Introduction
No validated reliable method of assessing the severity of prescribing errors exists. The aim of this study was to develop a method which would be reliable and valid for assessing prescribing errors and which could also be used to evaluate related pharmacists' interventions. Specific objectives were to explore the influence of the judges' profession on the reliability of the severity scores, to determine the minimum number of judges required to obtain reliable mean scores and to investigate the scoring method's validity.

Methods
A stratified random sample of eight pharmacists, eight doctors and eight nurses was selected from two hospitals. These participants judged the potential harm associated with fifty prescribing errors, both before and after pharmacists' interventions to correct them. Twenty-five of these cases were also scored on a second occasion. The scores were marked on a visual analogue scale numbered zero to ten with anchors of 'no harm' and 'death'. The reliability of the judges' scores was investigated using generalisability theory. Potential sources of variability were the case itself, the judge, the judge's profession and the scoring occasion. A generalisability coefficient of 0.8 was considered to represent an acceptable level of reliability. Seventeen prescribing errors with known outcomes were included to explore the validity of the scoring method.

Results
Generalisability studies showed that most of the score variance was attributable to the cases themselves. The reliability of the judges' mean scores was influenced by both the number of judges and the number of judges' professions, but the number of scoring occasions had little influence. In order to reach a generalisability coefficient of above 0.8, at least two judges of each of two professions (selected from doctors, nurses and pharmacists) would be required to assess prescribing errors and two judges of each of the three professions to evaluate the related pharmacists' interventions. The mean scores for the prescribing errors with known outcomes did not consistently reflect the severity of the outcomes.

Conclusions
A visual analogue scale was acceptable to doctors, nurses and pharmacists for assessing the severity of prescribing errors and related pharmacists' interventions. The method was found to be reliable if at least two judges from each of two professions assess prescribing errors and two judges of each of the three professions evaluate pharmacists' interventions. However involving all three professions in both assessments would facilitate ownership of the results and encourage multi-disciplinary management of the prescribing process. The results did not confirm the validity of the method, but it is unknown whether the scoring method itself is not valid or if the approach taken to assessing its validity was inappropriate. The problems of assessing validity where no 'gold standard' exists were highlighted and will be discussed in more detail.