Suicidal and nonsuicidal self-harm thoughts and behaviors (SHTBs) are a global public health concern, with estimates suggesting that >14.6 million individuals are affected by self-harm (defined as any form of deliberate self-injury irrespective of motivation or intent) each year and >700,000 deaths per year are attributable to suicide [1]. Understanding factors that contribute to the development of SHTBs is essential for prevention and early intervention.

The integrated motivational-volitional model (IMV) is one of the leading theoretical models of suicidal behavior, developed by O’Connor [2] and later refined by O’Connor and Kirtley [3] (Figure 1 [3]). The IMV model consists of 3 phases: the premotivational stage, describing the biopsychosocial context in which suicidal thoughts and behavior may emerge; the motivational phase, describing the factors that lead to the development of suicidal thoughts; and the volitional phase, describing the factors that predict the transition from thoughts to behaviors [3].

Core constructs of the motivational phase include feelings of defeat, humiliation, and entrapment, which drive the emergence of suicidal thoughts. These can be facilitated or impeded by the presence of moderating variables, termed “threat to self,” and “motivational” moderators. The transition from suicidal thoughts to behaviors is, in turn, influenced by the presence of “volitional moderators” [2,3]. Although the IMV model was developed in relation to suicidal thoughts and behavior, the central concepts of the model can also be applied to nonsuicidal SHTBs [4].

A recent systematic review of studies testing the IMV model of suicidal behavior yielded support for the central components of the model (ie, the defeat-entrapment–suicidal ideation [SI] pathway)—but called for more focus on the constructs referred to as threat-to-self and motivational moderators in the model [5]. The review identified extensive testing of the model using cross-sectional retrospective methods while highlighting the need for more prospective (including intensive longitudinal) testing of IMV constructs [5].

Recent technological advances have made it easier for researchers to gain insights into SHTBs in real time using intensive longitudinal methods [6-8]. These methods are commonly referred to in the literature as experience sampling methods (ESMs), ambulatory assessment, daily diaries, and ecological momentary assessment (EMA). From this point onward, for brevity, we use the term EMA to refer to this methodology. EMA is a diary-based method involving repeated and frequent assessment of feelings, behaviors, and contexts in an individual’s natural environment. This minimizes recall bias, maximizes ecological validity, and enables dynamic processes to be captured. Assessment may be once daily (daily diaries); repeated throughout the day at pseudorandomized or specific times in a signal-contingent sampling scheme; or repeated based on reporting of a specific event, such as an act of nonsuicidal self-injury (NSSI; event contingent) [6]. Despite concerns regarding the demands placed on research participants from intensive sampling, it has been found to be both acceptable and feasible, with generally good compliance reported [9]. While there have been further concerns about the repeated reporting of mental states having an influence on an individual’s mental state, there is no strong evidence of such iatrogenic effects [7,8].

Existing EMA studies of suicidal thoughts have shown them to be highly variable over time [10,11]; however, less is known about the extent to which proximal risk factors for SHTBs, such as those proposed by the IMV model, fluctuate in daily life. Understanding the dynamic nature of risk factors (within-person variability) and their moderators is essential to tailoring interventions and risk assessments. It is important to understand whether risk factors and moderators are better characterized by individual differences (between-person variability) or momentary changes in experiences (within-person variability) [11,12]. Examining intraclass correlation coefficients (ICCs—see the Data Analysis section for more details) enables distinction between trait-level risk factors (high ICC and high between-person variability)—supporting longer-term strategies for intervention—and state-level risk factors (low ICC and high within-person variability)—for which acute, timely, and situation-specific intervention may be more appropriate [11]. A recent proliferation of EMA studies in the field of suicide and self-harm has prompted the need for a comprehensive synthesis of this literature. While others have reviewed EMA literature on self-harm [13-15], suicidal thoughts [9,13,16,17], and interpersonal processes in an SHTB context [18], EMA studies specifically assessing key constructs across both the motivational and volitional phases of the IMV model have not yet been synthesized. In addition, existing reviews of EMA studies have typically focused on the relationship between risk factors and SHTBs, and less attention has been paid to the characteristics of the risk factors themselves.

We conducted a systematic review of the SHTB literature in which constructs from the motivational and volitional phases of the IMV model were assessed using EMA. We offer a narrative synthesis, describing how IMV constructs were assessed in daily life, characterizing their within-person variability, and summarizing the evidence of the proximal relationships between each IMV construct and SHTBs. We identified gaps in the evidence base and proposed directions for future research.

