Abstract
Background:
In patients with diabetes and multivessel coronary artery disease (CAD), FREEDOM demonstrated that, on average, CABG was superior to PCI for major acute cardiovascular events (MACE) and angina reduction. Nonetheless, multi-vessel PCI remains a common revascularization strategy in real world.
Objectives:
To translate the results of FREEDOM to individual patients in clinical practice, risk models of the heterogeneity of treatment benefit are built.
Methods:
Using patient-level data from 1900 FREEDOM patients, we developed models to predict 5-year MACE (all-cause mortality, non-fatal MI and non-fatal stroke) and 1-year angina after CABG and PCI using baseline covariates and treatment interactions. Parsimonious models were created to support clinical use. The models were internally validated using bootstrap resampling and the MACE model was externally validated in a large real-world registry.
Results:
5-year MACE occurred in 346 (18.2%) patients, and 310 (16.3%) had angina at 1 year. The MACE model included 8 variables and treatment interactions with smoking status (c=0.67). External validation in stable CAD (c=0.65) and ACS (c=0.68) demonstrated comparable performance. The 6-variable angina model included a treatment interaction with SYNTAX score (c=0.67). PCI was never superior to CABG and CABG was superior to PCI for MACE in 54.5% of patients and in 100% of patients with history of smoking.
Conclusions:
To help disseminate the results of FREEDOM, we created a personalized risk prediction tool for patients with diabetes and multi-vessel CAD that could be used in shared decision making for CABG vs PCI by estimating each patient’s personal outcomes with both treatments.
Keywords: coronary artery disease, diabetes, multivessel disease, personalized risk estimate, percutaneous coronary intervention, coronary artery bypass graft, shared decision making
Condensed abstract:
Patient-level data from the FREEDOM trial were used to create models to predict 5-year MACE and 1-year angina after CABG and PCI. The models were internally and externally validated. PCI was never superior to CABG and CABG was superior to PCI for MACE in 54.5% of all patients and in 100% of patients with history of smoking. Thus, the newly developed personalized risk prediction tools for patients with diabetes and multi-vessel CAD could be used in engaging patients and physicians in shared decision making for CABG vs PCI by estimating patient’s personal outcomes with both treatments.
Based on a number of clinical trials of patients with diabetes and multi-vessel coronary artery disease (CAD)(1), guidelines recommend treatment with coronary artery bypass graft surgery (CABG) over multivessel percutaneous coronary intervention (PCI)(2). As PCI techniques have improved with fewer procedural complications and less restenosis, this recommendation has continued to be challenged. Most recently, however, the Future Revascularization Evaluation in Patients with Diabetes Mellitus: Optimal Management of Multivessel Disease (FREEDOM) trial(3) again demonstrated superiority of CABG over PCI among these patients for both long-term major adverse cardiovascular events (MACE) and patients’health status; their symptoms, function and quality of life(4). Despite these results, and continued guideline recommendations, multivessel PCI has remained a common revascularization strategy in patients with diabetes and multi-vessel disease.(5,6)
Barriers to the adoption of this guideline may include patients’ preference for a less invasive approach(7,8), lack of appropriate shared-decision making, and over-estimation of surgical risk by physicians.(9) Moreover, treatment options are usually presented by the treating physician, who may be biased in over- or under-estimating the benefits and harms of a particular revascularization strategy (PCI and/or CABG), depending on their specialty, experience and background. A more evidence-based infrastructure is therefore needed so that the risks and benefits of each approach could be transparently shared and used as a foundation for shared decision-making. As such, we leveraged the FREEDOM trial data to develop prediction models for both long-term MACE and angina, after both CABG and PCI, among patients with diabetes and multi-vessel CAD and sought to validate the former of these in a contemporary observational registry.
METHODS
The FREEDOM Trial.
The FREEDOM trial(3) was an international, multicenter, randomized clinical trial that compared multi-vessel PCI to CABG in patients with diabetes and multi-vessel CAD who were on optimal medical therapy. The design, protocol, and methods of the FREEDOM trial(10) and the clinical(3) and health status(4) outcomes have been previously published. Briefly, between April 2005 and April 2010, adult patients with diabetes and angiographically confirmed multivessel CAD (70% stenosis in two or more major epicardial vessels involving at least two separate coronary territories but without left main stenosis) and an appropriate indication for revascularization were randomized on a 1:1 basis to undergo revascularization by CABG or multi-vessel PCI with first-generation DES. All patients were recommended to continue aspirin and clopidogrel for at least 12 months after stent placement. Patients with prior cardiac surgery and patients with PCI or stroke within the last 6 months were excluded. The median follow up for the FREEDOM study was 3.8 years (minimum of 2 years). Each participating site obtained Institutional Research Board approval, and all patients provided informed consent. The FREEDOM trial was registered at www.clinical-trials.gov (NCT00086450).
Assessment of outcomes in FREEDOM.
The primary outcome of the FREEDOM trial was MACE, which was defined as a composite of all-cause death, non-fatal myocardial infarction (MI), and non-fatal stroke(3)—events that were formally adjudicated by an independent committee. All patients underwent routine assessment for neurologic status and cardiac markers at each of the follow up visits to facilitate adjudication. Adjudicated 5-year MACE was the outcome of the first model created from FREEDOM.
