We tested this interaction because the effect on prognosis of the severity of disease at baseline, expressed in the scores of the questionnaires and substitute questions, may depend on the treatment received. For the substitute questions that were at least as good as their questionnaires in predicting outcome, the test-retest reliability was assessed by using the Pearson correlation coefficient. It is suggested that a reliability coefficient of 0.7 or higher

is acceptable (Cicchetti 1994). As the natural SB431542 concentration course of sciatica is favourable, we chose the measures at 3 and 6 weeks follow-up for calculation of the test-retest correlations as these were assumed to be the least influenced by the favourable natural course of sciatica. Also, the participants were already used to the trial setting, the treatment determined by randomisation and to answering the substitute questions and questionnaires. Table 1 shows the baseline characteristics of the 135 participants and the outcomes at 1 year follow-up; 18 participants

were lost to follow-up or had incomplete data at 1 year, necessitating carry forward of the last available score. Testing the correlation between the Tampa Scale for Kinesiophobia and its unique substitute question at baseline resulted in a correlation coefficient of 0.46 (Table Nutlin-3a mouse 2). Table 3 shows the explained variation of the three separate models on global perceived effect and severity of leg pain at 1 year follow-up, as well as the p values of the contribution of the substitute question and the original questionnaire to their models. Both the Tampa Scale for Kinesiophobia and its substitute question had prognostic properties to predict global perceived effect and pain at 1 year followup. The substitute question explained more of the variation in pain severity in the leg than did the Tampa Scale for Kinesiophobia. The interaction term between treatment and the score of the substitute question contributed significantly to the pain model. The mean score of the substitute

question at 3 weeks follow-up was 3.7 (SD 2.8) and at 6 weeks follow-up was 3.6 (SD 2.9). The Pearson correlation coefficient between these scores of the substitute questions was 0.65, indicating acceptable test-retest reliability, taking into Rolziracetam account that the reliability coefficient is directly dependent on the number of items. In classical test theory, a test with a limited number of items has a lower reliability, which limits the obtainable reliability for a single question (Cronbach 1990). The correlation coefficient between the Roland Morris Disability Questionnaire and its unique substitute question was 0.32 (Table 2). Table 4 shows the explained variation of the models predicting global perceived effect and pain. The substitute question did not have a prognostic ability to predict global perceived effect and pain severity in the leg at 1 year follow-up.