations among variables from different thematic domains and their influence on suicidal behaviors, in order to: (i) evaluate the direct and indirect associations between variables from distinct thematic domains (i.e., sociodemographic, clinical, and psychopharmacological) and suicidal behavior, as well as the reciprocal relationships among these variables; and (ii) estimate an additional, comprehensive network composed of the variables most strongly associated with suicidal behavior across the three thematic domains, in order to identify the most influential factors contributing to suicide risk. Methods Participants and procedure The present study’s participants were adult and older adult individuals with a psychiatric disorder (>18 years old) belonging to a multi-affected family, defined as a household with at least one first-degree relative with a documented history of psychiatric or neurodegenerative disorders, or who died by suicide. A convenience retrospective sampling strategy based on chart review was used, including participants admitted to the Department of Mental Health of the Fondazione IRCCS Ca’ Granda, Policlinico Hospital of Milan (Italy) between May 2018 and December 2019. At the time of data collection, participants were receiving different types of psychiatric care: the majority were outpatients receiving follow-up visits at mental health services, while a minority were inpatients admitted for acute psychiatric care. The decision to focus on this convenience sample was based on the unique opportunity to study individuals with lived experience of psychiatric disorders within multi-affected families in a community-based mental health setting. The following self-report data were considered for this study: biographical data, education, work activity, ICD-9-CM or DSM-5 diagnosis and comorbidities, number of hospitalizations, psychopathological and clinical characteristics (e.g., age onset of the mental disorder, duration of treated and untreated illness, current and past psychopharmacological therapy, history of a SA), family history of psychiatric and neurodegenerative disorders, family history of first-degree relatives who died by suicide, current and/or past substance use, and partum complications. To note, in the context of the present study, the term “suicidal behavior” always refers to a self-inflicted potentially injurious behavior with the deliberate intention to procure one’s death with a non-fatal outcome. Therefore, we did not refer to all those self-injurious behaviors without a clear intent to end their own life (e.g., deliberate self-harm or non-suicidal self-injury) or with a fatal outcome. Conversely, the term “family history of suicide” refers to a relative’s death by suicide. Chart review data collection was approved by the Ethics Committee of the Fondazione IRCCS Ca’ Granda, Policlinico Hospital of Milan (Ref: 6955). Given the retrospective design of the study, active informed consent was not required, and data use was permitted under the terms of the ethical approval. Statistical approach To present the descriptive characteristics of our sample, we collected and extracted information on twenty-six variables of interest, categorized into sociodemographic, clinical, and psychopharmacological domains, extracted from the health records of the patients. Subsequently, we conducted preliminary analyses to identify the primary predictors of SAs to be included in the three thematic networks. Specifically, we used logistic regression models to select predictors of SAs (Supplementary Table 1). Given the considerable heterogeneity of the collected variables, it was deemed appropriate to include only a subset of them in the networks, in order to focus on those most influential for the suicide outcome under investigation. Variable selection for inclusion in the networks was guided by both empirical associations identified through preliminary logistic regression analyses and theoretical relevance based on major suicide risk factors reported in the extant literature. For example, variables such as peri-partum complications and inpatient status were included because previous research has linked obstetric complications to long-term mental health vulnerability and elevated suicide risk33,34, while inpatient status has consistently been associated with heightened short-term risk, particularly during and immediately after hospitalization35. This combined empirical–theoretical rationale ensured that less frequently examined but potentially meaningful factors were not excluded solely because of their low prevalence in the sample. Similarly, the variables included in the final network were selected from those identified as most strongly related with prior SAs across the three thematic networks (sociodemographic, clinical, and psychopharmacological). This approach enabled the estimation of a final network composed of variables exhibiting strong associations with a lifetime history of a SA, thereby facilitating a comprehensive evaluation of the network associations. All analyses were performed using JASP software (version 0.19.1), which uses the latest version of R 4.1.3. The network estimations, including centrality