Symptom trajectories in the months before and after a suicide attempt in individuals with bipolar disorder: A STEP-BD study

Abstract

Objectives:

The suicide crisis is a relatively short-lived psychiatric emergency, with transient symptoms that ebb and flow around the suicide attempt. Understanding the dynamic processes of symptoms before and after suicide attempt may aid future prevention efforts.

Methods:

Data were drawn from the NIMH STEP-BD study, which followed 4,360 patients with bipolar disorder; a subset attempted suicide during the trial (245/4100 or 5.97% of the sample eligible for analysis). This analysis focused on change in suicidal ideation (SI) in the 120 days before and 120 days after suicide attempt; similar analyses were conducted for other depressive symptoms. Generalized linear mixed models with a two-piece linear function of time corresponding to pre- and post-suicide attempt trends were used.

Results:

SI ratings from 216 individuals were analyzed (n = 1,231 total; n = 395 pre-attempt, n = 126 circa-attempt, n = 710 post-attempt) and compared to data from a matched sample of 648 non-attempters. SI worsened in the 120 days pre-attempt but improved afterwards, reaching non-attempter levels by 90 days post-attempt. A similar pattern was found for other depressive symptoms, including depressed mood, loss of interest, guilt, and self-esteem. Pre/post differences in tension/activating symptoms of depression—anxiety, agitation, and irritability—were less pronounced and more time-limited.

Conclusions:

The suicide crisis is dynamic, and the days before and after suicide attempt may be particularly critical. The findings extend previous research on proximal symptoms of suicide and underscore that some SI and affective/cognitive symptoms of depression can remain elevated up to 90 days post-attempt in individuals with bipolar disorder.

1 |. INTRODUCTION

Suicide is a major public health concern and a leading cause of death.¹ In the United States, suicide rates have increased 33% over the past 15 years,^1,2 and an estimated 400 000 individuals present to emergency department settings annually with suicidal thoughts and behaviors.³ Clinicians have few proven resources at their disposal to treat these psychiatric emergencies. The time immediately following hospitalization for suicidal thoughts or behavior—particularly the following weeks and months—is a particularly high-risk time^4,5 and, unfortunately, few established effective interventions are available. Current interventions—which can take weeks or months to exert an effect on suicide risk—include lithium,⁶ dialectical behavioral therapy,⁷ cognitive behavioral therapy,⁸ clozapine,⁹ and electro-convulsive therapy (ECT).¹⁰ Consequently, a critical need exists for interventions that can prevent suicidal behavior in the short term.

The development of short-term, rapid-acting treatments necessitates an understanding of the potential outcome of interest: the active suicide crisis. As compared to psychiatric disorders, which are codified in the DSM,¹¹ the symptomatology and course of an active suicide crisis is relatively unknown. Current psychiatric theories of suicide, such as the diathesis-stress model,¹² or psychological theories in an ideation-to-action framework,¹³ both acknowledge the existence of and difference between chronic and acute suicide risk factors. In other words, some risk factors place an individual at long-term, or chronic, risk of suicide over the course of his/her lifetime,¹⁴ while others—such as suicidal thoughts, plan, or intent—are fairly short-lived risk factors that indicate acute suicide risk in the moment. In this context, the fluid vulnerability theory^15,16 suggests that suicide risk is a dynamic rather than a static process, and that certain symptoms—such as mood or anxiety—ebb and flow around the time of suicidal behavior. These dynamic symptoms, in addition to suicidal ideation, could represent ideal targets for rapid-acting treatments aimed at the short-term prevention of suicidal behavior. Indeed, some researchers have suggested that the active suicide crisis could be described as a distinct clinical entity or diagnosis, such as Acute Suicidal Affective Disturbance or Suicide Crisis Syndrome.^17,18

However, very little is known about the suicide crisis. For example: what does a “typical” suicide crisis look like? How long do symptoms last? What is the symptom trajectory before a suicide attempt? What is the symptom trajectory after an attempt—do symptoms immediately dissipate or do they persist, potentially requiring increased intervention? Is the suicide crisis distinct from a psychiatric episode such as depression? New methods such as timeline follow-back interviews,¹⁹ ecological momentary assessment,²⁰ and biological measures such as hair cortisol concentrations²¹ are beginning to illuminate some of the processes involved in the suicide crisis.

