https://orcid.org/0009-0006-8321-5737
J. Sunil Rao’s new book Statistical Methods in Health Disparity Research discusses statistical methods for health disparity estimation, covering approaches ranging from non-model to linear-based models, as well as the application of machine learning in health disparity estimation. Emphasis is placed on the need for validation of findings and a discussion of the advantages and limitations of each method. In addition, this book can be used as reading material for postgraduate courses in health and statistics and as a guide for practitioners interested in health disparities research. In-depth explanations are presented in seven chapters.
Chapter 1, “Basic Concepts,” outlines the basic concept of health disparities as differences in health status outcomes or access, utilization, or quality of health services between populations, which are not only related to race but can also involve geographic, gender, socioeconomic status, or age factors. A brief history of health disparities research includes the contributions of scientists such as Will Farr, Parent-Duchâtelet, and Villermé in the 19th century, who examined differences in mortality and social and sanitary factors. Examples of health disparities, such as differences in mortality rates and incidence of certain diseases between racial and social groups, are discussed, with factors such as education, geographical location, and socioeconomic status playing key roles. This chapter highlights the role of social, environmental, and biological factors in determining health disparities but only provides an overview. At the same time, more information may be available in subsequent sections or other relevant sources.
Chapter 2, “Overall Estimation of Health Disparities,” details the estimation of health disparities, focusing on two main sections: data and measurement and disparity indices. The first section discusses the data sources used, including experimental, observational, and complex survey data, which require specialized calculations for different sampling designs. The second section describes disparity indices used to measure health differences between population groups, such as total indices that measure differences across the population and group comparison indices that help understand the causes of disparities. Examples of indices include IMD and IID, which calculate individual health differences within populations, and other indices, such as absolute and relative concentration indices. This chapter provides an in-depth look at health disparity measurement methods that are important for understanding health differences in populations and measuring progress in reducing disparities.
Chapter 3, “Domain-Specific Estimates,” details estimates specific to a particular domain or subgroup, including geographic areas and socioeconomic groups. This chapter introduces two types of estimates: direct and indirect. Direct estimates use sample data specific to the domain, while indirect estimates use strengths from other domains or sources. Examples of direct estimates include using summary statistics to estimate essential features in the domain, such as health rates based on survey data. Meanwhile, indirect estimation involves methods such as synthetic estimation and model-based estimation for small areas. This chapter comprehensively overviews the various estimation methods used for specific domains.
Chapter 4, “Causality, Moderation, and Mediation,” outlines the concepts of causality, moderation, and mediation in the context of health disparities using a socioecological framework. An explanation of the complexity of the interactions between factors that affect health at the individual and community levels, with factors such as neighborhood poverty and educational attainment considered necessary. The chapter highlights the challenges in identifying the causes of health disparities and describes two common approaches in health disparities research: experimental and observational. Experimental studies are recognized as ideal for evaluating causality but face certain constraints in the context of health disparities. In contrast, observational studies are more commonly used but require adjustment for confounding factors. Specific causal factors, such as race/ethnicity, are also discussed with consideration of social constructs and the extension of specific analytical tools. In addition, this chapter describes the method of estimating the Average Treatment Effect (ATE) and the assumptions associated with ATE identification.
Chapter 5, “Machine Learning Based Approaches to Disparity Estimation,” discusses Machine Learning (ML) based approaches to disparity estimation, introducing the concept of ML and classifying it into supervised and unsupervised learning. The chapter explains the relevance of ML in health disparity research by considering relevant big data categories, such as genomic data, geospatial data, medical records, wearable devices, and electronic data. The use of ML in health disparities research can facilitate the integration of data from multiple sources and address the complexity of such research. Various ML methods, including tree-based models such as decision trees, are used for classification and regression, helping to build accurate decision rules and understand them intuitively. This chapter provides a basic understanding of the use of ML in disparity estimation and includes information on some of the algorithms used in this context.
Chapter 6, “Health Disparity Estimation under a Precision Medicine Paradigm,” is an in-depth review of the application of the precision medicine paradigm to address health disparities. The chapter elaborates on the concept of precision medicine, which involves customizing medical treatment based on individual patient characteristics to classify them into different subpopulations in terms of response to treatment, disease prognosis, or susceptibility to certain diseases. Furthermore, it highlights the importance of using genomic data in precision medicine, which enables an in-depth understanding of the multilevel determinants of disease, including factors such as lifestyle, nutrition, environment, and access to care, apart from the molecular aspects of disease. The chapter also discusses statistical methods such as tree-based methods, which identify subtypes of disparities within populations and can help develop more precise models in predicting response to treatment or disease prognosis. In addition, this chapter presents the results of analyses related to epigenetic age prediction and the use of classified mixture models to predict DNA methylation in cervical cancer. This analysis aims to understand health disparities between races in a precision medicine paradigm. This study used data from TCGA and involved M-value and RAU transformations to transform DNA methylation values before applying mixture models. The results show that classified mixture models, particularly elastic net models, can accurately predict DNA methylation variation in cervical cancer. Efforts to extend the mixture model to include additional data on cancer type and fixed effect covariates were also made to improve the understanding of health disparities between races.
Finally, Chapter 7, “Extended topics,” discusses additional topics related to health disparities research. The discussion begins with the effect of biased sampling on estimates of disease prevalence in the population and how to correct such errors, with the example of COVID-19 testing demonstrating the potential for overestimation of prevalence due to imperfect testing. Next, the chapter discusses the role of geocoding in evaluating contextual risk in multilevel models, with a walkthrough of several large-scale geocoding studies and ideas on filling in missing contextual variables in cancer studies. Finally, it discusses the impact of differential privacy in health disparities research and ideas for conducting multilevel modeling that accounts for differential privacy errors, including technical evidence related to the topics discussed. Overall, this chapter discusses corrections for sampling bias, the role of geocoding, contextual variable filling, and the impact of differential privacy in health disparities research.
The book has notable weaknesses in the consistency of the structure, where some sections may feel disconnected, disrupting the reading flow. For example, the R package for each chapter is given in Chapter 7 rather than in every chapter. However, the book also has some significant advantages. First, the diversity of topics covering basic concepts to technical applications in addressing health disparities gives the reader a comprehensive insight. Second, the in-depth presentation of complex concepts allows the reader to understand the material. Finally, the book is supplemented with relevant references and a complete code package, allowing readers to pursue further research according to their interests or needs.
This book is recommended for some readers with specific interests and backgrounds. First, for students or professionals in the public health or health policy field, this book will provide an in-depth understanding of health disparities and strategies to address these challenges. In addition, this book can be a valuable reference source for researchers or academics interested in studying health disparities. Furthermore, health practitioners and social workers working with vulnerable or at-risk populations can also benefit from the insights presented in this book to improve their practice.
Department of Mathematics, Institut Teknologi Bandung,
Indonesia
[email protected]
Dewi Syitra Rumadaul
Department of Public Health, Universitas Gadjah Mada,
Indonesia
Esther Yolandyne Bunga
Department of Mathematics, Institut Teknologi Bandung,
Indonesia
Safril
Department of Biology, Universitas Gadjah Mada, Indonesia
Sri Redjeki Pudjaprasetya
Industrial and Financial Mathematics Research Group,
Institut Teknologi Bandung, Indonesia
Warsoma Djohan
Combinatorial Mathematics Research Group,
Institut Teknologi Bandung, Indonesia