Time-varying comparative effectiveness of surgical or percutaneous revascularization on patient-centred outcomes Authors: Vikram Fielding-Singh, Christian O’Donnell, Jack H. Boyd, Meltem Tuna, Mamas A. Mamas, Marc Ruel, Louise Y. Sun Published: 2025 ABSTRACT Background:Little is known about the comparative risks and timing of patient-defined adverse cardiovascular and noncardiovascular events (PACE) after coronary artery revascularization. We investigated comparative risks of PACEs after coronary artery bypass grafting (CABG) or percutaneous coronary intervention (PCI).Methods:We conducted a retrospective cohort study of patients who underwent isolated index myocardial revascularization procedures between Oct. 1, 2008, and Dec. 31, 2018, in Ontario, Canada. The primary exposure was revascularization by CABG or PCI. The primary outcome was PACE, a composite of postoperative severe stroke, ventilator dependence, new-onset or worsening heart failure, long-term care admission, and new-onset dialysis. We modelled the association of revascularization strategy and PACE using an overlap-weighted, cause-specific hazard model, with death as a competing risk.Results:Of 140 519 patients included in the analysis, 54 018 (38.4%) underwent CABG and 86 501 (61.6%) underwent PCI. The groups were well balanced after overlap weighting. During a median follow-up duration of 4.8 (interquartile range 2.5 to 7.6) years, a total of 22 926 (16.3%) patients experienced PACE, including 9725 (18.0%) in the CABG group and 13 201 (15.3%) in the PCI group. We found no significant between-group difference in the cumulative incidence of PACE over the entire study period (average hazard ratio [HR] 0.97, 95% confidence interval 0.94 to 1.01). However, the HR for PACE varied over time. The HR comparing CABG to PCI was elevated in the first year, reached a minimum of around 0.7 at years 3 and 4, and then rose, favouring PCI again after year 8.Interpretation:The comparative risk of PACE after CABG versus PCI varied significantly over time. These findings provide granular data to support physicians and patients engaged in shared decision-making about revascularization strategies. Historically, the choice between coronary artery bypass grafting (CABG, commonly called bypass surgery) and percutaneous coronary intervention (PCI, commonly called stenting) for coronary artery disease (CAD) has been evaluated using superiority approaches that emphasize outcomes associated with traditional cardiovascular clinical trials.1–7 However, neither treatment approach has been definitively shown to be superior for all patients. Although European and American guidelines recommend CABG over PCI in select patients with heart failure, left main disease, or complex multivessel disease,8–12 practice patterns for coronary revascularization vary considerably.13–16 Some have suggested that the 2 approaches be treated as complementary rather than antagonistic, considering both the risks and timing of outcomes that are highly valued by patients when engaging in shared decision-making.17 To achieve this, granular data on the timing and risks of specific patient-centred outcomes following CABG and PCI are needed. Recently, patient-defined adverse cardiovascular and noncardiovascular events (PACE)18,19 were developed in Canada using a national consensus-based approach — involving patients with cardiovascular disease, caregivers, and multidisciplinary clinicians — to capture outcomes of cardiovascular interventions through a patient-centric lens.20 Consistent with an increasingly robust literature suggesting that patients most highly value quality of life,21 these studies found that the ability to live independently and free from severe stroke, heart failure, dialysis, and ventilator assistance were critical outcomes for patients and their caregivers. Large trials comparing revascularization methods have emphasized stroke and myocardial infarction and have generally used short- to medium-term follow-up periods.1–4,6 We sought to investigate the comparative time-varying risks of surgical versus percutaneous