Cardiovascular Risk Stratification for Patients Treated With Drug-Eluting Stents: Development and Validation of the DESIRE Score

Abstract: Background. We sought to develop a risk score to estimate the risk of major adverse cardiac event (MACE) occurrence during the in-hospital and long-term follow-up periods after percutaneous coronary intervention (PCI) with drug-eluting stent (DES) implantation. Methods. This score was developed and validated in a single-center database encompassing all consecutive patients treated with DES between 2007 and 2014 (n = 4061). For the development of the score, we analyzed all patients treated between January 2007 and December 2012 (n = 2863) while the validation was conducted in a cohort treated between January 2013 and December 2014 (n = 1198). MACE was defined as the combination of cardiovascular death, myocardial infarction, and ischemia-driven target-lesion revascularization. Different stratification models were developed for the in-hospital (logistic regression) and late follow-up score (Cox model). Results. In-hospital scores ranged from 0 to 37 points and comprised: (a) age; (b) previous coronary artery bypass grafting (CABG); (c) acute coronary syndrome; (d) peripheral vascular disease; (e) treatment of saphenous vein graft; (f) long lesions; (g) small vessels; (h) multivessel disease; and (i) thrombus. The late scores ranged from 0 to 45 points and comprised: (a) previous CABG; (b) diabetes mellitus; (c) acute coronary syndrome; (d) multivessel disease; (e) small vessels; (f) ejection fraction <40%; and (g) treatment of saphenous vein graft. Patients were stratified into low-risk, moderate-risk, and high-risk groups. Both scores had close to 70% accuracy for predicting MACE. Conclusion. The present score was developed and validated based on contemporary models for assessing periprocedural and long-term MACE risk post PCI, throughout the full spectrum of patient risk, and important patient subgroups.

Key words: coronary artery disease, drug-eluting stents, percutaneous coronary intervention, prognostic, risk factors, score

Risk stratification remains a key element in cardiovascular medicine to guide therapeutic decisions and to ascertain the acute and long-term procedural risks to patients. Attempts to create clinical scores for patients with coronary atherosclerotic disease (CAD) date from many decades ago and precede the development of percutaneous coronary intervention (PCI).^1,2 In the past, several studies tried to establish prognostic predictors among patients kept on medical treatment or referred to coronary artery bypass grafting (CABG). For example, the Duke Jeopardy score^3,4 and the CASS⁵ score were attempts to correlate coronary lesion severity and long-term prognosis. 

Lately, these scores have become more sophisticated, with the inclusion of other clinical and angiographic variables and novel statistical models. For example, the SYNTAX 1 and SYNTAX 2 scores are commonly used for outcome estimations, especially after PCI with drug-eluting stent (DES) implantation.^6-8 However, most of these scores were developed and validated in the setting of controlled trials and/or with outdated devices, and therefore might not reflect the current real-world practice. We sought to develop a contemporary score able to discriminate the risk of major adverse cardiac event (MACE) occurrence in the immediate (acute) and long-term clinical follow-up of a real-world cohort treated predominantly with new-generation DES implantation.

Methods

Study population and design. From May 2002 to December 2016, all consecutive patients treated at a single tertiary center with ≥1 DES were included in the non-randomized Drug-Eluting Stents In REal world (DESIRE) registry; patients were divided into three cohorts according to the period of inclusion (Figure 1). Cohort A (2002-2006) represents a historical sample, since it includes our initial experience with the first generation of DES and dual-antiplatelet therapy (DAPT) for 3 months after sirolimus-eluting stent implantation and 6 months after paclitaxel-eluting stent implantation, according to the recommendations at the time. Cohort B (2007-2014)reflects the contemporary interventional cardiology practice with the incorporation of the new-generation metallic DES implantation and use of prolonged DAPT (for a minimum of 1 year). However, inclusion in the DESIRE registry is ongoing. Cohort C represents all patients enrolled after December 2014, including our initial cases with bioresorbable scaffolds and new antiplatelet drugs (ticagrelor and prasugrel). This cohort is not the focus of the present analysis. 

