Skip to main content

Foundational Trials

ADAPT-Sepsis

Image: Shutterstock / ChaNaWiT

Publication

  • Title: Biomarker-guided antibiotic duration for hospitalized patients with suspected sepsis: the ADAPT-Sepsis randomized clinical trial
  • Acronym: ADAPT-Sepsis
  • Year: 2025
  • Journal published in: JAMA
  • Citation: Dark P, Hossain A, McAuley DF, et al; ADAPT-Sepsis Collaborators. Biomarker-guided antibiotic duration for hospitalized patients with suspected sepsis: the ADAPT-Sepsis randomized clinical trial. JAMA. 2025;333(8):682-693.

Context & Rationale

  • Background
    • Sepsis care requires early effective antimicrobials, but unnecessarily prolonged courses increase drug toxicity, line/monitoring burden, Clostridioides difficile infection risk, and selection pressure for antimicrobial resistance.
    • Biomarkers (particularly CRP and PCT) are frequently measured in ICU sepsis, yet their use as explicit “stopping rules” varies and is often non-protocolised.
    • Pre-ADAPT systematic review and patient-level meta-analysis evidence supported modest reductions in antibiotic exposure with PCT-guided discontinuation in critical illness, but with heterogeneity in trial design, adherence, and outcome definitions.12
    • International sepsis guidelines preceding ADAPT provided only conditional support for PCT (with clinical judgement) to assist antibiotic discontinuation, reflecting uncertainty and variability in prior trials.3
    • Evidence for CRP-guided antibiotic discontinuation in ICU sepsis was limited, despite CRP being widely available, inexpensive, and embedded in routine practice.
  • Research Question/Hypothesis
    • Effectiveness: daily PCT-guided or daily CRP-guided antibiotic discontinuation advice would reduce total antibiotic duration by 28 days compared with standard care (superiority hypothesis).
    • Safety: each biomarker-guided strategy would be non-inferior to standard care for 28-day all-cause mortality (absolute non-inferiority margin 5.4%).
  • Why This Matters
    • ICU stewardship interventions are hard to evaluate rigorously because clinician behaviour is difficult to blind and outcomes can be confounded by co-interventions.
    • A pragmatic head-to-head comparison of PCT versus CRP addresses a key translational question: whether a cheaper, ubiquitous biomarker can safely substitute for PCT in antibiotic stopping decisions.
    • Even small reductions in duration (if safe) could translate into meaningful system-level gains across large sepsis populations, but only if they reflect true reductions in antibiotic exposure and do not increase relapse or harm.