revascularization choice on other outcomes important to patients, such as the need for long-term care or dialysis, over a longer follow-up period. Methods Study design and population We conducted a population-based cohort study of patients who underwent an isolated coronary revascularization procedure between Oct. 1, 2008, and Dec. 31, 2018, in Ontario, Canada. During the study period, Ontario was Canada’s most populous province and reimbursed all covered services and clinicians through a publicly funded health care system.22 Inclusion and exclusion criteria were designed to select patients with CAD who had not previously experienced a PACE outcome and who could plausibly be treated nonemergently with either CABG or PCI. Patients included in the study were those who underwent an index isolated CABG or PCI within the study period, were between 40 and 85 years of age, and presented with at least 1 of the following features: 50% or greater stenosis in the left main coronary artery, 70% or greater stenosis in the left anterior descending artery, or 70% or greater stenosis in 2 or more major epicardial arteries.23 Exclusion criteria were non-Ontario residency status, ineligibility for the Ontario Health Insurance Plan (OHIP), missing data regarding age or sex, death on the procedure date, concomitant procedures (including CABG and PCI) during the same admission, emergency revascularization within 24 hours of presentation for acute myocardial infarction, presence of metastatic malignant tumour within 5 years before the index procedure, American Society of Anesthesiologists Physical Status class 5, cardiogenic shock before procedure, dialysis or ventilator dependency within 90 days before revascularization, and long-term care facility residency. Data sources We used CorHealth Ontario clinical registry data and population-level administrative health care databases from ICES. Databases and linkage methods used in this study have been previously described and validated for multiple outcomes, exposures, and comorbidities.24–27 In brief, data sets were deterministically linked using a unique encrypted identifier to preserve patient confidentiality. The CorHealth registry (date and type of revascularization procedure, preoperative left ventricular ejection fraction (LVEF), comorbidities, and the location and severity of CAD) was linked with the Canadian Institute for Health Information’s Discharge Abstract Database (comorbidities and hospital admissions), the National Rehabilitation Reporting System (severe stroke), the Continuing Care Reporting System (long-term care admissions), the Canadian Organ Replacement Registry (dialysis use), the Ontario Congestive Heart Failure Database (heart failure), the Assistive Devices Program (ventilator dependence), the Ontario Diabetes Database (diabetes status), the Same Day Surgery database (additional comorbidities), the OHIP database (physician service claims), the Registered Persons Database (vital statistics), the Office of the Registrar General Database (cause-specific death), and the Canadian Census. We defined frailty using the Johns Hopkins Adjusted Clinical Groups frailty indicator.27 Exposure The primary exposure was performance of CABG versus PCI. Outcomes The primary outcome was PACE, a previously described composite of patient-defined adverse cardiovascular and noncardiovascular events, including postoperative severe stroke requiring hospital admission for at least 14 days or inpatient rehabilitation, ventilator dependence, new-onset or worsening heart failure, long-term care admission, and new-onset dialysis.18–20 Secondary outcomes were each of the individual PACE components. Consistent with previous studies,18–20 we identified severe stroke using the Discharge Abstract Database28 and National Rehabilitation Reporting System, ventilator dependence using the Assistive Devices Program and OHIP databases, heart failure using a validated algorithm (85% sensitivity and 97% specificity),29 long-term care admission using the Continuing Care Reporting System, and dialysis using the Discharge Abstract Database, Same