Because the 2 goals of CAD treatment are to prevent MACE and to improve angina, we also constructed a model of being angina-free 1 year after treatment. Angina was assessed at baseline (prior to randomization), at 1, 6 and 12 months after randomization, and annually thereafter using the Seattle Angina Questionnaire (SAQ). The SAQ is a reliable and valid 19-item questionnaire with a 4-week recall period that measures 5 domains of health in patients with CAD: angina frequency (SAQ AF), angina stability, quality of life, physical limitation, and treatment satisfaction.(11,12) Linguistically and culturally validated translations in each patient’s native language were used (http://cvoutcomes.org/licenses). Domain scores range from 0 to 100, with higher scores indicating fewer symptoms and better quality of life. For this particular analysis, we focused on the SAQ AF, which has been shown to correlate closely with daily angina diaries.(13) Congruent with prior work, angina was categorized as none (SAQ AF score=100) or any (SAQ AF score <100)(14), and we developed a model to predict having angina at 1 year.
Statistical analysis.
Demographic and clinical characteristics were compared between patients with and without MACE at 5 years and those with and without angina at 1 year using independent t-tests for continuous variables and chi-square tests for categorical variables. We developed separate multivariable models for MACE at 5 years (Cox proportional hazards regression) and for having angina at 1 year (logistic regression). Patients with missing angina frequency scores at 1 year (n=237) were excluded from the angina model. As the intent of the models was to guide treatment after the performance of coronary angiography, but before revascularization, only variables available at the time of intended decision-making were included. Candidate variables were selected a priori based upon published literature and clinical experience, and interactions between each candidate variable and treatment were explored. Candidate variables for the MACE model included: age, sex, race, body mass index (BMI), smoking history, history of MI, history of stroke, history of PCI, chronic obstructive pulmonary disease (COPD), peripheral vascular disease, insulin use, reason for revascularization (acute coronary syndrome [ACS] vs. stable CAD), estimated glomerular filtration rate (eGFR), hemoglobin, left ventricular ejection fraction (LVEF), and SYNTAX score (categorized as ≤22 vs >22 (15)). Candidate variables for the angina model included the same variables, in addition to baseline SAQ AF categories of daily/weekly [SAQ AF 0–60], monthly [SAQ AF 61–99], and no angina [SAQ AF=100], and the number of anti-anginal medications at baseline. Both the categorization of the SYNTAX score and SAQ AF scale were designed to support clinical use of the models, as it was felt that clinicians could estimate these categories even if formal assessments were not performed. Spline terms were considered for all continuous variables to test for linearity. The spline was significant for the association of age with MACE, and so age was categorized as <50, 50–59, 60–69 and ≥70. To improve clinical applicability, backward variable selection was used with a p-value of <0.2 for retention of co-variates. Discrimination of the models was assessed using c-statistics, and calibration was tested by plotting deciles of predicted risk against observed event rates and comparing the slope of the regression line as well as the intercept for significant deviations between 1 and 0, respectively.
Both models were internally validated using bootstrap resampling with 500 replications.(16,17) For each step of resampling, the model was refit using the method described above, and model discrimination and calibration were assessed on the bootstrapped data and then validated on the original dataset. The difference in performance between the two data sets was calculated and averaged over the 500 replications to calculate optimism-adjusted c-statistics.
Model external validation.
Once the MACE prediction model was built, its performance, in terms of discrimination and calibration, was tested on a cohort of patients from British Columbia (BC), Canada. A detailed description of the cohort and outcome assessments has been previously published.(6) In brief, the cohort was derived from a population-based registry that collects detailed clinical and procedural data on all adults undergoing cardiac catheterization, PCI, or CABG in BC. The validation cohort included all BC patients who were ≥20 years of age with DM and angiographically confirmed multivessel CAD (stenosis of >70% in 2 or more major epicardial vessels, excluding the left main coronary artery), who underwent either PCI or isolated CABG between October 2007 and January 2014 in BC. Exclusion criteria were then applied to create a cohort that mimics the FREEDOM trial cohort. The components of the MACE outcome were determined by International Classification of Diseasae-10th Revision diagnosis codes. The patients were followed from the time of procedure to MACE, end of 5-year follow-up, or end of study, March 2014, whichever came first. The external validation was performed overall and separately on patients with stable CAD and ACS.
Demonstrating the heterogeneity of benefit of revascularization strategy.
To understand the degree of variability in patients’ predicted MACE and angina outcomes with PCI vs CABG, we calculated (on the basis of our model) each patient’s individualized predicted probability of MACE and angina twice, first assuming treatment with multi-vessel PCI and second assuming treatment with CABG. Through bootstrap analyses, we assessed whether these personalized predictions for CABG were higher or lower than the personalized predictions for PCI (for that individual patients) with 95% confidence (p<0.05).
All statistical analyses were conducted using SAS version 9.4 and R version 3.3.1.(18)
RESULTS
Patient characteristics.
