September 24th, 2013
A New Model to Predict Post-PCI Bleeding
CardioExchange’s Harlan Krumholz interviews Sunil Rao about his research group’s risk model that uses an expanded definition of bleeding to predict which patients will have bleeding complications after percutaneous coronary intervention (PCI). The study is published in JACC: Cardiovascular Interventions.
Investigators analyzed clinical data from the CathPCI Registry to identify factors associated with bleeding complications within 72 hours after PCI. The data set comprised more than 1 million PCI procedures performed at more than 1100 centers from February 2008 through April 2011. The definition of bleeding included several events that previous definitions have not addressed, such as intracranial hemorrhage, tamponade, reductions in hemoglobin that account for potential hemodilution, and transfusions that account for severe anemia.
Full and simplified risk scores were validated. The full model included 31 variables; the risk score had 10 variables (see Tables 3 and 4 in the published article in JACC). Under the new definition of bleeding, the overall bleeding rate was 5.8%. The full model’s discriminatory value was consistent across a variety of prespecified subgroups and across the PCI risk spectrum.
Krumholz: Clinicians may be feeling risk-model fatigue. What should they take away from this article?
Rao: That is a fair criticism. Many published models have examined short- and long-term mortality, acute kidney injury, and bleeding. It’s important to understand the role of these risk models in contemporary PCI practice. First, we have the simple matter of estimating a patient’s risk for an adverse event so that the conversation with the patient can be more informative. Second is the issue of risk adjustment for PCI performance measures. An outstanding article published by Dr. Krumholz in Circulation in 2000 makes clear that a performance measure should reflect a meaningful outcome to patients and society — it should be valid, reliable, easily measured, modifiable through improvements in care processes, and amenable to risk adjustments. This last element is vital. Quality-improvement initiatives should avoid penalizing sites that perform PCI on sicker patients who have a higher rate of adverse events. Risk adjustment reduces the likelihood of that unfair penalty. The model in our paper is designed to accomplish both objectives: to inform discussions with patients about the risks of PCI, and to risk-adjust bleeding outcomes as applied to the CathPCI Registry cohort.
Krumholz: If clinicians want to apply your model in practice, how do you suggest they do it? The score is not simple enough for a mental calculation (and that is not meant as a criticism).
Rao: I tend to agree. An article by Gurm and colleagues, published earlier this year in JACC, makes an argument against simple risk scores, which the authors contend do not have good discriminatory value. In the era of smartphones and electronic health records (EHRs), more-complex scores are not necessarily burdensome to calculate. I think the best way to apply these in practice is to integrate them into daily workflows. EHRs need to capture the pertinent patient variables and should be programmed to calculate the risk for relevant outcomes without requiring much physician brain power. Unfortunately, we are probably not there yet, but professional societies are in the process of developing apps that can calculate these risks at the bedside.
Krumholz: How many scores should we calculate for PCI patients?
Rao: Great question. Building on my previous answers, I think patients want to know their risk for any adverse event, a portion of which we cannot quantify because some events are rare and may depend on unknown factors. That said, we can calculate some of the major risks such as bleeding, acute kidney injury, and mortality. In an ideal world, our EHR would calculate these automatically for a patient who is scheduled for catheterization and PCI.
Krumholz: Are you using these risk models in your practice?
Rao: We can certainly do better. We use them in cases that are deemed to be “complex,” but doing it routinely is the way to integrate it into practice. Like the physical examination, it should become part of the patient evaluation. Again, the role of the EHR is paramount. If robust models generate the risk for adverse events electronically, we can have a more informed conversation with the patient, which will then become part of our practice. We should aim to get to the point where not quantifying a patient’s risk is rare.
JOIN THE DISCUSSION
Are you ready for a new risk model for post-PCI bleeding? What do you consider the strengths and limitations of the model from Dr. Rao’s study group?