measures and the weight matrix, were conducted using default parameters based on least absolute shrinkage and selection operator (LASSO) regularization combined with the extended Bayesian information criteria36 (EBIC). The LASSO algorithm automatically selects the most appropriate operator through the EBIC, which is regulated by the tuning parameter γ, set to 0.5. Pairwise deletion was used to handle missing data, maximizing statistical power by using all available data on an analysis-by-analysis basis. Results Sociodemographic and clinical characteristics of participants Table 1 presents the sociodemographic characteristics of the 92 individuals receiving psychiatric care included in the study. Among these, most were females (54.35%), and the mean age was 48.66 years (age range = 19–87 years; standard deviation [SD] = 14.58). The vast majority of the participants reported being single (55.4%) or married (33.7%). Furthermore, a significant proportion of the sample had received an elementary/junior school (37%) or a high school (48.9%) education, resulting in more than 85% of the sample having attained one of these two levels. Most patients were engaged in either stable (30.4%) or occasional occupation (34.8%), while a minority were students (3.3%), retired (13%), or unemployed (17.4%).Table 1Socio-demographic characteristicsN%SexMales4245.65%Females5054.35%Age (years)Mean (SD)48.66(14.58)Marital statusSingle5155.4%Married3133.7%Divorced/separated99.8%Widowed11.1%Education levelElementary/junior high school3437%High school4548.9%Professional certificate22.17%University degree77.60%Missing data44.35%Working statusStable occupation2830.4%Occasional occupation3234.8%Studying33.3%Not working nor studying1617.4%Retirement1213%Missing data11.1%NationalityItalian8289.1%Foreigner1010.9%Total number of participants92100% Table 1Socio-demographic characteristicsN%SexMales4245.65%Females5054.35%Age (years)Mean (SD)48.66(14.58)Marital statusSingle5155.4%Married3133.7%Divorced/separated99.8%Widowed11.1%Education levelElementary/junior high school3437%High school4548.9%Professional certificate22.17%University degree77.60%Missing data44.35%Working statusStable occupation2830.4%Occasional occupation3234.8%Studying33.3%Not working nor studying1617.4%Retirement1213%Missing data11.1%NationalityItalian8289.1%Foreigner1010.9%Total number of participants92100% Socio-demographic characteristics As shown in Table 2, the majority of patients (83.7%) were treated as outpatients, while the remaining ones (16.3%) were admitted to the psychiatric ward during the assessment. Furthermore, more than half of the participants had been hospitalized at some point in their lives due to psychiatric conditions, with some experiencing multiple admissions (37%). The most prevalent diagnosis was bipolar disorder (20.6%), followed by major depressive disorder (18.5%), schizophrenia spectrum disorder (16.3%), anxiety spectrum disorder (13%), personality disorder (8.7%), and obsessive-compulsive disorder (4.3%). Concerning psychiatric comorbidities, the most common were personality disorder (10.8%) and major depressive disorder (7.6%). A high degree of heterogeneity was observed in the duration of psychiatric disorders, with patients experiencing symptoms for up to five years (26.1%) and others reporting symptoms for over thirty years (16.3%). The majority of patients (43.5%) received psychiatric treatment for a period of up to ten years, and most (90.2%) reported a family history of psychiatric disorders. Additionally, only six participants (6.5%) reported having attempted suicide at least once during their lifetime, while more than a third of the sample reported a history of substance abuse, either alcohol (17.4%), cannabis (8.7%), and cocaine (8.7%). Furthermore, approximately one-third of the participants had a family history of suicide (9.8%) or neurodegenerative disorders (19.6%). Finally, a small percentage of participants (7.6%) reported experiencing complications during childbirth.Table 2Clinical characteristicsN%InpatientsYes1516.3%No7783.7%Hospitalizations11516.3%2–42426.1%4+1010.9%Missing data4346.7%DiagnosisBipolar disorder1920.6%Major depressive disorder1718.5%Personality disorder88.7%Obsessive-compulsive disorder44.3%Anxiety spectrum disorder1213%Schizophrenia spectrum disorder1516.3%Psychiatric comorbidityBipolar disorder22.2%Major depressive disorder77.6%Personality disorder1010.8%Obsessive compulsive disorder11.1%Anxiety spectrum disorder11.1%Psychosis22.2%No comorbidity6975%Psychiatric disorder age onset10–20 years old2223.9%21–30 years old2021.7%31–40 years old2527.2%41–50 years old1718.5%50+ years old88.7%Psychiatric disorder duration0–5 years2426.1%6–10 years1920.6%11–20 years1617.4%21–30 years1819.6%30+ years1516.3%Psychiatric treatment duration0–10 years4043.5%11–20 years1314.1%21–30 years88.7%30+ years1112%Missing data2021.7%Untreated psychiatric disorder duration0–10 years99.8%11+ years66.5%Missing data7783.7%History of a suicide attemptYes66.5%No8693.5%Family history of a psychiatric disorderYes8390.2%No99.8%Family history of suicideYes99.8%No8390.2%Family history of a neurodegenerative disorderYes1819.6%No7480.4%Substance