Skip to main content
  • Study Protocol
  • Open access
  • Published:

CO2 reactivity as a biomarker of exposure-based therapy non-response: study protocol



Exposure-based therapy is an effective first-line treatment for anxiety-, obsessive–compulsive, and trauma- and stressor-related disorders; however, many patients do not improve, resulting in prolonged suffering and poorly used resources. Basic research on fear extinction may inform the development of a biomarker for the selection of exposure-based therapy. Growing evidence links orexin system activity to deficits in fear extinction and we have demonstrated that reactivity to an inhaled carbon dioxide (CO2) challenge—a safe, affordable, and easy-to-implement procedure—can serve as a proxy for orexin system activity and predicts fear extinction deficits in rodents. Building upon this basic research, the goal for the proposed study is to validate CO2 reactivity as a biomarker of exposure-based therapy non-response.


We will assess CO2 reactivity in 600 adults meeting criteria for one or more fear- or anxiety-related disorders prior to providing open exposure-based therapy. By incorporating CO2 reactivity into a multivariate model predicting treatment non-response that also includes reactivity to hyperventilation as well as a number of related predictor variables, we will establish the mechanistic specificity and the additive predictive utility of the potential CO2 reactivity biomarker. By developing models independently within two study sites (University of Texas at Austin and Boston University) and predicting the other site’s data, we will validate that the results are likely to generalize to future clinical samples.


Representing a necessary stage in translating basic research, this investigation addresses an important public health issue by testing an accessible clinical assessment strategy that may lead to a more effective treatment selection (personalized medicine) for patients with anxiety- and fear-related disorders, and enhanced understanding of the mechanisms governing exposure-based therapy.

Trial registration Identifier: NCT05467683 (20/07/2022).

Peer Review reports


Anxiety-, obsessive–compulsive and trauma- and stressor-related disorders are prevalent and costly mental health disorders [1,2,3,4]. Exposure-based therapy (EBT) has demonstrated efficacy [5] and is a recommended first-line treatment for these disorders [6,7,8,9,10]. As the utilization of EBT has strengthened [11], it becomes more important to identify biomarkers of EBT non-response [12]—an outcome observed in a significant subset of patients [13,14,15,16,17]. Indeed, making available easy-to-implement screening tools to aid clinicians and patients in deciding whether to initiate EBT can reduce “unnecessary” prolonged suffering and treatment burden/costs, as well as increase the availability of EBT therapists for patients most likely to respond to this treatment.

Although there have been efforts to identify predictors of EBT non-response, these have been largely post-hoc or based on the secondary analyses of efficacy studies [18,19,20,21,22,23,24,25,26,27,28,29,30], and consequently they have been handicapped by low statistical power and replication failure. As we detail next, a fruitful approach to identifying “clinically meaningful” biomarkers of EBT non-response is to (1) focus on variables that are (theoretically) related to core mechanisms of action of EBT, especially when such variables can be readily assessed in clinical practice; and (2) employ a methodological approach that facilitates the goal of reproducibility.

Theory-informed biomarkers

EBT was derived from models of extinction learning [31]. The focus on developing safety memories/fear extinction is at the heart of EBT for anxiety-, obsessive–compulsive and trauma- and stressor-related disorders [31, 32]. Through guided experiences involving the presentation of feared cues, clinicians help patients undo the maladaptive anxiogenic beliefs, vigilance to harm, and avoidance that characterize these disorders. Importantly, consistent with the hypothesis that fear extinction is a core mechanism of EBT [31, 33, 34], a growing body of literature links subjective and neural indices of fear extinction to EBT outcomes [35,36,37,38,39,40]. Hence, identifying variables that predict fear extinction can aid clinical decision-making with respect to the selection of EBT [41].

Orexins are neuropeptides produced by neurons localized exclusively in the hypothalamus [42, 43]. Orexin-expressing neurons have extensive projections within the CNS and, as such, orexins play a role in several important functions, including sleep, feeding, and anxiety [44,45,46]. Experimental evidence suggests an important role of the orexin system in fear extinction [47,48,49,50,51]. Indeed, greater activation of orexin neurons in the medial hypothalamus is associated with poor fear extinction [52], blocking the (orexin/)hypocretin-1 (Hcrtr-1) receptor with the antagonist SB334867 facilitates fear extinction [48, 50], and administering Hcrtr-1 impairs fear extinction [48]. Furthermore, antagonism of orexin receptors increases the recruitment of basolateral amygdala (BLA) neurons that project to the infralimbic cortex (IL) during extinction [50]. Those very same neurons (the IL-projecting BLA neurons) are found to be active during fear extinction [53], underscoring the observation that orexin activation in the lateral hypothalamus (LH) accounts for individual differences in fear extinction [47]. Translation of these rodent findings to the clinic is aided by a study involving patients with panic disorder that has linked the Hcrtr-1 rs2271933 genotype to poor response to EBT, implicating orexin activity as a marker of EBT non-response [54].

The relation between orexin cell activity and the aforementioned processes has been well established, particularly in non-human animals, where the cell activity can be directly quantified [49]. In humans, this relation has been extended to patients in which microdialysis could be performed [55]; however, most studies conducted in humans have quantified orexin either in plasma, saliva, or cerebrospinal fluid (CSF). Whereas orexin quantified from CSF appears to correlate with orexin sampled from the LH, orexin quantified from plasma or saliva does not correlate with orexin in the CSF, suggesting that modulations occurring in line with the findings established in rodents are specific to the central nervous system (CNS). Fluctuations in plasma orexin A are also not differentially modulated by orexin receptor gene polymorphisms [56]. Furthermore, orexin A is present in the blood in low amounts, and its levels do not follow autonomic or neuroendocrine circadian rhythms [57], suggesting that evaluations of orexin quantifications outside of CNS should be interpreted very cautiously.

Given these limitations, carbon dioxide (CO2) reactivity emerges as perhaps the best non-intrusive index of orexin activity. Indeed, prepro-orexin knockout mice evidence reduced respiratory reactivity to hypercarbic gas exposure (CO2 challenge) and pre-CO2 challenge injection of the Hcrtr-1 receptor antagonist SB334867 reduces respiratory reactivity in wild-type mice [58]. Similarly, blocking the Hcrtr-1 receptor with the administration of Hcrtr-1 receptor antagonists (e.g., SB334867, JNJ-61393215) reduces fear and anxious responding to CO2 challenge both in rats and in humans [59, 60]. Collectively, these findings show that the orexin system (and particularly the Hcrtr-1 receptor) mediates hypercapnia-induced fear and sympathetic drive and, by extension, suggest that CO2 reactivity can serve as a proxy to index orexin activity [51, 59, 60].

The CO2 challenge is a safe and well-tolerated procedure that involves the inhalation of CO2-enriched air at various concentrations (e.g., 5%, 7%, 15%, 20%, or 35%)[61]. Important for studies translating non-human animal work is the observation that CO2 challenge fear reactivity (i.e., behavior in rats, subjective ratings and avoidance in humans) as well as respiratory and cardiovascular reactivity are comparable across species [62]. Stemming from physiological systems that are distinguishable from those subserving general trait anxiety and lie on a continuum with the extreme being panic [63], the CO2-driven subjective and behavioral responses in humans are dose-dependent [64] and elevated relative to healthy control subjects among individuals with panic disorder (PD) [65,66,67], social anxiety disorder (SAD) [68, 69], generalized anxiety disorder (GAD) [70, 71], obsessive–compulsive disorder (OCD) [72], and posttraumatic stress disorder (PTSD) [73, 74], while also demonstrating adequate intra-individual variability within each of these diagnostic categories.

Optimizing the methodological approach to validating putative biomarkers

We conceptualize reactivity to hyperventilation as a control biomarker because voluntary hyperventilation (VH): (1) reliably induces affective reactivity in those with anxiety and related disorders [75,76,77,78]; (2) shares method variance with the CO2 challenge (i.e., uses the respiratory system and an identical strategy for indexing subjective reactivity); (3) reactivity is predicted (as is CO2 reactivity) by psychological variables relevant to fears of somatic sensations [76,77,78]; and (4) has the opposite effect to CO2 challenge on pH and thus orexin cell firing in the LH: acidification (i.e., decrease in pH resulting from an increase in CO2) increases intrinsic excitability, whereas alkalinization (i.e., increase in pH resulting from a decrease in CO2) depresses it [79,80,81], and thus is not considered a marker of orexin activity. Hence, VH allows for disentanglement of reactivity due to fears of respiratory somatic sensations and to reactivity more directly reflecting orexin activity (via CO2 challenge).

We utilize converging self-report, behavioral, and physiologic methods to index CO2 reactivity, including: (1) self-reported peak anxiety reactivity and habituation/sensitization in anxiety, (2) behavioral latency between inhalations (avoidance/escape), and (3) physiologic reactivity during recovery (end-tidal CO2). All three indices are available from a single-session CO2 challenge; hence, we retain the efficiency of our procedures while providing a comprehensive set of subjective and objective indices.

CO2 reactivity can also be influenced by a number of psychological factors. We have selected control variables relevant to CO2 reactivity that are not related to orexin activity, including: (1) anxiety sensitivity (AS; the tendency to perceive that anxiety-related symptoms and sensations have catastrophic consequences [82]), which operates as a transdiagnostic risk factor for the maintenance of anxiety pathology [83] and a reliable predictor of reactivity to interoceptive challenges, including CO2 [84, 85] and VH [78, 86]; (2) distress intolerance (DI); notably, although AS can be conceptualized as a measure of distress intolerance (DI), there is variability in prediction and reactivity with alternative measures of DI [87,88,89]); (3) experiential avoidance (EA) [90]; (4) Intolerance of uncertainty (IU) [91, 92], a putative transdiagnostic risk factor for the maintenance of anxiety and related disorders and EBT non-response [93,94,95,96,97,98,99]; and finally, (5) a behavioral index of subtle avoidance behaviors that may interfere with extinction [100]. In sum, AS, DI, EA, IU, and related avoidance tendencies all share variance with CO2 and VH reactivity, but like VH, should be a poor marker of orexin activity. Finally, to demonstrate that the CO2 derived biomarker offers incremental predictive utility, we will also include standard, albeit unreliable, clinical and demographic predictors of EBT non-response [18,19,20,21,22,23,24,25,26,27,28,29,30].

