Abstract
Background
The mortality benefit of transradial primary PCI has been shown by several studies. Previous risk models have not considered access site as a candidate predictor and many of them were developed using low risk populations of randomized trials. We conducted a prospective cohort study to construct and validate an admission risk model including access site as candidate variable for predicting 30-day mortality after primary PCI.
Methods
We analyzed data of 1255 patients using variables readily available at presentation. Predictor selection was based on backward logistic regression combined with bootstrap resampling. The model has been validated internally and temporally externally.
Results
Thirty-day mortality was independently associated with older age, faster heart rate, need for life support on or prior to admission, and femoral access while it was inversely related to systolic blood pressure. ROC curve analysis revealed high discriminatory power, which was preserved in the validation set (c-statistic: 0.88 and 0.87, respectively). For the new score the acronym ALPHA (Age, Life support, Pressure, Heart rate, Access site) has been coined. Compared with previous models, our score achieved the highest c-statistic (0.87) followed by the GRACE 2.0 (0.86), APEX-AMI (0.86), and CADILLAC (0.85) models, the other scoring systems (TIMI, Zwolle, and PAMI) performed less well. The ALPHA, GRACE 2.0, APEX-AMI, and CADILLAC models predicted 30-day mortality better than the PAMI score (p = 0.005, 0.004, 0.01, and 0.02, respectively).
Conclusions
Using this tool, mortality risk may be precisely assessed at admission and patients who may benefit most from transradial access may be identified.
Highlights
- •
A prospective cohort study to construct a simple risk model including vascular access site for predicting 30-day mortality of patients undergoing primary PCI.
- •
Thirty-day mortality was independently associated with older age, faster heart rate, need for life support on or prior to admission, and femoral access while it was inversely related to systolic blood pressure.
- •
During comparative validation our score achieved the highest c-statistic.
- •
Using this tool, mortality risk may be assessed at admission and patients who may benefit most from transradial access may be identified.
1
Introduction
Mortality risk of ST-segment elevation myocardial infarction (STEMI) patients shows high variability. In order to assess individual risk, a number of scoring systems have been developed. Some of the models use parameters that are available at presentation like demographic and historical data, vital signs, ECG (“admission models”) , while others also make use of more time-consuming imaging/laboratory studies, and findings/results of the coronary intervention (“discharge models”) . The most common variables used in existing models are age , Killip class , heart rate and systolic blood pressure (SBP) at admission , ECG localization of the infarction , ischemia time , glomerular filtration rate , occurrence of triple vessel disease , final Thrombolysis In Myocardial Infarction (TIMI) flow . Despite this plethora of models, as treatment approaches evolve over time, there is a need to build new risk prediction systems to preserve/improve prognostic accuracy. One of the most relevant improvements in this context is transradial primary percutaneous coronary intervention (PPCI), since this technique is capable to reduce mortality . Previous risk models have not considered access site as a candidate predictor . A model employing access site as a variable would be able to identify patients who may benefit most from transradial PPCI. Furthermore, existing scoring systems may not be accurate for the whole risk spectrum of patients, since many of them were constructed using low risk populations of trials excluding cardiogenic shock . For the foregoing reasons, we conducted a single center, prospective cohort study to develop and comparatively validate a registry-based admission risk model including access site as candidate variable for predicting 30-day mortality of STEMI patients undergoing PPCI.
2
Methods
2.1
Study design
We analyzed prospectively collected data of 1255 consecutive STEMI patients treated with PPCI at our center from September 2007 through December 2011. Data were divided by time into development (September 2007 through March 2010, 750 patients, ≈60%) and validation (between April 2010 and December 2011, 505 cases, ≈40%) sets. Patients were followed-up for a minimum of one year by means of hospital records, follow-up visits, telephone interviews, and records of the National Health Insurance Fund. No patient was lost for follow-up and all cases were complete cases, i.e. no data were missing. We investigated the role of eight candidate variables readily available at or soon after presentation on 30-day all-cause mortality using logistic regression. We studied age, onset-to-door (ischemia) time, heart rate and systolic blood pressure at admission as continuous parameters, whereas gender, ECG localization of the infarction, need for life support on or prior to admission, and vascular access site were investigated as categorical variables. Vascular access site was considered as the artery from which the coronary intervention was actually performed regardless of femoral artery puncture in cases necessitating intra-aortic balloon pump counterpulsation (for further definitions see Supplemental Methods). To evaluate predictive stability over time, we tested the model’s predictive performance at 180 days, 1 year, and 1 year among 30-day survivors. All patients gave informed consent to be available for follow-up and the study was conducted according to the principles of the Declaration of Helsinki.
