SC.03

SC.03 Measuring the Impact of Nonignorable Missing Data

Sunday, 5 July 2020  |  9:00 - 17:00 | Location: TBD

Instructors:
  • Daniel F. Heitjan, PhD
  • Hui Xie, PhD


Summary:

The popular assumption of ignorability greatly simplifies analyses with incomplete data. Unfortunately, it is generally impossible to robustly assess the validity of this assumption (or the closely related missing-at-random assumption) with just the data at hand. The possibility that the missing observations in a study are the result of a nonignorable (or missing-not-at-random) mechanism casts doubt on the validity of conclusions based on the assumption of ignorability.

One way to address this problem is to conduct a local sensitivity analysis: Essentially, re-compute estimates of interest under models that slightly violate the assumption of ignorability. If the estimates change only modestly under violation of the assumption, then it is safe to proceed with an ignorable model. If they change drastically, then a simple ignorable analysis is of questionable validity.

To conduct such a sensitivity analysis in a systematic and efficient way, we have developed a measure named the index of local sensitivity to nonignorability (ISNI). Computation of ISNI is straightforward and avoids the need to estimate a nonignorable model. We have developed a suite of statistical methods for ISNI analysis, now implemented in an R package named ISNI. In this short course we will describe these methods and train users in their applications.


Prerequisites:
Working knowledge of the analysis of cross-sectional and longitudinal data, including generalized linear models (linear regression, logistic regression, etc.) and mixed-effects and generalized linear mixed-effects models for longitudinal/clustered data. Knowledge about missing data methods is a plus.

Outline:

1.

Overview of sensitivity analysis and the sensitivity index (ISNI).

2. ISNI in common statistical models:
2.a Independent outcomes — the generalized linear model.
2.b Longitudinal/clustered outcomes
2.b.1 Marginal multivariate normal model.
2.b.2

Linear mixed model.

2.b.3 Generalized linear mixed model.
2.b.4

Monotone vs. non-monotone missingness patterns.

3.  Description of the R package isni
4. Application of ISNI:
4.a SOS survey data
4.b SWOG QoL trial
4.c Cocaine Trail
4.d CARDIA smoking-trend data
4.e A smoking cession trail data set


Learning Outcomes:

By the end of this workshop, attendees will:

  1. Recognize the importance and potential value of analysis for sensitivity to nonignorability with missing data.
  2. Understand the sensitivity index measures, their underlying assumptions and interpretation.
  3. Be able to apply the R software to compute the sensitivity measures in real clinical trials and observational studies.
  4. Use the sensitivity index measures to judge the credibility and robustness of study findings when there are missing observations.

About the Instructors:

Daniel F. Heitjan, PhD, is Professor of Statistical Science at Southern Methodist University and Professor of Clinical Sciences at UT Southwestern Medical Center, both in Dallas, Texas, USA. He and his students and collaborators have created and developed ISNI methods for a broad array of missing-data settings. Dr. Heitjan is a Fellow of the American Statistical Association, the Institute of Mathematical Statistics, and the Society for Clinical Trials, and the winner of the 2017 Don Owen Award from the ASA San Antonio Chapter.

Hui Xie, PhD, is Milan and Maureen Illich/Merck Frosst Chair in Biostatistics for Arthritis & Musculoskeletal Diseases and Professor of Biostatistics at the Simon Fraser University and Arthritis Research Canada. Drs. Xie and Heitjan, together with their collaborators and students, have developed a set of tractable, flexible and robust methods for sensitivity measures in a range of common settings, as well as user-friendly software to make formal and principled sensitivity measures accessible.

Recommendations for the Course:

The course requires all participants bring their own laptops.