Author + information
- aDepartment of Cardiology, Geisinger Medical Center, Danville, Pennsylvania
- bDepartment of Biomedical and Translational Informatics, Geisinger Health System, Danville, Pennsylvania
- ↵∗Address for correspondence:
Dr. James C. Blankenship, Department of Cardiology 27-75, 100 N Academy Avenue, Geisinger Medical Center, Danville, Pennsylvania 17822.
In this issue of the Journal, Qintar et al. (1) report a personalized risk model to assess the likelihood of major adverse cardiovascular events (MACE) in diabetic patients with multivessel coronary artery disease enrolled in the FREEDOM (Future Revascularization Evaluation in Patients with Diabetes Mellitus: Optimal Management of Multivessel Disease) trial. Overall, results of the FREEDOM trial suggested that coronary artery bypass graft (CABG) surgery is superior to percutaneous coronary intervention (PCI) in these patients. However, the personalized risk model developed by Qintar et al. (1) suggests that 45% of FREEDOM patients would have similar 5-year MACE rates if they underwent CABG or PCI, and 35% would have similar likelihood of relief of angina at 1 year.
At first blush this might appear to be a transparent attempt by interventionalists to justify recommending PCI to diabetic patients with multivessel coronary disease, thumbing their noses at randomized controlled trials, cohort studies, meta-analyses, and editorials that uniformly support CABG in this patient population. Is the personalized risk model a legitimate tool to use in advising patients about revascularization options, or is it just a ploy that interventionalists will use to justify their oculostenotic reflexes?
Is the Personalized Risk Model a Useful Tool?
Perhaps. Personalized risk models can identify patients with the most to lose or gain from a particular treatment and are at the forefront of the move to personalized health care. These models are increasingly reported in the cardiovascular data literature, with over 800 published in the past 2 decades at a rate that is doubling each decade. Perhaps the most widely used clinical prediction tool is the CHA2DS2-VASc (Congestive heart failure, Hypertension, Age 75 years, Diabetes, Stroke/transient ischemic attack, Vascular disease, Age 65 to 74 years, Sex category) score to predict risk of stroke in patients with atrial fibrillation. Personalized risk models have been used in interventional cardiology to predict mortality (2), stent thrombosis (3), contrast-induced nephropathy (4), risk of dual antiplatelet therapy (5), and bleeding (6). Implementation of the bleeding risk model reduced bleeding events by 44% in 1 study (7). However, the utility of personalized risk models may depend on physician acceptance, which to date has been limited (8).
Is This Personalized Risk Model a Useful Tool?
Probably. A useful risk model has 3 requirements (9). It must identify differences in risk among individuals more reliably than can be done with clinical judgment. It should be implementable—able to be integrated into patient care workflow without disruption. Finally, it should produce actionable results—that is, it should be able to guide clinical decision-making (9). The personalized risk model reported by Qintar et al. (1) was not assessed for the first requirement, superiority over clinical judgment, although its predictive capability was proven by application to an external dataset. The second requirement, easy integration into clinical practice, awaits testing of the digital application promised by the authors. The third requirement, that it can be used to guide decision making, is met; the tool can be used to advise individual patients about their risk of MACE with PCI or CABG.
But, the clinician must consider how much the personalized risk model adds to clinical judgment. The quality of a personalized risk model is assessed by its C-statistic, a metric unfamiliar to many interventional cardiologists. Simply put, given 2 individuals (one who experiences an outcome of interest, in this case MACE, and the other who does not), the C-statistic is the probability that the model will predict higher risk for the patient with the outcome than for the patient who does not experience the outcome (10). It is a measure of concordance (hence, the name “C-statistic”) between model-based risk estimates and observed events. A C-statistic of 0.5 indicates that the model has a 50% chance, the same as a random coin flip, of correctly predicting risk in a pair of individuals. A C-statistic of 0.95 indicates a very high-quality model. The model described in this study had C-statistics in the 0.65 to 0.70 range, similar to those reported for the CHA2DS2-VASc score. In the context of this study, a C-statistic of 0.66 would indicate that if 1 individual were picked from the group who experienced a MACE and 1 were picked from the group without a MACE, the model would correctly assign higher risk to the individual with the MACE 2 out of 3 times. Although that level of accuracy is not ideal, it is likely to help in conversations with patients about risk.
Do Diabetic Multivessel Coronary Disease Patients as a Whole Do Better With CABG Than PCI?
Yes. This has been consistently observed in randomized controlled trials, real-world cohorts, and meta-analyses. The advantage of CABG has not changed over the past 25 years (11) and is maintained at least 7 years after the revascularization procedure (12). So, it is safe to conclude that in general, diabetic multivessel coronary disease patients have better outcomes, specifically less MACE, with CABG than with PCI. However, many diabetic multivessel coronary disease patients still undergo PCI (13).
