Introduction

In the past few years, precision public health (PPH) has emerged as a multidisciplinary field [1,2] that relies on big data and data science [3] to drive public health assessment, policy, and implementation activities. The use of the word “precision” in the context of population-level activities as opposed to individualized precision medicine interventions has generated a variety of reactions [4–7]. The word is traditionally applied to individuals and may seem at odds with public health efforts focused on improving population health. In addition, as the field of precision medicine has so far identified mostly with genomic medicine, there was concern about its silence on social and environmental determinants that are important drivers of population health. An online dialogue and a paper elaborated on the tension between precision medicine and population health and the need for both personalized and population-level interventions to improve health [8]. Increasingly, big data offer opportunities for wider implementation of PPH approaches together with methodological strategies to address potential limitations.

Recently, a special series of articles explored the topic of PPH in Frontiers in Public Health [9]. In their introductory editorial, Weeramanthri and colleagues [10] discussed how the term was introduced in Western Australia as a way to use genomics, spatial technology in health, and data linkages to complement the developments in “precision medicine,” a term used in a 2011 US National Research Council Report [11]. In 2015, the US government launched its precision medicine initiative [12], focusing on cancer and the development of a large cohort of a million or more participants for a longitudinal study of health and disease (AllofUs Research Program) [13]. In 2016, the concept of PPH was introduced in the peer-reviewed literature as a call to modernize surveillance and information systems, as well as targeted interventions from a population health perspective [1]. In 2019, we elaborated on the notion of PPH as much larger in scope than individual genomic variation [14], as well as the need to use novel data sources and analytics to modernize public health tracking and implementation by time, place, and persons [14]. While we recognize that PPH is a global issue, national public health systems and their approaches do vary significantly. Therefore, in this review we focus mainly on the US health and public health system to demonstrate important points and examples.

From genomics and big data to precision medicine and PPH

In the past decade, we have seen an explosion in health- and non-health-related data. With electronic health records, genomic and other biomarkers, and emerging use of wearable biomonitoring devices, the sheer amount of information available for population analysis is large and increasing exponentially [2]. Perhaps as opposed to “small data,” big data are “too complex and varied to be contained in one spreadsheet or sometimes even in one computer” [15]. Big data can be integrated from many sources related to people, places, and time, such as geographic information systems, electronic health records, digital biomarkers, online apps, and social media.

Banks [15] recently reported some staggering numbers about big data. Examples include an estimated worldwide health data of 2,314 exabytes (10¹⁸ bytes) in 2020 and 25 petabytes (10¹⁵ bytes) of genomic data by 2030. A single typical hospitalization generates about 150,000 pieces of data, and the market for wearable devices that can capture health related is projected to be at $16.3 billion in 2022 [15].

Big data enable the potential for more “precision” in medicine and public health. In theory, more data at the individual level can help redefine the meaning of healthy and the progression from health to disease, helping to uncover preventable disease risk factors and allowing more precision diagnostic and prognostic information. At the population level, big data can help integrate multiple social and environmental risk factors such as air pollution, neighborhood walkability, and access to healthy food.

Table 1 summarizes our take on the evolving definitions in medicine and public health in relation to how genomics and “precision” have influenced these fields. It is important to acknowledge at the outset that addressing health problems in a population requires the application of individual medical approaches [16] and population-level approaches [17,18]. Improving the health of populations is not only about delivering the most optimal healthcare but is about working across all determinants of health, ranging from genetic and biological factors to social and environmental determinants of health. The completion of the human genome project led to a new era of genomic medicine [19] driven by genetic information in individuals. In parallel, the field of public health genomics arose as an approach to use genomic information to improve the health of populations. However, the recent evolution of genomic medicine to include other types of information led to the rise of “precision medicine,” which seeks to personalize medical interventions at the individual level. PPH is more than the application of genomics in populations. Beyond genomics, all kinds of data can be used in measuring determinants of health and impact of interventions. If precision medicine is about delivering the right intervention to the right individual at the right time, PPH can be simply viewed as delivering the right intervention to the right population at the right time. Put in another way, PPH is “about using the best available data to target more effectively and efficiently interventions of all kinds to those most in need” [7].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Genomics, precision medicine, and PPH: Definitions and comparisons.

https://doi.org/10.1371/journal.pmed.1003373.t001

Current applications in PPH

There are several current or near-term potential applications of genomics and big data in PPH, including a focus on modernizing public health surveillance, development of targeted interventions to implement effective interventions to improve health and reduce health disparities, the use of machine learning in public health, and special applications of pathogen genomics in the public health response to infectious diseases. Recent studies, commentaries, tools, and applications in PPH can be searched in the curated and continually updated Centers for Disease Control and Prevention (CDC) Public Health Genomics and Precision Health Knowledge Base [20] for specific health conditions or public health issues.

