Impact of missing data due to drop‐outs on estimators for rates of change in longitudinal studies: a simulation study

摘要: Many cohort studies and clinical trials are designed to compare rates of change over time in one or more disease markers several groups. One major problem such longitudinal is missing data due patient drop-out. The bias efficiency six different methods estimate changes with incomplete observations were compared: generalized estimating equation estimates (GEE) proposed by Liang Zeger (1986); unweighted average ordinary least squares (OLSE) individual (UWLS); weighted OLSE (WLS); conditional linear model (CLE), a covariate type Wu Bailey (1989); random effect (RE), joint multivariate RE (JMRE) estimates. latter method combines for the underlying pattern marker log-normal survival informative drop-out process. performance these presence completely at (MCAR), (MAR) non-ignorable (NIM) compared simulation studies. Data generated under effects parameter values derived from realistic examples HIV infection. Rates drop-out, assumed increase time, allowed be independent depend either only on previous both current values. Under MACR all yielded unbiased group mean between-group difference. However, cross-sectional view GEE resulted seriously biased MAR NIM ranged 30 per cent 50 cent. degree increases severity non-randomness proportion data. MCAR other five performed relatively well. JMRE efficient(that is, had smaller variance) than UWLS, WLS CL NIM, particularly tended underestimate rate (bias approximately 10 cent). better terms (3-5 cent) giving most efficient Given that key variables related progression, likely MAR. Thus, may not appropriate analysing potential biases require greater recognition reports Sensitivity analyses assess drop-outs inferences about target parameters important.