Refinement of Anxiety Sensitivity Measurement: The Short Scale Anxiety Sensitivity Index (SSASI)

Abstract

Anxiety sensitivity, defined as the fear of anxiety and arousal-related sensations, has been among the most influential cognitive-based transdiagnostic risk and maintenance factors in the study and treatment of emotional and related disorders. The currently available anxiety sensitivity measures are limited by their length. Specifically, the length of these instruments discourages the adoption of routine anxiety sensitivity assessment in clinical or medical settings (e.g., primary care). The goals of this study were to develop and assess the validity and reliability of a short version of the Anxiety Sensitivity Index-3 (ASI-3; Taylor et al., 2007), entitled the Short Scale Anxiety Sensitivity Index (SSASI), using three independent clinical samples. Results indicated that the abbreviated five-item version of the SSASI had good internal consistency and a robust association with the ASI-3. Further, across the samples, there was evidence of unidimensionality and excellent convergent and discriminant validity. There also was evidence of partial measurement invariance across sex and full measurement invariance across time. Overall, the five-item scale offers a single score that can be employed to measure anxiety sensitivity. Use of the SSASI may facilitate screening efforts and symptom tracking for anxiety sensitivity, particularly within clinical settings where practical demands necessitate the use of brief assessment instruments.

1. Introduction

Anxiety sensitivity, defined as the fear of anxiety and arousal-related sensations (McNally, 2002), has been among the most influential cognitive-based transdiagnostic risk and maintenance factors in the study and treatment of emotional and related disorders (Leventhal and Zvolensky, 2015; McNally, 2002; Taylor, 2014). In theory, anxiety sensitivity predisposes individuals to the development of anxiety/depressive problems by amplifying negative mood states (e.g., anxiety). This construct may serve as a „meta-anxiety‟ indicator that encourages those with higher anxiety sensitivity to be more attentive to anxiety-related symptoms and sensations. To illustrate, when a person with high anxiety sensitivity experiences physiological sensations, she/he is likely to misinterpret the symptoms as signs of impending personal threat (e.g., “I‟m going crazy”) and experience them as emotionally toxic (e.g., “I can‟t stand this discomfort anymore”). Thus, anxiety sensitivity is an „amplifying factor,‟ enhancing the aversiveness and need to escape/avoid negative affective or somatic experiences (Otto et al., 2016; Taylor, 2014). Most of the past work on anxiety sensitivity suggests its maintains a dimensional latent structure (Bernstein et., 2010), although there are some reports that it may have a taxonic (class) structure with in-class dimensional variability (Bernstein et., 2011).

Although originally studied predominately in relation to panic psychopathology, anxiety sensitivity has now been implicated in a larger array of clinical health problems. For example, existing research supports an association between anxiety sensitivity and posttraumatic stress disorder (PTSD; Fedroff et al., 2000), major depressive disorder (Taylor et al., 1996), social anxiety disorder (Scott et al., 2000), hypochondrias (Eifert and Zvolensky, 2004), chronic pain (Asmundson et al., 2000), obsessive-compulsive disorder (OCD; Naragon-Gainey, 2010), and several medial disorders (e.g., respiratory illness; McLeish et al., 2011). Furthermore, AS has been related to addictive behaviors, such as smoking (Leventhal and Zvolensky, 2015), hazardous drinking (Schmidt et al., 2007), cannabis use problems (Johnson et al., 2010), and other forms of substance abuse (e.g., opiate addiction; Lejuez et al., 2008). The anxiety sensitivity findings have been observed across experimental, cross-sectional, and longitudinal designs (Hayward et al., 2000; Li and Zinbarg, 2007; Maller and Reiss, 1992; Marshall et al., 2010; Schmidt et al., 2010; Schmidt et al., 1997, 1999; Schmidt et al., 2006b).

Several studies have identified anxiety sensitivity as a malleable treatment target (e.g., Otto and Reilly-Harrington, 1999; Vujanovic et al., 2012). Research to date has found that anxiety sensitivity reduction interventions are efficacious in the prevention and treatment of panic-related psychopathology as well as anxiety symptoms more generally (Gardenswartz and Craske, 2001; Schmidt et al., 2008; Schmidt et al., 2007). Furthermore, anxiety sensitivity appears to be a chief explanatory element (mechanism) in positive treatment gains for intervention programs for certain emotional disorders. Smits et al. (2008) found, for example, large controlled effect sizes in change of anxiety sensitivity levels for treatment-seeking anxiety disorder samples, and anxiety sensitivity changes accounted for significant variance in emotion symptom reduction (Smits et al., 2004).

Anxiety sensitivity was originally measured using the 16-item Anxiety Sensitivity Index (ASI; Reiss et al., 1986). Efforts to improve the psychometric properties of the ASI have yielded additional variants of the measures, including 36-item Anxiety Sensitivity Index–Revised (ASI-R; Taylor and Cox, 1998b), 60-item Anxiety Sensitivity Profile (ASP; Taylor and Cox, 1998a), and the 18-item Anxiety Sensitivity Index–3 (ASI-3; Taylor et al., 2007). Among youth, anxiety sensitivity has been measured via the Childhood Anxiety Sensitivity Index, which has shown good psychometric properties (CASI; Silverman et al., 1991). Past work among adults suggests that the ASI-3 demonstrates the strongest psychometric properties (Taylor et al., 2007). Across measures, past work suggest that the anxiety sensitivity construct is comprised of one higher order factor and three specific lower-order factors (Taylor et al., 2007). The three lower-order factors represent Physical Concerns (e.g., “When I feel pain in my chest, I worry that I’m going to have a heart attack”), Cognitive Concerns (e.g., “When my thoughts seem to speed up, I worry that I might be going crazy”), and Social Concerns (e.g., “I worry that other people will notice my anxiety”). Further, there is evidence that the dimensional variability in the lower-order factors differential relate to domain-specific fears (Zvolensky et al., 2001; Zvolensky et al., 2002).

