Author + information
- Received September 6, 2001
- Revision received March 25, 2002
- Accepted April 30, 2002
- Published online August 7, 2002.
- Mandeep Singh, MD*,
- Ryan J Lennon, MS†,
- David R Holmes Jr, MD, FACC*,
- Malcolm R Bell, MB, BS, FRACP, FACC* and
- Charanjit S Rihal, MD, FACC*,* ()
- ↵*Reprint requests and correspondence:
Dr. Charanjit S. Rihal, Mayo Clinic, 200 First Street SW, Rochester, Minnesota 55905, USA.
Objectives Our goals were to identify clinical and angiographic risk factors associated with major cardiovascular complications of percutaneous coronary intervention (PCI) (in-hospital death, Q-wave myocardial infarction, urgent or emergent coronary artery bypass surgery and stroke) and to construct a simple score for risk stratification.
Background Both clinical and angiographic features influence risk of PCIs.
Methods Percutaneous coronary interventions performed between January 1, 1996, and December 31, 1999, were analyzed. Logistic regression and bootstrap methods were used to create an integer risk score for estimating the risk of procedural complications using baseline, angiographic and procedural characteristics. The risk score was tested in a validation-set consisting of all procedures performed in the year 2000.
Results Among 5,463 procedures, 5 clinical and 3 angiographic variables were significantly correlated with procedural complications: cardiogenic shock, left main coronary artery disease, severe renal disease, urgent or emergent procedure, congestive heart failure class III or higher, thrombus, multivessel disease and older age. In the validation-set, the model fitted the data adequately; the average receiver operating characteristic curve area was 0.782 (standard deviation, 0.018).
Conclusions Eight variables were combined into a convenient bedside risk scoring system that estimates the risk of complications after PCIs.
As percutaneous coronary intervention (PCI) techniques have improved, the number of clinical and angiographic indications for PCI has increased rapidly. Now PCI is frequently performed in high-risk subgroups, such as elderly patients and patients with left main coronary artery disease, diffuse disease or cardiogenic shock, in whom PCI was previously considered relatively contraindicated (1–3). The benefits of PCI among these subgroups may be partially offset by increased complication rates due to the inherent risk. Assessment of the potential for benefit and the potential for harm is important for having informed discussions with patients and families as various treatment options are considered. Recent studies have reported improvements in angiographic and clinical success rates of PCI among high-risk groups, largely due to the availability of stents and intravenous platelet glycoprotein (GP) receptor inhibitors (4,5). Although procedural failure and complications still occur, their causes may have changed (6–9).
The goals of the present study were: 1) to identify clinical and angiographic risk factors associated with major cardiovascular complications in a consecutive series of PCIs performed over a recent four-year period; 2) to construct a simple risk score (based on those risk factors) for identifying subgroups within various risk strata; and 3) to validate internally the risk score using a consecutive one-year sample of PCI procedures.
At the time of PCI, patient-specific data are entered into a prospective registry approved by the Mayo Foundation Institutional Review Board. The registry includes clinical, procedural and angiographic data on all patients undergoing PCI. All patients are contacted at 6 months, 12 months and yearly thereafter by a clinical research nurse. Medical records of all patients requiring hospitalization at Mayo Clinic, or elsewhere, are reviewed to further characterize any clinical events during follow-up. Separate approval for the current study was obtained from the Institutional Review Board. Patients who denied research authorization were excluded from the study in accordance with Minnesota law.
For data analysis and risk score construction, we included PCIs that were performed between January 1, 1996, and December 31, 1999 (study-set). Internal validation used PCIs performed during 2000 (validation-set). The period from 1996 to 1999 was chosen because: 1) stent deployment became routine rather than being used only as a bail-out treatment; 2) a thienopyridine was routinely added to aspirin therapy after stent deployment; and 3) parenteral GP IIb/IIIa receptor antagonists became available. Only first-time procedures were included.
The outcome of interest was major complications, defined as one or more of the following: 1) in-hospital death; 2) Q-wave myocardial infarction (MI); 3) urgent or emergent coronary artery bypass graft surgery (CABG) during the index hospitalization; and 4) cerebrovascular accident. Myocardial infarction was diagnosed in the presence of two of the following three criteria: 1) chest pain for at least 20 min; 2) elevation of creatine kinase (or the MB fraction) >2 times normal; and 3) new Q waves on electrocardiography. Other procedural complications, such as non–Q-wave MI and vascular access site problems, were not included in the present analysis. Patients who underwent elective bypass surgery during hospitalization (n = 20) for severe residual disease were not included.
