Results

Site recruitment

Recruitment was open between 15 December 2020 and 12 October 2022. The last sample from the last participant was collected on 25 November 2022.

Participant recruitment

In total, 545 participants were screened for the PRECISE study. Of these, 95 were not approached. The main reasons for this were that parents had already mentioned that they were not happy for additional blood samples to be taken from their child (n = 24), time constraints (n = 19) or parents were too distressed to take in information about the PRECISE study (n = 15). The number of participants approached but not recruited was 44, with 24 of these declining as they did not want additional blood samples to be taken. A total of 406 participants were consented into the PRECISE study.

Samples were also included from participants recruited into the main BATCH trial (n = 261) (from the PCT arm) who gave consent for their samples to be stored and used for future research.

A detailed PRECISE CONSORT flow diagram is provided in Figure 4.

Sample processing

A total of 406 participants were consented into the PRECISE study (203 from the PCT arm, 203 from UC arm). Not all participants who consented to take part in the PRECISE study had samples collected. The study was pragmatic and aimed to collect samples when routine blood samples were being taken. If routine blood collection was not due, permission was sought from parents before any additional samples were taken. This was not always given.

Additional samples were included from 261 participants recruited into the main BATCH trial prior to commencement of the PRECISE study, who gave consent for their samples to be stored and used for future research. These BATCH samples were from the PCT arm only within the BATCH trial.

In total, 1872 samples were collected and stored by the local NHS laboratories for transfer to the University of Liverpool laboratory for analysis. Of the 1872 samples collected, 150 were either not stored or not transferred to the University of Liverpool research laboratory due to there not being enough quantity leftover after routine tests or BATCH PCT tests had been carried out, or, for example, the sample had haemolysed.

The samples were transferred to the University of Liverpool research laboratory. The exact order of tests performed depended on the quantity of the sample and were carried out in accordance with the hierarchy described in Laboratory analysis.

There were several issues encountered in the research laboratory during the sample processing, including for example; with the sample identification (e.g. patient identifiers were not decipherable); where samples had to be discarded due to there not being enough sample to process, or calibration issues meaning some results could not be interpreted and therefore, could not be included in the statistical analysis. There were, however, 321 samples with valid MRpro-ADM results that were able to be included in the statistical analysis. The analyses were repeated for TRAIL, IP-10, CRP and MRpro-ADM tests where there was sufficient sample to process

The final PRECISE data set contained a total of 978 observations from 356 participants.

Table 2 shows the number of participants with one or more measurements. A total of 120 participants had two measurements from the samples collected, and 64 had three measurements.

The number of participants included in the final analysis with a baseline sample available (taken within ±1 day of the randomisation date) was 305 (235 in the PCT arm and 70 in the UC arm). The imbalance between arms is a result of the additional PCT arm only samples included from the BATCH participants. For each biomarker, the number of participants was 147 for MR-proADM, 255 for TRAIL, 256 for IP10 and 258 for CRP.

Statistical analysis

Descriptives

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. Patient baseline characteristics are summarised by trial arm in Table 3. Descriptive summaries of biomarker values at baseline are presented by trial arm in Table 4. Histograms of biomarker values at baseline are shown in Figures 5–8.

FIGURE 5 C-reactive protein and log(CRP) at baseline. (Figure not reproduced here; see the original article.)

FIGURE 6 Tumour necrosis factor-related apoptosis-inducing ligand and log(TRAIL) at baseline. (Figure not reproduced here; see the original article.)

FIGURE 7 Interferon gamma-induced protein 10 and log(IP-10) at baseline. (Figure not reproduced here; see the original article.)

FIGURE 8 Mid-regional proadrenomedullin and log(MR-proADM) at baseline. (Figure not reproduced here; see the original article.)

Main analysis

Table 5 presents the χ² test statistics and p-values associated with the test for interaction of intervention arm and biomarker for each outcome and each biomarker (either included as a continuous variable or dichotomised using pre-specified cut-offs). Unadjusted p-values suggest potential interaction effects between BATCH intervention arm and CRP on antibiotic duration (IV, total, broad-spectrum), indicating a tendency for antibiotic durations to be shorter in the PCT arm than in the UC arm when CRP is high, and vice versa when CRP is low. However, after a Benjamini–Yekutieli correction for multiple testing, none of these effects remains statistically significant at the 5% level. For the other biomarkers (TRAIL, IP-10, MR-proADM) and outcomes (time to discharge from hospital), not even unadjusted p-values are statistically significant. No analysis models could be fitted for SADR and HAI due to the very small numbers of such events in the data set.

Additional analyses

Descriptive summaries of locally determined CRP values at baseline are presented by trial arm for the full BATCH data set in Table 6. Histograms of biomarker values at baseline for the full BATCH data set are shown in Figure 9. The correlation between baseline log(CRP + 0.1) values analysed in local labs as part of the main BATCH trial versus centrally for the PRECISE study was 0.44 (Pearson’s correlation) or 0.51 (Spearman’s rank correlation).

FIGURE 9 C-reactive protein and log (CRP) longitudinally over time in relation to time of randomisation. (Figure not reproduced here; see the original article.)

FIGURE 10 Tumour necrosis factor-related apoptosis-inducing ligand and log(TRAIL) longitudinally over time in relation to time of randomisation. (Figure not reproduced here; see the original article.)

FIGURE 11 Interferon gamma-induced protein 10 and log(IP-10) longitudinally over time in relation to time of randomisation. (Figure not reproduced here; see the original article.)

FIGURE 12 Mid-regional proadrenomedullin and log(MR-proADM) longitudinally over time in relation to time of randomisation. (Figure not reproduced here; see the original article.)

FIGURE 13 Scatter plot and Bland–Altman plot of baseline log (CRP + 0.1) values analysed in local labs vs. centrally. (Figure not reproduced here; see the original article.)

FIGURE 14 C-reactive protein and log(CRP + 0.1) at baseline (full BATCH data set). (Figure not reproduced here; see the original article.)

Table 7 presents the χ² test statistics and p-values associated with the test for interaction of intervention arm and biomarker for each outcome for CRP (either included as a continuous variable or dichotomised using pre-specified cut-offs). Unadjusted p-values suggest potential interaction effects between BATCH intervention arm and CRP on broad-spectrum antibiotic duration only when using a cut-off of 25 mg/l, indicating a tendency for antibiotic durations to be shorter in the PCT arm than in the UC arm when CRP is high, and vice versa when CRP is low. However, after a Benjamini–Yekutieli correction for multiple testing this effect does not remain statistically significant at the 5% level. For the other outcomes, not even unadjusted p-values are statistically significant.