Skip to main content

Trial Summaries & Critiques

PRONTO

Image: Shutterstock / Saiful52

Point of care procalcitonin assay held in a gloved hand

Publication

  • Title: Procalcitonin testing combined with NEWS2 evaluation compared with usual care based on NEWS2 for identification of sepsis and antibiotic initiation in the emergency department in England and Wales (PRONTO): a multicentre, randomised, controlled, open-label, phase 3 trial
  • Acronym: PRONTO
  • Year: 2026
  • Journal published in: Lancet Respiratory Medicine
  • Citation: Todd S, Euden J, Condie J, Aston S, Barlow G, Brookes-Howell L, et al; PRONTO Trial Group. Procalcitonin testing combined with NEWS2 evaluation compared with usual care based on NEWS2 for identification of sepsis and antibiotic initiation in the emergency department in England and Wales (PRONTO): a multicentre, randomised, controlled, open-label, phase 3 trial. Lancet Respir Med. Published online March 22, 2026.

Context & Rationale

  • Background
    • Emergency department recognition of sepsis is a time-critical but diagnostically uncertain task: NEWS2 can identify physiological deterioration, but it is not a specific test for bacterial sepsis and was never intended to function as one in isolation.1
    • Earlier procalcitonin literature suggested that biomarker-guided strategies might shorten antibiotic exposure and perhaps improve survival, but most higher-quality evidence arose from ICU cohorts or lower respiratory tract infection populations rather than broad, undifferentiated suspected sepsis at first ED contact.2
    • Recent ED-focused synthesis still shows only moderate diagnostic performance for PCT as a sepsis test, with pooled sensitivity 0.62 and specificity 0.80, supporting its role as an adjunct to rather than a replacement for clinical judgement.3
    • The stewardship problem is substantial: in a 2025 seven-hospital ED cohort, 31.5% of broadly treated suspected sepsis cases were less likely or definitely unlikely to have bacterial infection, 79.1% of definite/probable bacterial infections received retrospectively over-broad therapy, and 17.3% developed possible antibiotic-associated complications.4
  • Research Question/Hypothesis
    • Whether rapid point-of-care PCT testing, added to NEWS2-based triage in adults managed as suspected sepsis in UK EDs, would reduce IV antibiotic initiation by 3 h while maintaining at least non-inferior 28-day mortality compared with NEWS2-based usual care alone.1
    • The embedded mechanistic hypothesis was that a biomarker-informed early algorithm might improve discrimination between bacterial infection, viral illness, and non-infectious sepsis mimics at the bedside.
  • Why This Matters
    • Few acute care decisions are more consequential than whether a patient with possible sepsis needs immediate broad-spectrum IV antibiotics, brief diagnostic clarification, or urgent evaluation for an alternative diagnosis.
    • A useful test at this stage could improve patient outcomes while simultaneously reducing unnecessary antimicrobial exposure at system scale.4
    • PRONTO therefore tested not just a biomarker, but an entire model of early ED decision support embedded in routine NHS practice.