The current available anxiety sensitivity measures, although valid and reliable, are limited by their length. Specifically, the length of these measures discourages the adoption of routine anxiety sensitivity assessment in clinical or medical settings, despite its robust association with clinically-relevant health outcomes (Mark et al., 2008). Indeed, in general, longer measures, such as those currently available to assess anxiety sensitivity, are less likely to be integrated into clinical settings (Mark et al., 2008). Efficiency is a particularly important attribute of assessment instruments in clinical settings that are busy and are marked by competing demands, leading researchers to develop short-form versions of instruments to increase their routine use in such settings (e.g., Kroenke et al., 2003). Following from that line of research, a shorter measure of anxiety sensitivity may increase the likelihood of anxiety sensitivity assessment, and thereby, allow for a greater understanding of the impact of anxiety sensitivity on health within the larger community. Additionally, some evidence suggests longer measures result in reduced response rates and lower data quality relative to their shorter counterparts (Diehr et al., 2005; Snyder et al., 2007). Thus, another potential benefit of a shortened anxiety sensitivity measure may be an increase in response quality. Increasing response rate and quality could help improve patient care and decrease costs as clinical decisions would be informed by more accurate data. Yet, to date, no short scale has been developed or validated to assess anxiety sensitivity.

Due to the clinical relevance of the anxiety sensitivity construct and its wide applicability to psychopathology, addictive behavior, and physical illness and disease, scales that have been developed for the construct have been increasingly applied in a variety of community contexts. Yet, the best performing scale, the ASI-3 along with earlier alternative instruments, are significantly lengthy for use in such contexts and limits their widespread adoption. Consequently, there is a need for a shorter scale to measure the construct in a reliable and valid fashion that would be more user-friendly for administration in community settings (e.g., dental clinics, primary care). Developing a shorter scale for the anxiety sensitivity construct also would increase its efficiency in clinical studies and eliminate redundancy within the measure. The goals of this study were to assess the validity and reliability of a short version of the ASI-3, entitled the Short Scale Anxiety Sensitivity Index (SSAI), using three independent clinical samples.

The development and validation of the SSASI was carried out in three studies:

Study 1 Aims and Hypotheses: Study 1 aimed to derive the SSASI from the ASI-3. The ASI-3 was administered to a community sample of treatment seeking adults in primary care. Data were evaluated to form the SSASI and provide preliminary evidence for construct validity. It was hypothesized that at 5-item SSASI measure would provide strong correlations with the ASI-3 (total score and subscales) and be associated with measures related to the ASI-3 total score.

Study 2 Aims and Hypotheses: Study 2 aimed to replicate the SSASI model structure derived in Study 1 among an independent clinical sample and explore construct validity. The second community sample of adults served to cross-validate the SSASI in an independent sample. For this study, the SSASI was evaluated for measurement invariance across sex and construct validity. It was hypothesized that the SSASI derived in Study 1 would be replicated in Study 2, and that the SSASI would demonstrate measurement invariance across sex as well as evidence for concurrent validity.

Study 3 Aims and Hypotheses: Study 3 aimed to replicate the SSASI structure in an independent sample of smokers recruited from the community and evaluate construct validity. The third community sample of treatment seeking smokers served to cross validate the SSASI in an independent sample of substance users. For this study, the SSASI was evaluated for measurement invariance across time, test-retest reliability, and construct validity. It was hypothesized that the SSASI structure would be replicated in the present sample, and that the SSASI would demonstrate measurement invariance across time, strong test-retest reliability, as well as evidence for concurrent and discriminant validity.

2. Method

2.1. Study 1

2.1.1. Participants and Procedures

The sample consisted of 536 (76.3% female; M_age = 45.54 years; SD = 17.45) adults presenting for medical treatment at a community health center. The racial/ethnic distribution of this sample was as follows: 34.1% Black, 32.1% Latino, 33.0% White, 0.6% Asian, and 0.2% Other. Participants reported an average ASI-3 total score of 21.05 (SD = 17.37).

Individuals presenting for treatment at a community health center were approached by a trained research assistant in waiting rooms, where the study purpose was described. Individuals who agreed to participate provided written informed consent and completed study measures individually in the waiting room. A more private location was made available, if requested. All participants stated that English was their preferred language for communication. Participants consented to have a research team member complete a medical record review. Participants were entered into a raffle to have a chance (≈10%) of winning a $20 gift card. The research was approved by an institutional review board serving the local medical community. Further, the investigation was carried out in accordance with the latest version of the Declaration of Helsinki.

2.1.2. Measures

Anxiety Sensitivity Index-3 (ASI-3; Taylor et al., 2007).

The ASI-3 is an 18-item measure that assesses anxiety sensitivity, including physical (e.g., It scares me when I become short of breath), cognitive (e.g., When I feel “spacey” or spaced out I worry that I may be mentally ill), and social (e.g., It is important for me not to appear nervous) concerns. ASI-3 items are rated using a 5-point scale (ranging from 0 to 4). The ASI-3 was adapted from the original ASI (Reiss et al., 1986) to provide a more stable assessment of the three most commonly replicated ASI subscales: physical concerns, cognitive concerns, and social concerns. The ASI-3 total score and respective facets demonstrated acceptable to excellent internal consistency in this study (total score; α = 0.92; physical subscale: α = 0.85; cognitive subscale: α = 0.85; social subscale: α = 0.77).

Patient Health Questionnaire-15 (PHQ-15; Kroenke et al., 2002).

The PHQ-15 is a 15-item self-report measure that assesses the severity of somatic symptoms (e.g., stomach pain, chest pain, dizziness) over the past month using a 3-point scale (ranging from 0 to 2). The PHQ-15 is considered a preferred measure to assess for the severity of somatic symptoms (Kroenke, 2007). There is one item of the PHQ-15 (i.e., menstrual cramps or other problems with your periods) that is applicable only to women. The PHQ-15 showed good internal consistency in this study (α = 0.83).

