Patients admitted for an infection, surgery or medical issue can experience sudden symptoms such as confusion, loss of time orientation and problems recognizing their environment. Sometimes these symptoms can be seen as just confusion; however, they can indicate the patient has developed delirium.
Sometimes delirium can be hard to spot as symptoms can come on quickly and often fluctuate throughout the day. Northwell Health's Feinstein Institutes for Medical Research are investigating whether AI can help hospitals detect these symptoms more quickly.
The Institute has received a $3.4 million grant from the National Institutes of Health (NIH), through the National Institute on Aging (NIA), to design machine learning models to detect and predict delirium in hospitals using electronic health records and wearable devices.
The project will be led by Dr Liron Sinvani, Dr Alex Makhnevich and Dr Theodoros P. Zanos.
Delirium is an acute disturbance causing problems with attention, awareness and cognition. Unlike dementia, the symptoms usually come on quickly, over hours or days. Symptoms can also fluctuate throughout the day. Delirium can also develop in people who already have dementia, making it important to recognize sudden changes from their usual cognitive and behavioral state.
Patients can display agitation, restlessness and abnormal withdrawal. Some patients will have problems understanding others, become confused or fail to recognize those around them. Some patients will experience hallucinations and abnormal sleeping patterns.
Not all patients will present with clear agitation, and a type of delirium called hypoactive delirium can be difficult to spot. Patients can appear unusually quiet, sleepy and unresponsive. This can often be seen as tiredness or attributed to their underlying illness.
Older patients admitted to the hospital are at increased risk of developing delirium. This can be due to severe infections, surgery, pain, dehydration, certain medications and metabolic issues.
Delirium can cause many issues for patients. It can increase the risk of falls, increase the length of hospital admission and cause ongoing cognitive issues. Some patients can suffer from memory and cognitive issues.
Even with an intensive screening program, more than 75% of delirium cases among older adults in the emergency department were missed in the researchers' previous work.
Researchers will design machine learning models to analyze data from electronic health records and continuous data from wearable devices.
Electronic health records will provide the researchers with information regarding the patient’s medical history, diagnoses, assessments and treatments. Wearable devices will allow the researchers to collect continuous data from the patient.
By using these resources together, researchers may be able to see patterns that may not be obvious to clinicians.
The project has three main focuses:
The first model will be for patients admitted to the emergency department with delirium.
As the emergency department is often the first place patients go for urgent medical care, this could allow clinicians to have a better understanding of how the patient is presenting.
The second model will help identify patients likely to develop delirium during their admission.
According to the NIH grant record, 20–40% of older adults admitted to hospital may develop delirium. This project hopes to identify patients that are at higher risk of developing delirium.
This could help the healthcare team look into risk factors, ensure those at risk are closely monitored and take preventative measures.
While designing an algorithm is important, the team will also work with doctors, nurses and hospital administration to understand how they can implement the models.
This is very important, as if the AI does not help clinicians make decisions in a timely manner, the models may not be of much use.
They will develop and validate models using a range of patient groups.
The NIH grant record provides further details about the planned validation process. The emergency department model will undergo prospective validation involving 1,800 patients, while the hospitalization prediction model will be prospectively validated in 1,000 patients. Researchers will use the Confusion Assessment Method (CAM) as a reference standard for assessing delirium. The project will also analyze structured electronic health record data and unstructured information, including clinical notes.
The project being funded by the NIH comes at a time where researchers are looking into ways to use AI to help with delirium.
A recent review published online in the Journal of the American Geriatrics Society looked at how new technologies could help with some of the issues relating to delirium detection and management. Published on 5th November 2026, the title was Artificial Intelligence as a Bridge to Gaps in Delirium Prediction, Detection, and Management.
This paper (PubMed) discusses how machine learning algorithms could be used to analyze data from electronic health records. This could include structured data and notes.
Other technologies discussed include Natural Language Processing, Large Language Models, Computer Vision, Wearables and Ambient Sensors. This could all help in monitoring patients in a more continuous manner; Reducing some of the workload associated with conducting frequent manual assessments.
However, they also mention that good models do not always equate to improved patient outcomes.
Some models have shown potential; however there are few models used within everyday practice. Many models have issues with false positives, different patients and generalizability.
