Published online: 2019-04-25

35
© 2019 IMIA and Georg Thieme Verlag KG

Role of Artificial Intelligence within the

Telehealth Domain
Official 2019 Yearbook Contribution by the members of IMIA
Telehealth Working Group
Craig Kuziemsky1, Anthony J. Maeder2, Oommen John3, Shashi B. Gogia4, Arindam Basu5,
Sushil Meher6, Marcia Ito7
1
Telfer School of Management, University of Ottawa, Ottawa, Canada
2
College of Nursing & Health Sciences, Flinders University, Adelaide, Australia
3
  George Institute for Global Health, University of New South Wales, New Delhi, India
4
  Society for Administration of Telemedicine and Healthcare Informatics, New Delhi, India
5
  University of Canterbury School of Health Sciences, Christchurch, New Zealand
6
All India Institute of Medical Sciences, New Delhi, India
7
IBM Research, Brazil

Summary Introduction come challenges of health service delivery

due to time, distance, and difficult terrains,
Objectives: This paper provides a discussion about the potential The IMIA Telehealth Working Group con- enabling cost-effectiveness and better access
scope of applicability of Artificial Intelligence methods within the sists of over 30 members worldwide, com- in both developed and developing world
telehealth domain. These methods are focussed on clinical needs prising clinicians and telehealth experts settings. It has a special role during emer-
and provide some insight to current directions, based on reports working as academics, data scientists, gencies like earthquakes and floods [2].
of recent advances. entrepreneurs, promoters, and researchers. Additionally, longer lifespan and associated
Methods: Examples of telehealth innovations involving Artificial This invited article outlines the potential rising incidence of chronic diseases have
Intelligence to support or supplement remote health care delivery scope of applicability of Artificial Intelli- increased care demand and complexity,
were identified from recent literature by the authors, on the basis gence (AI) methods within the telehealth necessitating longer interactions between
of expert knowledge. Observations from the examples were syn- domain. patients and providers (as well as among
thesized to yield an overview of contemporary directions for the We consider the importance, current, and providers), and thereby increasing the need
perceived role of Artificial Intelligence in telehealth. possible future uses of AI, and then outline for tele-healthcare support [3].
Results: Two major focus areas for related contemporary direc- the potential opportunities for addressing Telehealth has been classified histori-
tions were established. These were first, quality improvement for some of the system level challenges in cally as synchronous using real-time elec-
existing clinical practice and service delivery, and second, the implementation. We start by discussing the tronic communication, or asynchronous
development and support of new models of care. Case studies overall role of AI in telehealth after which we using store-and-forward communication.
from each focus area have been chosen for illustration purposes. provide some specific scenarios wherein AI Recently a third form has been recognised
Conclusion: Examples of the role of Artificial Intelligence in delivery can improve health outcomes and enhance as remote (tele) monitoring, involving
of health care remotely include use of tele-assessment, tele-diag- stakeholder experience and acceptance. We data collection through distributed devices
nosis, tele-interactions, and tele-monitoring. Further developments conclude with some comments on social and including Internet of Things (IOT). The
of underlying algorithms and validation of methods will be required ethical considerations. last World Health Organisation global
for wider adoption. Certain key social and ethical considerations also
eHealth observatory survey [4] noted
need consideration more generally in the health system, as Artifi-
four exemplary well-embedded telehealth
cial-Intelligence-enabled-telehealth becomes more commonplace.
services: tele-radiology, tele-pathology,
Keywords Telehealth Paradigm in tele-dermatology and tele-psychiatry; of
Telemedicine; Artificial Intelligence; quality of health care; Health Care Delivery these, the first three follow asynchronous
models of care and the fourth is synchro-
delivery of health care
Telehealth uses information and commu- nous. This demonstrates an inherent diffi-
Yearb Med Inform 2019:35-40 nication technologies to transfer medical culty in replacing or augmenting real-time
http://dx.doi.org/10.1055/s-0039-1677897 information for the delivery of clinical and clinician-delivered services. Furthermore,
educational services [1]. It attempts to over- the same survey revealed that over 60% of