Whiteley Index (WI; Pilowsky, 1967).

The original WI was a 14-item measure that assessed health anxiety (e.g., Do you worry a lot about your health?) using a true/false rating of items. We used a six-item short-form of the original WI that improves the factorial validity of the WI (Asmundson et al., 2008; Welch et al., 2009). We also used the recommended 5-point rating scale (ranging from 1 to 5) instead of the original dichotomous response option (Welch et al., 2009). The WI-6 correlates strongly (rs of 0.63 and 0.80) with other measures of health anxiety (Fergus and Bardeen, 2013). The WI-6 showed good internal consistency in this study (α = 0.83).

Discomfort Intolerance Scale (DIS; Schmidt et al., 2006a)

The DIS is a five-item scale that assesses discomfort intolerance using a 7-point scale (ranging from 0 to 6). Items 1 and 2 are reverse coded. Schmidt et al. (2006) found that the DIS assesses two dimensions of discomfort intolerance: a two-item scale assessing the inability to tolerate physical discomfort (“intolerance,” e.g., I can tolerate a great deal of physical discomfort) and a three-item scale assessing the avoidance of physical discomfort (“avoidance,” e.g., I am more sensitive to feeling discomfort compared to most persons). There is some evidence that the two DIS scales have distinct correlates (Bonn‐Miller and Zvolensky, 2009; Schmidt et al., 2007; Schmidt et al., 2006a), although the two-factor structure of the DIS has not been uniformly supported (Luberto et al., 2013). Consequently, a total score was used in this study. The DIS evidenced internal consistency below conventional guidelines (α = 0.62), although the DIS had an average inter-item correlation (0.25) that fell in recommended ranges (i.e., between 0.15 and 0.50; Clark and Watson, 1995).

2.1.3. Data Analytic Strategy

Informed by guidelines for short-form development (Smith et al., 2000), a priori criteria for the shortened scale were strong content validity, sufficient content coverage, a good coefficient alpha value, and a small number of items (3–5 items). We used Hayes‟ alphamax macro (Hayes, 2012), an SPSS-based scale-shortening algorithm, to inform item selection. This computation tool takes the responses to a set of indicators of a latent variable and computes the Cronbach‟s alpha for all possible “subscales” (α_sub), and the correlations between the average response to the indicators on the subscale and the average response to the indicators on the full scale (r_sub). Considering the aim of the short-form was to offer a brief alternative to the longer anxiety sensitivity assessments without sacrificing content or utility, we coupled our statistical analyses with a theoretical approach to item inspection and retention. Specifically, we inspected possible subscales for items that (a) demonstrated high factor loadings on the general factor relative to other items in their respective subdomain (Allan et al., 2015; Ebesutani et al., 2014; Osman et al., 2010) and (b) were not overly redundant in wording as to limited content coverage. We evaluated five subscales with the highest r_sub that also demonstrated good reliability as a composite scale (using Cronbach’s alpha 0.80 as the criterion; Hegarty et al., 2005) in a confirmatory factor analysis (CFA). Analyses were conducted using Mplus 8 (Muthén and Muthén, 2017). Robust weighted least squares estimator (WLSMV in Mplus) was employed. Overall model fit was assessed using the χ2 statistic and several fit indices. A nonsignificant χ² test statistic indicates good model fit. Other model fit statistics and association criteria included: root mean square error of approximation (RMSEA), with values less than 0.06 indicating excellent fit, values less than 0.08 indicating acceptable fit, and values above 0.10 suggesting poor fit and the Comparative Fit Index (CFI), with values between 0.95 and 1.00 indicating excellent fit and values between 0.90 and 0.94 indicating acceptable fit (Awang, 2012; Hu and Bentler, 1999). The model with the strongest model fit was evaluated for adequate content coverage.

Analyses were conducted to evaluate the construct validity of the selected SSASI items. Specifically, bivariate correlations were conducted between the SSASI total score, ASI-3 total, and the three ASI-3 subscales. Additionally, for comparison purposes, correlations were examined across the SSASI and ASI-3 total score with constructs that significantly relate to the ASI-3, including the PHQ-15, WI, and DIS (Fergus et al., 2018); observed scale scores were used to evaluate correlations. It was hypothesized that both the SSASI would positively and significantly relate to the ASI-3 total score as well as all ASI-3 subscales, and that the SSASI and ASI-3 total score would significantly and positively relate to the PHQ-15, WI, and DIS.

2.1.4. Results and Discussion

Possible subscales were ranked according to their correlation with the ASI-3 total scale. Subscales with the highest r_sub and good Cronbach‟s alpha were evaluated in sequential order (see Table 1 for evaluated models). Informed by model fit indices and adequate content coverage (i.e., demonstrated high factor loadings on the general factor relative to other items in their respective subdomain (Allan et al., 2015; Ebesutani et al., 2014; Osman et al., 2010) and were not overly redundant in wording), a five-item subscale was selected that consisted of ASI-3 items 6, 8, 12, 14, and 18. The subscale demonstrated good internal consistency (alpha = 0.81) and a high correlation with the ASI-3 total (r = 0.96). Moreover, the items demonstrated strong loadings in their respective facets and sufficient content coverage (i.e., items 8 and 12 represented the physical content, items 14 and 18 represented the cognitive content, and item 6 represented social content).

Table 1.

Evaluated SSASI models.

Potential SSASI Models	rsub	αsub	χ2 (df)	CFI	RMSEA
Model 1: Items 6, 8, 12, 14, 18	.956	.81	9.24 (5)	.998	.04
Model 2: Items 3, 5, 6, 14, 16	.956	.80	80.54 (5)	.960	.17
Model 3: Items 3, 5, 6, 14, 18	.956	.80	66.49 (5)	.967	.15
Model 4: Items 3, 5, 7, 9, 18	.955	.81	59.75 (5)	.972	.14
Model 5: Items 5, 7, 9, 11, 16	.955	.82	83.07 (5)	.965	.17

Note. N = 536. Bold font indicates the model with the best fit indices. r_sub = correlation between model items and full scale; α_sub = Cronbach‟s alpha for the model items; χ² = Chi-square test; df = degrees of freedom; CFI = comparative fit index; RMSEA = root mean square error of approximation.

Next, correlations were evaluated across the SSASI, ASI-3 total score, and ASI-3 subscales. The SSASI significantly and positive correlated with the ASI-3 total score and all subscales (total: r = 0.96, p < 0.001; physical subscale: r = 0.89, p < 0.001; cognitive subscale: r = 0.89, p < 0.001; social subscale: r = 0.86, p < 0.001). Lastly, associations between the SSASI and ASI-3 total and PHQ-15, WI, and DIS were examined. Table 3 presents correlations. As hypothesized, SSASI and ASI-3 significantly and positively related to the PHQ-15, WI, and DIS, as well as with one another. The magnitude of the observed correlations was comparable across the SSASI and ASI-3 (see Table 3).

Table 3.

Bivariate correlations and descriptive statistics.

	Study 1 (N = 531)			Study 2 (N = 770)
Variable	ASI3	SSASI	M (SD)	Variable	ASI3	SSASI	M (SD)
ASI-3	--		21.05 (17.37)	ASI-3	--		32.15 (15.50)
SSASI	0.96 (0.95, 0.96)	--	5.57 (5.41)	SSASI	0.93 (0.92, 0.94)	--	8.14 (5.13)
PHQ-15	0.48 (0.40, 0.55)	0.46 (0.39, 0.53)	11.84 (6.11)	Psychopathology	0.35 (0.28, 0.41)	0.29 (0.22, 0.36)	744 (88.0%)a
WI	0.54 (0.48, 0.60)	0.51 (0.44, 0.57)	16.00 (6.73)	BAI	0.69 (0.65, 0.73)	0.63 (0.58, 0.68)	22.74 (13.34)
DIS	0.36 (0.27, 0.44)	0.35 (0.26, 0.43)	15.11 (6.50)	BDI-II	0.60 (0.55, 0.65)	0.55 (0.49, 0.60)	23.56 (12.50)

Note. M (SD) = Mean (Standard Deviation). All correlations significant at p < 0.001. Upper and lower 95% bias-correct confidence intervals presented in parentheses next to correlations. Correlations subjected to 5000 bootstrapped samples. ASI-3: Anxiety Sensitivity Index-3 (Taylor et al., 2007); SSASI: Short-Scale Anxiety Sensitivity Index; PHQ-15: Patient Health Questionnaire-15 (Kroenke et al., 2002); WI: Whiteley Index (Pilowsky, 1967); DIS: Discomfort Intolerance Scale (Schmidt et al., 2006a); BAI: Beck Anxiety Inventory (Beck et al., 1988); BDI-II: Beck Depression Inventory-II (Beck et al., 1996)

^aN and percentage of participant who met criteria for psychopathology (0 = absent; 1 = present).

2.1.5. Brief Discussion

Findings provided evidence for a five-item short-scale ASI-3 measure comprised of items 6, 8, 12, 14, and 18. The short-scale, entitled SSASI, evinced excellent model fit as well as acceptable internal consistency. The SSASI also demonstrated strong associations with the ASI-3 total and subscale scores and measures of concurrent validity that tapped into psychological and physical health. To provide further evidence for the five-item SSASI measure, address questions regarding measurement invariance, and more comprehensively evaluate the construct validity of the SSASI, Study 2 was conducted.

2.2. Study 2

2.2.1. Participants and Procedures

The sample consisted of 846 (55.9% female; M_age = 34.46 years; SD = 15.72) outpatient adults seeking mental health services at a university-affiliated clinic. The clinic provides services to the greater community and only refers out individuals suffering from psychotic or bipolar-spectrum disorders and are not stabilized on medication, or if they are an immediate danger to themselves or others. Primary diagnoses for the sample included: anxiety disorders (42.8%), mood disorders (18.0%), and trauma and stressor related disorders (11.7%); 12% did not meet criteria for a mental illness. The racial/ethnic distribution of this sample was as follows: 64.6% White, 23.3% Black, 1.8% Asian, 0.6% American Indian/Native American, 0.4% Pacific Islander, and 9.3% Other. Regarding education, 2.5% did not complete high school, 13.0% high school degree, 2.5% trade or technical school degree, 52.6% some college, 18.8% four-year college degree, and 10.5% graduate school degree.

Individuals presenting for treatment at a university-affiliated clinic provided responses to measures for the current study. All measures were collected at a baseline screening appointment prior to receiving psychological services. Written consent to participate in clinic research was obtained for all participants, and all policies and procedures were approved by the university’s Institutional Review Board.

2.2.2. Measures

Structured Clinical Interview-Non-Patient Version for DSM-IV (SCID-I/NP).

Diagnostic assessments of past year psychopathology was conducted using the SCID-I/NP (First et al., 1994). All SCID-I/NP interviews were administered by trained research assistants or doctoral level staff and supervised by independent doctoral-level professionals. Data from the SCID-I/NP were used to describe psychopathology among the sample.

Short-Scale Anxiety Sensitivity Index (SSASI).

As reported in Study 1, the ASI-3 is an 18-item self-report measure used to assess fear of anxiety-related symptoms. The SSASI was derived from summing items 6, 8, 12, 14, and 18. In the current sample, the ASI-3 and SSASI demonstrated excellent and acceptable internal consistency, respectively (ASI-3: α = 0.92; SSASI: α = 0.79).

Beck Anxiety Inventory (BAI; Beck et al., 1988).

The BAI is a 21-item self-report measure developed to assess physiological symptoms of anxiety. Participants are asked to indicate the extent to which each item reflects their experience over the past week using a 4-point Likert scale ranging from 0 to 3 with higher scores indicating elevated physiological anxiety symptomology. The BAI has demonstrated strong psychometric properties, including excellent internal consistency (α = 0.92; Beck et al., 1988). In the current sample, the BAI demonstrated excellent internal consistency (α = 0.93).

