Lead Investigator: Yang Li, Renmin University of China
Title of Proposal Research: Missing data considerations for patient reported outcome measures in randomized controlled trials
Vivli Data Request: 9910
Funding Source: None
Potential Conflicts of Interest: None
Summary of the Proposed Research:
Background
Patient-reported outcome (PRO) measures are increasingly included in clinical trials to provide insights into patients’ perspectives on treatment, including their symptoms and quality of life. These measures typically consist of multiple items grouped into various domains. A total score or domain score, calculated according to the questionnaire’s scoring rules, is often used as an endpoint in clinical trials. However, missing data in PRO measures, which can occur due to patients skipping some questions (missing item) or entire questionnaires (missing visit), is a common issue, especially when data is collected over multiple visits.
Missing data can significantly impact the calculation of domain and total scores. Some scoring manuals allow for missing items if more than 50% of items are completed, using methods such as the “half rule” or item mean imputation. However, many manuals do not provide guidance for calculating scores when there are missing items.
Research Necessity
There are two primary approaches to handling missing PRO data: imputing missing data at the item level or at the domain score level. Item-level imputation uses observed items and other predictors like age or disease severity to estimate missing values. Domain-level imputation, on the other hand, treats the domain score as missing and does not rely on individual items, using alternative predictors instead.
The FDA advises specifying imputation methods in analysis plans but does not recommend specific approaches. Previous research, such as that by Rombach et al., has explored the performance of item-level versus domain-level imputation in single-visit scenarios. However, there has been limited progress in understanding PRO endpoints within a longitudinal framework, where data is collected over extended periods and missing data patterns include both missing items and missing visits.
Study Overview
We will conduct a simulation study to evaluate various imputation methods for missing PRO data within a longitudinal framework. This study will compare multiple imputation (MI) on item scores, MI on domain scores, and item mean imputation methods. The PALOMA-2 trial has been selected for this research due to its use of widely-recognized PRO endpoints, such as the Functional Assessment of Cancer Therapy (FACT)-Breast and EuroQOL 5 dimensions (EQ-5D), and its collection of PRO data at multiple time points, encompassing both missing item and missing visit patterns. The trial’s sample size of 666 participants allows us to assess the performance of these methods across different sample sizes.
This research aims to provide trial statisticians with a comprehensive evaluation of imputation methods, leading to improved handling of missing PRO data. Better handling of missing data results in more accurate information on patients’ conditions, contributing to a deeper understanding of trial results and ultimately enhancing patient treatment outcomes.
Requested Studies:
A RANDOMIZED, MULTICENTER, DOUBLE-BLIND PHASE 3 STUDY OF PD-0332991 (ORAL CDK 4/6 INHIBITOR) PLUS LETROZOLE VERSUS PLACEBO PLUS LETROZOLE FOR THE TREATMENT OF POSTMENOPAUSAL WOMEN WITH ER (+), HER2 (-) BREAST CANCER WHO HAVE NOT RECEIVED ANY PRIOR SYSTEMIC ANTI CANCER TREATMENT FOR ADVANCED DISEASE
Data Contributor: Pfizer Inc.
Study ID: NCT01740427
Sponsor ID: A5481008