Adopting a stage model approach for biomarker research

The identification of biomarkers for psychosocial treatment response has been marked by difficulties in replication. For this reason, we believe that establishing reliability for a biomarker is more important than establishing the treatment specificity of that marker in the first stage of biomarker validation. This is analogous to the stage approach to treatment development; we are starting with a Stage 1 (open) evaluation of biomarker adequacy. Accordingly, our design maximizes reliability by including a replication sample in the design to ensure consistency of prediction for the putative biomarker. Consistent with a Stage 1 approach [101], our design does not include a test to ensure that the non-response predicted by the biomarker is specific to EBT vs. some alternative treatment. Indeed, the selection of which treatment should serve as an alternative treatment is a premature decision for two reasons: (1) we have not yet determined the level at which CO2 reactivity becomes predictive of non-response (hence, it is unclear what cut-off can be used to stratify based on biomarker status [+ or -]); and (2) it is unclear at this time what intervention can be used effectively either because it does not rely on orexin function or because it can manipulate the orexin system.

Sample size and data analytic approach

Simulation work has shown that large sample sizes are required to develop accurate predictive models of treatment response [102, 103]. Our own statistical simulation shows that given the two-site, split-half validation design, we require 300 participants per site—at least 300 to develop the model and another 300 to adequately validate it in this Stage 1 model. The reality is that underpowered studies likely detect only the largest effects, and multivariate models based on them yield unstable predictions that overfit the sample data and perform poorly out-of-sample [104, 105]. As pointed out by Kessler and colleagues, “The inevitable conclusion is that samples much larger than those in existing mental disorder randomized clinical trials are required to develop useful personalized treatment recommendations” [106]. The overarching message of the statistical learning literature and our own simulations used to develop this proposal is “go big or go home”.

Our primary objective is to build a model predicting EBT non-response using a multivariate combination of candidate predictors. One set of predictors constitutes what we consider our “control” model; these include “traditional” prognostic factors and psychological variables (see Table 1). The addition of a second set of predictors, corresponding to our proposed measures of reactivity to CO2 and VH, constitutes our “experimental” model. We have two key hypotheses that we aim to test: (1) the experimental model will outperform the control model when predicting never-before-seen data, establishing the additive predictive value of assessing reactivity measures prior to initiating treatment; and (2) when predicting never-before-seen data, a model that excludes VH reactivity measures will outperform a model that excludes CO2 reactivity measures, establishing that the additive predictive value is specific to CO2 reactivity.

Table 1 Schedule of assessments

Importantly, cross-validation is used during model development to prevent overfitting, and all statistical tests of model predictions will be performed using data collected from a completely independent site located in a different geographic region of the country. This ensures that our findings will not just be “statistically significant” in the sense of being improbable under a null hypothesis, but will demonstrate real-world predictive utility in two respects: (1) the biomarkers must out-perform other measures and (2) they must predict new data, showing that predictions based on one clinical sample will generalize to another. We have powered our proposed study to meet this higher standard. If we are successful, it is our further objective to translate our model predictions into a set of decision rules to facilitate their implementation in clinical practice.



We aim to recruit a transdiagnostic sample (N = 300 at each of two collaborating sites [The University of Texas at Austin (UT) and Boston University (BU)]) presenting with one or more DSM-5 anxiety disorders, OCD, or PTSD. Eligible participants will complete 20% CO2 and VH challenges, as well as a comprehensive set of clinician-rated and patient-rated measures prior to starting open, transdiagnostic, EBT. Assessment of non-response will occur weekly during treatment, at 1-week posttreatment (i.e., primary endpoint), and at 3-month follow-up. This study has been registered with (NCT05467683; 20/07/2022). The UT Institutional Review Board has approved the study protocol for both participating sites (STUDY00001631).


The transdiagnostic sample will consist of 600 participants. Inclusion criteria include: (1) a primary DSM-5 diagnosis of PD (with or without agoraphobia), SAD, GAD, OCD or PTSD; (2) a score of 8 or greater on the Overall Anxiety Severity and Impairment Scale (OASIS) [107]; (3) ages 18–70; (4) willingness and ability to provide informed consent and comply with the requirements of the study protocol; and (5) proficiency in English (because many assessment instruments have only been validated in English). Exclusion criteria include: (1) a lifetime history of bipolar or psychotic disorders; substance use disorders (other than nicotine) or eating disorder in the past 6 months; serious cognitive impairment; (2) active suicidal ideation with at least some intent to act with or without a specific plan (i.e., a rating of 4 for suicidal ideation on the Columbia-Suicide Severity Rating Scale; CSSRS) [108] or suicidal behaviors (actual attempt, interrupted attempt, aborted or self-interrupted attempt, or preparatory acts or behavior) within the past 6 months; (3) medical conditions contraindicating CO2 inhalation or VH (e.g., cardiac arrhythmia, cardiac failure, asthma, lung fibrosis, stage 2 high blood pressure, epilepsy, or stroke); (4) pregnancy or lactation; (5) ongoing psychotherapy directed toward the primary disorder; or (6) pharmacological treatment started within 8 weeks prior to the screen (participants “stable” on their medication regimen will be included and their medication status will be included as a variable in the model).


Participants will be recruited from our outpatient clinics specializing in the treatment of fear- and anxiety-related disorders, which helps ensure adequate flow for the proposed study. To complement the natural flow, we will advertise through numerous community organizations, social media platforms, and other internet-based referral sources.


We include an incentive-based approach that includes; (1) obtaining multiple methods for contacting participants; (2) offering flexibility in scheduling appointments; (3) personalized connections around scheduling; (4) providing reminders of appointments; and (5) weekly monitoring of recruitment and retention and quality control across sites. Also, assessment adherence and study completion will be aided by study compensation. Compensation is based on one biomarker assessment session, one baseline session, 12 weekly assessment sessions, 1 posttreatment and 1 follow-up assessment at the following levels: $70 for the biomarker assessment session, $30 for the baseline assessment, $10 for each weekly assessment ($120), and $30 for posttreatment and follow-up ($60), plus a reward of $20 for completing all assessments, for a total of $300 per participant.



An internet prescreen will be conducted for all potential participants. Persons who appear eligible will be invited to complete diagnostic screening. Participants will receive an informed consent form explaining the details of the study, potential benefits and risks of participation, and the procedures they will undergo if they choose to participate. If the individual provides informed consent, they will begin the psychiatric evaluation process, which will be conducted during an in-person visit with a clinician.

Biomarker visit

Prior to the first treatment visit, participants will complete two distinct respiratory challenges to assess the putative biomarker and relevant control variables.

CO2 challenge

The CO2 challenge comprises two 20-min trials: the 1st trial involves 3 vital capacity (VC) breaths from a bag of compressed air and the 2nd trial involves 3 VC breaths from a bag containing 20% CO2-enriched air (participants will be blind to the content of the inhalation mixture in the bag). Participants will first view a video recording that provides information about CO2 inhalation and modeling of the CO2 challenge procedures. The integrated system built by Hans Rudolph, Inc. and customized for this study includes a pulse oximeter (to assess heart rate and blood oxygen saturation; exploratory measures), a breathing mask with sensors (to measure tidal volume, respiratory rate, minute ventilation, end-tidal O2, and end-tidal CO2), a dial (for continuous ratings of anxiety) and a button to initiate inhalation. Prior to the first trial, participants will practice VC inhalation (exhaling completely and inhaling to maximum lung capacity). Participants will then be allotted 20 min to complete 3 VC inhalations at their own pace by using the button to initiate each inhalation. They will be asked to rate how much anxiety they feel, moment by moment, using the rating dial. The two trials—involving VC breaths of room air followed by 20% CO2-enriched air—are each preceded by a 2-min baseline period and followed by a 2-min recovery period.

VH challenge

The VH challenge will be administered within 30 min following the CO2 challenge. It comprises one 20-min trial involving 3 two-minute VH provocations (matching the CO2 challenge procedures). Consistent with recommendations for standardization [109], participants will view a video recording that explains hyperventilation and the challenge procedures and models proper VH (i.e., breathing in pace with pacing tones signaling inspiration and expiration to guide the rate of 18–24 breaths per minute) [109] and then complete a 15-s practice trial supervised by staff. Participants will be fitted with the same pulse oximeter (to assess heart rate) and breathing mask which will allow assessment and monitoring of tidal volume, respiratory rate, minute ventilation, end-tidal O2, and end-tidal CO2, and provide the staff with the necessary feedback to potentially modify breathing rate to ensure that the participant stays at the target end-tidal CO2 level (i.e., 20 mmHG) for 2 min [109]. They will rate anxiety using the dial continuously, and will be told that they have 20 min to complete three VH provocations at their own pace with a button press to initiate each trial, mirroring the CO2 challenge procedures.

Open exposure-based therapy

Transdiagnostic EBT will be delivered by experienced, license-eligible clinicians. To aid generalization to EBT delivered in clinical practice, the study clinician will develop a personalized assessment and treatment plan for each participant. Assessment algorithms will (1) guide the case formulation, which emphasizes threat appraisals as maintaining factors to be targeted during treatment; and (2) provide the data for tracking success and progress. The case formulation guides the clinician in the development of personalized exposure exercises, while tracking success and progress allows for updating of the case formulation and fine-tuning of the treatment plan. Consistent with contemporary models of EBT [31, 110], exposure practice aims to help patients relearn a sense of safety around feared cues. Hence, exposure exercises are planned to ensure violation of threat expectancies. In addition to ensuring sufficient activation of the “fear network” and a focus on repetition to provide disconfirmatory evidence, exposure practice will be planned and delivered keeping in mind that fear extinction tends to be context specific. Specifically, practice will occur across relevant contexts both within and outside the session (i.e., homework) and clinicians will guide participants in processing their exposure practice to facilitate consolidation of safety learning.