Beck Depression Inventory-II (BDI-II; Beck et al., 1996).

The BDI-II is a 21-item self-report measure developed to assess general symptoms of depression. Participants are asked to indicate the extent to which each item reflects their experience over the past two weeks using a 4-point Likert scale. Scores on each item range from 0 to 3, with higher scores indicating elevated depressive symptomology. The BDI-II has demonstrated strong psychometric properties, including high internal consistency (α = 0.85) and good to excellent test-retest reliability (ICC’s = 0.73–0.96; Wang and Gorenstein, 2013). In the current sample, the BDI-II demonstrated excellent internal consistency (α = 0.93).

2.2.3. Data Analytic Strategy

A CFA was conducted to evaluate the SSASI in the present sample. A model-fitting (CFA) approach was used instead of a purely exploratory factor analytic approach because of findings from Study 1. However, model modification indices were used in this study, and so confirmatory factor analysis was used in an exploratory mode for the purpose of model modification. Analyses were conducted similarly as presented in Study 1 and adhered to the same evaluative statistics and criteria, including χ² statistic, RMSEA, and CFI (Awang, 2012; Hu and Bentler, 1999).

Measurement invariance (Meredith, 1993) was assessed across sex. Specifically, configural, metric, and scalar invariance were assessed by constraining parameters and comparing nested models. Configural invariance assesses for similar factor-indicator patterns across sex. Metric invariance assesses for consistency in the strength of the association between items and factors across sex. Scalar invariance assesses for consistency in item means across sex. A non-significant change in χ² provided statistical evidence for invariance across models. Sex differences in the mean SSASI scale scores also was evaluated.

Finally, construct validity of the SSASI was evaluated. Correlations were conducted between the SSASI total score and ASI-3 total score and constructs that significantly relate to the ASI-3, including the presence of psychopathology, BAI, and BDI-II (Beck et al., 1996; Beck et al., 1988; Taylor et al., 2007). Limited missing data were present in the current sample (ASI-3: 1.2%; presence of psychopathology: 0.1%, BAI: 3.0%, BDI-II: 5.2%).

Correlations between the SSASI, ASI-3, and measures to evaluate construct validity were examined. It was hypothesized that both the SSASI and ASI-3 would significantly and positively relate to each other and the presence of psychopathology, BAI, and BDI-II.

2.2.4. Results

The one-factor structure comprised of five items demonstrated mixed fit across indices (X²[5] = 325.57, p < 0.001; RMSEA = 0.28 [90% CI: 0.25, 0.30], p < 0.001; CFI = 0.908). Inspection of standardized factor loadings indicated strong, significant factor loadings (standardized factor loadings range: 0.53–0.84) across all items; see Table 2. Modification indices suggested allowing a residual covariance between items 8 and 12. Considering that both items loaded onto the physical subscale of the ASI-3 and maintain strong factor loadings on the general and physical subfactor when evaluated in bifactor models (Allan et al., 2015; Ebesutani et al., 2014; Osman et al., 2010), their residual covariance was included. After adding a residual covariance between items 8 and 12, the model demonstrated excellent fit (X²[4] = 12.10, p = 0.02; RMSEA = 0.05 [90% CI: 0.02, 0.08], p = 0.46; CFI = 0.998). Standardized factor loading remained strong and significant after adding the residual covariance (see Table 2).

Table 2.

Standardized factor loadings.

Item	Study 1	Study 2
	Factor Loadings	Factor Loadingsa	Factor Loadingsb
6. When I tremble in the presence of others, I fear what people  might think of me.	0.66	0.53	0.58
8. When I feel pain in my chest, I worry that I‟m going to have a  heart attack.	0.71	0.79	0.49
12. When I notice my heart skipping a beat, I worry that there is  something seriously wrong with me.	0.78	0.84	0.59
14. When my thoughts seem to speed up, I worry that I might be  going crazy.	0.85	0.74	0.79
18. When my mind goes blank, I worry there is something  terribly wrong with me.	0.87	0.77	0.87

Note. N for Study 1 = 536; N for Study 2 = 846. Standardized factor loadings presented. All factor loading significant at p < 0.001.

^aModel contains no residual covariances.

^bModel contains a residual covariance between items 8 and 12.

Next, the measurement model for SSASI was evaluated for measurement invariance across sex. The residual covariance between items 8 and 12 was constrained to be equal across men and women. The five-item structure demonstrated configural invariance across sex (X²[9] = 19.00, p = 0.03; RMSEA = 0.05 [90% CI: 0.02, 0.08], p = 0.43; CFI = 0.997). Full metric invariance was observed. Partial scalar invariance was achieved after allowing intercepts 1–4 for item 8 and intercepts 1–3 for item 12 to vary across sex (see Table 4). Furthermore, men (M = 8.01, SD = 5.13) and women (M = 8.25, SD = 5.13) did not differ on their SSASI scores (t (799.14) = −0.68, p = 0.50).

Table 4.

Measurement invariance of the five-item single factor SSASI measure across sex.

Models	χ2	df	CFI	RMSEA	Δχ2	Δdf	p-value for Δχ2
1. Configural	19.00	9	0.997	0.05	--	--	--
2. Metric	19.95	13	0.998	0.04	4.09	4	0.39
3. Scalar	95.94	32	0.982	0.07	74.72	19	<0.001
3a. Partial Scalar	67.83	28	0.989	0.06	48.16	15	<0.001
3b. Partial Scalar	39.62	25	0.996	0.04	20.20	12	0.07

Note. N = 846. χ² = Chi-square test; df = degrees of freedom; CFI = comparative fit index; RMSEA = root mean square error of approximation; Δχ2 = change chi-square from previous model; Δdf = change in degrees of freedom from previous model. Model 3a allowed item 8 intercepts 1–4 to vary. Model 3b allowed items 8 intercepts 1–4 to vary and 12 intercepts 1–3 to vary.

