Cognitive Anxiety Sensitivity: Invariance, Longitudinal Course, and Associations With Suicide Risk in a Large Military Sample

Abstract

This study examined cognitive anxiety sensitivity’s invariance, longitudinal course, and associations with suicidal thoughts and attempts within a large military sample (N = 1,147). First, multiple group confirmatory factor analyses assessed the latent structure of cognitive subscale from the anxiety sensitivity index (ASI-C) by group (i.e., Active Military/Veterans and men/women). Second, free-loading latent growth curve modeling assessed the stability of cognitive anxiety sensitivity across each group over four time points. Third, multiple linear regressions tested if cognitive anxiety sensitivity at the previous time point predicted suicidal thoughts and number of attempts at the following study visit, above and beyond previous suicidal thoughts, generalized anxiety, and thwarted belongingness and beyond previous lifetime attempts, respectively. The ASI-C displayed a very well-fitting unifactorial structure and metric invariance across all groups. Overall, cognitive anxiety sensitivity appeared to significantly decrease over time across all groups, significantly more so for Active Military personnel than for Veterans and significantly more so for women than for men. Cognitive anxiety sensitivity predicted future suicidal thoughts at Time Point (T) 4 above and beyond control variables among Active Military and T2 and T3 among men but not for Veterans or women. Further, cognitive anxiety sensitivity predicted future suicide attempts at T3 beyond control variables among Active Military, Veterans, and men but was not for women. Findings suggest that cognitive anxiety sensitivity is worth including in suicide risk screening and, due to its malleability, may be a viable treatment target to reduce suicide risk among Active Military and Veteran men.

Suicide continues to be a leading cause of death in the United States among the general population (Centers for Disease Control and Prevention, 2023). Importantly, it is the leading cause of death among Active Military (i.e., those enlisted in the military [Active Duty], reserves, and national guard; Defense Casualty Analysis System, 2023), the 13th most common cause of death among Veterans, and the second leading cause of death for Veterans aged 45 and younger (U.S. Department of Veteran Affairs, 2023). Statistically, death by suicide has a low base rate, which can make it particularly challenging to understand, predict, and prevent. In 2022, reports of suicide deaths among Active Military members were similar to demographically adjusted rates in the general population (CDC, 2023; Department of Defense, 2023), in contrast to a quarter century ago, when military rates were far lower than civilian rates. In 2021, reports of suicide deaths among Veterans indicated a fourfold increase relative to suicide deaths in the general U.S. population (CDC, 2023; U.S. Department of Veteran Affairs, 2023).

Two of the top three risk factors for death by suicide in the general population, albeit with low predictive power in an absolute sense, include previous suicide attempts and suicidal ideation (Franklin et al., 2017). Estimations indicate that both Active Military members and Veterans are at an increased risk for suicide attempts compared to those in the general population (CDC, 2023; Department of Defense, 2023; Pruitt et al., 2019; Schafer et al., 2022). Among military personnel, global studies estimate that 14% of military personnel indicate a desire for suicide, compared to 4.8% of U.S. adults (CDC, 2023; Moradi et al., 2021). Taken together, there is a need to identify suicide risk factors which effectively translate into treatment efforts to reduce suicidal ideation and attempts in both Active Military service members and Veterans.

Military related stressors may, in part, explain heightened suicide risk. Military personnel, regardless of gender, report high levels of job-related stress than personal stress; studies suggest 26% of Active Military report significant levels of routine work stress (e.g., combat, periodic change of station, lack of control over duty assignments, unexpected deployments), and another 15% report significant emotional distress from work-related stressors (Pflanz, 2001; Pflanz & Sonnek, 2002). When analyzed by gender, both Active Military men and women were nearly twice as likely to report job-related stressors compared to personal stressors (Bray et al., 2001). Moreover, research indicates that low levels of social support during and after work-related stressors exacerbated adverse mental health outcomes within Active Military members (Han et al., 2014; Ozer et al., 2003). In addition, whether real or perceived, low social support may induce psychological states similar to thwarted belongingness, a robust predictor of suicidal ideation in military samples (Chu et al., 2018; Van Orden et al., 2010).

The stress-management model of job strain articulates that arousal related to anxiety increases when job demands are high and both job control and social support are low (Karasek, 1979). One form of reactivity to these repeated states of arousal is known as anxiety sensitivity, the thought or belief that symptoms of anxiety or arousal can have harmful consequences (i.e., the “fear of fear”; Olatunji & Wolitzky-Taylor, 2009). Anxiety sensitivity is a known risk factor for anxiety disorders (Olatunji & Wolitzky-Taylor, 2009), posttraumatic stress disorder (Marshall et al., 2010; Taylor, 2003), depression (Allan, Capron, Raines, & Schmidt, 2014), and suicide risk (Allan et al., 2023). Empirically, there is meta-analytic evidence of small-to-moderate and moderate associations between anxiety sensitivity and suicidal ideation (r = .24, 95% confidence interval, CI [.21, .26], p < .001) and suicide risk more broadly (r = .35, 95% CI [.31, .38], p < .001; Stanley et al., 2018). However, it is important to determine if anxiety sensitivity confers risk for suicidal ideation above and beyond generalized anxiety and known risk factors for suicide such as thwarted belongingness.

Importantly, associations between anxiety sensitivity and suicide risk appear to be driven primarily by the cognitive subscale of the Anxiety Sensitivity Index (ASI-C; Reiss et al., 1986), which reflects fears specific to a loss of cognitive control (e.g., “when I have trouble thinking clearly, I worry that there is something terribly wrong with me”), rather than the social or physical subscales of anxiety sensitivity (Capron, Fitch, et al., 2012). Oglesby et al. (2015) found that cognitive anxiety sensitivity significantly predicted future categorization of individuals at low versus high risk for suicide; however, these findings did not hold regarding the other anxiety subscales. In the meta-analysis referenced above, associations between cognitive anxiety sensitivity and suicidal ideation were significant with small-to-moderate effects while the relationship between suicide risk and cognitive anxiety sensitivity was moderate in magnitude (Stanley et al., 2018). Similar findings have emerged in military cadets undergoing basic training, such that high cognitive anxiety sensitivity and low physical anxiety sensitivity predicted future suicidal ideation above and beyond other psychopathology (Capron, Cougle, et al., 2012). Raines et al. (2017) also reported that, among Veterans, cognitive anxiety sensitivity partially mediated the relationship between posttraumatic stress disorder symptom severity and higher suicide risk (i.e., ideation, plans, and impulses). Studying cognitive anxiety sensitivity in military populations is essential, as it can impair focus and decision making in Active Military members and reduce distress tolerance and increase rumination among Veterans.

Though some work has examined cognitive anxiety sensitivity in military samples (Capron, Cougle, et al., 2012; Jeon et al., 2025; Raines et al., 2017), further work is needed. First, although previous research within the general population suggests that the Anxiety Sensitivity Index-3 (Taylor et al., 2007) is invariant across sex, age, race/ethnicity, and sexual minority status (Jardin et al., 2018), to our knowledge, no studies have assessed cognitive anxiety sensitivity’s invariance across military-related groups (i.e., Active Military vs. Veterans and men vs. women military personnel). Second, given that cognitive anxiety sensitivity appears to be a relevant risk factor for suicidal thoughts and suicide attempts, it is important to consider its longitudinal course (i.e., suggesting malleability and potential viability as a treatment target). While some research suggests anxiety sensitivity is relatively stable (Hovenkamp-Hermelink et al., 2019; Rodriguez et al., 2004), other studies have shown mild reductions in anxiety sensitivity over time (Allan, Capron, Lejuez, et al., 2014) and across the lifespan (Mahoney et al., 2015). Among Active Military and Veterans, some research suggests that extended service duration is associated with reduced anxiety (Rice & Schroeder, 2019). This effect may stem from increased familiarity and competence within military roles, as well as the development of robust coping strategies and resilience among those who remain in service. Another plausible explanation is that individuals in leadership positions, or Veterans who have attained higher ranks—potentially with greater financial security through retirement or disability benefits—experience a heightened sense of control over their futures, thereby mitigating anxiety and stress related to postservice life. Third, past research on cognitive anxiety sensitivity’s relationship to suicidal thoughts and suicide attempts is predominately cross-sectional and, when prospective, is not specified by the group (i.e., Active Military vs. Veterans and men vs. women military personnel); thus, establishing longitudinal associations within and across groups is warranted.

