Document resource
WHAT IS ALREADY KNOWN ON THIS TOPIC Procalcitonin measurements, when embedded in diagnosis specific clinical algorithms, reduce the duration of antibiotic use in critically ill adults and in neonatal early onset sepsis, without increasing mortality in critically ill adultsNo evidence based universal protocols currently exist to guide antibiotic treatment duration in neonatal late onset sepsis, resulting in treatment durations ranging from five days to up to three weeksWHAT THIS STUDY ADDS Procalcitonin-guided treatment can safely reduce antibiotic use in neonatal late onset sepsis (meningitis, septic shock, and deep-seated infection excluded)HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE, OR POLICY This strategy represents a possible personalised tool in antibiotic stewardship approach to reduce the burden of antibiotic exposure at bedside in early lifeIntroduction Neonatal sepsis is one of the most common neonatal diseases, and its incidence is rising worldwide. The Global Burden of Disease 2019 study, conducted across 204 countries and territories in the past 30 years, reported that the number of incident cases of neonatal sepsis has increased by 12.8% annually, from 5.59 million in 1990 to 6.31 million in 2019. 1 Late onset neonatal sepsis is an infection occurring in newborns after 72 hours of life, typically acquired postnatally from the environment, healthcare-associated sources, or caregivers. It is distinct from early onset sepsis, which occurs within the first 72 hours of life and is usually a result of vertical transmission from the mother.Preterm and low birthweight infants are particularly vulnerable to late onset sepsis, with reported rates as high as 40%.2 3 However, the body of evidence of the harm associated with prolonged or unnecessary antibiotic exposure in children is growing, particularly in newborns. The short term consequences are the development of antimicrobial resistance, identified by the World Health Organization (WHO) as one of the greatest threats to human health for years,4 and adverse neonatal outcomes like necrotising enterocolitis and increased mortality.5 6 Long term consequences include obesity,7 atopy, and asthma,8 possibly related to the specific impact of antibiotics on microbiota implantation in very preterm infants.9 Current antibiotic prescribing practices vary widely internationally.10 The optimal duration of antimicrobial therapy in neonatal late onset sepsis is not well established11 and, although French diagnostic and therapeutic protocols are based on international recommendations,12 13 many are not standardised and practice remains heterogeneous. To date, no evidence based universal protocol exists to guide antibiotic treatment in such situations.10Procalcitonin-guided decision making has been used to reduce antibiotic treatment in critically ill adults; procalcitonin measurements embedded in diagnosis specific clinical algorithms reduce antibiotic duration without increasing mortality.14 Consequently, international guidelines recommend the use of procalcitonin levels to support the discontinuation of empiric antibiotics in adult patients who are initially suspected of having sepsis.15 The Neonatal Procalcitonin Intervention Study (NeoPInS) evaluated the efficacy of procalcitonin-guided decision making on antibiotic duration in suspected early onset neonatal sepsis and reported a reduction of about 20%.16The efficacy of procalcitonin-guided decision making for antibiotic duration in late onset sepsis among preterm and full term newborns has never been studied. We conducted a national, multicentre, randomised intervention trial to assess whether procalcitonin-guided management of suspected late onset sepsis could safely reduce the duration of antibiotic treatment (superiority hypothesis) without increasing mortality in the first month after infection (non-inferiority hypothesis).Methods Study design The Procalcitonin and duration of AntiBIotherapy in late onset Sepsis of newborns (ProABIS) study was a prospective, multicentre, randomised, open label intervention trial conducted in 36 neonatal intensive care units in France. The study protocol and statistical analysis plan are provided in the online supplemental appendix. The study protocol was prospectively registered at ClinicalTrials.gov (NCT03730636) on 2 November 2018, before enrolment of the first patient. This article has been written in accordance with the CONSORT 2025 statement (https://www.equator-network.org/reporting-guidelines/consort/) (online supplemental file 4).SP110.1136/bmjmed-2026-002602.supp1Supplementary dataSP210.1136/bmjmed-2026-002602.supp2Supplementary dataSP510.1136/bmjmed-2026-002602.supp5Supplementary dataParticipants Eligible participants were newborns in neonatal intensive care units of postconceptional age 24-45 weeks and after 4 days of life, with suspected or proven late onset sepsis (defined as onset three days after birth, 17 and for whom the treating physician had planned to continue antibiotic treatment beyond 48 hours’ duration). Eligible infants had to weigh at least 1000 g at inclusion initially and at least 700 g after 19 July 2019 (for substantial changes to the protocol, see online supplemental table S1a). Randomisation was permitted only if the first antibiotic doses had been taken by the participant within the preceding 48 hours.SP310.1136/bmjmed-2026-002602.supp3Supplementary dataExclusion criteria were the absence of an indication to continue antibiotic treatment after the initial 48 hour period, antibiotic treatment in the 48 hours preceding the current suspected episode of infection, severe infections such as meningitis, conditions requiring prolonged antibiotic therapy (eg, endocarditis, bone infection, deep-seated infection, or abscess), septic shock (defined by fluid resistant hypotension requiring vasopressor therapy), treatment with extracorporeal membrane oxygenation or extracorporeal circulation in the past 72 hours, infections not contracted in hospital in the period after birth, and infection diagnosed more than 48 hours after discharge from hospital.Randomisation and masking Patients were randomly assigned in a 1:1 ratio to receive either procalcitonin-guided treatment (procalcitonin-guided group) or usual care (usual care group). The investigator allocated participants at 48 (±12) hours after initiation of antibiotic treatment, using the electronic case report form (CleanWeb, Telemedicin Technologies, SAS). Randomisation was stratified by participating centre. Centralised block randomisation, with permuted blocks of variable size, was performed by an independent statistician from the Clinical Research Unit (URC-EST).Procalcitonin-guided strategy In the procalcitonin-guided group, a nurse working in the neonatal intensive care unit drew a sample of 0.2 mL of blood (or 0.5 mL when additional biological assays were required from the same sample) by venepuncture that was sent immediately to the biochemistry laboratory. Procalcitonin concentration was measured at randomisation (the second day after the start of antibiotic treatment) and then every two days, to minimise blood loss. In this group, if a participant’s procalcitonin level at randomisation was 0.5 µg/L or lower, treatment was continued until the next assay performed 24 hours later. If procalcitonin level was above 0.5 µg/L, procalcitonin concentration was measured every two days until the procalcitonin value fell to 0.5 µg/L or lower. In both situations, the physician was strongly encouraged to discontinue antibiotics as soon as the procalcitonin value was 0.5 µg/L or lower.We did not modify the intervention protocol during the study. Adherence was assessed as the proportion of newborns whose antibiotic treatment was managed in accordance with the described protocol. In cases of recurrence or new infection, antibiotic treatment duration in the experimental group was no longer guided by procalcitonin levels.In the usual care group, antibiotics were discontinued according to the attending physician's practice or local guidelines, and procalcitonin concentration measurement was permitted before randomisation only. Apart from procalcitonin assays, patient monitoring was identical in both groups, and we limited blood sampling to the usual frequencies of standard neonatal care. We measured procalcitonin concentrations using one of the following BRAHMS PCT assays: sensitive KRYPTOR (Thermo Fisher Scientific), Elecsys (Roche), Atellica IM and ADVIA Centaur (Siemens), Alinity i and ARCHITECT (Abbott). The intervention is described in accordance with the TIDieR checklist (available in online supplemental file 5).SP410.1136/bmjmed-2026-002602.supp4Supplementary dataOutcomes The primary outcome was the duration of antibiotic treatment, defined as the total number of days from initiation to completion of the treatment for the primary sepsis episode, plus any additional days required for recurrence (under the superiority hypothesis). We defined primary sepsis as the initial infection that led to study inclusion and initiation of the first course of antibiotic therapy. Recurrence of infection was defined as a new infection occurring within 72 hours after completion of the initial course of antibiotic treatment for primary sepsis that required an additional course of antibiotic treatment. We considered any infection requiring antibiotics within this 