{"schema_version":1,"research_id":"eoa_inverse_innovation_exp04_external_evaluation_20260802","source_assessment_id":"invariant_mode_decomposition_design__psychology:RETRIEVAL_FIRST:v0","cell_id":"invariant_mode_decomposition_design__psychology","search_queries":["site:pmc.ncbi.nlm.nih.gov exposure therapy fear reduction safety behaviors avoidance inhibitory learning Craske","site:nice.org.uk exposure therapy guideline anxiety disorder exposure recommendations clinician","site:healthquality.va.gov anxiety exposure therapy guideline measurement based care PTSD 2023 PDF","site:fda.gov clinical decision support software guidance clinician independently review basis recommendations 2022","site:nimh.nih.gov strategic plan personalized mental health treatment measurement based care computational 2025","site:nih.gov funding opportunity adaptive interventions mental health personalized just-in-time 2025","exposure therapy dropout rates prolonged exposure safety behavior fear reduction primary study PMC","adaptive exposure therapy physiological reinforcement learning virtual reality 2025 paper","\"An adaptation algorithm for personalised virtual reality exposure therapy\" full text","site:ojs.aaai.org AIIDE 2021 automated personalized exposure therapy physiological measures","site:patents.google.com exposure therapy biofeedback titration physiological adaptive difficulty","site:jmir.org control systems engineering adaptive mobile health interventions system identification tutorial","Craske Treanor Conway Zbozinek Vervliet 2014 PubMed 24864005","Comparing Dropout Cognitive Processing Therapy Prolonged Exposure 916 veterans 2026 PubMed","site:healthquality.va.gov PTSD CPG 2023 full guideline prolonged exposure PDF","site:ptsd.va.gov prolonged exposure treatment first line VA 2026"],"sources":[{"source_id":"S1","title":"Maximizing exposure therapy: an inhibitory learning approach","publisher":"Behaviour Research and Therapy / PubMed","url":"https://pubmed.ncbi.nlm.nih.gov/24864005/","source_class":"AUTHORITATIVE_SECONDARY","publication_date":"2014-05-09","accessed_at":"2026-08-02","claims_supported":["Exposure is effective, but a substantial proportion of patients do not obtain clinically significant relief or experience return of fear.","Within-session fear reduction and ending fear levels are not reliable predictors of later fear, supporting the concern that fear alone is an incomplete pacing signal.","Safety behaviors are clinically relevant but their effects are context-dependent, and gradual removal may be used to reduce attrition."]},{"source_id":"S2","title":"Comparing dropout from cognitive processing therapy versus prolonged exposure: Results from a randomized clinical trial","publisher":"Journal of Consulting and Clinical Psychology / PubMed","url":"https://pubmed.ncbi.nlm.nih.gov/41926191/","source_class":"PRIMARY_RESEARCH","publication_date":"2026-03-01","accessed_at":"2026-08-02","claims_supported":["In a 916-veteran randomized trial, dropout was 52.31% for prolonged exposure and 45.77% for cognitive processing therapy.","The authors identify treatment dropout as a continuing problem and call for intensive longitudinal research using time-varying predictors.","This supports the importance of retention, but does not establish that fear-score pacing causes dropout or pacing errors."]},{"source_id":"S3","title":"Prolonged Exposure for PTSD","publisher":"U.S. Department of Veterans Affairs, National Center for PTSD","url":"https://www.ptsd.va.gov/professional/treat/txessentials/prolonged_exposure_pro.asp","source_class":"OFFICIAL_GUIDANCE","publication_date":"2026-02-20","accessed_at":"2026-08-02","claims_supported":["VA identifies prolonged exposure as strongly recommended and widely studied, making VA and exposure-treatment clinics credible adopter settings.","The page reports continuing work to improve effectiveness, access, and retention.","PE remains clinician-delivered treatment; this supports preserving therapist authority rather than autonomous pacing."]},{"source_id":"S4","title":"Using Just-in-Time Adaptive Interventions to Optimize Established Adolescent Mental Health Treatments (R61/R33 Clinical Trial Required)","publisher":"National Institute of Mental Health","url":"https://grants.nih.gov/grants/guide/rfa-files/RFA-MH-22-150.html","source_class":"GOVERNMENT_OR_REGULATOR","publication_date":"2021-11-15","accessed_at":"2026-08-02","claims_supported":["NIMH explicitly sought adaptive, sensor-informed augmentations to established mental-health treatments with empirical mechanisms, decision points, tailoring variables, and decision rules.","The program required milestone-driven measurement validation, safety, usability, acceptability, target engagement, clinical outcomes, stakeholder involvement, and human-subject protections.","NIMH intended $2 million for up to five phased awards, providing a historical resource benchmark and evidence of funder pull, although the opportunity is expired and adolescent-specific."]