Large clinical trials, such as the NIMH Systematic Treatment Enhancement Program for Bipolar Disorder (STEP-BD) trial, are another key opportunity to study symptom trajectories before and after a suicide crisis. In particular, focusing on the suicide crisis in bipolar disorder (BD) is especially useful, given that BD is associated with a significantly higher risk of suicide than many other psychiatric diagnoses²² and, as such, a number of pre-existing comprehensive multidimensional models of suicide and BD can be used to guide future investigations and treatment studies.²³ In fact, recent neuroimaging studies have suggested that suicide attempters with BD demonstrate changes in regions associated with emotional regulation and self-referential thought processing, which has implications for the symptomatology associated with a suicidal crisis.^24,25 The STEP-BD trial followed over 4,000 patients in both clinical trial and naturalistic treatment settings; over 100 participants in this trial attempted suicide and continued to be followed by study clinicians. Because these BD patients were regularly assessed for a number of psychiatric symptoms, including suicidal ideation (SI), these data permit a naturalistic examination of the active suicide crisis.

A previous analysis by our group demonstrated that SI and loss of interest significantly increased in the months before suicide attempt in the STEP-BD trial²⁶; however, the analysis did not address symptom trajectories after the attempt. The present study sought to describe the course of SI in the 120 days before and after suicide attempt for participants in the STEP-BD trial. The study also explores whether symptom fluctuations surrounding the suicide attempt were suicide-specific or if other depressive symptoms showed similar trajectories. Thus, depressive symptom clusters—including affective/somatic symptoms (depressed mood, loss of interest, sleep difficulties), activating/tension symptoms (anxiety, agitation, irritability), and cognitive symptoms (guilt and decreased self-esteem)—were also evaluated; all of these have previously been linked to SI and suicidal behavior.^14,27 It is important to note that because all participants in the STEP-BD trial were receiving high-quality clinical care and were regularly seen as outpatients, any symptom trajectories identified here would not reflect an inability to access psychiatric services. This analysis, which seeks to define a “typical” course of symptomatology surrounding suicide attempt in individuals with BD, may help provide foundational knowledge for developing future rapid-acting treatments for suicide risk.

2 |. PATIENTS AND METHODS

2.1 |. Participants

The data used in this exploratory analysis were drawn from the prospective STEP-BD study,²⁸ which followed 4,360 patients with BD in 22 academic psychiatry centers over seven years (NCT00012558). A subset of 864 participants (216 attempters, 648 non-attempters; details described below) was included in the analysis (see Table 1 for demographics). The STEP-BD study combined clinical data from both naturalistic treatment settings and randomized clinical trials; data were obtained by clinicians trained and certified in assessment and evidence-based treatments. The study was reviewed and approved by the institutional review board (IRB) of each participating institution.

TABLE 1.

Demographic information for attempters vs non-attempters

	Attempters (N = 216)	Non-attempters (N = 648)
	M ± SD (range)	M ± SD (range)	Comparison
Age (years)	34.86 ± 11.46 (16 – 77)	40.35 ± 12.67 (16 – 81)	F(1,862) = 31.91, P < .0001
	n (%)	n (%)
White or caucasian	188 (87%)	580 (90%)	F(1,862) = 1.00, P = .32
Previous suicide attempt	165 (76%)	214 (33%)	F(1,862) = 108.40, P < .0001
Baseline Clinical scores
Suicidal ideation (more than usual)	99 (46%)	180 (28%)	N/A
Agitation (more than usual)	110 (51%)	242 (38%)	N/A
Depressed mood (> 50% of the time)	122 (56%)	286 (44%)	N/A
Anxiety (> 50% of the time)	93 (43%)	232 (36%)	N/A
Global assessment of function	55.2 ± 13.15	60.97 ± 11.31	N/A

Note: Differences between groups in clinical variables are not hypothesized to be similar and were therefore not tested (N/A). The statistical significance of demographic differences are to be interpreted with caution, as the large sample size affords power to detect even inconsequential differences. The covariates age, gender, and history of suicide attempt were selected a priori and were included regardless of whether the difference between groups was statistically significant.

