Definition, clinical use, and regulatory boundary
A CBT-I digital therapeutics system is a software-delivered clinical intervention based on cognitive behavioral therapy for insomnia and is classified in WSIC 4.1.1. It is generally intended to improve chronic insomnia symptoms and related daytime functioning through structured assessment, sleep diaries, stimulus control, individualized adjustment of time in bed, cognitive strategies, relaxation or arousal-reduction training, sleep education, and relapse prevention. A system may be self-directed or supported by a coach, therapist, or prescriber. Digital delivery, however, is only a mode of delivery; it does not reduce the requirements for treatment integrity, safety management, or evidence.
This entry does not describe a general sleep-education app, meditation library, sleep tracker, ordinary relaxation workbook, teleconsultation portal, or a course that merely moves printed material onto a screen. Nor does it encompass every broadly labeled “cognitive behavioral system for insomnia.” The terms “digital therapeutic,” “prescription-grade,” and “clinical-grade” do not by themselves establish regulatory authorization or efficacy. Whether a product is software as a medical device, who may prescribe it, whether it may be offered directly to consumers, and what supervision is necessary depend on its intended use, functions, risks, and jurisdiction. Product information should state the regulatory status, applicable region, indication, and version without implying authorization from that of a similar product.
Treatment model, core components, and fidelity
The system should disclose its treatment model, intended population, course structure, and the function of each component. A complete program is ordinarily more than sleep hygiene plus reminders. It translates diary data into an explainable behavioral plan: going to bed when sleepy, leaving the bed after extended wakefulness, stabilizing rise time, and cautiously restricting or compressing time in bed to consolidate sleep. It also addresses catastrophic interpretations and excessive effort around sleep, teaches cognitive alternatives and arousal-reduction skills, and includes a maintenance plan. Versions may adjust order and intensity, but a product that removes core mechanisms cannot claim equivalence by continuing to borrow evidence for complete CBT-I.
Digital implementation requires a defined treatment dose, module-release rules, handling of incomplete work, review opportunities, and limits of human support. An automated system should identify whether a recommendation comes from fixed rules, a statistical model, or an adaptive algorithm. A therapist-supported system should additionally state who reviews diaries, the expected response interval, and when a professional may override the software. Fidelity is not measured only by logins, clicks, or course completion: it also asks whether core components were correctly presented, understood, and enacted, and whether deviations represented appropriate safety adaptations. Reminders and gamification should not use shame, streak pressure, or punitive sleep scores to drive adherence.
Eligibility screening, risk controls, and clinical escalation
Before treatment begins, the system should establish the insomnia concern, duration, daytime impact, treatment goals, and ability to use the digital service. Screening should identify situations that may warrant prior or concurrent assessment, including prominent snoring or breathing pauses, uncontrollable daytime sleepiness, shift-work or circadian disorders, parasomnias, mania or bipolar risk, seizures, substance use, pregnancy, pain and complex medical conditions, and self-harm or suicide risk. Screening is not an automated exclusion diagnosis, and a negative questionnaire is not proof that disease is absent. Results should lead to an explicit path to proceed, adapt, defer, or refer.
Restriction of time in bed can temporarily increase sleepiness, and incorrect execution can endanger driving, work at height, machinery operation, or other safety-critical duties. Sleep reduction can also increase risk for some people with bipolar disorder or seizure disorders. The system therefore needs minimum safety limits, symptom review, pause rules, and conditions for human escalation, while allowing a clinician to adapt treatment to comorbidity, occupation, and caregiving duties. It should not autonomously stop or switch medication, change doses of melatonin or other supplements, or alter PAP or oxygen settings, and it should not blame deterioration on insufficient effort. If real-time monitoring or crisis care is unavailable, that limitation must be prominent, with locally appropriate routes to urgent and professional help.
Data inputs, algorithmic logic, and personalization
The sleep diary is usually a central treatment input. It should preserve the distinct meanings of going to bed, attempting sleep, estimated sleep onset, wakefulness during the night, final awakening, getting out of bed, naps, medication, and other modifiers. It also needs explicit handling for dates that cross midnight, time-zone and daylight-saving changes, shift work, caregiving interruptions, and missed entries. Wearables can supplement activity or physiological signals but should not silently replace a patient's report of wakefulness and distress. Non-wear, failed synchronization, and indeterminate algorithm output should be marked as missing or uncertain, never filled in as zero wakefulness or high-quality sleep.