Data science: Machine learning, predictive analytics, and health disparities

An important scientific foundation for the use of big data in PPH is the emerging potential for using novel forecasting and predictive analytic methods. Under the rubric of data science, novel approaches to complex analysis are now increasingly used to integrate data from many areas to arrive at more precision in measuring health and disease outcomes, which in turn can be used to improve forecasting and prediction, aiding in decision-making for resource allocation and implementation strategies. Machine learning denotes a general approach to the processing of big data, to learn patterns in the data, and validate patterns for decision makers (e.g., these approaches can be deployed to doctors, patients, or policy makers) [36].

One example that demonstrates the potential of machine learning to improve the accuracy of disease diagnosis comes from medical image analysis, such as automating screening for diabetic retinopathy [37]. Patients with diabetes are at increased risk of eye disease. Manual analysis of image data is currently a rate-limiting step that can slow down screening for eye disease in diabetes. A new branch of machine learning—deep learning—has emerged and promises to enhance health care decision-making with automated image analysis.

Cancer management and prevention provide examples [38] of important current applications for the use of big data analytics. One example is the combination of data from multiple sources (e.g., DNA germ line and tumor sequencing, gene expression, epigenetics, proteomics) along with clinical patient information and environmental exposures [39] to determine individualized cancer therapeutic strategies. A specific application of machine learning analytics is in breast cancer screening. A recent study [40] developed a machine learning approach to identify breast cancer using mammograms in large UK and US data sets. The study compared the system’s cancer predictions and clinical radiologists’ original decisions based on biopsy-confirmed breast cancer. In the US data set, the approach led to 5.7% fewer false positives and 9.4% fewer false negatives than radiologists. In the UK data set, the results were more mixed. While promising, this machine learning approach will require further replication and prospective evaluation in multiple populations.

Overall, so far, data forecasting and predictive analytics have not provided better quantitative risk prediction models than classic statistical methods such as logistic regression analysis [41]. There are several methodologic issues that need to be addressed before big data analytics can be used in PPH. These include unrepresentative or selected populations, data inaccuracy and missing information, measurement issues, and substantial concerns about confounding and inference about causality. Other issues include deficiencies in model calibration and the lack of or insufficient data sharing [42].

As discussed in a recent commentary [43], predictive analytics need to have a clear purpose. The current literature contains studies on machine learning approaches that have undergone retrospective testing but not prospective evaluation and validation. As a result, the current applications of machine learning to healthcare systems remain limited. These limits also apply to public health activities that are concerned with measuring disease- and health-related information in population subgroups outside the healthcare delivery system.

One important limitation of machine learning is its potential impact on health disparities. An important tenet of PPH is to reduce health disparities by using big data to quantify health-related outcomes in subpopulations and use this knowledge to reduce health disparities [2]. However, there are current methodologic deficiencies such as systematic bias that could result in prediction models inadvertently contributing to a widening of health disparities, especially in racial and ethnic minority populations. For example, recent studies have consistently shown that genetic risk prediction models based on genome-wide association studies are less accurate in non-European populations. This is because most genome-wide association studies have been conducted in populations of European descent [44].

Another recent example [45] is a study of a machine learning risk assessment tool that is widely used in US healthcare organizations that has erroneously assigned a large proportion of Black patients to the same level of health risk as White patients. The authors estimated that racial bias can reduce the number of Black patients identified for extra care by more than half. Bias is present because the machine learning algorithm uses health costs as a proxy for healthcare needs. As less healthcare resources are spent on Black patients who have the same level of need, the algorithm falsely concludes that Black patients are healthier than equally sick White patients. This is a real example of biased predictive analytics that can widen existing health disparities in the population. In PPH, we need to have a strong emphasis on preserving privacy and confidentiality, as well as continuously evaluating the ethical, legal, and social implications of big data and predictive analytics.

The limitations of predictive analytics are even more pronounced in the context of global health data due to incompleteness and limitations of data for measuring exposures, health outcomes, and implementation challenges around the world [46]. In this context, resource constraints can influence the development and applications of these methods. In addition, issues of fairness, accountability, transparency, and privacy preservation provide important challenges to ensure that the promise of big data and predictive analytics to improve health and reduce health disparities does not lead to unintended effects.

The goal of big data analytics is to improve decision-making both in the clinical setting and in the population context. Providing outcome probabilities may or may not change physician, patient, or health system behavior. We need to evaluate the balance between health benefits and potential harms of specific implementation interventions that use big data analytics [47].

Pathogen genomics and PPH

Since the advent of next-generation genomic sequencing, pathogen genomics has rapidly transformed infectious disease public health, allowing for more precise detection and investigation of outbreaks, providing insights into disease emergence and transmission, enabling more accurate and efficient phenotyping of microorganisms, and thus providing more information for a PPH response, as compared with existing technologies [48,49].