There were 1900 patients with diabetes and multi-vessel CAD who were enrolled in FREEDOM and randomized to CABG (n=947) or PCI (n=953), all of whom were included in the MACE models. Comparisons of patient characteristics for patients in the angina analytic cohort compared with those alive but with missing data are shown in Supplemental Table 1. Among patients in the analytic cohorts, 346/1900 (18.2%) experienced a MACE in the 5 years after randomization (CABG vs. PCI: 15.4% vs. 20.9%) and 310 (16.3%) reported angina at 1 year (CABG vs. PCI: 14.1% vs. 18.4%). Comparisons of patients with vs. without MACE and with vs. without angina are shown in Table 1. Patients with MACE were more likely to be older, with history of smoking, to have had a prior MI or cerebrovascular event, to have three vessel disease, to be on insulin, to have a lower LVEF, lower hemoglobin, lower eGFR, higher Euroscore, and higher SYNTAX scores. Patients who reported angina at 1 year were more likely to be female, to have lower hemoglobin, and to have angina at baseline (Table 1).
Table 1.
Demographic and clinical characteristics of patients with or without MACE at 5 years and of patients with or without angina at 1 year
| Major Adverse Cardiovascular Events at 5 years | Any Angina at 1 year* | |||||
|---|---|---|---|---|---|---|
| n=346 | n=1554 | n=310 | n=1353 | |||
| Assigned to CABG arm | 146 (42.2%) | 801 (51.5%) | 0.001 | 134 (43.2%) | 677 (50.0%) | 0.030 |
| Age (y) | 64.9 ± 9.7 | 62.7 ± 8.9 | < 0.001 | 62.2 ± 9.1 | 63.1 ± 8.9 | 0.107 |
| >=70 | 120 (34.7%) | 349 (22.5%) | 77 (24.8%) | 317 (23.4%) | ||
| Male | 243 (70.2%) | 1113 (71.6%) | 0.604 | 201 (64.8%) | 999 (73.8%) | 0.001 |
| White race | 269 (77.7%) | 1182 (76.1%) | 0.517 | 247 (79.7%) | 1024 (75.7%) | 0.135 |
| Body mass index (kg/m2) | 29.5 ± 5.7 | 29.8 ± 5. | 0.306 | 30.0 ± 5.7 | 29.7 ± 5.2 | 0.347 |
| History of smoking | 65 (18.8%) | 233 (15.0%) | 0.027 | 51 (16.5%) | 207 (15.3%) | 0.436 |
| Prior myocardial infarction | 106 (30.6%) | 381 (24.5%) | 0.018 | 70 (22.6%) | 345 (25.5%) | 0.284 |
| Prior cerebrovascular event | 24 (6.9%) | 41 (2.6%) | < 0.001 | 12 (3.9%) | 39 (2.9%) | 0.362 |
| History of peripheral vascular disease | 35 (12.5%) | 161 (9.9%) | 0.201 | 27 (8.7%) | 141 (10.4%) | 0.367 |
| Chronic lung disease | 21 (6.1%) | 62 (4.0%) | 0.086 | 10 (3.2%) | 55 (4.1%) | 0.491 |
| Recent acute coronary syndrome | 121 (35.0%) | 462 (29.7%) | 0.055 | 84 (27.1%) | 422 (31.2%) | 0.157 |
| Three vessel disease | 298 (87.1%) | 1275 (82.5%) | 0.038 | 254 (81.9%) | 1121 (83.4%) | 0.532 |
| On insulin | 148 (42.8%) | 467 (30.1%) | < 0.001 | 109 (35.2%) | 410 (30.3%) | 0.095 |
| LVEF (%) | 55.8 ± 13.1 | 59.0 ± 11.2 | < 0.001 | 59.9 ± 11.4 | 58.7 ± 11.3 | 0.090 |
| Hemoglobin (g/dL) | 13.3 ± 1.9 | 13.8 ± 1.6 | < 0.001 | 13.5 ± 1.7 | 13.8 ± 1.6 | 0.006 |
| Hemoglobin A1C < 7 g/dL | 99 (31.4%) | 531 (37.0%) | 0.060 | 101 (36.7%) | 465 (36.7%) | 0.986 |
| eGFR MDRD, mL/min/1.73 m2 | 63.6 ± 19.8 | 70.9 ± 17.4 | < 0.001 | 69.1 ± 17.6 | 70.6 ± 17.8 | 0.176 |
| Euroscore | 3.4 ± 3.0 | 2.5 ± 2.3 | < 0.001 | 2.5 ± 2.1 | 2.6 ± 2.3 | 0.606 |
| Median (interquartile range) | 26.0 (21.0, 32.0) | 26.0 (19.8, 31.0) | 26.0 (19.0, 31.0) | 26.0 (20.0, 31.5) | ||
| SYNTAX score >22 | 238 (68.8%) | 987 (63.5%) | 0.063 | 198 (63.9%) | 880 (65.0%) | 0.697 |
| Number of antianginal medications | 1.5 ± 0.9 | 1.5 ± 0.9 | 0.489 | 1.5 ± 0.9 | 1.4 ± 0.9 | 0.825 |
| No Angina | 55 (19.8%) | 369 (23.0%) | 35 (11.4%) | 346 (25.7%) | ||
| SAQ Angina Frequency | 72.0 ± 23.8 | 71.0 ± 25.1 | 0.491 | 62.3 ± 26.2 | 73.4 ± 24.2 | < 0.001 |
Data on angina frequency at 1 year was missing in 237 patients
Risk Model for MACE over 5 years:
The final MACE model included 8 covariates and a treatment interaction with a history of smoking (p for interaction=0.04; Figure 1, Table 2), where patients with history of smoking had a lower risk of MACE with CABG compared with PCI. The c-statistic was 0.67 in the derivation cohort, with good internal validation (bootstrap c-statistic of 0.65) and excellent calibration of predicted and observed risk (Supplemental Figure 1). The predicted risk, by deciles, ranged from 9% to 51% over 5 years.