72 hour period a recurrence, even if the initial episode of sepsis was culture negative. The cumulative duration of antibiotic therapy was measured as the number of days from treatment initiation (counted from two days before randomisation) to treatment completion, or to day 28 if treatment extended beyond this time. We added duration of treatment for recurrent sepsis to that of the primary episode of sepsis.Mortality at day 28 after randomisation was the key secondary outcome. Other secondary outcomes were the proportion of patients with recurrence of the primary sepsis, the proportion of patients with other infections within 28 days, the proportion of patients with bronchopulmonary dysplasia at day 28, and the types of bacteriological species identified in blood or other samples. We define other infections as any bacterial infection requiring antibiotic treatment more than 72 hours after discontinuation of antibiotics prescribed for primary sepsis.During the trial, data from individual patient records collected by local investigators were entered into the electronic case report form. Parents were contacted by telephone, 28 (±2) days after randomisation to assess adverse outcomes if the newborn had been discharged before day 28. The sponsor's clinical research associate verified all source data during site visits to ensure data quality and completeness. An independent adjudication committee, comprising two neonatologists, a paediatrician, and a bacteriologist, reviewed recurrences of primary sepsis and analysed causes of death, including those attributed to sepsis.Serious adverse events were reported immediately by the investigator to the sponsor and monitored until the patient's clinical condition had stabilised. An independent data and safety monitoring board—comprising two neonatologists and one paediatric intensivist who were not involved in the design, conduct, or recruitment of the study—regularly reviewed masked patient safety data.Statistical analysis We designed the trial to assess the superiority of the procalcitonin-guided strategy in reducing the duration of antibiotic therapy for treating primary sepsis (primary outcome) and to demonstrate non-inferiority for mortality at day 28 compared with usual care (key secondary outcome).We calculated sample size to demonstrate both superiority for antibiotic duration and non-inferiority for mortality at day 28. In high income settings, late onset sepsis in preterm infants is associated with a mortality of 7% to 16%, depending on gestational age, cohort, and definitions of late onset sepsis.18 As reported in the NeoPIns study,16 the choice of the non-inferiority margin was based on clinical and feasibility considerations, given limited available evidence. A margin of less than 3% would have compromised feasibility, whereas a margin of more than 3% would have been clinically unacceptable. Therefore, with a 3% non-inferiority margin (as a compromise between feasibility and safety of the study) and an expected death proportion of 6% in the usual care group according to Cantey et al19 (6.1% (95% confidence interval (CI) 3.1% to 9.1%)), a one sided α=0.025, a β=0.20 and the addition of 10% of non-evaluable patients, our study required a total of 2162 randomised patients. Assuming a mean antibiotic duration of 11 days and a standard deviation (SD) of 5.5 days in the usual care group, a sample size of 2162 patients would provide a power level of more than 90% to demonstrate a reduction in antibiotic duration of at least 15% in the procalcitonin-guided group (9.35 days) under the superiority hypothesis, with a two sided α of 0.05 and 2.5% of patients being non-evaluable (ie, as a post hoc example, 510 patients would be sufficient).We defined three population sets: the intention-to-treat population dataset including all randomised patients, the intention-to-treat population dataset with available data for the primary outcome, and the per protocol population dataset (defined as all randomised patients without major protocol deviation).The duration of antibiotic therapy for primary sepsis was compared between groups using the Wilcoxon-Mann-Whitney test and assessed in the intention-to-treat population. We performed sensitivity analyses in the intention-to-treat dataset with available data and per protocol dataset, and distribution plots of antibiotic duration were generated for both groups. We performed additional sensitivity analyses using a linear quantile mixed model considering centre as a random effect in the intention-to-treat and per protocol populations. Median CIs were calculated using bootstrap resampling with replacement (1000 samples).We performed non-inferiority analysis for mortality