The 2025 ATS focused update suggests prescribing empiric antibiotics for two virus-positive groups — outpatients with comorbidities and inpatients with nonsevere CAP — out of concern for bacterial-viral coinfection. IDSA agreed with 8 of the 10 recommendations but declined to endorse the guideline, disagreeing with 2 of the 4 antibiotic recommendations, both concerning patients with positive viral assays.1,2
Neither society has randomized or high-quality observational data to settle the question. Cited estimates of bacterial coinfection among hospitalized virus-positive patients span 3% to 39% — a range wide enough to underline the uncertainty. IDSA argues the upper estimates are inflated: bacterial testing is not systematic, sputum is collected preferentially from sicker patients with purulent secretions, and airway colonization is mistaken for infection. The pro-antibiotic case leans in part on 1918-pandemic autopsy series showing frequent bacterial coinfection — data limited by definition to patients who died.2
IDSA's position is that the harms of unnecessary antibiotics are established while the benefit is not: Clostridioides difficile colitis, arrhythmia, drug-drug interactions, allergic reactions, microbiome disruption, and selection for resistance. Antibiotics are the leading cause of medication-related emergency-department visits. For nonsevere illness it is safe to briefly withhold antibiotics to clarify the diagnosis; the default assumption that treating is the safer path, IDSA writes, is not warranted.2
The practical fault line is severity: treat the severe patient; individualize the nonsevere one, using rapid diagnostics and reliable follow-up to withhold-and-observe when the balance of features argues against coinfection.
Ascertainment drives the numerator. Coinfection rates are computed only among patients who were tested, and testing is not random. A patient with copious purulent sputum or one who is intubated is both more likely to have a sample sent and more likely to grow an organism, inflating the apparent rate. Colonization is not infection. Respiratory cultures grow upper-airway flora, especially in chronic lung disease, so a positive culture over-counts true coinfection. The result is a range — 3% to 39% — that reflects who was studied and how, not a single biological truth.
CAPE COD was a phase 3, double-blind RCT of adults admitted to the ICU for severe CAP, randomized to intravenous hydrocortisone (200 mg/day for 4 or 7 days by clinical response, then tapered; total 8 or 14 days) or placebo. It was stopped at the second interim analysis after 800 patients. Day-28 mortality was 6.2% with hydrocortisone versus 11.9% with placebo (absolute difference -5.6 percentage points, 95% CI -9.6 to -1.7; P=0.006). Hydrocortisone also reduced intubation and vasopressor initiation (hazard ratio 0.59 for each).3
SONIA (2025) tested low-dose oral glucocorticoids for 10 days versus standard care in 2180 adults hospitalized with CAP across 18 non-ICU public hospitals in Kenya. Thirty-day mortality was 22.6% versus 26.0% (hazard ratio 0.84, 95% CI 0.73 to 0.97; P=0.02). The signal held in a population markedly different from CAPE COD — younger (median age 53 vs 67), sicker overall (mortality 24.3% vs 9.1%), and with more immunosuppression (HIV in roughly 16%) — though the effect size was smaller (HR 0.84 vs about 0.53).3,4
Torres 2015 randomized 120 patients with severe CAP and a high inflammatory response (C-reactive protein above 150 mg/L) to methylprednisolone or placebo for 5 days. Treatment failure fell from 31% to 13% (odds ratio 0.34, 95% CI 0.14 to 0.87; P=0.02); in-hospital mortality did not differ. The patients most likely to benefit are those with the highest inflammatory response.5
CAPE COD was halted at its second interim analysis once a mortality difference emerged. Stopping early for benefit is ethical, but it tends to overestimate the true effect: a random high point is more likely to cross a stopping boundary than a low one, so the published effect size is, on average, larger than it would have been had the trial run to its planned end. It is a reason to read a single stopped-early result alongside the wider body of trials — which, here, point the same direction.
Uranga randomized 312 hospitalized CAP patients at day 5 to a stability-guided stop — a minimum of 5 days, discontinued once the patient was afebrile (37.8 C or less) for 48 hours with no more than one sign of clinical instability — versus a physician-determined, longer duration. Clinical success at day 30 was 91.9% versus 88.6% (P=0.33), with equivalent symptom scores: the short, stability-guided course was non-inferior while cutting antibiotic exposure.7
The 2025 ATS update carries this forward, suggesting less than 5 days (minimum 3) of antibiotics once clinical stability is reached, rather than 5 or more days. The operative guardrail in every short-course trial is clinical stability — resolution of fever and hemodynamic or respiratory instability — not a fixed calendar.1
A non-inferiority trial like Uranga does not ask whether the short course is better — it asks whether it is not unacceptably worse by a pre-specified margin. You read it from the confidence interval of the difference: if the entire interval stays on the acceptable side of the margin, non-inferiority is met, even when the P value for superiority is non-significant. A non-significant superiority test (here P=0.33) is therefore not a failure — it is the expected result when two strategies are genuinely equivalent.
PEN-FAST is a point-of-care clinical decision rule derived and validated against formal allergy testing (derivation cohort 622 patients; external validation 945). Four features carry the score: an allergy event five or fewer years ago (2 points), anaphylaxis or angioedema or a severe cutaneous adverse reaction (2 points), and treatment required for the reaction (1 point), for a maximum of 5. A score below 3 identifies low risk: only 3.7% (17 of 460) of these patients had a positive allergy test, a negative predictive value of 96.3% (95% CI 94.1 to 97.8%).8
The large majority of reported penicillin allergies are disproved on formal evaluation. A retained label pushes patients to second-line agents, which raises surgical-site-infection and treatment-failure risk and drives resistance. A low-risk PEN-FAST history supports a direct oral challenge without prior skin testing — the shift that lets primary teams, not only allergists, begin delabeling.
An IgE penicillin allergy — even anaphylaxis — is specific to the culprit drug's side chain, not the shared beta-lactam ring. Cephalosporins with distinct side chains are therefore safe in penicillin allergy; only a severe non-IgE reaction (SJS/TEN, DRESS) to a beta-lactam calls for avoiding the class entirely. This is the reasoning behind offering a cephalosporin, rather than reflexively a fluoroquinolone, to the penicillin-allergic patient.
A 96.3% negative predictive value means that among low-risk histories, about 96 of 100 truly have no allergy. But NPV is not a fixed property of the rule — it rises as true allergy becomes rarer in the tested population. Because genuine IgE penicillin allergy is uncommon (most labels are false), a low-risk PEN-FAST score is highly reassuring here; the same score would carry a lower NPV in a population enriched for real reactions. Read NPV as a statement about this population, not a universal constant.