To develop the DESIRE score, cohort B was subdivided in two groups: (1) cohort B1 (2007-2012), which aimed to develop the score with in-hospital and late models; and (2) cohort B2 (2013-2014), which comprises the patients in whom the score was prospectively validated, with a minimum clinical follow-up of 2 years.

The design and main features of this registry have been published elsewhere.^9-11 In brief, clinical inclusion criteria were “all comers” for routine or emergency PCI who were ≥18 years of age. Angiographic inclusion criteria were the presence of at least 1 documented stenosis ≥70% (by visual estimation) in a native coronary vessel or graft (arterial or venous) suitable for PCI with DES implantation. There were no protocol-prespecified limitations concerning the number of target lesions and/or target vessels that could be treated with DES implantation.

The study was approved by the institutional ethics committee. The institution and study participants did not receive any financial support to develop this research.

Stenting procedure. All interventions were performed according to the current standard guidelines. The stent type(s) deployed as well as predilation and/or postdilation strategies were left to the operator’s discretion. Multiple DES procedures were allowed. For cohort B, DAPT including a loading dose of aspirin (200-325 mg) and clopidogrel (300 mg) was started at least 24 hours before elective procedures; otherwise, a loading dose of 600 mg of clopidogrel was given immediately before the intervention. Postprocedural aspirin was continued indefinitely, and clopidogrel was maintained for a minimum of 12 months post stent deployment. New antiplatelet agents such as prasugrel and ticagrelor were not commercially available during cohort B enrollment; therefore, they are not represented in this study.

During the procedure, intravenous heparin (70-100 IU/kg) was administered after sheath insertion to maintain an activated clotting time of 250 seconds. Use of additional medications during the procedure, including glycoprotein IIb/IIIa inhibitors, was left to the operator’s discretion. A 12-lead electrocardiogram (ECG) was obtained before the procedure, immediately afterward, and 24 hours later.

Blood sample laboratory analysis included creatine-kinase cardiac enzymes (CK-MB mass) before the procedure (<24 hours) and 12-18 hours after treatment.

Study objectives and definitions. The primary objective of the present analysis was to develop an individual risk score able to estimate the acute (in-hospital) and late risk (up to 2 years) of combined MACE, defined as cardiac death, myocardial infarction (MI), and ischemia-driven target-lesion revascularization (TLR). Additionally, these scores would be validated for accuracy.  The secondary objective was to report the acute and long-term rates of the individual MACE components.   

All deaths were defined as cardiac in nature unless a non-cardiac origin could be clearly established by clinical and/or pathological study. A diagnosis of MI was based on either the development of new pathological Q-waves in ≥2 contiguous ECG leads, and/or elevation of CK-MB mass >5 times the upper normal limit post procedure during index hospitalization (periprocedure or type 4a MI), or cardiac enzyme elevation >2 times the upper normal limit thereafter (spontaneous MI). A diagnosis of TLR was only based on the presence of symptoms and/or signs of ischemia.

Stent thrombosis was classified as definite, probable, or possible according to definitions proposed by the Academic Research Consortium,¹² and was stratified as acute (<24 hours), subacute (24 hours to 30 days), late (1 to 12 months), or very late (>12 months).

Angiographic success was defined as attainment of <20% residual stenosis by quantitative coronary angiography (QCA) in the treated segment post DES treatment. Procedural success was defined as angiographic success plus absence of MACE during hospitalization. During the enrollment period, detailed demographic, clinical, angiographic, and procedural data, including complications, were gathered for each patient.

Clinical follow-up, by office appointment or phone call, was scheduled at 1 month, 6 months, and 12 months after stent implantation, and then annually based on information entered on case report forms at the time of the office visit/telephone contact. At the time of follow-up, data were collected pertaining to current clinical status, concomitant drug therapy (with special emphasis on the antiplatelet regimen), and interim occurrence of the predefined adverse events. All phone follow-up data were collected by the same person, who was specially trained for this task and blinded to the procedure results. Individual patient data were coded to prevent the identification of study participants.

Adverse events were adjudicated by an independent committee of three cardiologists not involved in the procedures.

