Practitioner AI Use Cases in Healthcare
Predictive analytics is the analysis of data to be able to foresee future trends and events. In healthcare, predictive analytics has played a pivotal role in the prevention of chronic diseases and illnesses by allowing practitioners to detect early-warning signs and intervene when necessary. This approach leverages sophisticated algorithms and vast amounts of data to predict potential health issues before they become severe, ultimately improving patient outcomes and reducing healthcare costs.
Patient Records: These documents contain a comprehensive overview of an individual's patient histories, plans of treatment, diagnosis codes, and other crucial information and can pull information from all sources. EHRs integrate information from different healthcare providers, ensuring a holistic view of the patient's health history, which is essential for accurate predictions. By aggregating data from various sources, EHRs help in building a complete picture of patient health over time.
Example: One famous such instance of patient records was lead by a team at Johns Hopkins to detect high probability instances of sepsis, which was ultimately implemented within the emergency room [1]. This tool pulled data directly from the electronic medical record to determine confidence of sepsis risk and alert a staff member accordingly. This staff member would then administer antibiotics accordingly, saving the patients life.
Vitals & Lab Results: ECGs, weight, height, blood tests, urine tests, and other lab examination results provide crucial details for evaluating a patient’s state of health. Consistent monitoring of lab results over time can reveal trends that signal the onset of diseases, allowing for preemptive measures. These results can be continuously tracked to monitor the effectiveness of treatments and adjust them as needed.
Example: Echocardiograms offer time series channels (usually 12) that evaluate the electrical signals within the cardiovascular system of a patient. Multiple datasets have been growing to support training of these systems across different ECG setups and multiple conditions. Convolutional neural networks are uniquely suited to handle this data as evidenced by the team at Mayo Clinic in Rochester Minnesota [2].
Imaging Data: MRIs, X-rays, and CT scans offer valuable information about a patient’s anatomy, enabling machine learning techniques to analyze these images and provide data-driven insights. Advanced imaging analytics can detect patterns and anomalies that may not be visible to the human eye, leading to earlier diagnosis and intervention. These insights can help in the early detection of conditions such as tumors or other abnormalities that might otherwise go unnoticed.
Example: Given the accuracy of neural networks with imaging / matrix data, it is no secret that convolutional neural nets are particularly of value here. However the devil is in the details. Simply detecting a tumor is one thing - often describing why a system believes one exists is another. Frequently AI papers cite difficulty in machine image variance, or if an object such as a ruler is detected in the image being analyzed. A team from Germany led by Franziska Mathis-Ullrich published a paper in 2022 describing a constructed system that would visually cue the practitioner [3]. In medical AI applications explainability is huge.
Practitioner Notes: Practitioners are required to record patient diagnosis, risks, medications, and all sorts of other info during a visit or file review. This can be particular tedious on top of medical billing. Given that this takes so much time, many tools have evolved to assist with it, such as speech-to-text, summarization, and general guidance on risk vectors for a patient.
Example: A more nuanced use case of this is actually recording and documenting conversations as they happen - a technology that is seeing increasing performance called diarization shown below. Tenasol constructed this for a federal client seeking derive insights from medical discussions. Rather than just performing speech-to-text, diarization also seeks to assign the text to one of the speakers in the room.
Social Determinants of Health (SDoH): The lifestyle of a patient, including information such as their income, education, ethnicity, and other non-clinical factors, provides useful data points that can influence providers' actions. By understanding social risk factors, providers can offer personalized recommendations to prevent diseases linked to certain behaviors. This data helps in creating targeted wellness programs and interventions tailored to individual patient needs.
Example: Geographic Information Systems (GIS) offer the ability to triangulate problematic areas on an overall or per-capita basis. The below visualization was created using the public COVID-19 data by Tenasol in a demo for the federal government. Insights derived by these methods can drive actions and policy in the same way a weather forecast informs you what you should be wearing for the day. In this example, we see reported per-capita covid mortality in Baltimore was highest in the lower income areas surrounding the city shown in a detailed fashion, despite the fact that there are only 3 counties reporting individual counts. This was constructed using triangulation and AI methodologies for more localized regions based on demographic data.
The Future of Predictive Analytics
Predictive analytics is already providing numerous benefits to healthcare providers, but in the future, the use cases of predictive analytics are set to explode with continuous advancements in machine learning and real-time data collection. For instance, wearable devices will become even more valuable as their ability to monitor real-time health improves, allowing timely and personalized care. These devices will continuously track vital signs and other health metrics, providing immediate feedback and alerting both patients and providers to potential health issues. In addition, utilizing the genetic information of an individual will contribute to more accurate predictions and enable more tailored treatment plans based on a person’s genes.
In the AI realm, deep learning algorithms used in medical imaging tasks and natural language processing (NLP) for clinical notes will continue to evolve and greatly improve. These advancements will help doctors in interpreting complex medical images and diagnosing diseases and other illnesses at earlier stages with increased accuracy and precision. Deep learning models will be trained on extensive datasets, enhancing their ability to recognize patterns and anomalies indicative of specific conditions. As for NLP, improvements will allow for the collection of many more key insights from unstructured clinical notes, facilitating a more comprehensive understanding of a patient’s health and allowing doctors to make better-educated medical suggestions for them.
If would like more information on how Tenasol might be able to help, reach out to us in the Contact section of this website.
References
Henry, Katharine E., et al. "A targeted real-time early warning score (TREWScore) for septic shock." Science translational medicine 7.299 (2015): 299ra122-299ra122.
Siontis, Konstantinos C., et al. "Artificial intelligence-enhanced electrocardiography in cardiovascular disease management." Nature Reviews Cardiology 18.7 (2021): 465-478.
Zeineldin, Ramy A., et al. "Explainability of deep neural networks for MRI analysis of brain tumors." International journal of computer assisted radiology and surgery 17.9 (2022): 1673-1683.