Introduction

Suicide is a global phenomenon that accounts for 1.4% of premature deaths worldwide (Bachmann, 2018). In mood disorders lifetime risk for suicide is estimated to range from 2 to 8% (Blair-West et al., 1999; Bostwick and Pankratz, 2000; Nordentoft et al., 2011) and mortality from suicide is more than 20 times higher than in the general population (Osby et al., 2001). Among depressive symptoms, suicidal ideation (SI), which has been reported in up to 60% of depressive outpatients (Sokero et al., 2003; Trivedi et al., 2013), is known to substantially increase suicidal risk: a meta-analysis has estimated that suicide was 2.3 times more likely in mood disorder patients who had previously expressed SI than in their counterpart who had not (Hubers et al., 2018). Based on this evidence, treating SI becomes a priority. Although antidepressant medications (ADs) have been debated about self-harm and SI in children and adolescent (Scahill et al., 2005; Hammad et al., 2006; Högberg et al., 2015), in adults they have univocally shown superiority to placebo in reducing SI (Gibbons et al., 2012; Näslund et al., 2018). Despite this, at least one in five patients who take ADs are not likely to obtain a complete SI remission (Zisook et al., 2011). We argue that, if these subjects could be identified as earlier as possible and their treatments enhanced or modified to improve their effectiveness, this strategy would allow to minimise suicidal risk. Prior research indicates several factors that are associated with the occurrence of SI in depressive episodes such as depression severity, age at onset and course of depressive disorder (Zisook et al., 2011), abuse and violence in childhood (Chou, 2012; Zisook et al., 2022) and mental disorder comorbidities (Hardt et al., 2015; Wiebenga et al., 2021). Moreover, SI has been related to depression psychopathology, in particular hopelessness (Wolfe et al., 2017; Ribeiro et al., 2018; Wiebenga et al., 2021) and self-blame (Olgiati et al., 2006), irritability (Jha et al., 2019), insomnia (Pigeon et al., 2012; Wang et al., 2019) and bipolar features (Akiskal et al., 2005; Weinstock et al., 2016; Rihmer and Rihmer, 2019). Little is known about the role of these clinical variables in predicting SI response during antidepressant treatment. To address this question, we analysed data from the Combining Medications to Enhance Depression Outcome (CO-MED) study (Rush et al., 2011) in order to ascertain which characteristics of major depressive disorder (MDD) assessed at baseline were associated with a lack of SI remission after six treatment weeks.

Methods Subjects

In CO-MED trial, eligible subjects were aged 18–75 years, with a primary DSM-IV diagnosis of nonpsychotic MDD and a HAM-D17 score of at least 16 (Rush et al., 2011). To be included in the current analysis, two additional criteria were required: (a) to endorse SI at baseline and (b) to have no missing visit during follow-up period until week 6.

Treatment and data collection

The CO-MED trial was characterised by a single-blind placebo-controlled design and three treatment arms: (a) escitalopram plus placebo; (b) bupropion SR plus escitalopram; (c) venlafaxine XR plus mirtazapine. The trial included a short-term (12 weeks) treatment followed by a continuation phase (weeks 12–28) (Rush et al., 2011).

Research data were collected by a variety of assessment tools as reported in our previous publications (Olgiati et al., 2022; Olgiati and Serretti, 2022; Serretti et al., 2021): (a) a socio-demographic form including age, sex, ethnic group, education and monthly income; (b) the Mini International Neuropsychiatric Interview (M.I.N.I) (Sheehan et al., 1998) for retrospective evaluation of depressive disorder (chronic or recurrent course of depression; number of depressive episodes and age at onset of first episode) and to ascertain the lifetime occurrence of subthreshold hypomanic episodes (see below); (c) a battery of scales for cross-sectional clinical assessment including the 30-item Inventory of Depressive Symptomatology-Clinician Rating (IDS-C30) (Corruble et al., 1999) and the 16-item Quick Inventory of Depressive Symptomatology (Rush et al., 2003), the Concise Associated Symptoms Tracking (CAST) (Trivedi et al., 2011b) for irritability, the Altman Self-Rating Mania Scale (ASRM) (Altman, 1998) and the Work and Social Adjustment Scale (WSAS) (Mundt et al., 2002) to ascertain functional impairment; (d) the Psychiatric Diagnostic Screening Questionnaire (Zimmerman and Mattia, 1999) to investigate comorbid mental disorders including panic disorder, generalised anxiety, obsessive-compulsive disorder, social phobia, posttraumatic stress disorder, and alcohol and substance use disorders; and (e) two questionnaires that were specifically developed for CO-MED project to investigate lifetime suicidal behaviour and experiences of neglect and abuse during childhood (Serretti et al., 2021; Olgiati and Serretti, 2022).