Therefore, this study aims to address all three of these points: First, we hypothesized that the ASI-C will display measurement invariance across groups (i.e., Active Military vs. Veterans and men vs. women). Second, as previous research suggests anxiety sensitivity declines over time and more generally researchers have not observed iatrogenic effects due to repeated measurement—if anything, the opposite (Finn & Tonsager, 1992; Hom et al., 2018)—we hypothesize that there will be a statistically significant reduction in cognitive anxiety sensitivity across time points for all groups. Nevertheless, we will examine if there are group-based differences in slope reductions as an exploratory aim. Third, we expect cognitive anxiety sensitivity at the previous time point to predict suicidal ideation at the following time point for all groups above and beyond robust predictors of suicidal ideation in military samples (i.e., the previous time point’s thwarted belongingness and generalized anxiety; Chu et al., 2018; Van Orden et al., 2010) and the previous time point’s suicidal ideation. Moreover, we expect that cognitive anxiety sensitivity at the previous time point will predict suicide attempts in between study visits for all groups, controlling for generalized anxiety and lifetime suicide attempts.

Method

Procedure

All data were collected through the Military Suicide Research Consortium (MSRC) Common Data Elements (CDE) project between 2016 and 2023 (Gai et al., 2021). These data were an amalgamation of several studies representing heterogeneous populations and recruitment methodologies. Each study administered the CDE questions, which took approximately 40 min to complete. In total, 1,437 participants from 12 study sites completed the questions and were included in our study. These data were representative of naturalistic settings among military-related populations (i.e., given that the study sites included a combination of inpatient psychiatric samples receiving treatment, outpatient psychiatric samples receiving treatment, and nonclinical samples not receiving treatment).

All 12 sites employed a study assessor who explained the study’s purpose to participants, acquired informed consent, and administered T1 and follow-up measurements. Measurements were administered in-person on study laptops in a randomized order to buffer potential ordering effects. Participants were encouraged to complete all measures. Of note, each study site varied regarding the number of follow-up appointments and time intervals between visits. Study interval lengths by site averaged a span of months (between time points (T) 1 and T2: M = 2.66 [6.19]; between T2 and T3: M = 3.97 [1.48]; between T3 and T4: M = 9.03 [3.53]).

Participants

The sample (N = 1,437) predominately included Active Military service members (79.8%, n = 1,147), with 290 Veterans (20.2%). Among the total sample, most participants identified as men (71.4%, n = 1,026), with the remaining participants (28.6%) identifying as women (n = 411). Most participants were Caucasian (66.6%), non-Hispanic (78.3%), married (61.87%), and ranged in age from 18 to 82 years (M = 33.09, SD = 9.96). Overall, those active in the military served in the following divisions: 56.7% served in the Army, 17.9% Reserves, 8.5% Airforce, 8.3% Marines, 7.9% Navy, and 0.8% Coast Guard. The majority of Veterans (49.8%) served in the Army, 19.4% Reserves, 11% Air Force, 10.6% Marines, 8.4% Navy, and 0.7% Coast Guard. See Table 1 for detailed socio-demographic and military service characteristics for the sample stratified by current service members, Veterans, men, and women. Of note, there was a small amount of missing data for the demographic and military experiences variables, accounting for relevant percentages above not totaling 100%.

Table 1.

Summary of Demographic Characteristics of Sample

Group (N = 1,437) Demographic	Active Military (n = 1,147)		Veteran (n = 290)		Men (n = 1,026)		Women (n = 411)
	n (range)	M (SD)	n (range)	M (SD)	n (range)	M (SD)	n (range)	M (SD)
Age	1,147 (18–59)	31.2 (8.9)	290 (22–82)	40.7 (10.1)	1,026 (18–82)	33.7 (10.0)	411 (18–59)	31.6 (9.7)
Active duty (years)	1,142 (0–41)	7.0 (5.8)	284 (0–26)	7.6 (6.3)	1,016 (0–41)	7.4 (6.1)	410 (0–26)	6.4 (5.3)
Reserves (years)	1,072 (0–42)	3.4 (5.1)	286 (0–27)	2.5 (4.7)	964 (0–42)	3.3 (5.2)	394 (0–32)	4.2 (4.5)
Deployed (times)	1,135 (0–20)	2.3 (2.6)	290 (0–20)	2.2 (3.1)	1,020 (0–20)	2.5 (2.8)	405 (0–20)	1.9 (2.3)
Demographic	n	%	n	%	n	%	n	%
Gender
Men	806	70.3	220	75.9	1,026	100.0	0	100.0
Women	341	29.7	70	24.1	0		411
Race
Caucasian	761	66.6	193	66.6	713	69.7	241	58.8
Black/African American	233	20.4	54	18.6	180	17.6	107	26.1
Asian	36	3.2	3	1.0	26	2.5	13	3.2
American Indian/Native American	26	2.3	14	4.8	26	2.5	14	3.4
Native Hawaiian/Pacific Islander	16	1.4	2	0.7	13	1.3	5	1.2
Other race	71	6.2	24	8.3	65	6.3	30	7.3
Missing	4		0		3		1
Ethnicity
Non-Hispanic/Latino	886	77.2	239	83.0	827	80.8	298	72.5
Hispanic/Latino	261	22.8	49	16.9	197	19.2	113	27.5
Missing	0		2		2		0
Marital status
Married	713	64.4	152	52.4	636	63.9	229	57.0
Widowed	4	0.4	3	1.0	4	0.4	3	0.8
Divorced/separated	70	6.3	53	18.3	79	7.9	44	11.0
Cohabitating	25	2.3	25	8.6	29	2.9	21	5.2
Single	296	26.7	57	19.7	248	24.9	105	26.1
Missing	39		0		30		9
Division
Army	650	56.7	136	49.8	571	56.5	215	52.6
Airforce	97	8.5	30	11.0	80	7.9	47	11.5
Marines	95	8.3	29	10.6	99	9.8	25	6.1
Navy	91	7.9	23	8.4	72	7.1	42	10.3
Coast guard	9	0.8	2	0.7	10	1.0	1	0.2
Reserves	205	17.9	53	19.4	179	17.7	79	19.3

Measures

The MSRC CDE include 91 items; their respective scales have demonstrated acceptable psychometric properties (see Ringer et al., 2018) assessing suicide risk-related thoughts, behaviors, and conditions. The assessment includes full and partial existing scales (75 items; with permission to utilize items obtained from the scale’s copyright holders), as well as items developed specifically for the CDE (16 items). All measures selected for the CDE assessment were designed and approved by the MSRC directors, senior advisors, and experts from the field of suicide research. Thus, the CDE reflect a collection of items assessing a broad range of suicidal symptoms and suicide-related constructs (e.g., anxiety sensitivity and thwarted belongingness) that have been empirically linked to suicide-related behaviors across multiple studies. See below for descriptions of the measures used in this study; for a full description of the development of the first wave of the MSRC CDE, please refer to Ringer et al. (2018).