Design & Methods

  • Research Question: In adults managed as suspected sepsis in the emergency department, does adding a rapid point-of-care procalcitonin result to NEWS2-guided clinical assessment reduce IV antibiotic initiation by 3 h without worsening 28-day mortality?
  • Study Type: Parallel, two-arm, open-label, individually randomised, pragmatic, multicentre phase 3 trial with co-primary effectiveness and safety endpoints, adaptive group-sequential stopping rules, and recruitment across 20 hospital emergency departments within 17 NHS Trusts or Health Boards in England and Wales.
  • Population:
    • Adults aged ≥16 years treated by local teams as suspected sepsis according to site sepsis pathways; enrolment was irrespective of baseline NEWS2.
    • Sites used local sepsis pathways aligned to NICE suspected sepsis guidance, with NEWS2 mandatory and additional local triggers permitted.
    • Exclusions: IV antibiotics already started at screening; current immunosuppressive/myeloablative chemotherapy; solid-organ transplantation or allogeneic bone marrow/stem-cell transplantation within 3 months; urgent surgical intervention; and end-of-life pathways or directives that included withholding antibiotic treatment.
    • Patients not for CPR could still be enrolled if other active treatment remained appropriate.
  • Intervention:
    • Point-of-care procalcitonin testing at presentation, with assay turnaround under 20 min, integrated into a clinician-facing NEWS2-based risk algorithm.
    • PCT thresholds were <0.5 μg/L, 0.5–1.9 μg/L, and ≥2.0 μg/L; these were used to down-classify or up-classify risk relative to NEWS2 alone.
    • Algorithm recommendations: high risk—urgent senior review plus IV antibiotics within 1 h; medium risk—urgent senior review and antibiotics within 3 h if sepsis remained suspected; low risk—clinical assessment and treatment according to judgement.
    • Patients with NEWS2 ≥7 but PCT <0.5 μg/L were specifically flagged for consideration of non-infectious causes.
    • Two CE-marked devices were used during the trial: BRAHMS PCT direct initially and PathFast BRAHMS PCT after the former was removed from the market in 2023.
    • Clinicians retained full discretion to ignore or over-ride the algorithm.
  • Comparison:
    • Usual care based on NEWS2 plus local standard sepsis assessment, without trial-mandated PCT testing.
    • Patients remained managed according to local NICE-derived risk pathways: IV antibiotics within 1 h for high-risk patients, within 3 h for medium-risk patients if sepsis remained likely, and after clinical assessment for low-risk patients.
    • Additional local tools beyond NEWS2 were allowed, reflecting pragmatic NHS practice.
    • Clinicians could over-ride the pathway at any time.
  • Blinding: Open-label. Participants, research staff, outcome assessors, and statisticians were unblinded. The major implication is risk of performance bias rather than detection bias, because the co-primary outcomes were objective.
  • Statistics: The trial was powered primarily on 28-day mortality: 7002 participants were required to exclude more than a 2.5% absolute increase in mortality (assumed control mortality 15.0%; non-inferiority threshold 17.5%) with 90% power at a one-sided 5% significance level. For the co-primary superiority endpoint of IV antibiotics by 3 h, 532 participants would have detected a 10% absolute reduction (90% to 80%) with 90% power at a two-sided 5% level. After one O’Brien–Fleming interim analysis and 5% inflation for dropout, the recruitment target was 7676. The primary analysis used a modified intention-to-treat complete-case population: all consenting randomised participants with both co-primary outcomes available, analysed according to allocation with two-level logistic regression adjusted for baseline NEWS2 and random site effects.
  • Follow-Up Period: Daily inpatient review from triage until discharge or day 28, whichever came first; mortality, readmission, and safety follow-up to day 90.

Key Results

This trial was not stopped early. A planned interim analysis at 43% recruitment and an IDMC-requested second interim analysis at 57% both crossed the superiority stopping boundary for 28-day mortality, but the IDMC recommended continuation to the full planned sample size.

Outcome Procalcitonin-guided care Usual care Effect p value / 95% CI Notes
IV antibiotics initiated within 3 h (co-primary) 1325/2738 (48.4%) 1308/2715 (48.2%) Adjusted RD −0.08 percentage points 95% CI −2.58 to 2.42; P=0.95 No superiority for the stewardship endpoint.
28-day mortality (co-primary) 372/2738 (13.6%) 450/2715 (16.6%) Adjusted RD −3.12 percentage points 90% CI −4.68 to −1.57; P=0.0009 Met both non-inferiority and superiority criteria.
Mortality within 90 days 598/2988 (20.0%) 693/2909 (23.8%) Adjusted OR 0.804 95% CI 0.701 to 0.922; P=0.0018 Mortality benefit persisted to day 90.
Time until death Median 12 days (IQR 4–35) Median 12 days (IQR 4–34) Adjusted HR 0.835 95% CI 0.748 to 0.932; P=0.0013 Secondary survival analysis favoured intervention.
IV antibiotic initiation after 3 h 614/2635 (23.3%) 631/2606 (24.2%) Adjusted OR 0.966 95% CI 0.848 to 1.101; P=0.61 No signal that PCT delayed antibiotics.
Days on any antibiotic during first 28 days Median 5 days (IQR 1–9) Median 5 days (IQR 1–9) Adjusted IRR 1.005 95% CI 0.952 to 1.060; P=0.86 No reduction in overall antibiotic exposure.
ICU or HDU admission 192/2925 (6.6%) 161/2847 (5.7%) Adjusted OR 1.139 95% CI 0.913 to 1.420; P=0.25 Numerically higher in the intervention arm, not statistically significant.
Adverse antibiotic outcomes 56/2768 (2.0%) 53/2738 (1.9%) Adjusted OR 1.056 95% CI 0.721 to 1.547; P=0.78 No evidence of excess antibiotic-related harm.
All recorded adverse events over 28 days 84 events in 66/3042 participants (2.2%) 96 events in 57/2968 participants (1.9%) Not reported P=0.23 One serious adverse event, probably or definitely attributable to PCT testing, occurred in the intervention group; the nature of the event was not reported.
  • No prespecified subgroup significantly modified either co-primary treatment effect on formal interaction testing, including baseline NEWS2, infection site, suspected COVID-19, PCT platform, recruitment epoch, or ED crowding.
  • The mortality benefit persisted to 90 days and was reported across both infectious and non-infectious discharge diagnoses; a post-hoc deprivation analysis suggested greater benefit in more deprived groups (interaction P=0.0064).
  • In the primary analysis population, clinicians had a PCT result available and seen in 1771/2738 (64.7%) intervention patients and reported agreement with the algorithm in 1293/2738 (47.2%), yet 3-h IV antibiotic use remained 48.4% versus 48.2%, underscoring the disconnect between biological reclassification and antibiotic delivery.

