Therapeutic Alliance is Linked with Suicidality Trajectories in Psychotherapy for Late-Life Depression

Abstract

Objectives.

Late-life suicidality is rapidly increasing; psychotherapy may reduce suicidality, but trajectories of change in suicidality remain unclear. A strong therapeutic alliance may protect against persistent suicidality. We examined the association of alliance with suicidality trajectories during brief, remote interventions for depression.

Methods.

Sixty middle-aged and older adults with major depression were randomized to brief remote psychosocial interventions. Suicidality and alliance were assessed during treatment. Patients’ suicidality trajectories were classified as: a) absent; b) improved; c) persistent, based on pre- to post-treatment changes. All three groups were included in mixed-effects models to examine the association between suicidality and alliance.

Results.

At baseline, 53% of patients endorsed suicidality, with 45% in the absent group, 45% improved; and 10% persistent. Patients in the absent and improved groups showed stable, strong alliance while the persistent group showed worsening alliance.

Conclusion.

Simple, remote psychosocial interventions are promising first-line treatments for late-life suicidality. A strong alliance may reduce suicidality and enhance outcomes, guiding timely, personalized interventions.

Objectives

Suicide rates are the highest among older adults and rapidly increasing in recent decades¹. Psychotherapy for depression can reduce late-life suicidality,² yet trajectories of change in suicidality during treatment remain largely understudied. Identifying how individual patient’s suicidality changes can inform precise and personalized risk detection and intervention strategies.

A strong therapeutic alliance - an emotional bond and agreement on therapeutic goals and tasks between the patient and therapist - may protect against suicidality.³ Therapeutic alliance is a key predictor of response to psychotherapy for depression,^4,5 but its relationship with suicidality in late-life is unknown. A strong alliance may be especially important for mitigating late-life suicidality. It provides a meaningful social connection during a period of heightened isolation and unmet social need - factors linked to persistent suicidality.⁶ Uncovering the association between suicidality and alliance can inform clinical recommendations for this high-risk population.

To our knowledge, this study is the first to examine the association between alliance and late-life suicidality during interventions for depression. In a randomized controlled trial, older adults received 9-weeks of simple, remote psychosocial interventions. We aimed to a) assess the rates of late-life suicidality at the outset of treatment and changes in suicidality during treatment; b) examine the association between therapeutic alliance and changes in suicidality over time.

Methods

Design.

Older adults with major depression were enrolled in a randomized controlled trial and assigned to one of two 9-week, remotely delivered interventions: 1) Engage & Connect (n=29), an adapted version of Engage behavioral activation psychotherapy,⁷ aimed to improve functions of the reward system by increasing engagement in rewarding social activities⁸; 2) Symptom Review and Psychoeducation (n=31), an intervention focused on symptom monitoring and patient education. Patients were recruited using online advertisements and flyers. Trained study personnel conducted initial phone screens to determine potential eligibility. Potential participants provided informed consent and were then evaluated by a licensed or license-eligible clinician who determined study eligibility. Enrolled participants attended weekly remote therapy sessions and weekly assessments as well as study visits at weeks 0, 3, 5, and 9. Therapists were master’s level, license-eligible or licensed social workers or mental health counselors. One therapist was a master’s level predoctoral psychology intern. License-eligible clinicians completed all educational requirements for licensure and had either completed or were in the process of completing necessary supervised hours for licensure. The study was approved by the Weill Cornell Medical College Institutional Review Board and is registered on ClinicalTrials.gov (Trial Registration #: NCT04487730).

Participants.

The sample included 60 patients, aged 50–85 (mean age = 63.3, SD = 7.57; Table 1) with unipolar major depression without psychotic features, determined by a licensed or licensed eligible clinician (SCID-V: Structured Clinical Interview for DSM Disorders – V; Score of ≥ 20 on the MADRS; Montgomery-Åsberg Depression Rating Scale⁹). Patients were unmedicated or on stable antidepressants. Exclusion criteria included intent or plan to commit suicide in the near future, past or current psychiatric disorders other than Major Depressive Disorder, Generalized Anxiety Disorder, Persistent Depressive Disorder, or specific phobia, use of psychotropic drugs or cholinesterase inhibitors other than ≤ 0.5 mg of lorazepam daily up to seven times per week, neurological disease, cognitive impairments (Mini Mental Status Exam (MMSE) ≤ 24), or acute or severe medical illness, and any contraindications for MRI scanning.

