Sleep-EVAL Aims

Written by Maurice M. Ohayon, MD, DSc, PhD

When the first epidemiological survey was launched in 1992, we were looking for an assessment tool that could be used by interviewers with little knowledge about sleep disorders.

We also wanted a tool that allowed the broad coverage of sleep habits, including sleep/wake schedule and sleep hygiene.
The tool also had to permit the identification of mental disorders most frequently associated with sleep problems; and, most of all, the tool should allow the formulation of sleep diagnoses according to several classifications.
Tools that met all these requirements were nonexistent.
Questionnaires to assess sleep disorders covered only some disorders.
None of them was designed to allow a differential diagnosis making.


- in 1983, creation of Adinfer (©M Ohayon, 1983) a level 0+ expert system devoted to the assessment of psychiatric disorders.
- from 1983 to 1991, Adinfer went through several changes to increase its diagnostic abilities.
- in 1990, creation of Sleep-EVAL (©M Ohayon, 1990), a level-2 expert system endowed with a causal reasoning mode.

The integration of a neural network in an expert system gives "reasoning" aibilities that are the closest to human reasoning at this time.

Indeed, neural networks are able to find solutions to problems that usually require human observations or thought processes.

When used in diagnostic processes, they allow incorporation of subjectivity in answers provided by the subject and manage the resulting uncertainty in the assessment of a disorder.


Sleep-EVAL was developed with clear objectives in mind:

  • to improve the quality of collected data,
  • to find new ways to analyze risk factors associated with some abnormalities, and finally,
  • to provide some kind of validation of the usefulness of existing classifications such as the DSM-IV (APA, 1994), the International Classification of Sleep Disorders (ASDA, 1990, 1997), and the International Classification of Disease (ICD-10, WHO).

The use of fuzzy logic reasoning managed by a neural network was allowing the inclusion of the richness of clinical experience in a tool that can be used by inexperienced interviewers.