Why are missing quality of life data a problem in clinical trials of cancer therapy?

Abstract

Assessment of health related quality of life has become an important endpoint in many cancer clinical trials. Because the participants of these trials often experience disease and treatment related morbidity and mortality, non-random missing assessments are inevitable. Examples are presented from several such trials that illustrate the impact of missing data on the analysis of QOL in these trials. The sensitivity of different analyses depends on the proportion of assessments that are missing and the strength of the association of the underlying reasons for missing data with disease and treatment related morbidity and mortality. In the setting of clinical trials of cancer therapy, the assumption that the data are missing completely at random (MCAR) and analyses of complete cases is usually unjustified. Further, the assumption of missing at random (MAR) may also be violated in many trials and models appropriate for non-ignorable missing data should be explored. Recommendations are presented to minimize missing data, to obtain useful documentation concerning the reasons for missing data and to perform sensitivity analyses.

Quality of life as a clinical trial endpoint: determining the appropriate interval for repeated assessments in patients with advanced lung cancer

“…Both investigators and patients may lose interest in QOL assessment when it is performed at intervals too often to be perceived as being relevant. Problems arise in serial measurement when observations are not missing at random [17,1,7,10,34]. In a 673-patient data set from two combined trials from 30 cancer centers for patients with advanced non-small cell lung cancer (NSCLC) [6,9], the loss of patients from evaluation was not random and correlated highly with the baseline patient-determined symptom burden and QOL [17].…”

“…Many statistical methods have been suggested for the problem of missing observations using QOL as an endpoint, including imputation [34]. Nonetheless, a better strategy is to plan at the outset to avoid loss of data [17,1,7,10,30].…”

Analysis of messy longitudinal data from a randomized clinical trial

“…Although the missing completely at random assumption was made after exploring our data, it may not be appropriate if the missing mechanisms were di!erent across the two treatments, for example, if the distributions of number of forms completed, patient dropout and missing items on forms were obviously di!erent across the two treatments. Some authors, for example Fairclough et al [12,13] and Troxel [14], have considered non-random, non-ignorable and informative missing data in mixed models for quality of life studies. Their analyses were based on normally distributed measures.…”

“…Clinicians were asked to complete a symptom assessment form before starting treatment, then at each attendance 2-weekly for the ACE#G-CSF group and 3-weekly for ACE group after each cycle of chemotherapy, and then monthly to 6 months. If we denote the "rst cycle of chemotherapy to be given at week 0, then for the ACE#G-CSF arm the symptom assessment forms were completed at week 0 (pre-treatment), weeks 2, 4, 6, 8, 10, 12 and then monthly to 6 months and for the ACE arm at week 0 (pre-treatment), weeks 3, 6,9,12,15,18 and then monthly to 6 months. Ideally, during these "rst 18 weeks clinicians would have completed eight clinical reports for each patient in the ACE#G-CSF arm and seven for each patient in the ACE arm.…”

Influence of Gender and Life Stressors on Longitudinal Depression Treatment Outcomes Among Older Primary Care Patients

“…Research suggested the majority of missing data techniques are subjective and biased the missing values of the research participants as overly healthy (Engels & Diehr, 2003). However, when working with older adults, missing data are likely related to deteriorating health (Andrew, Mitnitski, Kirkland, & Rockwood, 2012;Engels & Diehr, 2003;Fairclough, Peterson, & Chang, 1998). Research suggested the most accurate missing longitudinal data method when working with a sample of older adults is the last and next method or the last observation carried forward (Engels & Diehr, 2003).…”