Figure 1. Risk prediction model for major adverse cardiovascular events (MACE) at 5 years in patients with diabetes and multi-vessel disease from FREEDOM trial.
Hazard Ratios (HRs) are presented separately for PCI vs CABG for variables with significant interaction with revascularization strategy.
Table 2: Predicting risk of 5-year major adverse cardiovascular events (MACE) model (The reduced model).
This table contains the estimated Cox regression coefficients for the significant variables that are entered into the 5-year MACE model.
| Parameter | Regression Coefficient | Standard Error | P value |
|---|---|---|---|
| Revascularization strategy (PCI) | 0.01860 | 0.17253 | 0.9142 |
| Age (50–59)* | −0.14618 | 0.24227 | 0.5463 |
| Age (60–69)* | −0.02349 | 0.23427 | 0.9201 |
| Age (70 and above)* | 0.39064 | 0.23720 | 0.0996 |
| Body mass index (BMI) (+1 unit) | −0.01665 | 0.01034 | 0.1073 |
| History of Smoking | 0.08001 | 0.16811 | 0.6341 |
| History of myocardial infarction | 0.28813 | 0.11889 | 0.0154 |
| History of stroke | 0.61882 | 0.21482 | 0.0040 |
| On Insulin | 0.43909 | 0.11114 | <0.0001 |
| eGFR (+10 mL/min per 1.73 m2) | −0.1708 | 0.00298 | <0.0001 |
| LVEF (<50%) | 0.47982 | 0.12478 | 0.0001 |
| Revascularization with PCI*Smoking | 0.44995 | 0.22391 | 0.0445 |
Reference group for Age is (less than 50)
The 5-year individualized MACE predicted risk can be calculated as follows (substitute 1 or 0 for presence or absence of any categorical variable: age category, smoking, history of MI, history of stroke, on insulin and LVEF<50%. For the variable “treatment” substitute 1 for PCI and 0 for CABG. For all continuous variables (BMI and eGFR) plug the actual value in the equation)= 1–0.48903êxp(0.01860*treatment-0.14618*(if 50<=age<=59) −0.02349*(if 60<=age<=69)+ 0.39064*(if age>=70) −0.01665*BMI+0.08001*(if history of smoking) +0.28813*(if history of MI)+ 0.61882*(if history of stroke)+ 0.43909*(if on insulin) −0.01708*eGFR+0.47982*(if LVEF<50%)+0.44995*treatment*(if history of smoking)).
External Validation of the MACE model:
The external validation cohort included 1358 patients with stable CAD and 2242 patients with stabilized ACS (Supplementary table 2) over a median follow-up of 3.48 years, (interquartile range = (1.92, 4.98). MACE events occurred in 19.4% of the stable CAD cohort and in 33.8% of the ACS cohort. In the stable CAD cohort, the MACE model showed moderate discrimination (c-statistic 0.65) and good calibration, with an intercept of 0.04 (p=0.69) and a slope of 0.60 (R2=0.90) with slightly higher predicted than observed rates in the highest decile of risk in the stable CAD cohort. In the ACS cohort, the MACE model also showed moderate discrimination (c-statistic 0.68) and good calibration, with an intercept of 0.07 (p=0.90) and a slope of 0.83 (R2=0.92) with slightly lower predicted than observed rates in the stabilized ACS cohort (Supplement Figure 3).
Risk Model for Angina at 1 year:
The final angina model included 6 variables and a treatment interaction with SYNTAX score (p for interaction=0.02; Figure 2, Table 3) such that patients with intermediate/high SYNTAX scores had less angina with CABG versus PCI. The discrimination of the angina model was 0.67 in the development cohort, which reduced to 0.64 after optimism correction with internal validation. The model had excellent calibration with a predicted angina, across deciles, of 7% to 33% (Supplement Figure 2).
Figure 2. Risk prediction model for angina at 1 year in patients with diabetes and multi-vessel disease from FREEDOM trial.
Odds Ratios (ORs) are presented separately for PCI vs CABG for variables with significant interaction with revascularization strategy.
Table 3: Predicting risk of 1-year angina model (The reduced model).
