Table 3.
Logistic regression prediction models of health-care utilization for respiratory problems
| Derivation Cohort* |
Internal Validation* |
|||||
|---|---|---|---|---|---|---|
| Hospitalization | Emergency Department Visit | Two or More Outpatient Visits | Hospitalization | Emergency Department Visit | Two or More Outpatient Visits | |
| Model 1: Demographics, comorbidities, & tobacco history† | ||||||
| Model p value‡ | .018 | .63 | .46 | .046 | .051 | .0052 |
| C-index** | 80% | 64% | 59% | 79% | 73% | 70% |
| Model 2: Model 1 + COPD Severity Score | ||||||
| Odds ratio for COPD Severity Score§ | 1.66 (p<.001) | 1.77 (p<.001) | 1.91 (p<.001) | 1.99 (p=.001) | 1.54 (p=.001) | 1.67 (p<.001) |
| Model p value‡ | <.0001 | .0005 | <.0001 | .0003 | .0012 | <.0001 |
| C-index** | 87% | 82% | 77% | 91% | 82% | 78% |
| p-value for C-index difference between Model 1 & Model 2¶ | .03 | .0031 | <.0001 | .048 | .021 | .029 |
Derivation cohort used baseline data from 2002 to predict outcomes reported one year later in 2003. Internal validation used baseline data from 2003 to predict outcomes reported one year later in 2004
Covariates in Model 1 consist of age, race, educational attainment, tobacco history, and medical comorbidities (heart failure, coronary artery disease, diabetes, and sleep apnea). Covariates chosen to maximize C-index of Model 1.
Based on likelihood ratio test of global null hypothesis. Model 1 has 8 df. Model 2 has 9 df.
C-index is area under the receiver operator characteristics curve
Odds ratios expressed per ½ standard deviation increase in COPD Severity Score (p-value)
p-value from χ2 comparing C-index