## Adjusting for nonignorable drop-out using semiparametric nonresponse models (with discussion (1999)

Venue: | Journal of the American Statistical Association |

Citations: | 51 - 10 self |

### BibTeX

@ARTICLE{Scharfstein99adjustingfor,

author = {Daniel O. Scharfstein and Andrea Rotnitzky and James M. Robins},

title = {Adjusting for nonignorable drop-out using semiparametric nonresponse models (with discussion},

journal = {Journal of the American Statistical Association},

year = {1999},

volume = {94},

pages = {1096--1146}

}

### Years of Citing Articles

### OpenURL

### Abstract

Consider a study whose design calls for the study subjects to be followed from enrollment (time t = 0) to time t = T,at which point a primary endpoint of interest Y is to be measured. The design of the study also calls for measurements on a vector V(t) of covariates to be made at one or more times t during the interval [0,T). We are interested in making inferences about the marginal mean µ0 of Y when some subjects drop out of the study at random times Q prior to the common fixed end of follow-up time T. The purpose of this article is to show how to make inferences about µ0 when the continuous drop-out time Q is modeled semiparametrically and no restrictions are placed on the joint distribution of the outcome and other measured variables. In particular, we consider two models for the conditional hazard of drop-out given ( ¯ V(T), Y), where ¯ V(t) denotes the history of the process V(t) through time t, t ∈ [0,T). In the first model, we assume that λQ(t | ¯ V(T), Y) = λ0(t | ¯ V(t)) exp(α0Y), where α0 is a scalar parameter and λ0(t | ¯ V(t)) is an unrestricted positive function of t and the process ¯ V(t). When the process ¯ V(t) is high dimensional, estimation in this model is not feasible with moderate sample sizes, due to the curse of dimensionality. For such situations, we consider a second model that imposes the additional restriction that λ0(t | ¯ V(t)) = λ0(t) exp(γ ′ 0W(t)), where λ0(t) is an unspecified baseline hazard function, W(t) = w(t, ¯ V(t)), w(·, ·) is a known function that maps (t, ¯ V(t)) to Rq, and γ0 is a q × 1 unknown parameter vector. When α0 � = 0, then drop-out is nonignorable. On account of identifiability problems, joint estimation of the mean µ0 of Y and the selection bias parameter α0 may be difficult or impossible. Therefore, we propose regarding the selection bias parameter α0 as known, rather than estimating it from the data. We then perform a sensitivity analysis to see how inference about µ0 changes as we vary α0 over a plausible range of values. We apply our approach to the analysis of ACTG 175, an AIDS clinical trial. KEY WORDS: Augmented inverse probability of censoring weighted estimators; Cox proportional hazards model; Identification;

### Citations

1323 | Statistical Analysis with Missing Data - Little, Rubin - 1987 |

374 | Efficient and Adaptive Estimation for Semiparametric Models - Bickel, Klaassen, et al. - 1993 |

241 |
A Generalization of Sampling without Replacement from a Finite Population
- Horvitz, Thompson
- 1952
(Show Context)
Citation Context ...special case in which b(¯v(t),t;µ) is chosen to be identically 0, we refer to ˜µ(b) as an inverse probability of censoring weighted (IPCW) estimator. This is a generalization of the Horvitz–Thompson (=-=Horvitz and Thompson 1952-=-) estimator used in the sample survey literature. When b(¯v(t),t;µ) is nonzero, we refer to ˜µ(b) as an augmented IPCW (AIPCW) estimator. The regularity condition 2 of Appendix A guarantees that π( ¯ ... |

175 | A new approach to causal inference in mortality studies with a sustained exposure period—application to control of the healthy worker survivor effect - Robins - 1986 |

102 | Estimation of regression coefficients when some regressors are not always observed - Robins, Rotnitzky, et al. - 1994 |

96 |
Good Thinking: The Foundations of Probability and Its Applications
- Good
- 1983
(Show Context)
Citation Context ...and the estimated (i.e., imputed) mean in the drop-outs and a plot of the estimated weights as a function of Y . At first, this might seem totally unacceptable. However, as pointed out by I. J. Good (=-=Good 1983-=-), if the expert were to carry out this reassessment for multiple simulated datasets before seeing the actual data, then this would be a perfectly valid method for eliciting the expert’s actual prior ... |

93 | Analysis of Semi-parametric Regression Models Under the Presence of Missing Data - Robins, Rotnitzky, et al. - 1995 |

72 |
Semiparametric efficiency bounds
- Newey
- 1990
(Show Context)
Citation Context ...stimation of η not to affect the asymptotic variance of ˆµ(b) is � −1 that n i ∂h† (Oi;µ0,η0;b)/∂η converge to 0 in probability, or, equivalently, that E[∂h † (O;µ0,η0;b)/∂η] =0. But it can be shown (=-=Newey 1990-=-) that E[∂h † (O;µ0,η0;b)/∂η] = E[h(O;µ0,Λ0;b)Sη], where Sη = ∂ ln L(µ0,η0;O)/∂η is the derivative of the observed-data log-likelihood ln L(µ0,η;O) for a single subject with respect to η. Thus we conc... |

71 | Informative dropout in longitudinal data analysis - Diggle, Kenward - 1994 |

70 |
Non- and semi-parametric maximum likelihood estimators and the von Mises method (Part 1
- Gill
- 1989
(Show Context)
Citation Context ...ered further in this article. The third approach is to recognize that if ˆ ψ(b) is a RAL estimator, then we can obtain a consistent estimate of its asymptotic variance by the nonparametric bootstrap (=-=Gill 1989-=-). Because in conducting a sensitivity analysis it is necessary to calculate confidence intervals for µ0 for many values of the selection bias parameter α0, bootstrap variance estimation may require i... |

