The inter-rate reliability of a generic measure of severity of illness

There has been increasing interest in the development of measures to quantify baseline severity of illness and thus provide a more meaningful interpretation of health outcomes.

We aimed to assess the inter-rater reliability of a generic measure of illness severity, the Duke University Severity of Illness (DUSOI) checklist.

Selected general practices and hospital outpatient departments across Tyneside and Humberside in the UK were used as the setting. Thirty-three clinicians were posted copies of the same set of 14 patient records and asked to rate the severity of illness of each patient using the DUSOI. The subjects were 27 GPs, three consultant chest physicians and three diabetologists. The main outcome measures were: (i) intraclass correlation coefficients used to express clinician agreement upon severity scores; and (ii) a 'gold standard', constructed in order to assess clinician agreement upon the health problems identified in the patient record to be considered as constituents of overall severity.

It was found that the degree of inter-rater reliability of severity scoring was satisfactory, with an intraclass correlation coefficient of 0.43. In terms of identification of problems to be rated, there was considerable agreement between raters and a specially compiled 'gold standard' of health problems.

The DUSOI has potential use in routine clinical practice, but strategies should be developed in order to maximize the reliability of rating illness severity on such a generic measure.