2.2 |. Suicidal behavior outcome

Data on suicidal behavior during the study were collected via the Serious Adverse Events report (SAE); this information was reviewed by an independent safety officer and committee for serious adverse events, which recorded deaths by suicide and suicide attempts. Data were also collected using the Care Utilization (CU) form, which queried whether participants had attempted suicide in the last three months. In the present study, data from the CU and SAE forms were collapsed, as detailed in the Data Preparation section below. Other analyses of suicide attempts and deaths in this sample have been reported using similar methods.^29,30

2.3 |. Affective disorder evaluation (ADE)

All BD patients were assessed at baseline using the Affective Disorder Evaluation (ADE) form.^28,31 Variables were selected as outcomes a priori, including overall SI (“Were there times when you were feeling so bad that you felt your life was not worth living? What about actually thinking about suicide or harming yourself?”) and psychomotor agitation (“Were there times you were so fidgety or agitated it was hard for you to stay still?”). Each of these symptoms was rated on a scale from + 2 (much more than usual) to −2 (much less than usual). These symptoms were assessed in the same manner at every clinic visit using the Clinical Monitoring Form (CMF), total scores of which are strongly correlated with depression rating scales such as the Hamilton Depression Rating Scale and the Montgomery-Asberg Depression Rating Scale.³² The frequency of abnormal anxiety and depressed mood was assessed differently depending on the form; the ADE asked patients to respond with the number of days in the past two weeks that they had experienced the symptom in question, while the CMF requested the percentage of the preceding 10 days. To combine these into a single scale, the ADE ratings were converted to a percentage (ie, a proportion of 14 days). Finally, for all patients, Global Assessment of Functioning(GAF) rating scale scores for the preceding month were obtained at each visit (range 0–100).

2.4 |. Data preparation

Among the 4,360 participants, eight died by suicide and were excluded from the analysis. A further 252 were excluded because no CMF data were available for their study participation. The SAE and CU forms from the remaining participants were reviewed to identify suicide attempts, of which there were 245/4100 or 5.97%. Because these forms did not list the specific date of the event (only whether it had occurred in the past three months), it was possible for the SAE and CU to separately identify the same event on separate study days. Thus, where the SAE and CU identified attempts that were within three months of one another (25 instances), the SAE report was retained and the CU report discarded. Participants with multiple suicide attempts (n = 29) were excluded because the statistical method could not incorporate multiple events. The final sample included 216 individuals who attempted suicide only once during the STEP-BD study. A subset of the 3,855 non-attempters was randomly selected to serve as a comparison group (n = 648). Because the primary models of analysis used 120 days before and 120 days after the individual’s attempt, which by definition did not occur in non-attempters, a randomized nearest-neighbor matching strategy was used to select a centering point for each non-attempter. Non-attempters were first matched to attempters on the basis of total time spent in the study (within five days) and at a 3:1 ratio. Next, data from each non-attempter was centered at attempt day for the attempter to whom they were matched. Demographic data for the attempter (n = 216) and non-attempter (n = 648) samples (entire sample, n = 864) are described in Table 1.

As shown in Figure S1, the CMF and ADE rarely reflected the technically continuous nature of the response scale. Instead, raters tended to respond at intervals of 0.5 or 1, and responses clustered around just a few response options. As such, in practice, the ratings from the CMF and ADE forms were ordinal rather than continuous. However, because of the relatively few response options (eg, ≥60% of all ratings were 0 for several of the items), the response scales were effectively categorical. Based on these distributions, responses were recoded into binary variables. For SI, agitation, and guilt, responses were recoded into ≤0 (“less than usual” to “usual”) vs >0 (“more than usual”). Loss of interest, self-esteem, and sleep were recoded into <0 (“less than usual”) vs ≥0 (“usual” to “more than usual”). Depressed mood, anxiety, and irritability were recoded into <50% and ≥50% for the time period specified by the instrument. GAF scores were not transformed.

Seven suicide attempters were missing information on number of previous lifetime suicide attempts from the ADE form. Because prior attempts was a covariate in the primary analytic model, values of “no” were imputed rather than excluding these participants. Sensitivity analyses excluding these participants as well as analyses imputing the values as “yes” indicated that this imputation was inconsequential and did not affect the results. No other missing data were imputed.