Anxiety Sensitivity Index–Cognitive Subscale (Taylor et al., 2007)

The ASI-C (adapted from the ASI-3; Taylor et al., 2007) is a five-item self-report measure adapted from the original six-item cognitive subscale of the full ASI-3. These five items assess fears of cognitive anxiety-related symptoms (i.e., cognitive anxiety sensitivity). Items are rated on a 5-point Likert-type scale ranging from 0 to 4, with higher scores indicating increased cognitive anxiety sensitivity. The five items were: “When my thoughts seem to speed up, I worry that I might be going crazy”; When my mind goes blank, I worry that there is something terribly wrong with me;” “When I feel ‘spacey’ or spaced out, I worry that I may be mentally ill”; “When I have trouble thinking clearly, I worry that there is something wrong with me”; and “When I cannot keep my mind on a task, I worry that I might be going crazy.” Internal consistency was excellent for the total sample and the subsamples (i.e., current service members, veterans, men, and women), with Cronbach’s α coefficients ranging from .92 to .96 (all 95% CIs fell between .89 and .97) across all four time points. To ensure the five-item self-report measure was empirically comparable to the full six-item cognitive subscale, we compared these data to another data set of active-duty service members (N = 175). Results of the additional data set suggested that the five-item index was the better fitting model, Δχ²(4) = 21.63; p < .001.

Depressive Symptom Index–Suicidality Subscale (Joiner et al., 2002)

The Depressive Symptom Index–Suicidality Subscale (DSI-SS) is a four-item self-report measure assessing the presence and severity of suicidal thoughts, plans, and urges in the past 2 weeks. Items are rated on a 4-point scale ranging from 0 to 3, with higher scores reflecting greater severity of suicidal thoughts and urges. Internal consistency was excellent for the total sample and the subsamples, with Cronbach’s α coefficients ranging from .90 to .95 (95% CIs fell between .84 and.97) across all subsamples and all time points.

Generalized Anxiety

To increase confidence in the specificity of cognitive anxiety sensitivity, as opposed to anxiety symptoms more generally, we included two items which covered generalized anxiety phenomenology: “Being ‘super alert’ or watchful or on guard,” and “Feeling jumpy or easily startled.” The correlations between these two items ranged from .73 to .84, depending on which subsample was examined (e.g., Active Military, Veterans). These coefficients are reasonably consistent with the view that these two items form a reliable index. Relevant to the index’s validity, its correlation with ASI-C at the same time point ranged between .34 and .68 and is consistent with figures reported throughout the literature (e.g., Baek et al., 2019; Intrieri & Newell, 2022). Cronbach’s α coefficients ranged from .85 to .91 (95% CIs fell between .82 and .93) across all subsamples and all time points.

Interpersonal Needs Questionnaire–Thwarted Belongingness (Van Orden et al., 2012)

The Interpersonal Needs Questionnaire–Thwarted Belongingness is a 15-item self-report questionnaire assessing perceived burdensomeness and thwarted belongingness. Participants rate each item on a 7-point scale, ranging from 1 to 7. Higher scores indicate higher levels of perceived burdensomeness and thwarted belongingness. The CDE assessment used five INQ-TB items, which have been shown to correlate highly with the parent measure (Ringer et al., 2018). The present study used the INQ-TB as a control variable.¹ Internal consistency was excellent both for the total sample and for subsamples (i.e., current service members, veterans, men, and women), with Cronbach’s α coefficients ranging from .91 to .96 (95% CIs fell between .89 and .98) across all subsamples and all time points.

Suicide Attempts

The presence and number of suicide attempts were measured using single-item self-report questions. At the first time point, lifetime suicide attempts were assessed with the question: “How many times in your lifetime have you made an attempt to kill yourself during which you had at least some intent to die?” Participants entered the number of times that they had previously attempted suicide. Follow-up questions asked about suicide attempts in between study visits with the question: “Since you enrolled in this study or since the last assessment (whichever is most recent), how many times have you made an attempt to kill yourself during which you had at least some intent to die?” For the purposes of this study, answers were coded on a scale ranging from 0 (no suicide attempt) to 4 (four or more suicide attempts).

Data-Analytic Plan

All study variables were examined for skewness, kurtosis, linearity, and outliers. Any outliers outside of appropriate scale ranges, of which there were only two instances of scores below appropriate scale ranges in the ASI-C at T2, were winsorized (i.e., brought to the highest or lowest identified number in the acceptable range; Beaumont & Rivest, 2009). Skewness and kurtosis values were considered acceptable if between −2 and 2 (George & Mallery, 2019). All analyses were conducted in R Version 4.1.2 (R Core Team, 2023) using the lavaan package (Rosseel, 2012). First, in order to test for measurement invariance, two multigroup confirmatory factor analyses (CFA) of ASI-C were conducted across Active Military and Veterans and again across men and women. While conducting multiple-group CFA analyses, a five-step process was followed, as recommended by Brown (2006). In Step 1, to assess configural invariance, the model with no constrained parameters (i.e., the baseline model) was fit separately to data from (a) Active Military and Veterans and (b) men and women. In Step 2, we assessed for metric invariance by running multiple-group CFA, constraining factor loadings to be equivalent across Active Military and Veterans and men and women, while intercepts were allowed to vary across groups. In Step 3, we ran a model testing for intercept-only invariance. In Step 4, we assessed for scalar invariance by constraining the loadings and intercepts to be equal across the groups (i.e., Active Military and Veterans and men and women). In Step 5, we assessed for strict factorial invariance (i.e., full uniqueness), fixing the residual variances to be the same across the groups. All models were assessed by the appropriate fit indices: Root-mean-square error of approximation (RMSEA), comparative fit index (CFI), Tucker–Lewis index (TLI), standardized root-mean-square residual (SRMR), and chi-square difference tests were used to determine which level of invariance (i.e., metric [weak], scalar [strong], or residual [strict]) was met; Hu & Bentler, 1999). Measurement invariance was also examined across time before proceeding to latent growth curve modeling (LGCM).

Next, LGCM was used to examine the degree to which ASI-C scores were predominately stable or changed across four time points in (a) Active Military, (b) Veterans, (c) men, and (d) women. LGCM uses observed or measured variables (i.e., indicators) and models two latent variables—the intercept and the slope. The estimates of the mean intercept and slope represent the average starting point and the average rate of change, respectively, with these two defining the average trajectory of change across time. In addition, LGCM estimates between-subjects variability in individual intercepts and slopes, thereby providing information about how much individuals vary in their starting point and how much they vary in terms of change (Curran et al., 2010). In addition to answering whether the construct changes over time, by assessing the relationship between the intercept and the slope, LGCM can answer whether the degree of change (i.e., slope) is associated with the initial levels of the construct (i.e., intercept). Unlike traditional methods for analyzing repeated-measures data (e.g., repeated-measures analysis of variance), LGCM is more flexible as it is equipped for handling partially missing data and unequal intervals between time points. These models also allow one to assess the fit of various growth functions (e.g., linear, quadratic) and typically have greater statistical power (Curran et al., 2010).

Spaghetti model plots of a random sample of 100 participants were examined first as a whole data set, then by group to observe discernable patterns in trajectories. Linear, quadratic, cubic, and logistic functions were analyzed to assess the best fitting form of growth, and results were compared to free-loading LGCM to control for time variability between each site’s study visits. First, the overall model fit was assessed across the four models using the following fit indices: RMSEA, CFI, TLI, and SRMR (Hu & Bentler, 1999). Second, the main parameter estimates were interpreted regarding the relationships between the slope and intercept, latent means, latent variances, and residuals. For all models, the intercept was centered at T1. As this study was conducted within a military setting, there was a substantial reduction in retention across study visits. While this is expected due to a number of reasons (e.g., being relocated, reassigned, or promoted), sensitivity analyses were conducted among participants with a minimum of two study visits across all models.