abuseCannabis88.7%Cocaine88.7%Alcohol1617.4%Peri-partum complicationsYes77.6%No8592.4%Total number of participants92100% Table 2Clinical characteristicsN%InpatientsYes1516.3%No7783.7%Hospitalizations11516.3%2–42426.1%4+1010.9%Missing data4346.7%DiagnosisBipolar disorder1920.6%Major depressive disorder1718.5%Personality disorder88.7%Obsessive-compulsive disorder44.3%Anxiety spectrum disorder1213%Schizophrenia spectrum disorder1516.3%Psychiatric comorbidityBipolar disorder22.2%Major depressive disorder77.6%Personality disorder1010.8%Obsessive compulsive disorder11.1%Anxiety spectrum disorder11.1%Psychosis22.2%No comorbidity6975%Psychiatric disorder age onset10–20 years old2223.9%21–30 years old2021.7%31–40 years old2527.2%41–50 years old1718.5%50+ years old88.7%Psychiatric disorder duration0–5 years2426.1%6–10 years1920.6%11–20 years1617.4%21–30 years1819.6%30+ years1516.3%Psychiatric treatment duration0–10 years4043.5%11–20 years1314.1%21–30 years88.7%30+ years1112%Missing data2021.7%Untreated psychiatric disorder duration0–10 years99.8%11+ years66.5%Missing data7783.7%History of a suicide attemptYes66.5%No8693.5%Family history of a psychiatric disorderYes8390.2%No99.8%Family history of suicideYes99.8%No8390.2%Family history of a neurodegenerative disorderYes1819.6%No7480.4%Substance abuseCannabis88.7%Cocaine88.7%Alcohol1617.4%Peri-partum complicationsYes77.6%No8592.4%Total number of participants92100% Clinical characteristics Psychopharmacological information The 92 patients included in the study were undergoing psychopharmacological treatment at the time of evaluation. Table 3 shows that the most frequently used medications were antidepressants, accounting for 62% of cases, followed by antipsychotics (57.6%) and mood stabilizers (56.5%). Among the patients treated with antidepressant medications, the duration of treatment ranged from up to five years, except for one patient who was treated for more than ten years. The distribution of antipsychotic medication duration across the different time categories exhibited a slightly more varied pattern, with the majority of patients receiving treatment for up to five years (62.3%), a substantial proportion treated for six to ten years (17%), and a smaller percentage treated for more than ten years (11.3%). A similar trend was observed among patients treated with mood stabilizers, with the vast majority receiving treatment for up to five years (46.1%), and only a small percentage receiving treatment for more than six years (7.7%).Table 3Psychopharmacological informationN%Antipsychotic medication (lifetime)Yes5357.6%No3942.4%Antipsychotic medication duration0-1 year1834%2-5 years1528.3%6-10 years917%10+ years611.3%Missing data59.4%Antidepressive medication (lifetime)Yes5762%No3538%Antidepressive medication duration0–1 year2747.4%2–5 years1424.6%6–10 years00%10+ years11.8%Missing data1526.3%Mood stabilizers medication (lifetime)Yes5256.5%No4043.5%Mood stabilizers medication duration0–1 year1426.9%2–5 years1019.2%6–10 years35.8%10+ years11.9%Missing data2446.2%Total number of participants92100% Table 3Psychopharmacological informationN%Antipsychotic medication (lifetime)Yes5357.6%No3942.4%Antipsychotic medication duration0-1 year1834%2-5 years1528.3%6-10 years917%10+ years611.3%Missing data59.4%Antidepressive medication (lifetime)Yes5762%No3538%Antidepressive medication duration0–1 year2747.4%2–5 years1424.6%6–10 years00%10+ years11.8%Missing data1526.3%Mood stabilizers medication (lifetime)Yes5256.5%No4043.5%Mood stabilizers medication duration0–1 year1426.9%2–5 years1019.2%6–10 years35.8%10+ years11.9%Missing data2446.2%Total number of participants92100% Psychopharmacological information Network analyses The first network consists of SA and sociodemographic information variables, including age, sex, education level, work status, and marital status. Figure 1 shows that age, sex, and marital status were directly related to SA, with marital status being the central variable of the network and also associated with both age and sex. Additionally, the education variable was isolated from the network. More specifically, SA exhibited the highest positive expected influence (EI = −1.50) in the network (Supplementary Table 2), meaning that changes in the node should produce changes in the overall network in the same direction. Regarding the weights matrix, the variable with the strongest association with SA was sex (weight = -0.3−23), followed by age (weight = −0.226) and marital status (weight = −0.152), with all associations being negatively related to each other (Supplementary Table 3).Fig. 1Gaussian graphical model networks of variables associated with lifetime suicide attempts across sociodemographic, clinical, psychopharmacological, and integrated domains.a Network including sociodemographic variables. “Age”: group age, “Ed”: educational level, “MS”: marital status, “SA”: history of a suicide attempt, “Sex”: biological sex, “WS”: working status. b Network including clinical variables. “AA”: alcohol abuse lifetime, “FHS”: family history of suicide, “Ho”: number of hospitalizations, “Inp”: inpatient status, “PDO”: psychiatric disorder’s age onset, “PC”: peri-partum complications, “UPD”: untreated psychiatric