This prospective study used EMA and focused on psychiatric outpatients who were identified as being at high risk of suicide. Participants were recruited from 4 hospitals in the Community of Madrid (Spain). This study is part of the SmartCrisis 2.0 project, a randomized controlled trial (RCT) with a parallel-group design. The trial includes patients with a recent history of suicidal behavior, who are randomly assigned to 1 of 2 groups. Participants in the control group receive treatment as usual together with EMA, whereas participants in the intervention group receive treatment as usual plus EMA and an ecological momentary intervention (EMI). Both groups are followed for up to 12 months, with repeated clinical assessments and continuous smartphone-based monitoring. EMA data is collected continuously throughout the follow-up period in both groups; therefore, for participants in the intervention arm, EMA monitoring occurs concurrently with the EMI, whereas in the control group, EMA is conducted without intervention. The EMI is a smartphone-based intervention that includes a personalized safety plan, an enhanced contact system with supportive messages, and a toolbox of brief therapeutic exercises delivered via the app. Full details of the SmartCrisis 2.0 RCT protocol are available in a separate publication [7]. Although the broader SmartCrisis 2.0 project includes an interventional element, the present study focuses on the observational EMA monitoring data collected during the study period and does not evaluate the effectiveness of the intervention.

The original protocol planned to recruit 220 participants across 5 sites, including centers in France. However, the French sites were ultimately withdrawn, and the study proceeded with the 4 Spanish sites included in the present study. In addition, although the target sample size for the RCT is 220 participants, the present analysis includes a smaller sample because sufficient statistical power had already been achieved to address the aims of this study.

The study adhered to the ethical principles set forth in the Declaration of Helsinki and was approved by the Research Ethics Committee of Fundación Jiménez Díaz (EC005-21_FJD). Written informed consent was obtained from all participants. To ensure confidentiality, the usernames and personal information of participants were pseudonymized using coded identifiers. All data were encrypted using AES-256 with 256-bit keys and were managed within a professional key management system, which underwent external security compliance audits. No financial compensation was offered to participants.

In this study, STBs are used as an umbrella term encompassing different suicide-related experiences, including:

Definitions were based on the Columbia–Suicide Severity Rating Scale [22].

Negative affect refers to a broad dimension of emotional distress that includes unpleasant emotional states such as anger, guilt, fear, worry, shame, and nervousness, as originally described by Watson et al [23]. In the present study, this construct was operationalized using EMA items assessing stress, restlessness, hopelessness, anxiety, sadness, psychological pain, and anger directed toward oneself or others, along with a positively worded happiness item that was coded inversely.

The study sample comprised psychiatric outpatients who had recently experienced STBs. The inclusion and exclusion criteria are listed in Textbox 1.

For the final analysis, only those participants who responded to at least 3 EMA prompts related to STB were included.

All statistical analyses were performed using the R software package (R Foundation for Statistical Computing). The intraindividual mean and variability of responses were calculated for each participant. Variability was assessed using the root mean square of successive differences. Missing data were not imputed; mixed models were chosen for statistical analysis as they are robust in dealing with missing values [29].

Mixed-effects models were used to evaluate within-person variability and temporal trends. Between-person reliability (R-between), within-person reliability (R-within), and intraclass correlation coefficients (ICCs) were calculated from the null model to quantify variability [30]. Reliability coefficients were interpreted according to Shrout (1998) as follows [31]: 0.00‐0.10 indicates no reliability, 0.11‐0.40 slight reliability, 0.41‐0.60 fair reliability, 0.61‐0.80 moderate reliability, and 0.81‐1.00 substantial reliability. Time was then entered as a fixed effect, participants as random effects, and each EMA category (eg, negative affect or sleep) as the dependent variable.

Mixed-effects models were also used to examine the effect of EMA categories (eg, negative affect, interpersonal difficulties, sleep, and eating behavior) on SI over time. Each model included fixed effects for the EMA category, time, and their interaction. Random intercepts were specified for participants, with random slopes for time. An additional model was run that included all significant EMA categories together.

Multiple linear regression analyses were conducted to assess the association between EMA variables and demographic covariates (age and gender). Further regressions examined whether baseline levels of depression and psychological distress predicted the mean and variability of SI and negative affect throughout the EMA follow-up period. Where appropriate, posthoc simple slope analyses were performed.

All tests were 2-tailed and significance was set at P<.05, with 95% CIs.

