Disease Propensity

Pain/Problem

Physicians and other healthcare providers often have challenging amounts of information to digest from disparate sources. This challenge can be a liability when making urgent decisions or identifying patients who are at risk, wasting precious time, delaying diagnosis, and—potentially—resulting in poor outcomes. 

Diagnosis is a multi-faceted process during which healthcare providers analyze patients’ vitals and symptoms to identify their disease, condition, or injury. Patients have the greatest chance of recovery when their problem is diagnosed correctly. 

Misdiagnosis can lead to serious health issues and result in patients undergoing unnecessary and expensive treatments—while their undiagnosed condition goes untreated. Studies reviewed by the Agency for Healthcare Research and Quality suggest that thousands of hospitalized patients die every year due to diagnostic errors. 

In an article about serious misdiagnosis-related harms published in the journal Diagnosis, around 1 in 10 patients with a “Big Three” disease is misdiagnosed, and approximately 50% of those misdiagnosed die or are permanently disabled as a result.

Identifying patients with a higher propensity of suffering from or developing a particular disease or condition and determining the appropriate treatment is key to managing both disease and costs.

Solution 

AI is an ideal way to help physicians and other healthcare providers supplement their diagnostic process by predicting a propensity to disease. Analyzing large amounts of patient data from multiple sources can deliver statistical probabilities of specific diagnoses and health outcomes. This capability can help providers avoid misdiagnosis—and focus on the root cause of a disease rather than its symptoms. 

Payers who are able to predict certain conditions in advance and intervene are more likely to save costs by preventing patients’ health from deteriorating. People are healthier; insurers (payers) are wealthier. For example, it’s far better to predict cancer and pay for preventative treatment than pay for chemotherapy, MRIs, and other procedures down the road. 

Automated machine learning models built using DataRobot can extract deep insights from demographic and historical patient data. This enables physicians and other medical providers to identify at-risk patients using results that can be easily explained to colleagues and other stakeholders. Providers can be alerted that a patient shows early indicators that suggest a specific diagnosis. 

AI can be used to augment clinical decision making, to help predict the likelihood of any particular disease, as long as historically labeled data for that disease exists. Top risk drivers—for example, a patient’s symptoms, past diagnosis, personal background, and admission history—can be highlighted so providers can create customized treatments. The AI can be integrated into the clinical workflow within an Electronic Health Record, or decision-support intelligence can be provided through intelligent apps and dashboards. 

Predictions made by DataRobot’s models are intended to supplement diagnoses made by physicians and other healthcare providers—and are in no way meant to replace their medical expertise. The intention is to bring in patients who are high risk or patients who might potentially be overlooked to the providers’ attention for possible follow-up.

Why DataRobot

DataRobot AI Platform is designed to unify data of all types, from all sources, into a single system of record across an organization. It delivers rapid results, with vital transparency, security, and built-in governance. 

DataRobot uses AI to enhance disease propensity modeling, with the goal of identifying patients who are at risk, improving outcomes, and decreasing the overall cost of care. By understanding the specific pathways a patient’s disease falls into and how severely it might develop over time, physicians and other medical providers can leverage simulations that reveal the effectiveness of different treatments for heading off severe outcomes. These predictions can be generated early in the process and updated each time new information comes in, helping providers customize treatment to avoid adverse reactions and increase the likelihood of a patient’s recovery.