Table 2.
Diagnostic value of published models to identify radiographic pneumonia in derivation (original) studies and in GRACE participants
| Study population (setting*) | Pneumonia prevalence (derivation study) | Linear predictor | Discrimination ROC area (95% CI) | Calibration‡ (P value) | |||
|---|---|---|---|---|---|---|---|
| Derivation study | Validation† | Validation in GRACE participants | |||||
| Diehr, 19842 | 1819 (ED) | 3% | −2 for rhinorrhoea, −1 for sore throat, +1 for night sweats, +1 for myalgia, +1 for sputum, +2 for respiratory rate >25, +2 for temperature >37.7°C | — | — | 0.67 (0.62 to 0.72) | 7816 (<0.001) |
| Singal, 198912 | 225 (ED) | 16% | –3.539, +0.884 for cough, +0.681 for fever, +0.464 for crackles, +0.030 for 5.0 (pretest probability of pneumonia§) | 0.75 (0.71 to 0.79) | 0.58 (0.45 to 0.70) | 0.68 (0.62 to 0.73) | 250 (<0.001) |
| Heckerling, 19906 | 1134 (ED) | 5% | –1.705, +0.494 for temperature >37.7°C, +0.428 for pulse >100 beats/min, +0.658 for rales, +0.638 for decreased breath sounds, +0.691 for absence of asthma | 0.82 (0.78 to 0.86) | 0.63 (0.50 to 0.75) | 0.65 (0.59 to 0.70) | 832 (<0.001) |
| Melbye, 19929 | 402 (OHD) | 5% | + 4.7 for fever (reported by patient) with duration of illness of >1 week, –4.5 for coryza, –2.1 for sore throat, +5.0 for dyspnoea, +8.2 for chest pain, lateral +0.9 for crackles | — | 0.49 (0.37 to 0.62) | 0.65 (0.60 to 0.70) | 1890 (<0.001) |
| Hopstaken, 20037 (signs and symptoms) | 246 (GP) | 13% | –2.74, +1.02 for dry cough, +1.78 for diarrhoea, +1.13 for temperature ≥38°C | 0.76 (NA) | 0.62 (0.50 to 0.75) | 0.55 (0.50 to 0.61) | 394 (<0.001) |
| Hopstaken, 20037 (signs, symptoms, CRP) | 246 (GP) | 13% | –4.15, +0.91 for dry cough, +1.01 for diarrhoea, +0.64 for temperature ≥38°C, +2.78 for C reactive protein ≥20 mg/L | 0.80 (NA) | 0.69 (0.58 to 0.80) | 0.71 (0.66 to 0.76) | 334 (<0.001) |
NA=not available.
*Setting: ED=emergency department, OHD=out of hours department, GP=general practice.
†In Graffelman.13
‡ Hosmer and Lemeshow test. After adjustment of intercept (to correct for difference in prevalence between derivation and validation cohorts), calibration remained poor (P<0.01).
§Pre-test probability of pneumonia was proportion of patients with pneumonia (5%) in our dataset.