Table 3:

Performance of the basic and extended (with CRP measurement) prediction models in the diagnostic risk classification of pneumonia in a hypothetical cohort of 1000 patients*

Predicted risk of pneumoniaObserved pneumoniaTotal
n = 1000
Yes
n = 130
No
n = 870
Basic model
Low (< 2.5%)4244248
Intermediate (2.5%–20%)44513557
High (> 20%)82113195
Extended model
Low (< 2.5%)4313317
Intermediate (2.5%–20%)35470505
High (> 20%)9187178
  • Note: CRP = C-reactive protein.

  • * Numbers are based on the median prevalence of community-acquired pneumonia of 13% across studies and the pooled sensitivities and specificities of the basic and extended models at low (2.5%) and high (20%) risk thresholds. At the low risk threshold, the pooled sensitivity was 0.97 (95% CI 0.95–0.98) for the basic and extended models; the pooled specificities were 0.28 (95% CI 0.27–0.29) and 0.36 (95% CI 0.34–0.37), respectively. At the high risk threshold, the pooled sensitivities were 0.63 (95% CI 0.59–0.66) for the basic model and 0.70 (95% CI 0.66–0.73) for the extended model; the pooled specificities were 0.87 (95% CI 0.86–0.88) and 0.90 (95% CI 0.89–0.91), respectively. To calculate the number of patients with observed pneumonia who will be classified as low risk, we used the formula: (1 − sensitivity) × prevalence × 1000; for the number without pneumonia who will be correctly classified as low risk, we used the formula: specificity × (1 − prevalence) × 1000. For example, at the low risk threshold, (1 − 0.97) × 0.13 × 1000 = 4 patients with observed pneumonia will get a predicted risk of < 2.5% using the basic model; 0.28 × (1 − 0.13) × 1000 = 244 patients without pneumonia will get a predicted risk of < 2.5% using the basic model.