To achieve these ends, study clinicians will use a manual that describes these procedures for treatment planning and delivery. The manual “Personalized Exposure Therapy: A Person-Centered Transdiagnostic Approach [32]” includes clear guidance on the conceptual model of EBT, assessment planning and strategies, and separate chapters on the planning and delivery of in vivo, imaginal and interoceptive exposure practice, respectively. The treatment dose will be set at 12 one-hour sessions, delivered over the course of 12 weeks. The quality assurance protocol for treatment implementation involves requiring all clinicians to (1) complete a 6-h training workshop; and (2) attend weekly supervision meetings.


Table 1 provides an overview of assessment targets and measures by study phase.


The online questionnaire first asks potential participants to provide standard demographic information and to indicate whether they have experienced or been diagnosed with any of the psychiatric or medical exclusion criteria. Participants who do not endorse exclusion criteria will then be asked to complete the OASIS [107]. Participants who endorse experiencing anxiety-related symptoms and impairment will then complete the DSM-5-TR Self-Rated Level 1 Cross-Cutting Symptom Measure [111]. Based on their responses to this measure, participants will also complete relevant Level 2 or diagnosis-specific measures, which for the current study include the Severity Measure for Panic Disorder [112], Mobility Inventory (MI; alone) [113], Social Phobia Inventory (SPIN) [114], PROMIS Emotional Distress—Anxiety—Short Form [115], Dimensional Obsessive Compulsive Scale [DOCS; OCD] [116], and the PTSD Checklist for Diagnostic and Statistical Manual of Mental Disorders-Fifth Edition (PCL-5) [117] as well as measures of constructs that aid the case formulation (e.g., anxiety/fear cues, core threat appraisals, and safety behaviors). Participants that do not meet any exclusion criteria after this screening will be given the option to schedule an in-person screening visit.

The clinician assigned to the participant will complete the in-person screen visit, which includes administration of the Structured Clinical Interview for DSM-5 (SCID-5) [118] and the Columbia Suicide Severity Rating Scale (C-SSRS) [108].

Biomarker measures

For each VC inhalation, mean and peak reactivity will be assessed across different epochs (i.e., baseline, anticipation, peak response, and recovery). For the subjective anxiety index, difference scores will be calculated between the 1st trial (bag of compressed air) and 2nd trial (bag of CO2). The primary subjective measure of CO2 peak anxiety reactivity will be the participant’s peak real-time ratings of subjective anxiety (reported throughout the trials using the rating dial). The primary measure of the degree of habituation vs. sensitization will be the change in peak subjective anxiety from the 1st to the 3rd of the CO2 provocations. The primary behavioral measure will be avoidance responses (latency in seconds between inhalations 1–2, and 2–3). The primary physiological measure will be the difference in end-tidal CO2 during the resting baseline vs. recovery phase.

For each of the three VH provocations, mean and peak reactivity will also be assessed across epochs (i.e., baseline, anticipation, peak response, and recovery). For the subjective index, difference scores will be calculated between VH baseline (at rest) ratings and the trial (VH) ratings. The primary subjective measure of VH peak anxiety reactivity will be the participant’s peak real-time ratings of subjective anxiety (reported throughout the trials using the rating dial). The primary measure of the degree of habituation vs. sensitization will be the change in peak subjective anxiety from the 1st to the 3rd of the VH provocations. The primary behavioral measure will be avoidance responses (latency in seconds between VH provocations 1–2, and 2–3). The primary physiological index of VH reactivity will be the mean difference in end-tidal CO2 during the resting baseline vs recovery phase.

Theoretically-relevant and general competing predictor variables

Participants will complete the following self-report measures of constructs related to CO2 reactivity and/or expected to be predictive of EBT non-response: ASI-3 [119], Distress Tolerance Index (DTI) [88], Brief Experiential Avoidance Questionnaire (BEAQ) [120], IUS [91], and Safety Behavior Assessment Form (SBAF) [121]. Additional control predictor variables include sex assigned at birth, gender identity, number of diagnoses (s measured by the SCID-5), and comorbidity symptom severity (as measured by the DSM-5-TR Self-Rated Level 1 Cross-Cutting Symptom Measure).

Symptom severity

Prior to each treatment visit (Weeks 1–12), at posttreatment (Week 13) and at follow-up (week 24), an independent evaluator (IE; telehealth) will administer the Clinical Global Impressions (CGI) scales [122]. Participants will also complete the OASIS and the Patient Health Questionnaire (PHQ-9) as well as the symptom severity measure corresponding to their primary diagnosis (Panic Disorder Severity Scale-Self Report Version [PDSS-SR; PD] [123], SPIN [SAD] [114], GAD-7 [GAD] [124], DOCS [OCD] [116], or PCL-5 [PTSD] [117]) prior to the meeting with the IE. Every 3 weeks participants will also be asked to complete several treatment process measures that are not related to the primary study aim. IE training will involve completion of a 3-h workshop and reliable rating (> = 80%) of interviews with test subjects. IE’s will also complete quarterly ratings of test cases to prevent rater drift.

Definition of non-response

Participants will be classified as non-responders if their CGI—Improvement (CGI-I) score is 3 or above OR if their OASIS score has not improved by at least 4 points.

Treatment integrity and acceptance

Participants will complete the Credibility/Expectancy Questionnaire (CEQ) [125] which is a widely used 6-item measure that assesses treatment credibility and expectancy, after the first treatment session. Participant adherence to each intervention will be assessed as the number of total sessions attended.

Data analysis

Rationale for statistical learning approach

Our primary goal of identifying mechanistic non-response indicators that can be readily assessed in clinical practice requires a machine learning approach that has (as its end product) a single model that any clinician can easily understand and adopt. Whereas there is evidence supporting the potential relevance of all our candidate predictors for individual bivariate relationships with non-response, we do not yet know how to optimally combine them into a single model to maximize prediction accuracy. This is especially true for the CO2/VH reactivity measures, which are multimodal, including physiology, behavior, and self-report. It is unlikely that all of these measures will provide a unique predictive signal; it is very likely that some may be redundant, and others may constitute noise. We therefore require a data-driven method of variable selection that reveals the best subset of predictors. A traditional approach would be to use a generalized linear model (GLM) and a stepwise selection algorithm, in which model terms are iteratively added or deleted and the model is repeatedly refit until some information criterion settles into a local optimum. Numerous weaknesses of such stepwise selection have been noted [126] including model selection bias, which can exaggerate the apparent strengths of relationships [127, 128].

The essential problem is that, with or without variable selection, an ordinary logistic regression is very likely to overfit the data [129] (meaning the model is trained to predict sample noise, which leads to inflated estimates of how well the model predicts the sample it was trained on, at the expense of generalizability when predicting other samples). Thus, a statistical learning technique is required that can handle the potential for highly correlated covariates and discourage overfitting. In developing a statistical learning approach for our objectives, we are mindful of the tradeoff between prediction accuracy and model interpretability that exists within the large umbrella of machine learning approaches. Generally speaking, the approaches with the best prediction performance (e.g., stacked ensembles that blend a diversity of machine learners) are the most successful in revealing comprehensible mechanisms. This has led us to select regularized regression, using the elastic net penalty, as our primary approach.

The chief advantage of the elastic net [130] is that it is still a GLM, and therefore its model output is as easy to understand and interpret as any GLM. Importantly, elastic net regression functions just as a GLM would in terms of handling covariates. For instance, if CO2 reactivity is entered into the model alongside other predictors, its estimated regression coefficient will reflect its incremental contribution to the prediction, controlling for all other covariates. The only difference is that the optimization procedure that fits the model, in addition to maximizing the likelihood of the observed data, works to minimize the size of the model coefficients, which is variously referred to in the literature as penalizing, regularizing, or shrinking the coefficients. The elastic net penalty is a mixture of “lasso” (L1) and “ridge” (L2) penalties. The lasso component favors sparsity by allowing the coefficients of the least influential covariates to be shrunk to 0, effectively selecting them out of the model entirely, while the ridge component favors inclusivity by enabling highly correlated variables to be shrunk together instead of arbitrarily picking one and discarding the rest. Importantly, in contrast to stepwise selection, variable selection is a by-product of coefficient shrinkage and the models are fit using all of the predictors. Cross-validation (tenfold) is used to tune the optimal combination and magnitude of penalties for a “just right” fit—flexible enough to capture real complexity, but constrained enough to avoid capturing noise—by choosing the amount of regularization that maximizes out-of-sample generalization instead of in-sample fit.

A potential disadvantage of the elastic net is that it does not provide an automatic search for potential nonlinear relationships or higher-order interactions among predictors, which other machine learning algorithms (e.g., random forests) would supply. However, in our publications and experience building machine learning ensembles to predict mental health outcomes [131], the elastic net has been the major workhorse of these ensembles, and the gains from adding other machine learners have been minimal at best (likely because the accurate capture of nonlinear, high-order interactions will require enormous sample sizes much larger than the ones we have worked with to date [132]). However, to assess how much loss of predictive accuracy an elastic-net may entail relative to more black-box approaches, we will also use superlearning/stacked ensembles, which have been touted as an optimal way to discover treatment rules that maximize response outcomes [133].

Model development and validation strategy

We will maintain the data collected at each site (UT and BU) as independent data sets. Two sets of models will be developed, one trained to each site’s data, which will then be used to predict the other site’s data. All statistical tests of model predictions will be performed using data from a completely independent sample. This ensures that findings are not merely “statistically significant” in the sense of being improbable under a null hypothesis, but also demonstrates real-world predictive utility by directly showing that predictions based on one clinical sample are likely to generalize to other clinical samples of the same population. We refer to the site used to fit models as the “train site” and the site used to validate models as the “test site”.