Internal Validity

  • Randomization and Allocation: Secure web-based 1:1 randomisation used site and NEWS2 minimisation with an 80% random element, making allocation concealment strong and reducing predictable assignment.
  • Drop out or exclusions: Deferred consent improved inclusiveness but introduced substantial post-randomisation attrition: 1548/7667 (20.2%) never consented, 109 withdrew by day 28, and the primary analysis included 5453/7667 (71.1%). Most complete-case exclusions were due to missing 3-h antibiotic data (304 intervention, 252 control; plus 1 control patient missing 28-day mortality). Mortality among withdrawn participants to day 90 was balanced at 74/809 (9.2%) versus 67/739 (9.1%).
  • Performance/Detection Bias: The study was open-label, and even outcome assessors and statisticians were unblinded. Because the co-primary outcomes were objective, detection bias is probably modest; performance bias is the central concern, as a clinician-facing biomarker can alter behaviour in ways not fully captured.
  • Protocol Adherence: Separation of exposure was incomplete. In the intervention arm, a PCT result was available in 2930/3092 (94.8%), seen by clinicians in 1962/3092 (63.5%), and the algorithm was reported as followed in 1430/3092 (46.2%). CACE analyses strengthened the 28-day mortality effect as adherence increased (OR 0.790 when PCT was available; 0.779 when seen; 0.695 when seen and the recommendation was accepted).
  • Baseline Characteristics: Groups were well balanced—median age 72 versus 73 years, mean Charlson Comorbidity Index 3.9 versus 4.0, and NEWS2 ≥7 in 35.2% versus 34.8%. This was a genuinely ill cohort, but not a septic-shock-only population.
  • Heterogeneity: Clinical heterogeneity was substantial and deliberate: blood cultures were positive in 207/3092 (6.7%) versus 197/3027 (6.5%), and final discharge diagnoses were non-infectious in 528/2738 (19.3%) versus 521/2715 (19.2%). That broadens relevance but weakens mechanistic specificity. Reassuringly, site-level heterogeneity for mortality was low (random-intercept variance 0.062 unadjusted; 0.074 adjusted).
  • Timing: The intervention was delivered early enough to plausibly influence first-line decisions—triage to randomisation 19 versus 20 min, triage to clinical risk assessment 46 versus 49 min, and antibiotic prescription to administration 27 versus 28 min. Senior review still occurred relatively late in both groups (median 3 h 12 min versus 3 h 4 min), reflecting real ED constraints rather than protocol failure.
  • Dose: Only a single baseline PCT measurement was used. That is biologically plausible for triage, but less optimal if the real therapeutic target is serial reassessment or de-escalation.
  • Separation of the Variable of Interest: Risk classification changed meaningfully, but care processes did not separate cleanly. Among 1944 patients initially high risk by NEWS2 alone, 479 (24.6%) were reclassified low risk and 641 (33.0%) medium risk. Despite this, IV antibiotics within 3 h were 48.4% versus 48.2%; any antibiotic days over 28 days were median 5 versus 5; IV antibiotic days were median 3 versus 3.
  • Key Delivery Aspects: Clinician discretion was intentionally preserved, improving pragmatism but weakening protocol contrast. One site (158/7667 recruits; 2.1%) incorporated PCT into routine sepsis care late in the trial, introducing small contamination risk. Device change mid-trial did not appear to modify effect in prespecified interaction testing.
  • Outcome Assessment: The co-primary outcomes were objective, clinically important, and clearly defined. Missingness clustered around the timing of 3-h antibiotic administration rather than mortality ascertainment.
  • Statistical Rigor: The adaptive design, interim boundaries, and analysis plan were prespecified; only 30 protocol deviations were recorded; complete-case, covariate-adjusted, and multiple-imputation analyses were directionally concordant. The main interpretive caveat is not statistical fragility but unexplained causality.