2.5 |. Statistical analysis

General linear mixed models with a two-piece linear function of time (in days) corresponding to pre- and post-event (suicide attempt) trends were used.³³ In these models, separate slopes were calculated for the data pre- and post-event by using two separate time variables, calculated as the difference in time between the date of assessment and the event. The first variable, the slope of which reflected the population-level pre-attempt trend, was fixed to zero for all assessments post-attempt. The second variable, the slope of which was the population-level post-attempt trend, was fixed to zero for all assessments pre-attempt. The amount of available pre-and post-attempt data varied by participant (see Figure S2), and the density of the pre-attempt data was sparser than the post-attempt data. For these reasons, a symmetric window of 120 days was implemented prior to analysis.

Random effects for intercept and linear slopes with an unstructured covariance matrix were used. Because random slopes for quadratic terms did not differ from zero, causing infinite likelihood, they were excluded. Because they have been linked to risk of suicidal behavior,^1,34 patient gender (male/female), previous history of suicide attempt (yes/no), and age at enrollment (in years) were entered as covariates, and their interactions with time values were evaluated and discarded if non-significant. Site was initially entered as a random effect, but it did not differ from zero and was therefore excluded from the model.

As described in the Data Preparation section, the response scales for the dependent variables were not continuous. In most cases, ratings clustered at one value (typically zero, which reflected “usual” levels of symptomatology) (see Figure S1). Attempts to model these data with a normal distribution yielded residuals thatconfirmed that this assumption was violated for all outcomes. Although dichotomizing continuous data nearly always decreases the power of a statistical test, in this case the response scale was not valid. Thus, given their extreme clustering, we elected to dichotomize responses. For these binary variables, a generalized linear mixed model with a binary distribution and logit link was used to estimate the predicted probability of the more severe score at a given timepoint (at the average value of the gender and history of suicide attempt covariates). GAF scores did have a slight left skew, but the normal distribution was tenable and a standard linear mixed model was used. The parameters of interest in these models were the pre- and post-attempt slopes and their interaction with group (attempter vs non-attempter). Contrasts of simple slopes were used to test the difference between pre- and post-slopes at given time points. Given the exploratory nature of these analyses, no adjustment to alpha was made.

3 |. RESULTS

3.1 |. Suicidal ideation

Altogether, 1,231 SI ratings from 216 attempters were analyzed (n = 395 pre-attempt, n = 126 circa-attempt, and n = 710 post-attempt). Table 2 lists the parameter estimates, or the pre/post slopes and the difference in slopes between groups, from the piecewise linear mixed model. Table 3 lists the difference in outcome between groups at 30 days pre-attempt, circa-attempt, 30 days post-attempt, 60 days post-attempt, and 90 days post-attempt. Among the attempters, SI worsened significantly prior to the attempt (t(2155) = 2.16, P = .03)and improved significantly after the attempt (slope at 30 days post-attempt: (2155) = −5.41, P < .0001) (see Figure 1). Among the attempters, SI ratings were also significantly better at 30 days post-attempt than at 30 days pre-attempt(t(2155) = −3.54, P = .0004) (see Table S1). For all points along the pre-attempt axis, the SI score of attempters was significantly higher than non-attempters (eg, 30 days pre-attempt, t(2155) = 6.36, P < .0001; see Table 3); the slope prior to the attempt did not differ between attempters and non-attempters (t(2155) = 0.88, P = .38). The post-attempt SI slopes differed significantly between attempters and non-attempters (difference in linear slope, t(2155) = −2.03, P = .04), such that attempters were no more likely than non-attempters to have worse than usual SI between 60 and 90 days following the attempt (group difference at 30 days post-attempt, t(2155) = 4.04, P < .0001; 60 days post-attempt, t(2155) = 2.15, P = .03; 90 days post-attempt, t(2155) = 1.20, P = .23).

TABLE 2.