Baseline clinical characteristics
Baseline clinical characteristics included age, gender, body mass index, history of hypertension, diabetes, hypercholesterolemia (serum cholesterol >6.21 mmol/l [240 mg/dl]), peripheral vascular disease, prior MI, prior CABG, Canadian Cardiovascular Society angina class, New York Heart Association (NYHA) heart failure class, smoking status and severe renal disease (patients with serum creatinine >265 μmol/l [3 mg/dl] or patients receiving dialysis).
Indications for PCI
Indications for PCI were classified as stable angina or unstable angina (defined as rest pain or post-MI angina), acute MI or cardiogenic shock (defined as systolic blood pressure <95 mm Hg or <110 mm Hg with inotropic or intra-aortic balloon pump support, unresponsive to fluid challenge or vasopressor therapy). Indications for PCI were subclassified as elective or emergent (patient with acute coronary syndrome brought into the cardiac catheterization laboratory because of ongoing chest pain or hemodynamic compromise or both).
PCI angiographic and procedural variables
The PCI angiographic and procedural variables included: location of the lesion; presence of multivessel disease (stenosis diameter ≥70% of vessel diameter in two or more epicardial coronary arteries or their major branches); multivessel angioplasty; type of lesion defined by the operator before the intervention (American College of Cardiology/American Heart Association [ACC/AHA] type A, B1, B2, or C); graft or native vessel angioplasty; left main coronary artery disease (stenosis diameter ≥70% of vessel diameter); intervention of unprotected left main coronary artery; intracoronary thrombus or presence of calcium at the lesion; PCI on a moderate (45° to 90°) or severe (>90°) bend; left ventricular function (≤0.40); and use of intracoronary stents and GP IIb/IIIa inhibitors.
Logistic regression was used to model the incidence of procedural complications and estimate odds ratios. All p values were two-tailed. Characteristics significant in univariate analysis were combined into an initial model, and the bootstrap method was used to remove variables to avoid overfitting the data (10–12). Two hundred bootstrap samples were selected. Backward selection at the 0.05 significance level was used to eliminate extraneous variables in each sample. Variables that were selected in at least 140 of the samples (70%) were included in the final multivariate model.
To develop a simple risk prediction score, the risk factors identified through multivariate modeling were assigned an integer coefficient. Integers were chosen to be approximately proportional to the estimated continuous coefficients from the logistic model. The score starts at 0, and each risk factor’s corresponding coefficient is added. The final score is typically between 0 and 25. The patient population was classified into five risk categories: 1) very low, 0 to 5; 2) low, 6 to 8; 3) moderate, 9 to 11; 4) high, 12 to 14; and 5) very high, ≥15. Patients were ordered by the predictive risk and separated into eight groups of similar size. Observed and expected numbers of events were calculated within each group. Model adequacy of the scoring system was then evaluated with the Hosmer-Lemeshow goodness-of-fit test (13,14). Within the study-set, discriminatory ability of the score was assessed with another 200 bootstrap samples. The risk score was calculated for each patient, and the area under the receiver operating characteristic (ROC) curve was determined for each sample. To study the performance of the prediction rule on patient subgroups, the data were stratified according to various clinical, procedural, and angiographic variables.
For the validation-set of procedures performed during 2000, the predicted probabilities of in-hospital death, Q-wave MI, stroke or need for emergent CABG were calculated from the integer risk score. Model discrimination was assessed by ROC curve analysis, and goodness-of-fit was tested with the Hosmer-Lemeshow statistic.
A total of 5,463 patients undergoing first PCI procedures were included in the study-set performed between January 1, 1996, and December 31, 1999. There were 112 deaths (2.1%), 54 Q-wave MIs (1.0%), 50 urgent or emergent CABG operations (0.9%) and 28 strokes observed (0.5%) during the index hospitalization. The composite end point (i.e., any major complication) occurred in 220 procedures (4.0%).