This table contains the estimated logistic regression coefficients for the significant variables that are entered into the 1-year angina model.
| Effect | Regression Coefficient | Standard Error | P value |
|---|---|---|---|
| Intercept | −0.1099 | 0.8079 | 0.8918 |
| Revascularization strategy (PCI) | −0.07646 | 0.2149 | 0.7220 |
| Age | −0.01316 | 0.007276 | 0.0707 |
| Male | −0.2904 | 0.1458 | 0.0466 |
| Presentation as acute coronary syndrome | −0.2818 | 0.1442 | 0.0509 |
| Hemoglobin (+1 g/dL) | −0.06585 | 0.04219 | 0.1188 |
| Daily/Weekly angina at baseline* | 1.1275 | 0.1972 | <.0001 |
| Monthly angina at baseline* | 0.6244 | 0.2009 | 0.0019 |
| SYNTAX SCORE (>22) | −0.3790 | 0.1963 | 0.0536 |
| Revascularization with PCI*SYNTAX (>22) | 0.5900 | 0.2700 | 0.0290 |
Reference group for angina is (no angina)
The 1-year individualized angina predicted risk can be calculated as follows (substitute 1 or 0 for presence or absence of any categorical variable: presenting as ACS, history of MI, history of peripheral vascular disease, male and angina at baseline. For the variable “treatment” substitute 1 for PCI and 0 for CABG. For all continuous variables (BMI, eGFR, LVEF, hemoglobin and age) plug the actual value in the equation)= 1/(1+exp(− (−1.1504–0.02040* treatment*BMI + 0.01612*BMI - 0.00727* treatment*eGFR + 0.002129*eGFR + 0.2870* treatment *presenting as ACS - 0.2912* treatment *history of MI + 0.05326* history of MI - 0.5213* treatment *history of peripheral vascular disease + 0.08552* history of peripheral vascular disease + 0.006867*LVEF - 0.07286*hemoglobin - 0.01575*age - 0.4273* presenting as ACS + 1.4594*treatment - 0.2648*Male + 0.8988*angina at baseline)))
Describing the Heterogeneity of Treatment Effect.
As described above in the statistical methods, we calculated (on the basis of our model) each patient’s individualized predicted probability of MACE and angina twice, first assuming treatment with multi-vessel PCI and second assuming treatment with CABG. Approximately half of patients (54.5%) would be expected to have a lower risk of MACE with CABG and 45.5% would be expected to have statistically similar MACE risk with CABG and PCI. No patients were predicted to have lower MACE with PCI than with CABG. We also note, based on the interaction with history of smoking, that all patients who had history of smoking are expected to have a lower risk of MACE with CABG vs PCI. Similarly, 35% of patients would be expected to have a similar angina relief with CABG and PCI and the rest of patients would be expected to have better angina relief with CABG. We also note, based on the interaction with the SYNTAX score, that all patients who had SYNTAX score >22 are expected to have angina relief with CABG vs PCI and all patients with SYNTAX score ≤ 22 are expected to have a similar angina relief with CABG and PCI.
Figure 3 displays theoretical patient scenarios for which these models could be used for shared-decision making. In example #1, a 55-year old diabetic patients (not on insulin) with no prior medical history with normal LVEF and BMI of 24 who presented with weekly angina would have a predicted 5-year MACE of 5.1% (95% CI 1.9%−13%) with CABG vs 5.2% (95% CI 1.8%−14%) with PCI and a 1-year predicted risk of angina of 25.0% (95% CI 18%−34%) with CABG vs. 24.0% (95% CI 17%−32%) with PCI. In contrast, a 55-year old patient with insulin-dependent diabetes and a history of prior stroke, who smoked and had had a prior myocardial infarction with depressed LVEF of 20% and a BMI of 35 who presented with weekly angina would have a predicted 5-year MACE of 25% (95% CI 8%−68%) with CABG vs. 38% (95% CI 12%−83%) with PCI and a 1-year predicted risk of angina of 25.0% (95% CI 18%−34%) with CABG vs. 24.0% (95% CI 17%−32%) with PCI.
Figure 3. Potential model output for shared-decision making.
This bar graph figure shows theoretical patient scenarios for which our models can provide individualized predictions (along with 95% CI) for 5-year MACE and 1-year angina for PCI vs CABG. See above text for actual numbers
DISCUSSION
The promise of precision medicine is to illuminate how a patient would be expected to fare with one treatment vs. another. While clinical trials, such a FREEDOM, provide the highest-quality evidence of the relative efficacy of alternative treatments, they report the mean effects of treatments across populations of randomized patients. Since no patient is ‘average’, these results can be difficult to apply to individual patients in routine clinical practice. Moreover, treatment options are usually presented by the treating physician, who may over- or under-estimate the benefits or harms of a particular treatment, depending on their specialty, experience and background. To overcome this challenge, personalized risk tools can be used to estimate an individual patient’s outcomes with different treatments. The case of PCI vs. CABG for patients with multi-vessel disease and diabetes provides an ideal setting for this approach, as there are trade-offs in terms of peri-procedural risks, length of recovery, and long-term outcomes. In this study, we derived and validated risk models for long-term MACE and angina after CABG and PCI, based on data from the FREEDOM trial, that could be used to support patient decision making. Furthermore, in a separate cohort of consecutive patients from British Columbia, we found the MACE model to perform exceedingly similarly to what was observed in the FREEDOM trial, even among patients presenting with an ACS. We found that patients with history of smoking did better with CABG, with respect to long-term MACE outcomes. We also found that patients with higher SYNTAX scores were more likely to be angina-free at 1 year after CABG; although, SYNTAX score did differentiate MACE rates between CABG and PCI. While no patients were predicted to have better outcomes (either MACE or angina) with PCI, there were a number of patients with statistically similar outcomes with CABG and PCI, in whom more discretion in the use of PCI might be reasonable; after an informed discussion with patients.
