Summary of results
The PRECISE analysis data set included 305 patients, 235 of which were randomised in the BATCH PCT arm and 70 in the BATCH UC arm. The imbalance is due to the fact that in the intervention arm of the BATCH trial, participants had an extra blood sample for PCT testing whereas the UC participants did not. The participants consented to samples being stored for further research so they were able to be used in the intervention arm participants, whereas UC participants did not have extra blood samples which could be stored. This created an imbalance in the intervention and UC arms.
Median age across both arms was similar, but there was a female preponderance in the UC arm versus the PCT arm (51% vs. 40%).
Using the plasma measurements of individual biomarkers TRAIL, IP10, CRP and MR-proADM, we observed an interaction between the BATCH PCT intervention arm and CRP for antibiotic duration. For CRP values measured on the MSD multiplex immunoassay in the research laboratory, the interaction was significant for high versus low CRP (both when using CRP as a continuous measure and when dichotomising using cut-offs of CRP > 25 mg/l or CRP > 100 mg/l) for IV antibiotic duration, total antibiotic duration (oral and IV) and duration of IV broad-spectrum antibiotics. However, after statistically adjusting for multiple testing, these effects were no longer statistically significant at the 5% type I error level. For CRP values measured as part of routine care using different NHS laboratory analysers at each recruitment site, the interaction was significant only for CRP > 25 mg/l for broad-spectrum antibiotic duration, but again did not remain significant following multiplicity correction. These results are clinically plausible from an infectious diseases clinician’s point of view; when the CRP value remains high, clinicians are more inclined to continue antibiotics for longer, and conversely when CRP values are low, clinicians are more confident to stop antibiotics.
The correlation between NHS laboratory measured CRP values and CRP values analysed centrally in the research laboratory using MSD multiplex immunoassay was weak. This may explain why the previously observed interaction was not replicable and suggests that combining CRP values from several different laboratory analysers might not be robust.
There was no significant interaction observed for IP-10, TRAIL or MR-proADM, but especially the number of participants with MR-proADM measurements was too low to make any definite conclusions.
Previous research
A randomised controlled interventional multicentre study of adults with suspected infection presenting to four EDs in Spain demonstrated that a MR-proADM-guided algorithm resulted in lower hospital admission in the intervention arm (58% vs. 41%) with no significant difference in safety end points.46 ED length of stay and hospital length of stay was not significantly different between intervention arms. There was a weak but significant correlation between the MR-proADM and hospital length of stay in a small study of adults with sepsis and septic shock.47 MR-proADM predicted hospital length of stay in COVID-19-hospitalised adult patients.48 Admission MR-proADM correlated with hospital length of stay in children with heart failure.49
Feedback of CRP-guided and PCT-guided treatment algorithms reduced the number of days on antibiotic in the first 30 days after hospital admission in adults with community-acquired pneumonia.50 In a small trial of PCT-guided early discontinuation of antibiotics in low-risk febrile neutropenia, the median duration of antibiotics in the intervention arm was significantly shorter (3 days vs. 7 days).51
A recent network meta-analysis compared PCT-guided versus CRP-guided and UC and reported that PCT- or CRP-guided antimicrobial cessation strategies may shorten antimicrobial therapy without increased mortality or recurrence. From 18 RCTs, they found that a PCT-guided method reduced antibiotic treatment by 1.9 days, and a CRP-guided approach reduced antibiotic treatment by 2.6 days. In particular, the usefulness of PCT guidance for sepsis where antimicrobials are used for more than 7 days was supported. There were very few RCTs where CRP-guided strategies were compared directly with UC and with PCT-guided strategies; therefore, it concluded that the evidence was not robust.52
It has been reported that absolute concentrations of biomarkers including CRP obtained by single-biomarker versus multiarray techniques (such as MSD) differed considerably, which indicates that, when comparing absolute values, realignment of the single-biomarker data to the multiarray data (or vice-versa) is warranted.53
Strengths and weaknesses
The PRECISE substudy was embedded within the HTA-funded BATCH trial, and, therefore, represented significant added value, to identify potential subgroups of patients who are likely to benefit from a PCT-guided antibiotic algorithm. Given the negative finding in the main BATCH trial, this additional information could have been particularly useful. The initial results showing an interaction between CRP and the intervention arm suggest that stratification by CRP might be helpful in reducing IV antibiotic duration. However, it is to be noted that these findings were no longer statistically significant after correcting for multiple testing.
Weaknesses of the study are that samples were taken at different time points after randomisation, so it was difficult to determine if values were contemporaneous and how results might have influenced clinical decisions if they had been available to clinicians at the same time.
In hindsight, the assumptions around effect sizes made for the sample size calculation may have been overly optimistic, assuming quite large differences in opposite directions for the different subgroups. The imbalance between PCT (n = 235) and UC arm (n = 70) will have affected the study power further.
