Table 5. Any adverse outcome or unplanned readmission model performance with sequential addition of variable groups and measures of prediction increment.
| Variable | No. of data elements | C−statistic (%) | Event NRI (>0) (%) | Nonevent NRI (>0) (%) | Combined NRI (>0) | P value, HL test |
|---|---|---|---|---|---|---|
| Severe adverse event-models§ (derivation cohort) | ||||||
| Age | 1 | 0.54 | – | – | – | 0.87 |
| + Gender | 2 | 0.55 | 15 | −11 | 0.04 | 0.29 |
| + ASA class 3/4 | 3 | 0.58 | 23 | 6 | 0.29 | 0.79 |
| + Intraoperative variables | 4 | 0.63 | 1 | 30 | 0.31 | 0.12 |
| + Laboratory values | 12 | 0.66 | −26 | 54 | 0.28 | 0.97 |
| + Comorbid conditions (full model) | 16 | 0.66 | −33 | 53 | 0.20 | 0.78 |
| Severe adverse event-models§ (validation cohort) | ||||||
| Age | 1 | 0.57 | – | – | – | 0.06 |
| + Gender | 2 | 0.58 | 12 | −8 | 0.04 | 0.20 |
| + ASA class 3/4 | 3 | 0.61 | 32 | 1 | 0.33 | 0.24 |
| + Intraoperative variables | 4 | 0.65 | 2 | 29 | 0.31 | 0.20 |
| + Laboratory values | 12 | 0.68 | −29 | 59 | 0.30 | 0.06 |
| + Comorbid conditions (full model) | 16 | 0.69 | −24 | 53 | 0.29 | 0.62 |
| Unplanned readmission-models§ (derivation cohort) | ||||||
| Age | 1 | 0.54 | – | – | – | 0.86 |
| + Gender | – | – | – | – | – | – |
| + ASA class 3/4 | 2 | 0.57 | 17 | 5 | 0.22 | 0.85 |
| + Intraoperative variables | 3 | 0.58 | −12 | 23 | 0.11 | 0.06 |
| + Laboratory values | 9 | 0.61 | −29 | 52 | 0.23 | 0.98 |
| + Comorbid conditions (full model) | 13 | 0.62 | −29 | 43 | 0.14 | 0.77 |
| Unplanned readmission-models§ (validation cohort) | ||||||
| Age | 1 | 0.59 | – | – | – | 0.10 |
| + Gender | – | – | – | – | – | – |
| + ASA class 3/4 | 2 | 0.62 | 33 | 0 | 0.33 | 0.85 |
| + Intraoperative variables | 3 | 0.62 | −10 | 20 | 0.10 | 0.49 |
| + Laboratory values | 9 | 0.64 | −38 | 58 | 0.20 | 0.30 |
| + Comorbid conditions (full model) | 13 | 0.65 | −47 | 68 | 0.21 | 0.58 |
§, as groups of variables were sequentially added. A P value of >0.05 for the Hosmer-Lemeshow goodness-of-fit test demonstrates good model calibration; a P value of <0.05 for the Hosemer-Lemeshow goodness-of-fit test suggests a poor model fit.