Table 2.
Mortality rates and goodness-of-fit to data of multiple high volume threshold models†.
| Volume Threshold | Mortality Rate at Centers with Volume Less than Volume Threshold (%) | Mortality Rate at Centers with Volume Equal to or Greater than Volume Threshold (%) | P-value | McFadden’s Pseudo r2 (shown as %) | AUC |
|---|---|---|---|---|---|
| Baseline | - | - | - | 3.23 | 0.6354 |
| 1 | - | 9.94% | - | 3.31 | 0.6364 |
| 2 | 13.31% | 8.63% | <0.001 | 3.43 | 0.6431 |
| 3 | 12.38% | 8.03% | <0.001 | 3.46 | 0.6430 |
| 4 | 11.54% | 7.71% | <0.001 | 3.35 | 0.6392 |
| 5 | 11.31% | 7.36% | <0.001 | 3.44 | 0.6414 |
| 6 | 11.04% | 7.09% | <0.001 | 3.43 | 0.6412 |
| 7 | 10.96% | 6.54% | <0.001 | 3.52 | 0.6425 |
| 8 | 10.86% | 6.35% | <0.001 | 3.56 | 0.6418 |
| 9 | 10.68% | 5.98% | <0.001 | 3.57 | 0.6435 |
| 10 | 10.58% | 5.37% | <0.001 | 3.73 | 0.6440 |
| 11 | 10.50% | 5.22% | <0.001 | 3.73 | 0.6440 |
| 12 | 10.35% | 5.31% | <0.001 | 3.79 | 0.6449 |
| 13* | 10.26% | 5.39% | <0.001 | 3.80 | 0.6450 |
| 14 | 10.22% | 5.50% | <0.001 | 3.76 | 0.6444 |
| 15 | 10.16% | 5.30% | <0.001 | 3.87 | 0.6465 |
| 16 | 10.14% | 5.26% | <0.001 | 3.87 | 0.6461 |
| 17 | 10.08% | 5.39% | 0.001 | 3.76 | 0.6441 |
| 18 | 10.01% | 5.57% | 0.002 | 3.67 | 0.6425 |
| 19 | 9.97% | 5.57% | 0.003 | 3.69 | 0.6428 |
| 20 | 9.95% | 5.10% | 0.002 | 3.64 | 0.6417 |
| 21 | 9.92% | 5.14% | 0.003 | 3.64 | 0.6417 |
| 23 | 9.92% | 4.87% | 0.002 | 3.69 | 0.6427 |
| 24 | 9.91% | 4.60% | 0.002 | 3.71 | 0.6437 |
| 25 | 9.83% | 4.33% | 0.004 | 3.70 | 0.6449 |
| 26 | 9.75% | 4.43% | 0.012 | 3.60 | 0.6430 |
| 27 | 9.71% | 4.52% | 0.021 | 3.50 | 0.6408 |
| 28 | 9.67% | 4.67% | 0.040 | 3.50 | 0.6408 |
| 29 | 9.63% | 4.92% | 0.080 | 3.40 | 0.6387 |
| 30 | 9.61% | 1.56% | 0.029 | 3.51 | 0.6390 |
| 34 | 9.54% | 2.94% | 0.191 | 3.31 | 0.6359 |
Multivariable logistic regression model adjusted for patient age, gender, race and Charlson Index of comorbidities, procedure year and hospital teaching status, and a sequentially changing independent variable of dichotomized annual hospital resection volume.
indicates the definition of “high volume” as suggested by the Leapfrog Group(5).
Table showing the relationship of different high volume thresholds to mortality rates and the goodness-of-fit [McFadden’s pseudo r2 and the Area Under the Curve (AUC)] of each model. An example of interpretation of the values for a volume threshold of 15 would signify that, on average, hospitals which perform greater than or equal to 15 esophageal resections per year have a mortality rate of 5.30%, compared to 10.16% at hospitals which perform less than 15 resections per year. The P-value for the χ2 statistic of these mortality rates is <0.001, and this model has a pseudo r2 of 3.87% and AUC of 0.6465.