disorder duration, “SA”: history of a suicide attempt. c Network including psychopharmacological variables. “Ad”: antidepressive treatment (lifetime), “AdD”: antidepressive treatment duration, “Ap”: antipsychotic treatment (lifetime), “ApD”: antipsychotic treatment duration, “MSt”: mood stabilizers treatment (lifetime), “MStD”: mood stabilizers treatment duration, “SA”: history of a suicide attempt, “TD”: treatment duration. d Integrated network including the variables most strongly associated with suicide attempts across the three thematic domains. “AA”: alcohol abuse lifetime, “Age”: group age, “MS”: marital status, “MStD”: mood stabilizers medication duration, “PDO”: psychiatric disorder’s age onset, “SA”: history of a suicide attempt, “Sex”: biological sex. In all networks, edge color indicates the direction of the association (blue = positive partial correlation, red = negative partial correlation), while edge thickness represents the strength of the association, with thicker edges indicating stronger relationships. Gaussian graphical model networks of variables associated with lifetime suicide attempts across sociodemographic, clinical, psychopharmacological, and integrated domains. a Network including sociodemographic variables. “Age”: group age, “Ed”: educational level, “MS”: marital status, “SA”: history of a suicide attempt, “Sex”: biological sex, “WS”: working status. b Network including clinical variables. “AA”: alcohol abuse lifetime, “FHS”: family history of suicide, “Ho”: number of hospitalizations, “Inp”: inpatient status, “PDO”: psychiatric disorder’s age onset, “PC”: peri-partum complications, “UPD”: untreated psychiatric disorder duration, “SA”: history of a suicide attempt. c Network including psychopharmacological variables. “Ad”: antidepressive treatment (lifetime), “AdD”: antidepressive treatment duration, “Ap”: antipsychotic treatment (lifetime), “ApD”: antipsychotic treatment duration, “MSt”: mood stabilizers treatment (lifetime), “MStD”: mood stabilizers treatment duration, “SA”: history of a suicide attempt, “TD”: treatment duration. d Integrated network including the variables most strongly associated with suicide attempts across the three thematic domains. “AA”: alcohol abuse lifetime, “Age”: group age, “MS”: marital status, “MStD”: mood stabilizers medication duration, “PDO”: psychiatric disorder’s age onset, “SA”: history of a suicide attempt, “Sex”: biological sex. In all networks, edge color indicates the direction of the association (blue = positive partial correlation, red = negative partial correlation), while edge thickness represents the strength of the association, with thicker edges indicating stronger relationships. The second network comprises the following clinical variables: hospitalizations, inpatient status, untreated disorder duration, alcohol abuse, partum complications, disorder’s age onset, and family history of suicide (Fig. 1). The analysis revealed that alcohol abuse, disorder age onset, partum complications, inpatient status, and untreated disorder duration were directly related to SA. Interestingly, these variables also constitute a cluster of highly related net of connections, with alcohol abuse and partum complications being the central ones. In this network, partum complications (EI = 1.86) and untreated disorder duration (EI = −1.45) had the highest expected influence (Supplementary Table 4), reflecting their overall directional strength in connecting to other variables. The weights matrix analysis revealed that the variables most strongly associated with SA were alcohol abuse (weight = 0.86), disorder age onset (weight = −0.820), and partum complications (weight= 0.559) (Supplementary Table 5). The third network included the following psychopharmacological variables: antipsychotic therapy and duration (lifetime), antidepressive therapy and duration (lifetime), mood stabilizers therapy and duration (lifetime), and overall psychopharmacological treatment duration (lifetime) (Fig. 1). The findings of this third network indicate that the duration of mood stabilizers was the only variable directly associated with SA. Antidepressant therapy (lifetime) is the central variable of the network, being related to antipsychotic and mood stabilizers therapies (lifetime) as well as to antipsychotic duration and overall psychopharmacological treatment duration (lifetime). The influence analysis within this network reveals that antidepressive therapy had the highest expected influence (EI = −1.90), followed by antipsychotic duration (EI = 0.69) (Supplementary Table 6). Moreover, SA and mood stabilizers duration exhibited a weak negative association (weight = −0.10) (Supplementary Table 7). The fourth network comprises the six variables most strongly related to SA among the nineteen included in the three previous thematic networks (Fig. 1). More specifically, of the five sociodemographic variables included in the first network, sex, age, and marital status were retained. Of the seven variables in the clinical network, alcohol abuse and disorder age onset were also included in this network. Additionally, duration of mood stabilizers was retained from the seven psychopharmacological