Is it Plausible That Despite the Overall Results of FREEDOM, Some Patients in FREEDOM Would Do Equally Well With Either Revascularization Strategy?
Yes. Subgroup comparisons of FREEDOM patients have shown that CABG is superior to PCI regardless of insulin requirement (any vs. none) (14) and chronic kidney disease (present vs. absent) (15). Other studies have concluded that CABG was superior to PCI regardless of presentation (acute coronary syndrome vs. stable ischemic heart disease) (13), age, sex, ejection fraction, chronic kidney disease, lung disease, or peripheral vascular disease (16). However, in the SYNTAX trials, low-intermediate SYNTAX score patients (73% of all patients) treated with PCI or CABG had similar MACE rates, whereas high SYNTAX score patients (27% of all patients) did better with CABG (16). These previous studies do not exclude the possibility that some FREEDOM patients might do equally well with PCI or CABG. Furthermore, quality of life was similar for PCI and CABG patients in FREEDOM (17).
Can the Personalized Risk Model Identify Patients Who Will Fare Similarly With PCI and CABG?
Yes, but not with perfect accuracy. Qintar et al. (1) developed multivariable models using only variables available at the time when a decision between PCI and CABG is made. The investigators used the model to calculate each patient’s individualized predicted probability of MACE and angina twice, first assuming treatment with multivessel PCI and second assuming treatment with CABG. They then assessed whether the personalized risk prediction for CABG was higher or lower than the personalized risk prediction for PCI for each patient. For MACE, smoking history interacted strongly with revascularization type such that all patients with a history of smoking were expected to have a lower risk of MACE with CABG compared to PCI. Conversely, all patients without a smoking history were predicted to have similar MACE with CABG versus PCI. Similarly, for angina relief, SYNTAX score interacted strongly with revascularization type such that all patients with SYNTAX score >22 were expected to have better angina relief with CABG. Conversely, all patients with SYNTAX score ≤22 were predicted to have similar relief with CABG compared with PCI.
Many patients prefer PCI over CABG, even when told the outcomes of PCI are worse than the outcomes of CABG (18). This is likely due to preference for the convenient and less painful course of PCI, which, to the patient at the time of decision making, might seem more important than a small but statistically significant future differences in outcomes. Every interventionalist has had the experience of counseling a patient that they are likely to live longer with CABG, only to have the patient request PCI. When counseling such patients, clinicians can use the personalized risk model of Qintar et al. (1) to provide estimates of risk using numbers.
Quintar et al. (1) provide examples of 2 hypothetical patients. For one, the personalized risk model predicts similar risks of angina and MACE. This patient can be counseled that other factors can be used to decide revascularization strategy. The other patient has a 38% risk of MACE with PCI and a 25% risk of MACE with CABG, which might persuade the reluctant patient to undergo CABG.
It is the clinician’s responsibility to present to the patient objective data regarding the merits of different approaches as well as clinical judgment. Use of the personalized risk model by Qintar et al. (1) may help clinicians provide more objective recommendations and may help patients make more rational decisions regarding revascularization strategies. The work of Qintar et al. (1) may provide a path to bring clinical practice more in line with evidence derived from clinical trials.
The authors are grateful for thoughtful review and critique from Greg Yost, DO, Department of Cardiology, Geisinger Medical Center.
↵∗ Editorials published in the Journal of the American College of Cardiology reflect the views of the authors and do not necessarily represent the views of JACC or the American College of Cardiology.
Both authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- 2019 American College of Cardiology Foundation
- Qintar M.,
- Humphries K.H.,
- Park J.E.,
- et al.
- Brennan J.M.,
- Curtis J.P.,
- Dai D.,
- et al.
- Tsai T.T.,
- Patel U.D.,
- Chang T.I.,
- et al.
- Mehta S.K.,
- Frutkin A.D.,
- Lindsey J.B.,
- et al.
- Spertus J.A.,
- Decker C.,
- Gialde E.,
- et al.
- Yeh R.W.,
- Kramer D.B.
- Herbison P.,
- Wong C.K.
- Farkouh M.E.,
- Domanski M.,
- Dangas G.D.,
- et al.
- Ramanathan K.,
- Abel J.G.,
- Park J.E.,
- et al.
- Dangas G.D.,
- Farkouh M.E.,
- Sleeper L.A.,
- et al.
- Cavalcante R.,
- Sotomi Y.,
- Mancone M.,
- et al.
- Corresponding Author
- Is the Personalized Risk Model a Useful Tool?
- Is This Personalized Risk Model a Useful Tool?
- Do Diabetic Multivessel Coronary Disease Patients as a Whole Do Better With CABG Than PCI?
- Is it Plausible That Despite the Overall Results of FREEDOM, Some Patients in FREEDOM Would Do Equally Well With Either Revascularization Strategy?
- Can the Personalized Risk Model Identify Patients Who Will Fare Similarly With PCI and CABG?