Conclusion on Internal Validity: Internal validity is moderate. Randomisation, allocation concealment, objective endpoints, and consistency across sensitivity analyses are strong, but open-label care, substantial post-randomisation non-consent, limited adherence, and minimal antibiotic-exposure separation leave the mortality signal credible yet mechanistically unresolved.

External Validity

  • Population Representativeness: PRONTO enrolled adults aged ≥16 years treated as suspected sepsis at ED triage, irrespective of baseline NEWS2, with median age above 70 years and substantial comorbidity. This is closer to real-world sepsis-alert populations than confirmatory infection cohorts.
  • Important Exclusions: Current IV antibiotic use, major recent transplantation, active myeloablative chemotherapy, urgent surgery, and end-of-life pathways that withheld antibiotics were excluded, so the findings should not be extrapolated uncritically to neutropenic, transplant, peri-operative, or source-control-dominant sepsis populations.
  • Applicability: Generalisability is strongest to health systems using NEWS2-like deterioration tools, nurse-led triage sepsis pathways, and rapid point-of-care PCT. It is less direct in systems that rely on other screening frameworks, lack fast biomarker platforms, or define suspected sepsis differently.
  • Demographic and Setting Constraints: The primary analysis population was predominantly White (84.5% versus 85.2%), and the study was conducted entirely within NHS workflow, staffing, and escalation structures. Those features may limit transportability of the exact effect size to more diverse populations and non-NHS health economies.
  • Implementation Caveat: Because the mechanism of benefit remains uncertain and antibiotic exposure did not fall, simple adoption of “single admission PCT” without careful implementation and process evaluation may not reproduce PRONTO’s outcomes elsewhere.

Conclusion on External Validity: External validity is moderate to good for UK-like emergency departments managing broad suspected sepsis populations through NEWS2-centred pathways. It is more limited for highly immunocompromised cohorts, urgent surgical sepsis, and health systems that cannot reproduce the same biomarker availability or care processes.

Strengths & Limitations

  • Strengths:
    • Largest randomised trial to date of baseline PCT-guided suspected-sepsis assessment in the emergency department.
    • Highly pragmatic multicentre NHS design with broad inclusion and deferred consent, enhancing real-world relevance.
    • Objective co-primary outcomes that captured both intended effectiveness and safety.
    • Prespecified adaptive design with consistent sensitivity analyses and full recruitment despite major pandemic-era operational pressures.
    • Mortality signal remained directionally stable across adjusted, complete-case, imputed, and compliant-average causal effect analyses.
  • Limitations:
    • Open-label design with unblinded assessors and statisticians, making performance bias the principal interpretive concern.
    • High post-randomisation non-consent and withdrawal reduced the analysable population.
    • Limited adherence and very little antibiotic-exposure separation, despite substantial biological risk reclassification.
    • Single-country, NEWS2-centric context with predominantly White participants limits portability.
    • Single baseline PCT measurement may not represent the optimal biological “dose” if serial biomarker-guided care is the more effective strategy.
    • The mortality mechanism was not directly measured, so the most clinically important finding remains the least well explained.

Interpretation & Why It Matters

  • Clinical interpretation
    • PRONTO did not establish baseline PCT as a tool for reducing early IV antibiotic initiation in suspected sepsis; the intended stewardship target was negative.
    • Its importance lies in the opposite finding: survival improved despite unchanged antibiotic exposure, suggesting that an early biomarker may exert value through better diagnostic orientation, prioritisation, or escalation rather than simple antibiotic withholding.5
    • For clinicians, the study justifies serious attention to rapid diagnostic support at presentation; for trialists, it argues that future biomarker studies must measure downstream diagnostic work-up, source control, escalation, and mimic detection, not antibiotic timing alone.