Finally, we conducted a series of multiple linear regression analyses to examine if cognitive anxiety sensitivity prospectively predicted suicidal ideation and suicide attempts within each group: Active Military, Veterans, men, and women, respectively. A total of five models were run involving the Active Military sample, one for each of the five clinical predictors/outcomes: suicidal ideation at T2, suicidal ideation at T3, suicidal ideation at T4, and suicide attempt between T1 and T2 and suicide attempt between T2 and T3.² Each suicidal ideation model controlled for respective clinical predictors at the previous time point (i.e., DSI-SS at the previous time point for all suicidal ideation models) and controlled for thwarted belongingness and generalized anxiety at the previous time point. Each suicide attempt model controlled for lifetime suicide attempts and generalized anxiety. Results were reanalyzed for Veterans, men, and women, respectively. Exploratory analyses including the same five regression analyses were conducted to further understand differences between Active Military men and Veteran men. Missing data were handled utilizing maximum likelihood estimations within R for invariance analyses, LGCM, and regression analyses. Sensitivity analyses were conducted among participants with a minimum of two study visits across all models.

Transparency and Openness

Preregistration

This study was not preregistered.

Data, Materials, Code, and Online Resources

Data and materials are publicly available from the Military Suicide Research Consortium (MSRC) and can be requested through a data use agreement with the Veteran’s Association Suicide Prevention Research Impact NeTwork. The code for all analyses is available by contacting the first author.

Reporting

All data were collected through the MSRC CDE project between 2016 and 2023. These data were an amalgamation of several studies representing heterogeneous populations and recruitment methodologies. Thus, given the nature of data collection procedures, a priori power analyses were not conducted. Likewise, measures were determined before this study was conceptualized. There were no experimental manipulations in this study, and we report how we determined our sample size, all data exclusions, and all measures in the study.

Ethical Approval

Study procedures were reviewed and approved by all investigators’ respective institutional review boards, relevant military institutional review boards, and the Department of Defense Human Research Protection Office.

Results

Descriptive Statistics

All descriptive statistics for key study variables can be found in Table 2. Across all groups, mean ASI-C levels were generally highest at T1, followed by a decline at T2 and T3. The Veteran group showed a consistent downward trend across all four time points. In contrast, Active Military participants exhibited a sharp increase in ASI-C scores at T4, reaching their highest recorded levels at that point. Similarly, both men and women showed slight increases in mean ASI-C levels at the final time point. Approximately 40.7% of Active Military and 23.9% of Veterans indicated above-zero suicidal ideation (i.e., DSI-SS) at T1. Moreover, 35.5% of men and 41.7% of women reported above-zero suicidal ideation (i.e., DSI-SS) at T1. Lifetime suicide attempt was reported by a total of 43.0% of Active Military, 26.4% of Veterans, 33.9% of men, and 50.3% of women. Zero-order bivariate correlations for key study variables can be found in Supplemental Table 1. Notably, correlations between ASI-C at T1 and ASI-C at T2 (Active Military: r = .64, p < .001, Veterans: r = .72, p < .001) indicate some evidence of group fluctuations across time. ASI-C at T1 was significantly related to suicidal ideation (i.e., DSI-SS) from T1 to T3 but not T4 for Active Military and was significant across all time points for Veterans. ASI-C at T1 was significantly related to past suicide attempts at T1, T1–T2, and T2–T3, but not T3–T4 for Active Military, and was significant for all time points except T2–T3 for Veterans.

Table 2.

Summary of Descriptive Statistics of Sample

Group (N = 1,437) Descriptive statistics	Active Military (n = 1,147)			Veteran (n = 290)			Men (n = 1,026)			Women (n = 411)
	n	Range	M (SD)	n	Range	M (SD)	n	Range	M (SD)	n	Range	M (SD)
ASI-C T1	1,147	0–20	7.5 (6.5)	290	0–20	5.8 (5.9)	1,026	0–20	6.8 (6.3)	411	0–20	8.2 (6.6)
ASI-C T2	301	0–20	5.0 (6.0)	234	0–20	4.4 (5.6)	401	0–20	4.4 (5.6)	134	0–20	5.8 (6.5)
ASI-C T3	249	0–20	4.0 (5.5)	167	0–20	4.1 (4.9)	306	0–20	3.9 (5.2)	110	0–20	4.5 (5.6)
ASI-C T4	58	0–20	9.2 (6.7)	160	0–20	3.8 (4.8)	156	0–20	5.2 (5.7)	62	0–20	5.4 (6.2)
DSI-SS T1	1,144	0–12	1.9 (2.8)	289	0–11	0.9 (2.1)	1,023	0–12	1.5 (2.6)	410	0–12	2.0 (3.0)
DSI-SS T2	319	0–11	1.3 (2.1)	238	0–11	0.7 (1.9)	417	0–11	1.1 (2.1)	140	0–8	0.9 (1.8)
DSI-SS T3	255	0–8	1.0 (1.9)	168	0–6	0.4 (1.1)	312	0–7	0.7 (1.6)	111	0–8	0.8 (1.8)
DSI-SS T4	58	0–7	2.8 (2.2)	159	0–8	0.5 (1.5)	156	0–8	1.1 (2.0)	61	0–7	1.1 (1.9)
GA T1	1,145	0–8	3.7 (2.8)	245	0–8	3.1 (2.6)	990	0–8	3.3 (2.8)	400	0–8	4.0 (2.9)
GA T2	319	0–8	2.9 (2.9)	241	0–8	3.4 (2.6)	420	0–8	3.0 (2.8)	140	0–8	3.4 (2.9)
GA T3	250	0–8	2.8 (2.9)	167	0–8	3.3 (2.6)	306	0–8	3.0 (2.8)	111	0–8	3.0 (2.8)
GA T4	58	0–8	3.4 (2.8)	160	0–8	3.1 (2.5)	156	0–8	3.3 (2.5)	62	0–8	3.0 (2.7)
INQ-TB T1	1,146	5–35	15.8 (8.5)	290	5–35	17.6 (8.4)	1,025	5–35	15.8 (8.4)	411	5–35	17.1 (8.7)
INQ-TB T2	319	5–35	15.9 (8.4)	238	5–35	17.1 (8.8)	417	5–35	16.4 (8.8)	140	5–35	16.3 (8.0)
INQ-TB T3	255	5–35	16.3 (9.0)	169	5–35	15.6 (8.7)	312	5–35	15.9 (9.0)	112	5–35	16.3 (8.5)
INQ-TB T4	58	5–35	19.5 (8.6)	160	5–35	16.0 (8.1)	156	5–35	16.6 (8.0)	62	5–35	17.7 (9.2)
Past suicide attempt	789	%	0.8 (1.1)	284	%	0.5 (0.9)	768	%	0.6 (1.0)	304	%	0.9 (1.1)
None	450	57.0		209	73.6		508	66.1		151	49.7
1	147	18.6		36	12.7		119	15.5		64	21.1
2	87	11.0		17	6.0		57	7.4		47	15.5
3	105	13.3		22	7.8		85	11.1		42	13.8
Suicide attempt T1–T2	268	%	0.02 (0.2)	93	%	0.2 (0.6)	270	%	0.1 (0.4)	91	%	0.02 (0.2)
None	263	98.1		84	90.3		258	95.6		89	97.8
1	4	1.5		6	6.5		8	3.0		2	2.2
2	1	0.4		2	2.2		3	1.1		0
4 or more	0			1	1.1		1	0.4		0
Suicide attempt T2–T3	210	%	0.1 (0.4)	40	%	0.2 (0.7)	182	%	0.1 (0.4)	68	%	0.1 (0.5)
None	199	94.8		36	90.0		170	93.4		65	95.6
1	8	3.8		1	2.5		7	3.9		2	2.9
2	2	1.0		2	5.0		4	2.2		0
3	0			1	2.5		1	0.6		0
4 or more	1		0.6	0			0			1		1.5
Suicide attempt T3–T4	16	%	0.1 (0.3)	34	%	0.1 (0.5)	32	%	0.1 (0.5)	18	%	0.1 (0.2)
None	15	93.8		32	94.1		30	93.8		17	94.4
1	1	6.3		0			0			1	5.6
2	0			2	5.9		2	6.3		0

Note. ASI-C = Anxiety Sensitivity Index–Cognitive Subscale; DSI-SS = Depressive Symptom Index–Suicidality Subscale; GA = Generalized Anxiety; INQ-TB = Interpersonal Needs Questionnaire–Thwarted Belongingness; T = time point.