IMIA Yearbook of Medical Informatics 2019

36
Kuziemsky et al.

respondents identified lack of information When coordinated care and continuum of health record systems, and automated
on telehealth-related clinical practice to be a care pathways are not followed, health care aggregation of patient information through
great impediment to wider uptake. equality as well as quality suffer. There is proliferation of healthcare information tech-
a need for seamless communications and nology. The availability of large data sets
connections to occur across different ele- combined with the rapid evolution of com-
ments of health care delivery: not all team putational data science (including AI-based
Artificial Intelligence in members can always be present, which machine learning methods) offer promising
Telehealth creates a need for tele-support or remote
care. AI again could help address this need,
opportunities for extracting new inferences
and actionable insights that have a potential
A recent analysis of contemporary trends in both by enabling the intelligent information to improve health outcomes significantly.
telehealth [5] proposed that two dominant and communication environment in which This more sophisticated data-enriched
drivers for change were emerging: i) high clinicians could interact as well as maintain environment in turn allows better clinical
volume demand, due to the increasing diffi- a detailed virtual knowledge base for the decision-making through support by auto-
culty to physically co-locate the patient, the progression of the patient’s condition and mated means, encouraging moves towards
clinician(s), and the associated data, and ii) his/her management. intelligent assistance and diagnosis.
high criticality applications, where special- Russell and Norvig [6] in 1995 described
ised expertise is necessary at the precise the field of artificial intelligence compre-
moment of clinical demand. Irrespective of hensively as including “problem solving Example: Clinical Assessment and
the modality used, actual delivery of care and search, logic and inference, planning,
inevitably necessitates some face-to-face probabilistic reasoning and decision mak- Evaluation
clinical interaction, with the required peri- ing, learning, communication, perception, Before routine use of high-end diagnostic
odicity depending on the circumstances. In and robotics”. They further contended that tools and imaging devices like MRI and CT
telehealth, Information and Communication computers, in the domain of artificial intel- became common, clinical evaluation largely
Technology (ICT) tools can be used to ligence, could serve as “intelligent agents” depended on a simple personal history and
address the issues of misdistribution of the and would mimic, if not match, humans in examination. Peterson et al. in 1992 [8]
demand versus supply of healthcare services. thought processes and reasoning to the extent reported a 76% contribution by history and
AI could assist with this issue by developing that computers might emulate human cogni- 11% by examination for an initial diagnosis.
algorithms to match the availability of care tive behaviour and mimic human cognitive Roshan and Rao in 2000 [9] found the role
providers with appropriate clinical skills to performance. Considering this wide range of history to be roughly the same but the
the need for such skillsets in the immediate of possibilities, Pacis et.al. [7] recently role of examination to be even less (7.6%).
vicinity. However, telehealth introduces summarised the potential impact of AI in As of today, the art of history taking and
several operational issues, such as when telehealth around four emergent trends based examination is getting compromised and
the telecommunication link fails, or when on distinctive health care purposes: patient health care is suffering as a result.
the remote care clinician is not available monitoring, healthcare information technol- Investigations make telecare easier as
remotely. AI could potentially alleviate such ogy, intelligent assistance and diagnosis, and such data is easy to collect and transfer. The
situations by providing mechanisms for information analysis collaboration. We argue contribution of high-end imaging Ultrasound
human or virtual interactions to occur, and that they can be merged to two main areas (gallstones, liver abscess), CT (tumor in
thereby address difficulties in timing and of focus and become topics which we dis- frontal area), MRI, etc., towards making a
availability of clinicians (such as the time cuss in the rest of this paper: firstly, quality conclusive diagnosis has downgraded the
taken to understand the patient’s problem or improvement for existing clinical practice importance of history taking and sometimes
taking a history). and service delivery, and secondly, the devel- even obviates the need for a cursory clinical
Increasing innovation and discovery opment and support of new models of care. examination. The quality increase in health-
is making care provision more complex, care delivery from this added sophistication
beyond the scope of any single clinical has not been commensurate with the costs
provider. AI can support the development incurred. History taking is painstaking and
of knowledge of clinical processes: for Telehealth, AI, and Quality consumes much of a physician’s time, so it
example, how to leverage new innovations in
therapies and procedures through personali- Improvement tends not to be used to the extent feasible via
telemedicine. This is regrettable, as that is a
sation to match patient cohorts or individual In recent years, there has been a rapid and component which is easily possible remotely
profiles. Increasing longevity and chronic exponential increase in the quantum of and requires no special equipment. High end
disease management in multi-morbidity health-related digital data that is generated investigations, on the other hand, require
situations needs a multi-disciplinary team by the citizens themselves as well as health- much more remote infrastructure making
approach with continued care being provided care providers. There has consequently been it costly and decreasing the cost benefit
within the community, ideally at home. a movement towards universal electronic advantage of telecare.

