The CRP-for-cough paper and the stewardship committee
A colleague forwards a paper: a prospective single-centre diagnostic accuracy study of point-of-care C-reactive protein in 312 adults presenting with acute cough of less than three weeks. The reference standard was pneumonia on chest radiograph, reported by a radiologist blinded to the CRP result; prevalence was 6%. At a cut-off of 20 mg/L, CRP had a sensitivity of 82% and a specificity of 68%. The authors conclude that CRP is "a useful rule-out test for pneumonia" and recommend it before antibiotics. Your practice sees about 600 adults with acute cough each year, and the device costs the equivalent of two consultation fees per test. A colleague argues that because the negative predictive value is high, a negative test safely avoids antibiotics, and a positive test justifies them. The antimicrobial stewardship committee has asked you to advise on whether to adopt CRP testing.
1. Reconstruct the 2×2 table and calculate the predictive values and likelihood ratios. What do they mean for your patients?
Read the worked answer
With 312 participants and 6% prevalence, there are about 19 cases of radiographically confirmed pneumonia and 293 without.
| Pneumonia present | Pneumonia absent | Total | |
|---|---|---|---|
| CRP ≥20 mg/L | 16 | 94 | 110 |
| CRP <20 mg/L | 3 | 199 | 202 |
| Total | 19 | 293 | 312 |
PPV = 16/110 = 14.5%. NPV = 199/202 = 98.5%. Positive likelihood ratio = 0.82/0.32 = 2.56; negative likelihood ratio = 0.18/0.68 = 0.26.
The NPV looks impressive but is mostly a product of the low prevalence. The clinically useful way to read it is as a shift in probability: a negative test moves a 6% pre-test probability to about 1.7%, which is a genuine and useful rule-out. A positive test moves 6% only to about 14%. That is below any sensible threshold for prescribing antibiotics, so the colleague's second claim is wrong: a positive CRP does not justify antibiotics, and treating on it would mean prescribing to roughly seven patients for every one with pneumonia.
An LR+ of 2.56 is weak. Tests that rule in disease generally need an LR+ above 5–10. CRP here is a rule-out test, and only in a population whose pre-test probability resembles 6%.
2. Appraise the study. What would you want to know before believing the authors' conclusion?
Read the worked answer
Several things are missing. Was the chest radiograph performed on all 312 patients, or only on those in whom the clinician suspected pneumonia? If the latter, there is verification bias and both sensitivity and specificity are distorted. Was the CRP measured and interpreted blind to the radiograph, and was the device calibrated and quality-controlled? A single centre with a single reading radiologist does not establish reproducibility; inter-observer agreement for pneumonia on plain film is moderate at best, so the reference standard is itself imperfect.
Was the 20 mg/L cut-off pre-specified, or chosen after looking at the data? Was it the only cut-off tested? Were confidence intervals reported around 82% and 68%, and around the predictive values? With only 19 cases, the sensitivity estimate is fragile - a handful of patients moves it substantially.
Spectrum matters: a study recruiting adults with cough in a single centre may have excluded the very patients in whom the test is hardest - the frail elderly, those with heart failure or COPD, those already on antibiotics.
Most importantly, diagnostic accuracy is a surrogate. The question the stewardship committee actually cares about is whether using CRP reduces antibiotic prescribing without increasing hospitalisation or death. That requires a randomised or cluster-randomised implementation trial with patient-important outcomes. Accuracy studies cannot answer it, and the authors' jump from accuracy to a prescribing recommendation is not supported by their data.
3. Would you adopt CRP testing in your practice? Set out your reasoning with numbers.
Read the worked answer
Not on this evidence, and not at that price without a local implementation study. The decision turns on pre-test probability.
If your case mix resembles the study - about 6% pneumonia - a negative test takes probability to roughly 1.7%, so a negative result can support withholding antibiotics in a patient who otherwise looks low risk. But clinical assessment plus a validated score such as CRB-65 already achieves a similar rule-out in low-risk patients at no cost. The marginal gain is small, and the 3 false negatives in 312 would translate to roughly 6 missed pneumonias per 600 patients if the same accuracy held.
If your practice sees a lower-risk population, say 2% prevalence, the PPV falls to about 5% - a positive test becomes almost uninterpretable. If you saw a higher-risk population at 15%, the PPV rises to about 33% and the test becomes more useful. So the test's value is population-dependent, and blanket adoption across a practice with mixed case mix is unjustified.
Cost: 600 tests a year at two consultation fees buys 600 results, of which about 110 would be positive in a 6%-prevalence population and only about 16 would have pneumonia. The money is better spent on clinical training in pneumonia risk scores and on audit of antibiotic prescribing.
My advice to the committee: do not adopt now. Offer to run a six-month implementation study with antibiotic prescribing and 30-day hospitalisation as outcomes, and revisit. If CRP is used at all, restrict it to patients where clinical assessment is genuinely equivocal.