},{"source_id":"S5","title":"Personalizing Exposure Therapy via Reinforcement Learning","publisher":"arXiv","url":"https://arxiv.org/abs/2504.14095","source_class":"PRIMARY_RESEARCH","publication_date":"2025-04-18","accessed_at":"2026-08-02","claims_supported":["A human-subject adaptive VRET system used electrodermal activity and reinforcement learning to modify spider attributes toward therapist-defined anxiety levels.","The system outperformed a rules-based method at reaching desired anxiety levels but did not test therapeutic outcomes.","This substantially overlaps personalized closed-loop pacing while leaving a contrast with explicit multivariate mode stability and sensitivity."]},{"source_id":"S6","title":"US11523773B2 — Biofeedback for therapy in virtual and augmented reality","publisher":"United States patent record via Google Patents","url":"https://patents.google.com/patent/US11523773B2","source_class":"GOVERNMENT_OR_REGULATOR","publication_date":"2022-12-13","accessed_at":"2026-08-02","claims_supported":["The patent discloses collecting motion and biometric data, using a learning system, and adjusting therapy protocols during or before later rounds.","For exposure therapy, intensity and duration may be scaled from biometric feedback, including reducing exposure when heart rate exceeds a target.","This creates substantial prior-art collision for multimodal, response-dependent exposure titration, while not visibly disclosing the proposed invariant-mode rule."]},{"source_id":"S7","title":"Tutorial for Using Control Systems Engineering to Optimize Adaptive Mobile Health Interventions","publisher":"Journal of Medical Internet Research","url":"https://www.jmir.org/2018/6/e214/","source_class":"AUTHORITATIVE_SECONDARY","publication_date":"2018-06-28","accessed_at":"2026-08-02","claims_supported":["The tutorial already describes dynamical modeling, intervention-option selection, open-loop system identification, controller design, and control-optimization trials for adaptive behavioral interventions.","It states that control methods require dynamic outcomes measured at sufficient temporal density and validated person-specific models.","It recommends comparison with standard care only after the controller meets predefined optimization criteria."]},{"source_id":"S8","title":"Clinical Decision Support Software: Guidance for Industry and Food and Drug Administration Staff","publisher":"U.S. Food and Drug Administration","url":"https://www.fda.gov/media/109618/download","source_class":"OFFICIAL_GUIDANCE","publication_date":"2026-01-29","accessed_at":"2026-08-02","claims_supported":["Non-device CDS must enable a health professional to independently review the recommendation basis and avoid primary reliance on the software.","Software that analyzes patterns or signals from medical signal-acquisition systems can remain a regulated device function.","The candidate's clinician override and displayed residuals help, but physiological-signal use, intended-use wording, and recommendation specificity require formal regulatory classification."]}],"problem_evidence":{"support":"MODERATE","rationale":"The general problem matters: fear reduction is not a sufficient index of exposure learning, safety behavior is independently relevant, and exposure-treatment dropout is substantial. However, no opened source directly demonstrates the candidate's precise failure mode—premature escalation caused by a growing avoidance mode or unnecessary repetition caused by harmless stable arousal—or establishes its prevalence relative to competent clinician judgment.","source_ids":["S1","S2","S3"]},"stakeholder_evidence":{"support":"MODERATE","rationale":"VA is an identifiable delivery system that uses prolonged exposure and seeks better retention and effectiveness. NIMH has explicitly funded milestone-driven adaptive mental-health augmentations involving sensing, decision rules, safety, and stakeholder involvement. Neither source expresses demand for this exact modal controller, the cited NIMH opportunity is expired and adolescent-specific, and no clinic has committed data, personnel, or adoption authority.","source_ids":["S3","S4"]},"prior_art":{"proximity":"SUBSTANTIAL_COLLISION","closest_analogues":[{"name":"Biofeedback-controlled VR exposure titration","similarity":"Collects biometric and motion feedback and uses a learning system to alter exposure intensity or duration during or between treatment rounds.","remaining_difference":"The opened disclosure uses learned outputs or physiological targets, not an explicit five-variable transition eigendecomposition whose stability and intervention sensitivity select advance, hold, or regression.","source_ids":["S6"]},{"name":"Physiology-driven reinforcement-learning VRET","similarity":"Personalizes exposure content in closed loop from physiological response and has been tested against a rules-based controller in humans.","remaining_difference":"It regulates toward a therapist-defined scalar anxiety level and did not test clinical outcomes; it does not visibly decompose fear, approach, avoidance, arousal, and safety behavior into action-relevant modes.","source_ids":["S5"]},{"name":"Idiographic control-system optimization of behavioral interventions","similarity":"Already combines deliberate intervention variation, within-person dynamical model estimation, controller design, validation, and later comparison with standard care.","remaining_difference":"It is domain-general guidance rather than an exposure-therapy implementation using modal stability and sensitivity as the decision rule.","source_ids":["S7"]},{"name":"Inhibitory-learning exposure practice","similarity":"Rejects fear reduction as the sole exposure endpoint and explicitly attends to expectancy, avoidance-related learning, and safety behavior.","remaining_difference":"It is a clinical learning framework, not a fitted multivariate transition operator or automated next-trial recommendation system.","source_ids":["S1"]}],"distinctive_claim_remaining":"The bounded remaining claim is that next-trial difficulty can be selected more effectively by fitting a local transition operator over prespecified fear, approach, avoidance, arousal, and safety-behavior measures and jointly using identifiable mode stability plus difficulty sensitivity, with residual and drift abstention, than by scalar fear/anxiety pacing, biofeedback thresholds, reinforcement learning without explicit modal structure, or clinician judgment alone.","confidence":"MODERATE"},"implementation_evidence":{"support":"WEAK","rationale":"Adaptive sensing, exposure titration, within-person system identification, and clinician-facing CDS are individually feasible. The decisive feasibility question is unresolved: a patient-specific five-state transition model plus difficulty input contains many parameters, while four sessions may supply too few sufficiently varied trials for identifiable and stable modes. Measurement scaling, missingness, near-degenerate modes, nonstationarity, residual thresholds, data protection, IRB oversight, and FDA device status also remain open. The shadow-mode design limits immediate treatment risk but does not solve statistical identifiability.","source_ids":["S4","S5","S6","S7","S8"]},"scores":{"meaningful_impact":{"score":4,"rationale":"Exposure therapy is important and widely used, while incomplete response and dropout are consequential. The candidate targets a clinically meaningful process, but its proposed pacing-error mechanism is not yet observed directly.","source_ids":["S1","S2","S3"]},"stakeholder_pull":{"score":3,"rationale":"VA and NIMH provide credible institutional interest in exposure-treatment improvement and adaptive mental-health interventions, but there is no current solicitation or named clinical partner for this system.","source_ids":["S3","S4"]},"incremental_advantage":{"score":2,"rationale":"The claimed 20% greater safety-behavior reduction and bounded dropout difference have no supporting comparative evidence. Existing adaptive VRET and control-system approaches already cover most workflow elements.","source_ids":["S5","S6","S7"]},"distinctiveness_plausibility":{"score":3,"rationale":"No opened source contained the complete five-variable modal stability-and-sensitivity rule, but the remaining distinction is narrow and could disappear under broader patent, bibliographic, product, Koopman/state-space, or non-English searches.","source_ids":["S5","S6","S7"]},"technical_implementability":{"score":2,"rationale":"The component methods are implementable, but reliable patient-specific modal estimation from the proposed short, noisy, deliberately varied clinical sequence is unproven and may be underidentified.","source_ids":["S5","S7"]},"adoption_authority_feasibility":{"score":3,"rationale":"Clinician authority, patient stopping rights, abstention, and shadow-mode evaluation make research feasible in principle. IRB approval, a clinical sponsor, privacy governance, and FDA classification remain prerequisites.","source_ids":["S3","S4","S8"]},"evidence_readiness":{"score":2,"rationale":"The candidate has a comparator, outcome threshold, falsifiers, and shadow protocol, but lacks a dataset, partner, sample-size or excitation justification, validated five-variable measurement protocol, and prespecified numerical identifiability and residual thresholds.","source_ids":["S4","S7"]},"safety_net_benefit":{"score":3,"rationale":"Hidden recommendations, clinician override, patient stopping rights, range restrictions, and reversion to usual pacing provide a credible initial safety net. Later treatment-changing use could still misdirect exposure if modes are unstable or important state variables are omitted.","source_ids":["S1","S8"]},"scalability":{"score":2,"rationale":"Software