IMIA Yearbook of Medical Informatics 2019

 37
Role of Artificial Intelligence within the Telehealth Domain

For an experienced clinician, many clues A well-known existing area for tele-diag- capabilities for health care offered by patient
pointing to a person’s problem lie in the nosis is tele-dermatology, which lends itself monitoring must therefore be exercised in
history. Not only does it guide the diag- well to automation through AI. Currently, an environment of information analysis and
nosis, but it also leads to a more focussed diagnostic accuracy for melanoma is depen- collaboration by care team members and
investigation approach. AI can make history dent on the experience and training of the health systems agencies.
taking easier by providing prompts to the treating doctor. A recent study has shown that
process and clues to the diagnosis, and also a computer algorithm using convolutional
asking the right next set of questions based neural networks outperformed the majority
on the answer saving the clinician time. For of 58 dermatologists tested in accurately
Example: Conversational Agents
example, a person suffering long-term with diagnosing melanoma, with a median area and Virtual Assistants
chronic, dull aching pain in the upper abdo- under the receiver operating characteristic A natural progression from human-to-hu-
men without interference with sleep is likely curve of 0.86 compared with 0.79, P < 0.01 man interaction in telehealth is the devel-
suffering from gastritis. A summary of such [12]. Another study demonstrated classifi- opment of computer-based generation and
questions being asked in sequence can be cation of skin lesions using a deep convo- understanding of conversation to enable
delivered as part of a telehealth application, lutional neural networks (CNN), trained computer-to-human interaction. Online
and may be implemented efficiently using end-to-end from images directly, using only therapeutic and health counseling interven-
mobile ICT. pixels and disease labels as inputs. The CNN tions have long acknowledged the utility
Such systems can even support decision achieves performance on par with all tested of the technology that allows for variations
making when it is impossible to contact a experts across tasks of identification of the between human-guided, patient-guided, and
clinician e.g., chest pain with likelihood most common cancers, and identification of computer-guided approaches [15]. The value
of myocardial infarction means an urgent malignant melanoma the deadliest skin can- of synchronous and task-oriented comput-
consideration for the use of streptokinase cer, demonstrating an artificial intelligence er-generated dialogue has been observed for
or at least Sorbitrate or Aspirin to be capable of classifying skin cancer with a mental health applications [16] and a broader
administered remotely. Benefits from such level of competence comparable to derma- range of applications has been noted in
questions and conclusions thereof can be tologists [13]. Similar promise is offered in recent years [17]. Automated conversational
availed of directly by the patient with the other areas of automated diagnosis such as interactions offer many other opportunities
assistance of a local nurse practitioner. Such breast cancer or cervical cancer screening. across the care spectrum to augment and in
questioning can be assisted by easy to use some cases replace human carer tasks. These
symbols within the user interface which all may include:
can understand. • reminders and motivational messages e.g.,
Telehealth, AI, and New for medication, nutrition, and exercise;
Models of Care • routine condition checks and health main-
Example: Tele-diagnosis of Clinical Globally there has been a significant increase
tenance, based on personal monitoring
data;
Conditions in burden of chronic diseases. This com- • answering of health queries and provision of
Medical diagnosis has moved from clini- bined with a growing base of aging popu- targeted health information and education;
cal-examination-based in earlier days, to lations with multiple morbidities, and the • providing a personalised means to
largely evidence-based supported by the existing models of healthcare delivery are address social isolation and community
doctor’s own interpretation based on his/ overstressed and unsustainable. Telehealth involvement;
her experience and skill. It is here that AI offers promising alternatives to utilize ICT • acting as an intermediary or broker
in last few years has begun to play a highly optimally for remote healthcare diagnosis, entity between multiple carers or service
supportive role to doctors when analyzing monitoring, and delivery of care. Neverthe- agencies.
the evidence. This has special relevance in less, a number of system level challenges
oncology [10]: considering the progress of have restrained transformative telehealth The nature and complexity of conversational
diseases that could relate to the formation of models that have been scaled up at national agent (or virtual assistant) solutions can vary
cancers, different disease diagnosis patterns or regional levels to exploit the full poten- considerably. For simple tasks, which are lit-
might reflect different chances of cancer tial of delivering healthcare over distance. tle more than a message or cue for informing
risks. Modeling disease progression and A recent review of telehealth interventions the recipient, an audio voice or visual text
variants in disease trajectories helps in pre- concluded that the implementation of a com- communication is often adequate, and can
diction [11]. By applying machine learning plex innovation such as remote care requires also convey a minimal complexity response
methods to large datasets of disease popula- it to organically evolve, be responsive and such as a confirmatory acknowledgement.
tions, AI is on its way globally to making a adaptable to the local health and social care Speech-text conversion utilities and chatbots
significant impact on the way doctors could system and driven by support from front-line capable of audio or typed inputs and outputs
diagnose a potential disease. staff and management [14]. The increased are examples of such technologies. They