Angiographic analysis. After intracoronary nitrate administration (100-200 µg), serial coronary angiography was obtained at baseline and post procedure. Offline QCA analysis was performed using the semiautomatic edge contour-detection computer analysis system QAngio XA, version 7.2 (Medis Medical Imaging Systems BV). The minimum lumen diameter (MLD) and mean reference diameter (RD), obtained from averaging 5 mm “non-diseased” segments proximal and distal to the target lesion location(s), were used to calculate the diameter stenosis (DS): DS = (1 – MLD/RD) x 100. Acute gain was defined as the change in MLD from baseline to post-stent implantation angiograms. 

In addition, a qualitative angiographic assessment was performed and the CAD extension was evaluated using the SYNTAX score calculator. 

All cardiac catheterization images were analyzed at the Hospital Angiographic Core Laboratory in São Paulo, Brazil by experienced senior operators blinded to procedural data.

Variables analyzed in the DESIRE score. Fifteen clinical and 12 angiographic variables were tested in the final model to develop the score in its two phases (in-hospital and late). 

(a) Clinical variables: age (≥18 years); sex; family history of CAD (first-degree relatives); diabetes mellitus (in use of oral medication and or insulin); hypertension; dyslipidemia; current smoking; obesity (body mass index >30 kg/m²); previous MI; previous PCI; previous CABG; previous stroke; presence of peripheral artery disease; renal insufficiency (CrCl <60 mL/min according to the Cockcroft & Gault formula); and initial clinical presentation (silent ischemia, stable angina, unstable angina, non-ST segment elevation MI, and ST-segment elevation MI).

(b) Angiographic variables: single-vessel and multivessel CAD extension (defined by the presence of lesions >70% by visual estimation); left ventricular ejection fraction (LVEF, analyzed with QAngio XA version 7.2 and classified as normal [LVEF >55%], mild deficit [LVEF >40% and < 55%], moderate deficit [LVEF >30% and <40%), and severe dysfunction [LVEF <30%]); target-lesion location (right coronary artery, left anterior descending coronary artery, circumflex coronary artery, left main coronary artery, venous and arterial graft); de novo lesions; in-stent restenosis (defined as >50% inside a previously deployed stent); ostial lesion location; bifurcations (according to Medina classification); chronic total occlusion; thrombus-containing lesions; presence and severity of lesion calcification; American College of Cardiology/American Heart Association lesion complexity classification; small vessels (<2.5 mm by QCA); and long lesions (>20 mm by QCA).

Statistical analysis. Statistical analyses were performed at two distinct moments: first, with retrospective data of patients enrolled from January 2007 to December 2012 (cohort B1), which were used to develop the score models; then, the scores were prospectively applied to all patients enrolled in the DESIRE registry between January 2013 and December 2014 (cohort B2). This second cohort was then followed for a minimum of 2 years (up to December 2016) for the score validation. Cohorts B1 and B2 were compared regarding baseline clinical and angiographic characteristics as well as procedure technical aspects and 2-year MACE rates.

Data are presented as mean ± standard deviation or frequencies. Categorical variables were compared with the Chi-square test. When the assumptions were broken, Fisher’s exact test was used. The Student’s t-test was used for continuous variable comparisons. All statistical analyses were performed with R software, version 3.3.1.

Risk score development. Due to major differences in type of MACE in the in-hospital and late follow-up periods, we decided to develop one score for each phase. For the in-hospital phase, a regression model was used,¹³ while Cox risk model was the choice for the late follow-up period, estimating the time to the first event.¹⁴ 

Initially, associations between variables were tested with a significance level of 5%. Variables that significantly correlated with MACE were included in a multiple regression model. The final selection of the variables, in both phases, was done with the stepwise/backward method, according to the Akaike information criterion (AIC).¹⁵ 

The simplified score scale was based on the odds ratio (OR) and hazard ratio (HR) estimated by the models, respectively. Estimations of probability of in-hospital MACE by the simplified model have a square correlation, while the probability of late MACE by the simplified model have a cubic correlation. 