Baseline clinical characteristics
Mean age (± SD) of the patients was 65.8 ± 12 years, 70.4% were men, 43.1% presented with unstable angina and 14.6% presented with acute MI (Table 1). At the time of the procedure, 4.2% of the patients were in cardiogenic shock. The prevalence of risk factors is shown in Table 1: 60.6% had hypertension, 22% had diabetes, 11.3% had peripheral vascular disease and 3.4% had severe renal disease.
Clinical correlates of procedural complications
Univariate associations between baseline demographic characteristics and adverse outcomes, with odds ratios and confidence intervals, are shown in Table 2. The significant correlates included preprocedural shock, NYHA class ≥III, congestive heart failure on presentation, severe renal disease, treatment of acute MI, peripheral vascular disease and age (by decade). The following variables were not significantly associated with procedural complications: unstable angina, diabetes, hypertension, hypercholesterolemia and current smoking status. Elective procedures and male gender were associated with reduced rates of complications.
Angiographic characteristics and procedural complications
High-risk characteristics such as presence of thrombus (37.7%), multivessel (51%) or left main coronary artery disease (1.6%) and type C treated lesions (46%) were frequent (Table 3). Variables significantly associated with procedural complications after PCI included left main coronary artery disease or intervention, thrombus, calcification, multivessel disease, and ACC/AHA type C lesion. Vein graft and proximal left anterior descending coronary artery lesions were also associated with procedural complications.
Five clinical and three angiographic variables were selected in at least 70% of the multivariable analyses of the bootstrap samples. The variables (fitted to the original data) were the following: presence of cardiogenic shock, left main coronary artery disease, severe renal disease, an urgent or emergent procedure, congestive heart failure ≥class III, thrombus, multivessel disease and age (Table 4). The data did not deviate significantly from the logistic model, as indicated by the nonsignificant Hosmer-Lemeshow test result (p = 0.93) (Fig. 1). The mean area (± SD) under the ROC curve of the bootstrap samples was 0.782 (±0.018), indicating a good ability to discriminate between patients who had complications during the index hospitalization and those who did not.
The discriminatory ability of the prediction equation as measured by the area under the ROC curve (c) was calculated for each group (Table 5). In general, model discriminatory power was better among high-risk subgroups (patients with diabetes or left main coronary artery interventions, for example) than among low-risk subgroups (patients who had elective procedures).
On the basis of the integer risk score, 2,145 procedures (39.3%) were classified as very low risk; 2,182 (39.9%), low risk; 809 (14.8%), moderate risk; 210 (3.8%), high risk; and 117 (2.1%), very high risk. The observed rates of procedural complications (and the expected range of risk) in these strata were 1.0% (≤2%) for very low-risk; 3.0% (>2% to 5%), low-risk; 6.2% (>5% to 10%), moderate-risk; 19.5% (>10% to 25%), high-risk; and 35.0% (>25%), very high-risk (Fig. 2).
A total of 1,781 PCI procedures were performed during 2000, with 58 adverse events among these procedures. The model predicted 72.3 events (a difference of 1 event per 100 procedures). In the validation data set, the area under the ROC curve was 0.755, and the Hosmer-Lemeshow statistic p value was 0.64 (χ26 = −4.25), indicating that the model fitted the data well and typically correctly ranked patient risk (Fig. 3). Generally, patients who died had higher risk scores than those who did not (c statistic, 0.88), although this did not hold as well for patients who had a Q-wave MI, stroke or emergency CABG (c statistic, 0.684, 0.719 and 0.641, respectively).
Among the 607 elective PCI procedures, there were 8 complications, whereas the model predicted 10 (c statistic, 0.595). Thus, although the model was accurate for elective patients as a group, individual patients who experienced complications were only slightly more likely to have higher risk scores than were patients who had no complications.
The present study identifies clinical, angiographic and procedural variables that are significantly associated with procedural complications after PCI in current practice. This model has been internally validated. Our study documents the usefulness of readily available clinical and angiographic variables in estimating procedural complication rates in the current era of stents and GP IIb/IIIa receptor inhibitors. We developed a risk score for procedural complications after PCI in a simple integer-scoring scheme that can be readily applied at the bedside.
