Updated ACTION-GWTG Risk Score Aims to Better Predict In-Hospital Mortality After Acute MI

Developing accurate risk models to predict in-hospital mortality after MI has long been hampered by rapidly changing practice and unmeasurable confounders. Now, an updated version of the ACTION Registry-Get With the Guidelines (GWTG) risk score is drawing on a much broader patient group undergoing modern AMI care.

The main reason for updating the model was to be able to include previously unavailable information on patients with cardiac arrest, which is highly predictive of mortality, lead author Robert McNamara, MD (Yale University School of Medicine, New Haven, CT), told TCTMD.

“In general, all risk models can be improved upon with more contemporary data,” McNamara added, noting that the number of included hospitals more than doubled in this version compared with its predecessor, “so it would seem to be more generalizable.”

To update their score, McNamara and colleagues looked at 243,440 acute MI patients who were treated at 655 ACTION Registry-GWTG hospitals between 2012 and 2013. Sixty percent of the population was used to derive the model, while the remaining 40% was used to validate it.

Details about the tool were published online yesterday ahead of print in the August 9, 2016, issue of the Journal of the American College of Cardiology.

Overall in-hospital mortality was reported in 4.6% of patients, and the following risk factors were found to be associated with a higher risk of death in both the derivation and validation cohorts:

 

• Older age
• Higher heart rate
• Lower systolic blood pressure
• STEMI on ECG
• Heart failure
• Cardiogenic shock
• Cardiac arrest
• Decreased creatinine clearance
• Higher troponin ratio

 

Using this information, the researchers built a risk model with high discrimination in both populations (C statistic = 0.88 for both) and “excellent” calibration, they write. Observed mortality rates ranged from 0.4% in the lowest risk group (score < 30) to 49.5% in the highest risk group (score > 59). Also, it performed well when tested in various subgroups including STEMI, NSTEMI, age, sex, race, transfer status, diabetes, renal dysfunction, cardiac arrest, and cardiogenic shock.

An Improvement but Not Definitive

“The model is really an improvement on certainly their prior model from the ACTION Registry, and really perhaps in many ways, the most contemporary and best available model for predicting mortality after myocardial infarction,” Robert W. Yeh, MD, MBA (Beth Israel Deaconess Medical Center, Boston, MA), who was not involved in developing the risk score, told TCTMD.

But its usefulness in clinical practice is unclear, he said. “I would use it mostly as a guide to talk to the team about how sick a patient is, maybe to talk to the patients to make sure that they’re fully informed about the gravity of [their] illness,” Yeh predicted, adding that there are too many “unmeasurables” like frailty for this model to reliably change his decision-making.

While obviously meant for different purposes, he compared this score with the DAPT Score, which has served as an aid for physicians to best determine how long to prescribe dual antiplatelet therapy after stent implantation. “That score was built definitively to help clinicians make a decision,” Yeh said, acknowledging that he himself was a contributor. “So if you have a high DAPT score, it’s supposed to serve as some evidence in conjunction with clinical judgement that perhaps extending dual antiplatelet therapy is better.” For mortality on the other hand, there remains no clear clinical decision to be informed based on this or any risk score, he added.

In the paper, McNamara and colleagues write that their model “should enable improved assessment of hospital quality and enhance research into best practices to further reduce mortality in patients with [acute MI].”

McNamara acknowledged that the risk model was primarily designed so that administrators could “have an even playing field” for comparing hospitals, but said it still has clinical usefulness. Developing a fully functional clinical risk score would require a prospective trial, which would be difficult, he added.

What might be a more valuable type of score, Yeh suggested, is one that, for example, could indicate whether a patient would have a better chance of surviving with early invasive treatment. “That’s the kind of risk model that would actually change clinical decision-making,” he commented.

But McNamara said much more than simply in-hospital mortality risk would need to go into a score like that—“a patient’s ejection fraction, ideally patient-reported outcomes, as well as more longitudinal data,” he suggested.

In the near future, McNamara said his team expects to publish an in-hospital bleeding risk score that “may have a little bit better capability for people to identify patients that are high-risk for bleeding, and to use different medications or different strategies to decrease the risk of that bleeding.”

No Replacing Judgement

In an accompanying editorial, Peter Wilson, MD (Atlanta VAMC and Emory Clinical Cardiovascular Research Institute, Atlanta, GA), and Ralph D’Agostino Sr, PhD (Boston University, MA), write that this risk model has “demonstrated the dynamic nature of health risk appraisals.” While clinicians have “extensive information” regarding ACS and MI patients, all of it can be “assessed and aid in developing risk models that may provide usefulness to clinicians concerning medications and interventions,” they add.

However, Yeh said that balancing data with “the gestalt feel for the level of illness of the patient” is really the “art of medicine.” How that all is classified “is a really important question for registries and not yet I think solved,” he continued. “To be fair to risk scores, I think we do know that risk scores are better prognosticators of things like mortality in critical illness than physician judgement most of the time.”

How to balance data and intuition is a “key question for anything we do,” McNamara agreed. “Certainly we have moved much more toward evidence-based medicine over last 20 to 30 years. We want to provide that evidence so that becomes more part of the equation. But overall, we’re still dependent on clinical judgment in any situation. . . . It’s difficult to make too many individual decisions based on a score.”

For now, harmonizing the information offered by risk scores with clinical judgement is “really the best we can do,” Yeh concluded.

 

---

 

 

 

Related Stories:

 

• Public Reporting Linked to Less PCI, Higher Mortality in Acute MI Patients
• Short-term Mortality After Acute MI Lower at VA Hospitals Than Elsewhere, but Readmissions Higher
• Falling Mortality for In-Hospital Cardiogenic Shock in Acute MI Still Leaves Room for Improvement