A total of 115 participants were recruited. Of the initial sample, 99 had sufficient data (at least 3 EMA answers) and were included in the final analysis. The mean age was 39.5 (SD 0.33) years (43.4, SD 2.24 y for men and 37.1, SD 1.74 y for women), and there was a majority of female participants (n=62, 62.6% females vs n=37, 37.4% males).

Most participants had completed secondary education (n=71, 72.1%), followed by those with university education (n=25, 25%), while a small proportion had not completed compulsory education (n=3, 2.9%). Regarding marital status, 49% (n=49) of participants were single, 17.1% (n=17) were separated, divorced, or widowed, and 33.9% (n=34) were in a relationship. Additionally, 31.3% (n=31) of the sample consisted of individuals who were employed or studying.

There were 87.9% (n=87) participants with a lifetime mood disorder, 67.7% (n=67) participants with a lifetime anxiety disorder, 63.6% (n=63) participants with a lifetime psychotic disorder, 43.4% (n=43) participants had borderline personality disorder, and 22.2% (n=22) participants had a history of substance or alcohol use disorders.

A total of 13,900 EMA responses were obtained, corresponding to an approximate overall response rate of 29.3%. Of these, 5638 belonged to the SI category. The slope of SI over time was not statistically significant (β=−0.006, SE 0.014; P=.68), showing no significant changes over follow-up. Similar results were obtained for the slopes of negative affect (β=−0.019, SE 0.017; P=.24), interpersonal difficulties (β=−0.027, SE 0.019; P=.15), sleep problems (β=0.007, SE 0.011; P=.52), and appetite (β=0.001, SE 0.018; P=.99).

SI had an R-between of 0.51, an R-within of 0.84, and an ICC of 0.17. Negative affect showed an R-between of 0.74, an R-within of 0.83, and an ICC of 0.24. Interpersonal problems had an R-between of 0.79, an R-within of 0.84, and an ICC of 0.26. Sleep problems showed an R-between of 0.63, an R-within of 0.77, and an ICC of 0.29. Finally, appetite showed an R-between of 0.53, an R-within of 0.62, and an ICC of 0.45. Between-person reliability ranged from fair to moderate, while within-person reliability was good to high across constructs. ICCs were generally low across constructs, except for appetite.

Of the total sample, 56% (n=55) of participants received the safety plan intervention during EMA monitoring. We tested whether the intervention had an effect on SI outcomes. The effect of the safety plan was not significant for SI (β=−1.24; t₉₈=−0.47; P=.64), nor for the intraindividual mean (β=−.32; t₉₈=−0.11; P=.91) or variability (β=−1.29; t₉₈=−0.54; P=.59) of SI. For this reason, the safety plan intervention was not included as a covariate in the remaining models.

Figure 1 illustrates the evolution of the EMA responses over time.

The variables statistically significantly associated with SI were negative affect (β=0.193, SE 0.03; t₉₈=6.21; P<.001) and interpersonal difficulties (β=0.17, SE 0.03; t₉₈=5.96; P<.001). The association was positive: the greater the negative affect and the interpersonal difficulties, the greater the SI. When negative affect and interpersonal difficulties were put together in the same model, both predictors maintained their significance (P<.001).

The effect of the interactions negative affect × time (P=.12) and interpersonal difficulties × time (P=.29) were not statistically significant. There were also no statistically significant effects on appetite (appetite, P=.54; appetite × time, P=.78) and sleep models (sleep, P=.97; sleep × time, P=.15).

Greater intraindividual means of negative affect (R²=0.39; β=0.502, SE 0.08; t₉₈=6.65; P<.001) and interpersonal difficulties (R²=0.182; β=0.359, SE 0.07; t₉₈=5.03; P<.001) were associated with greater intraindividual means of SI. No significant differences were found for sleep problems and appetite. None of the intraindividual variability indexes were associated with the intraindividual mean of SI.

Intraindividual variability of SI was directly associated with intraindividual variability in the rest of the EMA variables: the greater the SI variability, the greater the variability of negative affect (R²=0.15; β=0.477, SE 0.12; t₉₈=4.04; P<.001), interpersonal difficulties (R²=0.15; β=0.323, SE 0.07; t₉₈=4.49; P<.001), sleep problems (R²=0.12; β=0.259, SE 0.09; t₉₈=2.96; P=.004), and eating habits (R²=0.23; β=0.318, SE 0.09; t₉₈=3.54; P<.001).

A subsample of 37 (48.6% females, mean age 40.6, SD 2.41 y) participants had information about baseline psychological pain and depression. The mean baseline IDS-C score of the participants was 26.2 (SD 1.59). Their current and maximum psychological pain at inclusion was 6.76 (SD 0.48) and 7.36 (SD 0.59), respectively. Participants answered EMA questions for a mean of 146 (SD 21.2) days.

There was a negative association between baseline maximum psychological pain and variability of negative affect (R²=0.12; β=−1.40, SE 0.65; t₉₈=−2.16; P=.04) but psychological pain itself did not predict the mean of negative affect (P=.41). Thus, the greater the psychological pain, the lower the variability. The interaction psychological pain × days of follow-up significantly predicted intraindividual negative affect mean (R²=0.24; β=0.019, SE 0.007; t₉₈=2.58; P=.02). Simple slopes showed significant results for the –1SD psychological pain (β=−0.08, SE 0.037; P=.03) but not for the +1 SD of psychological pain (β=0.04, SE 0.039; P=.19); showing patients with lower psychological pain and longer follow-up experiments less negative affect (Figure 2). The interaction psychological pain × days of follow-up did not predict the intraindividual variability of negative affect (P=.21).

Psychological pain itself did not predict intraindividual variability (P=.15) or the mean of SI (P=.16). The interaction psychological pain × days of follow-up predicted the intraindividual mean of SI (R²=0.34; β=0.024, SE 0.001; t₉₈=3.95; P<.001). Simple slopes showed significant results for the –1SD psychological pain (β=−0.12, SE 0.030; P<.001) but not for the +1 SD of psychological pain (β=0.04, SE 0.023; P=.10), showing that patients with lower psychological pain and longer follow-up experience less SI (Figure 3). The interaction psychological pain*days of follow-up did not predict the intraindividual variability of SI (P=.08).

Regarding depression, the effect of depression at baseline itself did not predict intraindividual variability (P=.75) or mean of negative affect (P=.11). The interaction depression*days of follow-up predicted the intraindividual mean of negative affect (R²=0.17; β=0.006, SE 0.003; t₉₈=2.35; P=.03). Simple slopes showed significant trend results for the +1 SD of psychological pain (β=0.06, SE 0.03; P=.06) but not significant results for the –1 SD psychological pain (β=−0.06, SE 0.04; P=.18). Thus, patients with higher depression and longer follow-up experience greater negative affect (Figure 4). The interaction depression × days of follow-up did not predict the intraindividual variability of negative affect (P=.87).

Results for variability in SI (depression P=.69; depression × days of follow-up P=.99) or SI mean (depression P=.26; depression × days of follow-up P=.51) using depression as a predictor did not show statistically significant results.

This study examined SI and its associations with other psychological and behavioral variables in a sample of almost 100 high-risk outpatients who were monitored for 6-months. Trajectories (slopes) of SI, negative affect, interpersonal difficulties, sleep disturbances, and appetite issues did not demonstrate significant change over time. However, SI was significantly associated with negative affect and interpersonal difficulties. Additionally, substantial within-person variability in SI was observed. Notably, individuals who reported higher levels of psychological distress at the beginning of the study exhibited increased levels of negative affect and SI throughout the follow-up period.

Considerable within-person variability was observed for all EMA-tracked variables. Relatively low ICCs suggest that these constructs reflect dynamic, state-like fluctuations rather than stable traits [32]. The pronounced variability in SI and its related factors emphasizes the fluctuating nature of suicide risk. Bernanke et al [33] have proposed two subtypes of SI: one that is reactive to stress and tends to spike following acute stressors, and another that is more chronic and associated with persistent depressive states. This distinction may partly account for the variability observed in our data, as individual emotional responses are likely shaped by personal experiences.

Consistent with this, Bonilla-Escribano et al [34] conducted an EMA study involving 275 psychiatric outpatients and nearly 50,000 responses. Their findings indicated that high SI variability was associated with greater instability in social withdrawal, sleep, and perceived social support.

The dynamic interplay between negative affect, interpersonal difficulties, and SI, as observed in our study, supports the need for personalized, real-time monitoring strategies [10]. This reinforces recent critiques of traditional clinical assessments, which often fail to capture moment-to-moment fluctuations in suicide risk [6]. Our results emphasize the value of EMA in capturing these rapid shifts, which standard tools may overlook.

One of the key findings was the significant association between SI and negative affect, as well as interpersonal difficulties, as measured via EMA. These results are consistent with prior research identifying these variables as critical components in the development of suicidal thoughts. In particular, our findings align with existing literature linking SI to interpersonal distress, including feelings of loneliness and of being a burden to others [35,36]. Similarly, Bentley et al [37] identified emotions such as anxiety, shame, and self-hatred as predictors of subsequent SI using EMA. These results echo the core tenets of the Interpersonal Theory of Suicide, which posits that a lack of belonging and the perception of being a burden are fundamental drivers of SI [38].