Hypothesis testing will be based on a series of model comparisons. All models will be fit to the train site and used to predict the probability of treatment response at the test site. These probabilities will be used to generate an ROC curve for each model’s ability to discriminate treatment responders from non-responders. Significant differences between curves will be evaluated using DeLong’s test for two correlated ROC curves as implemented by the R package “pROC”. The following model comparisons will be performed:

  1. 1.

    All candidate predictors vs. excluding CO2/VH reactivity measures. This comparison addresses the question: do the novel measures of CO2/VH reactivity add predictive value beyond traditional—and easier to collect—measures like diagnosis or symptom severity? We hypothesize that the addition of the reactivity measures will significantly improve model performance. If CO2 reactivity is completely confounded with these competing predictors, then adding it to the model will not result in any predictive gains; CO2 reactivity can only improve model predictions if it provides unique information that is not captured by “confounders”.

  2. 2.

    CO2 vs. VH reactivity. Finally, if the above model demonstrates the value of collecting CO2 reactivity measures, we will next assess does CO2 reactivity specifically predict treatment non-response (by revealing something about the underlying biology), or is it merely heightened sensitivity to the physiological sensations that are common to both CO2 inhalation and hyperventilation? Qualitatively, if VH reactivity has no predictive utility, then the statistical learning algorithm will have penalized these measures out of the model entirely by this point. However, the hypothesis of mechanistic specificity will be more formally tested by comparing a model that includes just CO2 reactivity measures to one that includes just VH reactivity measures. Note that the ability to perform a head-to-head comparison like this is one of the advantages of comparing ROC curves in an independent test set; the models do not need to be nested as they would when performing likelihood ratio tests on model fits.

Two strengths of this validation strategy should be emphasized. First, the significance of the biomarkers will be determined by their ability to outperform self-report measures, not just that their coefficients are significantly different from zero. The former speaks to their real-world utility; the latter does not. Second, the biomarkers are being compared on their ability to predict new data, not just their ability to fit the same data. This design/analytic feature helps ensure that we do not propagate failure-to-replicate issues. If we do not replicate a simplified model across sites and our hypotheses are not supported by the data (and there is not support for an alternative model with adequate sensitivity and specificity), we will recommend to the field that reactivity to CO2 and/or VH challenge (and the confounder variables we assessed) should not be used for determining patients’ suitability for EBT. Our use of a replicated assessment and a large sample helps instill confidence in these recommendations.

Translating anticipated ROC gains into clinical impact

The comparison of the ROC model curves is a framework for hypothesis testing, but the model will ultimately need to be couched in other terms besides an improvement in the ROC to convince clinicians of its practical utility, which depends not only on the improved performance of the model, but also clinical judgment about the optimal tradeoff between sensitivity and specificity: one can always boost the gain in the correct number of non-responders at the cost of increasing the number of falsely identified non-responders, and individual clinicians and patient circumstances may dictate different cut points. Thus, a metric like ROC that captures the overall tradeoff between specificity and sensitivity across all possible decision thresholds is the best way to frame our minimal objective, keeping in mind that this is a conservative expectation; larger predictive gains are possible.

However, assuming the minimal gains in ROC performance that we used to determine sample size requirements, we can offer a working example for one potential cutoff: if we were to require greater than an 80% predicted probability of non-response to forego a trial of EBT, then, out of 5000 patients entering treatment, the standard baseline model would only identify 4 true non-responders and falsely identify 1 responder as a likely non-responder. In contrast, the addition of CO2 reactivity measures would identify 262 true non-responders while misidentifying 26 responders as non-responders. This would spare > 10% of non-responders from an unsuccessful trial of EBT vs. < 0.2% under the baseline model, while incorrectly excluding only 1% of would-be responders. The biomarker assessment burden in our protocol is a single session, balanced against 12 one-hour sessions of treatment. The ability to spare 1/10 non-responding patients from the burden of unsuccessful treatment, while only negatively impacting 1/100 responding patients, justifies the burden of a single assessment session, in our opinion.

Deriving treatment recommendation heuristics

The final data product is to translate the above models into algorithmic recommendations to determine whether a patient is very unlikely (e.g., < 30% chance) to respond to EBT, which could support the recommendation of interventions that do not rely on fear extinction (e.g., other psychosocial interventions, pharmacological interventions). Although we chose a statistical learning algorithm that avoids “black box” predictions, obtaining a probability of non-response will still require inputting a number of measurements into something like a web-based calculator, which we will disseminate if the models demonstrate good predictive value. But clinicians might be more likely to adopt a treatment recommendation for a patient if they are able to derive it from decision rules (e.g., if a patient maintains a latency less than 90-s between CO2 inhalations) based on the measurements that are easiest to acquire. We will attempt to derive such decision rules by applying a recursive partitioning algorithm to the data (similar to the approach used to assist hospital pharmacy staff in identifying patients at risk of medication errors)[134] using only easy-to-obtain measures. We will then use the same model comparison strategy outlined above to compare the predictive performance of the decision rules to both a no-information model and the best models derived from elastic-net regression. If the simplified heuristic model offers significant gains over no information and is not significantly worse than the best models, then this will provide the ideal mechanism for translating the knowledge gained from this study to clinical practice.

Statistical power

We ran 5,000 computer simulations of our study design under a wide range of total sample sizes. In brief, a different multivariate distribution was defined for each of the following five diagnostic categories: (1) PD(with or without agoraphobia), (2) SAD, (3) GAD, (4) OCD, and (5) PTSD. Data were simulated to match the disorder-specific response rates (PD = 0.53, SAD = 0.45, GAD = 0.47, OCD = 0.43, PTSD = 0.59) reported in a meta-analysis of 87 studies [135]. The specified N for a given simulation was divided equally across the 2 sites and 5 disorder groupings within sites, and random samples of size N/10 were drawn for each disorder group for each site, one of which was arbitrarily labeled “train” and the other “test”. Just as we specified in our analytic plan, elastic-net logistic regression models were fit to the train site, and the optimal penalty mix (parameter alpha) and magnitude (parameter lambda) were selected using the average of 10 repeats of a tenfold cross-validation procedure. We then used this model to predict the response at the test site. These predictions (which are individual probabilities of response) and the actual response values were then used to generate ROC curves for the different models to be compared. The control model, which includes only “traditional” self-report measures, was assumed to have a true AUC of 0.63, and the experimental model, which adds CO2 and hyperventilation reactivity measures, was simulated to have a true AUC of 0.73. These values and their differences correspond to our minimal effect size of interest.

Then, for each of the simulated data sets, area under the ROC curve (AUC) was compared using DeLong’s test for two correlated ROC curves, as implemented by the `roc.test` function in the “pROC” package in R, with the directional hypothesis that the experimental model has a greater AUC than the control model. At a total sample size of 600 (300 train site/300 test site), 92.2% of simulations found a significant (p < 0.05) difference in model performance. Applying the same approach to the comparison of models with only CO2 reactivity (no hyperventilation measures) vs. models with only hyperventilation reactivity (no CO2 measures), 82.7% of simulations found a significant (p < 0.05) superiority of CO2 over hyperventilation. The likely reason for this ~ 10% drop in power is that we modeled CO2 and hyperventilation reactivity measures as likely to be correlated (r = 0.03—0.25) such that hyperventilation would have a smaller, spurious relationship with treatment outcome by proxy. While this makes the mechanistic specificity of CO2 reactivity more difficult to detect than if we had assumed independence between CO2 and hyperventilation reactivity, these simulations show that we are reasonably well powered to detect the mechanistically specific predictive value of CO2 even under this noisy condition. Power curves for these model comparisons at all simulated sample sizes are shown in Fig. 1.

Fig. 1
figure 1

Power curve


Anxiety-, obsessive–compulsive, and trauma- and stressor-related disorders reflect a significant public health problem. This study is designed to evaluate the predictive power of a novel biomarker based on a CO2 challenge, thus addressing the central question “can this easy-to-administer assay aid clinicians in deciding whether or not to initiate exposure-based therapy?”.

We hypothesize that the subjective and behavioral indices of CO2 reactivity will be the dominant predictors of EBT non-response. We do, however, include a set of competing predictors in our model. Hence, if our hypotheses are not supported and/or psychophysiological measures emerge as important predictors, we will still be able to deliver to the field an accounting of the variables/model that best predicts treatment failure. In essence, while our preliminary data provided a good basis to move forward, the work proposed in this application would comprehensively address possible confounders in a novel way. Ultimately, the final application of our work will be to recommend a specific assessment procedure and a specific go/no go assessment to guide clinical decisions for initiating exposure-based treatment vs. pursuing a different alternative.

Support for the study hypothesis that patients with elevated CO2 reactivity will not do well with EBT would justify the consideration to move away from this treatment modality altogether and instead select an alternative (e.g., psychosocial or pharmacological) modality for these patients, if indeed the biomarker is specific to EBT vs. an alternative treatment. As discussed, we have adopted a stage model approach to biomarker validation; targeting reliability of prediction in this study, with treatment specificity of prediction relegated to future investigations (i.e., we do not believe it is a cost-effective strategy to do both in the same initial trial). Moreover, we have considered a number of comparison conditions for a future study in the context of a randomized trial of two or more treatment arms, evaluating both alternative treatments and rescue strategies (i.e., a strategy that effectively recalibrates the mechanism that underlies the biomarker − EBT non-response relation), but all of these design decisions would be clearly informed by the results of the investigation proposed here (e.g., identification of the pre-randomization stratification point for CO2 reactivity). Finally, should the findings from the proposed study be consistent with the hypotheses, we will work with Hans Rudolph Inc. to develop accessible technology for assessing the relevant CO2 reactivity parameters thus aiding dissemination efforts.

