An introduction to computer diagnosis in psychiatry

The concept of computer diagnosis in medicine is not new. As early as 1971, a computer program was dem onstrated to be more accurate than a senior clinician at diagnosing acute abdominal pain before surgery. In psychiatry, however, the problems surrounding diag nosisand classification are more complex than in other branches of medicine, depending as they do on the clinical interview and some agreed classificatory system in the absence of external validating criteria. Reliability has been improved by the application of standardised interview techniques and by the use of operationalised diagnostic criteria but such tools are lengthy and their use requires specialist training. Consequently they tend to be reserved for research purposes rather than routine clinical use. The potential contribution of computer technology to the vexed question of psychiatric diagnosis is here evaluated.

Interviewer administered computer questionnaires Computer programs able to generate standard ised psychiatric diagnoses from the data supplied by an interviewer have been available for some time. The CATEGO program uses information obtained from the Present State Examination (PSE), the most recent being CATEGO-5 for use with the Schedules for Clinical Assessment in Neuropsychiatry (SCAN). In addition to making ICD-10 diagnoses this modified program can pro duce DSM-III-R and Research Diagnostic Criteria (RDC) diagnoses (Wing et al, 1990). SCAN also contains a glossary which aids clinical examin ation providing definitions independent of any algorithms. Similarly, AGECAT (Automated Geri atric Examination) employs both decision trees or 'top down' and combination and comparison of levels of symptoms or 'bottom up' classifications to obtain diagnoses. These are examples of com puter diagnostic programs based on determinis tic diagnostic systems. Statistical techniques have been used, but have not received such wide acceptance, proving less reliable when compared with a clinician. Most current diagnostic pro grams employ a decision tree method.
A consultation aid currently being developed is 'Psyxpert' (Overby, 1987). This is an example of an expert system which does not require special ist training or the administration of a standard ised interview. Psyxpert asks the user a set of initial questions, the answers to which are the facts upon which all subsequent inference pro cedures are based. The program uses the DSM-III decision tree and produces a diagnostic report with certainty factors and treatment recom mendations. It is, however, limited in that it addresses only psychotic disorders.

Self-administered computer questionnaires
Computer-patient interviews eliminate observer bias both in eliciting and interpreting infor mation. Computer-elicited case histories have been found to be at least as accurate as clinicianelicited histories and computers found to be better than clinicians at predicting further suicide attempts. It may be that patients do not perceive computers as judgemental.
The next step has been the development of computer programs to administer the interview and to analyse the data. This appears to have several advantages. As well as reducing observer bias and information transfer errors, it is pos sible to standardise the diagnostic rules based on an agreed classificatory system. The earliest selfadministered diagnostic program was based on an extensively modified version of the Hamilton Depression Rating Scale (HDRS), and was found to discriminate between patients and controls when a suitable cut-off score was chosen. A later version of the program called INTERACT has been used in a variety of settings including psy chiatric hospitals, drug trials, general hospitals and general practice (Ancill et al, 1985).
In practical terms computers are well suited to branching and complex diagnostic algorithms and thus it is possible to reduce the number of unnecessary questions as well as to provide a comprehensive diagnostic interview. This tech nique is utilised in the more recent programs.
Computer diagnostic programs based on stan dardised interviews and operationalised diagnos tic criteria include the computerised Diagnostic Interview Schedule (cDIS), which is claimed to be as reliable as the standard DIS at generating Psychiatric Bulletin (1994), 18, 73-74 some DSM-IIIR diagnoses. However it is not comprehensive in its coverage of diagnoses. The cDIS has recently been updated and is available commercially (Levitan et al, 1991).
The screening version of the DIS, the DISSI, has been computerised in two forms, one a selfadministered version (cDISSI) and the other a version which prompts the interviewer to ask all the questions (pDISSI). These have been com pared with the DIS and the computerised DIS and found to have acceptable validity but no significant saving of administration time.
Although some self-administered computer interviews have been shown to have acceptable reliability and validity, for certain groups of patients they are obviously inappropriate. Cer tain abnormalities in mental state appear to affect scoring. Patients with severe retarded depression were found to have an artificially low score on the INTERACT program and the effec tiveness of self-administered questionnaires at eliciting current psychotic symptoms has yet to be demonstrated.

Computer diagnosis in primary care
In terms of psychiatric morbidity in primary care, it is known that a considerable proportion of cases of affective disorder remains unrecognised and there is evidence that knowledge of both the precise diagnosis and severity of depression is required to select appropriate treatment. PROQSY (Programmable Questionnaire Sys tem) has been developed as a screening instru ment to determine the likelihood of neurotic disorder in primary care (Lewis et al, 1988). It is based on the Clinical Interview Schedule (CIS) and its reliability and validity are comparable to a clinician administered CIS. The Barnet Program is a self-administered interview for the screening and diagnosis of mood disorders, currently being field tested in general practice. The program is based on DSM-III-R diagnostic algorithms and also provides a measure of the severity of depression using a self-administered version of the HDRS. Other potential applications are as a diagnostic aid in the out-patient clinic and as a way of improving the standardisation of entry criteria for clinical trials of treatments for depression.

Acceptability to patients
An important consideration is the acceptability of such computer administered interviews to patients. Patients found the cDIS generally easy to use and operate and after the cDIS felt their level of expertise in computing had improved. General questioning following both the INTER ACT program and the PROQSY program have suggested favourable attitudes.

Conclusion
Although computer-aided diagnosis is routinely employed by some researchers it has not gained wider acceptance, despite the generally acknowl edged need to improve diagnostic reliability. Computer technology has certainly not proved to be "the fourth quantum advance in psychiatry" as optimistically predicted in the 1960s. Some of the diagnostic programs described above have been demonstrated to have acceptable reliability, to save clinician time and to be acceptable to patients and yet their application has not been extended beyond the teams who developed them. One probable explanation for this apparent lack of enthusiasm, particularly among British psy chiatrists, is the limited access most clinicians have had to computer systems in their work place. With the introduction of computer systems in most hospitals and the pressure for accurate audit in psychiatry, now is the time to seize the opportunities provided by the new technology.