Suicidal ideation

SI was assessed by means of the Concise Health Risk Tracking-Self Report (CHRT) (Trivedi et al., 2011a). This version was composed of 12 items (overall score: 0–48), of which nine explored a generic suicidal propensity. Our analysis was based on the last three, which specifically investigated SI and plans over the last 24 h: scoring more than 1 at baseline was used as an inclusion criterion (see paragraph ‘subjects’) whereas a threshold less than 2 was selected to define SI remission at week 6.

Clinical variables

A number of clinical variables were compared between patients with and without SI remission: (a) chronic depression: depressive episode lasting for at least 6 months; (b) baseline depression severity (IDS-C30 total score) and symptom profile including negative self-outlook, negative outlook of future, loss of pleasure, irritability (CAST), anxious mood, interpersonal sensitivity; poor concentration, difficulty in falling asleep, middle nocturnal insomnia and early awakening. Subsequently, we achieved a combined variable by pooling baseline scores of SI and middle nocturnal insomnia together: subjects scoring more than 8 were classified as ‘SI plus insomnia’ group and compared with the rest of the sample (see below); (c) bipolar spectrum variables, including: (i) age of onset (Benazzi, 2009); (ii) mood disorder recurrence (number of episodes/illness years) (Mazzarini et al., 2018); (iii) hypomanic symptoms occurring within major depressive episode (ASRM items): cheerfulness; self-confidence; reduced need for sleep; talkativeness; and goal-oriented hyperactivity; (iv) mixed depression, defined as a major depressive episode with three or more concurrent hypomanic symptoms (Akiskal et al., 2005); (v) lifetime subthreshold hypomania: a period of elated or irritable mood with at least two concurrent hypomanic symptoms (MINI interview), which did not fulfil DSM criteria for hypomanic/manic episode (Angst et al., 2003; Serretti et al., 2021); (d) lifetime most severe suicidal behaviour, rated via a 0 (no suicidal tendency) to 7 (suicide attempt with definite intent to die) scale; (e) comorbid mental disorders; and (f) experiences of neglect and abuse during childhood.

Statistical analyses

A preliminary power analysis was conducted to estimate minimum detectable differences between SI remitter and nonremitter groups, considering a type I error (alpha level) of 0.05 and a type II error (1-power) of 0.20. Univariate analyses were performed using Student’s t-test and Chi-square test for continuous and categorical variables respectively. Significance threshold was set at alpha = 0.05, without a formal correction for multiple testing given the exploratory nature of our analysis (Amrhein et al., 2019). Multiple logistic regression analysis was used to test the association of each factor with SI remission by controlling for confounding variables (see Tables 1–3 for details). Statistical software was OpenStat version 8 December 2014 (https://openstat.info/OpenStatMain.htm). Power analysis was carried out via G*Power 3 program that was developed by a Düsseldorf University team (Germany) (Faul et al., 2007). Finally, to analyse the performance of ‘SI plus insomnia’ variable in predicting the lack of SI remission, a diagnostic test evaluation was conducted by means of Med-Calc free statistical calculator (https://www.medcalc.org/).