Prior literature.
Although several randomized clinical trials(19–24) have addressed the best revascularization strategy for patients with multi-vessel CAD, only a few risk tools have been created to attempt to personalize treatment choices. One such tool is the SYNTAX II score, which estimates long-term mortality for patients with complex CAD but neither MACE or angina outcomes were included in that model.(25) While there were a number of similarities between the two risk scores including some similar predictors, the SYNTAX II score estimated that 29% of patients with low SYNTAX score would have statistically better long-term mortality risk with PCI, which was not found in the diabetic patients enrolled in FREEDOM. Furthermore, CABG was superior to PCI for patients with intermediate/high SYNTAX scores in the SYNTAX trial but did not seem to be associated with better MACE outcomes in FREEDOM, although it was better for relieving angina at 1 year.
Clinical implications.
Although the overwhelming evidence favors CABG in patients with diabetes and multi-vessel disease, PCI still remains a common revascularization strategy in real world. Thus, we believe that our models can help disseminate best treatment recommendations by showing both physicians and patients the expected benefits compared to multi-vessel PCI. Clinicians can use an online calculator by visiting http://www.h-outcomes.com/freedomscore and personalized estimates based on our models will be calculated. On the other hand, numerous potential reasons exist for why patients with diabetes and multi-vessel CAD may prefer being treated with PCI, despite guideline recommendations and clear clinical trial results. To help objectify the discussion, our tool can transparently show the estimated risks and benefits of both PCI and CABG so that a more informed discussion can occur. We believe that using such tools can alleviate some of the barriers that exist in routine clinical care and can lead to more evidence-based, patient-centered care. For example, in many patients, the predicted risks of poor outcomes with CABG were substantially lower than with PCI, which could alter the strength of the recommendation to proceed with CABG, despite the longer recovery. In addition, in minority of cases where the predicted risks are similar with CABG or PCI, there may be more comfort in allowing a choice of multivessel PCI despite the overall results of FREEDOM showing superiority of CABG. The careful investigation of using these risk models in routine clinical care is needed to understand its impact on adoption of clinical trial results and patient outcomes.
While the discrimination of our models may seem modest, we would argue that they can be clinically useful. First, Hayward and Kent have argued that a c-statistic greater than 0.60 can help improve clinical trials and the use of risk models clearly enable the results of the FREEDOM trial to be applied to individual patients.(26,27) In addition, the ranges of risk that were estimated across the population were large and clinically important when trying to communicate such complex information to patients. Finally, previous work from our group has shown that the prospective application of a bleeding risk model for PCI with a similar c-statistic (0.72) resulted in a 44% reduction in bleeding.(28,29) To further define the value of these risk models, they should be prospectively studied in routine clinical care to understand their impact on shared decision-making and clinical outcomes.
Limitations.
Our findings should be interpreted in light of the following potential limitations. First, as this is a post hoc analysis from a randomized clinical trial, our models may not work as well in more general populations. However, the validation of the MACE model in an external cohort somewhat mitigates this concern. Second, we modeled our MACE prediction tool at 5 years and angina at 1 year. It is possible, that CABG might confer more (or less) benefit on symptom control over PCI over a longer time period. Third, the stents used in FREEDOM were first generation DES, and it is possible that the use of newer stents might provide better outcomes. However, the primary benefit of newer DES are avoidance of repeat procedures, which was not part of our MACE outcome.(30–32) Finally, although FREEDOM did not have narrow inclusion criteria and we externally validated the models in a real world registry, clinicians still face the conundrum of applying clinical trial results to patients who were not eligible for randomization (e.g. prohibitive surgical risk) and the use of an evidence-based model does not exacerbate this problem.
Conclusion.
Using data from a large randomized clinical trial of patients with diabetes and multi-vessel CAD, we created personalized risk prediction tools that can estimate long-term MACE and angina after revascularization. We found that CABG was the preferred strategy in the majority of patients, especially among those with a history of smoking. While PCI was never the preferred revascularization strategy, 45% of patients were expected to have similar risks of poor outcomes with CABG or PCI, and patient preferences may play a larger role in the treatment decision among these patients. This tool should be prospectively tested at the point of care to see whether or not it can improve the process of shared medical decision-making and increase the use of CABG in patients most expected to benefit from this strategy.
Supplementary Material
Figure 4. Central illustration figure. Proposed shared-decision making algorithm for patients with multi-vessel coronary artery disease and diabetes utilizing the FREEDOM score.
Utilizing our innovative FREEDOM score, a heart-team approach can be undertaken to engage patients and their physicians to make an informed decision about the best treatment strategy for their coronary artery disease.
Clinical Perspetives:
Competency in medical knowledge and patient care:
We created personalized risk prediction tools that can estimate long-term MACE and angina outcomes after revascularization in diabetic patients with multi-vessel disease. CABG was the preferred strategy in the majority of patients, however 45% of patients were expected to have similar risks of poor outcomes with CABG or PCI, and patient preferences may play a larger role in the treatment decision among these patients.
Translational outlook:
This tool should be prospectively tested at the point of care to see whether or not it can improve the process of shared medical decision-making and increase the use of CABG in patients most expected to benefit from this strategy.
Funding sources.