The weak correlation between local NHS laboratory CRP values and the central research laboratory MSD multiplex immunoassay may result from different antibody target capture of the different immunoassay methods or the lack of realignment of absolute values using for example, weighted Deming regression. The weak correlation between the multiplex CRP assays and the laboratory CRP assay does bring into question the validity of the multiplex assay, but the interaction effects observed for both assays are in the same direction.
There were a large number of samples with insufficient plasma or with failed laboratory assay runs, but as these samples were missing at random, it is unlikely that this affected the results.
Findings of interest
The CRP subgroup findings prior to statistical adjustment for multiplicity suggest that in children with a CRP > 25 mg/l, the addition of a PCT-guided algorithm could potentially lead to a reduction in duration of IV antibiotics. If the CRP is high, clinicians might be inclined to continue IV antibiotics, but the presence of a low or falling PCT might give them the confidence to stop IV antibiotics and thus result in a shorter IV antibiotic duration. Conversely, if the CRP is low, clinicians might be inclined to stop IV antibiotics, but the presence of a high, rising or static PCT might make them a bit more cautious about stopping, therefore resulting in a longer duration of IV antibiotics.
Impact and learning
The issue with this trial was not the operational challenge of embedding a mechanistic study within a trial, but that:
- additional sampling in children is invasive and distressing so we had to be opportunistic with collecting samples with other routine blood tests
- the original assay that we planned to use was the ImmunoXpert multiplex test from MeMed, which was only valid on serum samples and we only collected plasma samples. We therefore had to use a custom-made multiplex immunoassay from Mesoscale discovery, which was fraught with technical challenges and issues to do with replicates and calibration issues.
Future trials embedding mechanistic studies should ensure that robust Good Laboratory Practices are followed and that all biomarkers measurements are standardised.
Antimicrobial stewardship is a multilevel activity, influenced by a range of societal, regulatory, organisational, service-level, professional and individual factors.54 The use of diagnostic tests to support antibiotic prescribing decisions requires a detailed understanding of the use of the diagnostic tests in context, including fidelity of intervention, how clinicians use and interpret the tests and algorithms, how patients are involved, and how services adapt to accommodate the new diagnostic tests. Biomarker-guided antimicrobial prescribing algorithms are complex interventions, where the clinical context, patient and sample flow and timing of testing in the care pathway need to be considered.55 Pragmatic clinical trials need to build the evidence base for complex interventions to increase clinical confidence in diagnostic testing and improve AMS, with embedded economic and implementation evaluation. It is imperative that diagnostic tests are offered at the right time, to the right person, in the right way and that results are interpret correctly, to improve clinical decisions about the management of severe infection.
Recruiting to large trials is costly and time-consuming, and traditional trial designs frequently produce inconclusive results due to a fixed sample size. Adaptive platform designs adopt a long-term strategy for a perpetual integrated research system, which builds on the initial investment (in time and money) required to set up a single trial (investigator network, governance and regulatory framework, data management systems), to create a long-term ‘single point of access’ for collaboration, discovery and improvements in care.56 Clinical and methodological expertise are layered on top of this foundation and applied across multiple trials to enhance processes and the generation of knowledge. The perpetual platform trial design provides the opportunity to rapidly learn what interventions are effective and eliminate those that are not effective, avoiding further wasting of money, and facilitating earlier adoption into clinical practice. Consequently, a larger upfront investment is required for adaptive platform trials, but as a perpetual collaborative research system, they provide much more efficient and cost-effective design for collaboration, discovery and improvements in care over the long term.
Implications for decision-makers
As the main BATCH trial did not demonstrate any effect on IV antibiotic duration, there are limited implications for policy.
Research recommendations
Embedded mechanistic studies provide value for money in that they use existing trial infrastructure and can help to identify subgroups of patients for whom the intervention may or may not be effective. A better understanding of the complex interactions influencing whether/how/why clinicians act on information from diagnostic tests to make antibiotic prescribing decisions will improve trial intervention fidelity and facilitate implementation and adoption of tests shown to be effective. Future studies of biomarker-guided interventions should utilise adaptive platform trial designs embedded in routine clinical care to comprehensively evaluate multiple diagnostic tests, to establish clinical utility, safety, cost-effectiveness and implementation outcomes robustly and rapidly.
Conclusions
In hospitalised children treated with IV antibiotics for suspected or confirmed SBI, a PCT-guided algorithm is not effective in reducing IV antibiotic duration overall, but there is some preliminary evidence that in the subgroup of children with a CRP > 25 mg/l, the addition of a PCT-guided algorithm could potentially lead to a reduction in duration of IV antibiotics.