Results

Out of 665 participants in the CO-MED trial, our analysis involved 198 outpatients who endorsed SI and attended all scheduled visits until week 6. Their mean age was 44.19 + 11.53 years; 61 patients (31%) were male and 140 (71%) were of Caucasian origin. Overall depression score (IDS-C30) was 40.45 + 9.13 at baseline and 22.65 + 12.14 after 6 weeks of AD use (t = 18.72; P < 0.0001). SI score (CHRT items 10–12), which was equal to 4.10 + 2.34 at baseline, decreased to 1.78 + 2.25 at week 6 (t = 10.05; P < 0.0001), although 78 patients (39%) failed to achieve SI remission. Power analysis suggested that this sample might have an adequate power (0.8) to detect a small-medium effect size (d = 0.45) or differences of at least 10.5% between comparison groups.

Role of demographic characteristics and depression severity

Patients in whom SI remitted were similar to their counterparts without remission as regard demographic features. Conversely, at clinical level, no remission group was characterised by higher baseline scores of SI (t = 3.34; P = 0.0010) and middle nocturnal insomnia (t = 2.55; P = 0.0114). These inverse correlations with SI remission were confirmed by multivariate analysis [SI: odds ratio (OR), 0.82; 95% CI, 0.71–0.96; middle nocturnal insomnia: OR, 0.71; 95% CI, 0.53–0.95] along with a direct correlation between lack of SI remission and lower levels of negative self-outlook (OR, 1.56; 95% CI, 1.07–2.27) after controlling for overall depression score (Table 1).

Table 1 - Patients with and without suicidal ideation remission (week 6), demographic variables and depressive symptomatology (baseline) Features SI remission (N = 120) Lack of remission (N = 78) P Age 44.33 ± 11.56 43.96 ± 11.55 0.8252 Sex (M; %) 33 (0.27) 28 (0.36) 0.2110 Ethnic group (Caucasian; %) 89 (0.74) 51 (0.65) 0.1850 Education (years) 13.79 ± 3.22 13.59 ± 2.92 0.6613 Chronic depression (>6 months) 62 (0.51) 45 (0.57) 0.4060 IDS-C30 Total score 39.77 ± 9.10 41.50 ± 9.14 0.1925 Negative self-outlook 2.09 ± 0.84 1.90 ± 0.91 0.1249 Negative outlook of future 1.71 ± 0.86 1.76 ± 0.87 0.7033 Loss of pleasure 1.61 ± 0.87 1.69 ± 0.92 0.5172 Anxious mood 1.90 ± 0.80 1.79 ± 0.74 0.3558 Psychomotor agitation 0.65 ± 0.64 0.64 ± 0.58 0.9208 Slowing 0.78 ± 0.64 0.63 ± 0.61 0.0895 Interpersonal sensitivity 1.50 ± 1.04 1.62 ± 1.11 0.4574 Difficulty in falling asleep 1.80 ± 1.25 1.95 ± 1.29 0.4194 Middle nocturnal insomnia 1.91 ± 1.12 2.31 ± 1.01 0.0114* Early awakening 1.35 ± 1.25 1.44 ± 1.32 0.6447 Poor concentration 1.78 ± 0.69 1.72 ± 0.68 0.5129 Irritability (CAST) 0.78 ± 0.41 0.76 ± 0.43 0.6606 Suicidality (CHRT) Propensity 19.94 ± 6.78 21.42 ± 6.48 0.1280 Ideation/planning 3.66 ± 2.41 4.77 ± 2.08 0.0010* Multiple logistic regression: dependent variable: SI remission; Chi-square = 22.23; df = 5; P = 0.0005 Predictors OR (95% CI) Negative self-outlook 1.41 (0.98–2.03) Slowing 1.49 (0.92–2.42) Middle nocturnal insomnia 0.71 (0.53–0.95)* Suicide propensity 0.98 (0.94–1.04) Suicidal ideation/planning 0.82 (0.71–0.96)* Only variables with P < 0.15 were included in MLR model MLR: dependent variable: SI remission; confounding factor; overall depression Chi-square = 7.23; df = 2; P = 0.0269 Predictors OR (95% CI) Overall depression score 0.96 (0.92–0.99) Negative self-outlook 1.56 (1.07–2.27)

CAST, Concise Associated Symptoms Tracking; CHRT, Concise Health Risk Tracking-Self Report; CI, confidence interval; MLR, multiple logistic regression; OR, odds ratio; SI, suicidal ideation.

