Comparing Methods to Handle Missing Data in the Estimation of Population Attributable Factors of Anemia in Preschool Children Open Access

Abstract

Introduction: Researchers frequently ignore missing data and include only subjects with complete data in analysis. However, ignoring missing data can lead to systemic bias in the effect estimation and inference as well as loss of power. The challenges of handling missing data include the lack of a general method to handle missing data, unknown missing data mechanisms, and applying appropriate methods corresponding to the missing data mechanism.

Objectives: The objective of the study is to evaluate the impact of missing data on the estimation of the population attributable fraction (PAF) of anemia in children based on three commonly used approaches.

Methods: The prevalence, relative risks, and PAF for proximal risk factors of anemia were estimated in preschool children accounting for complex survey design using national survey data from Nicaragua (NI2005), United States (US2006), and Pakistan (PK2011). Three approaches were used to handle missing data: 1) complete case analysis, 2) inverse probability weighting, and 3) multiple imputation.

Results: In this study, 32.75%, 13.49% and 4.48% were missing SF in NI2005, US2006, and PK2011, respectively. The estimates of PAF were similar across different methods in US2006 and PK2011. The estimated PAF values were substantially smaller using multiple imputation in NI2005 compared to those using complete case and inverse probability weighting. Specifically, the estimated PAF for inflammation, iron deficiency, and vitamin A deficiency were respectively 3%, 29%, and 2-3% using complete case and inverse probability weighting; however, they were 1%, 7%, and 1%, respectively, using multiple imputation. Overall, the estimates using complete case were similar to those using inverse probability weighting,

Conclusions: Different ways of handling missing data can affect the estimate of PAF. Greater impact is observed with a larger proportion of missing data (e.g., >30%). The findings were based on three national surveys and may not be generalized to other PAF estimations. Although the results of inverse probability weighting method and complete case analysis are similar, we recommend to use multiple imputation method in this study. 

Table of Contents

Table of Contents

1. Background........................................................................................................................................1

1.1 Missing data in survey data analysis...........................................................................................1

1.2 Missing data mechanisms and assumptions................................................................................1

1.3 Current approaches......................................................................................................................3

1.3.1 Complete case analysis.............................................................................................................3

1.3.2 Available case analysis.............................................................................................................3

1.3.3 Missing indicator method.........................................................................................................4

1.3.4 Inverse probability weighting method....................................................................................4

1.3.5 Single imputation method.........................................................................................................5

1.3.6 Multiple imputation method.....................................................................................................5

1.4 Anemia in children......................................................................................................................6

1.5 Population attributable fractions..................................................................................................7

2. Introduction.......................................................................................................................................8

2.1 Reasons and significance of missing data.............................................................................8

2.2 Challenges of handling missing data...........................................................................................9 2.3 Objective....................................................................................................................................10

3. Method............................................................................................................................................11

3.1 Data selection............................................................................................................................11

3.2 Primary outcomes......................................................................................................................11

3.3 Potential risk factors of anemia and covariates.........................................................................12

3.4 Population attributable fractions................................................................................................13

3.5 Complete case analysis..............................................................................................................13

3.6 Inverse probability weighting method.....................................................................................14

3.7 Multiple imputation method.....................................................................................................15

3.8 Statistical analysis.....................................................................................................................16

4. Results.......................................................................................................................................17

4.1 Population characteristics..........................................................................................................17

4.2 Comparisons of variables with or without missing pattern.......................................................17

4.3 Summary of select anemia risk factors......................................................................................18

5. Discussion....................................................................................................................................19

6. Conclusion...................................................................................................................................22

7. Reference.....................................................................................................................................24

8. Appendix.....................................................................................................................................27

About this Master's Thesis

Rights statement

• Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.

School	Rollins School of Public Health
Department	Biostatistics
Subfield / Discipline	Biostatistics - MPH & MSPH
Degree	M.P.H.
Submission	Master's Thesis
Language	English
Research Field	Biology, Biostatistics
Keyword	Nutrition
Committee Chair / Thesis Advisor	Yi-an Ko, Emory University
Committee Members	Melissa Young, Emory University

Last modified

Primary PDF

Thumbnail	Title	Date Uploaded	Actions
	Comparing Methods to Handle Missing Data in the Estimation of Population Attributable Factors of Anemia in Preschool Children ()	2021-04-26 11:44:59 -0400	Press to Select an action Download

Supplemental Files

Thumbnail	Title	Date Uploaded	Actions