ll

Leading Edge

Commentary
Precision medicine in 2030— seven
ways to transform healthcare
Joshua C. Denny1,3,* and Francis S. Collins2
1
All of Us Research Program, National Institutes of Health, Bethesda, MD, USA
2
National Institutes of Health, Bethesda, MD, USA
3
Present address: Bldg. 1 Room 228, 1 Center Drive, Bethesda, MD 20814, USA
*Correspondence: [email protected] https://doi.org/10.1016/j.cell.2021.01.015

Precision medicine promises improved health by accounting for individual variability in genes, environment, and
lifestyle. Precision medicine will continue to transform healthcare in the coming decade as it expands in key areas: huge
cohorts, artificial intelligence (AI), routine clinical genomics, phenomics and environment, and returning value across
diverse populations.
Ever since the completion of the first human power of amalgamated healthcare data. researchers can start using the All of Us
genome sequence in 2003, clinicians Pooling Research Program’s data cloud in as little as
haveanticipateda data-driventransformation two hours after initial login.
in healthcare. New troves of molecular and The next step is clear: make it easier for
phenotypicinterrogation would lead to data from existing research cohorts enabled researchers to merge data from multiple
refined diagnoses, more rational treatment, rapid genomic studies that have identified cohorts. Currently, this requires painstaking
and prevention of disease. In 2011, an ad hoc loci associated with disease susceptibility manual phenotype adjudication and building
committee at the National Research Council and patient outcomes. COVID-19 has also large consortia including experts from each
argued for a ‘‘new taxonomy of human called attention to the need for longitudinal cohort. Fortunately, there are efforts
diseases’’ based on the emerging field of cohorts to identify clinical and biologic risk underway to improve this process. Groups
precision medicine (US National Research factors and long-term sequelae for acute such as the Global Alliance for Genomics and
Council, 2011). infectious disease. Many of the elements of Health (GA4GH) are working to develop and
Today, some of that promise has already the response to COVID-19 are basic to coordinate common data models and file
been realized. Researchers are routinely capabilities underpinning precision medicine. formats to facilitate collaboration and
using healthcare data for discovery, At the same time, COVID-19 has interoperability. In recognition of the need
identifying genomic underpinnings of cancer highlighted the need for precision medicine for better collaboration, the International
and many other common and rare diseases, to move further and faster. In this paper, we Hundred
introducing transformative molecularly suggest seven opportunities to accelerate an Thousand Plus Cohort Consortium (IHCC) has
targeted therapies, and leveraging massive equitable realization of the promise of brought together more than 100 cohorts in
computational capabilities with new machine precision medicine (Figure 1). Their impacts 43 countries comprising more than 50
learning methods. We are beginning to see are outlined in Table 1. million participants—nearly two orders of
the fruits of these efforts. magnitude bigger than the biggest single
There is perhaps no more poignant Huge, interoperable, longitudinal cohorts cohort today (Manolio et al., 2020). It would
example than the response to the COVID-19 Over the last two decades, national cohorts be hard to overstate the impact this work
pandemic. Genomics and molecular such as the UK Biobank, the Million Veteran could have on global research efforts.
technologies were key in identifying the Program, FinnGen, and the All of Us
etiologic agent, developing diagnostics and Research Program have amassed huge Improved diversity and inclusion in science
treatments, and creating vaccine candidates. populations with genomic, laboratory, and One of the biggest challenges (and
Rapid case reporting quickly exposed vast lifestyle assessments as well as longitudinal opportunities) before the biomedical
health disparities with COVID-19 and follow-up on health outcomes. The depth enterprise today is the lack of diversity in
highlighted the importance of capturing a and breadth of the data are staggering, as populations involved in research studies.
more detailed understanding of social are the opportunities for discovery across Less than 3% of the participants in published,
determinants of health. Large-scale consortia every area of medicine. genome-wide association studies are of
based on healthcare data quickly assembled In order to maximize the impact of these African or Hispanic or Latin American
huge datasets for rapid investigations of risk resources, an ‘‘open science’’ approach is ancestries, and 86% of clinical trial
factors and outcomes, demonstrating the emerging. For example, the UK Biobank has participants are white (Knepper and McLeod,
opened its doors to more than 19,000 ‘‘bona 2018; Mills and Rahal, 2020). The lack of
fide researchers’’ from 80 countries, and diversity in research risks exacerbating
Cell 184, March 18, 2021 Published by Elsevier Inc.
Figure 1. Seven opportunities for precision medicine by 2030 health disparities and also impoverishes standing, inn
1415
 ll
biologic discovery that could be applicable sensitivity; is more likely to pursue questions availability of devices and wearables to
to all populations. relevant to different audiences; and provide highresolution data streams will
With a growing depth of data, we have an ultimately delivers better research ( dramatically expand the availability of
opportunity to replace adjustments for race et al., 2020). detailed phenotype and environmental data
and ethnicity with more specific measures. As international collaborations grow, not previously available at this scale.
In particular, ‘‘race’’ conflates a plethora of researchers will also need to consider the Applications of machine learning
social, cultural, political, geographic, and ethics of international collaboration and approaches could result in new taxonomies
biologic factors together and can perpetuate rotate leadership, authorship, and resources of disease through genomic, phenomic, and
systemic racism. Routine collection of social to ensure that research benefits developing environmental predictors.
determinants of health in both research and countries as well as more advantaged ones.
clinical care in combination with more Establishing international infrastructures Routine clinical genomics to guide
precise measures of environmental and science facilities—not just access to prevention, diagnosis, and therapy Today,
influences, habits, and genetic ancestry can samples and data—will produce long-term clinical genomic analysis is typically
provide more rational, etiology-based benefits that accelerate health and performed only when evaluating certain
adjustments and yield better risk capabilities. cancers or when a rare genetic disease is
stratifications and treatments (Wilkins et al., suspected, and many commonly ordered
2020). Big data and artificial intelligence tests only evaluate a few genetic loci.
As we work toward increasing the and artificial intelligence (AI) are Moving forward, wholegenome approaches
diversity of populations in studies, we transforming previously intractable will become a routine, early step in the
should also increase the diversity of the problems such as search optimization, understanding, prevention, detection, and
biomedical research workforce. A more language translation, image interpretation, treatment of common and rare diseases.
diverse workforce—in culture, ancestry, and autonomous driving. Many accrued Rare diseases will increasingly be
beliefs, scientific backgrounds, and biomedical data sets meet all ‘‘5 V’s’’ of big diagnosed using genomic investigation as a
methodological approaches—brings data since they are voluminous, high cheaper and more efficient alternative to
increased under- targeted approaches. Early genome
sequencing can solve diagnostic dilemmas
and uncover ‘‘hidden’’ Mendelian diseases
such as unexplained kidney disease, atypical
diabetes, or unexplained development
delay (Turro et al., 2020). Some of these
Mendelian diseases point to specific new
1416 Cell 184, March 18, 2021 treatments and screening strategies that
could dramatically improve health, such as
Commentary sulfonylureas for young diabetic patients
with HNF1A mutations or specific causes of
velocity, come in many varieties, have significant variability, liver or kidney failure.
and have intrinsic value. However, AI approaches in medicine The last decade has also shown that many
have been limited by the (un)availability of large, commonly common conditions, such as diabetes or
structured datasets. hypertension, can be associated with genetic
Looking forward, biomedical datasets will become risks at thousands of loci, often found using
increasingly ready for analyses. As we discuss in the following huge genetic studies aggregating data across
sections, the growth of clinical data (including image, hundreds of thousands of participants. While
narrative, and real-time monitoring data), molecular many of these genetic loci may have very
technologies (genomics principal among them), and the small
Commentary
Table 1. Envisioning how precision medicine will affect clinical medicine and research in the next decade
Where we are today Where we will be in 2030
Clinical applications