Results of primary linear mixed model of suicidal ideation before and after suicide attempt

Generalized linear mixed model, outcome = “worse-thanusual” suicidal ideation	Slope	SE	DF	t value	P
Fixed Effects
Intercept	−1.40	0.37	846	−3.74	.0002
Female (vs Male)	−0.16	0.17	2155	−0.92	.36
No previous suicide attempt (vs yes)	−0.79	0.18	2155	−4.30	<.0001
Age at enrollment (years)	0.00	0.01	2155	0.25	.80
Pre-event time (in months)	0.10	0.11	513	0.96	.34
Attempter (vs Non-Attempter)	1.56	0.27	2155	5.84	<.0001
Pre-event time * Attempter (vs Non-Attempter)	0.12	0.13	2155	0.88	.38
Post-event time (in months)	−0.34	0.20	628	−1.73	.08
Post-event time (quadratic)	0.06	0.05	2155	1.16	.25
Post-event time * Attempter (vs Non-Attempter)	−0.63	0.31	2155	−2.03	.04
Post-event time (quadratic) * Attempter (vs. Non-Attempter)	0.08	0.09	2155	0.89	.37
Rate of change 30 days pre- and post-event, by group
Non-attempters at 30 days pre-event	0.10	0.11	2155	0.96	.34
Non-Attempters at 30 days post-event	−0.21	0.12	2155	−1.86	.06
Attempters at 30 days pre-event	0.22	0.10	2155	2.16	.031
Attempters at 30 days post-event	−0.69	0.13	2155	−5.41	<.0001

Note: Model included random intercept and slopes for pre-event and post-event time. Interactions between covariates (age, gender, and history of suicide attempt) and time variables were tested and discarded as non-contributory. Parameters are in log-odds. Predicted probabilities for each group, by time point, are shown in Table 3. Trajectories are illustrated in Figure 1.

Abbreviations: SE, standard error; DF, degrees of freedom.

TABLE 3.

Model-estimated probabilities of suicidal ideation before and after suicide attempt by group and time

	Attempters		Non-attempters
Time relative to attempt	Probability of “worse-than- usual” SI	95% CI	Probability of “worse-than- usual” SI	95% CI	Comparison
−30 days	0.38	(0.31–0.47)	0.13	(0.10–0.16)	t(2155) = 6.36, p ≤ .0001
0 days	0.44	(0.34–0.54)	0.14	(0.10–0.19)	t(2155) = 5.84, p ≤ .0001
30 days	0.25	(0.19–0.34)	0.11	(0.08–0.14)	t(2155) = 4.04, p ≤ .0001
60 days	0.16	(0.11–0.24)	0.10	(0.07–0.13)	t(2155) = 2.15, P = .03
90 days	0.13	(0.08–0.20)	0.09	(0.06–0.14)	t(2155) = 1.2, P = .23

Note: Predicted probabilities obtained from linear mixed model described in Table 2. Probabilities are estimated at the average of the covariates (age, sex, and suicide attempt history). Probabilities are illustrated in Figure 1.

Abbreviation: SI = suicidal ideation.

FIGURE 1.

Model-estimated trajectory of suicidal ideation and secondary outcomes among attempters (n = 216) and a randomly matched sample of non-attempters (n = 648). Non-attempters were matched at a 3:1 ratio to attempters based on duration of study participation. “Attempt day” for each non-attempter was equal to the attempter to whom they were matched. All models include age, gender, and history of suicide attempt as covariates. Shaded band is 95% confidence interval of the predicted probability of the more severe outcome

3.2 |. Affective/somatic symptoms of depression

The full results for all secondary outcomes are shown in Table S2; slopes at 30 days pre- and post-event, by group, are shown in Table S3. Predicted probabilities by group and time are shown in Table S4 and illustrated in Figures 1 and 2. Among the attempters, the affective somatic symptoms of depression (depressed mood, loss of interest, and sleep) were stable, with a non-significant slope, leading up to the attempt. Depressed mood improved significantly post-attempt (slope at 30 days post-attempt: t(2161) = −4.15, P < .0001, see Table S3), such that the attempters no longer had worse scores than the non-attempters by 90 days post-attempt (t(2161) = 1.64, P = .10, see Table S4). Sleep increased post-attempt (slope at 30 days post-attempt: t(2160) = 3.39, P = .0007, see Table S3), such that the attempters did not differ from the non-attempters by 30 days post-attempt (t(2160) = −0.44, P = .66, see Table S4). Loss of interest, which was significantly lower among attempters relative to non-attempters both before and after the attempt, improved post-attempt (slope at 30 days post-attempt for attempters: t(2158) = 2.23, P = .03, see Table S3).

