Modern Biostatistical Methods for Evidence-Based Global Health Research

Preface
Contents
Contributors
List of Chapter Reviewers
About the Editors
Sub-Saharan African Region Strategies to Improve Biostatistics Capacity: Exploring Collaborations Between Training and Research Institutions
1 Introduction
2 SSACAB
3 Collaborations with Training Institutions
4 Conclusion
References
Part I Data Harmonization and Analysis
Diagonal Reference Modelling of the Effects of Educational Differences Between Couples on Women's Health-Care Utilization in Eritrea
1 Introduction
2 Data: Hospital Delivery Among Eritrean Women
3 Diagonal Reference Models
4 Application: Educational-Mobility Effects on Hospital Delivery
4.1 Measures
4.2 Results
4.2.1 Results from Conventional Logistic Regression
4.2.2 Results from Diagonal Reference Models (DRM)
5 Summary and Concluding Remarks
Appendix: R-Codes Used for Computing the Results in Some of the Tables in This Chapter
References
Sequential Probit Modeling of Regional Differences in the Effects of Education on Parity Progression Ratios in Ethiopia
1 Introduction
2 Probit and Sequential Probit Models
2.1 Probit Model
2.2 Sequential Probit Model
2.3 Parameter Estimation in the Sequential Probit Model
3 Application to Parity Progression Among Ethiopian Women
3.1 Data Set and Measures
3.2 Sequential Probit Model for Propensities of Parity Progression
3.3 Results
4 Summary and Concluding Remarks
Appendix: aML Code Used for Computing the Results in the Upper-Left Panel of Table 2 in This Chapter (Except Those in the First Column)
References
Propensity Score Approaches for Estimating Causal Effects of Exposures in Observational Studies