computation is scalable, but repeated multivariate measurement, sufficient within-patient excitation, clinician review, model abstention, and regulatory quality controls may impose substantial operational burden.","source_ids":["S4","S7","S8"]}},"score_confidence":"MODERATE","costs":{"first_evidence":{"band_2026_usd":"250K_TO_1M","scope":"Single-site, approximately 20–30 participant shadow-mode feasibility study with four clinician-directed sessions per participant, instrument selection, software prototype, secure data capture, statistical simulation, analyst and clinician effort, IRB oversight, and adverse-event monitoring.","confidence":"LOW","assumptions":["An existing exposure clinic, clinical protocol, and basic research data platform are available.","No autonomous treatment changes occur.","At least 12–20 analyzable exposure trials per participant can be obtained without unacceptable burden.","The historical NIMH phased-award scale is only a rough resource benchmark, not a quotation."],"source_ids":["S4","S7"]},"initial_deployment_startup":{"band_2026_usd":"1M_TO_5M","scope":"Production-grade clinician-facing software, validated instruments, security and privacy engineering, usability work, quality management, regulatory analysis or submission preparation, integration, training, and a prospective supervised pilot.","confidence":"LOW","assumptions":["Deployment remains clinician-facing and is limited to approved exposure protocols.","Physiological sensors, if used, are legally marketed and integrated consistently with their labeling.","Regulatory classification may require device-grade documentation."],"source_ids":["S5","S6","S8"]},"operational_launch":{"band_2026_usd":"1M_TO_5M","scope":"Multisite randomized comparison against fear-score pacing and clinician-managed usual care, including recruitment, clinical delivery, coordinating-center functions, safety monitoring, software operations, and analysis of safety behavior, symptoms, dropout, and subgroup performance.","confidence":"LOW","assumptions":["Three to five sites and roughly 150–300 participants are required for preliminary effectiveness and safety estimates.","The modal controller has already passed shadow-mode identifiability and usability gates.","This is not a definitive market-wide effectiveness trial."],"source_ids":["S2","S4","S7"]},"annual_recurring":{"band_2026_usd":"250K_TO_1M","scope":"Per-program annual software hosting, security review, model monitoring, support, clinician training, measurement licensing or hardware replacement, incident management, and periodic performance and equity audits.","confidence":"LOW","assumptions":["One health-system program with multiple clinics is supported.","Clinical labor for ordinary therapy is excluded; incremental measurement and review time are included.","No estimate of commercial pricing, reimbursement, or market size is implied."],"source_ids":["S4","S8"]}},"verified_pipeline_gates":{"externally_supported_problem":{"status":"YES","reason":"Independent clinical literature shows that fear reduction is an incomplete exposure-learning signal, safety behavior matters, and dropout remains substantial, although the candidate's exact pacing-error mechanism is not yet quantified.","source_ids":["S1","S2","S3"]},"externally_credible_adopter_or_authorizer":{"status":"YES","reason":"VA is a credible exposure-treatment delivery system and NIMH has explicitly solicited adaptive, mechanism-based mental-health intervention research. This establishes institutional plausibility, not a commitment to adopt or fund this candidate.","source_ids":["S3","S4"]},"distinct_testable_incremental_claim":{"status":"YES","reason":"The explicit modal stability-and-sensitivity decision rule can be compared with clinician usual care, scalar fear pacing, and a regularized non-modal multivariate predictor, with abstention and clinical endpoints prespecified.","source_ids":["S5","S6","S7"]},"bounded_next_evidence_step":{"status":"YES","reason":"A prospective four-session shadow study can test data completeness, mode identifiability, prediction, decision disagreement, burden, and abstention without changing care.","source_ids":["S4","S7"]},"no_unresolved_safety_or_authority_stop":{"status":"UNCERTAIN","reason":"Shadow mode and clinician authority control immediate risk, but a partner, IRB approval, consent and data plan, adverse-event governance, and FDA classification—especially if physiological signal patterns are analyzed—are unresolved.","source_ids":["S3","S4","S8"]},"credible_cost_scope_and_range":{"status":"YES","reason":"The ranges are broad, scope-specific resource estimates anchored loosely to a historical NIMH phased-research program and the documented requirements for clinical, technical, safety, and regulatory work. Confidence remains low because no site quotations or protocol-based staffing budget exists.","source_ids":["S4","S8"]}},"next_evidence_step":"Run a