*P < 0.05.

Comorbidities, childhood traumas and lifetime suicidal behaviour

Further characteristics of no remission group included more comorbid anxiety disorders (panic disorder: t = 2.990; P = 0.0031; social phobia: t = 2.323; P = 0.0212), more traumatic experiences during childhood (t = 1.980; P = 0.0491) and increased lifetime suicidality (t = 2.196; P = 0.029). After controlling for baseline depression score and SI, only panic disorder (OR, 0.93; 95% CI, 0.87–0.99) confirmed its negative association with SI remission (Table 2).

Table 2 - Patients with and without suicidal ideation remission Features With SI remission (N = 120) Without SI remission (N = 78) P Work/social impairment 27.58 ± 8.52 29.86 ± 7.24 0.0531 Comorbidities Panic disorder 3.85 ± 4.57 5.99 ± 5.40 0.0031* GAD 6.85 ± 2.82 7.35 ± 2.53 0.2095 PTSD 3.29 ± 4.09 4.12 ± 3.99 0.1640 Social phobia 4.91 ± 5.10 6.62 ± 4.97 0.0212* Alcohol 0.56 ± 1.28 0.62 ± 1.83 0.2080 Substance 0.22 ± 0.81 0.38 ± 1.16 0.2333 Lifetime suicidal behaviour 2.33 ± 2.36 3.11 ± 2.41 0.0293* Childhood traumas (total N°) 1.32 ± 1.41 1.73 ± 1.47 0.0491* Neglect 51 (0.42) 40 (0.51) 0.2268 Emotional abuse 53 (0.44) 45 (0.58) 0.0636 Physical abuse 27 (0.22) 24 (0.31) 0.1947 Sexual abuse 27 (0.22) 26 (0.33) 0.0933 Multiple logistic regression analysis Predictors OR (95% CI) Work/social impairment 0.97 0.93–1.01 Panic disorder 0.93 0.87–0.99* Social phobia 0.95 0.89–1.01 Lifetime suicidal behaviour 0.92 0.81–1.04 Number of childhood traumas 0.88 0.71–1.08 Emotional abuse 0.69 0.34–1.25 Sexual abuse 0.69 0.35–1.34

Comorbidities, lifetime suicidal behaviour and experiences of abuse and neglect during childhood.

MLR was performed to test all variables with P < 0.15 separately. Each MLR model included the following parameters: dependent variable: SI remission; covariates: baseline SI and overall depression scores.

CI, confidence interval; MLR, multiple logistic regression; OR, odds ratio; PTSD, post traumatic stress disorder; SI, suicidal ideation.

*P < 0.05.

Bipolar spectrum variables

The distributions of hypomanic symptoms and bipolar spectrum markers are displayed in Table 3. Patients who failed to achieve SI remission endorsed higher levels of self-confidence (t = 2.528; P = 0.0123) and reduced need for sleep (t = 2.430; P = 0.0165) during depressive episode in comparison with remission group. The negative association between hypomanic symptoms and SI remission was confirmed after controlling for confounding effects exerted by depression severity and baseline SI (self-confidence: OR, 0.52; 95% CI, 0.32–0.82; reduced need for sleep: OR, 0.75; 95% CI, 0.58–0.99; cheerfulness: OR, 0.57; 95% CI, 0.33–0.98).