1 Introduction
1.1 Infant and Young Child Feeding Interventions and Child Nutritional Outcomes
2 PS Methods to Minimise Confounder Bias in Estimating Exposure Effects
2.1 Propensity Score Definition
2.1.1 Propensity Score Estimation
2.1.2 Propensity Score Model Misspecification
2.2 Propensity Score Matching
2.3 Inverse Probability Weighting (IPW) on the PS
2.4 Assessing Confounder Balance
2.4.1 Sensitivity Analysis
3 Illustrative Examples Using Nutritional Outcome Data in Children
3.1 Data
3.1.1 Causal Pathway Framework
3.1.2 Exclusive Breastfeeding and Complementary Feeding Indicators
3.2 Results
3.2.1 The Effect of the Nutritional Interventions on Child Nutritional Outcomes Before PS Application
3.2.2 Estimating the Propensity Score
3.2.3 PS Matching
3.2.4 PS Weighting
3.2.5 Effect of Mothers' HIV Status on the Causal Effect of Exclusive breastfeeding for Zambian Children
3.2.6 Effect of Appropriate Complementary Feeding on Child Growth
3.2.7 Sensitivity Analysis
4 Discussion
5 Future Work
6 Further Reading
Appendix
STATA Code
References
Part II Systematic Review and Statistical Meta-Analysis
Evidence-Informed Public Health, Systematic Reviews and Meta-Analysis
1 Introduction to Evidence-Based Public Health, Systematic Reviews and Meta-Analysis
1.1 What Is Evidence-Based Public Health?
1.2 Why Do We Need Evidence-Informed Decision-Making?
1.3 Overview of Research Synthesis
1.4 Types of Research Synthesis
1.5 What Is So Special About Systematic Reviews?
1.6 The Value of Systematic Reviews in Public Health Decision-Making
1.7 Conclusion
2 Methods of Data Synthesis
2.1 Meta-Analysis
2.2 Unit of Analysis Issues
2.2.1 Cluster Randomised Trials
2.2.2 Studies with Multiple Arms
2.3 Assessment of Heterogeneity
2.4 Methods for Dealing with Heterogeneity in Meta-Analysis
2.4.1 Subgroup Analysis
2.4.2 Meta-Regression
2.5 Sensitivity Analysis
3 Emerging Techniques in Systematic Reviews and Meta-Analysis
3.1 Network Meta-Analysis
3.1.1 Introduction to Network Meta-Analysis: The Extension to Pairwise Meta-Analysis
3.1.2 Assumptions of NMA: Transitivity and Inconsistency
3.1.3 NMA Analysis in STATA and Presenting Results
3.1.4 Presenting and Evaluating and Assumptions of NMA
3.1.5 Conclusions
3.2 Meta-Analysis of Diagnostic Test Accuracy Studies
3.2.1 Introduction to Meta-Analysis of Diagnostic Test Accuracy Studies
3.2.2 Basic Concepts of Diagnostic Test Accuracy
3.2.3 When to Conduct a Meta-Analyses of Diagnostic Test Accuracy Studies
3.2.4 Methods for Meta-Analyses of Diagnostic Test Accuracy Studies
3.2.5 Steps in Conducting a Meta-Analyses of Diagnostic Test Accuracy Study
3.2.6 Conclusion
3.3 Data Synthesis when Meta-Analysis Is Not Possible
3.3.1 Reasons That May Render Meta-Analysis Impossible or Inappropriate
3.3.2 Recommended Alternative Ways of Synthesizing Data
4 Overview of Certainty of Evidence and GRADE
4.1 What Is GRADE and Why Is It Needed?
4.2 GRADE Judgements and Domains
4.2.1 Domains for downgrading evidence
4.2.2 Domains for upgrading observational studies
4.2.3 Making and presenting GRADE judgements
5 Useful Resources
References
Statistical Meta-Analysis and Its Efficiency: A Real Data Analysis and a Monte-Carlo Simulation Study
1 Introduction
2 Overview of Meta-Analysis with Summary Statistics
2.1 Summary Statistics and the Sources of Variations
2.2 Fixed-Effects Meta-Analysis
2.3 Random-Effects Meta-Analysis
2.4 Quantify Heterogeneity in Meta-Analysis
2.4.1 The τ2 Index
2.4.2 The H Index
2.4.3 The I2 Index
2.5 Meta-Regression
3 Simulation Study on Relative Efficiency Between SS-MA and IPD-MA
3.1 Continuous Data
3.1.1 Simulation Settings
3.1.2 Results
3.2 Categorical Data
3.2.1 Simulation Settings
3.2.2 Results
4 Real Data Analysis
4.1 Introduction to the Bacillus Calmette-Guerin Vaccine Data
4.2 Fixed-Effects and Random-Effects Meta-Analysis
4.2.1 Effect-Size Calculation
4.2.2 Fixed-Effects Meta-Analysis
4.2.3 Random-Effects Meta-Analysis
4.3 Meta-Regression
4.4 Relative Efficiency Between SS-MA and IPD-MA
4.4.1 Individualization Procedure
4.4.2 IPD-MA
4.4.3 SS-MA1
4.4.4 Relative Efficiency Between IPD-MA and SS-MA
5 Summary and Discussions
References
Meta-Analysis Using R Statistical Software
1 Introduction
1.1 Citation of R Statistical Software
1.2 Installation of R Software and R Studio
1.3 Basics of Meta-Analysis
1.4 Effect Measures for Continuous Outcome Data
