Research Aims

Sleep-EVAL is one of the earliest operational clinical artificial intelligence systems in sleep medicine, developed since 1987 and continuously used in epidemiological studies worldwide.

The program was created in response to a fundamental scientific gap: for decades, most knowledge about sleep disorders came from clinical samples recruited in specialized sleep clinics. Such samples are invaluable for mechanistic and therapeutic research, but they are not representative of the community because referral patterns, access to care, comorbidity and socioeconomic factors strongly shape who reaches a specialty clinic. As a consequence, prevalence estimates, clinical profiles, comorbidity patterns and treatment pathways inferred from clinic populations can substantially differ from those observed in the general population.

The Sleep-EVAL program addresses this limitation by combining large-scale population surveys with a standardized, AI-driven clinical interview system that reproduces clinical diagnostic reasoning. This enables rigorous epidemiological investigations while preserving the clinical richness of a diagnostic evaluation.

Sleep-EVAL constitutes the diagnostic interview and phenotyping engine of the Ad-Infer clinical AI platform. Ad-Infer is currently implemented as a hybrid AI architecture that combines a knowledge-driven inference core with statistical and machine learning components for population-scale analyses, and a modern LLM communication layer that supports natural-language interaction, clarification, and reporting. The LLM layer improves usability; the diagnostic decision authority remains governed by the Ad-Infer inference core.

The Ad-Infer inference core is a knowledge-based diagnostic reasoning system designed to generate clinically interpretable outputs. It relies on (i) a structured clinical knowledge base integrating international diagnostic frameworks (DSM, ICSD, ICD), (ii) rule-based and non-monotonic reasoning to manage complex diagnostic dependencies and exclusions, and (iii) probabilistic models and fuzzy logic to represent uncertainty, severity, and graded symptom thresholds commonly used in clinical practice.

Beyond categorical diagnoses, the platform produces standardized clinical representations that can be used for epidemiology, longitudinal research, and translational studies. These representations include symptoms, timing, severity, frequency, treatment exposure, impairment, risk factors and comorbidities, enabling high-dimensional analyses while preserving clinical interpretability.

A central aim of the Sleep-EVAL program is to generate AI-driven phenotypes of sleep and mental disorders in the general population. Traditional epidemiology often relies on a small set of screening questions or broad symptom counts. In contrast, Sleep-EVAL captures detailed, clinically structured information and builds multidimensional phenotypes that encompass:

• Sleep symptoms and sleep disorder criteria (initiation, maintenance, timing, breathing-related events, parasomnias, motor phenomena, hypersomnolence, circadian patterns)
• Psychiatric symptom dimensions (mood, anxiety, irritability, anhedonia, suicidality, cognitive and behavioral features)
• Neurological symptoms relevant to sleep and vigilance disorders
• Medical comorbidities, medications, substance exposure and treatment history
• Functional impairment, health-care use, and real-world outcomes

This phenotyping framework enables the identification of clinically meaningful subtypes, comorbidity structures, and trajectories that cannot be captured by diagnostic categories alone. It also provides a principled basis for transdiagnostic research by allowing disorders to be studied as organized configurations of symptoms, impairment and biological or environmental determinants.

A distinctive feature of the Ad-Infer platform is its family investigation framework. During the interview, participants can be asked about relevant health information in family members and, critically, the platform is designed to facilitate direct assessment of relatives through the same standardized interview procedure. The system can proceed through family roles such as spouse/partner, mother, father, siblings, children, and grandparents, creating a structured family map that supports epidemiological and familial aggregation studies.

This approach serves two complementary scientific goals. First, it increases the likelihood of accessing multiple family members over time, because the interview flow creates explicit pathways to invite relatives. Second, it enables a consistent, standardized phenotyping of each participating individual, allowing the study of familial clustering, intergenerational transmission, and shared environmental exposures using comparable measures.

Because Sleep-EVAL interviews can be deployed on the web and delivered in multiple languages, participation is not constrained by geography or language. Family members can live in different regions or countries and still complete the same interview in their preferred language, enabling truly international family-based epidemiology.

The following schematic summarizes how population-based interviews, clinical knowledge and hybrid AI reasoning are integrated to produce standardized diagnoses and phenotypes suitable for large-scale epidemiological research.

The scientific aims of the Sleep-EVAL program are intentionally broad, because the platform is designed as an infrastructure for population-based clinical intelligence. The core aims include:

• To characterize sleep disorders in the general population using AI-driven, clinically grounded diagnostic procedures
• To develop standardized, multidimensional phenotypes of sleep, psychiatric and neurological disorders
• To quantify prevalence, incidence and longitudinal evolution of disorders and key symptom dimensions
• To analyze comorbidity structures linking sleep disorders with psychiatric and medical diseases
• To identify environmental, behavioral and medical risk factors associated with sleep and mental disorders
• To study treatment pathways, recognition of disorders in the community, and determinants of health-care utilization
• To assess functional impairment, quality of life and broader public health consequences
• To examine familial aggregation and intergenerational transmission using the family investigation framework
• To enable international and multilingual studies through web-based deployment and standardized phenotyping

Together, these aims support an integrated program that links clinical reasoning, phenotyping, family structure and epidemiological inference. This is essential for modern sleep medicine and psychiatry, where the goal is not only to measure prevalence, but to understand how symptoms, comorbidity, trajectories and familial patterns organize disease expression in the community.

To fulfill these objectives, a new investigation tool was created: Sleep-EVAL. Sleep-EVAL is a knowledge-based expert system specialized in sleep medicine and psychiatry, capable of conducting structured clinical interviews and generating standardized diagnoses and phenotypes. Its knowledge base integrates sleep disorders, psychiatric conditions, medical diseases and pharmacological treatments, enabling comprehensive assessment within a standardized framework.