Table 3 - Bipolar spectrum validators and suicidal ideation remission Features SI remission (N = 120) No remission (N = 78) P Age of depression onset 23.03 ± 13.84 14.01 ± 12.35 0.2967 N. episodes/illness years 0.44 ± 0.72 0.33 ± 1.16 0.4352 Mixed depression (MxD) 14 (0.11) 12 (0.15) 0.4523 Subthreshold hypomania (SH) 15 (0.12) 7 (0.09) 0.4410 Irritability (CAST) 0.78 ± 0.41 0.76 ± 0.45 0.6606 Cheerfulness (ASRM) 0.16 ± 0.47 0.31 ± 0.73 0.0807 Self-confidence (ASRM) 0.14 ± 0.44 0.37 ± 0.84 0.0130* Increased talkativeness (ASRM) 0.42 ± 0.81 0.26 ± 0.57 0.1284 Reduced need for sleep (ASRM) 0.45 ± 0.93 0.83 ± 1.28 0.0165* Increased activity (ASRM) 0.09 ± 0.43 0.24 ± 0.78 0.0798 Multiple logistic regression models were performed for each variable with P < 015; each model included SI remission as dependent variable and baseline SI and depression severity as covariates. Predictors of SI remission OR (95% CI) Cheerfulness 0.57 (0.33–0.98)* Self-confidence 0.55 (0.33–0.91)* Reduced need for sleep 0.76 (0.56–0.99)* Increased talkativeess 1.43 (0.91–1.25) Increased activity 0.61 (0.35–1.06)

ASRM, Altman Self-Rating Mania Scale; CAST, Concise Associated Symptoms Tracking; CI, confidence interval; OR, odds ratio; SH, suicidal ideation.

*P < 0.05.

Suicidal ideation plus insomnia

Fifty-six patients (28%) had baseline SI plus insomnia; in comparison with the rest of the sample, this group was younger at illness onset (P = 0.0069) and associated with higher levels of overall depression (P = 0.0007), comorbid generalised anxiety (P = 0.0079) and post traumatic stress disorder (P = 0.0006), traumatic experiences during childhood (P = 0.0009) and worse suicidal behaviours (0.0004). In addition, these patients were less likely to experience SI remission at week 6 (P = 0.0014) (Table 4). The combination of baseline SI and insomnia predicted the lack of SI remission with a specificity of 80% (95% CI, 72–87%) and a negative predictive value of 68% (95% CI, 63–72%).

Table 4 - Characteristics of suicidal ideation (SI) plus insomnia (In) group Features SI+In (n = 56) Rest of the sample (n = 142) P Age of onset 18.20 ± 12.95 23.33 ± 13.11 0.0069* Depression severity 43.93 ± 9.08 39.08 ± 8.81 0.0007* Generalised anxiety 7.86 ± 2.53 6.73 ± 2.73 0.0079* PTSD 5.18 ± 4.19 3.00 ± 3.86 0.0006* Childhood traumas 2.02 ± 1.58 1.27 ± 1.34 0.0009* Lifetime suicidal behaviour 3.62 + 2.36 2.26 ± 2.32 0.0004* SI remission (week 6) 24 (0.43) 96 (0.68) 0.0014* Lack of SI remission Parameter Value 95% CI Sensitivity 41.03% 30.01–52.75% Specificity 80.00% 71.72–86.75% Positive likelihood ratio 2.05 1.31–3.20 Predicted (SI+In test) Negative likelihood ratio 0.74 0.60–0.91 Observed No Yes Total Lack of SI remission prevalence 39.00% No 96 24 120 Positive predictive value 56.74% 45.64–67.20% Yes 46 32 78 Negative predictive value 67.97% 63.34–72.27% Total 142 56 198 Accuracy 64.80% 57.71–71.44%

Baseline SI + middle nocturnal insomnia scores ≥8.

CI, confidence interval; PTSD, post traumatic stress disorder; SH, suicidal ideation.

*P < 0.05.

