The impact of missing data on estimation of health-related quality of life outcomes in a longitudinal clinical trial
Keywords: treatment effectiveness, longitudinal study, missing data, time-to-dropout, joint modeling, pattern mixture model
Purpose: Missing responses for health-related quality of life (HRQL) outcomes are common in clinical trials and may introduce bias as such data are often not missing at random. We evaluated the missingness (dropout) effect when comparing treatment groups (I vs. II) in a longitudinal randomized trial. Methods: We analyzed the Functional Assessment of Cancer Therapy Trial Outcome Index (TOI) change over 12 months for newly diagnosed patients with chronic myeloid leukemia. HRQL assessment was expected at baseline and months 1, 2, 3, 4, 5, 6, 9 and 12. We defined completers as those with baseline and month 12 TOI, and dropouts as all others as long as they had a baseline score. We defined censoring time as the time interval between baseline and scheduled month 12 visiting date, and approximate time-to-dropout as the time interval from baseline to the midpoint between date of last reported TOI and scheduled next visit date. A mixed-effects model (MEM) was first built to assess treatment effect, a joint model (JM) and a pattern mixture model (PMM) were then built to account for non-ignorable dropout. To fulfill the assumption about normality and heterosckedasticity, TOI scores were transformed to their square root. Results: A MEM fit the data well and revealed that TOI in group II declined significantly, whereas it remained constant in group I. Between-group differences at each visit were significant (p<.001) except for baseline. Completion rate at month 12 was different between group I and II (82.8% vs. 63.5%, p<.001); 90% time-to-dropout (month) was different also (9.7 vs. 4.2, p<.001); patients in group II were more likely to dropout (HR: 12.7, 95%CL: 10.7-15.1, p<.001). A JM simultaneously modeling longitudinal data and dropout time via a shared random effect vector with a 4th-degree polynomial time generated similar estimates (<10% change in magnitude mostly) as the separate longitudinal and survival sub-models, the significant association parameter (slope:1.99, 95%CL: 0.09-3.88, p=0.040) indicated that TOI change was positively associated with dropout. A PMM with non-significant 7th-degree polynomial time generated dramatically changed parameter estimates of 5th-, 6th- and 7th-degree polynomial time (>100% in magnitude) comparing to MEM, PMM also failed to fit a 4th-degree polynomial time model. Conclusion: treatment effect was different between groups; JM detected a significant effect of dropout on the longitudinal TOI; single PMM may misspecify dropout.