1.5 Effect Measures for Dichotomous Outcome Data
2 Methods for Pooling the Effect Measures
2.1 The Mantel Haenszel Method
2.2 The Peto Odds Ratio
3 Steps of Meta-Analysis for Intervention Studies with R Statistical Software
3.1 Fixed Effects Versus Random Effects Meta-Analysis
3.2 Some Study Designs in Clinical Trials
3.3 Meta-Analysis of Dichotomous Outcome Data
3.4 Meta-Analysis of Continuous Outcome Data
4 Heterogeneity
4.1 Handling Heterogeneity Between Studies
4.2 Estimation of Study Bias
4.3 A Discussion on Measures of Heterogeneity
4.4 Subgroup Analysis or Sensitivity Analysis
5 Further Topics in Meta-Analysis
5.1 Meta-Regression
5.2 Network Meta-Analysis
References
Longitudinal Meta-Analysis of Multiple Effect Sizes
1 Introduction
2 Statistical Meta-Analysis Model
3 Estimation of Parameters
3.1 Maximum Likelihood (ML) Estimation
3.2 Restricted Maximum Likelihood (REML) Estimation
4 Modeling Covariance Structures
4.1 Model 1—Independent m=2 Effect Sizes, Independent Random Time Effects, and Independent Within-Study Effect Sizes
4.2 Model 2—Correlated m=2 Effect Sizes, Independent Random Time Effects, and Independent Within-Study Effect Sizes
4.3 Model 3—Independent m=2 Effect Sizes, Correlated Random Time Effects, and Correlated Within-Study Effect Sizes
4.4 Model 4—Correlated m=2 Effect Sizes at Each Time Point, Correlated Random Time Effects, and Correlated Within-Study Effect Sizes
5 Example: Antiretroviral Drugs in Treatment-Experienced HIV-Infected Patients
6 Summary
Appendix: SAS Code
References
Part III Spatial-Temporal Modelling and Disease Mapping
Measuring Bivariate Spatial Clustering in Disease Risks
1 Introduction
2 Spatial Autocorrelation
2.1 Spatial Weights
3 Univariate Spatial Autocorrelation
3.1 Univariate GISA
3.2 Univariate LISA
4 Bivariate Spatial Autocorrelation
4.1 Bivariate GISA
4.2 Bivariate LISA
5 An Application
5.1 Data
5.2 Statistics Methods in Rate Estimation
6 Results
6.1 Univariate Cluster Analysis
6.1.1 Univariate Global Spatial Autocorrelation
6.2 Univariate ``Hot-Spot'' Analysis
6.3 Bivariate Analysis
6.3.1 Bivariate Association of Individual CVD Maps Over Time
6.3.2 Bivariate Spatial Association Between Two CVDs at a Point in Time
7 Discussion
References
Bivariate Copula-Based Spatial Modelling of Health Care Utilisation in Malawi
1 Background
2 Copula-Based Methods
2.1 General Copula Theory
2.2 Marginal Regression Models
3 Copula-Based Spatial Methods
4 Application to Antenatal Care Data
4.1 Methodology
4.2 Bivariate Copula Spatial Modelling
4.3 Results
4.4 Bivariate Model Results
4.5 Spatial Variation Across Malawi
5 Conclusion and Recommendation
Appendices
References
Part IV Bayesian Statistical Modelling
Bayesian Survival Analysis with the Extended Generalized Gamma Model: Application to Demographic and Health Survey Data
1 Introduction
2 Parametric Models for Survival Data
2.1 Background
2.2 Accelerated Failure-Time Models
2.3 The Extended Generalized Gamma (EGG) Model
2.4 Further Extensions of the EGG Model
3 Bayesian Inference in the Extended Generalized Gamma Model
3.1 Prior and Posterior
3.2 MCMC: Random Walk Metropolis–Hastings Algorithm with Block Sampling
3.3 Posterior Statistics and Convergence Diagnostics
3.4 Bayesian Model Comparisons
4 Application: Educational and Residential Differences in Marriage Timing Among Eritrean Men and Women
4.1 Data and Variables
4.2 Results from Bayesian Analysis of the Data Using the EGG Model
5 Summary and Concluding Remarks
Appendix 1: Density Functions, f(t), and Survival Functions, S(t), of Special Cases in the Extended Generalized Gamma Model
Appendix 2: Proof of Lemma 1
Appendix 3: R Program Codes for Bayesian Inference
References
Dynamic Bayesian Modeling of Educational and Residential Differences in Family Initiation Among Eritrean Men and Women
1 Introduction
2 The Data Set and Preliminary Analyses with Standard Models
3 Dynamic Bayesian Modeling of Survival Data
3.1 The Likelihood Under Piece-Wise Exponential Distribution
3.2 Prior Specification
3.3 Sampling and Inference from the Posterior Distribution
4 Application: Dynamic Bayesian Modeling of Time to Family Initiation
5 Summary and Concluding Remarks
Appendix: R Program Codes for Dynamic Bayesian Survival Modeling Used in the Chapter
References
Bayesian Spatial Modeling of HIV Using Conditional Autoregressive Model
1 Introduction
2 Data and Models