Invariance Testing of the ASI-C

The one-factor ASI-C (measuring fears specific to losing cognitive control) model demonstrated excellent fit regarding all indices, except RMSEA, which indicated close to adequate fit in the Active Military sample, χ²(5) = 48.31, CFI = .990, TLI = .981, SRMR = .012, RMSEA = .087, and men, χ²(5) = 54.29, CFI = .988, TLI = .976, SRMR = .014, RMSEA = .098. In Veteran, χ²(5) = 46.22, CFI = .965, TLI = .929, SRMR = .024, RMSEA = .169, and women groups, χ²(5) = 26.53, CFI = .986, TLI = .972, SRMR = .017, RMSEA = .102, all indices indicated excellent fit, except for RMSEA, which indicated poor fit (i.e., above 0.10). Although all models had higher than anticipated RMSEA values, it is worth noting that “for models with small df (degrees of freedom), the RMSEA can exceed cutoffs very often, even when the model is correctly spec-ified” (Kenny et al., 2015, p. 501). Kenny et al.’s (2015) examination concluded that RMSEA fit improves as more variables are added to a model, so the small number of variables in our model (i.e., five items) may partially explain the elevated RMSEA values.³ More details on the fit of all CFA models are available in Supplemental Table 2, and standardized factor loadings of all CFA models are available in Supplemental Table 3.⁴

Invariance testing results for T1 are presented in Table 3. Chi-square difference tests were conducted to determine best model fit, indicating that metric invariance was supported for Active Military and Veteran groups, Δχ²(4) = 3.27; p = .514. Scalar invariance was supported for men and women groups, Δχ²(8) = 15.50; p = .050. Of note, invariance testing results did not substantially vary for any other time point (i.e., T2 through T4). Scalar metric invariance was met across time, Δχ²(24) = 32.32; p = .119; please see Supplemental Table 4 for results.

Table 3.

Multiple-Group Confirmatory Factor Analysis Measurement Invariance Testing Results

Group	Active Military versus Veteran				Men versus women
	Configural	Metric	Scalar	Residual	Configural	Metric	Scalar	Residual
χ2	94.5	97.8	139.0	167.0	94.5	97.8	139.0	167.0
df	10	14	18	23	10	14	18	23
CFI	0.985	0.985	0.979	0.975	0.985	0.985	0.975	0.975
TLI	0.97	0.977	0.976	0.978	0.97	0.977	0.976	0.978
RMSEA	0.108	0.095	0.097	0.093	0.108	0.095	0.097	0.093
SRMR	0.015	0.018	0.026	0.025	0.015	0.018	0.026	0.025
Δδφ		4	8	13		4	8	13
Δχ2		3.27	44.8	72.1		3.27	44.8	72.1
p		.514	<.001	<.001		.514	<.001	<.001

Note. Invariance testing across military status was conducted on n = 1,147 Active Military versus n = 290 Veterans. Invariance testing across gender was conducted on n = 1,026 men versus n = 411 women. Maximum likelihood estimations were used to account for missing data. Values in bold indicate the best fitting model. CFI = comparative fit index; RMSEA = root-mean-square error of approximation; SRMR = standardized root-mean-square residual; TLI = Tucker–Lewis index.

Anxiety Sensitivity Over Time

None of the four functions (linear, quadratic, cubic, and logistic) showed an appropriate fit to the data when all groups were pooled together or when groups were split (i.e., Active Military, Veterans, men, and women). Due to the poor fit of these models and the time variability between each site’s study visits, a free-loading latent growth curve model was analyzed where we fixed the slope loadings at T1 and T2 and allowed the loadings at T3 and T4 to be freely estimated. By freeing the loadings of two parameters rather than fixing their growth trajectory, the model is allowed to capture the actual form of change that may not follow a conventional pattern (i.e., linear, quadratic). This allows for better representations of the data rather than forcing a pattern of growth that does not fit the data (Kline, 2023). All of the free-loading latent growth curve models suggested superior model fit to linear models, Δχ²(2) = 18.94–67.92; p < .001.

The Active Military model demonstrated good fit (see Table 4 for fit indices and parameters for all LGCM models). The relationship between slope and intercept was statistically significant, suggesting that initial levels of ASI-C were correlated with the changes in ASI-C scores. The intercept and slope were significant, indicating that Active Military participants declined at an average rate of 2.5 points on the ASI-C across the first two time points.

Table 4.

Free-Loading Growth Curve Modeling

Conditional model	Active Military	Veterans	Men	Women
χ2(df)	19.07 (3)	9.19 (3)	3.60 (3)	12.61 (3)
RMSEA [CI]	0.07 [0.068, 0.099]	0.07 [0.000, 0.125]	0.000 [0.000, 0.148]	0.088 [0.042, 0.141]
CFI	0.963	0.988	0.999	0.950
TLI	0.927	0.981	0.998	0.900
SRMR	0.078	0.037	0.025	0.077

Model parameter	Parameter	SE	Parameter	SE	Parameter	SE	Parameter	SE
Intercept	7.55***	0.193	5.57***	0.157	6.76***	0.198	8.232***	0.327
Intercept variance	17.69***	2.338	24.24***	3.108	25.23***	2.960	32.123***	7.570
Slope	−2.52***	0.267	−0.65***	0.157	−1.95***	0.223	−2.278***	0.404
Slope variance	−1.68	1.413	0.31	0.685	3.18	2.960	8.301	6.109
Covariance	3.54*	1.662	−1.35	1.27	−3.18	2.414	−7.911	6.455

Note. Maximum likelihood estimations were used to account for missing data. SE = standard error; RMSEA = root-mean-square error of approximation; CI = confidence interval; AIC = Akaike’s information criterion; BIC = Bayesian information criterion; CFI = comparative fit index, TLI = Tucker–Lewis index; SRMR = standardized root-mean-square residual.

^*p < .05.

^***p < .001.

The Veteran model demonstrated excellent model fit. The relationship between slope and intercept was not statistically significant. The slope mean was significant, indicating that Veteran participants declined at an average rate of 0.7 points on the ASI-C across the first two time points. We also conducted a chi-square difference test by constraining the model to specify the slope to be equal across groups to evaluate whether the Active Military slope was negative and larger in magnitude compared to the Veteran slope, which it was, Δχ²(2) = 27.92; p < .001.⁵

The model for men demonstrated excellent model fit. The relationship between slope and intercept was not statistically significant. The slope was statistically significant, indicating men declined at an average rate of 2.0 points on the ASI-C across the first two time points.

The model for women demonstrated adequate model fit. The relationship between slope and intercept was not statistically significant. The slope was statistically significant, indicating women declined at an average rate of 2.3 points on the ASI-C across the first two time points. Using chi-square difference testing, we found the slope was significantly higher in women than men, Δχ²(2) = 15.28; p < .001, indicating that women experience larger reductions in ASI-C than men across the first two time points. All sensitivity analyses, among participants with a minimum of two study visits, replicated these results.

The Relationship Between ASI and Suicidal Thoughts and Behaviors

As presented in Table 5, in Active Military, higher levels of ASI-C at the previous time point significantly predicted suicidal ideation at T4, above and beyond the previous time point’s suicidal ideation, generalized anxiety, and thwarted belongingness scores. However, ASI-C at T1 and T2 was not predictive of suicidal ideation at T2 and T3. Regarding suicide attempts, higher levels of ASI-C at the previous time point significantly predicted suicide attempts at T2 above and beyond lifetime suicide attempts and generalized anxiety but did not significantly predict suicide attempts at T3.

Table 5.

Linear Multiple Regression Models (Active Military vs. Veterans)

Outcome						Outcome
Predictor	b	SE	β	z	p	Predictor	b	SE	β	z	p
Active Military (n = 1,141; R2 = 0.26)						Veteran (n = 244; R2 = 0.60)
DSI-SS T2						DSI-SS T2
DSI-SS T1	0.26	0.05	0.35	5.23	<.001***	DSI-SS T1	0.75	0.05	0.74	13.81	<.001***
GA T1	0.06	0.05	0.08	1.14	.254	GA T1	−0.06	0.05	−0.07	−1.27	.205
INQ-TB T1	0.04	0.01	0.16	3.13	.002**	INQ-TB T1	0.02	0.01	0.06	1.02	.308
ASI-C T1	0.03	0.02	0.08	1.17	.243	ASI-C T1	0.03	0.02	0.08	1.24	.216
Active Military (n = 300; R2 = 0.69)						Veteran (n = 233; R2 = 0.52)
DSI-SS T3						DSI-SS T3
DSI-SS T2	0.69	0.04	0.74	15.59	<.001***	DSI-SS T2	0.49	0.06	0.68	8.94	<.001***
GA T2	0.04	0.03	0.07	1.24	.213	GA T2	0.00	0.03	0.01	0.11	.913
INQ-TB T2	0.01	0.01	0.03	0.65	.514	INQ-TB T2	0.00	0.01	0.00	0.01	.989
ASI-C T2	0.03	0.02	0.09	1.87	.061	ASI-C T2	0.02	0.02	0.09	1.18	.238
Active Military (n = 249; R2 = 0.55)						Veteran (n = 166; R2 = 0.37)
DSI-SS T4						DSI-SS T4
DSI-SS T3	0.40	0.14	0.35	2.84	.005**	DSI-SS T3	0.66	0.10	0.54	6.54	<.001***
GA T3	−0.37	0.13	−0.47	−2.87	.004**	GA T3	−0.05	0.05	−0.09	−1.07	.284
INQ-TB T3	0.11	0.03	0.46	3.50	<.001***	INQ-TB T3	0.01	0.01	0.09	1.13	.259
ASI-C T3	0.17	0.08	0.42	2.21	.027*	ASI-C T3	0.04	0.03	0.14	1.52	.129
Active Military (n = 787; R2 = 0.15)						Veteran (n = 244; R2 = 0.17)
Attempt T2						Attempt T2
Lifetime attempt	0.05	0.01	0.28	3.47	.001**	Lifetime attempt	0.11	0.04	0.28	2.42	.012*
GA T1	−0.01	0.01	−0.10	−1.20	229	GA T1	−0.03	0.02	−0.18	−1.10	.272
ASI-C T1	0.01	0.00	0.23	2.54	.011**	ASI-C T1	0.02	0.01	0.29	2.01	.044*
Active Military (n = 299; R2 = 0.16)						Veteran (n = 229; R2 = 0.09)
Attempt T3						Attempt T3
Lifetime attempt	0.13	0.04	0.31	3.35	.001**	Lifetime attempt	0.12	0.09	0.18	1.33	.183
GA T2	0.03	0.01	0.20	1.98	.047*	GA T2	0.01	0.05	0.06	0.25	.800
ASI-C T2	0.00	0.00	−0.05	−0.56	.574	ASI-C T2	0.02	0.02	0.17	1.00	.320

Note. DSI-SS = Depressive Symptom Index–Suicidality Subscale; ASI-C = Anxiety Sensitivity Index–Cognitive Subscale; GA = generalized anxiety; INQ-TB = Interpersonal Needs Questionnaire–Thwarted Belongingness; T = time point; SE = standard error.

^*p < .05.

^**p < .01.

^***p < .001.

For Veterans (Table 5), higher levels of ASI-C at the previous time point were not significantly predictive of suicidal ideation at any time point. However, higher levels of ASI-C at T1 were significantly predictive of suicide attempts at T2 above and beyond lifetime suicide attempts and generalized anxiety but were not predictive of suicide attempts at T3.

Regarding men (Table 6), higher levels of ASI-C at the previous time point were significantly predictive of suicidal ideation at T2 and T3 when controlling for the previous time point’s suicidal ideation, generalized anxiety, and thwarted belongingness. However, ASI-C at T3 was not predictive of suicidal ideation at T4. Higher levels of ASI-C at T1 were significantly predictive of suicide attempts at T2 above and beyond lifetime suicide attempts and generalized anxiety but not at T3.

Table 6.

Linear Multiple Regression Models (Men vs. Women)

Outcome						Outcome
Predictor	b	SE	β	z	p	Predictor	b	SE	β	z	p
Men (n = 986; R2 = 0.40)						Women (n = 399; R2 = 0.35)
DSI-SS T2						DSI-SS T2
DSI-SS T1	0.43	0.05	0.51	9.36	<.001***	DSI-SS T1	0.36	0.06	0.57	5.88	<.001***
GA T1	−0.04	0.04	−0.05	−0.90	.367	GA T1	0.01	0.06	0.02	0.16	.870
INQ-TB T1	0.03	0.01	0.12	2.74	.006**	INQ-TB T1	0.02	0.02	0.07	0.84	.402
ASI-C T1	0.06	0.02	0.16	2.79	.005**	ASI-C T1	0.00	0.03	0.00	0.00	.997
Men (n = 399; R2 = 0.67)						Women (n = 134; R2 = 0.67)
DSI-SS T3						DSI-SS T3
DSI-SS T2	0.62	0.04	0.72	16.36	<.001***	DSI-SS T2	0.74	0.07	0.75	10.31	<.001***
GA T2	0.01	0.03	0.02	0.49	.623	GA T2	0.06	0.05	0.11	1.32	.187
INQ-TB T2	0.00	0.01	0.01	0.13	.900	INQ-TB T2	0.01	0.01	0.07	1.00	.314
ASI-C T2	0.04	0.01	0.14	2.88	.004**	ASI-C T2	0.00	0.02	−0.01	−0.17	.867
Men (n = 306; R2 = 0.52)						Women (n = 109; R2 = 0.47)
DSI-SS T4						DSI-SS T4
DSI-SS T3	0.70	0.09	0.59	7.68	<.001***	DSI-SS T3	0.69	0.15	0.66	4.47	<.001***
GA T3	−0.19	0.06	−0.26	−3.05	.002**	GA T3	−0.08	0.09	−0.12	−0.90	.370
INQ-TB T3	0.06	0.02	0.26	3.64	<.001***	INQ-TB T3	0.04	0.02	0.19	1.79	.074
ASI-C T3	0.06	0.04	0.17	1.70	.088	ASI-C T3	−0.01	0.05	−0.02	−0.13	.895
Men (n = 736; R2 = 0.21)						Women (n = 295; R2 = 0.17)
Attempt T2						Attempt T2
Lifetime attempt	0.08	0.02	0.33	4.81	<.001***	Lifetime attempt	0.05	0.02	0.31	2.51	.012*
GA T1	−0.01	0.01	−0.15	−1.84	.066	GA T1	−0.01	0.01	−0.20	−1.55	.122
ASI-C T1	0.01	0.00	0.30	3.22	.001**	ASI-C T1	0.01	0.00	0.26	1.89	.058
Men (n = 396; R2 = 0.11)						Women (n = 132; R2 = 0.13)
Attempt T3						Attempt T3
Lifetime attempt	0.10	0.03	0.25	3.18	.001**	Lifetime attempt	0.13	0.08	0.22	1.53	.125
GA T2	0.01	0.01	0.07	0.65	.516	GA T2	0.05	0.04	0.26	1.51	.141
ASI-C T2	0.01	0.01	0.10	1.05	.293	ASI-C T2	−0.01	0.01	−0.09	−0.61	.540