More delirium cases detected does not necessarily mean patients will recover quicker or have less complications. The clinicians must be able to use the information.
One interesting aspect of this project is that they want to utilize continuous data capture from wearables. While patients would be monitored regularly in hospital, wearables may help provide more data in between.
This can be used alongside electronic patient records to help understand if there are any changes in patients condition and whether there are any patterns related to delirium.
Wearable devices can provide continuous physiological measurements, such as heart rate and respiratory rate, which may help identify changes in a patient's condition. However, these measurements cannot independently establish a diagnosis of delirium.
They will use initial data from the largest inpatient dataset in the United States. This was collected by the Division of Health AI with help from Scott and Debbie Rechler.
It is highlighted in the announcement that previous work led by Dr Zanos has been published in Nature Communications. It found that a wearable, AI powered monitoring system could predict deterioration in patients up to 17 hours in advance.
This provides some evidence that wearables may be useful in delirium research; however, it is not specific to delirium so may not be as relevant. We will have to see if we can accurately identify changes related to delirium.
Delirium can cause problems for older patients. Some patients will have long-term cognitive problems, will need additional support following discharge and struggle to get back to their usual level of independence.
By identifying delirium sooner, we can give patients more time to investigate what could be causing the delirium. We can review medications, ensure they are not dehydrated and treat any infection. We can implement preventative measures.
Patients who may have been missed can be closely monitored by nurses and doctors. AI is unlikely to be the answer. The models will have to be accurate for different types of patients, and those who are alerted need to be managed well. There are also issues around patient confidentiality, data security, false positives and ensuring humans are monitoring patients.
Delirium is also an important concern in Indian hospitals. A prospective observational study conducted at AIIMS, New Delhi, involving 200 older patients admitted to a geriatric ward, reported a delirium prevalence of 20% and an incidence of 5% during hospitalization.
The researchers also found that patients with delirium had significantly higher in-hospital and post-discharge mortality, along with greater functional dependency at the one-year follow-up. The findings highlight the importance of recognizing delirium early and monitoring older hospitalized patients.
The $3.4 million funding from the NIH will help the team create and validate the models. More research will be needed to see whether they can help reduce instances of delirium. Overall, this is an attempt to target an often overlooked condition. If the technology works, it could be another tool for hospitals to identify early signs.
References:
1. National Institute on Aging. “Machine Learning Models to Identify and Predict Delirium in Hospitalized Older Adults.” National Institutes of Health, 2026. https://taggs.hhs.gov/Detail/AwardDetail?arg_AwardNum=R01AG100193.
2. Ajmera, Yamini, Karandeep Paul, Maroof Ahmad Khan, Bharti Kumari, Nand Kumar, Prasun Chatterjee, Aparajit Ballav Dey, and Avinash Chakrawarty. “The Evaluation of Frequency and Predictors of Delirium and Its Short-Term and Long-Term Outcomes in Hospitalized Older Adults.” Asian Journal of Psychiatry 94 (2024): 103990. https://doi.org/10.1016/j.ajp.2024.103990.
3. Pagali, Sandeep R., Ariba Khan, Sunyang Fu, Mfon E. Umoh, Minji Kwak, Heidi Lindroth, Tru Brynes, Lara Magnabosco, David Karol, Marie Pavini, Esther S. Oh, and American Delirium Society Innovation in Delirium Special Interest Group. “Artificial Intelligence as a Bridge to Gaps in Delirium Prediction, Detection, and Management.” Journal of the American Geriatrics Society, November 5, 2026. https://pubmed.ncbi.nlm.nih.gov/42988738/.
4. Grover, Sandeep, and Ajit Avasthi. “Clinical Practice Guidelines for Management of Delirium in Elderly.” Indian Journal of Psychiatry 60, Suppl. 3 (2018): S329–S340. https://doi.org/10.4103/0019-5545.224473.
5. Scheid, Michael R., Beth Friedmann, Michael Oppenheim, Jamie S. Hirsch, and Theodoros P. Zanos. “Development and Validation of a Clinical Wearable Deep Learning Based Continuous Inhospital Deterioration Prediction Model.” Nature Communications 16 (2025): 9513. https://doi.org/10.1038/s41467-025-65219-8.
Rh/MF/MSM