with a margin of 3%. Concordant findings in the intention-to-treat and per protocol groups were required to establish non-inferiority. We performed additional sensitivity analyses using a generalised linear mixed model with a binary distribution and logit link and considering centre as a random effect in the intention-to-treat and per protocol groups, and other secondary and safety outcomes were analysed in the intention-to-treat population with available data, under the assumption of superiority. We calculated median CIs using bootstrap resampling with replacement (1000 samples), and CIs for proportions using the exact method. Post hoc analyses compared the number of patients with another infection between groups using the χ2 test and described the proportion receiving antibiotics for more than three days to treat the infection.In case of missing data, we applied single imputation for antibiotic therapy duration (using the maximum observed duration) and for vital status (imputed as death) in intention-to-treat analyses. Other missing data were not replaced. All tests were two sided, with P values of 0.05 considered statistically significant. We made no adjustment for multiplicity, therefore 95% CIs should not be interpreted as confirmatory for hypothesis testing. Analyses were performed with SAS (version 9.4) and R lqmm and boot packages (version 4.5.2).Patient and public involvement Although we recognise the importance of patient and public involvement, no patients were directly involved in the design, conduct, reporting, or dissemination of this study. The study was designed when patient and public engagement in healthcare research was less common in France, and structured processes for such involvement by national authorities, professional organisations, and research departments were not in place. Nevertheless, the impetus for this study arose from parental concerns regarding the potential misuse of antibiotics in newborn infants and the associated short term and long term health consequences. These issues were explained to parents when their infants were considered eligible for the study. We did not plan to send our results to the research participants; when the study was designed in 2019, no such measure was planned in France. Also, data are anonymised, so it was not possible to send results to all individual patients. However, on publication of the results we plan to disseminate them to public communities and patient organisations.Results Between 15 February 2019 and 7 February 2023, 513 eligible newborns with suspected or proven late onset sepsis were included in 33 centres out of the 36 participating centres. Two newborns did not meet the eligibility criteria at randomisation, leaving 511 patients randomised. Seven participants lacked parental consent, and their data could not be used. As a result, the intention-to-treat population comprised 504 patients: 248 patients were assigned to the procalcitonin-guided group and 256 to the usual care group ( figure 1). Baseline characteristics did not differ between groups (table 1). The per protocol population comprised 215 patients in the procalcitonin-guided group and 248 in the usual care group (figure 1). Reasons for exclusion from the per protocol population are reported in online supplemental table S2.Figure 1Procalcitonin-guided decision and antibiotic treatment duration in late onset sepsis of newborns (ProABIS) trial flow chartTable 1Baseline characteristics of the intention-to-treat populationCharacteristicsProcalcitonin-guided group(n=248)Usual care group(n=256)Sex Male131 (52.8)140 (54.7) Female117 (47.2)116 (45.3) Median (IQR) gestational age (weeks)28.3 (26.4-30.6)28.4 (26.7-30.9) Median (IQR) birthweight (g)1007.5 (800.0-1310.0)1007.5 (793.0-1335.0)Class of gestational age (weeks) <28115 (46.4)107 (41.8) ≥28 to <3290 (36.3)106 (41.4) ≥32 to <3732 (12.9)35 (13.7) ≥3711 (4.4)8 (3.1)Median (IQR) gestational age at inclusion (weeks)31.0 (28.9-34.0)31.3 (29.1-33.6)Median (IQR) weight at inclusion (g)1215.0 (952.5-1647.5)1227.5 (977.5-1617.5)Median (IQR) number of days of life at inclusion14.0 (10.0-24.0)13.0 (9.0-23.0)Main symptoms at diagnosis of primary sepsis Respiratory distress114 (46.0)116 (45.3) Apnoea/bradycardia151 (60.9)162 (63.3) Hypothermia/hyperthermia67 (27.0)57 (22.3) Hypotonia*68 (27.4)79 (31.0) Digestive signs (vomiting, meteorism)68 (27.4)98 (38.3) Rectal bleeding19 (7.7)29 (11.3) Haemodynamic signs (mottling, hypotension, discolouration)85 (34.3)87 (34.0)Invasive treatments at inclusion Central line196 (79.0)211 (82.4) Median (IQR) duration of central line (days)9.5 (6.0-15.0)8.0 (6.0-14.0) Invasive ventilation68 (27.4)67 (26.2) Thoracic drainage5 (2.0)4 (1.6) Other prosthesis52 (21.0)52 (20.3) Surgery in the preceding 72 hours3 (1.2)5 (2.0)Presumed infection site Septicaemia173 (69.8)166 (64.8) Enterocolitis20 (8.1)29 (11.3) Pulmonary19 (7.7)17 (6.6) Urinary tract3 (1.2)5 (2.0) Other2 (0.8)0 (0.0) Unknown site31 (12.5)39 (15.2)Inflammation before randomisation Procalcitonin assay151 (60.9)166 (64.8) Median (IQR) maximal procalcitonin value (µg/L)†1.6 (0.7-6.1)1.7 (0.7-8.7) Median (IQR) maximal C reactive protein value (mg/L)‡39.5 (11.1-69.0)33.2 (12.0-76.6)Data are number (%) unless otherwise stated*One missing value in the usual care group.