Controversies & Other Evidence

  • The core controversy is the discordant result profile: no reduction in 3-h IV antibiotics or total antibiotic exposure, yet a clear mortality benefit. Cabral and Rhee argued that unmeasured improvements in diagnostic clarity, prioritisation of escalation, or earlier pursuit of alternative diagnoses are more plausible explanations than stewardship itself, but also noted that those intermediate processes were not captured.5
  • The open-label design matters. The same editorial highlighted that clinician-facing biomarker trials are intrinsically vulnerable to performance bias, and that the one-tailed non-inferiority framework with 90% confidence intervals for the mortality endpoint, although prespecified and methodologically legitimate, warrants careful interpretation.5
  • The stewardship endpoint was probably harder to move than originally assumed. The trial was designed around an expected 90% control-arm IV antibiotic initiation rate by 3 h, yet observed control use was only 48.2% at 3 h and 65.5% by 12 h, implying major secular change in practice and much less room to demonstrate antibiotic reduction.
  • Low adherence cuts both ways: it weakens causal contrast, yet the compliant-average causal effect analysis strengthened rather than erased the mortality signal. That supports a real intervention effect, but does not prove its mechanism.
  • Subsequent and related interventional evidence remains mixed. In ICU sepsis, ADAPT-Sepsis showed that serial daily PCT safely shortened antibiotic duration by 0.88 days versus standard care, whereas the MULTI-CAP ICU CAP trial failed on its primary endpoint of antibiotic-free days at day 28 despite a modest 3-day reduction in cumulative antibiotic duration.67
  • Broader synthesis still supports caution rather than simple extrapolation. A 2023 meta-analysis found shorter antibiotic courses and lower 28-day mortality with PCT-guided strategies, but also higher recurrent infection and marked between-trial heterogeneity; a 2025 ED diagnostic meta-analysis reported pooled sensitivity only 0.62 and specificity 0.80, reinforcing that PCT is an adjunct, not a stand-alone rule-out test.23
  • The most recent international adult Surviving Sepsis Campaign guideline was published on March 23, 2026, one day after PRONTO appeared online. It is therefore the current guideline backdrop, but not yet a mature consensus response to PRONTO-specific implementation questions.8
  • Independent of whether PRONTO’s exact algorithm is adopted, newer ED evidence showing frequent overtreatment and appreciable antibiotic-associated harm in suspected sepsis means the underlying diagnostic-stewardship problem remains clinically urgent.4

Summary

  • PRONTO was a large, pragmatic NHS emergency department trial testing whether rapid point-of-care PCT plus NEWS2 could improve suspected-sepsis management.
  • The intended stewardship endpoint was negative: IV antibiotics within 3 h were essentially identical between groups (48.4% versus 48.2%), and overall antibiotic exposure did not fall.
  • Despite this, 28-day mortality fell from 16.6% to 13.6% (adjusted risk difference −3.12 percentage points), with benefit persisting to 90 days.
  • Internal validity is strengthened by concealed randomisation, objective outcomes, and concordant sensitivity analyses, but weakened by open-label care, 20.2% post-randomisation non-consent, limited adherence, and minimal separation in antibiotic delivery.
  • The trial is agenda-setting because it suggests that early sepsis biomarkers may matter most through decision support and diagnostic clarification, not simply through reducing antibiotic use.

Overall Takeaway

PRONTO overturned the simplest expectation for admission PCT in suspected sepsis: antibiotics did not fall, yet mortality did. It therefore shifts the field from asking whether a biomarker can merely suppress antibiotic prescribing to asking how rapid diagnostic support changes early clinical reasoning, prioritisation, and rescue. Until replicated and mechanistically explained, PRONTO is better viewed as practice-informing and research-directing than as a mandate for unqualified universal implementation.

Overall Summary

  • PRONTO is the largest randomised ED biomarker trial in suspected sepsis to date.
  • It showed no reduction in early IV antibiotic use or total antibiotic exposure, but it reduced 28-day and 90-day mortality.
  • The signal is clinically important and statistically robust enough to change the research agenda, but still requires mechanistic explanation and replication before universal adoption.

Bibliography