Table 1.

Baseline Demographic and Clinical Characteristics of Suicidality Groups

Group	Absent suicidality (N=27, 45.00%)	Improved suicidality (N=27, 45.00%)	Persistent suicidality (N=6, 10.00%)	Overall (N=60)
Age	62.7 (6.56)	64.7 (8.13)	59.7 (8.98)	63.3 (7.57)
Gender
Female	21 (77.80%)	22 (81.50%)	5 (83.30%)	48 (80.00%)
Male	6 (22.20%)	5 (18.50%)	1 (16.70%)	12 (20.00%)
Ethnicity
Hispanic	3 (11.10%)	2 (7.40%)	0 (0.00%)	5 (8.30%)
Non-Hispanic	24 (88.90%)	25 (92.60%)	6 (100.00%)	55 (91.70%)
Race
Black/African-American	3 (11.10%)	1 (3.70%)	0 (0.00%)	4 (6.70%)
White	21 (77.80%)	23 (85.20%)	6 (100%)	50 (83.30%)
Asian	1 (3.70%)	3 (11.10%)	0 (0.00%)	4 (6.70%)
Other	2 (7.40%)	0 (0.00%)	0 (0.00%)	2 (3.30%)
MADRS Total	23.4 (3.32)	25.7 (3.50)	23.2 (4.12)	24.4 (3.61)
	24.00 [17.00, 32.00]	26.00 [20.00, 33.00]	22.00 [18.00, 29.00]	25.00 [17.00, 33.00]
MADRS Item 10	0 (0)	1.59 (0.84)	0.83 (0.41)	0.80 (0.95)
	0 [0, 0]	1.00 [1.00, 3.00]	1.00 [0, 1.00]	1.00 [0, 3.00]
Alliance (WAI)-Week1	5.61 (1.11)	5.27 (1.26)	4.26 (0.99)	5.32 (1.21)
	5.78 [2.58, 7.00]	5.50 [2.08, 7.00]	4.79 [2.42, 4.92]	5.50 [2.08, 7.00]
Alliance (WAI)-Week9	6.18 (0.86)	5.85 (1.24)	3.71 (1.46)	5.77 (1.32)
	6.30 [4.17, 7.00]	6.00 [2.75, 7.00]	3.79 [1.75, 5.25]	6.04 [1.75, 7.00]
Therapy
E&C	13 (48.10%)	14 (51.90%)	2(33.30%)	29 (48.30%)
SRP	14 (51.90%)	13 (48.10%)	4 (66.70%)	31 (51.70%)

Note: Table presents frequency for categorical variables (n; %) and mean and standard deviation for numerical variables (Mean; SD) and median and range (Median; [Min, Max]). ¹ WAI: Patient-reported Working Alliance Inventory; ²MADRS: Montgomery-Asberg Depression Rating Scale; ³E&C: Engage & Connect; ⁴SRP: Symptom Review and Psychoeducation.

Interventions.

Interventions were delivered weekly, remotely via videoconference, over 9 weeks, by trained and certified licensed or license-eligible clinicians (e.g., masters-level social workers, mental health counselors). Therapists completed training by the PI and a supervising clinician and delivered both interventions. Interventions included: a) Engage & Connect (E&C), a modified version of Engage⁷, which is a behavioral activation psychotherapy aimed at improving reward system functions by enhancing engagement in rewarding social activities⁸. Patients who did not respond by week 3 receive targeted interventions to address key barriers to response; b) Symptom Review and Psychoeducation (SRP), a psychosocial intervention designed to track patients’ symptoms and deliver psychoeducation on topics related to late-life depression.

Measures.

