Comment

Reducing just-in-case primary care antibiotic prescribing in children

Alastair D Hay, Emily Brown

Lancet 2025; 406: 1538-1540

As health-care workers around the world will attest, infections are extremely common in children in primary care and community settings.1, 2 In The Lancet, Jan Yvan Verbakel and colleagues' pragmatic, cluster-randomised, controlled trial (ARON),³ set in a high-income primary care system, reports the culmination of years of rigorous research by a team seeking what many said was impossible: to safely rule out serious infection in children who present with acute illness to primary care, where its incidence is around 1%,⁴ without giving children unnecessary antibiotics or increasing hospital referrals.

Solutions to balance patient safety and antimicrobial stewardship can be considered in two groups: clinical decision tools to identify the children requiring treatment and prescribing strategies to allow parents to start antibiotics when necessary. Neither solution is perfect. To keep patients safe, clinical decision tools must be highly sensitive, whereas reducing unnecessary antibiotic use requires high specificity. However, few clinical decision tools achieve both. For example, if used in primary care, the UK's National Institute for Health and Care Excellence traffic-light clinical decision tool⁵would result in 32% of feverish children (in the red [high-risk] category) being immediately referred to hospital, and a further 63% (in the amber [intermediate-risk] category) being considered for admission.⁶ Prescribing strategies include delayed prescribing (otherwise known as backup or standby prescribing), in which patients with self-limiting infections are given a prescription but advised to use them only if symptoms worsen. Although internationally recommended,7, 8 the key weakness of this approach is that once dispensed, there is no constraint on when or to whom the antibiotic is given. This factor is also true of just-in-case prescribing, in which non-treatment is perceived as too risky by clinicians, particularly among clinically vulnerable groups such as children.⁹

So, what value does the ARON trial³ add? The ARON clinical decision tool consists of a three-item validated decision tree¹⁰—the clinician's gut feeling that something is wrong, dyspnoea, or a temperature of 40°C or higher—to guide the use of C-reactive protein (CRP) point-of-care testing, used together with safety-netting advice. If decision-tree responses are negative, and the clinician is not considering antibiotics, only safety-netting advice is provided. If the decision tree is negative but the clinician is considering antibiotics, CRP is tested, and safety-netting advice is provided if the result is <5 mg/L. If the CRP is ≥5 mg/L, the clinician gives safety-netting advice and only prescribes antibiotics if indicated by local guidelines. If any decision-tree response is positive and the CRP is <5 mg/L, the clinician provides safety-netting advice and only prescribes according to local guidelines. If the CRP is ≥5 mg/L, the clinician is advised to consider additional testing or specialist referral.

Effectiveness was investigated with a multicentre, parallel-group, pragmatic, cluster-randomised, controlled trial. Children were recruited if aged 6 months to 12 years and presenting with acute illness (ie, ≤10 days) to one of 171 in-hours general practices (approximately 90%) or community paediatricians (approximately 10%) in Belgium. Children were excluded if they had received antibiotics within 1 week, had a chronic condition (eg, asthma), had previously participated in ARON, needed immediate care, were immunosuppressed, or presented with trauma. There were 82 practices randomised to the intervention group and 89 to the usual care group. The coprimary outcomes assessed initial antibiotic prescribing (superiority tested) and four patient safety outcomes (recovery time, additional testing, follow-up visits, and subsequent antibiotic prescribing within 30 days, all tested for non-inferiority). Data were analysed for 2988 children in the intervention group and 3762 in the control group (3447 [51%] boys, 3302 [49%] girls, and one [<1%] who did not specify a gender, with a median age of 3·8 years [IQR 1·8–7·0]). Unfortunately, the authors were unable to report ethnicity.

The initial antibiotic prescribing reduced from 817 (22%) in the usual care group to 466 (16%) in the intervention group (adjusted absolute risk reduction 5·1% [number needed to treat approximately 20], odds ratio 0·72 [95% CI 0·55–0·94]). Safety outcomes were all non-inferior and adherence to the clinical decision tool was 72–99%. An exploratory analysis found initial hospital referrals were higher in the intervention group compared with the usual care group (24 [1%] of 2988 vs 13 [<1%] of 3762), suggesting those requiring admission might have been identified earlier.

In our view, the strengths of the study include rigorous design and conduct, addressing an important clinical question and providing effect estimates likely to reflect what would be observed in routine care in Belgium. Additionally, the clinical decision tool was rigorously developed and there was no evidence of recruitment bias after randomisation. The intervention also has three key attractive features. First, it avoids indiscriminate CRP testing. CRP was tested after the clinical assessment with the decision tree, ensuring the history, which is known to provide more diagnostic information than examination and investigation combined,¹¹ remained front and centre of the process. This aspect resulted in lower rates of CRP use (733 [25%] of the 2988 children in the intervention group) than if it was promoted for all clinical presentations. Second, decision-tree guided CRP testing resulted in a much higher proportion (535 [73%]) of positive tests than the 10–20% positive rate seen in previous trials,¹² albeit using a lower positivity threshold (≥5 mg/L vs ≥20 mg/L). We suggest that the key strength of ARON was the sequential use of the decision tree followed by CRP testing in children responding positively on the decision tree. The decision tree is known to be highly sensitive,⁹ resulting in almost no false negatives (ie, no missed cases of serious infection), but at the cost of increased false positives (ie, some children incorrectly identified as possible serious infection). We speculate that in children responding positively on the decision tree, CRP provided specificity to reduce the false-positive rate and thereby reduce just-in-case prescribing. However, this mechanism cannot be proven without CRP testing all participants. Finally, the intervention refers clinicians to local guidelines if they are considering antibiotic treatment, meaning it could be tailored to other settings.

However, there are potential barriers to implementation. First, if and how clinicians (including wider primary-care team members) with different degrees of experience can learn gut feeling is unclear. Second, not all primary-care services have access to CRP point-of-care testing. Third, there will be uncertainty around likely effect sizes (and therefore cost-effectiveness) in other countries. Finally, ensuring any safety-netting advice is suitable for the locality would be necessary.

Future research should establish the safety, clinical effectiveness, and cost-effectiveness of the intervention in other countries and settings. For example, antibiotic prescribing rates are higher in the UK than in Belgium, where around 50% of children presenting with a respiratory infection receive an antibiotic,¹³ so a similar relative reduction in prescribing would result in a larger absolute effect size (and a smaller number needed to treat). Understanding how the intervention works would also be valuable; for example, confirming that the decision tree is providing sensitivity and CRP specificity.

In the meantime, we suspect just-in-case antibiotic prescribing rates can be safely reduced in many primary-care settings, and we thank the ARON team for providing a tool—and the evidence to support its effectiveness—that can keep children safe today while preserving antibiotic effectiveness for the future.