FIGURE 2.

Model-estimated trajectory of secondary outcomes among attempters (n = 216) and a randomly matched sample of non-attempters (n = 648). Non-attempters were matched at a 3:1 ratioto attempters based on duration of study participation. “Attempt day” for each non-attempter was set equal to the attempter to whom they were matched. All models include age, gender, and history of suicide attempt as covariates. Shaded band is 95% confidence interval

3.3 |. Activating/Tension Symptoms of Depression

A less distinct pattern of results was observed across the activating/tension symptoms of depression (anxiety, irritability, and agitation), which were relatively stable prior to the event (see Figure 1, Tables S2–S4). When compared to non-attempters, these symptoms were generally worse among attempters prior to the attempt (eg, 30 days pre-attempt: anxiety, t(2157) = 2.66, P = .008; agitation, t(2142) = 2.36, P = .02; irritability, t(2159) = 1.65, P = .10), but no different from the non-attempters by 30 days post-attempt (anxiety, t(2157) = 1.97, P = .05; agitation, t(2142) = 1.01, P = .31; irritability, t(2159) = 0.66, P = .51).

3.4 |. Cognitive Symptoms of Depression

The cognitive symptoms of depression were also relatively stable during the pre-event period (see Figure 2, Tables S2–S4). While guilt decreased post-event among attempters, the level at 30 days post-attempt did not differ from 30 days pre-attempt (t(2142) = −1.34, P = .18), and it did not remit to the level of non-attempters until 120 days post-attempt (t(2142) = 0.20, P = .84). Self-esteem was somewhat increased post-attempt (30 days pre-attempt vs 30 days post-attempt, t(2148) = 2.24, P = .02), however, levels were not comparable to those of non-attempters until approximately 120 days post-attempt (t(2148) = −1.08, P = .28).

3.5 |. General functioning

While the individual symptom ratings of attempters were generally stable prior to the attempt, GAF scores did worsen during this period (slope at 30 days prior to attempt: t(371.6) = −2.39, P = .02) (Figure 2). After the attempt, however, GAF ratings improved significantly (slope at 30 days post-attempt: t(1462) = 7.84, P < .0001). While the GAF scores of the attempters approached that of the non-attempters, they did not converge until around 120 days post-attempt (t(907.6) = −1.49, P = .14).

4 |. DISCUSSION

The acute suicide crisis is a dynamic state, and these results from the STEP-BD trial highlight specific changes in SI in the 120 days before and 120 days after a suicide attempt. In particular, attempters exhibited a global pattern of stability or slight worsening in SI severity during the 120 days leading up to a suicide attempt and a subsequent improvement in SI following the attempt, such that SI levels did not differ between attempters and non-attempters by 90 days post-attempt. These findings underscore that SI was elevated both before and after suicide attempt, suggesting a limited timeframe in which careful identification, assessment, and intervention could be useful in reducing suicidal thoughts and potentially preventing suicidal behavior. When other depressive symptoms were similarly examined before and after the attempt, variability in symptom trajectories was observed. Certain symptoms, such as depressed mood and loss of interest, demonstrated a relatively similar timecourse as SI. Others, such as anxiety, agitation, and irritability, demonstrated no such pronounced separation from non-attempters. Lastly, cognitive depressive symptoms, such as guilt and low self-esteem, did not remit to non-attempter levels until 120 days post-attempt, a slightly longer trajectory than the other depressive symptoms. The descriptive results from this naturalistic clinical trial of individuals with BD suggest that the trajectory of SI in the days and weeks surrounding a suicide attempt is not necessarily distinct from the affective symptoms of depression, but that its timecourse may differ from activating or cognitive symptoms of depression.