Design & Methods

  • Research Question: In critically ill adults with suspected sepsis started on intravenous antibiotics, does a concealed daily biomarker-guided discontinuation protocol (PCT or CRP) reduce antibiotic duration (superiority) while remaining non-inferior for 28-day mortality (safety) versus standard care?
  • Study Type: Multicentre, pragmatic, investigator-initiated, randomised, parallel-group, concealed-intervention trial in 41 UK adult ICUs; 1:1:1 allocation; clinicians received standardised written advice in all groups (time-delayed/neutral advice in standard care to support concealment); protocol published a priori.4
  • Population:
    • Setting: UK National Health Service adult intensive care units.
    • Inclusion criteria: ICU admission; suspected sepsis (suspected infection treated with intravenous antibiotics, with consideration of ≥2 SIRS criteria); started antibiotics <24 hours before screening; clinician expectation that antibiotics would continue beyond the following day.
    • Key exclusions: expected antibiotic requirement >21 days; local specialist infection advice recommending a specific duration at enrolment; current antibiotic treatment for the present suspected sepsis episode for ≥24 hours; prophylactic antibiotics only; prior participation in the trial.
  • Intervention:
    • Daily PCT-guided protocol: daily serum PCT used to generate concealed written advice (“usual care”, “supports stopping”, “strongly supports stopping”) for discontinuing antibiotics for the suspected sepsis episode.
    • Decision thresholds (protocolised advice rules): usual care if PCT >2.0 µg/L; supports stopping if PCT 0.25–2.0 µg/L with ≥80% reduction from peak; strongly supports stopping if PCT <0.25 µg/L or PCT 0.25–2.0 µg/L with ≥90% reduction from peak.
    • Daily sampling: biomarker testing repeated daily during the intervention phase until stopping criteria were met or the intervention phase ended (with clinical decisions remaining clinician-led).
  • Comparison:
    • Daily CRP-guided protocol: daily serum CRP used to generate concealed written advice (“usual care”, “suggest stop”, “strongly suggest stop”) for discontinuing antibiotics for the suspected sepsis episode.
    • Decision thresholds (protocolised advice rules): usual care if CRP >100 mg/L; suggest stop if CRP 25–100 mg/L with ≥50% reduction from peak; strongly suggest stop if CRP <25 mg/L or CRP 25–100 mg/L with ≥75% reduction from peak.
    • Standard care: antibiotic duration determined by clinicians using usual clinical assessment and locally available tests; trial systems delivered neutral/time-delayed advice to support group concealment.
  • Blinding: Clinicians and participants were blinded to trial group allocation and to trial biomarker results; clinicians received standardised daily advice in all groups (concealed intervention), reducing performance and detection bias for clinician-driven outcomes.
  • Statistics: A total of 2760 patients were planned to detect a 1-day reduction in total antibiotic duration (assumed mean 7 days; SD 6) with 90% power at the 5% significance level (allowing 5% withdrawal), and to assess non-inferiority for 28-day mortality with a 5.4% absolute margin using 90% power with a 2.5% one-sided significance level (allowing 5% withdrawal; assumed 15% mortality); primary analyses were intention-to-treat with multiplicity control for two pairwise comparisons to standard care (P≤0.025 for primary endpoints).
  • Follow-Up Period: Primary endpoints to day 28; mortality additionally assessed to day 90.

Key Results

This trial was not stopped early. Recruitment was paused (March–August 2020) due to SARS‑CoV‑2 lockdown and then resumed to completion.