variables in the third network. Results indicated that the SA variable was central in the network, being connected to all the other variables except for age. Additionally, sex, disorder age onset, alcohol abuse, and SA resulted in a cluster composed of variables that were all highly related to each other, while marital status, age, and mood stabilizers therapy (lifetime) were more distant but not isolated. The expected influence analysis highlights that alcohol abuse (EI = 2.06) and SA (EI = -0.85) were the most influential variables in the network (Supplementary Table 8). Supplementary Table 9 shows that alcohol abuse (weight = 0.93), sex (weight = −0.89), and disorder age onset had the strongest associations with SA. Discussion The present study employed network analyses to investigate the relationships between sociodemographic, clinical, and psychopharmacological variables and a history of a SA in a sample of individuals suffering from a mental disorder and belonging to multi-affected families. Among the various risk factors examined, alcohol abuse, biological sex, and the age of onset of psychiatric disorders were found to be the most significant variables related to SAs, indicating their critical role in influencing suicidal risk. Comparisons between individuals receiving psychiatric care and the general population have already demonstrated a higher risk of exhibiting suicidal behaviors among those with mental disorders20. However, the extent to which belonging to a multi-affected family increases the risk of suicide among individuals with a psychiatric diagnosis, and the specific variables involved, remain to be elucidated. In this regard, beyond factors strictly related to genetic transmission, the impact of certain life events on each individual may play an important role in modulating suicide risk. For example, adults who have experienced bereavement by suicide are 65% more likely to attempt suicide than those bereaved by natural causes37. Moreover, the relatives of individuals with chronic mental disorders or neurodegenerative diseases often experience a lack of tangible support, emotional support, kindness, and positive social interactions38,39, which could be exacerbated by suffering from a psychiatric disorder. Consequently, it is crucial to take these factors into account when assessing the impact of specific variables on the risk of suicide. The network analysis on sociodemographic information and SAs revealed age and sex as the main relevant factors. Specifically, the findings indicated a negative correlation between these two variables and the propensity for SAs, suggesting that being an adult (18-40 years old) and being female were more frequently observed among individuals with a history of SA in this sample. These results are consistent with the literature on suicide rates and age, which indicates that, although suicide rates generally increase with age, several countries are facing rising suicide rates among young people40. Furthermore, compared to the general population, individuals suffering from psychiatric disorders are more likely to attempt suicide at a younger age41. Interestingly, the observed association between female sex and a history of non-fatal SAs aligns with the gender paradox in suicide, which posits that, despite males dying more from suicide, females attempt suicide more frequently42,43. The gender differences observed in SAs, with a higher prevalence among females, are also consistent with previous evidence that included individuals receiving psychiatric care from Italian healthcare facilities44. Furthermore, this network showed a direct correlation between marital status and SA, suggesting that individuals who have never been married and those who are divorced or widowed face an elevated risk of attempting suicide compared with those who are married45. Notably, marital status functions as a mediator in the association between working status and SAs, highlighting the potential of both occupational and marital status as indicators of social integration and loneliness, which may ultimately contribute to the onset of suicidal behavior. Furthermore, contrary to prior findings46, education level was isolated in our initial network, not exerting any influence on SA as well as on the other variables of the network. Notably, the majority of participants in this study were assessed in outpatient or ambulatory care contexts, which present specific challenges and opportunities for suicide risk assessment. In such contexts, clinicians frequently face constraints in terms of time and the infrequency of interactions with patients. Consequently, there is a need to adopt approaches that are collaborative, person-centered, and sensitive to contextual factors influencing suicide risk47. Standardized assessment tools should, therefore, incorporate both static and dynamic risk factors, protective factors, and recent changes in psychosocial circumstances. Furthermore, outpatient services can be strategically positioned to implement ongoing monitoring, brief targeted interventions, and linkage to community resources. As an example of this approach, the Zero Suicide Model incorporates these principles, intending