For two decades the cefazolin-versus-antistaphylococcal-penicillin question rested on observational data. It now has a randomized answer. SNAP — the Staphylococcus aureus Network Adaptive Platform trial — randomized 1341 adults with penicillin-resistant MSSA bacteremia to cefazolin 2 g q8h or flucloxacillin/cloxacillin. 90-day mortality was 15.0% vs 17.0% (adjusted OR 0.81, 95% credible interval 0.59–1.12; probability of non-inferiority 99.2%). Acute kidney injury within 14 days was lower with cefazolin (13.9% vs 19.6%; adjusted OR 0.67, 95% CrI 0.50–0.89), as were serious adverse reactions and drug discontinuations.1
A second RCT agrees. CloCeBa randomized 292 patients (ITT) to cefazolin vs cloxacillin and — unlike SNAP — did not exclude endocarditis. The composite success endpoint was 75% vs 74% (difference −1%, 95% CI −11 to 9; non-inferiority margin 12%), and day-7 AKI was again far less with cefazolin (1% vs 12%).2 The 2025 meta-analysis of 30 observational studies had already pointed the same way — 30-day mortality OR 0.73 (95% CI 0.62–0.85) and nephrotoxicity OR 0.30 (0.20–0.46) — though at low certainty from confounding by severity.3
| Study | Design / N | Comparator | Primary result | Safety |
|---|---|---|---|---|
| SNAP1 | RCT, non-inferiority; 1341 | Cefazolin vs (flu)cloxacillin | 90-d mortality 15.0% vs 17.0% (aOR 0.81, 0.59–1.12); NI met | AKI 13.9% vs 19.6% (aOR 0.67, 0.50–0.89) |
| CloCeBa2 | RCT, non-inferiority; 292 (IE allowed) | Cefazolin vs cloxacillin | Composite success 75% vs 74% (diff −1%, −11 to 9); NI met | AKI d7 1% vs 12% |
| Prosty meta3 | 30 observational studies | Cefazolin vs ASP | 30-d mortality OR 0.73 (0.62–0.85) | Nephrotoxicity OR 0.30 (0.20–0.46) |
The live hedge is the cefazolin inoculum effect: some MSSA strains hyperproduce type-A β-lactamase that hydrolyzes cefazolin at the high bacterial burden of endocarditis or a large undrained focus, which can produce microbiologic failure. SNAP's isolates were not tested for it, and CloCeBa's authors report that non-inferiority was not shown for most outcomes in the blaZ type-A subgroup.1,2 This is why an antistaphylococcal penicillin (or nafcillin) is still preferred by many for high-inoculum endovascular MSSA, and why cefazolin remains a poor choice for CNS infection, where it penetrates poorly.4
A non-inferiority trial asks whether the new agent is not worse than the standard by more than a pre-set margin (Δ). You read it off the confidence interval: if the whole CI for the difference sits on the acceptable side of Δ, non-inferiority is met. SNAP set its margin at an adjusted OR of 1.2, chosen to approximate an absolute mortality difference under about 2.5 percentage points if control mortality is around 15%. The margin is only as meaningful as that assumed control event rate — if the comparator does better than expected, a fixed absolute margin becomes proportionally more permissive.
SNAP went further: the probability of superiority for cefazolin on mortality was 89.8%, and it was decisively better on kidney injury. So the modern read is not merely 'cefazolin is acceptable' but 'cefazolin is the default for most MSSA, with the inoculum-effect exceptions above.'
The 2020 consensus guideline moved MRSA vancomycin monitoring from trough-based (15–20 mg/L) to AUC-guided dosing, target AUC/MIC 400–600, ideally by Bayesian estimation. The driver was toxicity: troughs of 15–20 overshoot the exposure needed for efficacy and track with nephrotoxicity, so AUC targeting preserves efficacy while cutting kidney injury.5 Switch agents when the vancomycin MIC is ≥2 mg/L — above that, AUC/MIC ≥400 is often unachievable at tolerable doses.
