вход по аккаунту



код для вставкиСкачать
Volume 00, Number 00, 2017
ª Mary Ann Liebert, Inc.
DOI: 10.1089/pop.2017.0146
Downloaded by Tufts University package NERL from at 10/28/17. For personal use only.
Augmented Intelligence:
Enhancing the Roles of Health Actuaries
and Health Economists for Population Health Management
Ian Duncan, FSA, FIA, FCIA, FCA, MAAA,1
Karen Fitzner, PhD,2 and Karen Ezekiel Handmaker, MPP3
chieving desired population health and business
outcomes requires new levels of productivity, efficiency,
and risk minimization. Machine learning (ML), cognitive
computing, natural language processing, and other augmented intelligence (AI) tools are increasingly applied to
health care initiatives. AI applies to a plethora of artificial
intelligence technologies and tools currently used in health
care. AI implies any kind of modeling that supplements
humans, augmenting human intelligence, not replacing it.
Health care decision makers are embracing AI, thereby
leveraging information from multiple sources to enable improved individual and population outcomes.1 ML, for example, applies algorithms and decision-making tools to
enormous amounts of clinical data from electronic health
records, pharmaceutical databases, and unstructured text
data.2 AI analytics are in predictive risk assessment, clinical decision support, home health monitoring, finance, and
resource allocation. These applications are integral to population health management (PHM), Accountable Care Organizations (ACOs), Medicare Advantage, and public health
initiatives. A recent hospital survey suggests a promising
future for AI that is particularly relevant to PHM. The survey
found that, of 7 applications, AI is likely to have the greatest
initial impact on population health (24%), clinical decision
support (20%), and patient diagnostic tools (20%), followed
by precision medicine (14%), and hospital/physician workflow (8%).3
A variety of AI applications are being tested for health
care. But growth and promise can only be met if the additional information is accessible and useful to CEOs and key
policy makers. Clinical informatics underlies efforts to improve quality of care and PHM.4 Considerable work is being
done by data scientists, such as Google/Sanofi’s joint venture, Onduo, that will collect and monitor real-time data on
people with diabetes.5 This initiative, like many others, will
fail unless there is a way to interpret the data, identify opportunities for intervention, and deliver the information in
real time to clinicians and patients. Results are likely to be
suboptimal if complex health care AI applications rely only
on data scientists. Even with excellent abilities, most data
scientists lack industry knowledge and the particular skills
of health actuaries (HAs) and health economists (HEs), such
as risk analysis and behavioral economics.
HA and HE expertise is vital to maximizing value from the
use of AI. These players guide health care business decisions,
policy making and operations improvements, creating useful
outputs for pricing, coverage decisions, business advances,
and policy making by applying sophisticated statistical modeling and analytic tools. Although there is overlap between the
abilities of actuaries and economists (Table 1), they bring
distinct and essential skill sets to an analytic team.
AI has been called the world’s most valuable resource as
it allows users to extract increasing, ongoing actionable
insights and value from myriad sources and uses of data.
HAs and HEs can leverage AI to provide expanded and
more specific guidance to payers, providers, suppliers,
programs, and health systems. For example, AI-enhanced
HA expertise helps ACOs to assess and pinpoint risk under
capitated contracts. HEs can incorporate the HA evaluation
and other AI-enabled analytics to recommend the optimal
deployment of care management resources to achieve contracted clinical and financial outcomes.
Blending PHM and AI-driven precision medicine could
yield a new health care services paradigm.6 The shift to
value-based care has increased the risk borne by providers
significantly and increases pressure to identify and manage
risk and evaluate outcomes. Innovative use of AI may help
identify short-term clinical goals, reduce risk to increase
savings, or serve as a return on investment indicator.7 Opportunities to enhance PHM exist with predictive modeling
for identifying high-risk patients. AI tools aid prediction and
help identify patients with multiple chronic conditions who
are moving across disease states/trajectories that are often
associated with increased resource use and cost.
Gaps exist. Predictive modeling is hampered by missing
pieces – what works and how it works – and gaps in comprehensive information. The common practice of segmenting populations by specific conditions (eg, diabetes) is
highly inefficient because most people have multiple
chronic conditions. Although HAs use predictive modeling
Department of Statistics and Applied Probability, University of California Santa Barbara, Santa Barbara, California.
FH Consultants, Sawyer, Michigan.
4sight Health, Louisville, Kentucky.
Downloaded by Tufts University package NERL from at 10/28/17. For personal use only.
Table 1. Roles of Health Actuaries and Health Economists
Health Actuaries (HAs)
Actuaries have been involved with health insurance for more than 50 years.a Traditionally, actuaries have compiled and
analyzed statistics to design benefit policies, assess insurance risks, and calculate premiums at the individual and payer
levels. In recent years health actuarial science has developed in new directions:
Predictive modeling: identifying future risks through assessing patterns of utilization. Modeling has become more
complex because of new sources of data (‘‘Big Data,’’ often unstructured, unformatted, and unvalidated) and new
models such as machine learning (ML) and artificial intelligence.
Revenue transfer based on relative population risk (‘‘risk adjustment’’). Although risk adjustment, a specific
application of predictive modeling, has been in use in the United States for more than 20 years, it has become more
widespread, and poses increasing challenges for traditional actuarial models (as the experience with Affordable Care
Act risk adjustment has shown).
Behavioral economics: actuaries have always assessed the effect of benefit plan design changes on insured member
behavior. However, as benefit designs have become more complex and the range of different options and programs
available to members has grown, actuaries have needed to be able to make more sophisticated predictions about
member behavior.
Value-based programs: many new value-based programs (eg, accountable care organizations [ACOs], bundled
payments, pay for performance) require substantial investments by payers; actuaries need to evaluate the financial and
clinical outcomes and determine whether this investment is warranted, or alternatively, what changes should be made
to improve the return on this investment.b
HAs add value and provide essential information for health care managers, clinicians, and employers on pricing and risk.
They also help integrated health systems identify effective means to transfer some or all of their risk. Actuarial science
is a major user of risk prediction, which is central to both clinical medicine and population health. Actuarial models
tend to be more basic, typically relying on generalized linear models, in part because of the effort required to acquire,
clean and warehouse data. Although they are used for research, more sophisticated and modern models (eg, ML, survival
modeling) have not yet found widespread use in health actuarial operations.c
Health Economists (HEs)
HEs focus on resource allocation and efficiency issues related to the financing and delivery of health services, and have done
so for more than 40 years. During that time, the field of health economics has demonstrated expertise in forecasting,
modeling, optimizing resources, and evaluating costs. Today, HEs incorporate augmented intelligence to undertake:
Trend analysis and impact forecasting: economic guidance for business and policy making includes tracking trends in
dynamic markets, developing scenarios to predict future changes, and modeling the market impact of new products,
services, or devices.
Sector analysis: encompasses assessment of potential effects of regulatory and other changes on the health care sector
at the national, state, or local level.
Cost containment: HEs use mathematical and statistical modeling (econometrics) to analyze factors that impact
financial and clinical outcomes. Cost-effectiveness and comparative effectiveness analyses are a key tool on which
health care systems, payors, and pharmaceutical firms have based considerable decision making.
Big data analysis: sophisticated econometric modeling uses vast amounts of health care data belonging to ACOs or
health care systems to achieve improved health for populations.
Behavioral modeling: the field of behavioral economics helps care managers and population health management
(PHM)-related organizations establish and refine patient engagement, incentive setting, and meet quality metrics.
HEs add value to health care entities by applying a variety of microeconomic tools to, for example, guide the CEO suite or
advise policy makers. They also take a broad health system focus, incorporating ML into models to achieve efficiencies in
PHM – identifying what works and what does not.
a. Society of Actuaries. Historical background. Accessed August 1, 2017.
b. Dyson S, Hardy B, Leung B. The role of the actuary in healthcare: where are we, and where are we going?
Library/S%20Dyson%20-%20The%20role%20of%20the%20actuary%20in%20healthcare.pdf Accessed May 28, 2017.
c. Makar M, Ghassemi M, Cutler DM, Obermeyer Z. Short-term mortality prediction for elderly patients using Medicare claims data. Int
J Mach Learn Comput. 2015;5:192–197.
to focus on the whole person with a high predicted financial
risk score, many PHM workflows are single condition focused to align with ‘‘care gaps’’ and related quality measures, which challenges HE principles of efficient resource
allocation. Also, data are limited on select interventions and
prescription data may not be available. Missing or incomplete data pose a major problem, highlighting a need for AI
coupled with impactful modeling for health care.8 With the
application of AI tools/techniques to nurses’ notes and images, for example, modeling is challenging but possible.9,10
Considerable work in health care is done by data scientists whose models may appear compelling but are not
transparent and are much harder to evaluate than HAs’
statistical models or HEs’ econometric models. This is of
concern. Although data scientists outnumber HAs and HEs,
they often do not understand health care delivery or health
care data (eg, Google’s 2013 flu episode).11
With the advances AI offers, the collaboration of HAs,
HEs, and data scientists may prove to be a ‘‘silver bullet’’ to
advance PHM and health care systems’ overall sustainability. The missing piece is the continuous interaction between
the HA risk analysis and the HE efficient allocation of resources. HAs and HEs can use AI to inform and guide health
entities to recalibrate to address changes in population and
Downloaded by Tufts University package NERL from at 10/28/17. For personal use only.
individual risk based on new information as it enters the
system. This approach is agile, efficient, reduces waste, and
drives positive outcomes.
AI is being adopted rapidly by health care systems, providers, and payers for highly data-dependent and data-driven
solutions. The promise is that it achieves efficiencies by
integrating information from multiple streams. The health
care community can look to HAs and HEs who possess
unique expertise in mathematical and statistical sciences
that naturally fit with the application of AI. Actuaries and
economic experts working in health care possess analytic
capabilities along with policy-advising experience and are
increasingly prepared to collectively advise on the sustainability and effectiveness of health care systems.
Author Disclosure Statement
The authors declare that there are no conflicts of interest.
The authors received no financial support for the research,
authorship, and/or publication of this article.
1. Neill DB, Heinz HJ. Machine learning for population
health and disease surveillance. College Carnegie Mellon
ations/Neill.pdf Accessed July 6, 2017.
2. HealthIT Analytics. Machine learning in healthcare: defining the most common terms. https://healthitanalytics
2936&elqat=1&elqCampaignId=2721 Accessed July 6, 2017.
3. Sullivan T. Healthcare IT News. Half of hospitals to adopt
artificial intelligence within 5 years. 2017.
tqeWpBOHExd3MifQ%3D%3D Accessed July 22, 2017.
Aston G. Powering the information engine. Hosp Health
Netw 2014;88:46–49, 1.
Reuters. Deals. Sanofi, Google parent form $500 million
diabetes joint venture. 2016. Accessed July
17, 2017.
IBM Institute for Business Value. Precision health and
wellness—the next step for population health management.
Accessed July 6, 2017.
Deagon B. AI meets ROI: where artificial intelligence is
already smart business. Technology. 2016. www.investors.
com/news/technology/artificial-intelligence-weaves-its-waydeeper-into-daily-business Accessed July 17, 2017.
Pierce D, Young E. A whole new world of data challenges
and opportunities with electronic health records. 2016. www Accessed
July 22, 2017.
Makar M, Ghassemi M, Cutler DM, Obermeyer Z. Shortterm mortality prediction for elderly patients using Medicare
claims data. Int J Mach Learn Comput 2015;5:192–197.
Zheng L, Wang Y, Hao S, et al. Web-based real-time case
finding for the population health management of patients
with diabetes mellitus: a prospective validation of the
natural language processing-based algorithm with statewide
electronic medical records. JMIR Med Inform 2016;4:e37.
Butler D. When Google got flu wrong. US outbreak foxes a
leading web-based method for tracking seasonal flu. 2013.
Accessed August 1, 2017.
Address correspondence to:
Karen Fitzner, PhD
FH Consultants
554 Hillcrest Drive
Sawyer, MI 49125
E-mail: [email protected]
Без категории
Размер файла
64 Кб
2017, 0146, pop
Пожаловаться на содержимое документа