Determination of risk level (low, moderate, or high risk), was arbitrary and based on overall risk distribution. For the in-hospital phase, the accuracy of the models was assessed with the receiver operating characteristic (ROC) curve, while we used the Haegerty and Zheng survival models for the late events.¹⁶

Risk score validation. Validation of the in-hospital and late DESIRE scores was performed among patients enrolled in the registry between 2013 and 2014 (internal validation), using the ROC curve for both time points. Furthermore, using the three risk-level categories (low, moderate, and high risk), we compared the estimated and real MACE rates for each case using the adherence (Chi-square) test.

Results

Baseline clinical, angiographic, and procedural characteristics. Between January 2007 and December 2014, a total of 4061 patients were enrolled in the DESIRE registry and comprise the B cohort, which is analyzed in the present manuscript. This population corresponds to 57.4% of the entire registry and was divided into the B1 cohort (n = 2863), which was used to develop the score, and the B2 cohort (n = 1198), which was used to validate the score (Figure 1). 

Baseline clinical characteristics of the B cohort, as well as the comparisons of the B1 and B2 subgroups, are displayed in Table 1. When compared with the B2 cohort, the B1 cohort had significantly more risk factors for CAD and comorbidities, such as family history of CAD (P=.02), current smoking (P=.01), and chronic renal insufficiency (P=.03), as well as higher incidence of previous events and procedures such as previous MI (P<.001), previous PCI (P<.01), and previous CABG (P<.001). Conversely, the B2 cohort was older (P=.02) and had higher incidence of MI as the initial clinical presentation (P<.01).

Table 2 displays the most relevant baseline angiographic and procedure characteristics. The B1 and B2 cohorts differed in most of the baseline angiographic characteristics. While the B1 cohort had more patients with multivessel disease (P<.001), ventricular dysfunction (P<.001), SVG lesions (P<.01), and thrombus in the target lesion (P<.001), the B2 cohort had a higher prevalence of ostial lesions (P=.04), bifurcations (P<.001), chronic total occlusions (P<.01), calcified lesions (P<.001), and type B2/C lesions according to the American College of Cardiology/American Heart Association classification (P<.001). 

Regarding procedural characteristics, the B2 cohort had more DESs implanted per patient (P<.001), and the stents were longer (P<.001) and had smaller diameters (P<.001). Postdilation was more often performed among these patients (P=.03). First-generation DESs were more prevalent in the B1 cohort (53.4% vs 46.6% second-generation DESs; P<.001), while only second-generation DESs were deployed in the B2 cohort. Angiographic success was achieved in almost 100% of all cases in both groups.

As previously mentioned, in the B2 cohort, lesions were longer (P<.001) and located in thinner arteries (P<.001). Residual stenosis was very low in both groups (Table 3).

Clinical outcomes. Complete follow-up was achieved in 96.5% of the cases. Median follow-up time was 4 years in the B1 cohort and 2.1 years in the B2 cohort. 

During the in-hospital phase, MACE was equally distributed among the B1 and B2 cohorts. Periprocedure (type 4a) MI was the most frequent MACE during this period and occurred in 3.87%, with no difference between the two groups.

Since the B1 cohort had a longer median follow-up period, for the purposes of comparison and score development, we only evaluated MACE through 2 years in both groups. In the late follow-up, the occurrence of MACE was very low and comparable between the two cohorts (Table 4). There were only 34 cases of stent thrombosis in the overall population (0.84%), with no difference between the groups (0.84% in cohort B1 and 0.83% in cohort B2; P=.90). Half of the cases (0.42%) were classified as definite, and most (0.54%) occurred in the late or very late period. There were no differences between the two cohorts. 

Development and validation of the DESIRE score. In-hospital score. After being tested in a univariate model, all variables with a significance level ≤.50 were included in the multivariate model. Next, using the stepwise/backward method, according to AIC, the final model was built, including 5 clinical variables and 4 angiographic variables (Table 5).

To develop a simplified model, variables were then graded according to their ORs, ranging from 0 to 37. The correlation between the logistic and the simplified models resulted in an R² coefficient of 95.1%.