Rules that turn diary data into a time-in-bed window, module branch, or risk alert require a version, input conditions, boundary checks, a displayed rationale, and auditable output. Personalization is not permission for a black box to issue progressively more aggressive instructions. Users and authorized clinicians should be able to inspect source records, correct obvious errors, and understand why an adjustment occurred; insufficient data, conflicting results, or a risk signal should trigger a conservative path. Any machine-learning component requires disclosure of its development and validation populations, performance limits, drift monitoring, and human override. A few nights of data must not be used to infer personality, fitness for work, or a psychiatric disorder, and a composite “sleep score” should not become the sole treatment end point.
Clinical evidence, validation, and interpretation of benefit
Evaluation should separately establish that the software works to specification, users can understand and operate it safely, the treatment model has clinical grounding, and the specific product and version produce meaningful benefit in the intended population. Studies should report recruitment, diagnostic criteria, comorbidities, digital exclusion, comparator, amount of human support, the definition of completion, attrition, and adverse events. Primary outcomes should be specified in advance, distinguishing insomnia severity, sleep onset and nocturnal wakefulness, subjective sleep, objective sleep, daytime function, quality of life, medication changes, and durability.
Research on digital CBT-I as a class does not automatically validate an individual application. Brand, algorithm, support model, language, version, and population can change benefit and risk. Engagement, satisfaction, login counts, or short-term improvement in sleep efficiency alone do not demonstrate durable clinical benefit. Reports should include absolute effects and uncertainty, external validation, and access across age, disability, low digital literacy, cultural and language groups, complex comorbidity, and underserved populations. Null results, reasons for withdrawal, transient sleepiness, symptom worsening, and escalation or referral belong beside benefits. Real-world monitoring complements rather than replaces suitable premarket clinical validation.
Human factors, accessibility, privacy, and cybersecurity
The interface should assume that a user may be tired, anxious, and operating in dim light. Critical tasks should be brief; text should resize; contrast should be clear; status should not depend on color alone; and keyboard, screen-reader, caption, audio-alternative, and plain-language access should be supported. Night notifications, bright displays, repeated confirmation, and complex charts may themselves increase arousal, so users need scheduling controls, a quiet mode, and paper or offline alternatives. Localization is more than replacing words: scales, examples, risk language, and help routes require comprehension testing in the local context. Usability testing should include actual intended users and high-risk workflows.
Sleep diaries, psychological responses, device data, prescriptions, and clinical communications are sensitive. The system should apply purpose limitation, data minimization, role-based access, strong authentication, encryption in transit and at rest, tenant isolation, audit logs, retention limits, export, and deletion, while defining the responsibilities of the patient, care organization, platform, and third parties. Treatment data should not default to advertising, data brokerage, or unrelated profiling. Cybersecurity covers design, release, vulnerability intake, patching, dependencies, incident response, and end of support. When a wearable, health record, or messaging connection fails, provenance, time, and status should remain visible, and the system should fail safely rather than silently generate an erroneous recommendation.
Quality system, change management, and procurement acceptance
The developer should manage intended use, user needs, software requirements, risk controls, traceability, verification and validation, clinical evaluation, release approval, complaints, adverse events, and postmarket surveillance within a quality system. Before changing content, thresholds, a rules engine, model, interface, or third-party dependency, it should determine whether the change affects treatment dose, eligible population, performance, or regulatory status and perform proportionate regression, usability, and clinical-impact assessment. Users need visibility into the current version, material changes, effective date, and any renewed consent. Phased rollout and experimentation may not bypass safety controls or allow a product name to drift silently beyond the version supported by evidence.
Procurement or prescribing review should examine indication, contraindications and escalation, professional responsibilities, support response, evidence-version alignment, data location, interoperability, accessibility, cybersecurity documentation, service discontinuation, and data migration. Download counts, marketing “completion rates,” reimbursement eligibility, and platform availability are not substitutes. Contracts should address outage notice, vulnerability remediation, audit, subcontractors, return of data, and safe exit. Related entries include bedtime relaxation workbooks, cognitive behavioral systems for insomnia, digital CBT-I platforms, and remote sleep-care platforms. This system can serve as one element of stepped care, but automation does not replace necessary diagnosis, professional judgment, shared decision-making, or emergency support.