The review questions are as follows:

The secondary review question is as follows:

This review was preregistered on the PROSPERO database (CRD42022349514) and on the Open Science Framework (OSF) [19]. The PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines were followed (Multimedia Appendix 1), in addition to the Non-Intervention, Reproducible, and Open Systematic Review guidelines [20]. We searched the databases Web of Science, Embase, MEDLINE, PsycINFO, and Europe PMC Preprints. We also hand searched reference lists and citations of the included papers for additional papers not returned by the database searches.

The full search terms and strategy are available on the OSF [19]. We searched for studies that used intensive longitudinal methods, often referred to as EMA, experience sampling, ambulatory assessment, or daily diary methods. We did not establish a date limit on the search. The initial search was conducted in October 2022, yielding 40 papers, with an updated search in November 2023 yielding an additional 13 papers. A final presubmission updated search conducted in March 2024 yielded an additional 6 studies over 5 papers.

Records were exported to, stored, and managed using the application Rayyan (Rayyan Systems Inc). In total, 2 (blinded) authors independently screened the papers for inclusion based on the titles and abstracts simultaneously against the inclusion and exclusion criteria, with disagreements resolved through discussion. One author conducted full screening of the selected papers based on the full text. Full details are available in the corresponding OSF project page [19].

We included studies that had assessed at least one of the IMV constructs in daily life using intensive longitudinal data collection methods (ie, EMA). These factors include the core motivational phase factors (defeat, shame, humiliation, and entrapment), threat-to-self moderators (problem-solving, coping, memory bias, and rumination), motivational moderators (thwarted belongingness, burdensomeness, future thinking, goals, norms, resilience, social support, and attitudes), and volitional moderators (suicide planning, exposure to self-harm, impulsiveness, pain sensitivity, fearlessness about death, and imagery). To be included, the studies needed to report details of within-person variability in IMV constructs (ie, ICC; see the Data Analysis section). Where not reported, we contacted the authors to request this information. If ICCs were not available and no association between at least one IMV construct and SHTBs was reported, the study was excluded.

We included PhD theses published on the internet and excluded studies that were meta-analyses, reviews, editorials, or commentaries, as well as articles not written in English. We did not limit the inclusion of studies according to population or participant characteristics, and both clinical and nonclinical samples were included.

We will not describe the assessment of SHTBs in this study as this has been covered by previous reviews [8,14,16,17].

Data extraction (template available on the OSF [19]) included a range of descriptive data for each study, including demographic information about the sample, whether the study sampled a clinical or community-based population, study design, and information about the IMV constructs measured. ICCs were extracted to describe the level of within-person variability in each construct. ICCs indicate the proportion of a variable’s variance that is due to between- and within-person variability [21]. The within-person variability is calculated as 1 minus the ICC. When within-person variability is low, this means that the variability in a construct is mostly due to differences between people and there is little fluctuation in the construct in people (ie, the construct may be considered more traitlike than statelike). For example, a hypothetical ICC of 0.83 would indicate just 17% within-person variance, suggesting that the construct is more traitlike and shows little fluctuation in people. Conversely, an ICC of 0.26 would indicate 74% within-person variance, suggesting that the construct is more statelike and shows large fluctuation in people over time.

Where studies tested associations between IMV constructs and SHTBs—either concurrent or lagged—these associations were also extracted. Quality assessment of the reporting of the studies was conducted according to an EMA-specific quality assessment tool [18]. Example reporting criteria included participant training in the EMA protocol being detailed in the Methods section, justification of the sample size, compliance rate and reasons for noncompliance, discussion of EMA-specific limitations, and open code for analysis (the full criteria are available on the OSF [19]).

A total of 53 studies (unique samples) were included in this review (Figure 2 provides the PRISMA flowchart) across 58 papers, all from higher-income countries, with most from North America (n=40, 75%), the United Kingdom (n=3, 6%), and Germany (n=3, 6%). The studies varied substantially in terms of population, sample size, design, and constructs measured. Several papers (11/58, 19%) reported different analyses using the same study sample (ie, the same sample was used to report different IMV constructs across different papers). To avoid double counting of samples and designs, Table 1 reports a summary of 53 studies (58 papers reporting results from 53 independent samples).