Dr. Qintar is supported by The National Heart, Lung, and Blood Institute of the National Institutes of Health under Award Number T32HL110837. All data collection, data analyses, the preparation of the manuscript, and the decision to submit the manuscript for publication were done independently of the study sponsor.
Abbreviations:
- CAD
coronary artery disease
- MACE
major acute cardiovascular event
- CABG
coronary artery bypass graft
- PCI
percutaneous coronary intervention
- ACS
acute coronary syndrome
- MI
myocardial infarction
- SAQ
Seattle Angina Questionnaire
Footnotes
Disclosure of potential conflicts of interest. JAS: grant funding from NIH, Patient-Centered Outcomes Research Institute (PCORI), Abbott Vascular, Bayer and an equity interest in Health Outcomes Sciences; copyright to the SAQ. Consultant to Novartis, Bayer, United Healthcare and AstraZeneca. DJC: research grant support from Boston Scientific, Abbott Vascular and Medtronic and consulting income from Medtronic. ACS: research grant support from Boston Scientific and Gilead, honoraria from Medtronic and speaking fees from Abiomed. The remaining authors have no relevant relationships to disclose.
References:
- 1.Verma S, Farkouh ME, Yanagawa B et al. Comparison of coronary artery bypass surgery and percutaneous coronary intervention in patients with diabetes: a meta-analysis of randomised controlled trials. The lancet Diabetes & endocrinology 2013;1:317–28. [DOI] [PubMed] [Google Scholar]
- 2.Fihn SD, Blankenship JC, Alexander KP et al. 2014 ACC/AHA/AATS/PCNA/SCAI/STS focused update of the guideline for the diagnosis and management of patients with stable ischemic heart disease: a report of the American College of Cardiology/American Heart Association Task Force on Practice Guidelines, and the American Association for Thoracic Surgery, Preventive Cardiovascular Nurses Association, Society for Cardiovascular Angiography and Interventions, and Society of Thoracic Surgeons. The Journal of thoracic and cardiovascular surgery 2015;149:e5–23. [DOI] [PubMed] [Google Scholar]
- 3.Farkouh ME, Domanski M, Sleeper LA et al. Strategies for multivessel revascularization in patients with diabetes. The New England journal of medicine 2012;367:2375–84. [DOI] [PubMed] [Google Scholar]
- 4.Abdallah MS, Wang K, Magnuson EA et al. Quality of life after PCI vs CABG among patients with diabetes and multivessel coronary artery disease: a randomized clinical trial. Jama 2013;310:1581–90. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Ramanathan K A J, Park JE, Fung A, Mathew V, Taylor CM,, Mancini GBJ G M, Ding L, Humphries KH, Farkouh ME. Revascularization in Diabetics with Multivessel Disease A Population-based Evaluation of Outcomes. American Heart Association, Orlando: November 10, 2015 2015. [Google Scholar]
- 6.Ramanathan K, Abel JG, Park JE et al. Surgical Versus Percutaneous Coronary Revascularization in Patients With Diabetes and Acute Coronary Syndromes. Journal of the American College of Cardiology 2017;70:2995–3006. [DOI] [PubMed] [Google Scholar]
- 7.Kipp R, Lehman J, Israel J, Edwards N, Becker T, Raval AN. Patient preferences for coronary artery bypass graft surgery or percutaneous intervention in multivessel coronary artery disease. Catheterization and cardiovascular interventions : official journal of the Society for Cardiac Angiography & Interventions 2013;82:212–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Ohlow MA, Farah A, Kuntze T, Lauer B. Patients’ preferences for coronary bypass grafting or staged percutaneous coronary intervention in multi-vessel coronary artery disease. International journal of clinical practice 2018;72:e13056. [DOI] [PubMed] [Google Scholar]
- 9.Tariq Afnan R The FREEDOM Trial: Revascularization in Diabetics with Multivessel Disease: A Population-Based Evaluation of Outcomes ACC.org, 2016.