Trained research assistants administered the MADRS, a 10-item assessment of depression severity on a 7-point Likert scale, from 0 to 6, with higher scores indicating greater severity. The MADRS was administered at weeks 0,3,5, and 9, with high internal consistency in this sample (Cronbach’s α = .90, CI: [.86,.93]). Suicidality was measured with item 10 (“This past week, have you felt like life isn’t worth living, tired of living…thought that you would be better off dead…hurting or even killing yourself?”).

Therapeutic alliance was measured weekly using the 12-item version of the Working Alliance Inventory (WAI; Working Alliance Inventory¹⁰) on a 7-point Likert scale, from 1 to 7, with higher scores indicating a stronger alliance. The WAI was completed by patients and therapists after each session. The WAI assesses the emotional bond and agreement on therapy goals and tasks, with higher scores indicating stronger alliance. We found high internal consistency in this sample (Week 1: Cronbach’s α = .96, CI: [.94,.97] patient version; Cronbach’s α = .93, CI: [.91,.96] therapist version).

Statistical Analysis.

Analyses were conducted in R (Version 4.3.3) using the intent-to-treat sample with all available data (3 patients dropped out and their last assessment was used). Data from both interventions were used to increase statistical power. We calculated the proportion of patients endorsing a positive score (>0) on Item 10 of the MADRS at pre-treatment to assess rates of suicidality and its severity.

Patients were classified into three groups based on MADRS item 10 scores during treatment: a) absent suicidality: no suicidality throughout treatment, (score = 0 at all timepoints); b) improved suicidality: pre-post treatment suicidality reduction (score decreased from baseline to week 9); c) persistent suicidality: worsening or consistent suicidality pre-post treatment (score remained the same or increased from baseline to week 9). We opted for this approach, as opposed to data-driven clustering, because in a small sample, data-driven clustering may produce unstable clusters. Our approach aligns closely with clinical practice, and enhances clinical relevance, as the grouping strategy can be readily applied in real-world clinical settings.

We examined differences in therapeutic alliance between suicidality groups using mixed-effects models with a subject-specific random intercept. Alliance (WAI) was the dependent variable, and fixed effects included suicidality group, time, and their interaction. Time was coded numerically as sessions. The random intercept accounts for individual differences in alliance at the start of treatment (Week 1), allowing the model to estimate group-specific trajectories over time while controlling for these initial differences. Demographic variables (age, gender, race, ethnicity) were initially included but were nonsignificant and did not affect results, so they were removed for a more parsimonious model.

We focused on patient-reported alliance in this report. We tested the therapist-reported alliance and results were equivalent. We also tested whether results replicated using only the completers sample (n=57), to ensure dropouts did not bias the findings, and found no bias due to dropouts.

Results

At baseline, 53% of patients (n=32) reported suicidality (47% scored 0; 37% scored 1; 7% scored 2; 10% scored 3). Patients were classified into three trajectory groups: a) absent suicidality (45%; n=27); b) improved suicidality (45%; n=27); c) persistent suicidality (10%; n=6). We confirmed group differences in suicidality over time (F(2, 232.44) = 44.31, p < .001). The improved suicidality group showed a significant and sustained decline in suicidality (F(1, 104.31) = 94.89, p < .001), while the persistent group did not show change (F(1, 23) = .71, p = .41) (Fig. 1).

Fig. 1.

Patient-reported therapeutic alliance over time by suicidality trajectories groups: absent (purple, no suicidality; n=27), improved suicidality (blue; n=27), and persistent suicidality (red; n=6). WAI: Working Alliance Inventory. Confidence intervals reflect within-participant standard errors.

We found a significant main effect, reflecting that patient-reported therapeutic alliance significantly differed across suicidality groups (F(2, 66.81) = 5.52, p < .01). The alliance was weaker in the persistent suicidality group compared to both the absent (b = 1.63, SE = 0.49, t(67) = 3.31, p < .01) and improved suicidality groups (b = 1.25, SE = .49, t(67) = 2.53, p < .05) (Table 1; Fig. 1).

We also observed a significant interaction between suicidality group and time in patient-reported alliance (F(2, 431.78) = 5.04, p < .001). This stemmed from an increase in alliance over treatment in the absent and improved suicidality groups, and a worsening in alliance in the persistent group (Table 1; Fig. 1). Findings were similar in therapist-reported therapeutic alliance. We confirmed that results were replicated for both main effect and time-by-group interaction in the completers sample (n=57).

