Evaluating a complex intervention with a single outcome may not be a good idea: an example from a randomised trial of stroke case management

Design: a re-analysis of stratified, balanced, randomised clinical trial. Setting: five university-affiliated acute-care hospitals in Montreal, Quebec, Canada. Participants: a total of 190 persons (mean age 70 years) returning home directly from the acute-care hospital following a first or recurrent stroke with a need for health-care supervision post-discharge because of low function, co-morbidity or isolation. Intervention: for 6 weeks following discharge a nurse case manager delivered, depending on need, over 50 different nursing interventions (range 2-15 per person), which targeted physical, emotional and psychological impairments, role participation restrictions and health perception. Measurements: seven of the SF-36 subscales were used to measure the targeted constructs, at the post-intervention and 6 month evaluations. Seven binary response variables were created with a change of 10 points the criterion for individual response. Generalised estimating equations, equivalent to a logistic regression for multiple outcomes, were used. Results: the odds of responding to one or more outcomes was 41% greater in the intervention group than in the control group [odds ratio (OR): 1.41; 95% confidence interval (CI): 1.11-1.79]. Conclusion: an analysis considering the complexity of the intervention and outcomes targeted indicated effectiveness of the nurse case-management post-stroke, whereas the traditional one outcome analysis did not.