preregistered, single-site shadow-mode study in approximately 20–30 consenting patients receiving clinician-directed graded exposure. Before enrollment, simulate the proposed estimator to set a minimum number and pattern of trials, parameter regularization, excitation requirement, missing-data rule, mode-identifiability threshold, eigenvector/gain stability threshold, residual tolerance, drift threshold, and abstention rule. Collect the five prespecified measures after each trial for four sessions; train only on preceding trials and issue hidden increase/hold/reduce recommendations. Compare (1) out-of-sample prediction and recommendations from the modal controller, (2) scalar fear-score pacing, (3) a regularized non-modal multivariate predictor using the same variables, and (4) the treating clinician's actual decision. Primary feasibility falsifiers are: fewer than 80% of participants achieving prespecified identifiable modes; modal prediction no better than both algorithmic comparators; unstable mode identity under bootstrap or adjacent windows; fewer than 15% clinically reviewable recommendation disagreements; excessive missingness or burden; any recommendation outside the approved hierarchy; or safety/abstention failures. Do not estimate treatment benefit from hidden recommendations. Only after these gates pass should a clinician-supervised randomized study test the stated at-least-20% incremental safety-behavior reduction and no-more-than-5-percentage-point dropout increase.","blocking_evidence":["No direct evidence estimates how often fear-score pacing causes the specific premature-escalation or unnecessary-repetition errors.","No clinical partner, data-access agreement, IRB approval, or adopter commitment is identified.","The number and spacing of within-session trials required to identify a five-state patient-specific transition operator have not been justified.","Reliability, scaling, temporal resolution, and burden of the five proposed state measures are unspecified.","No prospective evidence shows that modal recommendations are reproducible, distinct from simpler multivariate models, understandable to clinicians, or predictive out of sample.","The 20% safety-behavior advantage and 5-percentage-point dropout margin require a treatment-changing randomized trial.","FDA device status, privacy architecture, cybersecurity requirements, and applicable professional-liability allocation remain unresolved.","Worldwide novelty, patentability, freedom to operate, market size, and realized impact remain unmeasured."],"research_disposition":"PARTNERED_RESEARCH_PROGRAM","world_novelty_boundary":"This evaluation establishes only that eight opened sources show substantial prior art for adaptive exposure titration, physiological feedback, reinforcement learning, clinician-directed exposure practice, and idiographic control-system development, while not visibly showing the complete five-variable invariant-mode stability-and-sensitivity decision rule. It does not measure worldwide novelty, patentability, freedom to operate, market size, or realized impact. Broader patent-family and claims review, bibliographic-database searching, product documentation, non-English searching, and terminology searches covering state-space models, latent dynamical systems, Koopman modes, spectral control, dynamic factor models, and switching systems could erase the remaining distinction.","arm":"RETRIEVAL_FIRST","candidate_version":0,"controller_recommendation":{"action":"STOP_EMPIRICAL_RESEARCH_NEEDED","repairable":true,"material_progress_observed":true,"progress_targets":["Secure an exposure-therapy clinical partner, treating-clinician sponsor, data governance approval, and IRB pathway.","Use simulation and pilot data to justify the minimum trial count, excitation schedule, regularization, and numerical identifiability thresholds for the five-state model.","Validate the reliability, timing, scaling, missingness, and patient burden of fear, approach, avoidance, arousal, and safety-behavior measurements.","Preregister modal, scalar-fear, non-modal multivariate, and clinician comparators plus prediction, abstention, stability, burden, and safety falsifiers.","Obtain a documented FDA/CDS classification analysis and define privacy, cybersecurity, audit, and professional-liability controls.","Complete the shadow study before any treatment-changing use; proceed to randomization only if modes are reproducible and outperform simpler models."],"reason":"Bounded web research verifies an important adjacent problem, credible institutions, extensive colliding practice, and a narrow falsifiable residual claim. It cannot determine whether patient-specific modes are identifiable from four sessions or whether modal pacing improves safety behavior without worsening dropout. Those decisive questions require proprietary clinical data, human-subject fieldwork, and eventually live randomized testing."}}