Note. DSI-SS = Depressive Symptom Index–Suicidality Subscale; ASI-C = Anxiety Sensitivity Index–Cognitive Subscale; GA = generalized anxiety; INQ-TB = Interpersonal Needs Questionnaire–Thwarted Belongingness; SE = standard error; T = time point.

^*p < .05.

^**p < .01.

^***p < .001.

For women (Table 6), higher levels of ASI-C at the previous time point were not significantly predictive of suicidal ideation at T2, T3, or T4. Additionally, higher levels of ASI-C at the previous time point were not significantly predictive of suicide attempts at T2 or T3 when controlling for lifetime suicide attempts and generalized anxiety.

Exploratory analyses were conducted to better understand the differences between Active Military men and Veteran men (see Supplementary Table 5). Regarding Active Military men, higher levels of ASI-C at the previous time point were significantly predictive of suicidal ideation at T3 (ASI-C T2: β = 0.16, p = .006) above and beyond the previous time point’s suicidal ideation, generalized anxiety, and thwarted belongingness scores but not at T2 (ASI-C T1: β = 0.15, p = .080) or T4 (ASI-C T3: β = 0.40, p = .081). Higher levels of ASI-C at the previous time point were not significantly predictive of suicide attempts at T2 (ASI-C T1: β = 0.14, p = .140) or T3 (ASI-C T2: β = 0.10, p = .364). Regarding Veteran men, higher levels of ASI-C at the previous time point were significantly predictive of suicidal ideation at T4 (ASI-C T3: β = 0.22, p = .045) above and beyond the previous time point’s suicidal ideation, generalized anxiety, and thwarted belongingness scores but not at T2 (ASI-C T1: β = 0.05, p = .105) or T3 (ASI-C T2: β = 0.10, p = .251). Higher levels of ASI-C at the previous time point were significantly predictive of suicide attempts at T2 (ASI-C T1: β = 0.51, p = .002) above and beyond lifetime suicide attempts and generalized anxiety but not at T3 (ASI-C T2: β = 0.41, p = .068). All sensitivity analyses, among participants with a minimum of two study visits, replicated these results.

Discussion

The aims of the present study were threefold. First, we aimed to assess the measurement invariance of the cognitive anxiety sensitivity subscale (i.e., fears specific to a loss of cognitive control; ASI-C), across Active Military personnel and Veterans as well as across gender. Second, we examined the construct stability of ASI-C across Active Military personnel and Veterans as well as across gender. Third, we examined if the previous study visit’s cognitive anxiety sensitivity predicted the following study visit’s suicidal ideation and number of suicide attempts, respectively. Our first hypothesis was supported: The ASI-C showed metric invariance by service status and scalar invariance by gender, indicating consistent construct measurement. These results support comparing construct relationships (e.g., correlations, regressions) across service status and latent mean levels across gender. Scalar invariance over time also supported testing our second hypothesis.

Examining change in cognitive anxiety sensitivity over time, our second hypothesis was partially supported. Examining change in cognitive anxiety sensitivity over time, our second hypothesis was partially supported. Growth curve models using maximum likelihood estimation indicated statistically significant negative slopes in ASI-C scores across all groups. Among Active Military participants, there was a significant negative correlation between slope and intercept, suggesting that those with higher initial ASI-C scores showed steeper declines. However, descriptive statistics at T4 revealed a different trend: Mean ASI-C scores increased among Active Military, men, and women, while Veterans continued to decline. This may reflect group differences in retention. Only 5% of Active Military participants (i.e., 58 of 1,147) remained at T4, compared to over 55% of Veterans (i.e., 160 of 290). Thus, model estimates for Active Military were predominately driven by data from T1 through T3—where scores consistently declined. Similarly, while retention among men and women were similar (i.e., around 15%), the majority of these subsamples were drawn from Active Military, with a smaller proportion of men and women identifying as Veterans. This may suggest that Active Military, both men and women, at T4 had higher symptom severity which is further supported by elevations in suicidal ideation and thwarted belongingness. These findings highlight the need to interpret model-based estimates alongside observed data.

Regarding our exploratory analyses, Active Military showed the steepest average decline in cognitive anxiety sensitivity (−2.5), followed by women (−2.3), men (−2.0), and Veterans (−0.7). Slopes differed significantly between groups and were independent of treatment effects (i.e., received treatment intervention, active treatment control, and no treatment).⁶ General age-related declines in anxiety sensitivity (Mahoney et al., 2015) or therapeutic effects of repeated measurement (Finn, 2020; Meyer et al., 2001) may help explain these reductions. An explanation for the observed decrease from T1 to T3 in both Active Military and Veterans may be the well-documented utilization of mental health services; 25.5% of Active Service Members and 49% of Veterans reported using at least one mental health service (Piro et al., 2023; U.S. Department of Veteran Affairs, 2023). However, the lack of slope differences by treatment suggests that treatment alone cannot explain the observed decrease in cognitive anxiety sensitivity. Other protective factors might contribute to it, especially in the Active Military. For instance, strong social support during service (i.e., group cohesion with fellow service members) may at least partially explain the greater decrease in Active Military compared to Veterans. Interestingly, women experienced a greater decrease in cognitive anxiety sensitivity than men, which seems to contradict previous literature surrounding gender differences in anxiety sensitivity. Previous research suggests men’s variability on cognitive anxiety sensitivity tends to be largely accounted for by environmental factors, which are more variable over time (Taylor et al., 2008), while in women, cognitive anxiety sensitivity tends to be more heritable. Therefore, future research should continue to explore the longitudinal course and potential malleability, of cognitive anxiety sensitivity, in particular among women.

Our third hypothesis was partially supported. Generally, T2 cognitive anxiety sensitivity predicted suicidal ideation at T3 beyond control variables for Active Military, while all other time points were nonsignificant for both groups (i.e., Active Military and Veterans). Elevated levels of cognitive anxiety sensitivity predicted future suicidal ideation beyond control variables among men but was not significant for women. Regarding suicidal behavior, the relationship between elevated levels of cognitive anxiety sensitivity at T1 predicted future suicide attempts at T2 beyond control variables for Active Military, Veteran, and male participants but not for women.

Among Active Military, higher levels of cognitive anxiety sensitivity at the previous time point predicted higher levels of suicidal ideation at T4 above and beyond the previous time point’s suicidal ideation, generalized anxiety, and thwarted belongingness, though not at T2 or T3.

Moreover, among both Active Military and Veterans, cognitive anxiety sensitivity at T1 also predicted higher numbers of suicide attempts at T2 above and beyond lifetime attempts and generalized anxiety, though not at T3. One possible explanation for this trend could be the time differences between study visits: On average, the length between T1 and T2 was around 1.5 months shorter than between T2 and T3. Thus, cognitive anxiety sensitivity may only impact suicidal behavior in the short term (i.e., from weeks up until 2 and a half months, but not beyond) but may impact suicidal ideation in the long term (i.e., beyond 2 and a half months). Thus, future research should consider examining the longevity of cognitive anxiety sensitivity’s effect on suicidal ideation and suicide attempts. Nevertheless, it is important to acknowledge that suicide attempts occurred at low base rates between study visits; therefore, these results should be interpreted with caution.

