AI in Personalized Health Management: Practices and Challenges

Zhanxu Jiang

Institute of Artificial Intelligence, Beihang University, Beijing, 100191, China

Keywords: Artificial Intelligence, Personalized Healthcare, Machine Learning.

Abstract: With the improvement of medical and technological level, people are no longer satisfied with the traditional

one-size-fits all inefficient medical model, but seek more consideration of individual differences and user

uniqueness of a new medical model called personalized health care. Artificial intelligence, especially deep

learning, has been used to analyze large amounts of medical data, and its potential in the health care field has

attracted much attention. In this article, the implement of artificial intelligence in personalized healthcare are

outlined, including health data analysis, genetic risk assessment, and diet and exercise intervention.

Specifically, this article first classifies health data analysis into two categories from the perspective of

technology, and then introduces two methods to deal with genetic risk: genetic risk prediction and drug

development, and then introduces how artificial intelligence can intervene in patients' daily life from two

aspects: diet intervention and exercise intervention. Finally, the article discusses the current problems

encountered in the development of artificial intelligence technology in the field of personalized health, and

provides a perspective on the prospects and solutions of artificial intelligence for doctors, healthcare

institutions, and governments in society. It is hoped that this article can provide theoretical support and

practical suggestions for doctors, health care institutions and government to cooperate to build a more

intelligent and personalized health care system.

1 INTRODUCTION

Personalized health management is a revolutionary

healthcare model, which analyzes users' medical data

from different sources, predicts and gives targeted

recommendations. Traditional healthcare models,

which tend to adopt standardized treatment regimens,

often do not well account for the diversity of

individuals in terms of health status, genetic

susceptibility, lifestyle choices, and environmental

influences (Goetz & Schork, 2018). The emergence

of personalized health management through

customized interventions and precision medicine

strategies to meet the individual needs of users while

ensuring their health status. Through the use of a large

number of health data analysis, prediction,

recommendation and other algorithms, this healthcare

model helps everyone to understand the possible

health risks and gives suggestions to help patients

actively participate in their own health management.

However, there are several challenges to

implementing this healthcare model. A major

https://orcid.org/0009-0003-5363-0305

challenge is the integration and analysis of a large

number of heterogeneous healthcare data sources,

including Electronic Health Records (EHRs),

genomic data, wearable data, and patient reports. As

Artificial Intelligence (AI) technology has advanced,

AI algorithms have shown outstanding performance

in identifying and discovering connections among

vast volumes of data. Through in-depth mining and

analysis of each patient's unique multi-dimensional

health data, machine learning, especially deep

learning, can detect probable health trends and risk

factors, thereby improving the accuracy of disease

prediction and providing scientific basis for personal

choices and medical decisions.

The first and most important step in most

personalized health processing processes is data

collection. EHRs, which hold personal health data

(such as diagnostic pictures, clinical notes, and past

medical histories), sensors in wearable devices that

capture physiological and biochemical indexes in

real-time (Pinto et al., 2017), microphones and

cameras for audio and video (Kim & Chung, 2015),

and social media rhetoric (Ahmed et al., 2020;

Jiang, Z.

AI in Personalized Health Management: Practices and Challenges.

DOI: 10.5220/0012897900004508

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Engineering Management, Information Technology and Intelligence (EMITI 2024), pages 46-51

ISBN: 978-989-758-713-9

Alotaibi et al., 2020; Sekulić & Strube, 2020) are

some of the most popular data resources. Much of the

current research focuses on disease prediction, where

researchers combine big data processing and Internet

of Things (IoT) techniques to process health

information, intending to predict whether a user has a

specific disease, such as diabetes (Krishnamoorthi et

al., 2022), chronic kidney disease (Abdel-Fattah et

al., 2022), Cardiovascular Disease (CVD) (Zhang et

al., 2023), cancer (Trivizakis et al., 2020; Cheerla &

Gevaert, 2017), and some chronic diseases (Tiwari &

Agarwal, 2023). In addition to processing the data of

electronic medical records, conducting personalized

health management based on real-time monitoring

also received much attention. A new online prediction

system using the Spark streaming framework was

proposed to predict health status (Hassan et al., 2020).