Genomics for disease Genomics is routine. Genetic causes and targeted therapies are
Primarily limited to rare disease
discovered for many ‘‘common’’ diseases.
and select cancers.
Microbiome measures are routinely included.

Pharmacogenomics (PGx) Common in cancer and within select Genome-aware EHRs make PGx easy and automatically update
applications of older medications at rules from central guidelines. New PGx associations discovered
select sites. from clinical data.
 ll
Genomics for healthy individuals In research, whole-genome ACMG59 grows to > 200, variant interpretation improved by
sequencing and search for huge, diverse sequenced populations.
mutations in one of the ACMG59 Cell-free DNA becomes a mainstay of cancer screening
genes, present in about 3% of
people. Variant interpretation is
hard.

EHRs Episodic capture from healthcare Genome- and device- enabled. Data can be easily moved between
without robust genomics support. EHRs and to participant apps.
EHR data is essentially not portable.
Environmental influences on health Patient-reported habits and exposures Geocode-based exposure linkage
Real time monitoring of multiple environmental exposures
Precision nutrition

Wearable sensors Ad hoc use of activity monitors Continuous monitoring of physical activity, sleep, metabolic
parameters
Research applications

Population demographics >80% European ancestry >50% non-European ancestry

Routinely available data Surveys of health conditions, lifestyle, Whole genomes, lab assays, surveys, full EHRs, environmental,
behavior, and diet. GWAS data, lab genomic and sensor data. Includes imaging, narrative,
assays, structured EHR data, and geocoded, and continuous monitoring approaches to clinical
geocoded exposure linkages. care, activity, precision nutrition, and environment.

