By Michael Vlessides
Originally published by our sister publication Anesthesiology News
Predicting which patients may develop delirium after surgery seems to have taken a step forward, thanks to the efforts of a Swiss startup company and an international research team.
Investigators from the United States, Canada and Europe validated the artificial intelligence–based risk assessment tool—the Pre-Interventional Preventive Risk Assessment (PIPRA AG; Zurich)—which divides patients into four risk categories for suffering the adverse postoperative event. Although the tool is only licensed for use in Europe, the clinicians behind the study believe FDA approval is not far off.
“Postoperative delirium is a significant problem, especially among older patients,” said Nicolai Goettel, MD, an associate professor of anesthesiology at the University of Florida College of Medicine, in Gainesville. “However, the adverse event is not only a major perioperative problem. It’s also a healthcare problem that can have repercussions on patients’ lives well after they experience it.
“Some patients will need to go into institutions or long-term care facilities and cannot return home,” Goettel continued. “In addition, postoperative delirium has been associated with early-onset dementia in some patients. So, we developed the PIPRA tool to help address what we see as a growing problem in healthcare today.”
In the current study, Goettel and his colleagues validated the scale by analyzing individual patient data from nine observational cohort studies, all of which systematically assessed postoperative delirium. The final data set comprised 2,250 patients, 444 of whom were diagnosed with postoperative delirium and 1,806 of whom were not.
Tool Approved for Use in Europe
In a virtual presentation during the 2022 annual meeting of the International Anesthesia Research Society, Association of University Anesthesiologists and Society of Critical Care Anesthesiologists (abstract 626), Goettel reported that the final statistical model comprised nine patient variables:
- age
- body mass index
- ASA physical status
- history of delirium
- history of cognitive decline
- medication usage
- C-reactive protein level
- type of surgery
- surgical risk
These variables allow the PIPRA tool to use AI to divide patients into four subgroups according to predicted risk: either low, medium, high or very high. Given further sensitivity and specificity analyses, however, the researchers proposed three thresholds for these subgroup distinctions.
“The limit between low and medium risk is where we attain a 90% sensitivity,” Goettel explained. “The limit between high risk and very high risk is where we attain 90% specificity, while the limit between medium and high risk is where sensitivity and specificity are equal.”
To help validate the scale, Goettel and his co-investigators tested the model using a training set of data. They found that the resulting area under the curve (AUC) was 0.826 (95% CI, 0.796-0.854), with a cross-validation AUC score of 0.81. The tool was also externally validated on a data set of 293 patients: 61 with postoperative delirium and 232 without. These analyses yielded an AUC of 0.75.
As Goettel discussed, the PIPRA tool is approved for use in Europe, and the company will soon be seeking FDA approval as well. Once that hurdle has been cleared, he believes the tool will make a palpable impact on clinical practice, but not without steps to prevent delirium once its potential has been identified.
“Obviously, just predicting the risk of delirium doesn’t mean you’re preventing it from happening,” he said in an interview with Anesthesiology News. “It simply tells us that a patient is at greater risk. Eventually, we would like to bundle the tool with an intervention, preferably nonpharmacologic management strategies, which have been shown to decrease its incidence of delirium by about 30%.”
Doing so, he added, will help clinicians and institutions target their efforts toward patients who need them most. “Treating postoperative delirium is a very complex intervention, and in healthcare, complex always means costly. If these interventions were affordable, we could do it with everyone. But since they’re costly, we want to limit them to those patients who are at greatest risk.”
Ehab Farag, MD, a professor of anesthesiology at the Cleveland Clinic Lerner College of Medicine, in Cleveland, said the study is a step in the right direction, but noted that more work needs to be done to fine-tune the tool. “We all know the risk factors for delirium, but the main issue is what can we do to either prevent or treat that delirium. In both cases, options are very limited.
“In addition,” Farag continued, “the incidence of delirium could be related to the type or duration of surgery, hemodynamic changes such as blood loss and intraoperative hypotension, postoperative pain, or even ICU admissions. None of these factors are included in the authors’ calculation. Finally, preoperative frailty is an important predictor of delirium.”
Farag reported no relevant financial disclosures. Goettel reported that he is a consultant to PIPRA AG, and the company’s head of research.