Heightened opioid risk predictable by algorithm prior to back surgery
Harvard researchers have used machine learning to predict which patients soon to undergo an operation for low-back pain should be monitored for opioid dependence down the road.
They further found their algorithms worked best when paired with patient-specific explanations for the automated predictions, such as a likely need for a long-term painkiller prescription combined with a prior diagnosis of depression.
In introducing their work, published online June 9 in The Spine Journal, Joseph Schwab, MD, and colleagues point out that spine surgery is a known risk factor for prolonged use of opioids.
“Preoperative prediction of opioid use could improve risk stratification, shared decision-making and patient counseling before surgery,” they write, underscoring that the main goal of their study was to develop predictive algorithms.
To begin, the team reviewed the charts of more than 5,400 patients who had surgery to relieve pain from herniated discs of the low back between 2000 and 2018.
Looking for patients who had opioid prescriptions of at least 90 to 180 days post-surgery, they found 416 patients (7.7%) fell into this target category.
From this data Schwab and team developed five algorithms as well as explanatory data models that applied both globally, meaning across all patients, and locally (to individual patients).
The best of their five AI tools used elastic-net penalized logistic regression. This model had good overall performance and identified various factors as the most important predictors of long-term opioid use.
The key predictors included opioid prescription prior to surgery and pre-existing depression at the time of the operation.
The authors noted a number of limitations to their study design. These included an assumption that the drugs were taken as prescribed and a lack of data on opioid dose in oral morphine equivalents in outdated electronic health records.
Still, they pointed out, the study has much value to offer clinicians caring for patients who have had, or will have, surgery for herniated discs.
“Preoperative prediction of increased risk of prolonged postoperative opioid prescription can result in management changes that provide more support and counseling to patients prior to surgery,” Schwab et al. write. “The decision curve analysis presented in this study clearly demonstrated that our models offer greater value than management decisions based solely on duration of preoperative opioid prescription alone.”
Patient-specific explanations can shed additional light on which patients are candidates for medical surveillance post-surgery and on what interventions can be introduced early enough to help head off dependence or addiction, the authors suggest.
“The ultimate use of these models will be subject to external validation,” they conclude, “but the implications for practice include a patient-centered approach to postoperative management and increased leverage of techniques such as machine learning to mitigate adverse events associated with prolonged opioid use following surgery.”
Schwab and colleagues have incorporated their final models into an open-access web application. It’s available for use or review here.
Dave P. has worked in journalism, marketing and public relations for more than 30 years, frequently concentrating on hospitals, healthcare technology and Catholic communications. He has also specialized in fundraising communications, ghostwriting for CEOs of local, national and global charities, nonprofits and foundations.