In the presence of dropout, valid statistical inferences based on longitudinal data can, in general, only be obtained from modeling the measurement process and the dropout process simultaneously. Many models have been proposed in the statistical literature, most of which have been formulated within the framework of selection models or pattern-mixture models. In this paper, we will use continuous data from a longitudinal clinical trial with a 24% dropout rate to illustrate some of the models frequently used in practice. We emphasize the underlying implicit assumptions made by the different approaches, and the sensitivity of the results with respect to these assumptions. The merits and drawbacks of the procedures are extensively discussed and compared from a practical point of view.
LittleRJARubinDBStatistical Analysis with Missing Data. New York: Wiley; 1987.
6.
WuMCCarrollRJEstimation and comparison of changes in the presence of informative right censoring by modeling the censoring process. Biometrics. 1988;44:175–188.
7.
WuMCBaileyKRAnalysing changes in the presence of informative right censoring caused by death and withdrawal. Stat Med1988;7:337–346.
8.
WuMCBaileyKREstimation and comparison of changes in the presence of informative right censoring: conditional linear model. Biometrics. 1989;4:939–955.
9.
DigglePKenwardMGInformative drop-out in longitudinal data analysis. Appl Stat. 1994;43:49–93 [With discussion].
10.
LittleRJAModeling the drop-out mechanism in repeated-measures studies. J Am Stat Assoc. 1995;90:1112–1121.
11.
HoganJWLairdNMMixture models for the joint distribution of repeated measures and event times. Stat Med.1997;16:239–257.
12.
MolenberghsGKenwardMGLesaffreE.The analysis of longitudinal ordinal data with non-random dropout. Biometrika. 1997;84:33–44.
13.
LairdNMDiscussion to Diggle, PJ and Kenward, MG: Informative dropout in longitudinal data analysis. Appl Stat. 1994;43:84.
14.
de BackerMde KeyserPde VroeyCLesaffreE.A 12-week treatment for dermatophyte toe onychomycosis: terbinafine 250mg/day vs. itraconazole 200mg/day—a double-blind comparative trial. Br J Dermatol. 1996;134:16–17.
15.
RobertsDTPrevalence of dermatophyte onychomycosis in the united kingdom: Results of an omnibus survey. Br J Dermatology. 1992;126Suppl. 39:23–27.
16.
RubinDBInference and missing data. Biometrika. 1976;63:581–592.
17.
LiangKYZegerSLLongitudinal data analysis using generalized linear models. Biometrika. 1986;73:13–22.
18.
KenwardMGMolenberghsG.Likelihood based frequentist inference when data are missing at random. Stat Sci.1998;13:236–247.
19.
VerbekeGMolenberghsG.“Linear mixed models in practice: A SAS-oriented approach.” Number 126 in Lecture Notes in Statistics. New York: Springer-Verlag; 1997.
20.
LairdNMWareJHRandom-effects models for longitudinal data. Biometrics. 1982;38:963–974.
21.
SmithDMRobertsonBDigglePJObject-oriented software for the analysis of longitudinal data in s. Technical Report MA 96/192. The United Kingdom: Department of Mathematics and Statistics, University of Lancaster; 1996.
22.
RubinDBDiscussion of the paper ‘Informative drop-out in longitudinal data analysis’ by Diggle and Kenward. Appl Stat. 1994;43:80–82.
23.
HoganJWLairdNMModel-based approaches to analysing incomplete longitudinal and failure time data. Stat Med.1997;16:259–272.
24.
KenwardMGSelection models for repeated measurements with nonrandom dropout: an illustration of sensitivity. Stat Med.1998;17:2723–2732.
25.
VerbekeGLesaffreEBrantLJThe detection of residual serial correlation in linear mixed models. Stat Med.1998;17:1391–1402.
26.
LesaffreEAsefaMVerbekeG.Assessing the goodness-of-fit of the Laird and Ware model, an example: The Jimma infant survival differential longitudinal study. Stat Med.1999;17:835–854.
27.
LittleRJAPattern-mixture models for multivariate incomplete data. J Am Stat Assoc. 1993;88:125–134.
28.
GlynnRJLairdNMRubinDBMultiple imputation in mixture models for nonignorable nonresponse with follow-ups. J Am Stat Assoc. 1993;88:984–993.
29.
de GruttolaVTuXMModelling progression of CD4-lymphocyte count and its relationship to survival time. Biometrics. 1994;50:1003–1014.
30.
DempsterAPLairdNMRubinDBMaximum likelihood from incomplete data via the EM algorithm. J Royal Stat Society, Series B.1977;39:1–38.
31.
MichielsBMolenberghsGLipsitzSRSelection models and pattern-mixture models for incomplete categorical data with covariates. Biometrics. 1999;55:978–983.
VerbekeGMolenberghsGThijsHLesaffreEKenwardMGSensitivity analysis for non-random dropout: A local influence approach. Biometrics. 2000;56.
34.
MolenberghsGVerbekeGThijsHLesaffreEKenwardMGMastitis in dairy cattle: Local influence to assess sensitivity of the dropout process. Computational Stat Data Analysis. 2001;36: in press.
35.
ThijsHMolenberghsGVerbekeG.The milk protein trial: Influence analysis of the dropout process. Biometrical J.2000;42:617–646.
36.
BakerSGLairdNMRegression analysis for categorical variables with outcome subject to non-ignorable nonresponse. J Am Stat Assoc. 1988;83:62–69.
37.
StasnyEAEstimating gross flows using panel data with nonresponse: an example from the Canadian Labour Force Survey. J Am Stat Assoc. 1986;81:42–47.
38.
BakerSGRosenbergerWFDerSimonianR.Closed-form estimates for missing counts in two-way contingency tables. Stat Med.1992;11:643–657.
39.
ConawayMRThe analysis of repeated categorical measurements subject to nonignorable nonresponse. J Am Stat Assoc. 1992;87:817–824.
40.
ConawayMRNon-ignorable non-response models for time-ordered categorical variables. Appl Stat. 1993;42:105–115.
41.
ParkTBrownMBModels for categorical data with nonignorable nonresponse. J Am Stat Assoc. 1994;89:44–52.