The accuracy of the model, measured with the ROC curve was 74.6% and 73.7% for the logistic and simplified scores, respectively (Figure 2A). Model calibration, estimated with the Hosmer-Lemeshow test, was 5,8 df 8 (P=.66).

The B1 cohort patients were divided in three risk groups (low, moderate, and high) based on the probability of MACE in each group. Patients were considered low risk when MACE rate was ≤4% and the simplified score was ≤11, moderate risk when MACE rate was between 4% and 8% and simplified score was between 12 and 15 points, and high risk when MACE rate was >8% and simplified score was between 16 and 37 points. For validation purposes, the predicated in-hospital MACE rates for each risk level were compared with the actual MACE rates observed among patients enrolled in the B2 cohort (Figure 2B). 

Long-term (late) score. To develop the late DESIRE score, we initially tested the association between clinical and angiographic variables and the occurrence of MACE in the long-term follow-up, using the Cox proportional risk model. Only events occurring after hospital discharge were considered for this analysis. Variables selected for the Cox regression model are displayed in Table 5. Despite not showing a prognostic value in the late phase, age was forced in the final model.

As with the in-hospital score, the AIC was used to build the final score. In the simplified score, grades ranged from 0 to 45 for the selected variables using the same methodology as in the in-hospital phase. The correlation between the logistic and the simplified models resulted in an R² coefficient of 95.3%.

The accuracy of the model, measured along the years, was 65.2% (Figure 3A). Model calibration, estimated with the Hosmer-Lemeshow test, was 2,4 df 8 (P=.96).

As in the in-hospital score, B1 cohort patients were divided into three risk groups (low, moderate, and high) based on the probability of MACE in each group. Patients were considered low risk when MACE rate was ≤6%, moderate when MACE rate was between 6% and 12%, and high risk when MACE rate was >12%. For validation purposes, predicated in-hospital MACE rates for each risk level were compared with the actual MACE rates observed among patients enrolled in cohort B2 (Figure 3B). Table 6 displays the final simplified in-hospital and late DESIRE registry scores.

Discussion

Using data from the DESIRE registry, we identified almost 30 demographic, clinical, and angiographic features associated with in-hospital and long-term MACE (up to 2 years) after PCI with DES implantation. These were summarized into a simplified 7-item in-hospital and 9-item long-term DESIRE risk score, to support both hospital and long-term outcome comparisons and patient-level preprocedural risk estimation, respectively. 

To the best of our knowledge, the DESIRE score represents the first attempt to create a pre-PCI score assessing MACE risk stratification at both in-hospital and long-term follow-up periods. Most of the currently available scores contemplate either one or the other follow-up periods and most are focused on mortality only (Table 7).^17-22 

The separation of different time frames was proposed to address the different mechanisms behind MACE in the immediate and long-term clinical follow-up periods. Only half of the MACE predictors are similar for both time points. Among the possible explanations, we highlight the fact that most of the in-hospital events are related to periprocedural cardiac marker increase (type 4a MI); of note, the relevance of cardiac markers as prognostic indicators has been intensely debated.^23-26 While it is well defined that the increase in myocardial necrosis markers is related to worse clinical outcomes, it is not clear which marker (CK-MB mass, troponin I, etc) or which cut-off point better predicts untoward events. The present study defined periprocedural MI as an increase in CK-MB mass >5x the upper limit. This choice was thought to include only periprocedural MI that negatively impacted PCI outcomes.

Of note, in our score development population (the B1 cohort), 19.3% and 8.6% of the patients were classified as moderate and high risk, respectively. Conversely, among patients enrolled in the validation group (the B2 cohort), 39% were classified as moderate or high risk, which demonstrates an absolute increase of 11% in PCI complexity over the years. Although these findings represent the expansion in terms of indications for catheter-based approaches to complex CAD, it also makes the task of developing and validating accurate risk-stratification models more cumbersome. Furthermore, the overall MACE rate was relatively low. Although a low MACE rate is expected with contemporary devices and optimized PCI techniques, it might have influenced the accuracy of our risk-estimation models. 