The sample sizes ranged from 10 to 743 participants (mean sample size 85.3, SD 105.4; median 54, IQR-38.5-94.5). Most studies (47/53, 89%) included largely female samples (69.8% of the total combined sample were female); in 4% (2/53) of the studies, all participants were female or women [22,23]; in 8% (4/53) of the studies, participants were mostly male or men [24-27]; 28% (15/53) of the studies included a small number of transgender, nonbinary, or gender nonconforming participants; and, in 2% (1/53) of the studies, all participants were transgender or gender diverse [28]. In 55% (29/53) of the studies, most participants were White; 19% (10/53) of the studies did not report the participants’ race or ethnicity. The mean sample age ranged from 15.0 to 47.7 years, with 28% (15/53) of the studies using a sample of young people (aged ≤25 years). In 26% (14/53) of the studies, participants were recruited from the general population, with 23% (12/53) of the studies recruiting from universities and the samples in the remaining studies being recruited from clinical settings or a mix of clinical and community settings.

Approximately a quarter of the studies (15/53, 28%) adopted a daily diary design with 1 assessment per day. Of the remaining 72% (38/53) of the EMA studies, the number of measurements ranged from 2 to 10 per day (mean 4.2, SD 2.3 measurements), and most (33/53, 62%) were delivered using a signal-contingent sampling scheme at pseudorandomized intervals. Study duration ranged from 3.4 to 90 days (mean 19.8, SD 18.5; median 14.0 days), with some studies (6/53, 11%) reporting varying durations based on length of hospitalization [24,25,29-32]. In the 85% (45/53) of the studies in which study duration was consistent for all participants, we found a small negative correlation (r=–0.14) between study duration (in days) and the number of assessments per day. EMA smartphone apps were most often used for data collection, including Illumivu, MetricWire, and movisensXS.

In this section, we refer to the 58 individual papers. In total, 3% (2/58) of the papers [25,31] each reported 2 independent samples. Several other papers (11/58, 19%) reported the same sample but different IMV constructs [33,34]. A total of 45% (26/58) of the papers reported more than one IMV construct. In 12% (7/58) of the papers [26,31,35-39], the study used a signal-contingent (pseudorandom) sampling scheme with multiple daily assessments, but one or more IMV constructs were assessed once per day, or an aggregated daily measure was used.

Across the 58 papers included in this review, the motivational moderators in the IMV model were most frequently assessed in EMA studies (Table 2). The constructs measured most frequently were thwarted belongingness (24/58, 41% of the papers), positive or negative thoughts about the future (20/58, 34% of the papers), and perceived burdensomeness (16/58, 28% of the papers). The least frequently measured constructs were humiliation, social problem-solving, mental imagery, physical pain sensitivity, and fearlessness about death. None of the included papers measured memory biases, goals, norms, or resilience using EMA.

As there are currently no gold standards for conducting EMA research, we assessed the studies against quality criteria for reporting EMA studies. The reporting quality of the studies varied considerably, although most were fully compliant with only a small minority of the quality criteria. The reporting criteria with the most compliance included referencing EMA (or equivalent) in the title and keywords, briefly introducing and justifying the use of EMA, reporting the full text of the items and response options, and describing data preparation and analysis in detail (Table 7). The criteria with the poorest compliance included justifying the sample size, justifying the sample design (eg, random or event based) and number of assessments, describing any design feature to address potential sources of bias or participant burden (eg, EMA questions appearing in different orders), reporting the number of EMA prompts that were planned to be delivered and the number that was actually received by participants (and any reasons why prompts were not sent out), and reporting whether EMA compliance was related to demographic or time-varying variables.

No study fully reported the amount of time from prompt signal to answering of the prompt. There was generally low use of open science practices, with only the studies reported in 3% (2/58) of the papers explicitly preregistering hypotheses and 2% (1/58) publishing study materials. The studies reported in 14% (8/58) of the papers made their analytic code either fully or partially publicly available, and those reported in 7% (4/58) of the papers made their data open access (other studies either reported data being available on request or did not include a data availability statement). However, it is worth noting that such open science practices are recent responses to the replication crisis and would not be expected in papers published before 2016. A detailed summary table of quality assessment data is available in Multimedia Appendix 4.

There has been a recent proliferation in EMA studies of SHTBs. We set out to summarize the state of the EMA literature pertaining to the constructs described in the IMV model. We selected the IMV model because it is one of the leading theoretical frameworks for understanding suicidal behavior that incorporates most constructs contained in other theoretical models, such as the interpersonal theory of suicide (IPTS) [80].

A total of 58 papers were included in this review. The IMV constructs measured most frequently were thwarted belongingness (24/58, 41% of the papers), thoughts about the future (20/58, 34% of the papers), and perceived burdensomeness (16/58, 28% of the papers). The higher number of studies focusing on thwarted belongingness and perceived burdensomeness likely reflects the key role of these constructs in the IPTS [80]. The least frequently measured constructs were humiliation, social problem-solving, mental imagery, physical pain sensitivity, and fearlessness about death. No included papers measured memory biases, goals, norms, or resilience using EMA.