An exercise app that uses a reinforcement learning

agent, was introduced to provide motion reminders at

appropriate times based on calendar information and

the user's instantaneous time (Wang et al., 2021).

Some researchers applied machine learning

algorithms on real-time data to find relationships

between blood pressure and lifestyle factors and to

provide precise advice for mitigating blood pressure

risks (Chiang et al., 2021).

After predicting possible health hazards,

personalized health management will also give

corresponding suggestions or reminders to guide

people to improve their personal health plans (Zhang

et al., 2023). Some work uses recommendation

systems to provide reference options for patients. In

2019, Subramaniyaswamy et al. introduced a

recommendation system, called ProTrip, which

supports travelers with long-term medical conditions

and users who need a strict diet to suggest food

availability is proposed by considering personal

preferences, the nutritional value of food, and climate

attributes (Subramaniyaswamy et al., 2019).

Moreover, an application using machine learning

algorithms is introduced to help patients with both

diabetes and depression by creating customized

messages with varying time and content based on

patients’ daily data (Aguilera et al., 2020). As

mentioned above, there have been a lot of AI-related

works in the field of personalized health

management, and this field is developing very fast at

present, so it is necessary to make a comprehensive

review of them.

The structure of the paper is as follows. First,

Section 2 sheds some light on the different parts of

the personalized health management process and

explains the role of artificial intelligence in

personalized health management. Then, current

challenges and limitations are summarized in Section

3. Finally, Section 4 makes a conclusion and an

outlook based on the previous discussion.

2 METHODS

2.1 Health Data Analysis and

Prediction

2.1.1 Classical Machine Learning-Based

Model

Chiang and Dey used exercise, sleep, and historical

Back Propagation (BP) measurements collected from

wearable devices and home BP monitors to predict

daily BP levels and estimate the influence of

individual health behaviors on BP (Chiang & Dey,

2018). They introduced a Random Forest (RF) with

feature selection (RFFS) model to filter out

unnecessary features and improve prediction

accuracy, aiming to provide personalized

recommendations for improving BP through sleep

and exercise. The proposed RFFS model has Mean

Square Error (MSE) and Mean Absolute Error (MAE)

of 47.33 and 5.18 for systolic blood pressure, 37.45

and 4.30 for diastolic blood pressure, respectively.

These values were lower compared to the results of

other deep learning models, providing better

performance in predicting blood pressure levels.

Lu et al. proposed a patient network and machine

learning based risk prediction model for Type 2

Diabetes Mellitus (T2DM) (Lu et al., 2022). A real-

world administrative claims dataset was used to

extract medical data from T2DM patients and non-

T2DM patients to construct a patient network. Using

patient network analysis and machine learning

algorithms, researchers extracted potential patient

characteristics, such as centrality measures, that

effectively predicted T2DM risk. Seven traditional

machine learning models were used for the

prediction. The centrality measure of the patient

network and the patient's age were the most

significant features in the random forest model, which

performed the best out of all the models, according to

the results data, which showed an accuracy of 83.98%.

2.1.2 Deep Learning-Based Model

A model named DeepRisk, based on attention

mechanisms and deep neural networks, was proposed

to automatically and efficiently select suitable

features from longitudinal and heterogeneous

electronic medical records from electronic medical

records, obtain accurate and robust patient

representations, and eventually estimate the patients'

AI in Personalized Health Management: Practices and Challenges

risk of developing cardiovascular disease (An et al.,

2019). The approximate working flow of the model is

as follows: Input data characteristics include age,

gender, patient type (inpatient, outpatient, and

emergency), number of visits, and surgical history.

These data are loaded respectively into their

embedding parts, producing independent embedding

vectors. For each input data, a deep neural network

based on the attention mechanism, or a plain deep

neural network is trained to generate a representation

vector of the patient. These representation vectors are

concatenated and then inserted into the softmax layer

to forecast patients at high risk. In comparison to

current methods, DeepRisk can greatly enhance the

accuracy of high-risk prediction of cardiovascular

disease, according to the findings of experiments

conducted on real medical datasets.

