Predicting Adverse Events in Developmental Disabilities Population

James P. McGlothlin

, Micah Price-Offerman

, Robbie Beyer

, George Casey

and John P. Barile

RSM US LLP, Chicago, IL, U.S.A.

University of Hawai‘i at Mānoa, Havaii, U.S.A.

Keywords: Artificial Intelligence, Development Disabilities, Population Health, Analytics, Business Intelligence,

Predictive Analytics, Machine Learning, Supervised Learning.

Abstract: Individuals with development disabilities can experience a variety of adverse events. We have found that

these events are often unreported. In this project, we work with a large government program which assists

such individuals. The goal of the project is to use artificial intelligence (AI) and other modern technologies to

predict adverse events. This will allow case managers to better avoid adverse events, prepare for them and

help the program participants. Our initial results show very good accuracy and precision in identifying risk

and predicting participant adverse events.

1 INTRODUCTION

The Developmental Disabilities Division (DDD)

supports ~3,600 active participants. DDD is the

operating state agency for the Medicaid 1915(c)

Home and Community Based Services (HCBS)

Waiver for Individuals with Intellectual and

Developmental Disabilities. The department’s

mission is to “Foster partnerships and provide quality

person-centered and family-focused services and

supports that promote self-determination.”. The

department’s vision is “Individuals with intellectual

and developmental disabilities have healthy, safe,

meaningful and self-determined lives.”

The goal of this product is to determine which

participants are most at risk for an adverse event. Our

study looks at 10 types of adverse events:

1. Suspected abuse, neglect or financial

exploitation

2. Behavior Change

3. Change in Health Requiring Medical

Treatment

4. Any Use of Restraints

5. Injury from a Known/Unknown Cause

Requiring Medical Treatment

6. Medication Errors and/or Unexpected

Reaction to Medications or Treatment

7. Participant's Whereabouts Unknown

8. Use of Prohibited Intervention

9. Use of Seclusion

10. Death

Our research has three goals. The first goal is to

identify the majority of adverse events. DDD uses a

custom participant records system called Inspire.

Anecdotally, it is expected that less than half of

adverse events are correctly documented in Inspire.

Therefore, we leverage Medicaid claims data to

augment the participant records. As the vast majority

of the participants have coverage through the

government Medicaid program, this method should

allow us to identify any hospital visits, clinical visits

or medication dispenses for the participants.

The second goal is to predict adverse events

before they happen. This is the central exercise of this

paper and experiment.

For the purpose of our data and models, the grain

is one row per participant per month. So if the

participant has been in the program for a year they

will have 12 rows. For each participant and month,

we predict yes there will be an adverse event or no

there will not be. Therefore, while our system does

not attempt to predict exactly when an adverse event

will occur, it does predict it within the calendar

month.

The third and final goal is to utilize these

predictions to take prescriptive action and prevent

adverse events. This will require making suggestions

for interventions and tracking the input of these

interventions on participant outcomes.

844

McGlothlin, J. P., Price-Offerman, M., Beyer, R., Casey, G. and Barile, J. P.

Predicting Adverse Events in Developmental Disabilities Population.

DOI: 10.5220/0013318200003911

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

In Proceedings of the 18th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2025) - Volume 2: HEALTHINF, pages 844-849

ISBN: 978-989-758-731-3; ISSN: 2184-4305

2 ARTIFICIAL INTELLIGENCE

The goal of this product is to predict adverse events

before they occur. To do this, we leverage machine

learning and predictive analytics.

Machine learning (ML) has become a

transformative tool in various sectors, and public

health is no exception. At its core, machine learning

involves algorithms that can learn from data, identify

patterns, and make decisions or predictions without

explicit programming for each task. In public health,

ML is applied to analyze large volumes of data such

as electronic health records, genomic information,

and social determinants of health. This enables the

identification of trends and patterns that may not be

immediately obvious to human researchers. With the

power of ML, public health systems can improve

outcomes through early disease detection, predictive

modeling, and more efficient resource allocation

(Jordan, 2015) (Bi, 2019).

