Introduction: the problem of unexplained variations in outcome Studies evaluating health services typically use patient-based indicators, such as morbidity or mortality, to assess quality of care or efficacy of treat­ ment. However, as earlier chapters have argued, interpreting such outcomes is not straightforward. A standard way of improving the interpretability of these data is to control for case mix, since differences in the characteristics of study populations — including age, gender and disease severity — can confound the results. Crude case-mix adjustment using merely age and sex is inadequate, as was demonstrated by a study of the effects of introducing a stroke unit in Scotland.1 Unadjusted data clearly showed the benefits of stroke-unit care, and the effects were less marked but still significant when age and sex were taken into account. However, all differences in mortality and morbidity were eliminated when adjustment was made for case-severity measures, including incontinence in the first week, coma, motor impairment and social class.