2.2 Genomic Analysis and Risk

Assessment

2.2.1 Genomic Data Processing and

Prediction

Dai et al. performed exon sequencing and gene

loading testing in 245 preterm infants (gestational age,

32 weeks) to identify two collections of risk genes

that are overexpressed in patients with

Bronchopulmonary Dysplasia (BPD) and severe BPD

(sBPD), named BPD-RGS and SBPD-RGS,

respectively (Dai et al., 2021). In the data analysis

phase, the authors used multiple tests to assess the

distribution of the data, followed by Ward

agglomeration method to cluster clinical

characteristics. Multivariate Logistic Regression

Analysis was then employed to evaluate the

independent association between clinical

characteristics and BPD or sBPD in 245 infants. The

prediction model was created using the Least

Absolute Shrinkage and Selection Operator

Regression (Lasso) technique. A new method for

accurate risk stratification of BPD in preterm infants

has been provided by the experimental results, which

demonstrated that prediction models combining

BPD-RG or sBPD-RGS with basic clinical risk

factors outperformed the models containing only

these factors in the independent test datasets.

2.2.2 Drug Development

The application of graph convolutional networks in

computational personalized medicine for medication

response prediction has attracted much attention.

Nguyen et al. discussed the limitations of existing

methods, namely representing drugs as strings, failing

to capture the native molecular structure, and lacking

an in-depth explanation of genetic mutations that

influence drug responses (Nguyen et al., 2021). The

suggested approach GraphDRP directly depicts cell

lines as binary vectors of genetic mutations and

pharmaceuticals as molecular graphs. By learning

features of drugs and cell lines through convolutional

layers and predicting response values of drug-cell line

pairs using fully connected neural networks,

GraphDRP outperformed tCNNS in all experiments.

Responses to drug-cell line pairs missing from the

GDSC dataset were predicted and analyzed. In

addition, using the saliency map, the authors

identified the ten most important genomic

abnormalities for the three cell lines with lower IC50

values for the corresponding drugs and their

contribution to drug sensitivity.

2.3 Nutrition and Exercise

Management

2.3.1 Individualized Nutrition Suggestion

Wang et al. investigated the application of machine

learning algorithms to assist in the early assessment

of Enteral Nutrition (EN) for patients in the Intensive

Care Unit (ICU) (Wang et al., 2023). The article

stated that malnutrition in intensive care patients may

lead to complications and poor outcomes; however,

the initiation of enteral nutrition often relies on the

awareness of the physician, resulting in poor feeding

proportions. The purpose of this study was to create

and verify a predictive model for enteral nutrition

initiation in intensive care unit patients using data

from the Critical Care IV database. This study

compared different machine learning models, and

found that XGBoost had the best prediction

performance. The model determined that the three

most significant parameters determining the

beginning of enteral feeding were acute kidney injury,

the score on the sequential organ function assessment,

and sepsis. This model was intended to help identify

those high-risk patients who may benefit from early

enteral nutrition, guide clinician decision making, and

enhance outcome outcomes for intensive care

patients.

2.3.2 Individualized Exercise Intervention

Determining the appropriate timing of intervention is

important for the effectiveness of exercise

recommendations. A reinforcement learn-based

mobile exercise application was introduced with the

aim of sending customized reminders to promote user

participation in physical activity based on the user's

temporary contextual information (Wang et al.,

EMITI 2024 - International Conference on Engineering Management, Information Technology and Intelligence

2021). Participants used the other functions in this

application for three weeks during the study period.

Every participant received no more than 14 reminders

from this app requesting physical exercise during the

fourth week, which was designated as the

intervention week. The collected data were analyzed

using questionnaires and one-on-one interviews. The

results showed that when sent at the right time, 83.3%

of users responded within 50 minutes and 66.7% of

users participated in physical activity within 5 hours.

In addition, the behavior of reminders could be traced

to detailed information, including the time stamp

when the reminder was sent, the notification click,

and the start of the campaign. In conclusion, by

building and testing intelligent reminder models, this

research investigated the viability of using

reinforcement learning models to send reminders in

smartphone sports applications. Based on the lessons

discovered, the timing of delivering reminders might

be further optimized.