Correlations across the SSASI, ASI-3, the presence of psychopathology, BAI, and BDI-II were evaluated. Table 3 presents correlations and associated confidence intervals. As hypothesized, SSASI and ASI-3 significantly and positively related to the presence of psychopathology, BAI, and BDI-II, as well as with one another. The magnitude of the observed correlations were comparable across the SSASI and ASI-3 (see Table 3).

2.2.5. Brief Discussion

Data supported a slightly modified version of the SSASI measure derived in Study 1. Specifically, the five-item SSASI measure demonstrated acceptable model fit allowing a residual covariance between items 8 and 12. Empirical and theoretical consideration was given to the inclusion of this covariance and ultimately was found to support it. Despite the need for the additional covariance, the present study demonstrated strong clinical utility for the SSASI. Indeed, the SSASI demonstrated partial invariance across sex, suggesting mean SSASI scores can be compared across men and women, equivalence in SSASI scores across men and women, as well as strong concurrent validity with measure of anxiety and depression. An important ‘next step’ to validate the SSASI is to examine the measure in a non-clinical sample, evaluate invariance over time, and explore additional indicators of construct validity, including discriminant and predictive validity.

2.3. Study 3

2.3.1. Participants and Procedures

A sample of 578 (48.3% female; M_age = 36.9; SD = 13.5) treatment-seeking adult smokers who responded to study advertisements (e.g., flyers, newspaper ads, radio announcements) were included in the current study. Exclusion criteria included suicidality and psychosis. The racial/ethnic distribution of this sample was as follows: 82.9% White/Caucasian; 9.9% Black/Non-Hispanic; 0.9% Black/Hispanic; 2.6% Hispanic; 1.0% Asian; and 2.8% „Other.‟ At least one current (past year) Axis I diagnosis was endorsed by 45.8% of the sample, most commonly social anxiety disorder (9.9%), alcohol or substance use disorder (7.6%), generalized anxiety disorder (5.4%), current major depressive episode (4.7%), and posttraumatic stress disorder (2.9%). On average, participants reported smoking 16.7 cigarettes per day (SD = 10.0) and being a daily smoker for 18.6 years (SD = 13.4). A moderate level of tobacco dependence was observed within the sample based on the FTCD (Fagerström Test for Cigarette Dependence; M = 5.2, SD = 2.3; Heatherton et al., 1991).

Data for the present study was collected during a large, multi-site randomized controlled clinical trial examining the efficacy of two smoking cessation interventions described in detail elsewhere (Schmidt et al., 2016). Interested persons responding to community-based advertisements (e.g., flyers, newspaper ads, radio announcements) contacted the research team and were provided with a detailed description of the study via phone. Participants were then screened for initial eligibility, and if eligible, scheduled for an appointment. Inclusion criteria included: (a) being between ages 18–65, (b) reporting smoking 8 or more cigarettes per day, and reporting motivation to quit rated as at least 5 or higher on a 10-point scale. Exclusion criteria included: (a) current suicidal ideation requiring immediate intervention or (b) active psychosis. After providing written informed consent, participants were interviewed using the SCID-I/NP and completed a computerized self-report assessment battery as well as biochemical verification of smoking status. After completing the baseline assessment, participants were informed of their eligibility. For the current study, baseline data was included from all participants regardless of eligibility for the larger trial. Participants eligible for the larger trial were randomized to one of two treatment conditions. Each treatment was delivered across four intervention sessions with session four being quit day. The current study is based on secondary analyses of baseline and quit day data from the larger trial. All study procedures were approved by the Institutional Review Boards.

2.3.2. Measures

Smoking History Questionnaire (SHQ; Brown et al., 2002).

The SHQ is used to assess smoking rate, years of daily smoking, and other characteristics (Brown et al., 2002). Smoking rate was obtained from the question, “Since you started regular daily smoking, what is the average number of cigarettes you smoked per day?” Furthermore, years as a daily smoker was assessed by the question, “For how many years, altogether, have you been a regular daily smoker?”.

Fagerström Test for Cigarette Dependence (FTCD).

The FTCD (Heatherton et al., 1991) is a six-item assessment of an individual smoker‟s tobacco dependence. Total scale scores range from 0 to 10, with higher scores reflecting high levels of physiological tobacco dependence on nicotine. In the current study, the FTCD total score was used to describe the smoking severity of the sample and included as a covariate (α = 0.64). The alpha for this measure, although low, is comparable to other studies (see Korte et al., 2013 for discussion).

Structured Clinical Interview-Non-Patient Version for DSM-IV (SCID-I/NP).

As with Study 2, the diagnostic assessments of past year Axis I psychopathology was conducted using the SCID-I/NP (First et al., 1994). Please see description above for more details. Data from the SCID-I/NP were used to characterize psychopathology among the present sample.

Short-Scale Anxiety Sensitivity Index (SSASI).

The SSASI was derived from summing items 6, 8, 12, 14, and 18 of the ASI-3. In the current sample, the SSASI was computed from the baseline ASI-3 and the quit week ASI-3. The SSASI demonstrated acceptable to good internal consistency at baseline and quit week (baseline: α = 0.80; quit week: α = 0.76).

Positive and Negative Affect Schedule (PANAS; Watson et al., 1988).

The PANAS measured the extent to which participants experienced 20 different feelings and emotions on a scale ranging from 1 (Very slightly or not at all) to 5 (Extremely). The measure yields two factors, negative affect (NA) and positive affect (PA), and has strong documented psychometric properties (Watson et al., 1988). Both factors were utilized in the current study (NA α = 0.92; PA α = 0.90).

Inventory of Depression and Anxiety Symptoms (IDAS; Watson et al., 2007).