Daptomycin entered SAB on Fowler 2006: 6 mg/kg was non-inferior to standard therapy (success 44.2% vs 41.7%; absolute difference 2.4%, 95% CI −10.2 to 15.1). Two signals shaped current practice. First, microbiologic failure was numerically higher with daptomycin, and 6 of 19 daptomycin failures grew isolates with reduced daptomycin susceptibility on therapy. Second, renal dysfunction was lower than standard-of-care (11.0% vs 26.3%).6 The emergent-resistance signal — amplified by cross-reduced susceptibility after prior vancomycin — is why SAB is now treated at 8–10 mg/kg rather than the labeled 6 mg/kg: higher exposure to suppress resistance.
ERADICATE randomized 387 patients (mITT) with complicated SAB to ceftobiprole vs daptomycin; day-70 overall treatment success was 69.8% vs 68.7% (adjusted difference 2.0 percentage points, 95% CI −7.1 to 11.1; NI margin 15%), giving ceftobiprole its SAB approval. Only about 25% had MRSA, so that subgroup is underpowered, and GI adverse events were more common with ceftobiprole.7 DOTS tested a two-dose dalbavancin strategy (1500 mg on days 1 and 8) against 4–6 weeks of standard IV in complicated SAB with cleared cultures: it was not superior by desirability-of-outcome ranking (47.7%, 95% CI 39.8–55.7), but clinically comparable (day-70 efficacy 73% vs 72%) with fewer discontinuations (3% vs 12%) — an option for the patient who cannot complete outpatient IV therapy, not a new standard.8
| Agent | Anchor trial | Typical SAB dose | What it showed |
|---|---|---|---|
| Vancomycin | Consensus guideline5 | AUC-guided, target 400–600 | AUC dosing preserves efficacy, less AKI than trough 15–20 |
| Daptomycin | Fowler 20066 | 8–10 mg/kg IV daily | NI at 6 mg/kg; resistance emerged on therapy → higher dose now |
| Ceftobiprole | ERADICATE7 | 500 mg IV q6h→q8h | NI to daptomycin (69.8% vs 68.7%); MRSA subgroup underpowered |
| Dalbavancin | DOTS8 | 1500 mg IV days 1 & 8 | Not superior by DOOR; comparable efficacy; fewer discontinuations |
Efficacy in MRSA tracks the area under the concentration–time curve relative to the MIC (AUC/MIC ≥400), not the trough. Trough was long used as a convenient surrogate, but the trough that reliably delivers AUC ≥400 (15–20 mg/L) sits higher than necessary and drives acute kidney injury. Bayesian AUC estimation from one or two levels lets you hit 400–600 at a lower trough — same exposure, less nephrotoxicity.
The practical fellow point: a 'therapeutic' trough of 18 can represent an AUC well over 600 in a patient with reduced clearance — over-exposed and at renal risk — which trough-only dosing hides.
DOOR ranks each patient on a global ordinal scale that combines benefit and harm (e.g., alive without complication > alive with an adverse event > dead), then asks the probability that a randomly chosen patient does better on treatment A than B. 50% means no difference. DOTS returned 47.7% — dalbavancin was not more desirable overall, even though simple efficacy was non-inferior — because the composite weighs harms and quality of life, not just cure.