2.2
Primary PCI
The choice of access site was left to the operator’s discretion. After insertion of the introducer unfractionated heparin (70 IU/kg, for PCI with planned use of the glycoprotein IIb/IIIa receptor inhibitor eptifibatide or 100 IU/kg, for PCI without planned use of eptifibatide) was given intra-arterially. Use of eptifibatide was at the operator’s discretion. PCI was then performed using standard techniques. The arterial sheath was removed immediately after the procedure. Bleeding from the radial artery was stopped using the TR Band (Terumo Europe, Leuven, Belgium), while the femoral artery was closed by the FemoSeal device (St. Jude Medical, St. Paul, MN). In cases of persistent femoral artery bleeding, manual compression was applied. All patients were pretreated with acetylsalicylic acid and a loading dose of 600 mg clopidogrel, i.e. none of them received prasugrel or ticagrelor, and discharged on dual antiplatelet therapy for at least 12 months. Interventional cardiologists were high-volume operators (> 200 PCIs/year) skilled in both femoral and radial techniques.
2.3
Statistical analysis
Variables in 2 × 2 contingency tables were assessed using Fisher’s exact test. Categorical data in 2 × k contingency tables were analyzed using the unordered chi-squared test or, to detect linear trend, the chi-squared test for trend. Continuous parameters were examined for normality with the D’Agostino–Pearson test. As none of the continuous variables showed normal distribution, the Mann–Whitney test was applied for comparisons. Survival was analyzed with the use of Kaplan–Meier curves, log-rank test, or log-rank test for trend, if number of groups was > 2 with a natural ordering. A two-tailed p value < 0.05 was considered statistically significant.
We used logistic regression analysis for model construction. We did not apply formal sample size calculations because there are no generally accepted approaches to estimate sample size for risk prediction models. Nevertheless, to avoid overfitting, we took into consideration the number of events per variable in the initial multivariate model . Based on the work of Vittinghoff et al. a number of 5 to 9 events per covariate was considered acceptable . Presence of non-linear relationships of the continuous variables to log odds of 30-day mortality were explored using non-linear functions (restricted cubic splines, RCS) which were evaluated graphically and by Wald testing for linearity . Since SBP has been proved to be non-linearly associated with the logit of the outcome, it was represented by a RCS with three knots (at 80, 120, and 160 mmHg) technically resulting in two variables: SBP and SBP’ (Supplemental Fig. S1). To determine the most relevant predictors, backward stepwise logistic regression (significance level for staying in the model: p = 0.1) was combined with bootstrap resampling . Variables selected in at least 60% of the samples were included in the final model. The likelihood ratio and the Hosmer-Lemeshow tests were used to assess model fit, whereas discriminative ability was evaluated by receiver operating characteristic (ROC) curve analysis .
Since the performance of a prediction model in the derivation data set may overestimate the true performance (the difference between the “apparent” and true performance is known as “optimism”), we conducted internal validation by bootstrapping . During temporal external validation, the Hosmer-Lemeshow test, calibration intercept, and slope were used to describe model fit, while the c-statistic was calculated to evaluate discriminatory power . Comparative validation of the predictive capacity was carried out according to DeLong et al. .
Model construction and validation studies were performed with R version 3.3.0 (R Foundation for Statistical Computing, Vienna, Austria) using the rms 4.5–0 and pROC 1.8 packages. Descriptive statistics and graphical interpretation of the results were partly carried out using MedCalc 16.8 (MedCalc Software, Ostend, Belgium).