The IDAS is a 64-item self-report instrument that assesses distinct affect symptom dimensions within the past two weeks. Items are answered on a 5-point Likert scale ranging from “not at all” to “extremely.” The IDAS subscales show strong internal consistency, convergent and discriminant validity with psychiatric diagnoses and self-report measures; and short-term retest reliability (r = 0.79) with both community and psychiatric patient samples (Buckner et al., 2015; Capron et al., 2014; Leventhal et al., 2011; Watson et al., 2007). In the present study, we employed the well-being subscale (8 items; α = 0.92), dysphoria (9 items; α = 0.91), panic (8 items; α = 0.88), and social anxiety (5 items; α = 0.87).

Sheehan’s Disability Scale (SDS; Sheehan, 2008).

The SDS is a 3-item self-report measure used to assess impaired functioning in key psychosocial domains of living due to illness. The SDS assessed work/school work, social life/leisure activities, and family life/home responsibilities. Items are rated on a 10-point Likert-type scale ranging from 0 (not at all) to 10 (extremely).

2.3.3. Data Analytic Strategy

A CFA was conducted to evaluate the SSASI in the present sample. A model-fitting (CFA) approach was used. Analyses were conducted similarly as presented in Study 2 and adhered to the same evaluative statistics and criteria presented in Study 1, including χ2 statistic, RMSEA, and CFI (Awang, 2012; Hu and Bentler, 1999).

Measurement invariance (Meredith, 1993) was assessed from baseline to quit week. Similar to the test of measurement invariance across sex conducted in Study 2, configural, metric, and scalar invariance were assessed by constraining parameters and comparing nested models. A non-significant change in χ² provided statistical evidence for invariance across models. Following tests of measurement invariance over time, test-retest reliability was evaluated.

A series of zero-order correlations between baseline variables were conducted to demonstrate convergent and discriminant validity. Variables to evaluate convergent and discriminant validity included PANAS-NA, IDAS-dysphoria, IDAS-panic, IDAS-social anxiety, PANAS-PA, and IDAS-well-being. It was hypothesized that the SSASI would positively relate to PANAS-NA, IDAS-dysphoria, IDAS-panic, IDAS-social anxiety, and negatively relate to PANAS-PA, and IDAS-well-being. Subsequently, the strength of the associations were compared using the Steiger method (Steiger, 1980; Zou, 2007); this approach appropriately considers the non-independent nature of the compared correlations. It was hypothesized that the SSASI would demonstrate significantly stronger correlations with measures of convergent validity (i.e., PANAS-NA, IDAS-dysphoria, IDAS-panic, IDAS-social anxiety, PANAS-PA, and IDAS-well-being) than measures of discriminant validity (i.e., PANAS-PA, and IDAS-well-being).

Lastly, three separate hierarchical regression models were conducted to evaluate the unique predictive validity of the SSASI on functional impairment after controlling for anxiety measures (i.e., IDAS-panic and IDAS-social anxiety). IDAS-panic and IDAS-social anxiety were entered on the first step and SSASI was entered on the second step. Functional impairment across (a) work/school work, (b) social life/leisure activities, and (c) family life/home responsibilities were the three dependent variables. Only participants who provided data for the SDS were included in these analyses (N = 485). Unstandardized regression coefficients are presented.

2.3.4. Results

Consistent with Study 2, the one-factor model demonstrated mixed fit across indices (X²[5] = 137.69, p < 0.001; RMSEA = 0.21 [90% CI: 0.18, 0.25], p < 0.001; CFI = 0.94). Subsequently, the modified model with a residual covariance between items 8 and 12 was evaluated. The one-factor structure comprised of five items, and a covariance between items 8 and 12, demonstrated excellent fit (X²[4] = 4.16, p = 0.38; RMSEA = 0.01 [90% CI: 0.00, 0.06], p = 0.86; CFI = 1.000). Next, the measurement model for SSASI was evaluated for measurement invariance across time (i.e., baseline to quit week). Because tests of measurement invariance over time require error terms of similar items to covary over time, the covariance between items 8 and 12 was constrained to be equal at baseline and quit week. The five-item structure demonstrated configural, metric, and scalar invariance from baseline to quit week (see Table 5). Test-retest reliability of the SSASI from baseline to quit week was strong (r = 0.63, p < 0.001).

Table 5.

Measurement invariance of the five-item single factor SSASI measure across time.

Models	χ2	df	CFI	RMSEA	Δχ2	Δdf	p-value for Δχ2
1. Configural	105.91	28	0.978	0.07	--	--	--
2. Metric	102.49	32	0.981	0.06	5.43	4	0.25
3. Scalar	121.87	49	0.980	0.05	21.93	17	0.19

Note. N = 846. χ² = Chi-square test; df = degrees of freedom; CFI = comparative fit index; RMSEA = root mean square error of approximation; Δχ2 = change chi-square from previous model; Δdf = change in degrees of freedom from previous model.

As hypothesized, the SSASI significantly and positively correlated with PANAS-NA (r = 0.59, p < 0.001), IDAS-dysphoria (r = 0.55, p < 0.001), IDAS-panic (r = 0.54, p < 0.001), and IDAS-social anxiety (r = 0.53, p < 0.001). The SSASI also significantly, but negatively, correlated with PANAS-PA (r = −0.23, p < 0.001) and IDAS-well-being (r = −0.23, p < 0.001). The strength of the correlations between the SSASI and measures of convergent validity were significantly stronger than the strength of the associations between the SSASI and measures of discriminant validity (all p‟s < 0.001).

IDAS-panic and IDAS-social anxiety accounted for significant variance in functional impairment at work/school work (F[2, 482] = 42.14, p < 0.001, R² = 0.15). SSASI accounted for additional variance in the criterion than accounted for by IDAS-panic and IDAS-social anxiety (ΔR² = 0.03, p < 0.001) and was positively associated with work/school work impairment (b = 0.13, SE = 0.03, t = 4.18, p < 0.001). Similar patterns emerged across impairment with social life/leisure activities (Step 1: F[2, 482] = 74.94, p < 0.001, R² = 0.24; Step 2: ΔR² = 0.02, p < 0.001) and family life/home responsibilities (Step 1: F[2, 482] = 45.02, p < 0.001, R² = 0.15; Step 2: ΔR² = 0.04, p < 0.001), such that SSASI accounted for addition variance in these variables and was positive associated with each (social life/leisure activities: b = 0.13, SE = 0.03, t = 3.89, p < 0.001; family life/home responsibilities: b = 0.15, SE = 0.03, t = 4.65, p < 0.001).