Outcome [Intervention] [Comparison] Effect p value / 95% CI Notes
Total antibiotic treatment duration to 28 days (primary effectiveness) — PCT protocol 9.8 (SD 7.2) days [n=898] 10.7 (SD 7.6) days [n=905] MD (SC − protocol): 0.88 days 95% CI 0.19 to 1.58; P=0.01 Unadjusted primary analysis; P≤0.025 prespecified for two comparisons.
Total antibiotic treatment duration to 28 days (primary effectiveness) — CRP protocol 10.6 (SD 7.7) days [n=892] 10.7 (SD 7.6) days [n=905] MD (SC − protocol): 0.09 days 95% CI −0.60 to 0.79; P=0.79 Unadjusted primary analysis; no evidence of effect on total duration.
28-day all-cause mortality (primary safety; non-inferiority margin 5.4%) — PCT protocol 184/879 (20.9%) 170/878 (19.4%) AD (protocol − SC): 1.57% 95% CI −2.18 to 5.32; P=0.02 (1-sided NI) Upper CI (5.32%) < 5.4% margin (non-inferior).
28-day all-cause mortality (primary safety; non-inferiority margin 5.4%) — CRP protocol 184/874 (21.1%) 170/878 (19.4%) AD (protocol − SC): 1.69% 95% CI −2.07 to 5.45; P=0.03 (1-sided NI) Upper CI (5.45%) > 5.4% margin (non-inferiority not demonstrated).
Antibiotic duration for initial sepsis period (secondary) — PCT protocol 7.0 (SD 5.7) days [n=893] 8.1 (SD 6.1) days [n=902] MD (SC − protocol): 1.13 days 95% CI 0.58 to 1.68; P=Not reported Secondary endpoint; no multiplicity-adjusted P values reported.
Antibiotic duration for initial sepsis period (secondary) — CRP protocol 7.4 (SD 6.0) days [n=889] 8.1 (SD 6.1) days [n=902] MD (SC − protocol): 0.71 days 95% CI 0.16 to 1.26; P=Not reported Secondary endpoint; direction favours biomarker protocols for sepsis episode only.
Unscheduled care escalation or readmission (≥1 event; secondary) — PCT protocol 208/888 (23.4%) 236/900 (26.2%) AD (SC − protocol): 2.80%; OR (SC vs protocol): 1.16 95% CI (AD) −1.16 to 6.76; 95% CI (OR) 0.94 to 1.44; P=Not reported Event-based endpoint; no statistically supported difference reported.
New infection or superinfection at a different site (≥1 event; secondary) — PCT protocol 29/908 (3.2%) 24/913 (2.6%) AD (SC − protocol): −0.57%; OR (SC vs protocol): 0.82 95% CI (AD) −2.13 to 0.93; 95% CI (OR) 0.47 to 1.42; P=Not reported Rare events; wide confidence intervals.
Suspected clinically relevant antibiotic-related events (≥1 event; secondary) — PCT protocol 71/888 (8.0%) 70/900 (7.8%) AD (SC − protocol): −0.21%; OR (SC vs protocol): 0.97 95% CI (AD) −2.81 to 2.30; 95% CI (OR) 0.69 to 1.37; P=Not reported No signal of increased antibiotic-related harms in biomarker groups.
90-day all-cause mortality (secondary) — PCT protocol 217/847 (25.6%) 215/842 (25.5%) AD (SC − protocol): −0.09%; OR (SC vs protocol): 1.00 95% CI (AD) −4.29 to 4.08; 95% CI (OR) 0.80 to 1.24; P=Not reported Secondary mortality endpoint; neutral estimate with wide CI.
  • PCT-guided advice reduced total antibiotic duration by 0.88 days and met the prespecified non-inferiority criterion for 28-day mortality (upper CI 5.32% vs 5.4% margin).
  • CRP-guided advice did not reduce total antibiotic duration and did not demonstrate non-inferiority for 28-day mortality (upper CI 5.45% > 5.4% margin).
  • Sensitivity analyses were directionally consistent: per-protocol antibiotic duration MD 0.86 days (95% CI 0.16 to 1.56) for PCT vs standard care, and per-protocol 28-day mortality AD 1.25% (95% CI −2.51 to 5.02) for PCT vs standard care.

Internal Validity

  • Randomisation and Allocation
    • Central randomisation with minimisation and prespecified factors (including site, surgical status, and septic shock) supported allocation concealment and baseline balance.
    • Three-arm design with two pairwise comparisons to standard care increased interpretability for PCT and CRP separately, but required multiplicity control (P≤0.025) for the co-primary endpoints.
  • Dropout and Exclusions
    • Randomised: 2760 participants (918 PCT; 924 CRP; 918 standard care).
    • Withdrew completely prior to day 28: 127/2760 (4.6%); an additional 11 requested data removal (excluded from analyses).
    • Withdrew from the intervention phase but remained in follow-up: 364/2760 (13.2%), potentially diluting biological separation between groups.
    • Missingness was low but present: antibiotic duration data missing for 54 participants (2.0%); 90-day mortality data missing for 130 participants (4.8%).
  • Performance and Detection Bias
    • Clinicians were blinded to allocation and trial biomarker results; advice was delivered in a standardised format, reducing the risk of differential co-interventions driven by knowledge of group assignment.
    • Standard care group did not receive stop/strong stop advice, creating a potential route for partial unblinding; primary outcomes (mortality, antibiotic duration) remain relatively objective, mitigating detection bias.
  • Protocol Adherence and Separation of the Variable of Interest
    • Advice production: stop advice was similar for PCT and CRP protocols; strong stop advice was more common and produced earlier for PCT than CRP.
    • Mean time to first stop advice: 3.2 days (PCT) vs 3.4 days (CRP) vs 5.5 days (standard care).
    • Mean time to first strong stop advice: 3.5 days (PCT) vs 5.1 days (CRP).
    • Observed antibiotic separation was statistically significant but modest: total duration 9.8 (SD 7.2) vs 10.7 (SD 7.6) days (MD 0.88); initial sepsis duration 7.0 (SD 5.7) vs 8.1 (SD 6.1) days (MD 1.13).
  • Baseline Characteristics and Illness Severity
    • Groups were well balanced: mean age 60.2–60.3 years; 60.2%–61.0% male; APACHE II mean 17.2–17.4; SOFA median 7 (IQR 5–10); septic shock 48.5%–49.3%.
    • Most common infection source was respiratory (31.2%–34.0%), with abdominal and urinary sources also represented.
  • Timing and Dose
    • Randomisation occurred early (within 24 hours of starting intravenous antibiotics), aligning with the decision window where discontinuation protocols might influence total exposure.
    • Biomarker “dose” was an intensive daily measurement programme with threshold and kinetics-based stopping advice; the requirement for daily assays may be a delivery constraint outside trial settings.
  • Outcome Assessment and Statistical Rigor
    • Primary endpoints were prespecified and objectively measurable (antibiotic duration to day 28; 28-day mortality).
    • Planned sensitivity analyses (per-protocol, imputation, complier average causal effect) were reported; Bayesian analyses quantified the probability of a clinically small but non-trivial effect (PCT: probability of >0.5-day mean reduction 0.85; CRP 0.13).