3 DISCUSSIONS

Although existing works have achieved superior

performance in tasks such as prediction, they are

often trained, validated and tested on specific

datasets, which is not sufficient to accurately judge

whether the model has overfitting. This question can

be better addressed by validating the generalization

ability of the model on other datasets. In the case of

lack of medical data, transfer learning can also be

used to learn the common feature representations in

the existing datasets and transfer these feature

representations to the target task.

In addition, the widely-used deep learning

algorithms often lack interpretability. Due to the

complexity and privacy of medical data, the lack of

interpretability often makes healthcare workers and

users reluctant to trust these emerging artificial

intelligence tools. Some novel interpretable AI

techniques, model-interpretable methods, and

visualization tools are being developed to allow

healthcare professionals and patients to understand

the principles of how AI technologies work in more

detail (Stiglic et al., 2020).

How to translate the research results into

applications in real industrial products is also a

challenging problem. For example, some intelligent

medical platforms based on EHRs may face

inconsistencies in data heterogeneity and operating

standards caused by multiple data sources in practical

applications (Si et al., 2021). This may require

hospitals, healthcare companies, governments, and

many other institutes to work together to develop

standards. More interdisciplinary collaboration may

be needed in the future to cultivate talents who

understand both healthcare and artificial intelligence

in order to better manage data and use AI to aid

medical decision making (Qiu et al., 2022).

Personalized health management involves the

collection and processing of a large amount of

personal health data, so data privacy and security

become an important challenge. Although there are

some information encryption technologies (Suneetha

et al., 2020; López Martínez et al., 2023) that can

protect personal privacy to a certain extent, there are

still risks of data leakage and abuse. Federated

learning may be one answer to this problem, which

can help multiple agencies to collaborate on AI while

meeting the requirements of privacy protection, data

security, laws and regulations.

Large language models (LLMs), which have

recently become popular, are increasingly becoming

the channel through which people get information,

and conversational chatbots have great potential to

become a personalized health assistant. LLMs have

shown strong power in helping people understand

medical knowledge, monitoring vital signs,

complying with prescription requirements, and even

assisting with individualized medical decision

making (Abbasian et al., 2023; Benary et al., 2023).

However, not many projects have been actually

applied due to the possible problems of false output

and illusion, which can be caused by a bias in the

training data or insufficient predictive power of the

model in large language models (Xu et al., 2024). As

a result, most of the existing LLMs applications in the

medical field focus on instruction fine-tuning and

dataset construction. One of the future directions that

large language models can be used in healthcare field

is to construct knowledge graphs. Through the

analysis and collation of massive medical literature,

as well as the extraction of the knowledge and

experience of medical experts, the big model can

build a complete and accurate medical knowledge

graph, and provide more comprehensive and reliable

medical knowledge support for the medical inquiry

intelligent assistant.

4 CONCLUSIONS

This study investigates AI's potential for assisting

personalized healthcare, with a focus on how AI can

be used for personalized diagnosis of various

diseases, risk prediction, and intervention in health

planning at all stages of personalized healthcare

AI in Personalized Health Management: Practices and Challenges

management. This article also discusses the

challenges and opportunities, such as data quality,

interpretability, privacy issues, collaboration across

disciplines, and recent advances in LLMs. At present,

AI cannot replace professional medical staff, but the

assistance of AI can bring great convenience to

medical staff and users. AI can provide significant

benefits for personalized healthcare management, but

it also requires careful evaluation and

implementation. In the era of rapid technological

development, it is also time to think about the safety

and ethics problems of artificial intelligence

technology. In any case, there is still much potential

for the further development and application of

artificial intelligence in personalized healthcare

management, and future results are expected to better

improve medical outcomes and patient experience.

REFERENCES

Abbasian, M., Azimi, I., Rahmani, A. M., & Jain, R. 2023.

Conversational health agents: A personalized llm-

powered agent framework. arXiv preprint

arXiv:2310.02374.