to foster continuity of care and facilitate early detection of escalating risk48. Given the complexity of suicidal behavior, integrating the network-analytic perspective into outpatient assessments could facilitate the identification of key interconnected risk nodes that may otherwise remain undetected in a purely symptom-focused evaluation. Additionally, a notable strength of network analysis, as compared with conventional regression methods, is its capacity to delineate both direct and indirect associations among variables, which enables the identification of interdependencies and potential mediating pathways that may be overlooked by single-predictor models31,32,49. For example, the present study elucidated how specific factors, including marital or working status, exert an indirect influence on SAs through their association with more central variables, such as age, ultimately providing a more dynamic and clinically informative perspective on suicide risk. However, it should be noted that the small sample size of this study significantly limits the stability and generalizability of the findings. Regarding the clinical variables associated with a history of a SA, the second network exhibited a greater number of direct and indirect associations in comparison to the sociodemographic network. Specifically, alcohol abuse and the age of onset of the psychiatric disorder were found to be strongly associated with a previous SA, followed by peri-partum complications, inpatient status, and a relative’s history of suicide death. While alcohol abuse has been identified as a well-established risk factor for a SA50,51, to our knowledge, no prior studies have examined the specific age onset of the mental disorder and peri-partum complications as predictors of suicidal behaviors, suggesting their potential as risk factors among those with a psychiatric disorder. Furthermore, inpatient status was associated with prior SAs and, in this network, was statistically positioned between the number of previous hospitalizations and suicidal behaviors, although this cross-sectional design precludes determining whether it mediates this relationship. This finding aligns with prior studies showing the elevated prevalence of suicide among inpatient individuals35, with the risk being particularly high in patients experiencing short hospital treatment52. However, these findings should be interpreted with extreme caution, as the probability of being hospitalized for a prolonged period increases exponentially if a person has been previously admitted to the hospital for a SA. This complicates the determination of whether the inpatient status in our sample was directly attributable to a previous SA. A family history of suicide exhibited direct associations with previous SAs, also acting as a mediator for both the number of hospitalizations and the untreated duration of the disorder’s indirect relationships with SAs. Prior studies have indicated the potential for familial aggregation of suicidal behavior to occur independently of the transmission of psychiatric disorders, thereby suggesting both genetic and environmental contributions6. In the present sample, a family history of suicide was associated with several clinical variables, including alcohol abuse, peri-partum complications, and inpatient status. These results suggest that a family history of suicide may serve as a contextual factor that exacerbates the development of a broader constellation of variables. Indeed, in multi-affected families, exposure to suicide within close relatives may lead to the normalization of self-destructive behavior, an increase in vulnerability through shared environmental stressors, or an intensification of grief and stigma irrespective of the prevalent psychiatric diagnosis in the family53. This, in turn, may influence both the onset and trajectory of psychiatric illness. Future studies with increased sample sizes may investigate whether family history of suicide alters the network structure of risk and protective factors, thereby potentially identifying family-level targets for prevention. Furthermore, the network analysis on psychopharmacological therapies highlighted that the only variable directly associated with a previous SA was the duration of the mood stabilizers therapy. This finding is consistent with the prevailing use of mood stabilizers therapies for the treatment of major affective disorders, which are the most prevalent diagnoses in our sample and are known to be associated with an elevated risk of suicide attempts11,12,54. However, this result also indicates that duration plays a crucial role in this regard, thus suggesting how chronicity of affective disorders has a substantial influence on suicide risk, as confirmed by previous studies55,56. Additionally, the analyses revealed an indirect correlation between prolonged antidepressant use and prior SAs, with the duration of mood stabilizers therapy acting as a potential mediator of this relationship. This pattern may reflect that individuals with a longer history of mood stabilizer use were more likely to have had a history of SA, whereas those who had taken antidepressants over their lifetime were more