Day Surgery, Ontario Health Insurance Plan, and Canadian Organ Replacement Registry databases. Missing data During cohort selection, we excluded 3 patients who were missing data on age or sex. In the final cohort, 5 variables had missing values (rural residence 0.1%, income quintile 0.3%, case priority 1.5%, LVEF 14.7%, estimated glomerular filtration rate 8%). We imputed these missing values using all candidate covariates with predictive mean matching for continuous variables and logistic regression for categorical variables.30,31 Twenty data sets were imputed; we pooled estimates across the 20 imputed samples using Rubin’s Rules.30 Statistical analysis We presented symmetrically distributed continuous variables as means (standard deviations) and compared them using the Student t test. We presented asymmetrically distributed continuous variables as medians (interquartile ranges [IQRs]) and compared them using the Wilcoxon rank-sum test. We presented categorical variables as counts (percentages) and compared them using the χ2 test. Patients were censored at the end of the follow-up period or when they no longer qualified for OHIP coverage. To account for death as a competing risk, we estimated the incidence of PACE and individual PACE components using cumulative incidence functions and compared adjusted event rates between groups using the Gray test. We presented unweighted outcomes as event rates. We used logistic regression to estimate propensity scores for treatment with CABG or PCI using all covariates listed in Appendix 1, eMethods, available at www.cmaj.ca/lookup/doi/10.1503/cmaj.250312/tab-related-content. When estimating the propensity score model, we treated hospitals as a random effect to account for clustering of patients and patterns of care within institutions.14,32 We modelled the association of revascularization strategy and PACE (and individual PACE components) using overlap-weighted, cause-specific hazard models.33 Although overlap weighting does not have a precisely defined target estimand and results in a smaller effective sample size, it places the highest weights on patients who could plausibly have been treated with either CABG or PCI. This approach results in estimates that focus on patients for whom there is clinical equipoise and where the comparison is most valid.34 We assessed the balance of baseline covariates between the CABG and PCI groups with and without overlap weighting using standardized mean differences.35 Death was treated as a competing risk.36 We evaluated the proportional hazards assumption using log-log plots. As the proportional hazards assumption was violated for most outcomes, we reported both the average and time-specific hazard ratios (HRs) at 30 days, 90 days, 180 days, and annually up to 10 years. We used PCI as the reference group. We estimated time-varying HRs using restricted cubic splines.37 We determined risk differences between the 2 exposure groups at each time point in the weighted samples. We estimated standard errors of absolute risk and risk differences using 200 bootstraps within each of the 20 imputations. We conducted analyses of subgroups according to patient sex, diabetes status, LVEF less than 35% versus ejection fraction of 35% or greater, isolated left anterior descending artery versus multivessel disease, completeness of revascularization, and age. Using overlap weighting based on propensity scores within each of these subgroups, we evaluated for their modification of the effect of revascularization strategy on PACE. We also performed a landmark analysis of long-term outcomes among patients who were alive 30 days postoperatively. We performed all analyses using SAS version 9.4. We defined statistical significance as a 2-sided p value less than 0.05. Ethics approval The use of data in this project was authorized under section 45 of Ontario’s Personal Health Information Protection Act, which does not require review by a research ethics board.38 Results We included a total of 140 519 patients in the analysis (Figure 1). Of these patients, 54 018 underwent CABG and 86 501 