Finally, the DESIRE score accuracy in both phases is good (~70%) and robust, but not excellent, especially when compared with other scores that have more complex execution, such as the New York score, which has an accuracy of 88.6%.²¹ These findings might be partially explained by the fact that the DESIRE is more inclusive by estimating the risk of composite MACE instead of a single component (such as death), as observed in most other registries (Table 7). Also, the continuous evolution of PCI indications, devices, and techniques tends to weaken the reproducibility of scores developed and validated at different time points. As an example, about 50% of DES options used in the development phase of the DESIRE registry were still from the first generation, while few of these devices were used in the validation phase. Conversely, the complexity observed in the second phase was considerably higher than in the first phase. 

The DESIRE score represents a user-friendly tool, since the final simplified model encompasses only 12 variables, most of which are easy to obtain. However, it is noteworthy that although good risk scores can adequately predict adverse outcomes after PCI procedures on average, they are not designed to precisely predict a single patient’s risk. In addition, an individual patient may have other medical problems not included in the model. Therefore, a risk score should be used as a tool to help physicians and patients make informed decisions, but not to foresee a patient’s specific outcome.

Our next step is to work with different institutions to perform an external validation of the DESIRE score, and we look forward to attempts to test the reproducibility of the risk score developed in this study as well as to compare its accuracy with other available scores. 

Study limitations. The main limitation of our score, as previously mentioned, is the lack of external validation, since it represents a single-center analysis. Therefore, extrapolation of our findings to other experiences should be viewed with caution. However, it is also noteworthy that most of the variables included in the statistical model used to build the DESIRE score are traditionally associated with negative outcomes in other CAD populations treated by PCI with DES implantation. Finally, the score model was built after the diagnostic procedure and therefore only applies to patients stratified with diagnostic cardiac catheterization and coronary angiographic assessment. 

Conclusions

Using data from the DESIRE registry, we developed and validated contemporary models for assessing periprocedural and long-term MACE risk post PCI, encompassing the full spectrum of patient risk and subgroups, for both in-hospital and long-term follow-up periods. We anticipate that these scores will be very helpful in guiding our therapeutic decisions and in helping patients make informed decisions.

From the ¹Hospital do Coração (HCOR), São Paulo, Brazil; and ²Instituto Dante Pazzanese de Cardiologia (IDPC), São Paulo, Brazil.

Disclosure: The authors have completed and returned the ICMJE Form for Disclosure of Potential Conflicts of Interest. The authors report no conflicts of interest regarding the content herein.

Manuscript submitted August 18, 2019, provisional acceptance given August 27, 2019, final version accepted September 9, 2019.

Address for correspondence: Adriana Costa Moreira, MD, Rua Desembargador Eliseu Guilherme, 123, Paraíso, São Paulo, SP – Brazil, CEP 04004-030. Email: moreira.adriana@uol.com.br 