IMIA Yearbook of Medical Informatics 2019

38
Kuziemsky et al.

may have highly constrained conversational Example: Remote Patient changing the contrast of the image to help
models embedded in them, capable of recog- the sizing and automatic calculation of
nising and constructing only a limited range Monitoring and Management boundary and area. Manual oversight and
of phrases for a prespecified purpose. These Remote monitoring (or tele-monitoring) correction are still deemed to be needed
solutions are better suited for interactions involves data acquisition using an appro- and may continue to be a necessary option
where the context of the situation and the priate sensor, transmission of data from for much AI-based tele-monitoring at least
user are simple and clearly established. AI patient to clinician, integration of data with in the near future.
mechanisms for these agents are typically other data describing the state of the patient
rule-based using expert systems or decision (e.g. sourced from the electronic health
tree logical constructs. record), synthesis of an appropriate action
Schumaker et al. [18] characterise a or response or escalation in the care of the
chatbot as a system that “seeks to mimic patient with associated decision support, and Other Considerations and
conversation rather than understanding it”. storage of data [26]. AI systems for tele-mon-
itoring depend on and also expand the scope
Issues
Examples of chatbots include responding
to sales enquiries, travel directions, or per- of other health system ICT components. While AI offers much potential for enhanc-
forming search engine tasks. These have They can potentially outperform humans in ing care delivery through telehealth tools,
been widely adopted by the business sector many ways. They consistently execute their it is also important that we consider the
as online assistants or first response agents. instructions in a mathematical fashion, with social and ethical aspects of using it. AI, like
The first chatbot called ELIZA [19] was able a fundamental reliance on inbuilt logic mod- other technological advances in healthcare,
to establish a conversation with human beings erated by statistical evidence extracted by will cause disruption to many aspects of
and mimic their conversational models. It was machine learning methods from large-scale healthcare delivery including workflows,
based on the rephrasing of input sentences, datasets. They can immediately incorporate communication, access to services, and the
when these matched a set of pre-defined rules. and co-ordinate data from additional tools engagement between different providers and
Another evolving technology is voice recog- like location finders (GPS), accelerometers, patients [32]. Implementation is always the
nition e.g. Cortana (MS Windows®), Alexa motion sensors, gyroscopes, …etc. Sourcing “last-mile” problem and our focus needs not
(Amazon®), Siri (Mac®), Google Home® such additional data by humans is tedious to be on developing new AI tools or algo-
…etc. [20, 21]. Such technologies make it and would need education and training for rithms but rather on developing approaches
possible to have a service that can attend incorporation in care delivery. for embedding AI in society.
to a patient and caregiver in need of help at Tele-monitoring has been evaluated for The medical informatics community
any time. Hybrid technologies where both the distant surveillance of patients with has long been studying the occurrence of
humans and chatbots interact with patients chronic disease, such as chronic heart failure unintended consequences (UICs) post HIT
and caregivers can also exist. [27], chronic obstructive pulmonary disease implementation. These UICs occur, not
Health applications where virtual assis- (COPD) [28], and diabetes mellitus [29]. In necessarily out of negligence on the part
tants can provide a viable supplementation or COPD, AI methods have been applied to the of system designers, but rather because of
alternative to traditional healthcare delivery management and surveillance of the condi- our inability to account for the new types of
models include dealing with cognitively tion. A Classification and Regression Tree connectivity and communication practices
impaired individuals [22], improving acces- (CART) algorithm for the early identification that arise post-HIT implementation [33-35].
sibility of online clinical information [23], of patients at high risk of an imminent exac- Like all new advancements, AI-enabled
or providing avatar-based patient agents for erbation has been validated using telehealth telehealth will go through a crest of excite-
the elderly [24]. Such cases rely upon a more measurement data recorded from patients ment and potential, followed by a trough of
sophisticated conversational purpose and with moderate/severe COPD living at home disappointment and fear before reaching
knowledge base, and the level of AI complex- [30]. Similar approaches could be used as stability. Our efforts should be to achieve
ity rises with a deeper understanding by the a real-time exacerbation event detector in a stability early. We offer four key social and
AI agent through data accumulation. It may number of chronic conditions. ethical considerations for enabling AI-en-
be necessary to incorporate aspects of affec- Apart from these above deterioration-ori- abled telehealth.
tive behaviour, using multimodal contextual ented approaches, management of recovery 1. Ensuring equity - While AI and technol-
awareness mechanisms to enable an authentic situations provide another patient monitoring ogy can enhance access to and delivery
conversational dialogue [25]. For example, if opportunity for telehealth. For example, soft- of services, it can also increase the divide
issues arising from a patient’s past interactions ware to measure wound size [31] can supple- between the have and have nots [36]. We
or past medical history need to be considered ment visual assessment to give improvement must ensure that technology-enabled
in making conversational decisions, a person- in performance and assist remote care. In healthcare delivery enables service deliv-
alised model of the individual’s context will the example case, a special computer read- ery for those who need it most, such as
be needed in addition to the context model able scale is provided, as well as applying people in rural and remote areas as well
for the current conversation. automated image processing methods for as in under-developed countries.

IMIA Yearbook of Medical Informatics 2019

 39
Role of Artificial Intelligence within the Telehealth Domain

2. Monitoring the technological divide consequences. Most AI methods require a 13. Goldstein BA, Navar AM, Pencina MJ, Ioannidis
– The world’s population is diverse. substantial learning phase which can only JP. Opportunities and challenges in developing risk
prediction models with electronic health records
Some populations are proficient users achieve reliability after a very long time, data: a systematic review. J Am Med Inform Assoc
of technology while others are not. and hence should be subject to continuous 2017;24(1):198-208.
Persons expected to benefit most from testing and refinement to better develop the 14. Hendy J, Chrysanthaki T, Barlow J, Knapp M, Rog-
ers A, Sanders C, et al. An organisational analysis
advances in telehealth like the aged and human-AI interchange. Within the telehealth of the implementation of telecare and telehealth:
extremely ill may be limited in their use domain, control is even more difficult as the whole systems demonstrator. BMC Health Serv
of technology. We must ensure that AI some of the service delivery components Res 2012 Dec;12(1):403.
applications do not increase the digital may not have a human immediately in the 15. Bara, A, Klein B, Proudfoot JG. Defining in-
ternet-supported therapeutic interventions. Ann
divide, but rather enhance our ability loop. In such cases if things do go wrong, Behav Med 2009 Aug 1;38(1):4-17.
to provide quality patient-centered care the question of who should be targeted for 16. Hoermann S, McCabe KL, Milne DN, Calvo RA.
delivery to all citizens [37]. redress will be a challenge for society. Application of synchronous text-based dialogue
3. AI is only a means – HIT implementa- systems in mental health interventions: systematic
review. J Med Internet Res 2017 Aug;19(8).
tion in healthcare has been described as 17. Laranjo L, Dunn AG, Tong HL, Kocaballi AB,
a journey, not a destination [38], and as
the digitization of healthcare proceeds at References Chen J, Bashir R, et al. Conversational agents in
healthcare: a systematic review. J Am Med Inform
record speeds this becomes more appar- 1. World Health Organisation. A health telematics Assoc 2018 Jul 11;25(9):1248-58.
policy in support of WHO’s Health-For-All strat- 18. Schumaker RP, Ginsburg M, Chen H, Liu Y. An
ent than ever. We must embrace the notion egy for global health development: report of the evaluation of the chat and knowledge delivery
of AI to be part of a learning health care WHO group consultation on health telematics, components of a low-level dialog system: The
system [39] with a pragmatic approach 11–16 December, Geneva, 1997. Geneva: World az-alice experiment. Decis Support Syst 2007
with regards to the design and delivery Health Organization; 1998. January;42:2236–46.
2. Ajami S, Lamoochi P. Use of telemedicine in 19. Weizenbaum J. ELIZA – A computer program
of AI-enhanced tools. for the study of natural language communication
disaster and remote places. J Educ Health Promot
4. People must come first – Healthcare 2014;3:26. between man and machine. Communications of
is first and foremost about people. AI the ACM 1966;9(1):36-45.
3. McLean S, Protti D, Sheikh A. Telehealthcare for
tools will change how interactions occur 20. Klopfenstein LC, Delpriori S, Malatini S, Bogliolo
long term conditions. BMJ 2011 Feb 3;342:d120. A. The rise of bots: a survey of conversational in-
between different health system agents 4. World Health Organization. Telemedicine: op- terfaces, patterns, and paradigms. In: Proceedings
and we need to ensure that the focus portunities and developments in member states. of the 2017 Conference on Designing Interactive
Report on the second global survey on eHealth. Systems 2017 Jun 10:555-65.
of these tools are outcomes such as the Geneva: World Health Organization; 2010. 21. Lisetti C, Amini R, Yasavur U. Now all together:
empowerment of patients and a reduction 5. Wilson LS, Maeder AJ. Recent directions in tele- overview of virtual health assistants emulating
in provider burnout [40]. medicine: review of trends in research and practice. face-to-face health interview experience. KI-Kün-
Healthc Inform Res 2015 Oct 1;21(4):213-22. stliche Intelligenz 2015 Jun 1;29(2):161-72.
6. Russell S, Norvig P. Artificial intelligence - a 22. Yaghoubzadeh R, Kramer M, Pitsch K, Kopp S.
modern approach. Prentice-Hall; 1995. Virtual agents as daily assistants for elderly or
7. Pacis DM, Subido Jr ED, Bugtai NT. Trends cognitively impaired people. In: International
Conclusions in telemedicine utilizing artificial intelligence.
In: AIP Conference Proceedings 2018 Feb
Workshop on Intelligent Virtual Agents 2013 Aug
29. Berlin, Heidelberg: Springer; 2013. p. 79-91.
The above discussion suggests that AI-en- 13;1933(1):040009. 23. Bickmore TW, Utami D, Matsuyama R,
8. Peterson,MC, Holbrook JH, Von Hales D, Smith Paasche-Orlow MK. Improving access to online
abled telehealth offers contributions in health information with conversational agents: a
the form of quality improvement and NL, Staker LV. Contributions of the history, phys-
ical examination, and laboratory investigation randomized controlled experiment. J Med Internet
enhancement of existing practice, as well in making medical diagnoses. West J Med 1992 Res 2016 Jan;18(1).
as the instigation of new models of care. 24. Shaked NA. Avatars and virtual agents–relation-
Feb;156(2):163–5.
ship interfaces for the elderly. Healthc Technol Lett
Examples provided of the role of AI in 9. Roshan M, Rao A. A study on relative contributions 2017 Jun 28;4(3):83.
remote delivery of health care include of the history, physical examination and inves- 25. Riccardi G. Towards healthcare personal agents.
the use of tele-assessment, tele-diagnosis, tigations in making medical diagnosis. J Assoc In: Proceedings of the 2014 Workshop on Road-
Physicians India 2000 ;48(8):771–5. mapping the Future of Multimodal Interaction
tele-interactions and tele-monitoring. Fur- 10. Liang CW, et al. Using Machine Learning to Pre- Research including Business Opportunities and
ther development of underlying algorithms dict Cancer Risk by Clinical Diagnosis History - A Challenges 2014 Nov 16. ACM; 2014. p. 53-6.
and validation of methods will be required Case Study of Liver Cancer. In preparation. 26. Nangalia V, Prytherch DR, Smith GB. Health
for wider adoption. 11. Haenssle HA, Fink C, Schneiderbauer R, Toberer technology assessment review: Remote monitoring
F, Buhl T, Blum A, et al. Man against machine: of vital signs-current status and future challenges.
Certain key social and ethical consid- diagnostic performance of a deep learning convo- Crit Care 2010 Oct;14(5):233.
erations also need to be taken into account lutional neural network for dermoscopic melanoma 27. Inglis SC, Clark RA, McAlister FA, Stewart S,
as AI-enabled telehealth becomes more recognition in comparison to 58 dermatologists. Cleland JG. Which components of heart failure
commonplace. Unlike humans, AI systems Ann Oncol 2018 May 28;29(8):1836-42. programmes are effective? A systematic review
12. Esteva A, Kuprel B, Novoa RA, Ko J, Swetter SM, and meta-analysis of the outcomes of structured
never lack motivation or drive, but without Blau HMet al. Dermatologist-level classification telephone support or telemonitoring as the primary
emotions there is an inability to assess moral of skin cancer with deep neural networks. Nature component of chronic heart failure management
good or bad, with little understanding of the 2017 Feb;542(7639):115. in 8323 patients: abridged Cochrane Review. Eur