One significant application of machine learning in

public health is disease prediction and prevention.

ML algorithms are capable of processing complex

datasets to predict the likelihood of diseases based on

various risk factors. For example, ML models have

been used to predict the onset of chronic diseases like

diabetes and cardiovascular conditions (Siontis,

2012) (Collins, 2012). By analyzing factors such as

age, lifestyle, genetics, and environmental influences,

ML can forecast the potential for disease in

individuals or populations, allowing for early

interventions. This predictive power is particularly

valuable in resource-limited settings where

preventive measures can save lives and reduce

healthcare costs. Machine learning is making strides

is in epidemiology, especially in tracking and

controlling infectious diseases (Ghosh, 2024)

(Adegoke, 2024). ML algorithms are being used to

analyze patterns in disease spread and to create

models for forecasting outbreaks. During the

COVID-19 pandemic, ML models were widely

employed to predict the spread of the virus, assess

healthcare system burdens, and identify effective

intervention strategies (Van der Schaar, 2021)

(Malik, 2021) (Heidari, 2022). These models relied

on a combination of epidemiological data, mobility

data, and demographic information. In addition, ML

has been applied to track the emergence of antibiotic-

resistant bacteria, thereby enhancing surveillance

efforts and informing public health responses

(Brenda, 2024) .

In addition to modeling, ML is enhancing

personalized medicine and treatment in public health

(Srinivasaiah, 2024). By analyzing vast datasets,

machine learning can tailor healthcare interventions

to individuals based on their unique characteristics.

This is particularly important in managing chronic

diseases, where treatment regimens can vary

significantly from one person to another. For

instance, ML algorithms can help determine the most

effective treatment plans for cancer patients by

analyzing genetic data and patient responses to

previous treatments (Rafique, 2021) (Quazi, 2022)..

This precision medicine approach not only improves

individual outcomes but also reduces the

inefficiencies of one-size-fits-all healthcare

strategies.

Despite its potential, machine learning in public

health comes with challenges. These include data

privacy concerns, ethical issues regarding algorithmic

biases, and the need for sufficient training of

healthcare professionals in data science. Moreover,

the success of ML models in public health is heavily

dependent on the quality of the data used for training

these models. Inaccurate, incomplete, or biased data

can lead to misleading predictions and decisions. As

such, there is an ongoing need for collaboration

between data scientists, healthcare professionals, and

policymakers to ensure that ML applications are

designed, tested, and implemented responsibly.

Supervised learning is a type of machine learning

where the model is trained on labeled data, meaning

each input is paired with the correct output. The goal

is to learn a mapping from inputs to outputs so that,

when presented with new, unseen data, the model can

predict the correct result (Osisanwo, 2017). The

process involves using a dataset with known labels to

train the algorithm, which then fine-tunes itself by

adjusting its internal parameters to minimize errors

between predicted and actual outcomes. This form of

learning is widely used in tasks such as classification

and regression, where the model learns to categorize

data or predict continuous values based on historical

examples (Kotsiantis, 2006).

In healthcare, supervised learning has shown

significant potential in improving diagnostic

accuracy, personalized treatment plans, and

predicting patient outcomes. For instance, machine

learning models can be trained on medical images

like MRIs or X-rays, where the labels correspond to

specific diagnoses, enabling the algorithm to assist

radiologists in detecting diseases such as cancer or

tuberculosis with high accuracy (Sharma, 2025).

Supervised learning is also used in predicting patient

risk factors, such as the likelihood of developing

chronic diseases like diabetes or heart disease, based

on historical health data, lifestyle choices, and genetic

factors (Islam, 2024). This application helps

Predicting Adverse Events in Developmental Disabilities Population

845

healthcare professionals provide more tailored

treatments and preventative measures, thereby

improving patient care and reducing overall

healthcare costs (Razzak, 2018).

Predictive analytics models use statistical

algorithms and machine learning techniques to

analyze historical data and predict future outcomes.