Our findings also indicate men may be at particularly high suicide risk due to experiencing cognitive-based anxiety stressors. For men, higher levels of cognitive anxiety sensitivity at the previous time point predicted higher levels of suicidal ideation at follow-up, above and beyond the previous time points’ suicidal ideation, generalized anxiety, and thwarted belongingness, for T2 and T3. Moreover, higher levels of cognitive anxiety sensitivity at the previous time point predicted future suicide attempts at T2 but not at T3, above and beyond the number of lifetime suicide attempts and generalized anxiety. For women, cognitive anxiety sensitivity at the previous time point did not predict any future suicidal thoughts or behaviors. This may suggest that higher levels of cognitive anxiety sensitivity at the previous time point are only predictive of future suicidal thoughts or behaviors among male military personnel. Thus, there may be a gender-specific difference in coping with external stressors (e.g., combat exposure, interpersonal stress, or financial distress) such as disclosure to others that may exacerbate the relationship between cognitive anxiety sensitivity and suicidal ideation at later time points. Taken together, future research should continue to explore how gender may confer unique suicide risk factors.

Though exploratory, results from this study suggest that suicide risk is predicted by previous cognitive anxiety sensitivity scores for Veteran men, more so than Active Military men. Higher levels of cognitive anxiety sensitivity at the previous time point were significantly predictive of suicidal ideation at T3, above and beyond control variables, for Active Military men and T4 for Veteran men. Additionally, higher levels of cognitive anxiety sensitivity at the previous time point were significantly predictive of suicide attempts at T2 among Veteran men.

While difficulties managing any form of anxiety can interfere with adaptive functioning, cognitive consequences, specifically, can disrupt mental comfort, focus, and control (Allan, Capron, Raines, et al., 2014). Troubles regulating cognitive anxiety sensitivity have likely implications for Active Military members in their role (e.g., health preoccupation and mental wandering causing interference during a particularly challenging training seminar or a more complex task). For Veterans, preoccupations surrounding cognitive anxiety sensitivity can exacerbate feelings of helplessness or hopelessness regarding one’s mental health, decrease tolerance for distress, or lead to excessive cognitive rumination, all of which are strong precursors of suicidal ideation (Capron et al., 2013; Van Orden et al., 2010). Thus, our findings may suggest that many military personnel, in particular Veteran men, may benefit from cognitive anxiety sensitivity interventions, which may have an array of helpful effects, including suicide risk mitigation.

Taken together, cognitive anxiety sensitivity appears to be a viable treatment target (i.e., due to its malleable nature) and could provide a proxy approach toward reducing suicidal thoughts and behaviors. Examples of effective, brief interventions for anxiety sensitivity have been supported in the literature, such as Anxiety Sensitivity Amelioration Training (Schmidt et al., 2007), Anxiety Sensitivity Education and Reduction Training (Keough & Schmidt, 2012), cognitive anxiety sensitivity treatment (Schmidt et al., 2014), cognitive bias modification focused on changing interpretation bias (Schmidt et al., 2017), and cognitive behavioral therapy interventions (Smits et al., 2008), all of which offer suggestions for future clinical research. Given the observed association between cognitive anxiety sensitivity and suicide risk and the particularly high suicide risk in Veterans (Defense Casualty Analysis System, 2023; U.S. Department of Veteran Affairs, 2023), studies of brief anxiety sensitivity interventions with this population are of particular interest. One such study found a brief intervention led to sustained reductions in anxiety sensitivity 3 years later (Schmidt et al., 2023). This suggests that not only is anxiety sensitivity modifiable, but interventions can have effects lasting beyond the collective time period for data collection in this study (i.e., summing the average time between study visits results in a study period of around 16 months).

Further emphasizing cognitive anxiety sensitivity’s potential utility as a modifiable risk factor for suicide is its overlap with existing suicide theories. For example, Capron et al. (2013) developed the Depression–Distress Amplification Model of Anxiety Sensitivity, in which mood pathology, combined with cognitive anxiety sensitivity, leads to increased distress, which may translate into even greater suicide risk.

Constraints on Generality

Though this study has its strengths, there are several limitations worth noting. First, due to the low base rates of nonbinary, transgender, and other self-identified genders, gender was examined as a binary variable, and gender minorities were not included in these analyses. Thus, future research should aim to continue to test these hypotheses within other gender identities. As these data were drawn across 12 study sites, future research should also examine how both individual and study site interval variability (i.e., weeks between study visits) may impact the change in cognitive anxiety sensitivity over time. Further, future research should consider including protective factors (e.g., unit cohesion and use of mental health resources) to robustly examine the effects of cognitive anxiety sensitivity. Due to sample size restrictions, models assessing Active Military men versus Veteran men may have been limited in detecting effects. When exploratorily analyzing Active Military women versus Veteran women, the models did not meet sufficient model fit criteria (i.e., elevated levels of RMSEA) and, thus, were excluded from exploratory analyses. As such, future research should replicate the present findings. Regarding model fit, RMSEA was elevated across most groups, which may be indicative of lower degrees of freedom (i.e., five) due to the number of observed variables and parameters to be estimated in each model (see Kenny et al., 2015); nevertheless, we believe this was somewhat tested within this article as we compared the five- versus six-item ASI-C. As such, future models may consider inclusion of the entire ASI to continue to test model fit. Finally, those participating at T4, which had lower sample size rates by group, may have had elevated levels of stress and impairment, as evidenced by the increase in mean scores of suicidal ideation and thwarted belongingness, which may explain the elevations in cognitive anxiety sensitivity in the final study visit. Moreover, suicide attempts occurred at low base rates between study visits.

It is also important to highlight the strengths of this study. The present study uses a sizeable sample of military personnel (both Active Military and Veteran) which are high-risk groups for suicidal thoughts and behaviors. Further, the study used four waves of data to examine the stability of the relationship between cognitive anxiety sensitivity and suicidal thoughts and behaviors. The prevalence of self-reported suicidal ideation and lifetime suicide attempts was elevated in the current sample, relative to general population estimates, evidencing the high-risk nature of this sample. This work provides support for a relatively novel risk factor, which may further enhance our ability to treat and prevent suicide among the U.S. Military.

Conclusion

Taken together, these results show the ASI-C is an acceptable measure to use across military-related (Active Military/Veterans) and gender (men/women) groups. Moreover, cognitive anxiety sensitivity demonstrated a statistically significant reduction over time across all groups, with the fastest reductions among women and Active Military and the slowest reductions in the Veteran group. Finally, cognitive anxiety sensitivity was shown to significantly predict future suicidal thoughts and behaviors among Active Military (particularly men) and behaviors among Veterans (particularly men).⁷

As suicide is a major public health concern and a clinically important issue, identifying constructs that may prospectively predict suicidal ideation and behavior is of key importance. The results of this study support the notion that elevated levels of cognitive anxiety sensitivity impact future suicidal risk in military samples. Moreover, cognitive anxiety sensitivity decreases over time, suggesting it is a malleable construct. Together, its malleability and relevance to suicide risk suggest that cognitive anxiety sensitivity may be a reasonable treatment target for suicide prevention and intervention initiatives.

Supplementary Material

Supplemental materials: https://doi.org/10.1037/pas0001406.supp

Public Significance Statement.

This study suggests that cognitive anxiety sensitivity may be a viable treatment target to reduce suicide risk among military men.