2
Methods
2.1
Study design
We analyzed prospectively collected data of 1255 consecutive STEMI patients treated with PPCI at our center from September 2007 through December 2011. Data were divided by time into development (September 2007 through March 2010, 750 patients, ≈60%) and validation (between April 2010 and December 2011, 505 cases, ≈40%) sets. Patients were followed-up for a minimum of one year by means of hospital records, follow-up visits, telephone interviews, and records of the National Health Insurance Fund. No patient was lost for follow-up and all cases were complete cases, i.e. no data were missing. We investigated the role of eight candidate variables readily available at or soon after presentation on 30-day all-cause mortality using logistic regression. We studied age, onset-to-door (ischemia) time, heart rate and systolic blood pressure at admission as continuous parameters, whereas gender, ECG localization of the infarction, need for life support on or prior to admission, and vascular access site were investigated as categorical variables. Vascular access site was considered as the artery from which the coronary intervention was actually performed regardless of femoral artery puncture in cases necessitating intra-aortic balloon pump counterpulsation (for further definitions see Supplemental Methods). To evaluate predictive stability over time, we tested the model’s predictive performance at 180 days, 1 year, and 1 year among 30-day survivors. All patients gave informed consent to be available for follow-up and the study was conducted according to the principles of the Declaration of Helsinki.
2.2
Primary PCI
The choice of access site was left to the operator’s discretion. After insertion of the introducer unfractionated heparin (70 IU/kg, for PCI with planned use of the glycoprotein IIb/IIIa receptor inhibitor eptifibatide or 100 IU/kg, for PCI without planned use of eptifibatide) was given intra-arterially. Use of eptifibatide was at the operator’s discretion. PCI was then performed using standard techniques. The arterial sheath was removed immediately after the procedure. Bleeding from the radial artery was stopped using the TR Band (Terumo Europe, Leuven, Belgium), while the femoral artery was closed by the FemoSeal device (St. Jude Medical, St. Paul, MN). In cases of persistent femoral artery bleeding, manual compression was applied. All patients were pretreated with acetylsalicylic acid and a loading dose of 600 mg clopidogrel, i.e. none of them received prasugrel or ticagrelor, and discharged on dual antiplatelet therapy for at least 12 months. Interventional cardiologists were high-volume operators (> 200 PCIs/year) skilled in both femoral and radial techniques.
2.3
Statistical analysis
Variables in 2 × 2 contingency tables were assessed using Fisher’s exact test. Categorical data in 2 × k contingency tables were analyzed using the unordered chi-squared test or, to detect linear trend, the chi-squared test for trend. Continuous parameters were examined for normality with the D’Agostino–Pearson test. As none of the continuous variables showed normal distribution, the Mann–Whitney test was applied for comparisons. Survival was analyzed with the use of Kaplan–Meier curves, log-rank test, or log-rank test for trend, if number of groups was > 2 with a natural ordering. A two-tailed p value < 0.05 was considered statistically significant.
We used logistic regression analysis for model construction. We did not apply formal sample size calculations because there are no generally accepted approaches to estimate sample size for risk prediction models. Nevertheless, to avoid overfitting, we took into consideration the number of events per variable in the initial multivariate model . Based on the work of Vittinghoff et al. a number of 5 to 9 events per covariate was considered acceptable . Presence of non-linear relationships of the continuous variables to log odds of 30-day mortality were explored using non-linear functions (restricted cubic splines, RCS) which were evaluated graphically and by Wald testing for linearity . Since SBP has been proved to be non-linearly associated with the logit of the outcome, it was represented by a RCS with three knots (at 80, 120, and 160 mmHg) technically resulting in two variables: SBP and SBP’ (Supplemental Fig. S1). To determine the most relevant predictors, backward stepwise logistic regression (significance level for staying in the model: p = 0.1) was combined with bootstrap resampling . Variables selected in at least 60% of the samples were included in the final model. The likelihood ratio and the Hosmer-Lemeshow tests were used to assess model fit, whereas discriminative ability was evaluated by receiver operating characteristic (ROC) curve analysis .
Since the performance of a prediction model in the derivation data set may overestimate the true performance (the difference between the “apparent” and true performance is known as “optimism”), we conducted internal validation by bootstrapping . During temporal external validation, the Hosmer-Lemeshow test, calibration intercept, and slope were used to describe model fit, while the c-statistic was calculated to evaluate discriminatory power . Comparative validation of the predictive capacity was carried out according to DeLong et al. .
Model construction and validation studies were performed with R version 3.3.0 (R Foundation for Statistical Computing, Vienna, Austria) using the rms 4.5–0 and pROC 1.8 packages. Descriptive statistics and graphical interpretation of the results were partly carried out using MedCalc 16.8 (MedCalc Software, Ostend, Belgium).