These models are built upon data mining and pattern

recognition principles, helping organizations forecast

trends, behaviors, and events. For example,

regression analysis is often used in predictive

analytics to identify relationships between variables,

while classification algorithms like decision trees or

random forests help categorize data points into

predefined classes. These predictive models are

widely used in various sectors, including finance,

healthcare, marketing, and supply chain management,

offering insights that guide decision-making and

improve operational efficiency (Berrar, 2019).

One of the key advantages of predictive analytics

is its ability to enhance decision-making by offering

actionable insights based on historical data. Machine

learning models, particularly deep learning and neural

networks, allow for complex, nonlinear relationships

within data to be understood, improving the accuracy

of predictions over traditional statistical methods. For

instance, in the healthcare sector, predictive models

can help identify patients at risk of developing

chronic conditions, thereby enabling early

intervention and personalized care plans (Chung et

al., 2018). These capabilities empower businesses to

proactively address issues, reduce costs, and increase

customer satisfaction by anticipating needs and

actions.

However, despite their power, predictive analytics

models also come with challenges. The effectiveness

of these models is highly dependent on the quality and

volume of the data being analyzed. Inaccurate,

incomplete, or biased data can lead to incorrect

predictions, which could be detrimental in areas like

finance or healthcare. Moreover, predictive models

can be resource-intensive, requiring significant

computational power and expertise to develop and

maintain. It is also crucial to continually update the

models with new data to ensure they remain relevant

and accurate. Addressing these challenges requires

robust data governance practices and collaboration

between data scientists and domain experts (Aguirre,

2019).

3 IMPLEMENTATION

APPROACH

This project is intended to be used in a commercial

setting by hospital providers, so that they can comply

with the requirements of patient registries with less

burden to hospital staff. Therefore, we wanted to only

use commercially available and respected software

products which have been approved to handle

protected health information (PHI) under the United

States’s HIPAA (Health Insurance Portability and

Accountability Act of 1996) (Moore, 2019).

Therefore, we chose to implement our project

using software available from Microsoft including

Azure, Azure Machine Learning (AML) (Barga,

2015) (Barnes, 2015) and OpenAI.

Azure Machine Learning is a cloud service from

Microsoft designed to streamline the machine

learning process. It provides a variety of tools for

building, training, and deploying machine learning

models, catering to data scientists, developers, and

organizations. The platform supports integration with

popular frameworks and offers features such as

automated machine learning (AutoML), model

versioning, and deployment in a secure, scalable

environment. Notable capabilities include automated

hyperparameter tuning, experiment tracking, and

seamless integration with Azure’s cloud

infrastructure for efficient model management. Azure

Machine Learning also supports collaborative

development through integrated notebooks and offers

monitoring and management tools post-deployment.

It accommodates both code-based and low-code

development, making it accessible for users with

different skill levels. This flexibility enables

businesses to advance their AI projects while

ensuring governance, security, and scalability in

production environments. (Barnes, 2015).

The department uses two source systems to

manage participant records: Inspire and MedQuest.

Inspire is a custom solution built in Microsoft

Dynamics. Med-Quest is a custom solution built to

manage Medicaid claims.

Our evaluation determined that adverse events

were overwhelmingly under-reported in Inspire.

Augmenting the participate record with

corresponding claims data allows us to identify any

adverse event that resulted in a medical claim. Figure

1 shows the data flow diagram for both the data

abstraction and the machine learning.

HEALTHINF 2025 - 18th International Conference on Health Informatics

846

Figure 1: Architectural data flow diagram.

One of the most important factors for successful

model is to determine which features are valuable to

the prediction and have a true causal effect. Data

fields which we evaluated and leveraged include

participant age, race, use of medical devices, care

setting, level of participation in program, history of

trauma, medications and history of schizophrenia,

4 RESULTS

This research is in early stages of development and

test.

Our model made 24,146 predictions. The

algorithm predicted adverse events in 6,181

instances. In 6,156 instances the adverse event

occurred. Therefore, when an adverse event is

predicted, there is a 99.6% chance the adverse event

will occur. There were only 25 false positives,

representing only 0.1% of the preductions. There

were 246 false negatives, where the model failed to

predict a adverse event. This represents 1.0% percent

of predictions. The predictions were correct overall

98.9% of the time.