underwent PCI. The median duration of follow-up was 4.8 (IQR 2.5 to 7.6, maximum 11.5) years in the overall cohort, 4.9 (IQR 2.5 to 7.7) years for patients treated with CABG, and 4.7 (IQR 2.5 to 7.5) years for patients treated with PCI. Cohort flow diagram. *Same-day and same-encounter procedure describes both a percutaneous coronary intervention (PCI) and coronary artery bypass grafting (CABG) on the same day or during the same hospitalization. Note: ASA = American Society of Anesthesiologists Physical Status Classification, DAD = Discharge Abstract Database, LAD = left anterior descending artery, LM = left main artery, LTC = long-term care, OHIP = Ontario Health Insurance Plan, SDS = Same Day Surgery. See Related Content tab for accessible version. Table 1 compares patient characteristics by revascularization strategy before and after overlap weighting. Before weighting, patients who underwent CABG were more likely to be male; to have multivessel disease, hypertension, or diabetes; to have a smoking history; and to have prior hospitalizations and specialist encounters than those who underwent PCI. They were also more likely to be classified as morbidly obese or as frail and less likely to have undergone previous revascularization procedures. After weighting, patient characteristics were well balanced between groups. Table 1:Baseline patient characteristics without and with overlap weightingCharacteristicUnweighted cohortsWeighted cohortsNo. (%) of patients with CABG*n = 54 018No. (%) of patients with PCI*n = 86 501ASDNo. (%) of patients with CABG*n = 16 094.25No. (%) of patients with PCI*n = 16 094.25ASDAge, yr, mean ± SD65.9 ± 9.465 ± 10.80.0865.9 ± 9.565.9 ± 10.50.00Sex0.180.00 Male43 353 (80.3)62 881 (72.7)12 469.7 (77.5)12 469.7 (77.5) Female10 665 (19.7)23 620 (27.3)3624.6 (22.5)3624.6 (22.5)Race and ethnicity† Other50 557 (93.6)81 036 (93.7)0.0015 164.4 (94.2)15 164.4 (94.2)0.00 Chinese929 (1.7)1448 (1.7)0.00289.2 (1.8)289.2 (1.8)0.00 South Asian2532 (4.7)4017 (4.6)0.00640.7 (4.0)640.7 (4.0)0.00Morbid obesity23 934 (44.3)25 564 (29.6)0.315397.7 (33.5)5397.7 (33.5)0.00Hypertension48 131 (89.1)70 127 (81.1)0.2313 916.7(86.5)13 916.7 (86.5)0.00Atrial fibrillation2278 (4.2)3661 (4.2)0.00726.8 (4.5)726.8 (4.5)0.00Heart failure10 078 (18.7)14 782 (17.1)0.043144 (19.5)3144 (19.5)0.00LVEF, % ≥ 5035 672 (66.0)60 380 (69.8)0.0810 774.2 (66.9)10 659.4 (66.2)0.02 35–4912 942 (24.0)18 647 (21.6)0.063427.9 (21.3)3654.8 (22.7)0.03 20–344611 (8.5)6331 (7.3)0.051553.5 (9.7)1444 (9.0)0.02 < 20793 (1.5)1143 (1.3)0.01338.7 (2.1)336.1 (2.1)0.00NYHA class 146 293 (85.7)74 942 (86.6)0.0313 555.4 (84.2)13 555.4 (84.2)0.00 24489 (8.3)6977 (8.1)0.011479.1 (9.2)1479.1 (9.2)0.00 32600 (4.8)3302 (3.8)0.05824 (5.1)824 (5.1)0.00 4636 (1.2)1280 (1.5)0.03235.7 (1.5)235.7 (1.5)0.00CCS anginal class 14310 (8.0)6467 (7.5)0.021479 (9.2)1479 (9.2)0.00 210 813 (20.0)16 189 (18.7)0.033725 (23.1)3725 (23.1)0.00 310 266 (19.0)11 010 (12.7)0.172983.5 (18.5)2983.5 (18.5)0.00 425 516 (47.2)44 707 (51.7)0.096574 (40.8)6574 (40.8)0.00Valvular heart disease1589 (2.9)2239 (2.6)0.02539.8 (3.4)539.8 (3.4)0.00PPM or ICD1132 (2.1)3060 (3.5)0.09468 (2.9)468 (2.9)0.00Myocardial infarction within 30 d15 597 (28.9)28 525 (33.0)0.094402.2 (27.4)4402.2 (27.4)0.00Remote myocardial infarction (> 30 d)12 128 (22.5)20 140 (23.3)0.023792.7 (23.6)3792.7 (23.6)0.00Diabetes25 751 (47.7)33 075 (38.2)0.196969.1 (43.3)6969.1 (43.3)0.00eGFR, mL/min/1.73 m2, mean ± SD75.9 ± 19.577.3 ± 20.10.0776 ± 19.376 ± 20.40.00Dialysis dependent293 (0.5)556 (0.6)0.0199.5 (0.6)99.5 (0.6)0.00COPD or asthma15 296 (28.3)25 436 (29.4)0.024753.5 (29.5)4753.5 (29.5)0.00Pulmonary circulatory disease465 (0.9)784 (0.9)0.00163.7 (1.0)163.7 (1.0)0.00Smoking status Current11 707 (21.7)19 601 (22.7)0.023584.4 (22.3)3584.4 (22.3)0.00 Former19 337 (35.8)23 939 (27.7)0.185237.3 (32.5)5237.3 (32.5)0.00 Never22 974 (42.5)42 961 (49.7)0.147272.6 (45.2)7272.6 (45.2)0.00Cerebrovascular disease5031 (9.3)5964 (6.9)0.091347.3 (8.4)1347.3 (8.4)0.00Dementia151 (0.3)424 (0.5)0.0364.9 (0.4)64.9 (0.4)0.00Frailty8652 (16.0)8509 (9.8)0.182053.3 (12.8)2053.3 (12.8)0.00Peripheral arterial disease6079 (11.3)5791 (6.7)0.161541.5 (9.6)1541.5 (9.6)0.00Liver disease343 (0.6)535 (0.6)0.00107.8 (0.7)107.8 (0.7)0.00Alcohol use disorder744 (1.4)882 (1.0)0.03198.3 (1.2)198.3 (1.2)0.00Mental health1289 (2.4)1914 (2.2)0.01387.1 (2.4)387.1 (2.4)0.00Paraplegia119 (0.2)205 (0.2)0.0039.4 (0.2)39.4 (0.2)0.00Primary malignancy2265 (4.2)3666 (4.2)0.00730.2 (4.5)730.2 (4.5)0.00Anemia9654 (17.9)12 140 (14.0)0.102681.6 (16.7)2681.6 (16.7)0.00Collagen vascular disease379 (0.7)690 (0.8)0.01131.1 (0.8)131.1(0.8)0.00Rural residence8319 (15.4)12 387 (14.3)0.032498.1 (15.5)2498.1 (15.5)0.00Income quintile Q1 (lowest)10 411 (19.3)17 164 (19.8)0.013179.9 (19.8)3179.9 (19.8)0.00 Q211 143 (20.6)17 768 (20.5)0.003252.7 (20.2)3252.7 (20.2)0.00 Q311 183 (20.7)17 924 (20.7)0.003234.4 (20.1)3234.4 (20.1)0.00 Q410 759 (19.9)17 025 (19.7)0.013226.6 (20.0)3226.6 (20.0)0.00 Q5 (highest)10 522 (19.5)16 620 (19.2)0.013200.7 (19.9)3200.7 (19.9)0.00 Operative characteristics Previous cardiac surgery610 (1.1)5912 (6.8)0.29414.6 (2.6)414.6 (2.6)0.00Previous PCI5563 (10.3)16 362 (18.9)0.252458.3 (15.3)2458.3 (15.3)0.00Number of diseased vessels LAD-only disease2520 (4.7)32 796 (37.9)0.891687.8 (10.5)1687.8 (10.5)0.00 Left main or 2-vessel disease28 727 (53.2)42 272 (48.9)0.0910 236.4 (63.6)10 236.4 (63.6)0.00 3-vessel disease22 771 (42.2)11 433 (13.2)0.684170.1 (25.9)4170.1 (25.9)0.00Case priority‡ 11913 (3.5)23 865 (27.6)0.701440.9 (9.0)1440.9 (9.0)0.00 211 112 (20.6)20 048 (23.2)0.064491.1 (27.9)4491.1 (27.9)0.00 317 112 (31.7)25 363 (29.3)0.054634.4 (28.8)4634.4 (28.8)0.00 423 881 (44.2)17 225 (19.9)0.545527.8 (34.3)5527.8 (34.3)0.00 Health care resource utilization (12 mo prior) Emergency department visits 035 474 (65.7)54 389 (62.9)0.0610 291.6 (63.9)10 291.6 (63.9)0.00 111 476 (21.2)19 067 (22.0)0.023485.2 (21.7)3485.2 (21.7)0.00 ≥ 27068 (13.1)13 045 (15.1)0.062317.5 (14.4)2317.5 (14.4)0.00Hospital admissions 031 204 (57.8)61 816 (71.5)0.2910 541.3 (65.5)10 541.3 (65.5)0.00 119 006 (35.2)20 166 (23.3)0.264522.9 (28.1)4522.9 (28.1)0.00 ≥ 23808 (7.0)4519 (5.2)0.081030.1 (6.4)1030.1 (6.4)0.00Family physician visits 04150 (7.7)8206 (9.5)0.061267 (7.9)1267 (7.9)0.00 1–521 777 (40.3)35 455 (41.0)0.016504.1 (40.4)6504.1 (40.4)0.00 5–919 095 (35.3)28 240 (32.6)0.065576.7 (34.7)5576.7 (34.7)0.00 ≥ 108996 (16.7)14 600 (16.9)0.012746.5 (17.1)2746.5 (17.1)0.00Specialist visits 05541 (10.3)17 010 (19.7)0.271995.8 (12.4)1995.8 (12.4)0.00 1–521 147 (39.1)39 326 (45.5)0.136763.8 (42.0)6763.8 (42.0)0.00 5–918 866 (34.9)20 213 (23.4)0.264974.3 (30.9)4974.3 (30.9)0.00 ≥ 108464 (15.7)9952 (11.5)0.122360.3 (14.7)2360.3 (14.7)0.00Note: ASD = absolute standard difference, CABG = coronary artery bypass graft, CCS = Canadian Cardiovascular Study, COPD = chronic obstructive pulmonary disease, eGFR = estimated glomerular filtration rate, ICD = implantable cardiac defibrillator, LAD = left anterior descending, LVEF = left ventricular ejection fraction, NYHA = New York Heart Association; PCI = percutaneous coronary intervention, PPM = permanent pacemaker, SD = standard deviation.*Unless indicated otherwise.†Ethnicity was determined using a validated surname-based algorithm for use in ICES data.39‡Defined by the Cardiac Care Network of Ontario for CABG (level 1 = access target [recommended maximum wait time] of ≤ 2 d; level 2 = access target of ≤ 14 d; level 3 = access target of ≤ 42 d; level 4 = access target of ≤ 180 d) and PCI (level 1 = access target of ≤ 2 d; level 2 = access target of ≤ 7 d; level 3 = access target of ≤ 14 d; level 4 = access target of ≤ 28 d). Baseline patient characteristics without and with overlap weighting Note: ASD = absolute standard difference, CABG = coronary artery bypass graft, CCS = Canadian Cardiovascular Study, COPD = chronic obstructive pulmonary disease, eGFR = estimated glomerular filtration rate, ICD = implantable cardiac defibrillator, LAD = left anterior descending, LVEF = left ventricular ejection fraction, NYHA = New York Heart Association; PCI = percutaneous coronary intervention, PPM = permanent pacemaker, SD = standard deviation. Unless indicated otherwise. Ethnicity was determined using a validated surname-based algorithm for use in ICES data.39 Defined by the Cardiac Care Network of Ontario for CABG (level 1 = access target [recommended maximum wait time] of ≤ 2 d; level 2 = access target of ≤ 14 d; level 3 = access target of ≤ 42 d; level 4 = access target of ≤ 180 d) and PCI (level 1 = access target of ≤ 2 d; level 2 = access target of ≤ 7 d; level 3 = access target of ≤ 14 d; level 4 = access target of ≤ 28 d). PACE A total of 22 926 (16.3%) patients experienced PACE, of whom 9725 (18.0%) underwent CABG and 13 201 (15.3%) underwent PCI. We found no significant difference in the cumulative incidence of PACE between groups (log-rank p = 0.97) (Appendix 1, eFigures 1 and 2; Figure 2). The most frequent individual PACE event was new onset heart failure; ventilator dependence was very uncommon. Over the entire study period, patients who underwent CABG had, on average, a similar risk of PACE as those who underwent PCI (HR 0.97, 95% confidence interval [CI] 0.94 to 1.01) (Figure 3). However, the time-specific HR of PACE followed a U-shaped pattern, such that in the CABG group, the risk of PACE was significantly elevated during the first year, significantly lower between years 2 and 7, and elevated beyond the eighth postoperative year, compared with the PCI group (Figure 4; Appendix 1, eTable 1). Cumulative incidence of (A) patient-defined adverse cardiovascular and noncardiovascular events (PACE) and individual PACE events — (B) severe stroke, (C) new-onset heart failure, (D) long-term care, (E) new dialysis requirement, and (F) ventilator dependence — after coronary artery bypass grafting (CABG) or percutaneous coronary intervention (PCI) in the overlap-weighted cohort. Estimated treatment effects of coronary artery bypass grafting (CABG) versus percutaneous coronary intervention (PCI) on 10-year absolute risk of patient-defined adverse cardiovascular and noncardiovascular events (PACE) and individual PACE events in the overlap-weighted cohort. Note: CI = confidence interval, HR = hazard ratio. See Related Content tab for accessible version. Time-specific hazard ratios (HRs) for patient-defined adverse cardiovascular and noncardiovascular events (PACE) and individual PACE events — (B) severe stroke, (C) new-onset heart failure, (D) long-term care, (E) new dialysis requirement, and (F) ventilator dependence — after coronary artery bypass grafting (CABG) versus percutaneous coronary intervention (PCI) in the overlap-weighted cohort. Note: CI = confidence interval. Individual PACE events Overall, groups did not differ significantly in the cumulative incidence of most PACE events, with the exception of severe stroke and long-term care admission (Appendix 1, eFigure 1). Over the course of follow-up, we observed higher averaged rates of severe stroke (average HR 1.25, 95% CI 1.12 to 1.40) and lower averaged rates of long-term care admissions (average HR 0.83, 95% CI 0.77 to 0.90) among patients who underwent CABG compared with those who underwent PCI (Figure 3). We observed time-specific patterns in the risks of severe stroke, new-onset heart failure, long-term care admissions, and new-onset dialysis (Figure 4; Appendix 1, eTable 1). Specifically, in the CABG group, the HR of stroke was significantly elevated in the first 2 years and was lower thereafter, compared with the PCI group. The time-specific HR of heart failure after CABG (v. PCI) followed a U-shaped pattern such that it was significantly elevated in the first year, significantly lower in years 2 to 7, and began to rise after year 8. The HR of needing long-term care also followed a U-shaped pattern, with a significantly lower observed HR in the CABG group from years 2 to 7. In contrast, the HR of new-onset dialysis was lower in the CABG group for the first 6 months and was thereafter similar between groups. Subgroup analyses We observed minimal evidence of heterogeneity between prespecified subgroups. For patients with LVEF of 35% or greater, the HR of PACE was significantly higher in the first