Conclusions

Our results show that late-life suicidality is common, with 53% of patients endorsing suicidality at baseline. After receiving nine weeks of remote psychosocial interventions, suicidality was reduced in 45% of patients. This finding suggests that simple, remote psychotherapies may be promising and cost-effective first-line treatments for at least a portion of patients who report experiencing suicidality.² Patients with persistent suicidality reported weaker therapeutic alliance compared to those with absent or improved suicidality, and their alliance worsened during treatment. In contrast, patients in the absent and improved groups showed strong, stable alliance over time. These findings highlight the importance of strengthening the alliance during interventions for this high-risk group.

The stable and strong therapeutic alliance in the absent and improved suicidality groups suggests the important role of alliance in late-life suicidality. Our prior work showed that rewarding social experiences predict response to psychotherapy for late-life depression.¹¹ The alliance may be especially beneficial for older adults, as it offers a meaningful rewarding social experience at a time of significant social isolation. Future research could examine how therapist-patient interactions during treatment, in combination with patient-related factors, such as interpersonal pathology, may affect the alliance-suicidality association.

The low quality of working alliance in the persistent suicidality group is aligned with prior work on the negative impact of chronic suicidality on alliance.¹² Differences between the groups in alliance quality were meaningful. The early alliance quality in the persistent group was ~1.5 points lower than the absent and improved groups, and this difference increased across treatment. Further, we observed a pre-post decrease of ~.5 points in the persistent suicidality group. Prior work has shown that 1 point increase in alliance predicts a 4.35 point decrease in depression severity the following week.¹³ Additionally, alliance literature defines a weekly drop of 1 point in alliance as a “rupture”, suggesting a significant worsening of the working relationship.¹⁴ Thus, the low alliance in the persistent suicidality group likely may reflect a more challenging treatment process.

The study interventions did not include targeted interventions to improve alliance but rather highlighted the importance of collaboration across the treatment process. It is possible that patients with persistent suicidality are less likely to develop strong emotional bonds, including with a therapist, and therefore may be more likely to have poor working alliance. Therapists may also struggle to maintain a strong working alliance with patients who are at risk, due to difficulties with engagement or repeated risk assessments that cause frustration and stress. Our findings suggest that specific interventions that enhance alliance may be especially beneficial with this group.¹⁵

Our study had several limitations. We relied on a single-item measure for suicidality and thus could not examine different aspects of suicidality. Further, we used the combined sample to enhance statistical power and thus could not examine treatment-specific effects. Our future research will examine differential treatment effects. Our study could not determine causality for the alliance-suicidality relationship. Finally, our groups were defined based on clinical criteria. This approach is clinically sensible, can be easily applied in clinical settings, and is appropriate for our relatively small sample. Future studies, with larger samples, could use data-driven approaches, such as latent profile analysis or machine-learning based clustering.

In summary, we found that late-life suicidality is common and can be reduced significantly during simple, remote psychosocial interventions for late-life depression. Our findings suggest that alliance is associated with late-life suicidality and its changes during treatment. Our results show that patients without suicidality and those who improve over time report strong and stable alliance, while those with persistent suicidality show low and worsening alliance. These findings highlight the importance of enhancing alliance and developing personalized interventions for patients with persistent suicidality. Taken together, our study could inform future risk-detection strategies and just-in-time alliance-targeted interventions, critical for preventing suicides in this vulnerable population.

Highlights.

• We examined rates and trajectories of late-life suicidality during brief remote psychotherapies for depression and their association with therapeutic alliance over time.

• At baseline, 53% of patients endorsed suicidality. During treatment, 45% improved and 10% showed persistent suicidality. Alliance remained strong and stable in the absent and improved groups but worsened in the persistent group.

• Simple brief psychotherapies can reduce late-life suicidality; timely assessment and interventions that improve the alliance are crucial for high-risk patients with persistent suicidality.

Data Sharing Statement

Available upon request from authors.