Table 1. Characteristics of clinical prediction models.
| Clinical prediction rule | Setting | Population | Original reference standard | Prevalence pneumonia | Statistical model | Predictor variables | Performance | Level of evidence** | QUADAS-2 | |||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Risk classification (high versus low risk) | Sensitivity | Specificity | LR+ | LR- | risk of bias / concern applicability | |||||||
| 1. Mahabee (2005)[23] | US | 2m - 5y, cough + 1 of following: labored/ rapid/noisy breathing; chest/abdominal pain; fever | radiographic pneumonia | 44/510 (8.6) | MLRM | age≥12 months, respiratory rate ≥50/min, oxygen saturation ≤96%, nasal flaring in age <12months | 63.6 | 77 | 2.8 | 0.5 | 1 | low / low |
| 2. Bruel, van den (2007)[20] | BE* | < 17y, acute illness | hospital admission for radiographic pneumonia | 15/3981 (0.4) | CART | dyspnea, 'something is wrong' | 93.8 | 93.2 | 13.9 | 0.07 | 3 | low / high |
| Verbakel (validation 1, 2013)[26] | NL | " | " | 17/506 (3.3) | 94.1 | 44.6 | 1.7 | 0.13 | NA, different datasets | |||
| Verbakel (validation 2, 2013) | UK | " | " | 131/2687 (4.9) | 92.4 | 41.4 | 1.58 | 0.18 | ||||
| Verbakel (validation 3, 2013) | NL | " | " | 114/1750 (6.5) | 65.8 | 43.1 | 1.16 | 0.79 | ||||
| Verbakel (validation 4, 2013) | NL | " | " | 54/595 (9.1) | 81.5 | 45.5 | 1.49 | 0.41 | ||||
| Verbakel (validation 5, 2013) | UK | " | " | 67/700 (9.6) | 26.9 | 89.1 | 2.46 | 0.82 | ||||
| 3. Neuman (2011)[21] | US | < 21, chest X-ray for suspected pneumonia | radiographic pneumonia | 422/2574 (16.4) | CART | oxygen saturation ≤92%, history of fever, wheezing, focal rales, chest pain, focal decreased breath sounds | 90.1 | 21.6 | 1.2 | 0.4 | 1 | some / low |
| Probability (predicted risk in %) | AUC | |||||||||||
| 4. Lynch (2004)[19] | US | 1-16y, chest X-ray for suspected pneumonia | radiographic pneumonia | 204/570 (35.8) | MLRM | fever, decreased breath sounds, crackles, tachypnea | 0.67 | 3 | some / low | |||
| Bilkis (validation, 2010)[27] | US | " | 179/257 (69.6) | 0.7 | some / some | |||||||
| 5. Oostenbrink (2013)[24] | NL | 1m - 16y, fever and cough | nodular infiltration or consolidation on radiograph / rule out pneumonia by noneventful followup / consensus | 78/504 (15.5) | MLRM | ill appearance, tachypnea, O2 <94%, CRP | 0.79 | 3 | some / low | |||
| Oostenbrink (validation 1, 2013) | NL | " | 58/420 (13.8) | 0.81 | ||||||||
| Oostenbrink (validation 2, 2013) | NL | " | 27/366 (7.4) | 0.86 | ||||||||
| 6. Craig (2010)[22] | AU | <5y, fever | consolidation on radiograph | 533/15781 (3.4) | MLRM | general appearance, cough, temperature, breathing difficulty, abnormal chest sounds, chronic disease, capillary refill time, urinary symptoms, elevated respiratory rate, crackles, pneumococcal vaccine status, elevated heart rate, felt hot, meningococcal vaccine state, infectious contacts, crying, fluid intake, respiratory symptoms, diarrhoea, bulging fontanelle, male sex, focal bacterial infection, abnormal ear/nose/throat signs, age, rash, stridor, wheeze | 0.84 | 2 | low / low | |||
| Craig (validation, 2010) | AU | " | 193/5584 (3.5) | 0.84 | low / low | |||||||
| 7. Nijman (2013)[17] | NL | 1m - 15y, fever | nodular infiltration or consolidation on radiograph; rule out pneumonia by noneventful followup | 171/2717 (6.3) | MLRM | age, sex, duration of fever, temperature, respiratory rate, heart rate, oxygen saturation, capillary refill, retractions, ill appearance, CRP | 0.81 | 4 | low / low | |||
| Nijman (validation, 2013) | NL | " | 59/487 (12.1) | 0.81 | low / low | |||||||
| De Vos (validation, 2015) [12] | NL | " | 33/439 (7.5) | 0.83 | low / low | |||||||
| 8. Irwin (2017)[25] | US | <16y, (history of) fever | respiratory symptoms, signs and focal consolidation on radiograph | 63/532 (12) | MLRM | CRP, respiratory rate, normal air entry, resistine, procalcitonin | 0.84 | 1 | low / low | |||
m = months, y = years, ED = emergency department, GP = general practice, US = United States of America, BE = Belgium, NL = the Netherlands, AU = Australia, UK = United Kingdom
CART = classification and regression tree, MLRM = multivariable linear regression model, LR+ = positive likelihood ratio, LR- = negative likelihood ratio, AUC = area under the receiver operating curve
aderived in general practice and emergency department, validated in ED
bas described by Reilly (range 1 (only derived) to 5 (proven by broad impact analysis)[11]