Figure 2: Precision and Recall.

Figure 2 shows the precision and recall for our model.

Figure 3 the calibiration curve and predicted

probability.

Figure 3: Calibration Curve.

5 NEXT STEPS

Currently, our research predicts the occurrence of any

adverse event in a given month. It would be

preferable if we could predict exactly which adverse

event will occur. For a next step, we will attempt to

model each adverse event separately. This will also

allow for a distinct set of features per event, instead

of forcing a common set of prediction features across

all adverse events.

Additionally, we realize that more model tuning,

training and rigor is needed to show the statistical

significance of our algorithm and process over time.

There is a large amount of free text data collected

for the program participants. As a future long term

step, we would like to edit the records management

software to utilize drop down lists and not allow free

text. In order to get value out of the vast amount of

text data already collected, we plan to use generative

AI to summarize the data.

Predicting Adverse Events in Developmental Disabilities Population

847

As stated in our introduction, the third and final

goal of the project is to utilize these prediction to take

prescriptive action and prevent adverse events. This

will require making suggestions for interventions and

tracking the input of these interventions on participant

outcomes. This is left for future work and has not yet

been attempted. To do this properly, we will need to

provide integrated data into the case management

system enabling our case managers to take

appropriate action.

6 CONCLUSIONS

It is vital that both case workers and researchers know

when participants have adverse events. By

augmenting the participant records with claims data,

we were able to almost double the number of known

adverse events.

The primary purpose of this initiative is to predict

adverse events before they happen. While this is a

preliminary evaluation, our early results show an

exceptional 98.9% accuracy across all predictions.

This shows the promise of AI to help these

participants.

REFERENCES

Aguirre, S., Mahr, D., & Rivas, M. (2019). Data-driven

predictive analytics in business decision-making.

Springer.

Berrar, D. (2019). Predictive modeling applications in

business and industry. Wiley-IEEE Press.

Chung, W. Y., Park, J. H., & Kim, H. (2018). Applications

of predictive analytics in healthcare. Springer.

Mahesh, B. (2020). Machine learning algorithms-a

review. International Journal of Science and Research

(IJSR).[Internet], 9(1), 381-386.

Jordan, M. I., & Mitchell, T. M. (2015). Machine

learning: Trends, perspectives, and prospects.

Science, 349(6245), 255-260.

Bi, Q., Goodman, K. E., Kaminsky, J., & Lessler, J. (2019).

What is machine learning? A primer for the

epidemiologist. American journal of

epidemiology, 188(12), 2222-2239.

Panch, T., Szolovits, P., & Atun, R. (2018). Artificial

intelligence, machine learning and health

systems. Journal of global health, 8(2).

Ravì, D., Wong, C., Deligianni, F., Berthelot, M., Andreu-

Perez, J., Lo, B., & Yang, G. Z. (2016). Deep learning

for health informatics. IEEE journal of biomedical and

health informatics, 21(1), 4-21.

Johnson, A. E., Pollard, T. J., & Mark, R. G. (2017,

November). Reproducibility in critical care: a mortality

prediction case study. In Machine learning for

healthcare conference (pp. 361-376). PMLR.

Shen, D., Wu, G., & Suk, H. I. (2017). Deep learning in

medical image analysis. Annual review of biomedical

engineering, 19(1), 221-248.

Bishop, C. M., & Nasrabadi, N. M. (2006). Pattern

recognition and machine learning (Vol. 4, No. 4, p.

738). New York: Springer.

Kotsiantis, S. B., Zaharakis, I., & Pintelas, P. (2007).

Supervised machine learning: A review of

classification techniques. Emerging artificial

intelligence applications in computer

engineering, 160(1), 3-24.

LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning.

Nature, 521(7553), 436-444.

Moore, W., & Frye, S. (2019). Review of HIPAA, part 1:

history, protected health information, and privacy and

security rules. Journal of nuclear medicine

technology, 47(4), 269-272.

Barga, R., Fontama, V., Tok, W. H., & Cabrera-Cordon, L.

