Missing outcomes data in clinical trials can be detrimental to identifying important treatment effects because power is reduced and uncertainty is increased. Although missingness at the scale level for patient-reported outcomes (PROs) (e.g., due to attrition) is a considerable challenge to measurement in longitudinal clinical trials, missingness at the item level for PROs (e.g., due to omission) can be more easily overcome and a reliable scale score calculated. The FDA PRO Guidance states that the maximum tolerable number of missing item-level responses should be determined during the instrument development process, but no particular method is advocated, and instrument developers often recommend arbitrary guidelines. Although a number of methods exist for examining the effect of missing data on scale precision, one simple approach is to calculate Cronbach’s coefficient alpha sequentially as each item is deleted from the item set. The order in which items are removed from the item set is based on deleting the item with the largest contribution to alpha (i.e., alpha-if-item-deleted). When Cronbach’s alpha for the set of remaining items falls below an a priori identified threshold (e.g., 0.70), the number of items deleted from the scale minus one is the maximum number of responses that can be missing for a scale score to be reliably calculated for a subject. We explored this approach with several validated instruments and found that the developer’s guidelines are often stricter than the alpha-if-item-deleted method. Broader application of the Cronbach’s alpha approach would result in fewer missing PRO scale scores, increased statistical power, reduced uncertainty, and additional information with which to assess treatment effects.