The front-line guideline says combination therapy is 'occasionally used, evidence limited, ID only.' Two large RCTs are why. ARREST randomized 758 patients to two weeks of adjunctive rifampicin vs placebo on a standard backbone: no effect on the primary endpoint of treatment failure, recurrence, or death (hazard ratio 0.96, 95% CI 0.68–1.35). Rifampicin modestly cut microbiologic recurrence (1% vs 4%; number needed to treat 29) but added drug interactions and adverse events, with no mortality benefit.9
CAMERA2 tested the more intuitive idea — add an antistaphylococcal β-lactam to vancomycin or daptomycin for MRSA bacteremia. It was stopped early for safety. The 90-day composite (mortality, persistent bacteremia, relapse, failure) was 35% vs 39% (absolute difference −4.2%, 95% CI −14.3 to 6.0) — no benefit — while acute kidney injury more than tripled with combination therapy (23% vs 6%; difference 17.2%, 95% CI 9.3–25.2). The combination did clear blood cultures faster (day-5 persistence 11% vs 20%), but that surrogate did not translate into a better outcome.10
| Trial | Intervention | Primary result | Harm signal |
|---|---|---|---|
| ARREST9 | + Rifampicin ×2 wk | Failure/recurrence/death HR 0.96 (0.68–1.35) | More drug interactions & AEs |
| CAMERA210 | + Antistaph β-lactam (MRSA) | 90-d composite 35% vs 39% (−4.2%, −14.3 to 6.0) | AKI 23% vs 6% — trial halted |
CAMERA2's combination arm cleared bacteremia faster yet did not improve the clinical composite and caused more kidney injury. Faster blood-culture clearance is a surrogate — plausibly on the causal path, but trustworthy only once a trial shows the hard outcome moves with it. Here it didn't. The lesson generalizes: a combination or de-escalation that improves a microbiologic marker still has to prove a patient-level benefit before it changes practice.
A positive blood culture now speciates in about an hour on a multiplex panel that also reports the mecA/mecC resistance target, and a negative mecA supports early de-escalation to a β-lactam for presumptive MSSA. In a two-hospital before-after study of 200 MSSA patients, the PCR panel plus stewardship review cut time-to-optimal therapy from 49 to 29.1 hours (−19.9 h, P<0.001) and halved empiric anti-MRSA exposure (44.2 → 20.9 h).11 A network meta-analysis of bloodstream infection found rapid molecular diagnostics paired with active stewardship — not the assay alone — associated with lower mortality.12
The safety of early de-escalation rests on testing the isolate itself (mecA on the positive culture, with PBP2a confirmation), not on a surrogate host marker. This is the reasoning behind the front-line 'do not use a MRSA-nares PCR to guide therapy in culture-positive bacteremia' rule — a nares swab characterizes colonization, not the bloodstream isolate.
A test's negative predictive value is not a fixed property — it rises and falls with disease prevalence in the population you apply it to. A MRSA-nares PCR has a high NPV for ruling out MRSA pneumonia, where MRSA is uncommon. In a patient whose blood is already growing S. aureus, the relevant question is the resistance genotype of that isolate, which mecA answers directly; the nares result adds noise, not signal. Match the test to the specimen that carries the answer.
A transthoracic echo goes to every patient; the real decision is the transesophageal study. TEE is warranted after a TTE when it will change management — persistent bacteremia or fever, an intracardiac device or prosthetic valve, a community-acquired or unknown source, or any metastatic focus. A genuinely low-risk patient — health-care-associated SAB, a removable source, rapid clearance, no device, and a good-quality negative TTE — can forgo it.13,14 Two nuances to hold: TEE is far more sensitive than TTE for vegetations and perivalvular abscess, but a negative TEE does not exclude endocarditis when pretest suspicion is high — repeat it in 3–5 days rather than clearing the patient. And prolonged bacteremia (≥48 h) is itself a red flag, carrying a 90-day mortality near 39%, and should push you toward TEE and a source hunt.13,14
Does whole-body [18F]FDG-PET/CT find things? Yes, when ordering is targeted. In a Dutch academic cohort of 397 SAB patients, PET/CT was ordered selectively in 143 (36%) — community-acquired SAB, cultures still positive at ≥48 h, multiple foci, persistent fever, or endocarditis. It detected a new infectious focus in 73 of 143 (number-needed-to-scan ≈ 2), triggered a new intervention in 33 of 143 (≈ 4), and changed antimicrobial therapy in 44 of 143 (≈ 3). A CRP over 200 mg/L at presentation and a positive 48-hour culture independently predicted a consequence-bearing scan.15
Does it improve outcomes? Unproven. Observational cohorts report lower mortality with PET/CT, but the signal is fragile — in the cohort re-examined by Okamura and colleagues, most of the mortality separation fell within the first 30 days and then the curves converged, the pattern that confounding by indication and immortal-time bias generate. There is no randomized trial. False positives cluster on native valves and on prosthetic valves implanted within three months, where post-surgical inflammation mimics infection.16 The honest synthesis: PET/CT is a high-yield case-finding tool in selected increased-risk patients with an occult source — order it to answer 'where is the focus?', not as a reflex and not as a proven mortality intervention.
