Data mining for response shift patterns in multiple sclerosis patients using recursive partitioning tree analysis

To examine evidence of QOL response shift in patients with multiple sclerosis (MS) using recursive partitioning tree analysis
(RPART) technique.

Subjects: MS patients from the NARCOMS registry assessed an average of 6 times at a median interval of 6 months. Outcomes:
SF-12v2 Physical & Mental Component Scores (PCS, MCS). Covariates: Patient-determined disease steps, Performance Scales, and
symptomatic therapies. RPART trees were fitted separately by 3 disease-trajectory groups: (1) relapsing (n = 1,582); (2) stable (n = 787); and (3) progressive (n = 639). The resulting trees were interpreted by identifying salient terminal nodes that showed the unexpected quantitative patterns of contrasting MCS and PCS scores (e.g., PCS deteriorates but MCS is stable or improves), using a minimally important
difference of at least 5 points on the SF-12v2. Qualitative indicators of response shift were different thresholds (recalibration), content (reconceptualization), and order (reprioritization)
of disability domains in predicting PCS change by group.

Overall, 20% of patients demonstrated response shift quantitatively, with 10% in the “progressive” cohort, 8% in the “relapsing”
cohort, and 2% in the “stable” cohort. RPART trees differed qualitatively across disease-trajectory groups in patterns suggestive
of recalibration, reprioritization, and reconceptualization. Disability subscales, but not symptom management, distinguished
homogenous groups.

PCS and MCS change scores are obfuscated by response shifts. The contingent true scores for PCS change scores are not comparable
across patient groups.