Max J. Lewis, 28 years old, Columbia, MD – Our beloved son and brother, Max, was well loved by almost everyone he met. Quick with a smile, a hug and a joke, and a heart as big as Texas, he will be missed… [change name, repeat] … By this point, almost everyone has witnessed it… an opioid-related deadly overdose among someone’s family member or friend in our Facebook news feed, and constant news in the paper about the mounting deaths from drugs like Fentanyl, heroin and OxyContin. Last year, an estimated 60,000 Americans died from opioid overdoses, more than the number of Americans that died in the Vietnam War. Opioid addiction does not discriminate by age, race, gender or socioeconomic status, although rural America has been especially hard hit.
What is driving this epidemic one might ask? Like any massive problem, there is not one cause, but a variety of factors have led to this situation. Experts have attributed blame to causes such as increased marketing by pharmaceutical companies, easy access to pills, socioeconomic issues in blighted communities causing folks to seek drugs to assuage pain both physical and emotional, and years of the medical community increasing focus on “pain as the fifth vital sign”. Pain is a common ailment, with some estimates finding that over 30% of Americans have one form of acute or chronic pain or another.
While opioids can be an effective tool to ease acute pain, they can also be a lethal salve, and their effectiveness for long-term pain management is feeble.
Clinical Decision Support Systems for Better Prescriptions
The CDC guidelines for prescribing opioids for chronic pain were updated in 2016 due in part to the greater acknowledgment of the dangers of opioid abuse. Guidelines provide clinicians a succinct description of the best scientific knowledge to date within a given practice area. Guidelines may help decipher: when to initiate or continue opioids for chronic pain; opioid selection, dosage, duration, follow-up, and discontinuation; and, assessing risk and addressing harms of opioid use. However, published guidelines can be notoriously slow to translate into clinical practice, inconsistently implemented, and difficult to align with existing workflows and customary practices. This presents an adoption challenge; one in which clinical decision support systems (CDSS) may help overcome.
The use of automated, well-designed CDSS has shown promise in increasing adherence to guidelines and influencing prescriber behaviors by illuminating meaningful patient information that may otherwise be hidden in digital records or paper charts, and left unanalyzed by the prescriber. The widespread adoption of electronic health records (EHRs) over the last decade is enabling ever more collection of data digitally, which can be analyzed by a CDSS to give clinical professionals more robust and rapid insights into their patients clinical and behavioral states over time.
Accordingly, in our current research project sponsored by the National Institute for Health Care Management, we are seeking to develop decision support for pain management that can help predict a patient’s risk of opioid dependency. An indication of a high risk of opioid dependency could assist physicians in carefully considering other treatment options for the patients most likely to follow undesirable treatment courses.
Early Insights from CHIDS
Using a large dataset provided by the Defense Health Agency, our Smith School CHIDS team has been working on developing data-driven prediction and risk scoring of opioid misuse that may drive CDSS algorithms. Our data analytic approaches applya range of machine learning methods such as logistic regression, adaptive boosting, LASSO, and classification trees. We have also implemented a variety of sampling methods to improve the performance of the data mining algorithms, experimenting with oversampling, undersampling and artificial data synthesis. We have also tested sparse linear models, which can be useful as a way to provide a simple point scoring mechanism for risk, thereby providing a benefit of being more easily interpretable for physicians than some other types of statistical data.
The early results show promise in helping predict chronic opioid dependency, and we hope to bring the methods out of the lab and into practice soon. We will be making the tools freely available, and are also seeking partners and sites to test the CDSS tool being developed for evaluation, refinement and applied use. The first results will be released in early 2018.
We hope to play a small part in helping empower healthcare professionals with deeper insights through data science applications. And, through these additional insights, influence opioid treatment alternatives, when appropriate.
Users who have LIKED this post: