How Artificial Intelligence Can Improve Outcomes and Reduce the Cost of Care Among Higher Risk Patients

Predictive modeling can drive targeted, yet comprehensive, medication management to prevent adverse events and lower utilization among members with polypharmacy regimens.

More than half of Medicaid beneficiaries take five or more prescribed medications. That level of polypharmacy, as any clinician involved in caring for this higher-risk population knows, is associated with higher rates of medication errors and adverse events, as well as increased emergency department utilization and hospitalizations. 

Trying to prevent such medication-related adverse events can be costly and time-consuming, not to mention frustrating for the patient. It involves assessing and reconfiguring the patient’s medication regimen, discontinuing drugs, titrating dosages and trying different therapies in an attempt to arrive at an optimal combination that will be safe, tolerated and ultimately help the patient manage their conditions.  

A two-year study led by a researcher from Stanford University, has shown how artificial intelligence (AI) algorithms that combine patient data, claims information and highly vetted medication evidence can help clinicians practicing remotely identify the highest-risk patients in need of intervention so they can prevent adverse events. More importantly, an AI platform helped simulate medication adjustment options so an optimal regimen can be discovered and deployed sooner in a patient’s health care journey.  

By utilizing available data from the payer and other sources beyond the electronic health record (EHR), researchers and clinicians were able to significantly impact many outcomes for a large group of patients. With just five full-time pharmacists and medical assistants doing outreach, they were able to reduce the total cost of care (TCoC), drug-related adverse event risk, and care utilization as defined by multiple parameters. 

Supporting Targeted Interventions

The study, recently published in the Journal of Managed Care & Specialty Pharmacy, focused on 2150 men and women aged 40 to 60 years who were Medicaid beneficiaries and members of the Inland Empire Health Plan—one of the top 10 largest Medicaid plans and the largest not-for-profit Medicare-Medicaid plan in the country. Each study participant took 10 medications on average for their chronic conditions and an average of 25 medications total. 

Historically, determining medication risk to patients has been conducted in isolation, factoring in individual medications against patient characteristics or evaluating drugs against each other with only a minimum of information about the patient. Ideally, the patient’s complete health record, combined with health plan claims, a vetted drug database with clinical decision support tools and perhaps data from a health information exchange, would be available for risk stratification. For health plans and other payers, as well as providers, the luxury of such a complete data set rarely exists to guide interventions and optimize medication regimens.

Researchers did not have complete EHR data either, but they were able to create a comprehensive, contextual patient profile for risk stratification using an AI cloud platform (from Surveyor Health) that used validated and integrated claims, enrollment, demographics, conditions, and laboratory data from the health plan with prescription fill data from the pharmacy benefit manager.  

The AI platform also utilized highly vetted and curated drug data from FDB’s MedKnowledge database to help identify potential drug-drug, drug-disease contraindications, drug-allergy, drug side effects and other potential patient health and safety risks.

Along with the generated contextual patient profile, the AI platform used predictive modeling to simulate potential change options for each patient’s medication regimen, which relied heavily on the continually updated drug database, and delivered risk scores based on potential changes. In effect, clinicians were able to determine the safety and efficacy of prospective regimen changes before they took effect, which saved time, improved outcomes, and improved patient experience.  

Clinicians were able to eliminate one or more medications due to duplicative therapies, which makes managing regimens easier, and in some cases, less costly for patients. 

Automated, But Personalized

Another way this comprehensive medication management approach improved patient experience was through personalization. Although numerous aspects of the clinical interventions were automated thanks to the AI platform, the specificity of the data, decision support and interaction with the patients, which included a one-on-one consultation with a pharmacist over the telephone, was highly relevant and focused on the individual patient’s health status, needs and goals.  

In addition, communication materials were automatically generated to ensure consistency and save time. Patients received a customized patient care plan and an updated medication list. A consultation summary with recommendations was also faxed to their primary care provider (PCP).  

Researchers determined that despite the comprehensive nature of the medication review and outreach, each intervention consumed only 30 minutes of the pharmacist’s time on average, roughly half the time the same clinical team required for medication reviews without the AI platform. This enabled the small team to double their productivity while increasing their focus on condition management. On average, each patient received 3.5 interventions within 14 months, ranging from one to nine interventions. 

Significant Risk and Cost Reductions

These targeted polypharmacy medication assessments and interventions, which took place over 14 months between 2018 and 2019, reduced the total cost of care by 19.3%, or $554 per member per month on average, according to the study results. This translates to approximately $1.2 million a month and over $14 million annually for the 2150 patients in the study.    

Cost savings were attributed to numerous factors, including medication spend, which was reduced by 17.4% or $192 per member per month. The vast majority of medication-related regimen changes (84.6%) included discontinuations, many of which were therapy overlaps and duplicate medications. As such, researchers noted a 13.1% reduction in duplicative therapies and a 15.2% reduction in serious drug-drug interactions.  

Further cost savings were experienced from a 15.1% reduction in emergency department visits, a 9.4% decrease in hospital admissions, and a 10.2% reduction in length-of-stay when comparing the same population’s utilization before the AI-powered remote interventions. 

Researchers estimate that if similar data-driven analysis and interventions occurred with patients with similar risks, the health plan could save $109 million annually and save California’s Medicaid program $1.2 billion a year. It was calculated that the return on investment for the program was 12.4:1 based on TCoC savings and program costs. 

Safer Polypharmacy

The aging Baby Boomer population along with the growth of chronic conditions among patients of all ages means high levels of polypharmacy will likely increase, placing a greater burden on healthcare provider organizations to review medication regimens, identify potential health and safety risks and intervene.

Along with the growth of prescriptions, the expansion and improvement of sophisticated AI algorithms and drug databases mean that comprehensive medication reviews can be significantly accelerated, improved and beneficial for patients and clinicians.  

As the study demonstrates, not only can these technologies help identify the neediest patients faster and more efficiently, but they can also offer insights into safe and effective regimens using predictive modeling that was not feasible a generation ago.  

More relevant and reliable medication insights mean clinicians can prevent drug-related adverse events to reduce unnecessary care utilization, but also better engage patients in their care and improve their outcomes. 

Disclaimer: The views and opinions expressed are those of the author(s) and do not necessarily reflect the official policy or position of the Population Health Learning Network or HMP Global, their employees, and affiliates. Any content provided by our bloggers or authors are of their opinion and are not intended to malign any religion, ethnic group, club, association, organization, company, individual, or anyone or anything.