Abdel-Fattah, M. A., Othman, N. A., & Goher, N. 2022.

Predicting Chronic Kidney Disease Using Hybrid

Machine Learning Based on Apache Spark.

Computational Intelligence and Neuroscience, 2022,

1–12.

Aguilera, A., Figueroa, C. A., Hernandez-Ramos, R.,

Sarkar, U., Cemballi, A., Gomez-Pathak, L.,

Miramontes, J., Yom-Tov, E., Chakraborty, B., Yan,

X., Xu, J., Modiri, A., Aggarwal, J., Jay Williams, J., &

Lyles, C. R. 2020. mHealth app using machine learning

to increase physical activity in diabetes and depression:

clinical trial protocol for the DIAMANTE Study. BMJ

Open, 10(8), e034723.

Ahmed, H., Younis, E. M. G., Hendawi, A., & Ali, A. A.

2019. Heart disease identification from patients ’

social posts, machine learning solution on Spark.

Future Generation Computer Systems, 111.

Alotaibi, S., Mehmood, R., Katib, I., Rana, O., & Albeshri,

A. 2020. Sehaa: A Big Data Analytics Tool for

Healthcare Symptoms and Diseases Detection Using

Twitter, Apache Spark, and Machine Learning. Applied

Sciences, 10(4), 1398.

An, Y., Huang, N., Chen, X., Wu, F., & Wang, J. 2019.

High-risk prediction of cardiovascular diseases via

attention-based deep neural networks. IEEE/ACM

transactions on computational biology and

bioinformatics, 18(3), 1093-1105.

Benary, M., Wang, X. D., Schmidt, M., Soll, D.,

Hilfenhaus, G., Nassir, M., ... & Rieke, D. T. 2023.

Leveraging large language models for decision support

in personalized oncology. JAMA Network Open, 6(11),

e2343689-e2343689.

Cheerla, N., & Gevaert, O. 2017. MicroRNA based Pan-

Cancer Diagnosis and Treatment Recommendation.

BMC Bioinformatics, 18(1).

Chiang, P.-H., & Dey, S. 2018. Personalized Effect of

Health Behavior on Blood Pressure: Machine Learning

Based Prediction and Recommendation. 2018 IEEE

20th International Conference on e-Health Networking,

Applications and Services (Healthcom) (pp. 1-6). IEEE.

Chiang, P.-H., Wong, M., & Dey, S. 2021. Using

Wearables and Machine Learning to Enable

Personalized Lifestyle Recommendations to Improve

Blood Pressure. IEEE Journal of Translational

Engineering in Health and Medicine, 9, 1–13.

Dai, D., Chen, H., Dong, X., Chen, J., Mei, M., Lu, Y., ...

& Zhou, W. 2021. Bronchopulmonary dysplasia

predicted by developing a machine learning model of

genetic and clinical information. Frontiers in

Genetics, 12, 689071.

Goetz, L. H., & Schork, N. J. 2018. Personalized medicine:

motivation, challenges, and progress. Fertility and

Sterility, 109(6), 952–963.

Hassan, F., E., M., & Sahal, R. 2020. Real-Time Healthcare

Monitoring System using Online Machine Learning

and Spark Streaming. International Journal of

Advanced Computer Science and Applications, 11(9).

Kim, S.-H., & Chung, K. 2015. Emergency situation

monitoring service using context motion tracking of

chronic disease patients. Cluster Computing, 18(2),

747–759.

Krishnamoorthi, R., Joshi, S., Almarzouki, H. Z., Shukla,

P. K., Rizwan, A., Kalpana, C., & Tiwari, B. 2022. A

Novel Diabetes Healthcare Disease Prediction

Framework Using Machine Learning Techniques.

Journal of Healthcare Engineering, 2022, 1–10.

Pinto, S., Cabral, J., & Gomes, T. 2017. We-care: An IoT-

based health care system for elderly people. 2017 IEEE

International Conference on Industrial Technology

(ICIT).

López Martínez, A., Gil Pérez, M., & Ruiz-Martínez, A.

2023. A comprehensive review of the state-of-the-art

on security and privacy issues in healthcare. ACM

Computing Surveys, 55(12), 1-38.

