While hospitals are eager to use electronic health record (EHR) data to inform drug utilization decisions, they face some significant hurdles. The root cause of most of these issues is data illegibility—most EHR implementations were myopically focused on meeting regulatory requirements and getting data recorded, with little thought given to future data analysis.
At Lumere, we experienced these issues firsthand while developing our new Pharmacy Solutions. Unlike most health systems, however, we were fortunate to have a team of data scientists dedicated to standardizing, normalizing and reducing noisy complexity within the data. Below I highlight five of the biggest challenges along with the innovative solutions we uncovered to ensure our clients have the most effective tools to tackle inappropriate drug utilization.
- Interoperability is nonexistent
The data from different EHRs, or even different instances of the same EHR, are often structurally incompatible. For example, one system may record a drug’s frequency numerically, while another requires words. At Lumere, we work around this limitation by being very selective about the data we use and relying on calculations when possible. The more data utilized, the more variability you face.
- Data are incomplete
Missing pieces in the data make it difficult to interpret how much medication a patient actually received and why. Crucial information such as frequency, dose and route of administration is often absent. We fill in these gaps by using reference data or averages from existing data. In rare cases when there is nothing to reference, we look to data from clinical trials and studies to guide us.
- Data records actions, not intentions
Electronic patient records do not explicitly link diagnoses with the drugs that were administered, making it extremely difficult to understand if drugs were used optimally. Any clinician can easily match drugs with diagnoses, but to interpret the data at scale with a computer is nearly impossible. We solve this problem with a proprietary algorithm that looks at both the patient’s diagnosis and each drug’s known uses to assign an indication. This allows us to screen out the noise of comorbidities and better understand whether or not each drug was appropriately used.
- Drugs’ units can be incompatible for comparisons
Different drugs rarely use the same dosage units, making it a struggle to accurately compare pricing and utilization volume. For example, comparing the price of a pill taken daily to an injection given monthly is especially difficult because both their units and their dosing regimen are incompatible. We overcame this by building a clinically relevant normalization system that allows for equivalence comparisons. We normalize all prices and utilization of drugs to an average patient’s treatment day, allowing for true apples-to-apples comparisons.
- Standard therapeutic classification isn’t useful for making trade-offs
Commonly used therapeutic classification of drugs is based on chemical structure or mechanism of action. These drug classes aren’t always helpful for choosing the best drug to treat a specific condition. For example, the therapeutic category of “antineoplastics” is too broad—it includes many drugs for different indications. To equip our clients to make clinically relevant comparisons, we created our own system for grouping drugs, based not only on therapeutic category, but also on indication for use. This means that financial and clinical trade-offs are readily apparent, without a lot of manual slicing and dicing of the data.
More complex EHRs and emerging technologies will continue to pose new data legibility challenges. By bringing together data science and clinical knowledge, however, we at Lumere can help health systems make the most of the immense amount of healthcare data available to uncover insight, communicate the story and ultimately improve patient care.