2.2.5. Brief Discussion

The current study replicated finding from Study 2. Specifically, the five-item SSASI measure demonstrated acceptable model fit after allowing a residual covariance between items 8 and 12. The SSASI demonstrated full measurement invariance over time, providing empirical evidence to compare SSASI means over time. Consistent with Studies 1 and 2, the SSASI strongly correlated with measures selected to evaluate concurrent validity. Importantly, the SSASI also demonstrated discriminant validity. Indeed, the pattern of associations across indicators of concurrent and discriminant validity further bolster the construct validity argument for the SSASI. Finally, evidence emerged for predictive validity. As such, the SSASI may serve as a community friendly instrument that can be used to identify individuals more likely to experience functional impairment.

3. General Discussion

Despite the theoretical and clinical importance of the anxiety sensitivity construct in contemporary models of psychopathology and health (Otto et al., 2016), there is no version of the scale that is short enough to be „practice friendly‟ and therefore adopted for use in medical and other community environments. Thus, the aims of the present investigation were to refine the measurement of anxiety sensitivity by developing a brief measure of anxiety sensitivity, to examine its replicability, and to explore its construct validity across three independent samples. The results favored an abbreviated five-item version of the ASI-3 (referred to as the SSASI). The SSASI demonstrated good internal consistency and a robust association with the ASI-3 total score. Further, across the samples, there was evidence of unidimensionality and excellent convergent and discriminant validity with established measures of anxiety, depression, discomfort intolerance, health complaints, health anxiety, affective states, and overall well-being. Additionally, the SSASI demonstrated predictive validity across functional impairment domains after controlling for anxiety symptoms. Data also provided evidence for configural, metric, and scalar invariance across sex (partial invariance) and time (full invariance) as well as evidence for equivalent SSASI score across men and women, which is consistent with extant work on the ASI-3 (Nillni, Berenz, Rohan, & Zvolensky, 2012). Overall, this five-item scale offers a single score that can be employed to measure AS. Indeed, because the full version of this measure correlated at comparable magnitudes with other measures of psychopathology and health symptoms/problems, the present results suggest that the brief version may be a more refined and promising alternative for use in real world contexts wherein there is frequently, a high-degree of time constraints.

As discussed, longer measures are less likely to be integrated into clinical settings (Mark et al., 2008), and thus researchers are now developing short-form versions to facilitate the more routine use of evidenced-based assessment in such contexts (e.g., Kroenke et al., 2003). The clinical relevance of the anxiety sensitivity construct and its transdiagnostic importance, indicates potential value in the routine assessment of anxiety sensitivity. The ability to more efficiently assess anxiety sensitivity using the SSASI, relative to longer measures of this construct, underscores the potential utility of the SSASI for use in a broad array of clinical settings. For example, use of the SSASI has the potential to help identify individuals at-risk for emotional disorders, addictive behavior, and physical illness and disease, as well as to track changes in a potential mechanism (i.e., anxiety sensitivity) for positive treatment gains of intervention programs.

Several study limitations should be considered. First, the current study did not test the SSASI across ethnoracial factors. This limitation was due, in part, to an artifact of the sample size in Study 1 and limited diversity in Study 2 and 3. Thus, future research would benefit from evaluating the validity of the SSASI across more ethnically/racially diverse samples, including tests of invariance. Such work would bolster the applicability and support the adoption of the SSASI in medical centers and related community portals that provide care to diverse populations. Second, the five-items retained to comprise the SSASI were not evaluated independently of the full ASI-3. Thus, additional work is needed to validate the proposed measure when delivered independently of the discarded ASI-3 items. Third, although extant work suggests that the selected physical items only share 35% of their variance (Taylor et al., 2007), across Studies 2 and 3, adequate fit for the SSASI was only demonstrated after adding a residual covariance between these items. Conceptually the inclusion of a residual covariance suggests that these items share more variance than is explained by the SSASI, which may, in part, be a byproduct of the original ASI-3 physical subfactor (Taylor et al., 2007). Indeed, this hypothesis is substantiated by the replicated need for a residual covariance across both Studies 2 and 3. Although evidence for redundancy (i.e., correlation greater than 0.80; Berry & Feldman, 1985) was not evident for these items across samples (sample 1: r = 0.43; sample 2: r = 0.70; sample 3: r = 0.67), the need for a residual covariance may, in part, have been the consequence of item wording and their reference to heart related problems. Thus, additional work may be warranted in this domain. Fourth, by design, we utilized independent clinical samples, as these are the populations to which we seek to generalize. Based upon the current findings, future research may benefit by replicating and extending the current results to other samples, including those from the general community. Such work will be particularly useful when probabilistic sampling is employed among a large sample so that normative data can be developed, and clinical cut-points explicated. Finally, stability of the SSASI over time was limited to a sample of treatment-seeking adults. Therefore, the current findings warrant future replication in an independent sample of adults.

Overall, the present study results provide empirical support for the usefulness of a short-scale measure of AS, entitled the SSASI, and demonstrated strong associations with a wide array of clinical measures of anxiety, depression, health symptoms, and functional impairment. These findings indicate that the SSASI is a reliable and valid brief measure of anxiety sensitivity that can be used in community contexts with increased efficiency.

HIGHLIGHTS.

• This study aimed to assess validity and reliability of a short version of the ASI-3

• The Short Scale Anxiety Sensitivity Index (SSASI) had good internal consistency.

• There was evidence of excellent convergent and discriminant validity.

• There was also evidence of full measurement invariance across time.