likely to have attempted suicide if they had also been using mood stabilizers for some time, although causality and temporal ordering cannot be inferred. These findings suggest that when performing suicide risk assessment, a patient’s psychopharmacological history may be a relevant aspect to take into account, along with the specific diagnosis and other well-established risk factors associated. Notably, an additional consideration not captured within the present dataset concerns the role of psychological interventions in suicide prevention for individuals with psychiatric disorders. The suicide risk management targeting clinical populations, therefore, should go beyond pharmacological management, with robust evidence supporting a range of psychotherapeutic approaches. Guidelines from the National Institute for Health and Care Excellence57 recommend structured psychological therapies, such as cognitive behavioral therapy, dialectical behavior therapy, as well as other psychosocial interventions for individuals at risk. These interventions have demonstrated efficacy in reducing suicidal ideation, SAs, and associated symptomatology transdiagnostically58,59. Within the context of multi-affected families, psychotherapeutic interventions may be particularly valuable in addressing relational dynamics, stigma, and chronic stressors that pharmacological treatment alone cannot fully target. Future research integrating network-analytic findings with data on psychological treatment engagement and outcomes may provide a more comprehensive understanding of protective processes and inform multidimensional care strategies. The combination of the sociodemographic, clinical, and psychopharmacological factors most associated with a previous SA enabled the estimation of the final overall network. This network indicated that a particularly influential role was played by the following variables: female sex, young age of onset of the disorder, and alcohol abuse, showing the strongest statistical relationships with each other and with a previous SA. The duration of the mood stabilizers therapy, as well as the marital status, were also directly associated with a history of a SA, even if their relationships were weaker. Interestingly, contrary to what emerged in the sociodemographic thematic network, in this final network, the variable age did not exhibit a direct association with SA but was only associated with it through the mediation of the variables age onset, mood stabilizers duration, and marital status. This result highlights that, in this case, the age variable seemed to exert a significant relevance concerning the risk of suicide when considering only sociodemographic variables in the network, but not when other highly influential variables were considered. Similarly, the influence exhibited by mood stabilizers’ duration and marital status decreased significantly in this overall network. Regrettably, although the present study identified variables most strongly associated with a history of SAs, these findings should be interpreted within a broader conceptual framework. Indeed, suicidal behavior is increasingly conceptualized as a complex, dynamic, and relational phenomenon, shaped not only by individual-level factors but also by structural and contextual determinants60,61. The observed associations between female sex, early onset of psychiatric disorder, and alcohol abuse may reflect the interaction of lived experience with gendered social roles, stigma, economic disadvantage, and inequitable access to mental health care. Conversely, being raised in a multi-affected family may result in protracted exposure to stressors, interpersonal losses, and diminished social support. These factors can influence the emergence and progression of psychiatric illnesses, as well as the individual’s willingness to adhere to treatment. Network analysis provides a methodological framework that enables the transition from a static “risk factor checklist” to the identification of interdependent nodes that may serve as actionable targets within a complex biopsychosocial system. Integrating these findings with therapeutic assessment approaches62 and recovery-oriented models of care aligns with calls to shift from prediction to preventive, collaborative, and contextually informed approaches63,64. The complex interactions between variables belonging to different thematic domains and the changing influence of each single variable within the other networks further corroborate the importance of a holistic approach when assessing suicide risk. Indeed, this process should encompass a range of preventive strategies and evaluation methods, thereby adopting a multidimensional approach65–67. In this context, contemporary perspectives in suicidology highlight the limitations of traditional suicide risk assessment in mental health services68 and advocate for a shift from risk prediction to a care-based, safety-oriented approach grounded in a biopsychosocial understanding of suicidal behavior47. Within this paradigm, reducing rates of suicidal behavior requires prioritizing the implementation of prevention policies and programs as part of a structured national prevention