Molecular subtyping of pathogens has played an important role in infectious disease public health for decades. Subtyping of pathogens into finer groups often makes outbreaks easier to detect, for example, and provides data to support or refute suspected transmission of a pathogen [48,50,51]. Almost all of these legacy subtyping technologies depend on changes in very small parts of the pathogen’s genome. Pulsed-field gel electrophoresis (PFGE), for example, produces a gel pattern that reflects polymorphisms (i.e., small changes in the genetic code) in perhaps two-dozen sites within a 5-million nucleotide bacterial genome [52]. Multi-locus sequence typing provides sequence data on only perhaps 7 of 3,000 genes in that genome. These technologies are generally very pathogen specific, may vary from laboratory to laboratory, and are usually uninformative about function.

Next-generation sequencing (also called high-throughput sequencing) increases by multiple orders of magnitude the amount of DNA that can be sequenced, allowing for the first time routine whole-genome sequencing of most pathogens. This not only enables extremely fine subtyping of those pathogens but also provides information about function—the antimicrobial resistance of a bacterial isolate, for example, or the presence or absence of a virulence factor that may affect clinical or public health decisions [52]. These data are inherently standardized. Moreover, the technology is applicable across the entire spectrum of microbes of public health importance, leading to a convergence of subtyping methods not just for bacteria but also for viral and eukaryotic pathogens [48].

In many high-income countries, next-generation sequencing is first being adopted in the bacterial foodborne disease domain, improving surveillance for such pathogens as Salmonella, Campylobacter, and Listeria [52]. The extremely fine subtyping provided by whole-genome sequencing allows prompt detection of outbreaks, usually signaled by the sudden emergence of a group of pathogens with identical or near-identical sequences. The technology improves investigation of those clusters by more accurately segregating outbreak from non-outbreak cases, allows for more precise confirmation of suspected food vehicles, and is a valuable tool in trace-back investigations. The switch from legacy technologies (mostly PFGE) to whole-genome sequencing in both the UK [53] and the US [54] has increased both the number of clusters of disease detected by roughly 2-fold and the number outbreaks solved. All of this is leading to a clearer picture of how pathogens are entering the food system and how they can be prevented.

The use of whole-genome sequencing in tuberculosis control is similar: extremely fine subtyping allows for detection of clusters that might otherwise be invisible [55]. Another example is Legionnaire’s disease—cluster of disease can be more confidently linked to sources such as cooling towers. In addition, the technology is providing insights into the ecology of the pathogen in water systems—insights that could eventually lead to better prevention [5]. For seasonal influenza, the paradigm is different: next-generation sequencing is accelerating characterization of the virus while providing a high-resolution picture of the dynamics of viral emergence, information that is now being used to inform vaccine strain selection [56]. Other areas of public health impacted by genomics include HIV, healthcare infection control, antimicrobial resistance monitoring, viral hepatitis outbreak surveillance and investigation, vaccine-preventable disease control, and many others [49].

Looking ahead: Prospects for PPH

Our expectations for big data to lead to more precision in public health depend on the continued ability to modernize the use of data in healthcare and public health, the conduct of rigorous studies to evaluate the validity and utility of new data-driven approaches, and the need for innovation in applications of data science and workforce development to help integrate data science in public health.

Concluding remarks: PPH in the era of COVID-19

The field of PPH is clearly in its infancy, and many challenges lie ahead. Perhaps the biggest challenge of our time is the current pandemic of coronavirus disease 2019 (COVID-19). The rapid emergence of a novel coronavirus [66] has facilitated an accelerated use of the applications of “big data” tools and technologies discussed in this paper to the investigation of COVID-19. Just to illustrate, we cite the use of whole-genome sequencing [67–69] to track the virus origin and spread; detailed geographic information to track spread at the global, country, and local levels [70]; the use of smartphone-based tracking and control [71]; and rapid characterization of risk factors related to severe disease such as age, underlying medical conditions, and smoking [(72]. The role of host genomic factors is beginning to be explored [73]. Use of machine learning and data science is contributing to prediction of diagnosis and complications [74–76]. A recent commentary summarized the potential applications of emerging digital technologies [77] in augmenting public health strategies for tackling COVID-19, including public health surveillance, detection and control, and mitigation of its impact on healthcare delivery. In addition, while digital approaches to large-scale data collection can aid the investigation and control of COVID-19, ethical and social issues need to be considered, such as privacy and public trust. This will provide the opportunity to develop best practices for responsible data collection and processing globally [78]. The COVID-19 pandemic provides both a challenge and a call to action for further evolution of PPH, as new tools and technologies will begin to complement medical and public health approaches to diagnosis, treatment, control, and prevention [79]. In these challenging times, further developments in the field will require global, national, and local leadership and commitment to enhance coordination of systems; sharing, harmonization, integration, and evaluation of data; robust stakeholder engagement; and support for the infrastructure and expertise needed to achieve the promise of PPH.

Acknowledgments

The findings and conclusions in this paper are those of the authors and do not necessarily represent the official position of Centers for Disease Control and Prevention.