†Two missing values in the procalcitonin-guided group.‡34 missing values in the usual care group and 28 in the procalcitonin-guided group. 52 of 62 missing value are values under detection threshold.IQR, interquartile range.At randomisation, 235 patients (94.8%) in the procalcitonin-guided group and 233 (91.0%) in the usual care group had samples taken for a blood culture, which was positive in 66.8% and 62.2%, respectively. The most frequently isolated bacteria in blood culture or other samples were coagulase negative Staphylococcus (56.5% and 51.2%, respectively), Staphylococcus aureus (9.2% and 9.1%) and Enterobacteriaceae (9.2% and 9.5%). A full list of species isolated is provided in online supplemental table S3.Primary outcome Procalcitonin-guided decision making significantly reduced the duration of antibiotic treatment for primary sepsis and recurrent sepsis. In the intention-to-treat population, the median duration was 8.0 days (interquartile range (IQR) 5.0-12.0) in the procalcitonin-guided group versus 10.0 days (IQR 8.0-13.0) in the usual care group (P<0.001). The absolute difference was −2.0 days (95% CI −3.8 to −1.0, P<0.001), representing a 20% reduction in favour of the procalcitonin-guided group. Data were missing for one patient in the usual care group, owing to premature discontinuation on day 2 after investigator decision for deep-seated infection with haemodynamic instability. Results in the intention-to-treat population dataset with available data for the primary outcome are identical as results in the intention-to-treat population. In the per protocol population, procalcitonin guidance was associated with a 30% reduction in antibiotic duration, with a median difference of −3.0 days (95% CI −4.0 to −1.0, P<0.001) compared with usual care group ( table 2, figure 2). Sensitivity analyses accounting for centre effects confirmed these findings, with a median difference of −2.0 days (95% CI −3.2 to −0.9, P<0.001) in the intention-to-treat population and −3.0 days (95% CI −3.5 to −1.0, P<0.001) in per protocol population. Post hoc subgroup analysis according to the periods of substantial changes of the protocol design, especially the changes in participants’ weight and the changes to the window of randomisation, led to similar results (online supplemental table S1b).Figure 2Dot plot distribution for the primary outcome (antibiotic duration at day 28) in the intention-to-treat population and per protocol population of the procalcitonin-guided decision and antibiotic treatment duration in late onset sepsis of newborns (ProABIS) trialTable 2Intention-to-treat and per protocol analyses for the primary outcome (antibiotic duration at day 28) and key secondary outcome (mortality at day 28) of the procalcitonin-guided decision and antibiotic treatment duration in late onset sepsis of newborns (ProABIS) trialOutcomesNoProcalcitonin-guided groupNoUsual care groupUnadjusted difference(95% CI)Adjusted difference(95% CI)Intention-to-treat analysis Median (IQR) duration of antibiotic treatment (days)2488.0 (5.0-12.0)25610.0 (8.0-13.0)−2.0 (−3.8 to −1.0)*−2.0 (−3.2 to −0.9)* Mortality at day 282486 (2.4)25610 (3.9)−1.5 (−5.0 to 1.8)†−1.5 (−4.7 to 1.8)†Per protocol analysis Median (IQR) duration of antibiotic treatment (days)2157.0 (5.0-12.0)24810.0 (8.0-13.0)−3.0 (−4.0 to −1.0)*−3.0 (−3.5 to −1.0)* Mortality at day 282155 (2.3)24810 (4.0)−1.7 (−5.3 to 1.8)†−1.7 (−5.0 to 1.6)†Data are number (%) unless otherwise stated.*P<0.001.†Units for difference are % (95% CI).CI, confidence interval; IQR, interquartile range.Key secondary outcome In the intention-to-treat population, six patients (2.4%) in the procalcitonin-guided group and 10 (3.9%) in the usual care group had died by day 28 after randomisation. Vital status was available for all patients. The absolute difference between groups was −1.5% (95% CI −5.0% to 1.8%). In the per protocol population, the absolute difference was −1.7% (95% CI −5.3 to 1.8) ( table 2). Non-inferiority was demonstrated in both populations, as the upper bound of the two sided 95% CI was below the prescribed 3% margin (figure 3). Sensitivity analyses accounting for centre effects confirmed these findings, with absolute difference of −1.5% (95% CI −5.0% to 1.6%) in the intention-to-treat population and −1.7% (95% CI −5.0% to 1.6%) in the per protocol population.Figure 3Visualisation of the key secondary outcome (difference of mortality percentage (procalcitonin-guided group minus control group at day 28) in intention-to-treat and per protocol population and assessment of non-inferiority (margin of 3%) in the procalcitonin-guided decision and antibiotic treatment duration in late onset sepsis of newborns (ProABIS) trialOther secondary outcomes In the intention-to-treat population with available data, recurrence of primary sepsis occurred in seven of 248 patients (2.8%) in the procalcitonin-guided group and 10 of 256 (3.9%) in the usual care group, a between group difference of −1.1% (95% CI −4.6% to 2.3%).The proportion of patients with another infection within 28 days requiring antibiotic treatment was 24.6% (n=61) in the procalcitonin-guided group and 21.1% (n=54) in the usual care group (difference 3.5%, 95% CI −3.9% to 11.0%, P=0.35). Among these, 23.1% of patients (n=56) in the procalcitonin-guided group and 20.0% (n=49) of patients in the usual care group received antibiotics for more than three days (difference 3.1%, 95% CI −4.3% to 10.5%). Treatment after day 28 was given to 27 patients (10.9%) in the procalcitonin-guided group and 31 (12.1%) in the usual care group. This included five patients (2.0%) and four patients (1.6%), respectively, who were treated for primary sepsis or recurrence (online supplemental table S4).Subgroup analyses In the analyses for newborns with culture positive sepsis and culture negative sepsis, procalcitonin-guided decision making resulted in a significant reduction in the duration of antibiotic treatment for primary sepsis and recurrent sepsis when blood culture was positive ( table 3). In both the intention-to-treat and the per protocol populations, the median duration of antiobiotic treatment was 9.0 days (IQR 6.0-13.0) in the procalcitonin-guided group versus 11.0 days (IQR 9.0–14.0) in the usual care group (between-group absolute difference −2.0, 95% CI −4.0 to −1.0). In the intention-to-treat population with culture positive sepsis, two patients (1.3%) in the procalcitonin-guided group had died by day 28 after randomisation, and five (3.4%) had died in the usual care group (between group absolute difference −2.2%, 95% CI −6.8% to 1.6%).Table 3Intention-to-treat and per protocol analyses for duration of antibiotic treatment and mortality for newborn babies with culture positive sepsis and culture negative sepsisOutcomeNo*Procalcitonin-guided groupNo*Usual care groupDifference (95% CI)Median (IQR) duration of antibiotic treatment (days) Blood culture positive sepsis (intention-to-treat)1579.0 (6.0-13.0)14511.0 (9.0-14.0)−2.0 (−4.0 to −1.0) Blood culture negative sepsis (intention-to-treat)787.0 (5.0-9.0)878.0 (7.0-10.0)−1.0 (−2.5 to −1.0) Blood culture positive sepsis (per protocol)1379.0 (5.0-13.0)14411.0 (9.0-14.0)−2.0 (−4.0 to −2.0) Blood culture negative sepsis (per protocol)696.0 (5.0-9.0)828.0 (7.0-10.0)−2.0 (−3.0 to −1.0)Mortality at day 28 (No (%) of newborn babies) Blood culture positive sepsis (intention-to-treat)1572 (1.3)1455 (3.4)−2.2 (−6.8 to 1.6)† Blood culture negative sepsis (intention-to-treat)784 (5.1)874 (4.6)0.5 (−7.0 to 8.9)† Blood culture positive sepsis (per protocol)1371 (0.7)1445 (3.5)−2.7 (−7.3 to 1.0)† Blood culture negative sepsis (per protocol)694 (5.8)824 (4.9)0.9 (−7.3 to 10.1)†*Where data were available.†Units for difference are % (95% CI).CI, confidence interval; IQR, interquartile range.Results were unchanged when we performed analyses according to the presence of coagulase negative Staphylococcus identified in blood cultures (online supplemental table S5).In subgroup analysis according to gestational age at birth in the intention-to-treat population, we observed no significant difference in duration of antibiotic treatment between groups when gestational age was less than 28 weeks (9.0 days (IQR 6.0-14.0) in procalcitonin-guided group versus 10.0 days (IQR 8.0-13.0), between group absolute difference −1.0 days (95% CI −2.0 to 1.0)). The difference of treatment duration was in favour of procalcitonin-guided group after 28 weeks of gestation (absolute difference between groups at gestation between 28 weeks and 33 weeks plus six days, −3.0 days (95% CI −4.0 to −2.0); and at ≥34 weeks’ gestation, −5.0 days (95% CI −6.5 to −2.0) (online supplemental table S6). Results were similar in the per protocol analyses. No differences in mortality were observed between groups at any gestational