Availability of data and materials

In line with NIMH guidance, we will share de-identified data derived from this study via the NIMH Data Archive (NDA;, along with supporting documentation to enable efficient and appropriate use of the data. Data will be available under collection #4334. We agree that data will be deposited and made available through NDA, and that these data will be shared with investigators working at an institution with a Federal Wide Assurance (FWA) and could be used for secondary study purposes. All submitted data (both descriptive/raw and analyzed data) will be made available for access by qualified members of the research community according to the provisions defined in the NIMH Data Repositories Data Access Agreement and Use Certification.

We agree to deposit and maintain the study data and secondary analysis of data (if any) at NDA. The repository has data access policies and procedures consistent with NIH data sharing policies.

Descriptive/raw data will be shared on a semi-annual basis (on or before January 15 and July 15, beginning six months after the award budget period has begun). Analyzed data will be submitted prior to publication/public dissemination (whether the findings are positive or negative) using the NDA study feature that links data deposited in the NDA with publications/findings. We will include the entire analyzed dataset even if a publication only focuses on a specific aspect of the dataset.

We will identify where the data will be available and how to access the data in any publications and presentations about these data, as well as acknowledge the repository and funding source. For each publication, a unique digital object identifier (DOI) will be created using the NDA Study feature, and this DOI will be included in the manuscript, linking the specific participants and data structures in the NDA that correspond to the analyses reported in each publication. The NDA has policies and procedures in place that will provide data access to qualified researchers, fully consistent with NIH data sharing policies and applicable laws and regulations.



Area under the receiver operating characteristic curve


Anxiety sensitivity


Anxiety Sensitivity Index—3rd version


Brief Experiential Avoidance Questionnaire


Basolateral amygdala


Clinical Global Impression


Clinical Global Impression—Severity


Clinical Global Impression—Improvement


Central nervous system

CO2 :

Carbon dioxide


Cerebrospinal fluid


Columbia-Suicide Severity Rating Scale


Dimensional Obsessive–Compulsive Scale


Distress intolerance


Distress Tolerance Index


Experiential avoidance


Exposure-based therapy


Generalized anxiety disorder


GAD-7 scale




Independent evaluator


Infralimbic cortex


Intolerance of uncertainty


Intolerance of Uncertainty Scale


Lateral hypothalamus

O2 :



Overall Anxiety Severity and Impairment Scale


Obsessive–compulsive disorder


Posttraumatic Stress Disorder Checklist for DSM-5


Panic disorder


Panic Disorder Severity Scale—Self-Report


Patient Health Questionnaire—9-item version


Posttraumatic stress disorder


Receiver operating characteristic


Social anxiety disorder


Structured Clinical Interview for DSM-5


Social Phobia Inventory


Safety Behavior Assessment Form


Voluntary hyperventilation


  1. Kessler RC, Chiu WT, Demler O, Merikangas KR, Walters EE. Prevalence, severity, and comorbidity of 12-month DSM-IV disorders in the National Comorbidity Survey Replication. Arch Gen Psychiatry. 2005;62:617–27.

    Article  Google Scholar 

  2. Kessler RC, Berglund P, Demler O, Jin R, Merikangas KR, Walters EE. Lifetime prevalence and age-of-onset distributions of DSM-IV disorders in the national comorbidity survey replication. Arch Gen Psychiatry. 2005;62:593–602.

    Article  Google Scholar 

  3. Stein MB, Roy-Byrne PP, Craske MG, Bystritsky A, Sullivan G, Pyne JM, et al. Functional impact and health utility of anxiety disorders in primary care outpatients. Med Care. 2005;43:1164–70.

    Article  Google Scholar 

  4. Wittchen HU, Jacobi F, Rehm J, Gustavsson A, Svensson M, Jönsson B, et al. The size and burden of mental disorders and other disorders of the brain in Europe 2010. Eur Neuropsychopharmacol J Eur Coll Neuropsychopharmacol. 2011;21:655–79.

    Article  CAS  Google Scholar 

  5. Carpenter JK, Andrews LA, Witcraft SM, Powers MB, Smits JAJ, Hofmann SG. Cognitive behavioral therapy for anxiety and related disorders: A meta-analysis of randomized placebo-controlled trials. Depress Anxiety. 2018;35:502–14.

    Article  CAS  Google Scholar 

  6. Clinical Practice Review for Social Anxiety Disorder | Anxiety and Depression Association of America, ADAA. Accessed 2 May 2020.

  7. American Psychiatric Association. Practice guideline for the treatment of patients with obsessive-compulsive disorder. 2007.

    Google Scholar 

  8. Practice Guideline for the Treatment of Patients With Panic Disorder. In: APA Practice Guidelines for the Treatment of Psychiatric Disorders: Comprehensive Guidelines and Guideline Watches. 1st edition. Arlington, VA: American Psychiatric Association; 2006.

  9. Katzman MA, Bleau P, Blier P, Chokka P, Kjernisted K, Van Ameringen M, et al. Canadian clinical practice guidelines for the management of anxiety, posttraumatic stress and obsessive-compulsive disorders. BMC Psychiatry. 2014;14(Suppl 1):S1.

    Article  Google Scholar 

  10. Management of Posttraumatic Stress Disorder and Acute Stress Reaction 2017 - VA/DoD Clinical Practice Guidelines. Accessed 2 May 2020.

  11. Karlin BE, Ruzek JI, Chard KM, Eftekhari A, Monson CM, Hembree EA, et al. Dissemination of evidence-based psychological treatments for posttraumatic stress disorder in the Veterans Health Administration. J Trauma Stress. 2010;23:663–73.

    Article  Google Scholar 

  12. Brehl AK, Kohn N, Schene AH, Fernández G. A mechanistic model for individualised treatment of anxiety disorders based on predictive neural biomarkers. Psychol Med. 2020;50(5):727–36.

  13. Davidson JRT, Foa EB, Huppert JD, Keefe FJ, Franklin ME, Compton JS, et al. Fluoxetine, comprehensive cognitive behavioral therapy, and placebo in generalized social phobia. Arch Gen Psychiatry. 2004;61:1005–13.

    Article  CAS  Google Scholar 

  14. Foa EB, Hembree EA, Cahill SP, Rauch SAM, Riggs DS, Feeny NC, et al. Randomized trial of prolonged exposure for posttraumatic stress disorder with and without cognitive restructuring: outcome at academic and community clinics. J Consult Clin Psychol. 2005;73:953–64.

    Article  Google Scholar 

  15. Foa EB, Liebowitz MR, Kozak MJ, Davies S, Campeas R, Franklin ME, et al. Randomized, placebo-controlled trial of exposure and ritual prevention, clomipramine, and their combination in the treatment of obsessive-compulsive disorder. Am J Psychiatry. 2005;162:151–61.

    Article  Google Scholar 

  16. Barlow DH, Gorman JM, Shear MK, Woods SW. Cognitive-behavioral therapy, imipramine, or their combination for panic disorder: A randomized controlled trial. JAMA. 2000;283:2529–36.

    Article  CAS  Google Scholar 

  17. Rauch SAM, Kim HM, Powell C, Tuerk PW, Simon NM, Acierno R, et al. Efficacy of Prolonged Exposure Therapy, Sertraline Hydrochloride, and Their Combination Among Combat Veterans With Posttraumatic Stress Disorder: A Randomized Clinical Trial. JAMA Psychiat. 2019;76:117–26.

    Article  Google Scholar 

  18. Keeley ML, Storch EA, Merlo LJ, Geffken GR. Clinical predictors of response to cognitive-behavioral therapy for obsessive-compulsive disorder. Clin Psychol Rev. 2008;28:118–30.

    Article  Google Scholar 

  19. Waters AM, Pine DS. Evaluating differences in Pavlovian fear acquisition and extinction as predictors of outcome from cognitive behavioural therapy for anxious children. J Child Psychol Psychiatry. 2016;57:869–76.

    Article  Google Scholar 

  20. Rauch SAM, Yasinski CW, Post LM, Jovanovic T, Norrholm S, Sherrill AM, et al. An intensive outpatient program with prolonged exposure for veterans with posttraumatic stress disorder: Retention, predictors, and patterns of change. Psychol Serv. 2020;:No Pagination Specified-No Pagination Specified.

  21. Teismann T, Brailovskaia J, Totzeck C, Wannemüller A, Margraf J. Predictors of remission from panic disorder, agoraphobia and specific phobia in outpatients receiving exposure therapy: The importance of positive mental health. Behav Res Ther. 2018;108:40–4.

    Article  Google Scholar 

  22. Brown TA, Barlow DH. Long-term outcome in cognitive-behavioral treatment of panic disorder: clinical predictors and alternative strategies for assessment. J Consult Clin Psychol. 1995;63:754–65.

    Article  CAS  Google Scholar 

  23. Chambless DL, Tran GQ, Glass CR. Predictors of response to cognitive-behavioral group therapy for social phobia. J Anxiety Disord. 1997;11:221–40.

    Article  CAS  Google Scholar 

  24. Heldt E, Gus Manfro G, Kipper L, Blaya C, Isolan L, Otto MW. One-year follow-up of pharmacotherapy-resistant patients with panic disorder treated with cognitive-behavior therapy: Outcome and predictors of remission. Behav Res Ther. 2006;44:657–65.

    Article  Google Scholar 

  25. Dow MGT, Kenardy JA, Johnston DW, Newman MG, Taylor CB, Thomson A. Prognostic indices with brief and standard CBT for panic disorder: I Predictors of outcome. Psychol Med. 2007;37:1493–502.

    Article  Google Scholar 

  26. Deckert J, Erhardt A. Predicting treatment outcome for anxiety disorders with or without comorbid depression using clinical, imaging and (epi)genetic data. Curr Opin Psychiatry. 2019;32:1–6.