Most of the included papers were assessed to be fully compliant with a number of reporting quality criteria, including referencing EMA (or equivalent) in the title and keywords (34/58, 59%), briefly introducing and justifying the use of EMA (41/58, 71%), reporting the full text of the items and response options (35/58, 60%), and describing data preparation and data analysis in detail (36/58, 62%). However, there was poor compliance with over half of the 25 criteria against which the reporting quality of the included papers was assessed.

The results predominantly showed at least moderate within-person variability across all IMV constructs included in this review whether assessed once or multiple times daily, highlighting the utility of intensive longitudinal study designs. Moderate to high within-person variability characterizes these constructs as at least somewhat statelike and dynamic. Where shown to be associated with SHTBs, these cognitive, social, and situational stressors and coping resources outlined in the IMV model could be important targets for intervention. Finding statelike constructs can identify risk factors that are potentially suitable for in-the-moment or just-in-time interventions before SHTBs develop. Although some constructs from the IMV model were well researched using EMA, namely, thwarted belongingness, future thoughts, and perceived burdensomeness, less—or no—information was available for other constructs, such as defeat, humiliation, social problem-solving, memory biases, resilience, access to means, fearlessness about death, and exposure to suicide. Some of these (social problem-solving, memory biases, and resilience) are likely to be more traitlike constructs and, thus, less suited to EMA research.

We did not find evidence of patterns in within-person variability as a function of whether the samples were drawn from community or clinical populations or of the number of daily assessments. It is worth noting that most community samples were “high-risk” populations with a recent history of SHTBs who may not have differed substantially from clinical populations. Our ability to make comparisons between sampling schemes was also limited by a quarter of the included studies (13/58, 22%) not reporting within-person variance.

We found evidence across several studies (15/58, 26%) of concurrent associations between IMV constructs and SHTBs and some—though inconsistent—evidence of lagged associations between entrapment and SI, shame and SI, rumination and SHTBs, future thoughts and SHTBs, perceived burdensomeness and SHTBs, impulsivity and NSSI, thwarted belongingness and SHTBs, social support and NSSI, and mental imagery and SI. However, not all findings were consistent, possibly due to heterogeneity across the studies included in our review in terms of the populations studied, study design, operationalization of both IMV constructs and SHTBs, and analytical approach. While the cross-lagged panel model—sometimes used in the included studies—is a popular tool in this and other fields, it may often not be the correct tool [81]. The importance of disaggregation when analyzing EMA data to study within-person processes has been recognized, and we are likely to see a continued transition to alternative, superior models, such as the random-intercepts cross-lagged panel model. Failure to disaggregate within- and between-person variability is likely to lead to much higher lagged effects, particularly when studying stable, traitlike constructs.

Inconsistencies in concurrent and lagged associations between IMV constructs and SHTBs preclude firm conclusions about support for these constructs as proximal risk factors for SHTBs. We did not identify studies explicitly testing hypotheses from the IMV model, namely, moderation of the association among defeat, humiliation, and entrapment; between entrapment and SI; or between SI and suicidal behavior. Our review shows that there is scope for further investigation of temporal relationships between IMV constructs and SHTBs, particularly with regard to the motivational phase (defeat, humiliation, and entrapment), threat-to-self moderators (social problem-solving, coping, memory biases, and rumination), and volitional moderators (access to means, planning, impulsivity, physical pain sensitivity, fearlessness about death, and imagery) in the model. While our review suggests that most IMV constructs are subject to momentary shifts, more evidence is needed as to the implications of these dynamic patterns for SHTBs and progression from SI to behavior.

Assessment of the reporting quality of the included studies suggested little routine use of open science principles—even in studies published since the replication crisis—reflecting the situation in the suicide research field more broadly [82]. Open science practices can improve the transparency, reproducibility, and replicability of scientific research [83]. We recommend that future EMA studies in the field consider preregistration of hypotheses using the registration template for ESM research [82] and making data, study materials, and analytic code available. Other key areas for improvement include justifying the sample size (eg, by reporting power analyses [84]), sample design, and the number of assessments used in a study; consideration of potential sources of bias or participant burden when designing EMA studies; and more comprehensive reporting of EMA compliance, including whether compliance is related to demographic or time-varying variables. This is crucial as aspects of EMA study design, such as great questionnaire length, are associated with increased burden, which affects the quality and quantity of EMA data [85].