1. Kimmel SE, Berlin JA, Strom BL, Laskey WK. Development and validation of simplified predictive index for major complications in contemporary percutaneous transluminal coronary angioplasty practice. The Registry Committee of the Society for Cardiac Angiography and Interventions. J Am Coll Cardiol. 1995;26:931-938.
2. de Belder AJ, Jewitt DE, Wainwright RJ, Thomas MR. Development and validation of a Bayesian index for predicting major adverse cardiac events with percutaneous transluminal coronary angioplasty. Heart. 2001;85:69-72.
3. Dash H, Johnson RA, Dinsmore RE, Harthorne JW. Cardiomyopathic syndrome due to coronary artery disease. I: relation to angiographic extent of coronary disease and to remote myocardial infarction. Br Heart J. 1977;39:733-739.
4. Califf RM, Phillips HR 3rd, Hindman MC, et al. Prognostic value of a coronary artery jeopardy score. J Am Coll Cardiol. 1985;5:1055-1063.
5. Coronary artery surgery study (CASS): a randomized trial of coronary artery bypass surgery. Quality of life in patients randomly assigned to treatment groups. Circulation. 1983;68:951-960.
6. Sianos G, Morel MA, Kappetein AP, et al. The SYNTAX score: an angiographic tool grading the complexity of coronary artery disease. EuroIntervention. 2005;1:219-227.
7. Garg S, Sarno G, Garcia-Garcia HM, et al. A new tool for the risk stratification of patients with complex coronary artery disease: the Clinical SYNTAX score. Circ Cardiovasc Interv. 2010;3:317-326.
8. 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;381:639-650.
9. Sousa A, Costa JR Jr, Moreira AC, et al. Long-term clinical outcomes of the drug-eluting stents in the real world (DESIRE) registry. J Interv Cardiol. 2008;21:307-314.
10. Costa JR Jr, Sousa A, Moreira AC, et al. Incidence and predictors of very late (>or=4 years) major cardiac adverse events in the DESIRE (drug-eluting stents in the real world)-late registry. JACC Cardiovasc Interv. 2010;3:12-18.
11. Costa JR Jr, Sousa A, Moreira AC, et al. Drug-eluting stents in the elderly: long-term (> one year) clinical outcomes of octogenarians in the DESIRE (drug-eluting stents in the real world) registry. J Invasive Cardiol. 2008;20:404-410.
12. Cutlip DE, Windecker S, Mehran R, et al; Academic Research Consortium. Clinical end points in coronary stent trials: a case for standardized definitions. Circulation. 2007;115:2344-2351.
13. Hosmer DW, Lemeshow S. Multiple logistic regression. In: Applied Logistic Regression. 2nd ed. United States of America: Wiley Interscience; 2000:31-46.
14. Colosimo EA, Giolo SR. Modelo semi-paramétrico de riscos proporcionais de cox. Análise de Sobrevivencia Aplicada: Blucher; 2006:111-143.
15. Venable WN, Ripley BD. An unbalanced four-way layout, subseção automated modelo selection. Modern Applied Statistics With S. 4th ed. Springer; 2002:175-177.
16. Heagerty PJ, Zheng Y. Survival model predictive accuracy and ROC curves. Biometrics. 2005;61:92-105.
17. Singh M, Rihal CS, Lennon RJ, Spertus J, Rumsfeld JS, Holmes DR Jr. Bedside estimation of risk from percutaneous coronary intervention: the new Mayo Clinic risk scores. Mayo Clin Proc. 2007;82:701-708.
18. Chowdhary S, Ivanov J, Mackie K, Seidelin PH, Dzavik V. The Toronto score for in-hospital mortality after percutaneous coronary interventions. Am Heart J. 2009;157:156-163.
19. Hamburger JN, Walsh SJ, Khurana R, et al. Percutaneous coronary intervention and 30-day mortality: the British Columbia PCI risk score. Catheter Cardiovasc Interv. 2009;74:377-385.
20. Peterson ED, Dai D, DeLong ER, et al. Contemporary mortality risk prediction for percutaneous coronary intervention: results from 588,398 procedures in the National Cardiovascular Data Registry. J Am Coll Cardiol. 2010;55:1923-1932.
21. Hannan EL, Farrell LS, Walford G, et al. The New York state risk score for predicting in-hospital/30-day mortality following percutaneous coronary intervention. JACC Cardiovasc Interv. 2013;6:614-622.
22. Wu C, Camacho FT, King SB 3rd, et al. Risk stratification for long-term mortality after percutaneous coronary intervention. Circ Cardiovasc Interv. 2014;7:80-87.
23. Lim CC, van Gaal WJ, Testa L, et al. With the “universal definition,” measurement of creatine kinase-myocardial band rather than troponin allows more accurate diagnosis of periprocedural necrosis and infarction after coronary intervention. J Am Coll Cardiol. 2011;57:653-661.
24. Thygesen K, Alpert JS, Jaffe AS, et al. Third universal definition of myocardial infarction. Circulation. 2012;126:2020-2035.
25. Ndrepepa G, Colleran R, Braun S, et al. High-sensitivity troponin T and mortality after elective percutaneous coronary intervention. J Am Coll Cardiol. 2016;68:2259-2268.
26. Ndrepepa G, Braun S, Cassese S, et al. Prognostic value of high-sensitivity troponin T after percutaneous coronary intervention in patients with stable coronary artery disease. Rev Esp Cardiol (Engl Ed). 2016;69:746-753.