year after CABG (v. PCI) compared with patients with reduced LVEF (Figure 5 and Figure 6; Appendix 1, eTable 2). The difference between LVEF groups was no longer present by year 2 of follow-up. In addition, although CABG (v. PCI) was associated with low risk of PACE (HR < 1) for all age groups in years 3 and 4 after revascularization, the HR was lower and suggested more benefit for older patients (aged > 75 yr) across a longer follow-up period (years 2 to 7) than for patients in younger age categories. Estimated treatment effects of coronary artery bypass grafting (CABG) versus percutaneous coronary intervention (PCI) on 10-year absolute risk of patient-defined adverse cardiovascular and noncardiovascular events in subgroups by sex, presence of diabetes, left ventricular ejection fraction (LVEF), left anterior descending (LAD) artery–only versus multivessel disease, and completeness of revascularization in the overlap-weighted cohort. Note: CI = confidence interval, HR = hazard ratio. See Related Content tab for accessible version. Time-specific hazard ratios (HRs) for patient-defined adverse cardiovascular and noncardiovascular events following coronary artery bypass grafting (CABG) versus percutaneous coronary intervention (PCI) in subgroups by (A) sex, (B) presence of diabetes, (C) left ventricular ejection fraction (LVEF), (D) left anterior descending (LAD) artery–only versus multivessel disease, (E) completeness of revascularization, and (F) age in the overlap-weighted cohort. Note: CI = confidence interval. Landmark analysis Findings of the landmark analysis among patients who survived to 30 days postoperatively were consistent with the primary models (Appendix 1, eTables 3 to 6). Interpretation In this large, real-world cohort study of patients who underwent revascularization procedures, the risk of PACE was significantly elevated for 1 year among patients who underwent CABG (v. PCI), significantly lower for years 2 to 7, and then equivalent for years 8 to 10 of the follow-up period. This risk profile was largely driven by increased early risks of severe stroke and new-onset heart failure among those who underwent CABG, with lower mid-term (years 2 to 7) risks of new-onset heart failure and need for long-term care, compared with those who underwent PCI. Of note, patients who underwent CABG had a lower risk of new-onset dialysis in the first 6 months after revascularization than those who underwent PCI. These data provide new granular evidence on the risk and timing of a unique set of outcomes that are important to patients. Although patient-centred outcomes in cardiovascular health research remain poorly understood, the outcomes emphasized in large revascularization trials do not seem to always align with patient preferences. For example, studies of patient preferences report that stroke may be considered a worse outcome than death and that the need for revascularization is substantially less important to patients than other common major adverse cardiac events.40–43 Other studies have confirmed that outcomes important to patients emphasize freedom from disability after cardiac surgery18,20 and that patients do not weigh the risk of death as heavily as physicians do when considering CABG versus PCI.44 Our study fills gaps left by the constellation of large randomized trials by focusing on specific outcomes that matter to patients and providing highly granular risk estimates that support shared patient-centred clinical decision-making. Our findings also support this patient-centred approach by providing a detailed picture of how the timing of specific outcomes varies by revascularization procedure over a long follow-up period. Timing is critically important to patients. Even when presented with double the risk of death with PCI compared with CABG, patients express a preference for PCI.44 In part, this may be because people assign greater value to short-term events than long-term events.45 Prior literature also suggests that risk changes over time, although follow-up periods have generally been shorter than the 10 years achieved in our