2.1 Data Description
2.2 Theoretical Model
2.2.1 Bayesian Spatially Varying Coefficient Parameter (BSVCP) Model
2.2.2 Implementation of Conditionally Autoregressive Model in R-INLA
3 Results
3.1 Model Comparison Based on Deviance Information Criteria
3.2 Nonlinear Effect of Age
3.3 Spatially Varying Effects
3.4 Spatial Effects
4 Discussion
References
Estimating Determinants of Stage at Diagnosis of Breast Cancer Prevalence in Western Nigeria Using Bayesian Logistic Regression
1 Introduction
2 Data and Methods
2.1 Ethical Consideration
3 Statistical Model
3.1 Binary Response Logistic Regression Model Formulation
3.1.1 Computations and Implementations
4 Results
4.1 Bayesian Logistic Regression
5 Discussion
6 Strengths and Limitations of the Chapter
7 Implications of the Chapter
References
Part V Statistical Applications
Identifying Outlying and Influential Clusters in Multivariate Survival Data Models
1 Introduction
1.1 The Multivariate Survival Model and Its Estimation
2 Outlier Analysis for Multivariate Survival Data
2.1 Proposed Outlier Statistic for Multivariate Survival Data
2.2 Simulation Study
2.2.1 Simulation Results
2.3 Application to Malawi Child Survival Data
2.3.1 Results on Effects of Some Variables on Under-Five Child Mortality
2.3.2 Results for Outlier Subdistricts on Under-Five Mortality
3 Influence Analysis for Multivariate Survival Data
3.1 Some Common Influence Statistics for Univariate Survival Data
3.2 Proposed Influence Statistic for Multivariate Survival Data
3.3 Numerical Example
3.3.1 Results of Simulations
3.4 Application of the Influence Statistic on Malawi Child Survival Data
3.4.1 Results for Influential Subdistricts on Effect of Being Female on Under-Five Mortality
3.4.2 Impact of the Identified Influential Subdistricts on Estimate of Effect of Being Female on Under-Five Mortality
4 Conclusion
Appendices
R Code for Applying Derived Outlier Statistic on Child Survival Data
R Code for Applying Derived Influence Statistic on Child Survival Data
References
Joint Modelling of Longitudinal and Competing Risks Survival Data
1 Introduction
1.1 Competing Risks Joint Models
1.1.1 Survival Submodel
1.1.2 Longitudinal Submodel
1.2 Joint Model
1.3 Parameter Estimation in Competing Risks Joint Model
2 Application to Malaria Data
2.1 Materials and Methods
2.2 Study Population and Sample Size
2.3 Statistical Analysis Methods
2.4 Ethical Approval
3 Results
3.1 Descriptive Analysis
3.2 Longitudinal Process Analysis
3.3 Survival Process Analysis
3.4 Competing Risks Joint Models for Parasite Count and Hemoglobin Level
3.5 Model Comparison
3.6 Model Diagnostics
4 Discussion
5 Conclusion
References
Stratified Multilevel Modelling of Survival Data: Application to Modelling Regional Differences in Transition to Parenthood inEthiopia
1 Introduction
2 The Data Set: Distribution Across Regions and Covariates and Clustering Within Households
2.1 Ethiopia and Its Regions
2.2 The 2016 Ethiopia Demographic and Health Survey
2.3 Response and Explanatory Variables
3 Proportional Intensity Models for Transition to Parenthood
3.1 The Standard Cox Proportional Intensities Model
3.2 Multilevel Cox Proportional Intensities Model: Accounting for Clustering of Women into Households
3.2.1 Multilevel (Frailty) Models
3.2.2 Cox Proportional Hazards Model with Gamma Distributed Frailty
3.2.3 Cox Proportional Intensities Model with Log-Normal Distributed Frailty
4 Results
4.1 Results from Standard and Frailty Models for the Entire Country
4.2 Results from Stratified Cox Models with No Frailty
4.3 Results from Stratified Cox Models with Gamma Frailty
4.4 Results from Stratified Cox Models with Log-Normal Frailty
5 Summary and Concluding Remarks
Appendix: R-Codes Used for Computing the Results Reported in the Tables of This Chapter
References
Application of Multiple Imputation, Inverse Probability Weighting, and Double Robustness in Determining Blood Donor Deferral Characteristics in Malawi
1 Introduction
2 Missing Data Patterns
3 Missing Data Nomenclature
4 Methods
4.1 Study Design and Population
4.2 Data Sources, Description, and Management
4.3 Study Outcomes
4.4 Statistical Software and Methods
4.5 Assumption for Missingness Mechanism
4.6 Multiple Imputation
4.7 Inverse Probability Weighting
4.8 Double Robustness
5 Results
6 Discussion
7 Limitations
8 Conclusion
9 Disclosure
Appendix
References
Index