Conclusion on Internal Validity: Overall, internal validity appears moderate-to-strong given concealed randomisation, standardised advice delivery, objective co-primary endpoints, and consistent sensitivity analyses, while recognising that modest antibiotic separation and withdrawals from the intervention phase could attenuate true effects.

External Validity

  • Population Representativeness
    • Broad adult ICU sepsis population across 41 UK sites supports representativeness for high-income, mixed medical–surgical critical care.
    • Key exclusions (anticipated prolonged antibiotics >21 days; specialist infection advice mandating duration; prior extended antibiotic exposure for the same episode) reduce applicability to immunocompromised patients, endocarditis/osteomyelitis, and other prolonged-course indications.
  • Applicability
    • Implementation requires daily biomarker assays and an advice-generation infrastructure; this is more immediately feasible for CRP than PCT in many systems, but the trial showed no clinical effectiveness signal for CRP on total duration.
    • Settings with shorter baseline antibiotic durations may see smaller or negligible absolute reductions; settings with longer default courses may see greater opportunity for benefit.
    • Resource-limited ICUs without reliable daily biomarker turnaround may not be able to replicate the intervention fidelity.

Conclusion on External Validity: Findings are reasonably generalisable to adult ICUs in high-income systems, but applicability is limited in prolonged-course infections and in settings where daily biomarker testing and decision-support infrastructure are not available.

Strengths & Limitations

  • Strengths:
    • Large, pragmatic, multicentre UK ICU trial with robust follow-up and low missingness for primary endpoints.
    • Concealed intervention with blinded clinician-facing advice reduced performance bias common to stewardship trials.
    • Head-to-head evaluation of PCT and CRP provided direct evidence against assuming CRP is an adequate substitute for PCT in stopping protocols.
    • Multiple sensitivity analyses (per-protocol, CACE, imputation, Bayesian) supported robustness of the main conclusions.
  • Limitations:
    • Effect size on antibiotic duration was modest (0.88 days) and did not translate into clear reductions in antibiotic dose metrics (DDD), potentially limiting system-level impact.
    • CRP protocol did not achieve non-inferiority for 28-day mortality; the point estimate was close to the margin, but confidence bounds crossed it.
    • Intervention delivery is resource-intensive (daily sampling and advice generation), and 13.2% withdrew from the intervention phase, which may dilute real-world effectiveness.
    • Standard care received no stop/strong stop advice, which may compromise full blinding and could influence clinician behaviour (direction uncertain).
    • Trial recruitment spanned the SARS‑CoV‑2 pandemic period, with a pause that may affect co-interventions and baseline prescribing behaviour over time.

