Medical care is fraught with uncertainty, especially during times of transition. Patients bear medical, mental, emotional, psychological, and financial burdens when they are readmitted shortly after being discharged from the hospital. What if there was a way to decrease the risk of rehospitalization by identifying at-risk patients earlier? There is.
Imagine you are a busy nurse responsible for several residents for the night. You visit them one by one, taking vital signs, passing meds, distributing dinner trays, and assisting each in preparing for bed.
You finally have a few minutes to catch up on your charting. You pull out your notes and start entering data into the electronic health record. Mrs. W. in room 202 was running a low-grade fever and did not seem like herself. Everyone else had no noticeable concerns.
As you enter your patient’s information, you are called away several times to attend to their needs. Finally, as your shift ends, you are tired but satisfied. All your patient data has been entered, everyone has had breakfast, and you’re transferring care to the next nurse.
You barely had an extra moment all night long, but this is not unusual. You wish you had the time to review Mrs. W’s medical record to see if there are any warning signs of impending illness. You make a mental note to check on her as soon as your next shift begins.
Mrs. W. has a slightly elevated heart rate two days later and appears to be having more trouble staying awake.
The next day, the charge nurse is called to Mrs. W’s room. She is incoherent. Her temperature is now 103 degrees Fahrenheit. She is struggling to breathe. She is immediately transferred to the nearest hospital by ambulance.
This is Mrs. W’s third admission this year. Was there a clue in her chart that could have alerted you three days ago that she was developing pneumonia and would soon be septic?
Artificial Intelligence Can Help You Find the Needle in a Haystack of Data
What if a computer could scan all your residents’ medical records every night, looking for patterns that could help predict the future? The computer would note the slightly increased temperature you recorded three days ago. It would flag the note the nursing assistant added the next day. Mrs. W. did not eat her dinner.
Mrs. W. slept the majority of the next day. Her family had visited her the night before, so no one thought anything of it. But, now it seems significant.
Three days later, you learn that Mrs. W. died from sepsis. She received all the best possible care in the hospital. Unfortunately, treatments were started too late. Sepsis is preventable, but only if it is treated promptly. Each hour makes a difference.
Sepsis is incredibly difficult to diagnose, even in the best hospital settings in the world.
Thoughts swirl through your mind. Was this preventable? Did you miss key information?
Artificial intelligence is a field of study that seeks to write programs that can mimic and even go beyond human thought processes. Computers can parse through the mountains of data in your patients’ charts and send you a report each morning that ranks all your patients.
The report identifies the patients who are most likely to require rehospitalization within the next 72 hours, as well as the factors that the system “believes” are the most important contributors to your patients’ risk.
The SAIVA report identifies Mrs. W. as the resident most likely to be hospitalized. It pulls together multiple data points such as her increased temperature, lack of appetite, and fatigue. When a nurse sees this information, the diagnosis becomes clear. Her doctor is notified, appropriate labs and X-rays are taken, and orders for antibiotics are executed.
Imagine if you were given this valuable information three days ago. Data points that could be easily explained away on their own suddenly make a much clearer picture.
Mrs. W. was developing a serious infection and needed antibiotics immediately.
Machine Learning Leads to Better Medical Care
An overloaded healthcare provider often does not have time to read through all the entries in every patient’s electronic health record and identify trends that may be buried in reams of data. However, buried in that data could be a critical piece of information that could alert you that your patient’s health is about to deteriorate.
Machine learning is a methodology used in computer science and artificial intelligence. It seeks to make predictions about future events. It uses retrospective data to build correlations between previous events, models them, and then uses prospective data to make predictions about future events, such as Mrs. W’s impending illness and subsequent hospitalization.
Computers can help you make sense of an overwhelming amount of patient data to make informed clinical decisions.
Types of data the system can evaluate:
- Vital signs
- Physician orders
- Lab results
- Assessment results
- Progress notes
How Computers Use Data to Predict Events
Machine learning uses the millions of records in your patients’ electronic health records to build a model that is then used to predict the likelihood of rehospitalization.
Here are the steps:
Step 1: Training
- State the goal and write it in a language that the computer can understand. Here, our goal is to identify those patients who are most likely to need rehospitalization soon.
- Analyze the thousands of variables that are available in the medical record to see if they will help in making an accurate prediction. Each variable is evaluated in the context of every other variable.
- Compute a correlation score for each variable. Variables that aid in prediction are given a higher weight than those that do not.
- Experiment with different models to see how the predictive score can be maximized over your historical hospitalization data.
Step 2: Testing
- Use data from the last couple of months, data that the model never saw during training, to ensure the models work.
Step 3: Deployment
- Use the models to start evaluating medical information and sending daily reports.
AI and Machine Learning Support Your Work
AI can process enormous amounts of data and make predictions, but it cannot (and never aims to) replace skilled professionals who use these predictions to prioritize care, evaluate treatment plans, and develop interventions to reduce the risk of rehospitalization. Rather than “artificial intelligence” think of it as “intelligent augmentation”. Experienced nurses using SAIVA’s system have reduced hospitalizations by up to 52%!
These exceptional results can only be achieved by using the data available in your residents’ medical records, applying the SAIVA model, reviewing and discussing the report, and intervening when necessary to decrease the risk of rehospitalization.