study. Consistent with our study, evidence exists that the risk of stroke is elevated in the short term, but not the long term, among patients treated with CABG compared with PCI.2,46,47 Our study addresses a knowledge gap left by differences in the timing and composition of primary end points from randomized controlled trials. These differences have made characterizing more specific patient-centred and quality of life outcomes challenging. An analysis of the FREEDOM (Future Revascularization Evaluation in Patients with Diabetes Mellitus: Optimal Management of Multivessel Disease) trial focused on quality of life after CABG versus PCI for patients with diabetes and multivessel disease.48 The authors reported that, although PCI was associated with short-term benefits, CABG was associated with small but significant improvements in scores on the Seattle Angina Questionnaire at 2 years of follow-up. Beyond that time point, outcomes after PCI and CABG did not differ. Short-term follow-up of the FAME 3 (Fractional Flow Reserve Versus Angiography for Multivessel Evaluation) trial showed faster recovery after PCI but no significant difference in quality of life at 1 year.49 However, these studies used substantially shorter follow-up periods than our study and relied on patient-reported outcome questionnaires. In contrast, our study identified the risks of occurrence of a set of validated patient-centred outcomes — including stroke, new-onset heart failure, need for long-term care, and dialysis — with greater temporal resolution and a substantially longer follow-up period (10 yr) than previously reported. The introduction of the PACE measure is an important development in cardiac surgery. Further investigation in clinical trials and comparative effectiveness studies is warranted to validate PACE and characterize its long-term incidence in diverse populations and health care settings. Such research could also identify how patients prioritize individual outcomes within the PACE measure, shaping future study designs. Additionally, studies examining how patients balance short-term versus long-term risk would offer insights into the importance of PACE outcomes for patient-centred care. Our findings raise the possibility of heterogeneity in PACE outcomes by age, facility type, or other patient- and system-level factors. Finally, prospective studies should consider systematic, extended follow-up periods to better capture evolving risk profiles after revascularization procedures. Limitations Despite careful risk adjustment, treatment allocation bias and residual unmeasured confounding may persist, given the observational design. Although our study was designed to include only patients who could plausibly have been treated with either CABG or PCI, unobserved differences may persist between the groups, even after adjustment. We could not estimate the importance of factors not available in administrative data, such as physiologic measures of frailty, preprocedural functional status, or anatomic data such as the SYNTAX (Synergy between PCI with Taxus and Cardiac Surgery) score. Findings may not be generalizable to patients in other health care systems or with different demographic characteristics. Further validation of PACE outcomes with additional patient and caregiver groups and in different data sets is warranted. Although we modelled clustering by hospital in the estimation of propensity scores to account for institutional variation in treatment choices, we did not do so in the outcome model, and bootstrapped estimates may not fully account for intrahospital correlation. Conclusion In this large, real-world cohort study, CABG was associated with increased short-term, reduced mid-term, and equivalent long-term risk of PACE, compared with PCI, largely driven by early risk of stroke and heart failure, as well as reduced mid-term risk of heart failure and need for long-term care. Our findings provide patients and clinicians with time-specific and patient-centred outcome data to guide decisions about choice of revascularization strategy. Supplementary Information