strategy, aimed at promoting mental health and social well-being both within and beyond the healthcare system. The present study presents important limitations that can guide future research in this field. First, a cross-sectional research design was employed, which entails the collection of variables at a single point in time, thereby precluding the possibility of prospective predictive utility. Therefore, all identified relationships should be interpreted as statistical associations rather than directional or causal effects. Future longitudinal works may investigate the temporal relationships between variables and potential causal pathways leading to SAs. Second, a considerable rate of missing data was registered, especially regarding psychopharmacological therapy information. Although pairwise deletion was employed to maximize the available information for each analysis, this approach assumes that data are missing completely at random, which may not hold in this context. If missingness is related to unmeasured clinical characteristics, such as illness severity, treatment adherence, or recall bias, the resulting networks could underrepresent or distort certain associations. Future studies should aim to reduce missingness through standardized data collection and consider alternative analytic strategies, such as multiple imputation, to assess the robustness of the findings. Additionally, no data were collected on psychosocial and psychotherapy interventions, despite their crucial role in the field of mental health care and suicide prevention. Third, despite its novelty in evaluating the risk of suicidal behavior in individuals with psychiatric disorders and multi-affected families, this study’s small sample size and low event rate of the suicidal outcome preclude the possibility of generalizing its results. Indeed, given the low base rates of SAs in our sample, which represents a common methodological challenge in suicidology research69,70, the statistical power to detect stable associations may have been limited. Consequently, the estimated network structure and centrality indices should be interpreted with caution, as even minor changes in the data could substantially affect the relationships between nodes as well as the overall replicability of the network structure, as indicated by previous studies71,72. In light of this limitation, the current results should be viewed as exploratory and primarily hypothesis-generating rather than confirmatory, highlighting the need for future studies with larger samples to confirm and refine these preliminary patterns. Fourth, the investigation into the family history of suicide only considered first-degree relatives’ death by suicide, without considering non-fatal SAs, deliberate self-harm, or non-suicidal self-injury. Fifth, given that the study exclusively relied on retrospective data, it is important to consider the potential for recall bias. Sixth, data collection followed a non-probabilistic convenience sample approach, making the sample size constrained by the availability of eligible patients during the study period, ultimately limiting the generalizability of the findings. Finally, a previous SA was the only outcome considered in the study, without considering all the other relevant suicide-related outcomes. In conclusion, the present study employed network analyses to examine the interplay among the factors associated with the risk of suicidal behavior in people with lived experience of mental disorders belonging to multi-affected families. The overall network showed alcohol abuse, female sex, and the age of onset of the psychiatric disorder as the factors most strongly associated with the risk of SA. The study’s findings carry significant clinical implications, underscoring the necessity of a nuanced, comprehensive understanding of the multifaceted nature of suicidal behavior. Indeed, from a clinical perspective, the network structures identified in this study may inform decision-making by highlighting variables both associated with SAs and embedded in broader constellations of interrelated risk and protective factors. For instance, the crucial role of alcohol abuse, early onset of psychiatric disorders, and female sex suggest that these factors, when in combination, may necessitate more intensive monitoring, targeted psychoeducation, and timely referral to specialized interventions. Network analysis has the potential to assist clinicians in transitioning from static, checklist-based approaches to dynamic, context-sensitive formulations. This approach may integrate a comprehensive set of factors, including sociodemographic, clinical, and treatment-related variables, as well as interpersonal and structural evaluations, which can be used to prioritize modifiable targets, guide collaborative safety planning, and inform the allocation of resources to those at heightened risk. In outpatient and ambulatory care settings, these insights could support brief, focused assessments that remain sensitive to proximal triggers while placing them within the individual’s broader lived experience and support networks73,74. Supplementary information Supplementary information Supplementary information