    Article  Google Scholar 

  27. Aaronson CJ, Shear MK, Goetz RR, Allen LB, Barlow DH, White KS, et al. Predictors and time course of response among panic disorder patients treated with cognitive-behavioral therapy. J Clin Psychiatry. 2008;69:418–24.

    Article  Google Scholar 

  28. Raffin AL, GuimarãesFachel JM, Ferrão YA, Pasquoto de Souza F, Cordioli AV. Predictors of response to group cognitive-behavioral therapy in the treatment of obsessive-compulsive disorder. Eur Psychiatry J Assoc Eur Psychiatr. 2009;24:297–306.

    Article  Google Scholar 

  29. Steketee G, Siev J, Yovel I, Lit K, Wilhelm S. Predictors and Moderators of Cognitive and Behavioral Therapy Outcomes for OCD: A Patient-Level Mega-Analysis of Eight Sites. Behav Ther. 2019;50:165–76.

    Article  Google Scholar 

  30. Murphy D, Smith KV. Treatment Efficacy for Veterans With Posttraumatic Stress Disorder: Latent Class Trajectories of Treatment Response and Their Predictors. J Trauma Stress. 2018;31:753–63.

    Article  Google Scholar 

  31. Craske MG, Kircanski K, Zelikowsky M, Mystkowski J, Chowdhury N, Baker A. Optimizing inhibitory learning during exposure therapy. Behav Res Ther. 2008;46:5–27.

    Article  Google Scholar 

  32. Smits JAJ, Powers MB, Otto MW. Personalized exposure therapy: A person-centered transdiagnostic approach. New York, NY: Oxford University Press; 2019.

    Book  Google Scholar 

  33. Myers KM, Davis M. Mechanisms of fear extinction. Mol Psychiatry. 2007;12:120–50.

    Article  CAS  Google Scholar 

  34. Monfils MH, Holmes EA. Memory boundaries: opening a window inspired by reconsolidation to treat anxiety, trauma-related, and addiction disorders. Lancet Psychiatry. 2018;5:1032–42.

    Article  Google Scholar 

  35. Ball TM, Knapp SE, Paulus MP, Stein MB. Brain activation during fear extinction predicts exposure success. Depress Anxiety. 2017;34:257–66.

    Article  CAS  Google Scholar 

  36. Forcadell E, Torrents-Rodas D, Vervliet B, Leiva D, Tortella-Feliu M, Fullana MA. Does fear extinction in the laboratory predict outcomes of exposure therapy? A treatment analog study. Int J Psychophysiol. 2017;121:63–71.

    Article  Google Scholar 

  37. Lange I, Goossens L, Michielse S, Bakker J, Vervliet B, Marcelis M, et al. Neural responses during extinction learning predict exposure therapy outcome in phobia: results from a randomized-controlled trial. Neuropsychopharmacology. 2020;45:534–41.

    Article  Google Scholar 

  38. Berry AC, Rosenfield D, Smits JAJ. Extinction retention predicts improvement in social anxiety symptoms following exposure therapy. Depress Anxiety. 2009;26:22–7.

    Article  Google Scholar 

  39. Raeder F, Merz CJ, Margraf J, Zlomuzica A. The association between fear extinction, the ability to accomplish exposure and exposure therapy outcome in specific phobia. Sci Rep. 2020;10:4288.

    Article  Google Scholar 

  40. Craske MG, Hermans D, Vervliet B. State-of-the-art and future directions for extinction as a translational model for fear and anxiety. Philos Trans R Soc Lond B Biol Sci. 2018;373(1742):20170025.

  41. Anderson KC, Insel TR. The Promise of Extinction Research for the Prevention and Treatment of Anxiety Disorders. Biol Psychiatry. 2006;60:319–21.

    Article  Google Scholar 

  42. de Lecea L, Kilduff TS, Peyron C, Gao X-B, Foye PE, Danielson PE, et al. The hypocretins: Hypothalamus-specific peptides with neuroexcitatory activity. Proc Natl Acad Sci. 1998;95:322–7.

    Article  Google Scholar 

  43. Sakurai T, Amemiya A, Ishii M, Matsuzaki I, Chemelli RM, Tanaka H, et al. Orexins and Orexin Receptors: A Family of Hypothalamic Neuropeptides and G Protein-Coupled Receptors that Regulate Feeding Behavior. Cell. 1998;92:573–85.

    Article  CAS  Google Scholar 

  44. Sakurai T. The role of orexin in motivated behaviours. Nat Rev Neurosci. 2014;15:719–31.

    Article  CAS  Google Scholar 

  45. Mahler SV, Moorman DE, Smith RJ, James MH, Aston-Jones G. Motivational activation: a unifying hypothesis of orexin/hypocretin function. Nat Neurosci. 2014;17:1298–303.

    Article  CAS  Google Scholar 

  46. Chen Q, de Lecea L, Hu Z, Gao D. The hypocretin/orexin system: an increasingly important role in neuropsychiatry. Med Res Rev. 2015;35:152–97.

    Article  CAS  Google Scholar 

  47. Sharko AC, Fadel JR, Kaigler KF, Wilson MA. Activation of orexin/hypocretin neurons is associated with individual differences in cued fear extinction. Physiol Behav. 2017;178:93–102.

    Article  CAS  Google Scholar 

  48. Flores Á, Valls-Comamala V, Costa G, Saravia R, Maldonado R, Berrendero F. The hypocretin/orexin system mediates the extinction of fear memories. Neuropsychopharmacol Off Publ Am Coll Neuropsychopharmacol. 2014;39:2732–41.

    Article  CAS  Google Scholar 

  49. Flores Á, Saravia R, Maldonado R, Berrendero F. Orexins and fear: implications for the treatment of anxiety disorders. Trends Neurosci. 2015;38:550–9.

    Article  CAS  Google Scholar 

  50. Flores Á, Herry C, Maldonado R, Berrendero F. Facilitation of Contextual Fear Extinction by Orexin-1 Receptor Antagonism Is Associated with the Activation of Specific Amygdala Cell Subpopulations. Int J Neuropsychopharmacol. 2017;20:654–9.

    Article  CAS  Google Scholar 

  51. Monfils MH, Lee HJ, Keller NE, Roquet RF, Quevedo S, Agee L, et al. Predicting extinction phenotype to optimize fear reduction. Psychopharmacology. 2019;236:99–110.

    Article  CAS  Google Scholar 

  52. Furlong TM, Richardson R, McNally GP. Habituation and extinction of fear recruit overlapping forebrain structures. Neurobiol Learn Mem. 2016;128:7–16.

    Article  Google Scholar 

  53. Senn V, Wolff SBE, Herry C, Grenier F, Ehrlich I, Gründemann J, et al. Long-range connectivity defines behavioral specificity of amygdala neurons. Neuron. 2014;81:428–37.

    Article  CAS  Google Scholar 

  54. Gottschalk MG, Richter J, Ziegler C, Schiele MA, Mann J, Geiger MJ, et al. Orexin in the anxiety spectrum: association of a HCRTR1 polymorphism with panic disorder/agoraphobia, CBT treatment response and fear-related intermediate phenotypes. Transl Psychiatry. 2019;9:75.

    Article  Google Scholar 

  55. Blouin AM, Fried I, Wilson CL, Staba RJ, Behnke EJ, Lam HA, et al. Human hypocretin and melanin-concentrating hormone levels are linked to emotion and social interaction. Nat Commun. 2013;4:1547.

    Article  Google Scholar 

  56. Tang S, Huang W, Lu S, Lu L, Li G, Chen X, et al. Increased plasma orexin-A levels in patients with insomnia disorder are not associated with prepro-orexin or orexin receptor gene polymorphisms. Peptides. 2017;88:55–61.

    Article  Google Scholar 

  57. Mäkelä KA, Karhu T, Jurado Acosta A, Vakkuri O, Leppäluoto J, Herzig KH. Plasma Orexin-A Levels Do Not Undergo Circadian Rhythm in Young Healthy Male Subjects. Front Endocrinol (Lausanne). 2018;9:710.

  58. Deng B-S, Nakamura A, Zhang W, Yanagisawa M, Fukuda Y, Kuwaki T. Contribution of orexin in hypercapnic chemoreflex: evidence from genetic and pharmacological disruption and supplementation studies in mice. J Appl Physiol Bethesda Md. 1985;2007(103):1772–9.

    Google Scholar 

  59. Johnson PL, Samuels BC, Fitz SD, Lightman SL, Lowry CA, Shekhar A. Activation of the orexin 1 receptor is a critical component of CO2-mediated anxiety and hypertension but not bradycardia. Neuropsychopharmacol Off Publ Am Coll Neuropsychopharmacol. 2012;37:1911–22.

    Article  CAS  Google Scholar 

  60. Salvadore G, Bonaventure P, Shekhar A, Johnson PL, Lord B, Shireman BT, et al. Translational evaluation of novel selective orexin-1 receptor antagonist JNJ-61393215 in an experimental model for panic in rodents and humans. Transl Psychiatry. 2020;10:308.

    Article  CAS  Google Scholar 

  61. Tural U, Iosifescu DV. A systematic review and network meta-analysis of carbon dioxide provocation in psychiatric disorders. J Psychiatr Res. 2020.

    Article  Google Scholar 

  62. Leibold NK, van den Hove DLA, Viechtbauer W, Buchanan GF, Goossens L, Lange I, et al. CO2 exposure as translational cross-species experimental model for panic. Transl Psychiatry. 2016;6:e885.

    Article  CAS  Google Scholar 

  63. Battaglia M. Sensitivity to carbon dioxide and translational studies of anxiety disorders. Neuroscience. 2017;346:434–6.

    Article  CAS  Google Scholar 

  64. Griez EJ, Colasanti A, van Diest R, Salamon E, Schruers K. Carbon Dioxide Inhalation Induces Dose-Dependent and Age-Related Negative Affectivity. PLoS ONE. 2007;2:e987.