(2015). Predictive analytics with Microsoft Azure

machine learning (pp. 221-241). Berkely, CA: Apress.

Barnes, J. (2015). Azure machine learning. Microsoft

Azure Essentials. 1st ed, Microsoft.

Siontis, G. C., Tzoulaki, I., Siontis, K. C., & Ioannidis, J. P.

(2012). Comparisons of established risk prediction

models for cardiovascular disease: systematic

review. Bmj, 344.

Collins, G. S., & Moons, K. G. (2012). Comparing risk

prediction models. Bmj, 344.

Ghosh, S., & Sharma, V. (2024). Tracking of Disease—A

Review of the State of the Art of Technology for Next

Generation Healthcare. Deep Learning in Internet of

Things for Next Generation Healthcare, 242-268.

Adegoke, B. O., Odugbose, T., & Adeyemi, C. (2024). Data

analytics for predicting disease outbreaks: A review of

models and tools. International journal of life science

research updates [online], 2(2), 1-9.

Van der Schaar, M., Alaa, A. M., Floto, A., Gimson, A.,

Scholtes, S., Wood, A., ... & Ercole, A. (2021). How

artificial intelligence and machine learning can help

healthcare systems respond to COVID-19. Machine

Learning, 110, 1-14.

Malik, Y. S., Sircar, S., Bhat, S., Ansari, M. I., Pande, T.,

Kumar, P., ... & Dhama, K. (2021). How artificial

intelligence may help the Covid‐19 pandemic: Pitfalls

and lessons for the future. Reviews in medical

virology, 31(5), 1-11.

Heidari, A., Jafari Navimipour, N., Unal, M., & Toumaj, S.

(2022). Machine learning applications for COVID-19

outbreak management. Neural Computing and

Applications, 34(18), 15313-15348.

Branda, F., & Scarpa, F. (2024). Implications of Artificial

Intelligence in Addressing Antimicrobial Resistance:

Innovations, Global Challenges, and Healthcare’s

Future. Antibiotics, 13(6), 502.

Rodríguez-González, A., Zanin, M., & Menasalvas-Ruiz,

E. (2019). Public health and epidemiology informatics:

can artificial intelligence help future global challenges?

An overview of antimicrobial resistance and impact of

HEALTHINF 2025 - 18th International Conference on Health Informatics

848

climate change in disease epidemiology. Yearbook of

medical informatics, 28(01), 224-231.

Srinivasaiah, B. (2024). The Power of Personalized

Healthcare: Harnessing the Potential of Machine

Learning in Precision Medicine. International Journal

of Science and Research, 426-429.

Rafique, R., Islam, S. R., & Kazi, J. U. (2021). Machine

learning in the prediction of cancer

therapy. Computational and Structural Biotechnology

Journal, 19, 4003-4017.

Quazi, S. (2022). Artificial intelligence and machine

learning in precision and genomic medicine. Medical

Oncology, 39(8), 120.

Osisanwo, F. Y., Akinsola, J. E. T., Awodele, O.,

Hinmikaiye, J. O., Olakanmi, O., & Akinjobi, J. (2017).

Supervised machine learning algorithms: classification

and comparison. International Journal of Computer

Trends and Technology (IJCTT), 48(3), 128-138.

Kotsiantis, S. B., Zaharakis, I. D., & Pintelas, P. E. (2006).

Machine learning: a review of classification and

combining techniques. Artificial Intelligence

Review, 26, 159-190.

Sharma, R., Kumar, N., & Sharma, V. (2025). Computer

Vision for Disease Detection—An Overview of How

Computer Vision Techniques Can Be Used to Detect

Diseases in Medical Images, Such as X‐Rays and

MRIs. AI in Disease Detection: Advancements and

Applications, 77-98.

Islam, R., Sultana, A., & Islam, M. R. (2024). A

comprehensive review for chronic disease prediction

using machine learning algorithms. Journal of Electrical

Systems and Information Technology, 11(1), 27.

Predicting Adverse Events in Developmental Disabilities Population

849