Staphylococcus aureus

Fosfomycin

BACSARM trial 2013-2017

• Evaluated fosfomycin’s role in combination therapy for MRSA bacteremia.

• Terminated before reaching its planned sample size, which may have limited its statistical power to detect differences in clinical outcomes.

• Daptomycin plus fosfomycin provided a 12% higher rate of treatment success than daptomycin alone.

SAFO trial 2019-2022

• Evaluated fosfomycin’s role in combination therapy for MSSA bacteremia.

• Terminated before reaching its planned sample size, which may have limited their statistical power to detect differences in clinical outcomes.

• Cloxacillin plus fosfomycin provided a 5.3% higher rate of treatment success than cloxacillin alone.

## Why were the original trials inconclusive?

• Both trials ended before reaching their planned sample sizes.

• Point estimates favoured fosfomycin, but confidence intervals were wide.

• Lack of statistical significance did not demonstrate absence of benefit.

• The central problem was limited precision rather than necessarily absence of effect.

BACSARM + SAFO -> BACSAFO

• We performed a pooled post hoc analysis of individual participant data from the BACSARM and SAFO trials.

• Applying complementary Bayesian and frequentist analytical approaches to refine the assessment of fosfomycin efficacy and safety.

Data

BACSARM:

• Adult patients with MRSA bacteremia
• Randomized 1:1 to receive daptomycin vs daptomycin + fosfomycin.
• Intention to treat: 81 vs 74 participants.

SAFO:

• Adult patients with MSSA bacteremia
• Randomized 1:1 to receive cloxacillin vs cloxacillin + fosfomycin.
• Intention to treat: 110 vs 104 participants.

All patients from the intention-to-treat populations were included.

Outcomes

• Treatment success at 8 weeks from randomization:

  Alive, without signs of relapse, and improvement in clinical signs and symptoms.

• Secondary outcomes: persistent bacteremia, adverse events leading to treatment discontinuation and all cause mortality.

• Subgroup analysis: Age, MSSA vs MRSA, site of acquisition, complicated bacteremia.

The study was registered at ClinicalTrials.gov (NCT06695832).

Why an individual-participant-data analysis?

Advantages:

• Harmonised outcome definition.
• Consistent subgroup definitions.
• Preservation of within-trial randomization.
• Modelling of trial and hospital structure.
• Improved precision

What pooling cannot solve:

• Differences in background antibiotics.
• Differences between MRSA and MSSA populations.
• Missing outcomes.
• Clinical heterogeneity.
• Premature trial termination.

BACSARM participants+SAFO participants ⟶One-stage individual-participant-data model

One-stage statistical model

\[Y_{ih} \sim \text{Binomial}(1,p_{ih})\] \[log(p_{ih})=\alpha_0 + u_h + \beta T_{ih} \]

• \(h\) is the hospital.
• \(\alpha_0\) is the hospital-specific baseline risk.
• \(u_h \sim N(0,\tau^2)\)
• \(exp(\beta)\) is the treatment risk ratio.

One model, two inferential frameworks

Statistical Analysis: Bayesian

Hierarchical Bayesian log-binomial model

\[Y_{ih} \sim \text{Binomial}(1,p_{ih})\]

\[log(p_{ih})=\beta_0 + u_h + \beta T_{ih} \rightarrow RR_{treatment}=\exp(\beta) \] \(i\) stands for participant and \(u_h \sim N(0,\tau^2)\) is the hospital-specific random intercept.

Minimally informative prior centered on no treatment effect

Statistical Analysis: Bayesian

Statistical Analysis: frequentist

Hierarchical log-binomial model

\[Y_{ih} \sim \text{Binomial}(1,p_{ih})\]

\[log(p_{ih})=\beta_0 + u_h + \beta T_{ih} \rightarrow RR=\exp(\beta) \] \(i\) stands for participant and \(u_h \sim N(0,\tau^2)\) is the hospital-specific random intercept.

Software

All analyses were performed using R software (v.4.4.1).

Relevant packages:

flowchart, tydiverse, ggplot2, gtsummary, rstanarm and lme4.

Flowchart and baseline characteristics

Primary outcome

Primary outcome

Table 3. Clinical success at week 8

ARD: Absolute Risk Difference (Combination minus Monotherapy).
NNT: Number needed to treat.

The point estimate corresponds to an NNT of approximately 14. However, the 95% credible interval includes no effect and spans both benefit and harm: its upper limit corresponds to an NNT of approximately 6, whereas its lower limit corresponds to one additional failure per approximately 50 patients treated.

Primary outcome: sensitivity analysis

Table 4. Clinical success at week 8

Primary outcome: subgroup

(*) The frequentist approach yielded similar results.

Secondary outcome

Adverse events: Gastrointestinal disturbances, Heart failure, Liver injury.
* P(RR>1∣data)

Adverse Events Leading to Treatment Discontinuation

Absolute Risk Difference (Combination minus Monotherapy)

Other: Combination = hepatic metastases, multiple organ dysfunction syndrome, pneumonia; Monotherapy = rectosigmoid cancer, acute coronary syndrome, skin rash, COVID-19, pneumonia, respiratory failure

Clinical discussion

• Fosfomycin may enhance treatment success at 8 weeks compared to monotherapy.

• Higher incidence of adverse events leading to treatment discontinuation.

• Subgroup analyses suggested potentially greater benefit in patients with MRSA bacteremia, and those with nosocomial acquisition.

• Heterogeneity in study design—including variations in dosing, duration, and fosfomycin administration protocols—may introduce variability in the observed effects.

Methodological discussion

• Pooling individual patient data from multiple trials increases precision and may provide more informative estimates of clinical effects.

• In this study, Bayesian and frequentist approaches gave similar results.

• The Bayesian and frequentist methods provided complementary insights into treatment effects.

• The posterior probability distribution provided additional insight, supporting a more detailed clinical interpretation.

Final message

• Adjunctive fosfomycin may improve early bacterial clearance and treatment success in Staphylococcus aureus bacteremia but at the cost of increased toxicity.

• Pooling individual participant data improved precision but did not eliminate the uncertainty caused by premature trial termination.

• Bayesian analysis allowed that remaining uncertainty to be expressed as clinically interpretable probabilities of benefit and harm, even in the absence of frequentist statistical significance.

Moltes gràcies!

,