A test can change management often and still not change survival. PET/CT's number-needed-to-scan of about 2 for a new focus is a yield statistic — it counts findings, not lives saved. The leap from 'found more foci' to 'patients did better' needs a trial that randomizes the imaging, because the alternative explanations are strong: confounding by indication (the sicker or socially marginalized get scanned less, so the scanned group looks healthier for unrelated reasons) and immortal-time bias (a patient must survive to be scanned, so pre-scan time mis-attributed to the imaged group manufactures a survival edge). An early-only mortality gap that later closes is the fingerprint of both.
Duration keys off whether a deep-seated or metastatic focus is present, not the old complicated/uncomplicated label: roughly 14 days from documented clearance for low-risk disease without a focus, and an individualized, site-driven course when a focus is found.13,17 Name this for a trainee: the 14-day minimum for uncomplicated SAB is consensus, not the product of a randomized duration trial. It rests on cohort data and the historically high relapse rate of under-treated S. aureus. The randomized trials that exist interrogate the route and the setting of therapy, not its total length.
Three randomized trials bracket the oral question — strongest to weakest for SAB specifically. Two are large, positive trials in adjacent deep gram-positive infections; the third is the only one in SAB itself.
| Trial | Population / N | Oral vs IV result | Read for SAB |
|---|---|---|---|
| OVIVA18 | Bone & joint infection; 1054 | 1-yr failure 13.2% vs 14.6% (diff −1.4 pp, 90% CI −4.9 to 2.2); NI margin 7.5% | Large & durable — but the syndrome is bone/joint |
| POET19 | Left-sided endocarditis; 400 | Composite 9.0% vs 12.1% (diff 3.1 pp, 95% CI −3.4 to 9.6) | Pre-stabilized patients switched after ≥10 days IV |
| SABATO20 | Low-risk SAB; 213 | 90-d complication 13% vs 12% (diff 0.7%, 95% CI −7.8 to 9.1) | The only SAB trial — small, stopped early, 8% MRSA |
Read together, oral step-down for deep gram-positive infection is well-supported in general — OVIVA and POET are large, positive, and durable — yet the SAB-specific evidence (SABATO) is the thinnest of the three: stopped early at 213 of a planned 430, only 8% MRSA, and both attributable deaths in the oral arm.20 That asymmetry is why the front-line guideline permits oral step-down only in a narrowly-selected, ID-vetted low-risk patient, and keeps endocarditis, prosthetic material, and undrained deep foci on IV. For the patient who cannot take or absorb orals, a long-acting IV lipoglycopeptide — two doses of dalbavancin (Appendix S2) — sidesteps the oral question rather than answering it.8
Persistent bacteremia (≥48–72 h on appropriate therapy) is a source problem, not a drug-potency problem. No salvage regimen — combination, adjunct, or agent switch — has improved mortality in an adequately powered trial (Appendix S3), so the response is to re-image, re-examine lines and devices, exclude endocarditis, and confirm adequate drug exposure before escalating.9,10,13
OVIVA and POET license oral step-down in bone/joint infection and in stabilized left-sided endocarditis — not, by themselves, in S. aureus bacteremia. Non-inferiority is proven in the population and margin that were studied; the biology of the treated syndrome — biofilm, source control, relapse risk — can differ enough that the same route is not equivalent elsewhere. SABATO is the trial that tests the SAB population directly, and it was small and stopped early, so its 95% CI for the survival difference reached −14.8%, a range that still admits real harm. 'Non-inferior' here means 'compatible with equivalence in a narrow, carefully-selected sample,' not 'proven equal in all comers.'