External validation of a risk score in the prediction of the mortality after esophagectomy for cancer.

ABSTRACT

This study was designed as an external evaluation of the Steyerberg score in the prediction of different categories of postoperative mortality after esophagectomy on a large nationwide database of thoracic surgeons. Data collection was obtained from the Epithor national database encompassing the majority of thoracic procedures performed in France. We retrospectively compared the predicted to the observed postoperative 30-day (30DM), 90-day (90DM) and in-hospital mortality (IHM) rate in each decile of equal patient. Patients included in the study were operated for an esophageal cancer and Gastroesophageal junction (GEJ). Steyerberg score was determined according to its logarithmic formula obtained from a sum score including age, comorbidities, neoadjuvant treatment and hospital volume. Deviation of observed from theoretically expected number of deaths was investigated using the calibration test of Hosmer-Lemeshow. Discrimination of the score was determined using the measure of the area under the receiver operating characteristic curve (AUC) of each category of mortality. Over a 9-year period, 1039 consecutive patients underwent an esophagectomy over 42 centers. Among them, 18 centers were considered as intermediate or high-volume institutions, and 24 were low-volume institutions. There were 841 males (81%) with a mean age of 62.3 ± 10 years. Preoperative treatment was allocated to 420 patients (40%). Numbers of comorbidity was 1 in 261 patients (25%), 2 in 264 patients (25%), 3 in 383 patients (36%) and 4 in 5 patients (1%). The 30DM, 90DM and IHM rate were, respectively, 5.6%, 9.2% and 9.6%. The main causes of postoperative deaths were related to pulmonary complications (44%), complications of the gastric interposition (28%), cardiologic and thromboembolism events (10%). For 30DM, there were significant differences between predicted/observed mortalities in four deciles, whereas there was no significant difference for 90DM and for IHM. In term of calibration, there was a fair agreement of the Steyerberg score with observed 30DM. Predictions were above 20% for seven deciles. Calibration seemed more adequate for 90DM and for IHM. Predictions were above 20% for only three deciles but deviations were not significant. In terms of discrimination, for the 30DM the Steyerberg score overpredicted, the observed mortality rate and AUC was 0.64 (CI 95% 0.57-0.71). For the 90DM, AUC indicated 0.63 (CI 95% 0.57-0.68). For the IHM, AUC indicated 0.63 (CI 95% 0.58-0.68). Steyerberg scoring system seems to be a moderate risk score of the prediction of the IHM and 90DM. This score appears to have a fair discrimination for the 30DM. Nevertheless, because of its simplicity, we believe that this simple predictive score is relevant and transportable to others institution performing such surgery for benchmarking purposes. A reappraisal of the score adapted to current surgical cohort is required.