Enhancing AI in Chronic Disease Management with Data Collection

To effectively use artificial intelligence in chronic disease management, providers must have high-quality and accurate data. According to University of Pennsylvania Medical Center (UPMC) Chief Health Care Data And Analytics Officer Oscar Marroquin, MD, data scientists strive for real-world data to derive insights to be integrated into clinical programs.

“We need to use real-world data, which is the patient-level data not collected in conventional randomized clinical trials, and it includes a lot of the data we know is being collected as we’re delivering care. If we apply the right analytics to those data, we can derive the insights that we need in order to drive our organization by coming up with new real-world based evidence,” Marroquin said.

Electronic health records (EHRs) have become an important tool for patient data storage. However, comparing data across EHR platforms presents such challenges. To address this issue, UPMC built data warehouses to derive real-world evidence from aggregated data assets. The data is then used explicitly for analytics.

Maintaining the data warehouse is a collaborative effort among IT, data engineers, clinical analysts, and business intelligence specialists. Through their work, UPMC has developed many self-service analytics tools that allow businesses and clinicians to access data, insights, and evidence derived by the medical center. The analytic tools include AI-based, statistical, and predictive models.

However, Marroquin pointed out that these data-created insights aren’t effective if they are not in the hands of clinicians to improve patient care or in the hands of administrators to build programs. To make the data usable for clinical, algorithms need to be properly developed and trained to collect accurate information.

“The way that we build a model, the way that we train it, and the way that we do all of the derivation that the training has to mimic exactly how is it going to be utilized in when it’s deployed,” Marroquin said.

Marroquin also spoke about the importance of training algorithms on the correct target. For example, if providers are trying to determine individuals at high risk for chronic disease, AI needs to accurately identify early signs of the illness.

With these early signs identified, providers can recommend prevention strategies for the patient. Once the patient information is documented, researchers can use it to better understand social determinants of health and develop machine learning models.

“Machine learning models can recognize patterns from large amounts of data in our case — the totality of all the data points that went into the model were about 50 million for this algorithm,” Marroquin continued.

“For some things, machine learning is the best thing to do that for, for some, it could be just typical statistical or descriptive statistics. In this case, they were more interested in prioritizing the prediction, rather they infer is the understanding of causality or the why.”

UMPC used the machine learning model to learn more about the complicated interaction between covariates. By taking a less traditional medical approach, UMPC could get insight into the social determinants of health perspective they may not have yet perceived.

For artificial intelligence to assist in chronic disease prevention, broad data collection and well-trained algorithms need to be used. Algorithms can then be used in machine learning to identify individuals at high risk for chronic disease. With this information, physicians can encourage preventative health strategies and examine the social determinants of health that could factor into the illness.