Lu, H., Uddin, S., Hajati, F., Moni, M. A., & Khushi, M.

2021. A patient network-based machine learning model

for disease prediction: The case of type 2 diabetes

mellitus. Applied Intelligence, 52(3), 2411-2422.

Nguyen, T., Nguyen, G. T., Nguyen, T., & Le, D. H. 2021.

Graph convolutional networks for drug response

prediction. IEEE/ACM transactions on computational

biology and bioinformatics, 19(1), 146-154.

Qiu, Y., Wang, J., Jin, Z., Chen, H., Zhang, M., & Guo, L.

2022. Pose-guided matching based on deep learning for

assessing quality of action on rehabilitation

training. Biomedical Signal Processing and

Control, 72, 103323.

Sekulić, I., & Strube, M. 2019. Adapting Deep Learning

Methods for Mental Health Prediction on Social Media.

Proceedings of the 5th Workshop on Noisy User-

Generated Text (W-NUT 2019), 322–327.

EMITI 2024 - International Conference on Engineering Management, Information Technology and Intelligence

Si, Y., Du, J., Li, Z., Jiang, X., Miller, T., Wang, F., ... &

Roberts, K. 2021. Deep representation learning of

patient data from Electronic Health Records (EHR): A

systematic review. Journal of biomedical

informatics, 115, 103671.

Stiglic, G., Kocbek, P., Fijacko, N., Zitnik, M., Verbert, K.,

& Cilar, L. 2020. Interpretability of machine learning‐

based prediction models in healthcare. Wiley

Interdisciplinary Reviews: Data Mining and

Knowledge Discovery, 10(5), e1379.

Subramaniyaswamy, V., Manogaran, G., Logesh, R.,

Vijayakumar, V., Chilamkurti, N., Malathi, D., &

Senthilselvan, N. 2018. An ontology-driven

personalized food recommendation in IoT-based

healthcare system. The Journal of Supercomputing,

75(6), 3184–3216.

Suneetha, V., Suresh, S., & Jhananie, V. 2020. A novel

framework using apache spark for privacy preservation

of healthcare big data. In 2020 2nd International

Conference on Innovative Mechanisms for Industry

Applications (ICIMIA) (pp. 743-749). IEEE.

Tiwari, S., & Agarwal, S. 2023. Empirical analysis of

chronic disease dataset for multiclass classification

using optimal feature selection based hybrid model with

spark streaming. Future Generation Computer Systems,

139, 87–99.

Trivizakis, E., Papadakis, G., Souglakos, I., Papanikolaou,

N., Koumakis, L., Spandidos, D., Tsatsakis, A.,

Karantanas, A., & Marias, K. 2020. Artificial

intelligence radiogenomics for advancing precision and

effectiveness in oncologic care (Review). International

Journal of Oncology, 57(1), 43–53.

Wang, S., Sporrel, K., van Hoof, H., Simons, M., de Boer,

R. D. D., Ettema, D., Nibbeling, N., Deutekom, M., &

Kröse, B. 2021. Reinforcement Learning to Send

Reminders at Right Moments in Smartphone Exercise

Application: A Feasibility Study. International Journal

of Environmental Research and Public Health, 18(11),

6059.

Wang, Y. X., Li, X. L., Zhang, L. H., Li, H. N., Liu, X. M.,

Song, W., & Pang, X. F. 2023. Machine learning

algorithms assist early evaluation of enteral nutrition in

ICU patients. Frontiers in nutrition, 10, 1060398.

Xu, Z., Jain, S., & Kankanhalli, M. 2024. Hallucination is

inevitable: An innate limitation of large language

models. arXiv preprint arXiv:2401.11817.

Zhang, D., Liu, X., Xia, J., Gao, Z., Zhang, H., & de

Albuquerque, V. H. C. 2023. A Physics-guided Deep

Learning Approach For Functional Assessment of

Cardiovascular Disease in IoT-based Smart

Health.

IEEE Internet of Things Journal, 1–1.

AI in Personalized Health Management: Practices and Challenges