Our findings are broadly consistent with a recent systematic review of interpersonal processes and SHTBs in daily life [18] in which the authors highlight a lack of consistency in the operationalization of key constructs [16]. Efforts to improve transparency and harmonization of measurement include the ESM Item Repository [86], a searchable and public bank of items used in previous EMA studies to which researchers can contribute their items.

Also consistent with previous reviews of the EMA SHTB literature [8,14,16,18], our findings indicate that EMA is a useful methodological approach to providing rich information about individuals’ experiences of both IMV constructs and SHTBs in daily life. The same reviews have similarly noted the challenges to synthesis of findings due to heterogeneity in study design. Despite the wide variation in results reported, we found evidence of at least moderate within-person variance in IMV constructs. This adds to existing EMA literature reporting SHTBs to be subject to short-term fluctuation [8,13]. While SHTBs have been well studied using EMA methodology, risk factors in the IMV model have received less attention. EMA is particularly relevant to the IMV model of suicide, enabling an ecologically valid “in-situ” and nuanced understanding of an individual’s transition through the motivational and volitional phases of the model, including identification of immediate situational triggers and critical moments of escalation from ideation to behavior [3,9].

Our findings of mixed evidence of associations between IMV constructs and SHTBs are also broadly consistent a review of (mainly cross-sectional) studies testing hypotheses from the IPTS [87], which found mixed support for a main effect of thwarted belongingness on SI, with stronger—though still inconsistent—support for a main effect of perceived burdensomeness on SI. Also in line with our findings, the aforementioned review [87] found lower support for an association between acquired capability for suicide (including pain sensitivity and fearlessness about death) and suicide attempt.

Our review is the first to comprehensively synthesize the EMA literature on key constructs in the IMV model of suicide [3], including a narrative synthesis of both within-person variance of constructs and their associations with SHTBs. It also adds to the findings from a recent systematic review of cross-sectional, case-control, and longitudinal studies of the IMV model [5]. The broad scope of this review, in addition to the methodological differences between studies, prohibited a meta-analysis of findings. Future meta-analyses of variability in key IMV constructs may be feasible with more focused and specific inclusion criteria.

We conducted a thorough search of 5 databases as well as citation searching and used blinded double screening of abstracts against the inclusion and exclusion criteria. We made extra efforts to include unpublished research by searching the PMC Europe database of preprints, therefore limiting the effect of possible publication bias on our results. An additional search in 3 months of submission for publication ensured that the review was as up-to-date as possible as the EMA literature is prolific. We limited the search to papers in English following consultation with colleagues but acknowledge that this may have resulted in the loss of papers published in other languages.

Our findings suggest that EMA methods can be valuable in providing real-time information about key risk factors for SHTBs, as outlined in the IMV model [3]. EMA may also be of potential value in clinical support settings, providing clinicians with detailed insights into an individual’s mental states and daily experiences of constructs described by the IMV model. Enhanced understanding of these risk factors in daily life may inform individualized interventions.

Further research is needed to better understand how some IMV constructs—including defeat, humiliation, social problem-solving, memory biases, resilience, access to means, fearlessness about death, and exposure to suicide—vary in daily life and over what timescale and whether there are proximal associations with SHTBs. Many of the factors we investigated are conceptualized in the IMV model as moderators; however, all studies in this review focused on individual relationships with SHTBs. Further research is needed to explore moderation to specifically test the hypotheses outlined in the IMV model.

We found large heterogeneity in the populations sampled, including gender (eg, 2/53, 4% of the studies sampled only female participants [22,23], and 1/53, 2% of the studies sampled only transgender or gender-diverse participants [28]), ethnicity (eg, 1/53, 2% of the studies sampled only Hispanic or Latino adults [75]), and mental health status (40/53, 75% of the studies sampled only those with a diagnosed mental health disorder or with recent experience of SHTBs). Female participants were overrepresented in EMA studies, representing 69.8% of the total combined sample. While beyond the scope of this review, future EMA studies or reviews might consider sociodemographic differences in the within-person variability of IMV constructs.

Overall, there is existing evidence suggesting that there is within-person fluctuation in the IMV constructs included in this review, suggesting that it is possible to study them using EMA methods. We also found evidence of concurrent relationships between almost all constructs and SHTBs in daily life, with some evidence that entrapment, shame, rumination, thwarted belongingness, hopelessness, social support, and impulsivity are additionally associated with SHTBs in lagged (ie, longitudinal) relationships. While EMA methods show promise in providing valuable information about individuals’ experiences of both IMV constructs and SHTBs in daily life, there is currently large methodological heterogeneity and a paucity of quality in studies using this approach. Efforts to enhance the quality of reporting and advance transparency and harmonization in the field may improve future syntheses of findings.