age.Serious adverse events Seventy-two patients experienced at least one serious adverse event: 37 (14.9%) in the procalcitonin-guided group and 35 (13.7%) in the usual care group (difference 1.2, 95% CI −5.0 to 7.5). Three patients in the procalcitonin-guided group and two in the usual care group had bronchopulmonary dysplasia at day 28. Serious adverse events, excluding death, are given in online supplemental table S7.Discussion Principal findings The ProABIS trial demonstrates that procalcitonin-guided decision can safely reduce the duration of antibiotic treatment in newborns with proven or suspected late onset sepsis. We observed a 20% reduction in antibiotic treatment duration in the procalcitonin-guided group, and a 30% reduction in the per protocol population, without evidence of increased 28 day mortality in any of the randomised populations. Mortality was also similar between groups among newborns with culture positive sepsis. Recurrence rates and secondary infection rates did not differ between groups.Strengths and limitations of this study A strength of our study was its multicentre randomised design and the relatively low rate of protocol non-adherence (10%), which overcomes a longstanding clinical concern about the risk of undertreating or overtreating sepsis. Although the required weight at inclusion was ≥700 g (because of blood volume requirements for procalcitonin assays), the cohort was representative of newborns admitted to intensive care, with gestational ages ranging from 24 to 41 weeks and birthweights from 510 g to 4500 g. We excluded newborns with very short treatment courses (<2 days) based only on non-specific clinical signs, and focused instead on patients with severe infection for whom clinicians decided prolonged antibiotic treatment was needed. This strategy likely explains the high proportion of culture positive cases of sepsis (almost two thirds of the total cohort) in our study, a greater proportion than that reported in the literature. Because late onset neonatal infections are frequently caused by coagulase negative staphylococci, clinicians rely on the severity and nature of the clinical symptoms, as well as their evolution over time to differentiate true infection from sample contamination. Additional investigations such as imaging and/or biomarkers are also used to better estimate the probability of infection after tests are completed. Paired blood cultures are not easy to carry out in newborns with very low birthweight, as their catheters rarely allow blood return, and the volume of blood that can be drawn must be kept to an absolute minimum.A limitation of our study is that, despite a reasonable initial estimate of each centre’s recruitment potential, inclusion reached only 23.7% of the initial target, with 511 randomised patients out of the 2162 required to demonstrate non-inferiority in terms of mortality. As underrecruitment is common in neonatal randomised controlled trials,20 protocol amendments were introduced to support enrolment (online supplemental table S1). We found that mortality was lower in the procalcitonin-guided group, and the non-inferiority margin for mortality (3%) was not reached. But as the number of patients randomised was lower than expected, non-inferiority for mortality at day 28 should be interpreted with caution. However, a recent study of 82 850 infants across nine population based networks in 10 countries reported a 20.4% incidence of late onset sepsis, with a mortality of 10.9% temporally related to late onset sepsis.18 Thus, even if the upper bound of the 95% CI for mortality exceeded the expected 6.1%, it could still be considered acceptable in this context.Secondly, we did not observe any benefit from the procalcitonin-guided strategy in post hoc analysis of the subgroup of patients born before 28 weeks gestation. Continuation of antibiotic treatment with procalcitonin value of 0.5 µg/L or lower was allowed if the patient's condition was deemed clinically unstable. Frequent non-specific signs in very preterm infants, such as respiratory distress, apnoea, and bradycardia, probably discouraged clinicians from discontinuing antibiotic treatment, despite low procalcitonin values in these patients.Thirdly, although the proportion of patients with recurrence of primary sepsis was very low and not different between groups (1.6 v 2.0%), we could not conclude to non-inferiority because our assay was not designed to study non-inferiority for recurrent sepsis. However, a multicentre randomised controlled non-inferiority trial is currently ongoing in India (NCT03280147), comparing the efficacy of a seven day versus a 14 day course of intravenous antibiotics to treat uncomplicated