Using customized PRO profiles to evaluate expected patient quality-of-life outcomes.

Abstract

e18607 Background: Patient Reported Outcomes (PROs) are routinely employed in patient-focused drug development programs. However, PRO data are often under-utilized. Better approaches could employ trial data to describe expected patient outcomes. The objective of this work was to present a method for leveraging PRO-based patient profiles to project and thereby anticipate expected patient outcomes. Methods: A model-based approach to analyzing PRO scores allows for a more tailored approach to evaluating outcomes of interest. To demonstrate this approach, we analyzed PRO data from a Phase 3 trial of patients with prostate cancer randomized to either Hypofractionated 3D-CRT/MRT (experimental arm) or Conventionally Fractionated 3D-CRT/MRT (comparator arm). The outcome of interest was the difference between treatment arms on the Expanded Prostate cancer Index Composite (EPIC) urinary domain scores, defined on the sum of responses to 12 items. A longitudinal regression model was fit to EPIC scores at 6, 12, 24, and 60 months post-baseline. Patient profile variables (baseline EPIC urinary scores, treatment arm, age, race, clinical T-stage, and Gleason score) were included as predictors. For a particular patient profile of interest, expected EPIC urinary score at 6 months was estimated, as well as associated 95% confidence intervals, which allowed for an understanding of the range of likely outcomes for that particular patient profile. Results: n = 550 patients in the experimental arm and n = 542 patients in the comparator were analyzed. The difference in EPIC urinary scores between treatment arms at 6 and 12 months were computed, with higher scores reflecting greater symptom severity. The model estimated a difference of 2.12 points (p = 0.009) at 6 months and 2.16 points (p = 0.013) at 12 months, favoring the experimental arm. To illustrate how customized patient profiles can be used to better understand patient outcomes, a single profile is presented. At baseline, a patient receiving the experimental treatment, EPIC urinary score 98.0, age 57, white, clinical T-stage T2, Gleason score 6, could be expected to report an EPIC urinary score of 94.3 (92.06, 96.00) at 6 months based on the statistical model. In the observed trial data, one patient with this profile reported an EPIC urinary score of 93.8 at 6 months. The expected EPIC urinary score overestimated the observed score by only 0.53%, indicating the accuracy of the expected EPIC urinary scores. This corresponds to approximately one category decrease on two of the 12 domain items (i.e., a decrease in symptom severity). Conclusions: Well-established statistical methods can lead to improved clinician understanding of expected outcomes specific to a given patient profile. This would facilitate communication between clinician and patient regarding anticipated oncology treatment outcomes. Recommendations to support the use of these methods should be considered.