    Article  Google Scholar 

  65. CO2 challenge of patients with panic disorder. Am J Psychiatry. 1987;144:1080–2.

    Article  Google Scholar 

  66. Woods SW, Charney DS, Goodman WK, Heninger GR. Carbon Dioxide—Induced Anxiety: Behavioral, Physiologic, and Biochemical Effects of Carbon Dioxide in Patients With Panic Disorders and Healthy Subjects. Arch Gen Psychiatry. 1988;45:43–52.

    Article  CAS  Google Scholar 

  67. Schmidt NB, Telch MJ, Jaimez TL. Biological challenge manipulation of PCO2 levels: a test of Klein’s (1993) suffocation alarm theory of panic. J Abnorm Psychol. 1996;105:446–54.

    Article  CAS  Google Scholar 

  68. Schutters SI, Viechtbauer W, Knuts IJ, Griez EJ, Schruers KR. 35% CO2 sensitivity in social anxiety disorder. J Psychopharmacol (Oxf). 2012;26:479–86.

    Article  CAS  Google Scholar 

  69. Caldirola D, Perna G, Arancio C, Bertani A, Bellodi L. The 35% CO2 challenge test in patients with social phobia. Psychiatry Res. 1997;71:41–8.

    Article  CAS  Google Scholar 

  70. Seddon K, Morris K, Bailey J, Potokar J, Rich A, Wilson S, et al. Effects of 7.5% CO2 challenge in generalized anxiety disorder. J Psychopharmacol (Oxf). 2011;25:43–51.

    Article  CAS  Google Scholar 

  71. Perna G, Bussi R, Allevi L, Bellodi L. Sensitivity to 35% carbon dioxide in patients with generalized anxiety disorder. J Clin Psychiatry. 1999;60:379–84.

    Article  CAS  Google Scholar 

  72. Perna G, Bertani A, Arancio C, Ronchi P, Bellodi L. Laboratory response of patients with panic and obsessive-compulsive disorders to 35% CO2 challenges. Am J Psychiatry. 1995;152:85–9.

    Article  CAS  Google Scholar 

  73. Kellner M, Muhtz C, Nowack S, Leichsenring I, Wiedemann K, Yassouridis A. Effects of 35% carbon dioxide (CO2) inhalation in patients with post-traumatic stress disorder (PTSD): A double-blind, randomized, placebo-controlled, cross-over trial. J Psychiatr Res. 2018;96:260–4.

    Article  Google Scholar 

  74. Muhtz C, Yassouridis A, Daneshi J, Braun M, Kellner M. Acute panicogenic, anxiogenic and dissociative effects of carbon dioxide inhalation in patients with post-traumatic stress disorder (PTSD). J Psychiatr Res. 2011;45:989–93.

    Article  Google Scholar 

  75. Rapee RM, Brown TA, Antony MM, Barlow DH. Response to hyperventilation and inhalation of 5.5% carbon dioxide-enriched air across the DSM-III—R anxiety disorders. J Abnorm Psychol. 1992;101:538–52.

    Article  CAS  Google Scholar 

  76. Zvolensky MJ, Eifert GH. A review of psychological factors/processes affecting anxious responding during voluntary hyperventilation and inhalations of carbon dioxide-enriched air. Clin Psychol Rev. 2001;21:375–400.

    Article  CAS  Google Scholar 

  77. Telch MJ, Jacquin K, Smits JAJ, Powers MB. Emotional responding to hyperventilation as a predictor of agoraphobia status among individuals suffering from panic disorder. J Behav Ther Exp Psychiatry. 2003;34:161–70.

    Article  Google Scholar 

  78. Brown M, Smits JAJ, Powers MB, Telch MJ. Differential sensitivity of the three ASI factors in predicting panic disorder patients’ subjective and behavioral response to hyperventilation challenge. J Anxiety Disord. 2003;17:583–91.

    Article  Google Scholar 

  79. Williams RH, Burdakov D. Hypothalamic orexins/hypocretins as regulators of breathing. Expert Rev Mol Med. 2008;10:e28.

    Article  Google Scholar 

  80. Williams RH, Jensen LT, Verkhratsky A, Fugger L, Burdakov D. Control of hypothalamic orexin neurons by acid and CO2. Proc Natl Acad Sci U S A. 2007;104:10685–90.

    Article  CAS  Google Scholar 

  81. Sunanaga J, Deng B-S, Zhang W, Kanmura Y, Kuwaki T. CO2 activates orexin-containing neurons in mice. Respir Physiol Neurobiol. 2009;166:184–6.

    Article  CAS  Google Scholar 

  82. Reiss S, Peterson RA, Gursky DM, McNally RJ. Anxiety sensitivity, anxiety frequency and the prediction of fearfulness. Behav Res Ther. 1986;24:1–8.

    Article  CAS  Google Scholar 

  83. Smits JAJ, Otto MW, Powers MB, Baird SO. Anxiety sensitivity as a transdiagnostic treatment target. In: Smits JAJ, Otto MW, Powers MB, Baird SO, editors. The clinician’s guide to anxiety sensitivity treatment and assessment. 1st ed. San Diego, C.A.: Academic Press; 2018.

    Google Scholar 

  84. Telch MJ, Smits JAJ, Brown M, Dement M, Powers MB, Lee H, et al. Effects of threat context and cardiac sensitivity on fear responding to a 35% CO2 challenge: a test of the context-sensitivity panic vulnerability model. J Behav Ther Exp Psychiatry. 2010;41:365–72.

    Article  Google Scholar 

  85. Telch MJ, Harrington PJ, Smits JAJ, Powers MB. Unexpected arousal, anxiety sensitivity, and their interaction on CO2-induced panic: further evidence for the context-sensitivity vulnerability model. J Anxiety Disord. 2011;25:645–53.

    Article  Google Scholar 

  86. Schmidt NB, Telch MJ. Role of fear of fear and safety information in moderating the effects of voluntary hyperventilation. Behav Ther. 1994;25:197–208.

    Article  Google Scholar 

  87. McHugh RK, Reynolds EK, Leyro TM, Otto MW. An Examination of the Association of Distress Intolerance and Emotion Regulation with Avoidance. Cogn Ther Res. 2013;37:363–7.

    Article  Google Scholar 

  88. McHugh RK, Otto MW. Refining the Measurement of Distress Intolerance. Behav Ther. 2012;43:641–51.

    Article  Google Scholar 

  89. McHugh RK, Otto MW. Domain-general and domain-specific strategies for the assessment of distress intolerance. Psychol Addict Behav J Soc Psychol Addict Behav. 2011;25:745–9.

    Article  Google Scholar 

  90. Bond FW, Hayes SC, Baer RA, Carpenter KM, Guenole N, Orcutt HK, et al. Preliminary Psychometric Properties of the Acceptance and Action Questionnaire-II: A Revised Measure of Psychological Inflexibility and Experiential Avoidance. Behav Ther. 2011;42:676–88.

    Article  Google Scholar 

  91. Buhr K, Dugas MJ. The intolerance of uncertainty scale: psychometric properties of the English version. Behav Res Ther. 2002;40:931–45.

    Article  CAS  Google Scholar 

  92. Wilson EJ, Stapinski L, Dueber DM, Rapee RM, Burton AL, Abbott MJ. Psychometric properties of the Intolerance of Uncertainty Scale-12 in generalized anxiety disorder: Assessment of factor structure, measurement properties and clinical utility. J Anxiety Disord. 2020;76:102309.

    Article  Google Scholar 

  93. Carleton RN, Mulvogue MK, Thibodeau MA, McCabe RE, Antony MM, Asmundson GJG. Increasingly certain about uncertainty: Intolerance of uncertainty across anxiety and depression. J Anxiety Disord. 2012;26:468–79.

    Article  Google Scholar 

  94. Bomyea J, Ramsawh H, Ball TM, Taylor CT, Paulus MP, Lang AJ, et al. Intolerance of uncertainty as a mediator of reductions in worry in a cognitive behavioral treatment program for generalized anxiety disorder. J Anxiety Disord. 2015;33:90–4.

    Article  CAS  Google Scholar 

  95. Khakpoor S, Mohammadi Bytamar J, Saed O. Reductions in transdiagnostic factors as the potential mechanisms of change in treatment outcomes in the Unified Protocol: a randomized clinical trial. Res Psychother. 2019;22(3):379.

  96. Boswell JF, Thompson-Hollands J, Farchione TJ, Barlow DH. Intolerance of uncertainty: a common factor in the treatment of emotional disorders. J Clin Psychol. 2013;69(6):630–45.

  97. Katz D, Rector NA, Laposa JM. The interaction of distress tolerance and intolerance of uncertainty in the prediction of symptom reduction across CBT for social anxiety disorder. Cogn Behav Ther. 2017;46:459–77.

    Article  Google Scholar 

  98. Knowles KA, Olatunji BO. Enhancing Inhibitory Learning: The Utility of Variability in Exposure. Cogn Behav Pract. 2019;26:186–200.

    Article  Google Scholar 

  99. Keefer A, Kreiser NL, Singh V, Blakeley-Smith A, Duncan A, Johnson C, et al. Intolerance of Uncertainty Predicts Anxiety Outcomes Following CBT in Youth with ASD. J Autism Dev Disord. 2017;47:3949–58.

    Article  Google Scholar 

  100. Blakey SM, Abramowitz JS. The effects of safety behaviors during exposure therapy for anxiety: Critical analysis from an inhibitory learning perspective. Clin Psychol Rev. 2016;49:1–15.

    Article  Google Scholar 

  101. Group F-NBW. Predictive Biomarker. Food and Drug Administration (US); 2016.

  102. Greenland S. Tests for interaction in epidemiologic studies: a review and a study of power. Stat Med. 1983;2:243–51.