3
Results
3.1
Patient characteristics
Demographic, clinical, and procedural characteristics of the derivation and validation cohorts are summarized in Table 1 . Although there were differences between the two populations with respect to medical history, concomitant diseases, and procedural characteristics, the majority of the parameters were similar.
| Variable | Derivation (n = 750) | Validation (n = 505) | p Value |
|---|---|---|---|
| Age, years, median (IQR) | 64.0 (54.0–73.0) | 62.0 (54.0–71.0) | 0.22 |
| Female, % | 35.9 | 34.7 | 0.67 |
| BMI kg/m 2 , median (IQR) | 27.3 (24.5–30.5) | 26.8 (24.1–30.1) | 0.11 |
| LVEF, %, median (IQR) | 50.0 (43.0–55.0) | 50.0 (42.0–56.3) | 0.99 |
| Hypertension, % | 71.1 | 66.5 | 0.09 |
| Diabetes mellitus, % | 25.7 | 23.0 | 0.29 |
| Verified dyslipidaemia, % | 46.3 | 30.1 | < 0.0001 |
| Current smokers, % | 34.8 | 44.2 | 0.0009 |
| Peripheral artery disease, % | 8.3 | 5.3 | 0.06 |
| Cerebrovascular disease, % | 6.7 | 9.7 | 0.05 |
| Congestive heart failure, % | 3.7 | 5.5 | 0.16 |
| Chronic renal failure, % | 4.4 | 2.0 | 0.03 |
| Previous myocardial infarction, % | 11.5 | 12.3 | 0.66 |
| Previous PCI, % | 6.0 | 8.7 | 0.07 |
| Previous CABG, % | 2.0 | 2.2 | 0.84 |
| Onset-to-door time, hours, median (IQR) | 3.5 (2.0–6.0) | 3.0 (2.0–5.0) | 0.05 |
| Door-to-balloon time, min, median (IQR) | 50.0 (33.0–75.0) | 41.0 (30.0–68.0) | 0.0001 |
| ECG localization, % | 0.10 | ||
| anterior | 44.1 | 39.0 | |
| inferior | 51.6 | 52.1 | |
| posterior/lateral | 4.3 | 8.9 | |
| Life support on or prior to admission, % | 7.6 | 9.1 | 0.35 |
| Heart rate, 1/min, median (IQR) | 79.0 (69.0–90.0) | 79.0 (67.0–90.0) | 0.83 |
| SBP, mmHg, median (IQR) | 130.0 (110.0–148.0) | 130 (110.0–145.0) | 0.84 |
| Cardiogenic shock, % | 8.5 | 8.7 | 0.92 |
| IABP use, % | 5.7 | 5.1 | 0.71 |
| Respirator use, % | 12.0 | 12.3 | 0.93 |
| Glycoprotein IIb/IIIa receptor inhibitor, % | 82.1 | 83.4 | 0.60 |
| Transradial primary PCI, % | 84.0 | 94.1 | < 0.0001 |
| TR primary PCI in cardiogenic shock, % | 59.4 | 79.5 | 0.04 |
| TR primary PCI with IABP, % | 58.1 | 65.4 | 0.62 |
| Vessel dilated, % | 0.07 | ||
| LAD | 39.1 | 33.1 | |
| D, IM | 1.6 | 1.4 | |
| LCX | 11.3 | 10.3 | |
| RCA | 38.8 | 41.2 | |
| LM and/or multivessel | 8.7 | 13.5 | |
| Graft | 0.5 | 0.6 | |
| Number of diseased vessels, % | 0.14 | ||
| 1 | 43.3 | 40.0 | |
| 2 | 27.3 | 26.7 | |
| 3 and/or LM stenosis | 29.3 | 33.3 | |
| Final TIMI flow, % | 0.21 | ||
| 0 | 0.9 | 1.0 | |
| 1 | 0.8 | 0.6 | |
| 2 | 6.0 | 3.4 | |
| 3 | 92.3 | 95.0 | |
| Primary angioplasty success, % | 96.7 | 97.6 | 0.40 |
| 30-day mortality, % | 7.6 | 8.1 | 0.75 |
| 1-year mortality, % | 14.9 | 14.5 | 0.87 |
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