IMIA Yearbook of Medical Informatics 2019

40
Kuziemsky et al.

J Heart Fail 2011 Sep;13(9):1028-40. 33. Coiera E, Ash J, Berg M. The Unintended Con- not a destination. J Am Med Inform Assoc
28. Bolton CE, Waters CS, Peirce S, Elwyn G. Insuf- sequences of Health Information Technology 2004:11:121-4.
ficient evidence of benefit: a systematic review of Revisited. Yearb Med Inform 2016:163-9. 39. Friedman C, Rubin J, Brown J, Buntin M, Corn M,
home telemonitoring for COPD. J Eval Clin Pract 34. Kuziemsky CE, Randell R, Borycki EM. Under- Etheredge L, et al. Toward a science of learning
2011 Dec;17(6):1216-22. standing Unintended Consequences and Health systems: a research agenda for the high-function-
29. Polisena J, Tran K, Cimon K, Hutton B, McGill Information Technology: Contribution from the ing Learning Health System. J Am Med Inform
S, Palmer K. Home telehealth for diabetes man- IMIA Organizational and Social Issues Working Assoc 2015;22:43–50.
agement: a systematic review and meta-analysis. Group. Yearb Med Inform 2016(1):53-60. 40. Epstein RM, Fiscella K, Lesser CS, Stange KC.
Diabetes Obes Metab 2009 Oct;11(10):913-30. 35. Gogia SB, Maeder A, Mars M, Hartvigsen G, Basu Why the nation needs a policy push on patient-cen-
30. Mohktar MS, Redmond SJ, Antoniades NC, Roch- A, Abbott P. Unintended Consequences of Tele tered health care. Health Aff (Millwood) 2010 Aug
ford PD, Pretto JJ, Basilakis J, et al. Predicting Health and Approaches for their Solutions. Yearb 1;29(8):1489-95.
the risk of exacerbation in patients with chronic Med Inform 2016;(1):41-6.
obstructive pulmonary disease using home tele- 36. Veinot TC, Mitchell H, Ancker JS. Good intentions
health measurement data. Artif Intell Med 2015 are not enough: how informatics interventions
Jan 1;63(1):51-9. can worsen inequality. J Am Med Inform Assoc Correspondence to:
31. Thawer HA, Houghton PE, Woodbury MG, Keast 2018:25:1080-8. Craig Kuziemsky
D, Campbell K. Computer-assisted and manual 37. Ohno-Machado L. Understanding and mitigating Telfer School of Management
wound size measurement. Ostomy Wound Manage the digital divide in health care. J Am Med Inform University of Ottawa
2002;48(10):46–53. Assoc 2017;24(5):881. Ottawa, ON
32. Loh E. Medicine and the rise of the robots: a 38. McDonald CJ, Overhage JM, Mamlin BW, Dex- Canada
qualitative review of recent advances of artificial ter PD, Tierney WM. Physicians, information Tel: +1 613 562 5800 ext 4792
intelligence in health. BMJ Leader 2018;2:59-63. technology, and health care systems: a journey, E-mail: [email protected]

IMIA Yearbook of Medical Informatics 2019