    Article  CAS  Google Scholar 

  103. Luedtke A, Sadikova E, Kessler RC. Sample size requirements for multivariate models to predict between-patient differences in best treatments of major depressive disorder. Clin Psychol Sci. 2019;7:445–61.

    Article  Google Scholar 

  104. Judd CM, Westfall J, Kenny DA. Experiments with More Than One Random Factor: Designs, Analytic Models, and Statistical Power. Annu Rev Psychol. 2017;68:601–25.

    Article  Google Scholar 

  105. Durand CP. Does raising type 1 error rate improve power to detect interactions in linear regression models? A simulation study PloS One. 2013;8:e71079.

    Article  CAS  Google Scholar 

  106. Kessler RC, Bossarte RM, Luedtke A, Zaslavsky AM, Zubizarreta JR. Machine learning methods for developing precision treatment rules with observational data. Behav Res Ther. 2019;120:103412.

    Article  Google Scholar 

  107. Norman SB, Campbell-Sills L, Hitchcock CA, Sullivan S, Rochlin A, Wilkins KC, et al. Psychometrics of a Brief Measure of Anxiety to Detect Severity and Impairment: The Overall Anxiety Severity and Impairment Scale (OASIS). J Psychiatr Res. 2011;45:262–8.

    Article  Google Scholar 

  108. Posner K, Brown GK, Stanley B, Brent DA, Yershova KV, Oquendo MA, et al. The Columbia-Suicide Severity Rating Scale: Initial validity and internal consistency findings from three multisite studies with adolescents and adults. Am J Psychiatry. 2011;168:1266–77.

    Article  Google Scholar 

  109. Meuret AE, Ritz T, Wilhelm FH, Roth WT. Voluntary hyperventilation in the treatment of panic disorder–functions of hyperventilation, their implications for breathing training, and recommendations for standardization. Clin Psychol Rev. 2005;25:285–306.

    Article  Google Scholar 

  110. Hofmann SG. Enhancing exposure-based therapy from a translational research perspective. Behav Res Ther. 2007;45:1987–2001.

    Article  Google Scholar 

  111. - DSM-5 Assessment Measures. Accessed 3 Nov 2022.

  112. Lebeau RT, Glenn DE, Hanover LN, Beesdo-Baum K, Wittchen H-U, Craske MG. A dimensional approach to measuring anxiety for DSM-5. Int J Methods Psychiatr Res. 2012;21:258–72.

    Article  Google Scholar 

  113. Chambless DL, Caputo GC, Jasin SE, Gracely EJ, Williams C. The Mobility Inventory for Agoraphobia. Behav Res Ther. 1985;23:35–44.

    Article  CAS  Google Scholar 

  114. Baker SL, Heinrichs N, Kim H-J, Hofmann SG. The liebowitz social anxiety scale as a self-report instrument: a preliminary psychometric analysis. Behav Res Ther. 2002;40:701–15.

    Article  Google Scholar 

  115. Pilkonis PA, Choi SW, Reise SP, Stover AM, Riley WT, Cella D. Item Banks for Measuring Emotional Distress From the Patient-Reported Outcomes Measurement Information System (PROMIS®): Depression, Anxiety, and Anger. Assessment. 2011;18:263–83.

    Article  Google Scholar 

  116. Steketee G, Frost R, Bogart K. The Yale-Brown Obsessive Compulsive Scale: interview versus self-report. Behav Res Ther. 1996;34:675–84.

    Article  CAS  Google Scholar 

  117. Blevins CA, Weathers FW, Davis MT, Witte TK, Domino JL. The Posttraumatic Stress Disorder Checklist for DSM-5 (PCL-5): Development and Initial Psychometric Evaluation. J Trauma Stress. 2015;28:489–98.

    Article  Google Scholar 

  118. First MB, Williams JB, Karg RS, Spitzer RL. Structured Clinical Interview for DSM-5® Disorders—Clinical Trials Version (SCID-5-CT). Arlington, VA: American Psychiatric Association; 2015.

    Google Scholar 

  119. Taylor S, Zvolensky MJ, Cox BJ, Deacon BJ, Heimberg RG, Ledley DR, et al. Robust dimensions of anxiety sensitivity: development and initial validation of the Anxiety Sensitivity Index-3. Psychol Assess. 2007;19:176–88.

    Article  Google Scholar 

  120. Gámez W, Chmielewski M, Kotov R, Ruggero C, Suzuki N, Watson D. The Brief Experiential Avoidance Questionnaire: Development and initial validation. Psychol Assess. 2014;26:35–45.

    Article  Google Scholar 

  121. Goodson JT, Haeffel GJ, Raush DA, Hershenberg R. The Safety Behavior Assessment Form: Development and Validation. J Clin Psychol. 2016;72:1099–111.

    Article  Google Scholar 

  122. Guy W, editor. Assessment manual for psychopharmacology. Rockville, MD: US Department of Health, Education, and Welfare Public Health Service Alcohol, Drug Abuse, and Mental Health Administration; 1976.

  123. Houck PR, Spiegel DA, Shear MK, Rucci P. Reliability of the self-report version of the Panic Disorder Severity Scale. Depress Anxiety. 2002;15:183–5.

    Article  Google Scholar 

  124. Spitzer RL, Kroenke K, Williams JBW, Löwe B. A brief measure for assessing generalized anxiety disorder: the GAD-7. Arch Intern Med. 2006;166:1092–7.

    Article  Google Scholar 

  125. Devilly GJ, Borkovec TD. Psychometric properties of the credibility/expectancy questionnaire. J Behav Ther Exp Psychiatry. 2000;31:73–86.

    Article  CAS  Google Scholar 

  126. Sainani KL. Multivariate regression: the pitfalls of automated variable selection. PM R. 2013;5:791–4.

    Article  Google Scholar 

  127. Taylor J, Tibshirani RJ. Statistical learning and selective inference. Proc Natl Acad Sci U S A. 2015;112:7629–34.

    Article  CAS  Google Scholar 

  128. Cawley GC. Over-Fitting in Model Selection and Its Avoidance. In: Hollmén J, Klawonn F, Tucker A, editors. Advances in Intelligent Data Analysis XI. Berlin, Heidelberg: Springer; 2012. p. 1–1.

    Google Scholar 

  129. Lever J, Krzywinski M, Altman N. Model selection and overfitting. Nat Methods. 2016;13:703–4.

    Article  CAS  Google Scholar 

  130. Zou H, Hastie T. Regularization and Variable Selection via the Elastic Net. J R Stat Soc Ser B Stat Methodol. 2005;67:301–20.

    Article  Google Scholar 

  131. Pearson R, Pisner D, Meyer B, Shumake J, Beevers CG. A machine learning ensemble to predict treatment outcomes following an Internet intervention for depression. Psychol Med. 2019;49:2330–41.

    Article  Google Scholar 

  132. McNamara ME, Zisser M, Beevers CG, Shumake J. Not just “big” data: Importance of sample size, measurement error, and uninformative predictors for developing prognostic models for digital interventions. Behav Res Ther. 2022;153:104086.

    Article  Google Scholar 

  133. Luedtke AR, van der Laan MJ. Super-Learning of an Optimal Dynamic Treatment Rule. Int J Biostat. 2016;12:305–32.

    Article  Google Scholar 

  134. Shea KM, Hobbs AL, Shumake JD, Templet DJ, Padilla-Tolentino E, Mondy KE. Impact of an antiretroviral stewardship strategy on medication error rates. Am J Health Syst Pharm. 2018;75:876–85.

    Article  Google Scholar 

  135. Loerinc AG, Meuret AE, Twohig MP, Rosenfield D, Bluett EJ, Craske MG. Response rates for CBT for anxiety disorders: Need for standardized criteria. Clin Psychol Rev. 2015;42:72–82.

    Article  Google Scholar 

Download references


Not applicable.

Consortium name

Exposure Therapy Consortium.


This study is funded by the National Institute of Mental Health (NIMH) grants 1/2: CO2 Reactivity as a Biomarker of Non-response to Exposure-based Therapy (R01MH125951) and 2/2: CO2 Reactivity as a Biomarker of Non-response to Exposure-based Therapy (R01MH125949). An external funding body, the NIMH is a United States federal agency that evaluates all grant applications for scientific and technical merit using the National Institutes of Health (NIH) peer review system.

Author information

Authors and Affiliations




Concept and design: Smits, Monfils, Otto, Telch, Shumake, Feinstein. Drafting of the manuscript: Smits, Monfils, Otto, Telch, Shumake. Critical revision of the manuscript for important intellectual content: All authors. Obtained funding: Smits, Monfils, Otto, Telch, Shumake. Administrative, technical, or material support: Khalsa, Cobb, Parsons, Long, McSpadden, Johnson, Greenberg. The author(s) read and approved the final manuscript.

Corresponding author

Correspondence to Jasper A. J. Smits.

Ethics declarations

Ethics approval and consent to participate

The University of Texas at Austin Institutional Review Board has approved the study protocol (STUDY00001631). This study will be conducted in accordance with the principles of the Declaration of Helsinki, and all participants will provide written informed consent prior to enrollment.

Consent for publication

Not applicable.

Competing interests

Dr. Smits reports personal fees from Big Health, Elsevier, the American Psychological Association, Academic Press, and Oxford University Press, and serves as advisor for Earkick. Dr. Monfils reports personal fees from Princeton University Press. Dr. Otto reports compensation as a Scientific Advisor for Big Health. Dr. Shumake reports consulting fees from Aiberry, Inc. The other authors report no competing interests or disclosures relevant to this work.

Additional information

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit The Creative Commons Public Domain Dedication waiver ( applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Smits, J.A.J., Monfils, MH., Otto, M.W. et al. CO2 reactivity as a biomarker of exposure-based therapy non-response: study protocol. BMC Psychiatry 22, 831 (2022).

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: