SeVA: An AI Solution for Age Friendly Care of Hospitalized

Older Adults

Chongke Wu

, Jeno Szep

, Salim Hariri

, Nimit K. Agarwal

, Sumit K. Agarwal

and Carlos Nevarez

NSF Center for Cloud and Autonomic Computing, The University of Arizona, Tucson, Arizona, U.S.A.

Department of Medicine, Banner, University Medical Center Phoenix, Phoenix, Arizona, U.S.A.

SevaTechnology LLC, Tucson, Arizona, U.S.A.

cwildhorse@gmail.com

Keywords: Artificial Intelligence, Chatbot, Healthcare, Patient Monitoring, Delirium, Internet of Things.

Abstract: As a dangerous syndrome, delirium affects more than 50% of hospitalized older adults and has an economic

burden of 164 billion US dollars per year. It is crucial to prevent, identify and treat this syndrome

systematically on all hospitalized patients to prevent its short and long-term complications. Currently, there

are no AI-based tools being utilized at a large scale focused on delirium management in hospital settings. The

advancement of the Internet of Things in the medical arena can be leveraged to help clinical teams managing

the care of patients in the hospital. The renaissance of Artificial Intelligence brings the chance to analyze a

large amount of monitoring data. Deep neural networks like Convolutional Neural Network and Recurrent

Neural Network revolutionize the fields of Computer Vision and Natural Language Processing. Deep learning

tasks like action recognition and language understanding can be incorporated into the routine workflow of

healthcare staff to improve care. By leveraging AI and deep learning techniques, we have developed a chatbot

based monitoring system (that we refer to as SeVA) to improve the workload of the medical staff by using an

Artificial Emotional Intelligence platform. The SeVA platform includes two mobile applications that provide

timely patient monitoring, regular nursing checks, and health status recording features. We demonstrate the

current progress of deploying the SeVA platform in a healthcare setting.

1 INTRODUCTION

Delirium affects more than 25% of hospitalized

patients and can be seen in more than 50% of

hospitalized older adults, impacting long-term

survival, and quality of life (Marcantonio, 2017). The

continuous and objective monitoring of delirium

similar to blood pressure checks can help medical

staff identify, prevent, and treat delirium and its many

compilations.

The Internet of Things (IoT) technology has the

potential of improving healthcare quality. It

empowers clinicians (physicians and nurses) to

review vast amounts of clinical data efficiently and

meaningfully for clinical decision making by

improving their workflows. Xu et al. present an IoT-

based system for emergency medical service by

providing data access timely and ubiquitously in a

cloud and mobile computing platform (Yu, Beam,

and Kohane, 2018).

As healthcare systems increasingly adopt

Artificial Intelligence (AI) in their decision making,

we are seeing a renaissance of AI in the field of

Healthcare. For example: in the case of patient

monitoring in the intensive care unit or emergency

rooms, an AI-assisted alert system can be helpful to

process a large amount of data generated by routine

monitoring devices (Xu et al., 2014). The vital signs

and Modified Early Warning Score systems can be

used to build a prediction model for cardiac arrest

(Churpek et al., 2012; Szep, Akoglu, Hariri, &

Moukabary, 2018).

The advancement of Natural Language

Processing makes it possible to create an expert

knowledge system to provide ubiquitous service.

Microsoft released the Healthcare Bot service to

empower healthcare organizations to build and

deploy the conversational health care experience at

scale (Microsoft, 2020). It combines medical

intelligence with natural language capabilities. The

Wu, C., Szep, J., Hariri, S., Agarwal, N., Agarwal, S. and Nevarez, C.

SeVA: An AI Solution for Age Friendly Care of Hospitalized Older Adults.

DOI: 10.5220/0010313605830591

In Proceedings of the 14th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2021) - Volume 5: HEALTHINF, pages 583-591

ISBN: 978-989-758-490-9

583

IBM question-answering computer system Watson is

utilized to help physicians with the treatment of

patients as a “diagnosis and treatment advisor” (IBM,

2020). Extracting the knowledge broadly and

returning the results promptly is an inherent feature

of the AI healthcare knowledge engines. It provides a

significant advantage compared to traditional medical

processes, especially in areas with limited medical

resources.

Healthcare data is very sensitive and requires

security protection. Secure identification is one of the

measures to mitigate the risk of identity theft.

Pacheco et al. propose an IoT security framework for

smart infrastructures against cyber-attacks (Pacheco

and Hariri, 2016). When conducting user group

estimation, local differential privacy can protect user

information without the assumption of the trusted

data server (Gu, Li, Cao, and Xiong, 2019;Gu, Li,

Cheng, Xiong, and Cao, 2020).

By analyzing data coming from monitored

patients we can create a system that can respond to

patient's needs in a timely manner. In this paper, we

present an Age-Friendly patient care platform

connecting seniors, caregivers, healthcare, and

community by leveraging AI and ML techniques:

SeVA (Senior’s Virtual Assistant). With the support

of a Natural Language Processing platform, the SeVA

platform achieves real-time and continuous

monitoring of the patient status as well as capturing

patient intent from the human-computer interactions.

The remaining sections of the paper are organized

as follows: Section II introduces the related research

of artificial intelligence application in the medical

field; In section III, we present the system design of

the SeVA platform; Section IV shows the

implementation details of the platform; Section V

summarizes the work in this paper and discusses

future research plan.

2 BACKGROUND AND RELATED

RESEARCH

2.1 Delirium: Insidious and Dangerous

Syndrome

Delirium is a dangerous syndrome commonly seen in

hospitalized patients. More than half of hospitalized

older adults (> 65 years of age) are affected (around

7 million patients annually) and most of them remain

undiagnosed. Delirium in hospitalized patients leads

to higher hospital length of stay, higher mortality rate,

loss of physical function requiring long term care, and

can even be a precursor to dementia. It costs more

than 164 billion US dollars per year to healthcare

(Inouye et al., 2016). There is no mandatory

prevention program as well as no reporting to the

Centres of Medicare & Medicaid Services (CMS). As

a comparison, around 24 billion dollars are lost due to

sepsis, and every hospital carries a mandatory sepsis

alert program and pathway (Paoli, Reynolds, Sinha,

Gitlin, & Crouser, 2018). The data for sepsis is also

reported to CMS as an adverse event. The gaps in

delirium care include delayed recognition, inadequate

risk modification and prevention, and ineffective

treatment. The major reason for this is the lack of a

standardized multidisciplinary approach for the

management of delirium across hospital systems. The

Hospital Elder Life Program (HELP) developed by

Inouye is a system that relies on volunteer healthcare

workers to engage patients (Inouye et al., 1999);

however, it has only been implemented in a few

hospitals. There certainly is a need for a system that

can be easily implemented, customized and is

scalable across all hospitals that can provide timely

screening, assessment, and recognition, so that the

cause or precipitating factors for delirium can be

removed, and the patient can receive appropriate and

early treatment.

2.2 Gaps in Patient Monitoring

Despite best efforts by nursing staff in hospital

systems, to decrease the risk of falling, management

of uncontrolled pain, even addressing basic patient

needs like using the bathroom, can be easily missed.

For the patient who has cognitive impairment either

as delirium or dementia, this risk becomes even more

profound. Best nursing practices include a systematic

approach to addressing these care needs, e.g.

performing timed nurse rounding checks or checking

for the 4 Ps (Pain, Position, Potty, Periphery).

However, these practices require dedicated nursing

staff and strict protocols which can be difficult to

implement at a large scale due to limited resources

and cost issues. The “Unsupervised Care Windows”

created due to lack of these practices or between the

hours of timed nurse rounding can lead to serious

events like falls. Patient falls have an enormous cost

on the healthcare system according to the data

reported by the National Database of Nursing Quality

Indicators (NDNQI) (AHRQ, 2020; Mitchell,

Lavenberg, Trotta, and Umscheid, 2014).

Prevention of these adverse events by integrating

technology for the detection of unexpected patient

behaviors like unintended falls is the subject of

significant research over the last several years.

HEALTHINF 2021 - 14th International Conference on Health Informatics

584

Various detection systems have been developed and

can be broadly divided into wearable based, non-

wearable based, and fusion-based systems (Chaccour,

Darazi, El Hassani, and Andres, 2016).

The wearable based systems can be placed on

different body parts like feet, knee, waist, etc. To

measure the body motion parameters like

acceleration, the sensors must be tied to the body. The

typical sensors include accelerometer, gyroscope,

magnetometer, etc. With the universal acceptance of

mobile devices, the smartphone-based solution could

be a very competitive alternative to the conventional

dedicated fall detection and prevention tools (Habib

et al., 2014). The shortcomings of wearable based

systems are that they are relatively inflexible and

uncomfortable.

2.3 AI-based Assistant in Healthcare

Many researchers and companies have introduced

artificial intelligence into their mobile medical

applications to make interactions with the patient

easier. The applications can be further categorized as

healthy lifestyle assistants, remote diagnosis systems,

and medical advisors.

Healthy lifestyle assistant applications will

perform more on disease prevention so that the

suggestion will be more general. Pact Care is a startup

that provides a patient-centric healthcare data

solution. Their mobile product Florence is a chatbot

based personal health assistant with medication

reminders and health trackers. It does not provide

medical advice and is only for personal usage (PACT,

2020). Fadhil et al. propose an AI-chatbot scenario

for healthy lifestyle promotion with nutrition

education and behavior change interventions (Fadhil

and Gabrielli, 2017).

Figure 1: SeVA System Framework.

Remote diagnosis applications will connect the

patient and the physician remotely to mitigate the

medical resource imbalance distribution. The AI part

of the application is focused more on assisting

patients to find the correct doctor. Babylon Health is

a health service provider that offers remote doctor

diagnosis. Their artificial intelligence platform uses a

probabilistic graphical model and natural language

processing to interpret medical questions (Babylon

health, 2020).

Medical advisor applications will give medical

suggestions based on their knowledge base. Buoy

Health makes a digital assistant application that helps

patients self-diagnose and triage for the selection of

appropriate care. The chatbot will ask a series of

medical questions to diagnose customer symptoms

(Buoy Health, 2020). Chung et al. propose a chatbot-

based service with a knowledge base (Chung & Park,

2019). The patient could consult the system with the

picture or text input from the mobile devices. It gives

a fast treatment plan in response to accidents as well

as the change of conditions of a patient with chronic

disease. Comendador et al. develop a pediatric

generic medicine consultant chatbot. It acts as a

medical consultant to suggest generic medicine for

children (Comendador, Francisco, Medenilla, & Mae,

2015).

Our patient care platform SeVA brings the AI

technologies by leveraging Natural Language

Processing and the real-time monitoring of peripheral

sensors. SeVA can be classified as a combination of

the medical advisor and the remote diagnosis system

which is different from the other AI health platforms

which search for often unreliable solutions from the

Internet or other databases, SeVA allows clinicians to

design personalized conversations directly within the

platform. The user interaction in SeVA is easy to use

for older patients who might have limited proficiency

in using technology, or for patients with cognition

issues like dementia or delirium who might not be

able to use plain text-based interactions. It uses

simple gestures like hand waving or simple voice

conversations. This communication mode allows for

the recognition of emotion which can, in turn, allow

interventions like soothing music to mitigate the risk

of delirium. The modular system design provides the

possibility for the integration of additional extra

sensors to accommodates the system in different

environments.

SeVA: An AI Solution for Age Friendly Care of Hospitalized Older Adults

585

3 SYSTEM DESIGN

3.1 SeVA Framework

As shown in Figure 1, the SeVA platform consists of

a five-layer framework: infrastructure layer, data

layer, scenario layer, application layer, and security

layer. This framework provides a general

methodology for building a chatbot-based healthcare

system by utilizing patient real-time data and medical

expert knowledge.

The infrastructure layer provides the basic

hardware requirement of the data collection unit. It

includes the minimum requirement for deploying

SeVA platform to a different environment because

most of SeVA functions are provided as cloud

services. The wireless network block works as a

communication module for real-time data

transmission. The mobile device, such as the tablet or

cell phone, shows the user interface and conducts the

conversation. The peripheral sensor collects patient

movement data to infer the patient's position status

without infringing user privacy. The voice data and

movement data will then be transferred to the upper

layer.

The data layer processes the incoming raw data

from the infrastructure layer. The main task here is

text-to-speech conversion, speech-to-text conversion,

and action recognition. Hence, we need the neural

network as the backbone technique to implement task

functions. For a sequence to sequence problem, the

RNN will provide the majority solution. For the

action recognition which is based on the temporal

movement data, we use a long-short-term memory

neural network to process it. The output of this layer

will be the conversation plain text and the result of the

user action classification.

The scenario layer contains the scenarios provided

by the professional medical expert and returns the

proper conversation to the user. It requires a chatbot

engine to support the scenario representation. More

specifically, the medical expert predefines the

scenario representation with the related incoming

conversation plain text or action class, then it will

provide the conversation and trigger the other

program in the application layer.

The application layer consists of two mobile

applications: the SeVA Patient Room (SPR)

application and the SeVA Master Control (SMC)

application. The first application will work as the

main interface for the patient. It does not only supply

the conversation but also has a predefined workflow

for a regular medical check. The second application

is designed for the medical staff to receive the

notification from the patient room and return quick

feedback to the patient. Behind the user, the mobile

application is the application control system, which is

connected to the SeVA backend server. It manages

the user account database for the authentication

process and controls the communication between

different user applications.

The security layer serves as the auxiliary

component of our system. It protects user data

security and privacy and guarantees SeVA system

robustness. In the infrastructure layer, all the data

collection is compliant with the privacy policy and

with the consent of the user. The data transmission

and storage will be encrypted to guarantee data

confidentiality. We also set up a crash report in the

application layer and system monitoring server to

enable us to achieve a quick response to any

anomalous behavior of the SeVA platform functions.

Figure 2: System Architecture of SeVA.

HEALTHINF 2021 - 14th International Conference on Health Informatics

586

Figure 3: Chatbot Engine skill and story design.

3.2 Components

The SeVA system architecture is shown in Figure 2.

It shows the connection between the main SeVA

components. The arrow represents the information

flow direction. The Chat-bot engine scenario is

defined by the medical staff. Let us consider the

“Waving Hand” event as an example. When a patient

has a request, he/she just waves a hand and then the

peripheral sensor recognizes the action and sends an

HTTPS request to the SeVA backend server. The

request is parsed and fetches the necessary

information, like the room number, from the database

then triggers the chatbot engine to start the

conversation. The SeVA patient room application has

a WebSocket connection with the chatbot engine,

then the conversation will be launched. The

conversation result will be sent to the SeVA backend

server as the format of SeVA Command, which will

be further sent to the SeVA Master Control

application in the nurse room. Then the nurse can

make timely interventions if a critical event happens.

The system running status is monitored by the

Monitoring Server, which guarantees system

reliability.

3.2.1 Peripheral Sensor

We use peripheral sensors to monitor the patient's

status and provide timely interventions in case of any

emergency or critical events. The currently available

sensor is a smart wristband that can detect user

movement sending out the event trigger.

3.2.2 Chatbot Engine

The Chatbot engine architecture is shown in Figure 3.

We build the engine by using the Google NLP

platform Dialogflow. The user's voice is transformed

into plain text and is processed in the NLP module.

The text is first being pre-processed and tokenized,

which result in discrete word and sent to the RNN.

The output of the RNN can be entries, which is the

matching word candidates, or the intent score. The

intent score then is being sent to the expert system

module in a dialogue engine and gives out the final

intent. The intent and the entries serve as the input of

the predefined skills and the skill logic decides the

key content of the conversation. The skill can be

triggered by a request or determined by the intent.

3.2.3 SeVA Backend Server

The SeVA backend server processes the incoming

requests and manages the user data. As a centralized

processing center, it facilitates the management of the

user account as well as providing the API for the

mobile application, database, and monitoring server.

The data sent to the SeVA backend server are

encrypted. When the user tries to use the application,

it must submit the login request to the SeVA backend

server and wait for the authentication token stored in

the database. With the token, it starts the service of

the chatbot engine. The SeVA backend is listening to

the event trigger from the peripheral sensor, which

will be routed to the Chatbot engine to start the

conversation or directly to the SeVA mobile

application.

3.2.4 SeVA Mobile Applications

We have developed two applications: SeVA Patient

Room App (SPR) and SeVA Master Control App

(SMC). As the name indicates, SPR is deployed in the

patient room. SPR uses the Apple built-in Speech

framework for both the Speech-to-Text and Text-to-

Speech conversion. It has the authorization and user

SeVA: An AI Solution for Age Friendly Care of Hospitalized Older Adults

587

Table 1: SeVA skills in the Dialogflow.

Type Skill Story Description

Movement Response Sensor Response

Fall Detection

Response to patient and notify nurse by

recognize patient movement.

Wave

Regular Check

Hourly Rounding

Feeling Check

Perform regular hourly check to fulfil patient

needs actively.

Restroom Check

Brace Check

Heat pack Check

Delirium Check

What day is today

Perform regular delirium check to evaluate

patient cognition ability.

Spell weekdays in reversed order

Relaxations

Soothing Music Play Music

Use music, jokes, and small talk to improve

patient’s mental state.

Small Talk

Random Talk

Joke

Tell me a joke

More jokes

configuration pages for customizing user service. The

SMC app is deployed in the nurse room. It monitors

the status of all patients by receiving the messages

regularly from SPR and provides feedback to the

nurse immediately.

3.2.5 Monitoring Server

The SeVA backend server provides the monitoring

service. It monitors the API availability, the number

of connected nurses and patients, as well as the

backend server running status. Any abnormal

behaviors, such as backend server API unavailable or

no nurse online, will be reported to the operator

through the email service in the notification module.

4 IMPLEMENTATION

The mobile applications are written using Apple iOS

native program language Swift on Xcode 11, which

runs on the iOS device with iOS 13. The SeVA

backend server has 4 dedicated ARM processors with

2GB memory and the Ubuntu Xenial system.

Figure 4: SPR application user interface.

We use the Chatbot engine Dialogflow developed

by Google, LLC (Google, 2020). Dialogflow is a UI-

based platform for creating smart and proactive

chatbots. Our team’s medical experts define the skills

by setting the intent, trigger, and replies. A skill is

composed of inputs, slots, replies, actions, and

stories. Inputs define events that a bot can react to.

Slots are the memory of the bot for remembering

some information during the conversation. Replies

are all the possible sentences that a bot can reply to a

user. Stories define the logic behind a skill. The input

is classified by the RNN which has been trained with

the sample inputs we supplied. Once the intent

matches, it will continue the conversation in a

predefined way.

As shown in Table 1, the SeVA skill set includes

“Sensor Response”, “Hourly Rounding”, “Delirium

Check”, etc. Each skill contains multiple stories.

Regarding the stories of the “Sensor Response” skill,

when the patient is waving a hand for help, the story

“wave” will be triggered by the HTTPS post, and then

the chatbot engine will send the sentence to the story

part. The story part may contain the trigger of the next

story part, which will facilitate the program reuse. For

the story part of “Morning Check”: when the chatbot

engine receives the trigger, it will send the first

sentence to the SPR, then waits for the returning

sentence. The patient response will then be classified

as the intent “Yes” or “No”, which will continue to

different conversations.

The SPR provides the interaction between the

SeVA platform and the user. The user interface is

shown in Figure 4. Every user is required to register

so the personal profile will be created. After logging

in, the SPR will connect to the SeVA Backend server

through a WebSocket for communication. The

HEALTHINF 2021 - 14th International Conference on Health Informatics

588

conversation can be initiated by hand waving, using

the wake-up word, and simply touching the screen.

For privacy consideration, the SPR microphone is

turned off when there is no conversation and the

wake-up word feature is optional. The SPR will ask a

question every hour (except for the rest hour) to make

sure patient needs are satisfied. The SPR also can play

some of the music selected by the therapist which will

create a calm atmosphere for the patient and relieve

anxiety.

(a) Game “Connect Nodes” (b) Game “Click Animal”

Figure 5: The sample delirium check games embeded in the

SPR. (a) Game “Connect Nodes” for testing patient

visuospatial and executive functions, it require patient

connect numbered nodes in a given order. (b) Game “Click

Animal” for testing patient attention, it require patient click

the animal image which will disappear quickly.

When triggered, SPR starts checking the patient

delirium status by asking questions or launching the

delirium check game. For example, we use a modified

version of Alternating Trail Making from the

Montreal Cognitive Assessment for mental status

assessment in older adults (Julayanont & Nasreddine,

2017). In Figure 5, we display two delirium check

games. The first game “Connect Node” is used for

testing the visuospatial and executive ability. In this

game, the patient is required to connect nodes in a

certain order. If the patient fails the test, a message

will be immediately sent to the SMC. The second

game “Click Animal” tests the patient’s attention.

Pictures will be prompted and then they disappear.

The patient must click at all the animal picture. The

results of the test will also be sent to the SeVA Master

Control application.

Figure 6: SeVA Mater Control application user interface.

The SMC lets the nurse monitor the status of multiple

patient rooms and process the incoming request, as

shown in Figure 6. The SMC shows 6 sub-panels on

the screen. Each sub-panel contains buttons and a

textbox. It will display the request by flashing the

displayed buttons, and the nurse could click the

button to send back the acknowledge message. For

preventing the patient from falling, the SMC will

display the critical movement information like

“Patient is sitting up” on the textbox. Then the nurse

could make a timely intervention.

This system is being deployed at the Banner -

University Medicine Rehabilitation Institute. It has

gotten the approval of the Institutional Review Board

(IRB) and also obtained positive feedback from the

nursing staff at preliminary demonstrations.

5 CONCLUSION

The IoT architecture and AI-based SeVA platform

can improve healthcare quality and nursing

workflows by automating traditional standard clinical

and nursing practices in hospital settings. SeVA

platform implementation uses Artificial Emotional

Intelligence to build a monitoring and diagnosing

system. The system features include: starting a

conversation with the patient to check for delirium;

perform regular round checks to improve nursing

workflows and provide actionable items for nursing

care; detect critical events such as falls, detect

gestures and patient’s motion like walking, waving

for help, detecting emotion and providing

interventions like playing relaxing music. The system

reduces gaps from “Unsupervised Care Windows”

and provides a customized healthcare experience

catered to Age-Friendly Care. We are currently

testing and evaluating the feasibility of the current

SeVA: An AI Solution for Age Friendly Care of Hospitalized Older Adults

589

SeVA platform implementation in a hospital setting.

We are also investigating innovative methods to

quantify cognition and emotion with the goal to

recommend non-pharmacological interventions to

reduce stress during the hospital stay. We will

evaluate the system with patient and nurse surveys as

well as the alarm statistical metrics including True

Positive Rate, False Positive Rate, and False Negative

Rate.

ACKNOWLEDGEMENTS

This work is partly supported by the Air Force Office

of Scientific Research (AFOSR) Dynamic Data-

Driven Application Systems (DDDAS) award

number FA9550-18-1-0427, National Science

Foundation (NSF) research projects NSF-1624668

and NSF-1849113, (NSF) DUE-1303362

(Scholarship-for-Service), National Institute of

Standards and Technology (NIST) 70NANB18H263,

and Department of Energy/National Nuclear Security

Administration under Award Number(s) DE-

NA0003946.

REFERENCES

AHRQ.(2020). How do you measure fall rates and fall

prevention practices? Retrieved from https://

www.ahrq.gov/professionals/systems/hospital/fallpxto

olkit/fallpxtk5.html

Babylonhealth. (2020). Babylon health UK - the online

doctor and prescription services app. Retrieved from

https://www.babylonhealth.com/

Chaccour, K., Darazi, R., El Hassani, A. H., & Andres, E.

(2016). From fall detection to fall prevention: A generic

classification of fall-related systems. IEEE Sensors

Journal, 17 (3), 812–822.

Chung, K., & Park, R. C. (2019). Chatbot-based healthcare

service with a knowledge base for cloud computing.

Cluster Computing, 22 (1), 1925– 1937.

Churpek, M. M., Yuen, T. C., Huber, M. T., Park, S. Y.,

Hall, J. B., & Edelson, D. P. (2012). Predicting cardiac

arrest on the wards: a nested case-control study. Chest,

141 (5), 1170–1176.

Churpek, M. M., Yuen, T. C., Winslow, C., Robicsek, A.

A., Meltzer, D. O., Gibbons, R. D., & Edelson, D. P.

(2014). Multicenter development and validation of a

risk stratification tool for ward patients. American

Journal of Respiratory and Critical Care Medicine, 190

(6), 649–655.

Comendador, B. E. V., Francisco, B. M. B., Medenilla, J.

S., & Mae, S. (2015). Pharmabot: a pediatric generic

medicine consultant chatbot. Journal of Automation

and Control Engineering Vol, 3 (2).

Fadhil, A., & Gabrielli, S. (2017). Addressing challenges in

promoting healthy lifestyles: the al-chatbot approach.

In Proceedings of the 11th EAI International

Conference on Pervasive Computing Technologies for

Healthcare (pp. 261–265).

Google. (2020). Dialogflow | Google Cloud. Author.

Retrieved from https://cloud.google.com/dialogflow

Gu, X., Li, M., Cao, Y., & Xiong, L. (2019). Supporting

both range queries and frequency estimation with local

differential privacy. In 2019 IEEE Conference on

Communications and Network Security (CNS) (pp.

124–132).

Gu, X., Li, M., Cheng, Y., Xiong, L., & Cao, Y. (2020).

PCKV: Locally differentially private correlated key-

value data collection with optimized utility. In 29th

USENIX security symposium (USENIX security 20) (pp.

967–984).

Habib, M. A., Mohktar, M. S., Kamaruzzaman, S. B., Lim,

K. S., Pin, T. M., & Ibrahim, F. (2014). Smartphone-

based solutions for fall detection and prevention:

challenges and open issues. Sensors, 14 (4), 7181–

7208.

Health, B. (2020). Retrieved from https://www.

buoyhealth.com/

IBM. (2020). AI healthcare solutions. Retrieved from

https://www.ibm.com/watson-health

Inouye, S. K., Bogardus Jr, S. T., Charpentier, P. A., Leo-

Summers, L., Acam-pora, D., Holford, T. R., & Cooney

Jr, L. M. (1999). A multicomponent intervention to

prevent delirium in hospitalized older patients. New

England journal of medicine,340(9), 669–676.

Inouye, S. K., Marcantonio, E. R., Kosar, C. M., Tommet,

D., Schmitt, E. M., Travison, T. G., Jones, R. N. (2016).

The short-term and long-term relationship between

delirium and cognitive trajectory in older surgical

patients.

Alzheimer’s & Dementia, 12 (7), 766–775.

Julayanont, P., & Nasreddine, Z. S. (2017). Montreal

cognitive assessment (moca): concept and clinical

review. In Cognitive Screening Instruments (pp. 139–

195). Springer.

Marcantonio, E. R. (2017). Delirium in hospitalized older

adults. New England Journal of Medicine, 377(15),

pp.1456-1466.

Microsoft. (2020, Jul). Microsoft health bot project - ai at

work for your patients. Retrieved from https://www.

microsoft.com/en-us/research/project/health-bot/

Mitchell, M. D., Lavenberg, J. G., Trotta, R., & Umscheid,

C. A. (2014). Hourly rounding to improve nursing

responsiveness: a systematic review. The Journal of

nursing administration,44(9), 462.

Pacheco, J., & Hariri, S. (2016). IoT security framework for

smart cyberinfrastructures. In 2016 IEEE 1st

International workshops on Foundations and

Applications of self* systems (fas* w) (pp. 242–247).

PACT.(2020). Florence your health assistant. Retrieved

from https://www.florence.chat/

Paoli, C. J., Reynolds, M. A., Sinha, M., Gitlin, M., &

Crouser, E. (2018).Epidemiology and costs of sepsis

in the united states—an analysis based on timing of

HEALTHINF 2021 - 14th International Conference on Health Informatics

590

diagnosis and severity level. Critical care

medicine,46(12),1889.

Szep, J., Akoglu, A., Hariri, S., & Moukabary, T. (2018).

Two-level autonomousoptimizations based on ml for

cardiac fem simulations. In 2018 IEEE International

Conf. on Autonomic Computing (ICAC)(pp. 101–110).

Wu, T., Rappaport, T. S., & Collins, C. M. (2015). Safe for

generations to come: Considerations of safety for

millimeter waves in wireless communications. IEEE

Microwave Magazine, 16 (2), 65–84.

Xu, B., Da Xu, L., Cai, H., Xie, C., Hu, J., & Bu, F. (2014).

Ubiquitous data accessing method in IOT-based

information system for emergency medical services.

IEEE Transactions on Industrial informatics, 10 (2),

1578–1586.

Yu, K.-H., Beam, A. L., & Kohane, I. S. (2018). Artificial

intelligence in healthcare. Nature biomedical

engineering, 2 (10), 719–731.

SeVA: An AI Solution for Age Friendly Care of Hospitalized Older Adults

591