Validation of the society of thoracic surgeons predicted risk of mortality score for long-term survival after cardiac surgery in Israel

This is a retrospective cohort analysis of prospectively collected clinical data on 1279 consecutive patients who underwent five cardiac surgical procedures with a calculable STS PROM at Hadassah Hebrew University Medical Center from January of 2008 to December of 2015. The procedures included isolated coronary bypass graft grafting (CABG), aortic valve replacement (AVR) with or without CABG and mitral valve replacement or repair (MVR, r) with or without CABG.

The operations were performed by a group of five surgeons via a median sternotomy, mostly (90%) using near-normothermic (34–35 °C) cardiopulmonary bypass and cardioplegic arrest. Cardiopulmonary bypass was established using ascending aortic cannula, and either a single two-stage venous cannula via the right atrial appendage, or two single-stage bi-caval cannulas for mitral valve operations. Myocardial protection was achieved using intermittent cold antegrade and retrograde blood cardioplegia. One hundred twenty-eight operations (16%) out of 798 isolated coronary artery bypass grafting were performed using standard off-pump techniques. Operative techniques remained uniform throughout the study period. Hadassah is a tertiary academic center with a training program. Two hundred eighty-one (22%) operations were performed by residents under direct supervision of a senior faculty. All faculty had gone through varied periods of post-residency advanced training in the US, partially adopting the American operative techniques and practice patterns.

Data were prospectively collected and entered into our STS-linked departmental database using the STS collection tool and definitions [11]. The collected data included patient demographics, risk factors, medications, extent of cardiac disease, clinical presentation, procedural details and 30-day outcomes [11]. The data were entered into the database by a specifically trained departmental database manager. To avoid bias and gaming, procedural outcomes were collected by a professional who was not affected by the results in any way. Periodical auditing against patients` medical records was routinely performed to ensure data accuracy and completeness. The STS ACSDB collection tool and risk-algorithms are updated periodically. To ensure that the calculated STS PROM for each patient in our study is accurate, we used three versions of the STS ACSDB collection tool – 2.61, 2.73 and 2.81 and the most updated risk-algorithm during the study period. The variables included in the STS PROM include the type of procedure, age, gender, height, weight, left ventricular ejection fraction, heart failure, race, renal failure, cardiac presentation, cardiac symptoms, prior myocardial infarction, cardiac arrhythmia, renal failure, peripheral vascular disease, cerebrovascular disease, peripheral arterial disease, diabetes mellitus, hypertension, immunocompromised state, endocarditis, endocarditis, coronary artery anatomy, status of surgery, cardiogenic shock, New York Heart Association functional class, intra-aortic balloon pump, inotropes, previous cardiac intervention, type and severity of aortic, mitral and tricuspid valve dysfunction and incidence of surgery (re-operation) [11]. The primary end-point of the study was all-cause mortality from the index operation to death, including perioperative deaths. Since the STS ACSDB collects 30-day mortality only, we obtained long-term survival data by linking to the National Israeli Social Security Mortality Database. We used three unique identifiers for cross-reference—the patient`s name, age and the national Identification Number that is equivalent to the social security number in the US. In Israel, the Identification Number is used as the patient`s medical record number in any medical institution throughout the patient life time.

Categorical variables were expressed as absolute numbers and percentages. Continuous variables were expressed as means ± SD, 95% Confidence Intervals (CI), or medians with interquartile range (IQR) as appropriate. To evaluate the ability of the PROM to predict long-term survival we categorized the patients into five sub-groups according to the STS PROM—A: 0–0.99%, B: 1.0–1.99%, C: 2.0–2.99%, D: 3.0–4.99% and E: ≥ 5.0%. The calculated STS PROM was less than 5.0% in 1114 patient—89% of the cohort. To allow a meaningful comparative analysis, we based our categorization upon calculation of the median of each range of scores and the percent of patients in each score. We used the Kaplan Meier method to plot actuarial survival for each risk sub-group, and the non-parametric Wilcoxon signed-rank test to produce 95% CI and compare survival among sub-groups. To assess the discriminative power of the STS PROM risk model, we plotted the Receiver Operating Characteristics curve (ROC) and calculated the area under the curve (AUC). Model discrimination indicates the ability of the model to distinguish between survivors and non-survivors. The values of the area under the curve range from 0.5 to 1.0. Higher values indicate better model discrimination, whereas values close to 0.5 indicate that model discrimination is random. Cox regression analysis was performed to identify predictors of long-term mortality. We first performed univariable analysis of mortality using Chi square test and the Mann–Whitney test, as appropriate. Variables with a P value of ≤ 0.1 were entered into the model. The variables included in the final model were age, gender, diabetes mellitus, renal failure, dialysis, left ventricular ejection fraction, type of operation, reoperation, priority of the operation (elective, urgent, emergent, emergency salvage) and the STS PROM. We used the Spearman correlation analysis to confirm that there was no co-linearity or interactions among the covariates entered into the final model. Data were complete for most of the covariates entered into the Cox regression analysis. Missing data included: diabetes mellitus—75 patients (5.9%), left ventricular ejection fraction—22 patients (1.7%) and priority of surgery—3 patients (0.2%). A P value of < 0.05 was considered significant. The statistical analyses were performed using the SPSS version 24 for Windows (SPSS Inc., Chicago, IL, USA).

The study cohort consisted of 1279 consecutive patients who underwent cardiac surgery at our institution between January of 2008 and December of 2015. The five surgical procedures with a calculable STS PROM that were included in the study comprised 78% of the adult cardiac operative volume in our institution during the study period. The baseline clinical and operative profiles of the study cohort are summarized in Table 1. The observed 30-day mortality was 1.9% (25 patients). The STS PROM for the entire cohort was 3.1 ± 5.9% (median 1.5%, range 0.1–96%, IQR 0.7–3.2%).

Complete follow-up was achieved for 1256 (98%) patients over a mean period of 62 ± 28 months (range 0–107 months, median 64, IQR 41–86). Mean survival of the entire cohort was 95 ± 1 months (95% CI 93–96). Higher STS PROM was associated with reduced survival: A—104 ± 1 (CI 103–106) months, B—96 ± 2 (CI 93–99) months, C—93 ± 3 (CI 88–98) months, D—89 ± 3 (CI 84–95) months, E—74 ± 3 (CI 68–81) months (Table 2, Fig. 1). Incremental increase in the STS PROM was associated with a significant decrease in survival (Wilcoxon signed-rank test, p < 0.0001 comparing each subgroup against A or E). Survival among groups B, C and D was similar with overlapping confidence intervals.

To assess model discrimination of the STS PROM for overall survival we plotted the ROC and measured AUC (Fig. 2). The AUC for the STS PROM model was 0.76 ± 0.02 indicating excellent model discrimination for overall survival.

In a cox regression analysis, advanced age, lower ejection fraction, reoperation, diabetes mellitus, preoperative dialysis and STS PROM were identified as independent predictors of long-term mortality (Table 3). Higher STS PROM was associated with a significant increase in the hazard ratio for long-term mortality.