Discussion

CO-MED data had already been used to analyse the prevalence and clinical correlates of SI (Zisook et al., 2011; Jha et al., 2020) as well as its prognostic impact along with other markers of depression severity (Zisook et al., 2011; Sung et al., 2013). Here, we focussed on baseline variables that could predict the persistence of SI during acute phase treatment. The severity of SI within the baseline assessment was strongly related to its lack of remission after 6 weeks. This was consistent with a large amount of literature displaying a negative correlation between the initial severity of depressive symptoms and the likelihood of remitting during treatment (Riedel et al., 2011; Zisook et al., 2019). Hence, it is arguable that depressed patients who report severe SI at baseline, similarly to those with higher levels of overall depression, may need augmentation or combination strategies to enhance antidepressant treatment response (Henssler et al., 2022). A somehow unexpected result of our study was to find a positive correlation between higher levels of negative self-outlook and SI remission after controlling for depression severity. At first glance, this finding seems to be in contrast with the large body of evidence supporting the connection between negative self-appraisal and suicidality (Madsen and Harris, 2021). Nonetheless, it becomes less difficult to understand thanks to the knowledge that some ADs (e.g. escitalopram and agomelatine) can quickly normalise self-referential processing in depressed patients, an effect which is thought to play an important role in their therapeutical activity (Delaveau et al., 2016; Komulainen et al., 2018). Thus, basically, we suggest that patients who exhibited higher levels of negative self-outlook might be more impaired in self-referential processing and, as such, more sensitive to antidepressant treatment.

The most clear-cut result of our analysis was the association between insomnia and SI response. The combination of insomnia and severe SI before starting treatment, which involved more than one-quarter of our sample, had the strongest connection with SI persistence. As a predicting test, it was characterised by a good level of specificity and a small number (24/198) of false-positive results, at the expense of a lower sensitivity, and it was correlated with lifetime suicidal behaviour. The relationship between insomnia and suicidality has become a matter of extensive research in recent years (Pigeon et al., 2012; Wang et al., 2019). Despite considerable differences in study designs, samples and assessment techniques, there is converging evidence for insomnia as an empirical risk factor for suicidal behaviour (Bernert et al., 2015; Lau et al., 2020; Simmons et al., 2021). On the other hand, it is possible to reduce SI by targeting insomnia with both pharmacological treatment (McCall et al., 2019) as well as cognitive behavioural therapy (Jernelöv et al., 2021). Although the association between insomnia and suicidality can be independent of depression (Simmons et al., 2020), it appears to be stronger in depressed patients, among whom a direct correlation has been reported between higher levels of insomnia and the intensity of suicidal thinking (McCall et al., 2010; Bryan et al., 2015). This connection might not be limited to a short time horizon as in our study; in a longitudinal analysis of the NESDA study, for example, insomnia was associated with SI persistence for 6–9 years (Kivela et al., 2019). Exact pathophysiological mechanisms are not known yet; it has been hypothesised that sleep loss could increase susceptibility to suicidal thoughts and behaviours via its association with executive function deficits (Fernandes et al., 2021) and cognitive hyperarousal (Fernandes et al., 2021; Grove et al., 2021). The negative role of panic disorder during SI treatment was also consistent with available literature. In fact, among depressed patients, comorbid panic disorder has emerged as a risk factor for suicidality (Katz et al., 2011; Scheer et al., 2020; Zhang et al., 2022) as well as a poor prognostic factor (Buckman et al., 2021). The contribution of bipolar symptoms to convey an additional risk for suicidal behaviours during depressive episode is yet unclear, with studies yielding mixed results. For example, two prospective analyses of the Finnish Jorvi Bipolar Study have shown higher estimates of suicide attempts with mixed symptoms relative to depressive ones (Valtonen et al., 2008; Holma et al., 2014). Conversely, in a series of works based on the National Network of Depression Centres Mood Outcomes Program, mixed depressive and manic symptoms were associated with no excess risk for SI or behaviour (Fiedorowicz et al., 2021; Persons et al., 2022). Notably, prior studies analysed patients with bipolar depression (Valtonen et al., 2008; Persons et al., 2022) or mixed bipolar and unipolar samples (Holma et al., 2014; Fiedorowicz et al., 2021), whereas we only included subjects with MDD.

The CO-MED trial was characterised by a well-implemented design and a thorough clinical assessment. Despite this, we must acknowledge some limitations of our post hoc analysis. First, the group of patients without SI remission was relatively small and, therefore, underpowered to detect less clearcut differences from remitting subjects which, however, should be included in reliable prediction models (Abraham and Russell, 2008). A set of caveats involved th