The Oxford orthopedic scores (Oxford Hip and Knee scores) are relatively short (12-item) PRO measures designed and developed to measure pain and function in those undergoing Total Hip or Knee replacement surgery. They have found widespread use globally in diverse applications, from national orthopedic outcome programs to routine clinical care assessment. As a patient progresses through an episode of care, the respondent may encounter numerous providers and organizations, and, if PRO measures are administered at each step, the burden, redundancy, and resulting frustration may well hinder treatment more than it helps. This difficulty has led to the creation of shorter questionnaires, some of which are abridged (short-form) versions of existing PRO measures and some of which are entirely new. In either case the new form may not be equivalent in validity, accuracy, or characteristics measured to years of data collected with the original form. We will demonstrate that our CAT system reduces the number of responses required, although it remains consistent with the original form. The CAT systems discussed here were devised using machine learning techniques: existing data (a “training set”) was analyzed to discover patterns that permit accurate scores to be predicted for a new patient without requiring that all questions be asked. For the Oxford Knee and Hip the training sets were 8819 and 4583 cases, respectively; the resulting CATs were tested on 5622 and 3218 independent instances, respectively. The correlation between CAT and full Oxford Knee / Hip scores was 0.98, and the CAT required 42% fewer questions for the average patient The CAT system we have developed, by intelligent selection of questions tailored to the individual patient, provides for a reduced respondent burden whilst maintaining equivalent measurement properties to the original version.