68 | Causal inference from complex longitudinal data. Latent Variable Modeling and Applications to Causality - ROBINS - 1997 |

63 | Survival models for heterogeneous populations derived from stable distributions - Hougaard - 1986 |

57 | Integral equations - Tricomi - 1985 |

45 |
Towards a Curse of Dimensionality Appropriate (CODA) Asymptotic Theory for Semi-parametric Models,” Statistics in Medicine
- Robins, Ritov
- 1997
(Show Context)
Citation Context ...re is no estimator of µ0 that has, under all laws allowed by model A(α0), anapproximately normal sampling distribution centered near µ0 with variance sufficiently small to be of substantive interest (=-=Robins and Ritov 1997-=-). This reflects the fact that to estimate µ0 under model A(α0), it is necessary to use multivariate nonparametric smoothing techniques, which would require impractically large samples when ¯ V(t) is ... |

43 | 1991] “Ignorability and coarse data - Hetijan, Rubin |

42 |
Recovery of information and adjustment for dependent censoring using surrogate markers
- Robins, Rotnitzky
- 1992
(Show Context)
Citation Context ...cient estimates of the mean µ0 of Y under model B(α0). Our approach is motivated by the observations that (a) if α0 =0, the semiparametric variance bounds in models A(α0) and B(α0) will be identical (=-=Robins and Rotnitzky 1992-=-), and (b) even when α0 �= 0,ifW(t)in (2) is high dimensional, the semiparametric variance bound in model B(α0) will be only slightly less than the variance bound for the larger model A(α0). Thus if w... |

41 | Sensitivity analysis for selection bias and unmeasured confounding in missing data and causal inference models - ROBINS, SCHARFSTEIN, et al. - 2000 |

32 | Estimation and comparison of changes in the presence of informative right censoring: conditional linear model - Wu, Bailey - 1989 |

24 | Semiparametric regression for repeated outcomes with non-ignorable non-response - Robins, Rotnitzky, et al. - 1998 |

24 | Modeling the relationship of survival to longitudinal data measured with error. Application to survival and CD4 counts in patients with AIDS - Tsiatis, DeGruttola, et al. - 1995 |

19 | Modeling Progression of CD4-Lymphocyte Count and its Relationship to Survival Time - DeGruttola, Tu - 1994 |

19 | Estimating exposure effects by modelling the expectation of exposure conditional on confounders - Robins, Mark, et al. - 1992 |

17 | Semiparametrical analysis of the additive risk model - Lin, Ying - 1994 |

16 | A note about models for selectivity bias - Little - 1985 |

14 |
A trial comparing nucleoside monotherapy with combination therapy in HIV-infected adults with CD4 cell counts from 200 to 500 per cubic millimeter
- HAMMER, KATZENSTEIN, et al.
- 1996
(Show Context)
Citation Context ...T + ddC); and (4) didanosine 200 mg twice daily (ddI). Enrollment began in December 1991 and was closed in October 1992. CD4 counts were obtained at baseline and again at weeks 8, 20, 32, 44, and 56 (=-=Hammer et al. 1996-=-). One goal of the investigators was to compare the four treatment arm–specific mean CD4 counts at week 56 had (possibly contrary to fact) all subjects complied with their assigned therapy through tha... |

10 | Model-based approaches to analyzing incomplete longitudinal and failure time data - Hogan, Laird - 1997 |

9 | Multivariate Logistic Models for Incomplete Binary Responses - Fitzmaurice, Laird, et al. - 1996 |

9 | Inference from nonrandomly missing categorical data: an example from a genetic study on Turner’s syndrome - Nordheim - 1984 |

8 | Regression models for longitudinal binary responses with informative drop-outs - Fitzmaurice, Molenberghs, et al. - 1995 |

6 | Logistic Regression Models for Binary Panel Data with Attrition - Fitzmaurice, Heath, et al. - 1996 |

6 | Analyzing Changes in the Presence of Informative Right Censoring Caused by Death and Withdrawal - Wu, Bailey - 1988 |

5 | Bounds on net survival probabilities for dependent competing risks - KLEIN, MOESCHBERGER - 1988 |

5 | Missing Data - Laird - 1988 |

5 | Modeling a marker of disease progression and onset of disease - Self, Pawitan - 1992 |

4 | On Energy Policy Models - Freedman, Rothenberg, et al. - 1983 |

4 | Dependent competing risks and summary survival curves - Slud, Rubinstein - 1983 |

3 | A Self-Consistent Estimator of Marginal Survival Functions Based on Dependent Competing Risk Data and an Assumed Copula - Zheng, Klein - 1994 |

2 | Efficient and Inefficient Estimation - Laan, J - 1996 |

1 | ClosedForm Estimates for Missing Counts - Baker, Rosenberger, et al. - 1992 |

1 | Application of Empirical Bayes Inference to Estimation of Rate - Mori, Woodworth, et al. - 1992 |

1 |
Likelihood Ratio–Based Confidence Intervals in Survival Analysis
- Murphy
- 1995
(Show Context)
Citation Context ...der Laan 1993), which is asymptotically equivalent to ˆµ( ˆb ∗ ). Indeed, it may be algebraically equivalent, depending on which of several possible “nonparametric likelihood functions” is maximized (=-=Murphy 1995-=-). However, as shown in Section 4, the AIPCW methodology generalizes straightforwardly to model B(α0) with ¯ V(t) high dimensional. In this latter setting, the NPMLE is undefined (Robins and Ritov 199... |

1 | Methods for the Analysis of Informatively Censored - Schluchter - 1992 |