However, sleep and appetite disturbances were not significantly associated with increased SI in this sample. This diverges from earlier studies conducted by our group [39], which focused specifically on passive SI. Differences in sample size and the more pronounced role of negative affect may explain these inconsistencies.

We also found that greater variability in negative affect, interpersonal experiences, sleep, and appetite within individuals corresponded with greater variability in SI. Whether such fluctuation in SI translates into an increased risk of suicide attempts remains an open question. However, evidence from studies such as that of Oquendo et al [40], which followed 51 patients with depression for two years, suggests that higher SI variability may increase reactivity to stressors and elevate the risk of impulsive suicidal behavior.

Baseline psychological pain was found to have a significant influence on SI throughout the EMA follow-up. The relationship between psychological distress and suicidal thoughts can be explained by several theoretical models. For example, Shneidman [41] identified unbearable psychological pain as a key factor in suicidal behavior, while Baumeister [42] proposed that suicide may be an attempt to escape distress caused by perceived failure or unmet expectations. More recently, Klonsky and May [43] 3-step theory emphasized the role of psychological pain and hopelessness in the development of SI.

Regarding depression, we found that baseline severity was associated with increased negative affect over time. However, no significant association was found between depression and SI. Since negative affect includes components such as hopelessness, anxiety, sadness, anger, and restlessness—many of which overlap with depressive symptomatology—this association is expected. Nevertheless, our findings suggest that SI is a more multifaceted construct that is not exclusively linked to depressive disorders. This highlights the importance of considering other psychiatric conditions, particularly those characterized by emotional dysregulation, when assessing suicide risk.

This study has several strengths. The 6-month follow-up period allows for the examination of longer-term patterns beyond acute suicidal crises. Participants were recruited from real-world mental health outpatient settings and represented a high-risk clinical population, enhancing ecological validity. EMA items were derived from validated instruments and administered using a rotating sampling strategy designed to improve feasibility and adherence over time. Finally, the use of mixed-effects models allowed for the analysis of repeated EMA measurements while accounting for individual differences in response patterns.

Despite these strengths, several limitations should be acknowledged. First, the observational design does not allow causal conclusions to be drawn about the associations between SI and EMA variables. Second, EMA items were administered using a rotating sampling strategy to reduce participant burden; however, this limited temporal continuity across constructs and prevented the examination of lagged relationships. In addition, although the EMA design aimed to improve feasibility and maximize long-term adherence [8,20], some features—particularly the variation in temporal framing of items (eg, momentary, since the last 24 h, in the past few days)—could be refined in future studies to improve alignment between the time frame, response scale, and expected construct variability. Furthermore, although EMA prompts were delivered at random times, participants were allowed to complete questionnaires until midnight, which may have introduced recall bias if responses did not reflect the moment of the prompt, or selection bias if participants reported on a different moment of their choosing. Third, baseline data on psychological pain and depression were only available for a subsample of 37 participants, reducing the statistical power of secondary analyses. Finally, response rates could not be calculated for individual EMA items due to the rotating and partially random sampling design. The overall response rate was relatively low, likely due to the extended follow-up period of our study. It is important to note that EMA studies are usually conducted over shorter periods, often lasting only a few weeks (8). Therefore, the longer duration of our study may have negatively affected long-term participant adherence and should be considered when interpreting the findings.

Future studies should explore integrating EMA into routine clinical practice. One promising approach is the use of EMIs, which can deliver personalized, context-aware feedback based on momentary risk profiles. Furthermore, research should aim to identify subgroups of patients who may benefit from specific therapeutic strategies. For instance, individuals with heightened emotional variability may respond better to interventions based on emotion regulation, while those experiencing ongoing interpersonal difficulties could benefit from targeted social support programs [17].

This study is one of the few to use smartphone-based EMA to monitor self-injury over an extended period [13]. EMA is a useful method for capturing short-term suicide risk factors. It is particularly important to consider negative emotions and interpersonal issues in suicide prevention. The high within-person variability observed in SI and its correlates emphasizes the dynamic and fluctuating nature of suicide risk. Moving forward, integrating EMA-based monitoring into clinical care could facilitate real-time, personalized interventions aimed at mitigating suicide risk in individuals at risk.