Predicting Adverse Drug Events with Artificial Intelligence-Driven Analytics
Have you ever read the label on the back of a bottle of medication you received from the pharmacy or that you purchased over the counter at the grocery store? You’ll quickly see that medication labels are covered with sections titled “Warnings”, “Do not use” and “Stop use.”
Such warning labels are not only listed for legal reasons. These warnings are to indicate the potential risk of overdose or allergies. In addition, drug warning labels alert a patient of potential risk if they have one of any number of listed pre-existing conditions (and sometimes to indicate risk for factors of which the patient is unaware). When a medication is contraindicated or may cause complications, patients are encouraged to seek advice from their healthcare provider. But when that consultation takes place, the healthcare provider faces a significant challenge. There are more than 600,000 unique national drug codes from U.S. Food and Drug Administration. When faced with so much information, how does a provider assess the risks when dispensing drugs to their patients? This question is especially important when getting it wrong can have dire consequences.
Adverse drug effects and adverse events caused by medication are often underestimated. Health.gov states that adverse drug events account for one-third of all inpatient adverse events and more than 3.5 million physician office visits in outpatient settings. Effectively predicting these events during or before prescribing drugs is critical and will save countless hours of inpatient stays and unexpected healthcare cost.
Data Is Telling Us Better Stories About Patients’ Risk
Data has been the underlying treasure driving the evolution of healthcare technologies for the past two decades. Data-driven analytics are boosting the efficiency and accuracy of the decision-making process at the point of care or even prior to physician-patient encounters. Significant insights can be derived from various sources such as clinical data from EMR systems as well as patients’ social determinants of health. Adverse drug events, on the other hand, which vary from headache, to nausea, to life-threatening conditions, can be found in structured or unstructured patients’ encounter data – a handwritten notes or other unstructured data sources pose a significant challenge. Having access to robust data from clinicians, clinical facilities and patients can help organizations quickly determine what story the data is telling them and them make decisions about a patient’s risk more effectively, enabling a more accurate intervention plan that will address a patient’s immediate need.
Combining Forces to Provide More Insights for Providers and Pharmacists
Artificial intelligence, or task-oriented machine learning and deep learning programs, have become a strong force to solve the puzzles that inevitably stem from healthcare data. Many factors drive healthcare outcomes for a patient, and artificial intelligence programs provide the power to analyze thousands of factors at the same time when assessing patients’ risk for a targeted health outcome.
Adverse drug events are commonly recorded as a diagnosis in claims, with most of them tagged as adverse events, which provides the potential and ground truth for concise research. With adequate instances of patients who suffer from adverse drug events after taking certain drugs, an artificial intelligence model can be built to measure the risk of adverse drug events for a patient who takes the targeted drug and provide clinical evidence supporting how it draws the conclusion. A recent study from a joint group of University of Maryland, New Jersey Institute of Technology and Inovalon has proven the effectiveness of such a solution. In this model, patient history is considered with the sequence and temporal differences of each medical event, such as medication subscription, procedures or diagnosis. Many use cases are applicable in either a clinical setting or at a healthcare analyst’s desk.
Future of Artificial Intelligence-driven Analytics
The influence of data-driven analytics grows rapidly from providers’ offices to health plans. With data flowing across the care continuum, endless potential can be expected for future healthcare analytics leveraging artificial intelligence to provide insights for healthcare professionals in almost every corner. Compared with traditional analytics, artificial intelligence programs consider thousands of factors at the same time, revealing unknown risk factors and filling risk gaps for providers. With artificial intelligence-driven analytics becoming more prevalent, it is foreseeable that a bright future is ahead of us, combining the forces of artificial intelligence and healthcare data.