- 10.Farkouh ME, Dangas G, Leon MB et al. Design of the Future REvascularization Evaluation in patients with Diabetes mellitus: Optimal management of Multivessel disease (FREEDOM) Trial. American heart journal 2008;155:215–23. [DOI] [PubMed] [Google Scholar]
- 11.Spertus JA, Winder JA, Dewhurst TA, Deyo RA, Fihn SD. Monitoring the quality of life in patients with coronary artery disease. The American journal of cardiology 1994;74:1240–4. [DOI] [PubMed] [Google Scholar]
- 12.Spertus JA, Winder JA, Dewhurst TA et al. Development and evaluation of the Seattle Angina Questionnaire: a new functional status measure for coronary artery disease. Journal of the American College of Cardiology 1995;25:333–41. [DOI] [PubMed] [Google Scholar]
- 13.Arnold SV, Kosiborod M, Li Y et al. Comparison of the Seattle Angina Questionnaire With Daily Angina Diary in the TERISA Clinical Trial. Circ Cardiovasc Qual Outcomes 2014;7:844–50. [DOI] [PubMed] [Google Scholar]
- 14.Spertus JA, Salisbury AC, Jones PG, Conaway DG, Thompson RC. Predictors of quality-of-life benefit after percutaneous coronary intervention. Circulation 2004;110:3789–94. [DOI] [PubMed] [Google Scholar]
- 15.Patel MR, Calhoon JH, Dehmer GJ et al. ACC/AATS/AHA/ASE/ASNC/SCAI/SCCT/STS 2017 Appropriate Use Criteria for Coronary Revascularization in Patients With Stable Ischemic Heart Disease: A Report of the American College of Cardiology Appropriate Use Criteria Task Force, American Association for Thoracic Surgery, American Heart Association, American Society of Echocardiography, American Society of Nuclear Cardiology, Society for Cardiovascular Angiography and Interventions, Society of Cardiovascular Computed Tomography, and Society of Thoracic Surgeons. Journal of the American College of Cardiology 2017;69:2212–2241. [DOI] [PubMed] [Google Scholar]
- 16.Kohavi R. A study of cross-validation and bootstrap for accuracy estimation and model selection Ijcai: Stanford, CA, 1995:1137–1145. [Google Scholar]
- 17.Steyerberg EW, Harrell FE, Borsboom GJ, Eijkemans M, Vergouwe Y, Habbema JDF. Internal validation of predictive models: efficiency of some procedures for logistic regression analysis. Journal of clinical epidemiology 2001;54:774–781. [DOI] [PubMed] [Google Scholar]
- 18.Team RC. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria: 2013 2014. [Google Scholar]
- 19.Hueb W, Lopes NH, Gersh BJ et al. Five-year follow-up of the Medicine, Angioplasty, or Surgery Study (MASS II): a randomized controlled clinical trial of 3 therapeutic strategies for multivessel coronary artery disease. Circulation 2007;115:1082–9. [DOI] [PubMed] [Google Scholar]
- 20.Kapur A, Hall RJ, Malik IS et al. Randomized comparison of percutaneous coronary intervention with coronary artery bypass grafting in diabetic patients. 1-year results of the CARDia (Coronary Artery Revascularization in Diabetes) trial. Journal of the American College of Cardiology 2010;55:432–40. [DOI] [PubMed] [Google Scholar]
- 21.Serruys PW, Morice MC, Kappetein AP et al. Percutaneous coronary intervention versus coronary-artery bypass grafting for severe coronary artery disease. The New England journal of medicine 2009;360:961–72. [DOI] [PubMed] [Google Scholar]
- 22.So SI. Coronary artery bypass surgery versus percutaneous coronary intervention with stent implantation in patients with multivessel coronary artery disease (the Stent or Surgery trial): a randomised controlled trial. Lancet 2002;360:965–70. [DOI] [PubMed] [Google Scholar]
- 23.Serruys PW, Ong AT, van Herwerden LA et al. Five-year outcomes after coronary stenting versus bypass surgery for the treatment of multivessel disease: the final analysis of the Arterial Revascularization Therapies Study (ARTS) randomized trial. Journal of the American College of Cardiology 2005;46:575–81. [DOI] [PubMed] [Google Scholar]
- 24.Serruys PW, Unger F, Sousa JE et al. Comparison of coronary-artery bypass surgery and stenting for the treatment of multivessel disease. The New England journal of medicine 2001;344:1117–24. [DOI] [PubMed] [Google Scholar]
- 25.Farooq V, van Klaveren D, Steyerberg EW et al. Anatomical and clinical characteristics to guide decision making between coronary artery bypass surgery and percutaneous coronary intervention for individual patients: development and validation of SYNTAX score II. Lancet 2013;381:639–50. [DOI] [PubMed] [Google Scholar]
- 26.Kent DM, Rothwell PM, Ioannidis JP, Altman DG, Hayward RA. Assessing and reporting heterogeneity in treatment effects in clinical trials: a proposal. Trials 2010;11:85. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Hayward RA, Kent DM, Vijan S, Hofer TP. Multivariable risk prediction can greatly enhance the statistical power of clinical trial subgroup analysis. BMC medical research methodology 2006;6:18. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Mehta SK, Frutkin AD, Lindsey JB et al. Bleeding in patients undergoing percutaneous coronary intervention: the development of a clinical risk algorithm from the National Cardiovascular Data Registry. Circulation Cardiovascular interventions 2009;2:222–9. [DOI] [PubMed] [Google Scholar]
- 29.Spertus JA, Decker C, Gialde E et al. Precision medicine to improve use of bleeding avoidance strategies and reduce bleeding in patients undergoing percutaneous coronary intervention: prospective cohort study before and after implementation of personalized bleeding risks. Bmj 2015;350:h1302. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Navarese EP, Kowalewski M, Kandzari D et al. First-generation versus second-generation drug-eluting stents in current clinical practice: updated evidence from a comprehensive meta-analysis of randomised clinical trials comprising 31 379 patients. Open heart 2014;1:e000064. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Alazzoni A, Al-Saleh A, Jolly SS. Everolimus-Eluting versus Paclitaxel-Eluting Stents in Percutaneous Coronary Intervention: Meta-Analysis of Randomized Trials. Thrombosis 2012;2012:126369. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.de Waha A, Cassese S, Park DW et al. Everolimus-eluting versus sirolimus-eluting stents: an updated meta-analysis of randomized trials. Clinical research in cardiology : official journal of the German Cardiac Society 2012;101:461–7. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.