Doody's Review Service

Reviewer:	Douglas Gunzler, PhD (Case Western Reserve University School of Medicine)

Description
This book covers different methods for epidemiological time-series prediction in the COVID-19 crisis. The models it introduces can be used with available daily case data to provide decision support for governments and healthcare decision-makers. Real data from five countries (China, the U.S., Brazil, Italy, and Singapore) are used throughout the book to illustrate the approaches and techniques. Most chapters first discuss methods of prediction (in a not too mathematically rigorous manner) and then present illustrative figures, tables, and analyses of the results for each of the five countries.

Purpose
The purpose is to present different methods for epidemiological time-series prediction in the COVID-19 crisis and illustrate these methods based on real data from five countries with some similarities and significant differences. As the COVID-19 pandemic is still unfolding, this objective is a worthy one in public health and books like this are essential for decision makers. The book does a very nice job of introducing and illustrating many different approaches to and techniques of prediction using data on daily new cases. The approaches include basic epidemiological models such as SIR and SEIR, time-series forecasting models such as ARIMA, state-space models for nonlinear prediction such as Quadratic Kalman Filter, and more advanced machine learning models. As information becomes available, new models taking additional events into account that were not relevant at the time the book was written (for example, vaccines) may prove useful. This book does not discuss prediction using hospitalization and mortality data, which has proven useful given that data on daily new cases can be influenced by the number of daily tests available among other confounding factors.

Audience
The book is written for students, researchers, and other stakeholders in the COVID-19 pandemic with an interest in epidemiology and public health decision-making. The prediction models are presented at an introductory level and mathematical details are not too rigorous. Results are then interpreted using accuracy evaluation criteria discussed in the introductory chapter. The book is a great reference for different approaches for prediction in the COVID-19 crisis. Readers interested in learning more about these methods can look into the references provided after the chapters. The four authors combine vast expertise and experience using the type of models discussed in the book.

Features
The first chapter briefly discusses the background of the COVID-19 pandemic and then introduces prediction models, methods for validation and evaluation, and the data used throughout the book. The remaining five chapters focus on specific methods (compartmental SIR and SEIR models, autoregressive models, state-space models, artificial intelligence, and geographical prediction) and the illustration of these methods using daily case data from five countries during the COVID-19 pandemic. The book is clearly written and provides a great, not too technical, introduction to these prediction approaches. The authors detail how to interpret results from these analyses. Graphical displays in color help readers understand and evaluate findings. Since the pandemic is still unfolding, stakeholders now have more timely information to consider in predictive modeling. However, the book is a great introduction to approaches to using data on daily cases for an unfolding pandemic (given limitations for such data) for prediction and decision support.

Assessment
The authors wrote this book to provide a comprehensive reference on predictive modeling and decision support during the unfolding COVID-19 Crisis. The timeliness of such a reference is valuable for public health stakeholders. To my knowledge there are many papers and websites on prediction for the COVID-19 Crisis. Most utilize a single approach. However, this book is one of the first books to detail different methods and leave the reader with an understanding of what can be concluded for each approach.