bacterial sepsis in newborns. The primary outcome measure is a definite or probable recurrence within 21 days after the end of antibiotic treatment. For this trial, the authors assumed a high rate of recurrence of 10%, which is much higher than the assumed recurrence rate we reported. Nevertheless, our study suggests that the proportion of patients with recurrence in our two study groups was not clinically concerning.Finally, newborns with severe infections such as meningitis, septic shock, or deep-seated infection were excluded for pragmatic and safety reasons, as early discontinuation in such cases would be considered unsafe and unlikely to be adopted in practice. These situations are rare in newborn babies and account for only a small proportion of antibiotic prescriptions.Comparison with other studies Our results are comparable to those reported in a population of newborns with early onset sepsis and also in populations of critically ill adults with a reduction in antibiotic duration and the risk of mortality. In the PRORATA trial of 630 adult patients in intensive care, patients in the procalcitonin-guided group had a 23% reduction in antibiotic exposure and received antibiotics for 10 days, compared with 13 days in the control group. 21 In the SAPS trial comprising 1546 adult critical care patients with assumed or proven infection, procalcitonin-guided decision making resulted in a 25% reduction in antibiotic duration from seven days in the standard care group to five days in the procalcitonin-guided group, without an increase in mortality.22 A recent meta-analysis of randomised controlled trials comparing the effects of procalcitonin-guided antibiotic therapy to standard care in critically ill adult patients with proven or suspected infection treated with antibiotics concluded that procalcitonin-guided antibiotic therapy may be associated with reduced antibiotic use (−1.79 days, 95% CI −2.65 to −0.92) and a 16% reduction in 28 day mortality.14 This meta-analysis also reported a 36% higher recurrence of infection in the procalcitonin-guided antibiotic treatment. In contrast, our study did not show any difference between groups in primary sepsis recurrence, nor did it detect differences in secondary infection.14 The Neonatal Procalcitonin Intervention Study (NeoPInS), an international randomised controlled trial, is the only study in newborns that evaluated the efficacy of procalcitonin-guided decision on the duration of antibiotic treatment. The procalcitonin-guided group showed about a 20% reduction in the duration of antibiotic therapy and about a 5% reduction in the length of hospital stay compared to control, although non-inferiority for reinfection and death could not be demonstrated owing to the small number of events.16 Newborns included in this study were born after 34 weeks of gestation and had suspected early onset sepsis only with a very low risk of recurrence, which differs from the ProABIS study, where the inclusion criteria were 24-41 weeks' gestation at birth, of which more than 80% of newborns were born before 32 weeks and two thirds had proven late onset sepsis.Policy implications Procalcitonin-guided antibiotic stewardship in late onset neonatal sepsis is a promising strategy to individualise treatment and limit antibiotic exposure in this vulnerable population. This strategy could reduce the length of antibiotic treatment in newborns, especially in neonatal units where routine antibiotic treatment duration exceeds seven days. The efficacy of this approach aiming to limit antibiotic treatment duration will vary depending on local practice and will be of less benefit in a setting with an already low background use of antibiotics. An unexpected finding of our study was the reduction in duration of antibiotic treatment by one day when the blood culture was negative and that the treatment lasted between seven and eight days by group. Antibiotic treatment is often continued even if the blood culture is negative, because clinicians may have doubts about the quality of the blood culture (owing to the volume sampled), but also because of the clinical condition of most preterm newborns, who present non-specific symptoms which are attributed by default to an infection. 23 The procalcitonin-guided strategy offers a way to optimise antibiotic therapy duration in late onset sepsis and could guide antibiotic stewardship in neonatology departments.10Conclusions This study demonstrates the efficacy of using a procalcitonin-based strategy to reduce the duration of antibiotic treatment in neonatal late onset sepsis, without increasing mortality, although patients with meningitis, deep-seated infections, and septic shock were excluded.