The reduction in psychiatric symptoms post-suicide attempt merits particular mention. It was not possible to evaluate the level of treatment that the STEP-BD patients received after the suicide attempt, which may have included hospitalization, more intensive therapy, more frequent clinic visits, or in-depth suicide risk assessment. However, one major treatment concern post-suicide attempt is continuity of care³⁵; because all patients were part of the STEP-BD trial, it can be assumed that they experienced some sort of clinical “safety net” that is not always available to outpatients in the community. In this context, it is important to note that even with these clinical resources, SI and other related depressive symptoms did not immediately dissipate, highlighting the need for careful assessment and monitoring post-suicide attempt. Additionally, critical differences in symptom trajectories were observed. Specifically, activating symptoms such as anxiety or irritability did not exhibit a pronounced peak around the time of suicide attempt, while SI, depressed mood, and loss of interest gradually declined by 90 days and differences in guilt and self-esteem dissipated by 120 days. These trajectories underscore the importance of repeated comprehensive assessment of all facets of suicide risk and depression, given that an overall “depression” rating scale score may not capture these nuanced changes based on symptom type.

It should be noted that the present analysis was not designed with a particular model of the acute suicide crisis in mind^17,18,23 and therefore could not determine causality or whether these symptoms directly influenced the suicide attempt. For example, it is possible that additional intensive treatment of SI and depression at the time of worsening SI or affective depressive symptoms could have prevented the suicidal behavior, but such analyses are not possible with the current data. The burgeoning area of research focused on brief interventions for recent suicidal thoughts or behavior, including pharmacologic interventions such as ketamine³⁶ or psychotherapeutic interventions such as safety planning,³⁷ motivational interviewing,³⁸ and “teachable moment” brief interventions³⁹ may begin to address whether these symptom trajectories—and, ultimately, the outcome of suicidal behavior—can be prevented. Furthermore, the decline in symptoms has implications for clinical trials in which BD patients are recruited just after a suicide attempt or excluded based on suicidal behavior.⁴⁰ As findings from the present analysis underscore, patients post-suicide attempt could still be at high risk for suicidal behavior even though they may not report severe levels of depression and SI.^4,41 Inclusion criteria and outcome selection should therefore be conducted carefully in order to adequately assess these at-risk patients and track their symptoms.

Strengths of this analysis include the large sample of BD patients who were assessed 120 days before and 120 days after suicidal behavior as part of a comprehensive clinical trial. Notably, the clinical trial data permitted evaluation of ratings-level data rather than relying on medical record reviews of retrospective patient reports. Results were fairly similar to a previous analysis of pre-suicide attempt symptomatology,²⁶ in that SI and loss of interest demonstrated the greatest difference in suicide attempters compared to non-attempters. Limitations include the exploratory use of data not collected for these research questions; as such, care must be taken in interpreting the results. In addition, the STEP-BD study did not provide a daily assessment of participants or techniques such as ecological momentary assessment, so the precision of measuring both attempts and symptomatology is limited by the frequency of assessment. The precision of the measurements was also affected by the CMF questionnaire itself because respondents did not use the response scale consistently and because the response scale for many of the items required that respondents rate the symptom relative to their “usual” symptoms. It would also have been useful to know the range of SI-related experiences, from thoughts that life was not worth living to passive or active SI, etc For this reason, the absolute symptom levels are unknown. In addition, we have no information about any putative interventions administered to patients following a suicide attempt; some participants may have been hospitalized or changed medications, which would presumably have affected their post-attempt trajectory. Thus, these results are descriptive in nature, perhaps reflecting what would occur in the community with treatment as usual. Additionally, individuals who made multiple attempts during the trial were excluded; future analyses could evaluate predictors of re-attempt in this sample. Finally, because the results are limited to individuals with BD, they may not be generalizable to other diagnoses, such as major depressive disorder, post-traumatic stress disorder, or substance use disorders.

In conclusion, these results from the STEP BD identified dynamic changes in SI and depressive symptoms in the 120 days before to 120 days after a suicide attempt in individuals with BD. Specifically, elevations in SI may be present up to 90 days post-suicide attempt, with similar trajectories found for affective depressive symptoms such as depressed mood and loss of interest. Further research describing the suicide crisis, in line with recent theoretical and neurobiological research, is needed to support more precise predictions and targeted interventions for these vulnerable individuals.