Size of cohorts used in analysis Up to 500K, data downloaded and >100M using cloud-based federated analyses facilitated by
manually harmonized to sets of common standards
several million

Largest genomic studies performed >1M (GWAS) >50M (GWAS)

on a trait >2M (WGS)
Cost of a whole genome $500 $20*

*
Sequencing costs have often fallen faster than Moore’s law. Using Moore’s law, sequencing costs would be 1/32 of US $500, or $15.63.

genetic effect sizes (with odds ratios < 1.01), classes of CFTR mutations with cystic fibrosis majority of which were previously unknown
they point to pathways involved in disease or SERPINA1 variants with alpha-1 to the patient. As genomic knowledge
pathogenesis that may have significant antitrypsin deficiency, both of which can
therapeutic implications. Furthermore, present with different manifestations and at
weighted aggregations of genetic variants varying ages given the genetic variant, habits Cell 184, March 18, 2021
into polygenic risk scores can achieve similar (e.g., smoking), and exposures (e.g., hepatitis increases, the number of actionable genes
predictive as rare Mendelian disease variants virus coinfections). and the fraction of the population affected
(Khera et al., 2018). Moreover, use of Routine use of sequencing will produce will significantly increase.
polygenic risk scores may allow providers to valuable datasets for secondary research, Furthermore, pharmacogenomics can
risk-stratify individuals who would otherwise driving a more comprehensive improve drug efficacy, reduce adverse
be missed by traditional screening understanding of disease penetrance, variant events, and reduce cost. In a 2009
approaches, thereby identifying new pathogenicity, and factors influencing interview, one of the authors of this article
populations for treatment or screening. variable expressivity of given genetic (F.S.C.) made the comment, ‘‘if
We anticipate that diverse genetic causes variants. It will also produce more patients everybody’s DNA sequence is already in
and targeted therapies will be uncovered for for whom incidental pathogenic variants are their medical record and it is simply a click
many common diseases, which could lead to discovered. The American College of Medical of the mouse to find out all the
more specific treatment and prevention for Genetics and Genomics has identified 59 information you need, then there is going
the patient and family members. We will genes for which incidental findings should be to be a much lower barrier to beginning to
likely also discover that many genetic considered for patient return (i.e., the incorporate that information into drug
diseases occur on a spectrum of severity, ‘‘ACMG59’’) (Kalia et al., 2017). These genes prescribing’’ (Collins, 2009). Over a decade
penetrance, and expressivity, guided by the include hereditary cancer syndromes, later, we still have a long way to go. While
severity of different genetic variants, cardiomyopathies, and potentially fatal genomics-guided therapies are becoming
lifestyle, and environmental interactions. arrhythmias, for which actions can be taken the standard of care for some cancers, use
This concept is captured by the scientific to mitigate their risk. Today, about 3% of of germline pharmacovariants to guide
agenda of the International Common Disease patients harbor pathogenic variants, the vast prescribing has been adopted by only a
Alliance. Classic examples include different few US medical centers. Implementation

1417
 ll
has been EHR data require cleaning and Moreover, integration of these tools could
hindered by a lack of ‘‘genomics-enabled’’ harmonization and can reflect clinical and produce a shift in which most health-related
electronic health records (EHRs), the insurance biases. Unstructured EHR data, data is derived outside of the healthcare
complexity of the genetics and such as narrative reports or imaging data, setting.
recommendations, and a lack of clear often require advanced methods like natural Despite clear evidence of the impact of
evidence. Synthesized evidence and language processing or machine learning to nutrition on health, diet is an environmental
recommendations from the Clinical be useful on a population scale. However, all exposure often ignored in much of clinical
Pharmacogenomics Implementation of these tools are increasingly available and practice and many research studies. When it
Consortium, ubiquity of EHRs supporting applicable, providing access to data on a is assessed, it is often through episodic and
complex decision support, and common scale, depth, and detail not feasible with cumbersome surveys (research) or
data standards offer promise to accelerate purely research-collected data. perfunctory summative questions (in most
adoption. Some countries have Clinical EHR data can also be combined clinical settings). Replacing dietary
substantially reduced drug-induced with participant-provided research data assessment with data linkages to grocery
Stevens Johnson Syndrome using genetic collections to provide a more complete stores, digital uploads from restaurants,
testing (White et al., 2018). Even picture of patient outcomes. Research laboratory and microbiome assessments, or
considering just five drug-genomic cohorts such as the UK Biobank and All of Us machine learning applied to food imaging
interactions, nearly everyone has a have integrated both data sources. would provide more feasible, comprehensive
pharmacovariant that would affect drug Further, as clinical sequencing grows, the capture of dietary habits. A future of
prescribing (Van Driest et al., 2014). number of genotypes derived from clinical precision nutrition, as a type of ‘‘drug,’’
care will rapidly grow to dwarf those offers a powerful new modality for treating
EHRs as a source for phenomic and available from research use cases. Many and preventing disease (Rodgers and Collins,
genomic research genomic studies may no longer need 2020).
The key to any longitudinal cohort is separate research biospecimen collection to
detailed phenotype, exposure, and health perform large-scale genetic studies. Privacy, participant trust, and returning
outcome assessment. Many site-based and Collection of research biospecimens could value
national research cohorts now use EHRs then shift toward measuring other The utility of precision medicine is
and other health data to provide up to biomarkers, cell-free DNA, exposures, and dependent on broad participation, and broad
decades of extant disease and treatment epigenomics. participation of large populations requires
information that can be repurposed for trust, protection of privacy, and a return of
research, and we only see this use Higher variety, higher resolution phenomics value to the participants. We recognize that
expanding. and environmental exposure data for both science has not always been trustworthy or
Already EHR-based studies have been clinical and research use honored all participants equally.
instrumental to some of the largest The next decade will see the continued Transparency, authentic engagement with
genomic studies of clinically relevant growth of research and clinical uses for communities, and including participants
findings, some of which are exceeding 1 different ways to measure clinical within research governance can improve
million individuals (Vujkovic et al., 2020). phenotypes, exposures, and lifestyle. Data trust, create participant advocates, and
By providing a systematic collection of linkages to health claims, national vital ensure a more thoughtful, culturally sensitive
health-related information, EHRs provide statistics, and geospatial resources will direction. All of Us has involved participants
become more common as will the use of in all levels of governance from the
wearable devices to measure activity, beginning and seeks to return value by giving
1418 Cell 184, March 18, 2021 participants generated research data
phenotypes and data and enable novel study Commentary wherever
designs often not available in research
collections. For example, one study physical measurements, and exposures. Commentary
demonstrated participants had an average of Surveys can then be more focused on
more than 190 clinical notes, 14 radiological elements not covered by other methods, possible, such as genomics results or
studies, and more than 700 lab tests over an thereby decreasing participant burden. upcoming COVID-19 serology results.
average of about 8 years of follow up Activity monitors that take a number of Participants also need to trust that their
(Robinson et al., 2018). The power to clinical measurements such as singlelead data is secure and private. Highly public data
discover specific endophenotypes (e.g., electrocardiograms and oxygen saturation breaches, fear of reidentification, and legal
cardiac ejection fraction) or emerging are becoming inexpensive and can be easily concerns about the availability of certain
phenotypes (e.g., COVID-19), rare and shared with providers. Since the vast types of information for factors such as
specific phenotypes (e.g., osteonecrosis of majority of a patient’s life is spent outside insurability can make this challenging. Clear
the jaw), or to understand specific the healthcare system, integration of and honest communication with participants
manifestations of disease (e.g., wearable devices and other patient-provided is essential in building trust. Legal
bronchiectasis) often requires access to information would augment the EHR and protections for the data and technological
complete EHR data. enable greater telehealth capabilities, approaches to ensure secure information
experienced first at scale during COVID-19. systems (such as deidentifying and blurring
 ll
data, controlling access via blockchain, global scientific challenges. Lancet Digit Health 2, Cell
linking data using privacy-preserving hashed e567–e568. 184, March 18, 2021
identifiers, and analyzing data using Mills, M.C., and Rahal, C. (2020). The GWAS Diversity
homomorphic encryption) also play a role. Monitor tracks diversity by disease in real time.
Nat. Genet. 52, 242–243.

Conclusion Robinson, J.R., Wei, W.-Q., Roden, D.M., and Denny,

The technologies undergirding precision J.C. (2018). Defining Phenotypes from Clinical Data
to Drive Genomic Research. Annu. Rev. Biomed.
medicine are already transforming care.
Data Sci. 1, 69–92.
Transformative molecular treatments have
been developed for rare diseases like cystic Rodgers, G.P., and Collins, F.S. (2020). Precision
Nutrition-the Answer to ‘‘What to Eat to Stay
fibrosis and spinal muscular atrophy.
Healthy’’. JAMA 324, 735–736.
Genomic investigation led to the
development of new drugs for Turro, E., Astle, W.J., Megy, K., Gra¨ f, S., Greene, D.,
Shamardina, O., Allen, H.L., Sanchis-Juan, A.,
hyperlipidemia. In this time of COVID-19,
Frontini, M., Thys, C., et al.; NIHR BioResource for
science has been the answer to an existential the 100,000 Genomes Project (2020).
medical threat. Yet we are reminded that Wholegenome sequencing of patients with rare
many of the benefits of medicine’s diseases in a national health system. Nature 583,
advancement have not always been available 96–102.
to all. Biomedical approaches, computation US National Research Council (2011). Toward
algorithms, and the availability of high- Precision Medicine: Building a Knowledge Network
resolution data will dramatically increase for Biomedical Research and a New Taxonomy of
over the next decade. Implementation of a Disease (Washington, DC: National Academies Press
(US)).
bold plan to collaborate internationally, to
engage diverse populations of participants Van Driest, S.L., Shi, Y., Bowton, E.A., Schildcrout,
J.S., Peterson, J.F., Pulley, J., Denny, J.C., and Roden,
and scientists, to deeply measure our
D.M. (2014). Clinically actionable genotypes among
populations, to make clinical and research 10,000 patients with preemptive pharmacogenomic
data broadly available, and to implement this testing. Clin. Pharmacol. Ther. 95, 423–431.
knowledge in clinical practice—in a true
Vujkovic, M., Keaton, J.M., Lynch, J.A., Miller, D.R.,
learning healthcare system—will allow us to Zhou, J., Tcheandjieu, C., Huffman, J.E., Assimes,
achieve the vision of precision medicine for T.L., Lorenz, K., Zhu, X., et al.; HPAP Consortium;
all populations. Regeneron Genetics Center; VA Million Veteran
Program (2020). Discovery of 318 new risk loci for
type 2 diabetes and related vascular outcomes
REFERENCES among 1.4 million participants in a multi-ancestry
meta-analysis. Nat. Genet. 52, 680–691.
Collins, F. (2009). Opportunities and challenges for the
White, K.D., Abe, R., Ardern-Jones, M., Beachkofsky,
NIH–an interview with Francis Collins. Interview by
T., Bouchard, C., Carleton, B., Chodosh, J., Cibotti, R.,
Robert Steinbrook. N. Engl. J. Med. 361, 1321–1323.
Davis, R., Denny, J.C., et al. (2018). SJS/TEN 2017:
Hofstra, B., Kulkarni, V.V., Munoz-Najar Galvez, S., He, building multidisciplinary networks to drive science
B., Jurafsky, D., and McFarland, D.A. (2020). The and translation. J. Allergy Clin. Immunol. Pract. 6,
Diversity-Innovation Paradox in Science. Proc. Natl. 38–69.
Acad. Sci. USA 117, 9284–9291.
Wilkins, C.H., Schindler, S.E., and Morris, J.C. (2020).
Kalia, S.S., Adelman, K., Bale, S.J., Chung, W.K., Eng, C., Addressing health disparities among minority
Evans, J.P., Herman, G.E., Hufnagel, S.B., Klein, T.E., populations: why clinical trial recruitment is not
Korf, B.R., et al. (2017). Recommendations for enough. JAMA Neurol. 77, 1063–1064.
reporting of secondary findings in clinical exome and
genome sequencing, 2016 update (ACMG SF v2.0): a
policy statement of the American College of Medical
Genetics and Genomics. Genet. Med. 19, 249–255.
Khera, A.V., Chaffin, M., Aragam, K.G., Haas, M.E.,
Roselli, C., Choi, S.H., Natarajan, P., Lander, E.S.,
Lubitz, S.A., Ellinor, P.T., and Kathiresan, S. (2018).
Genome-wide polygenic scores for common diseases
identify individuals with risk equivalent to monogenic
mutations. Nat. Genet. 50, 1219–1224.
Knepper, T.C., and McLeod, H.L. (2018). When will
clinical trials finally reflect diversity? Nature 557, 157–
159.
Manolio, T.A., Goodhand, P., and Ginsburg, G. (2020).
The International Hundred Thousand Plus Cohort
Consortium: integrating large-scale cohorts to address

1419