Interpretation & Why It Matters

  • Clinical and methodological implications
    • Daily PCT-guided discontinuation advice achieved a statistically significant, clinically small reduction in antibiotic duration with no clear signal of harm in secondary outcomes, supporting its use as an adjunct to clinical judgement in ICU sepsis stewardship.
    • Daily CRP-guided advice did not reduce total antibiotic duration and did not demonstrate non-inferiority for 28-day mortality, cautioning against CRP kinetics being used as a stand-alone stopping rule in ICU sepsis without additional evidence.
    • The concealed-advice approach offers a rigorous trial method for behaviour-dependent ICU interventions, but the modest separation underscores the challenge of shifting entrenched antibiotic prescribing patterns.

Controversies & Subsequent Evidence

  • Non-inferiority margin choice: The 5.4% absolute margin for 28-day mortality was criticised as permissive relative to baseline mortality and the modest antibiotic-duration effect size, raising concerns about the level of harm considered acceptable for stewardship gain.56
  • Antibiotic duration in standard care: The mean total duration of 10.7 days was highlighted as long relative to other contemporary antibiotic-duration trials, with implications for generalisability to settings with shorter baseline courses and for the magnitude of achievable benefit.75
  • Antibiotic exposure metrics: The duration endpoint improved, but defined daily dose did not, emphasising that “shorter duration” may not equate to lower overall antibiotic exposure, and complicating inference about downstream antimicrobial resistance benefits.869
  • Relapse/recurrence trade-off: Infection relapse or recurrence requiring further antibiotics was numerically higher in the PCT arm (11/908; 1.2%) than standard care (5/913; 0.5%), with wide confidence intervals; this was framed as a key clinical trade-off given the small absolute reduction in antibiotic duration.76
  • Resistance and value proposition: Potential system-level benefits on antimicrobial resistance and costs were emphasised as important for implementation decisions, but ADAPT-Sepsis was not powered for resistance endpoints and did not demonstrate reduced antibiotic dose; further cost-effectiveness and implementation work was identified as necessary.10116
  • Positioning within the prior evidence base: ADAPT-Sepsis adds a large, concealed-intervention ICU RCT to a literature where reductions in antibiotic exposure with PCT guidance were previously supported by systematic review and meta-analysis, but with heterogeneous implementation and varying adherence.12

Summary

  • ADAPT-Sepsis randomised 2760 adults in 41 UK ICUs with suspected sepsis to daily PCT-guided advice, daily CRP-guided advice, or standard care.
  • PCT-guided advice reduced total antibiotic duration to 28 days by 0.88 days (MD [SC − protocol] 0.88; 95% CI 0.19 to 1.58; P=0.01) and was non-inferior for 28-day mortality (AD [protocol − SC] 1.57%; 95% CI −2.18 to 5.32; P=0.02; margin 5.4%).
  • CRP-guided advice did not reduce total antibiotic duration (MD 0.09; 95% CI −0.60 to 0.79; P=0.79) and did not demonstrate non-inferiority for 28-day mortality (upper CI 5.45% > 5.4% margin).
  • Secondary outcomes (escalation/readmission, new infection, suspected antibiotic-related events, 90-day mortality) were similar across groups, with wide confidence intervals.
  • The trial demonstrates feasibility of concealed, behaviour-dependent stewardship interventions in ICU, but the antibiotic-duration effect was modest and its value depends on baseline prescribing patterns and implementation costs.

Overall Takeaway

ADAPT-Sepsis is a landmark ICU stewardship trial because it combined a pragmatic, multicentre design with a concealed, behaviour-dependent intervention and co-primary effectiveness and safety endpoints. It shows that daily PCT-guided discontinuation advice can reduce antibiotic duration modestly while meeting a prespecified non-inferiority criterion for 28-day mortality, whereas a CRP-based stopping protocol did not deliver comparable effectiveness or safety assurance.

Overall Summary

  • PCT-guided stopping advice produced a small (≈1 day) reduction in antibiotic duration without excess short-term mortality; CRP-guided advice did not, and did not meet non-inferiority for 28-day mortality.

Bibliography