Comparison with previous studies
Ellis et al. (8) proposed an angiographic risk assessment scheme and found a strong correlation between complications after PCI and intervention in degenerated vein grafts and recent total occlusions. Our goal was to develop a risk score incorporating clinical and angiographic evaluations that are commonly available to clinicians before PCI. For risk prediction we found that clinical variables were as important as angiographic variables. In particular, age, the presence of shock or severe congestive heart failure, severe renal disease and procedural urgency were correlated with procedural complications, along with angiographic variables such as stenosis of the left main coronary artery, thrombus and presence of multivessel disease.
The Northern New England Cardiovascular Disease Study Group performed a multivariate analysis of in-hospital mortality (but not other major complications) in a data set of 15,331 patients who had PCI performed between 1994 and 1996 (6). Since 1996, however, PCI practice has changed considerably, with higher utilization rates of stents and glycoprotein IIb/IIIa inhibitors. Nonetheless, factors associated with in-hospital mortality were similar to our findings; they included old age, lower ejection fraction, cardiogenic shock, treatment of an acute MI, urgent and emergent priority, type C lesions and use of a preprocedure intra-aortic balloon pump.
In two prior studies from our laboratory, we documented the potential of the New York State multivariate risk score for predicting overall procedural mortality—but not other clinically important complications (9,15). The current study further demonstrates the utility of simple clinical prediction rules for complications after PCI. Several other studies have analyzed the risk factors associated with PCI, but they were performed in an era antedating current practice (16–18).
When considering treatment options, the potential for benefit must be weighed against the potential for harm for each option. For invasive procedures, in general, long-term benefit can be partially offset by procedural risk. Our risk score allows rapid bedside stratification of patients into five risk strata. It should be recognized that no model can predict which individual patients will experience complications (the “crystal ball approach”); however, the model can estimate the likelihood of a major PCI complication. Insofar as probabilistic estimates of risk and benefit are useful in clinical decision making, clinicians may find the model contributory.
Models such as ours, if externally validated, may also be useful as benchmarking mechanisms for hospitals and operators. Over large numbers of procedures, expected rates of complications can be calculated and compared with observed rates.
Although no specific subsets of patients were excluded, the current analysis is derived from the data set of a single referral center, and broader applicability is open to question. Operator volume was not considered in the current analysis (19). Independent validation in other data sets is needed. Perhaps the greatest limitation of any model in prediction of PCI complications is that discriminatory ability for individual elective patients is limited because of the overall low probability of an event. Over 95% of the elective interventions in the validation-set had predicted complication rates under 4%. As PCI procedures become safer, discriminatory ability will inevitably erode.
Clinical and angiographic characteristics are equally important in determining procedural risk with PCI. We developed and internally validated a simple integer risk score for prediction of major in-hospital complications after PCI. Five clinical variables (age, cardiogenic shock, congestive heart failure, renal failure and urgent or emergent PCI) and three angiographic variables (thrombus, multivessel disease and left main disease) can be used to accurately risk-stratify subgroups of patients undergoing PCI. This score was developed from a data set with low rates of missing values and with high rates of stent use and parenteral antiplatelet-agent use. This model may help clinicians assess procedural risk.
- American College of Cardiology/American Heart Association
- coronary artery bypass graft surgery
- myocardial infarction
- New York Heart Association
- percutaneous coronary intervention
- receiver operating characteristic
- Received September 6, 2001.
- Revision received March 25, 2002.
- Accepted April 30, 2002.
- American College of Cardiology Foundation
- Ellis S.G.,
- Tamai H.,
- Nobuyoshi M.,
- et al.
- Singh M.,
- Rihal C.S.,
- Berger P.B.,
- et al.
- O’Connor G.T.,
- Malenka D.J.,
- Quinton H.,
- et al.
- McGrath P.D.,
- Malenka D.J.,
- Wennberg D.E.,
- et al.
- Ellis S.G.,
- Guetta V.,
- Miller D.,
- Whitlow P.L.,
- Topol E.J.
- Agresti A.
- Efron B.,
- Tibshirani R.J.
- Johnson R.A.,
- Wichern D.W.
- Hosmer D.W. Jr.,
- Lemeshow S.
- Holmes D.R. Jr.,
- Berger P.B.,
- Garratt K.N.,
- et al.
- Multivessel Angioplasty Prognosis Study Group,
- Ellis S.G.,
- Vandormael M.G.,
- Cowley M.J.,
- et al.
- Kimmel S.E.,
- Berlin J.A.,
- Strom B.L.,
- Laskey W.K.