In-vitro diagnostic point-of-care tests in paediatric ambulatory care: A systematic review and meta-analysis

Introduction Paediatric consultations form a significant proportion of all consultations in ambulatory care. Point-of-care tests (POCTs) may offer a potential solution to improve clinical outcomes for children by reducing diagnostic uncertainty in acute illness, and streamlining management of chronic diseases. However, their clinical impact in paediatric ambulatory care is unknown. We aimed to describe the clinical impact of all in-vitro diagnostic POCTs on patient outcomes and healthcare processes in paediatric ambulatory care. Methods We searched MEDLINE, EMBASE, Pubmed, Cochrane Central Register of Controlled Trials, Cochrane Database of Systematic Reviews, and Web of Science from inception to 29 January 2020 without language restrictions. We included studies of children presenting to ambulatory care settings (general practice, hospital outpatient clinics, or emergency departments, walk-in centres, registered drug shops delivering healthcare) where in-vitro diagnostic POCTs were compared to usual care. We included all quantitative clinical outcome data across all conditions or infection syndromes reporting on the impact of POCTs on clinical care and healthcare processes. Where feasible, we calculated risk ratios (RR) with 95% confidence intervals (CI) by performing meta-analysis using random effects models. Results We included 35 studies. Data relating to at least one outcome were available for 89,439 children of whom 45,283 had a POCT across six conditions or infection syndromes: malaria (n = 14); non-specific acute fever ‘illness’ (n = 7); sore throat (n = 5); acute respiratory tract infections (n = 5); HIV (n = 3); and diabetes (n = 1). Outcomes centred around decision-making such as prescription of medications or hospital referral. Pooled estimates showed that malarial-POCTs (Plasmodium falciparum) better targeted antimalarial treatment by reducing over-treatment by a third compared to usual care (RR 0.67; 95% CI [0.58 to 0.77], n = 36,949). HIV-POCTs improved initiating earlier antiretroviral therapy compared to usual care (RR, 3.11; 95% CI [1.55 to 6.25], n = 912). Across the other four conditions, there was limited evidence for the benefit of POCTs in paediatric ambulatory care except for acute respiratory tract infections (RTI) in low-and-middle-income countries (LMICs), where POCT C-Reactive Protein (CRP) may reduce immediate antibiotic prescribing by a third (risk difference, -0.29 [-0.47, -0.11], n = 2,747). This difference was shown in randomised controlled trials in LMICs which included guidance on interpretation of POCT-CRP, specific training or employed a diagnostic algorithm prior to POC testing. Conclusion Overall, there is a paucity of evidence for the use of POCTs in paediatric ambulatory care. POCTs help to target prescribing for children with malaria and HIV. There is emerging evidence that POCT-CRP may better target antibiotic prescribing for children with acute RTIs in LMIC, but not in high-income countries. Research is urgently needed to understand where POCTs are likely to improve clinical outcomes in paediatric settings worldwide.


Introduction
In LMICs, these factors exist alongside the risk of serious communicable diseases and high childhood mortality rates. Population-level interventions, for example, like the mass roll-out of antibiotics may reduce mortality, but remains controversial and is likely unsustainable in resource-poor settings [20][21][22]. One factor which could help reduce this diagnostic uncertainty is POC technology. POCTs may help to improve diagnostic precision, optimise prescribing and improve the quality of care for children, and, indirectly, relieve pressure on healthcare systems [17][18][19]. Likewise, the benefit of POCTs for long-term conditions shorten the feedback loop by providing an immediate result that allows timely adaptation of treatment [10]. This mitigates against the impact of tardy laboratory results, or results only being actioned at the next consultation. Therefore, when treatment decisions lag behind "real-time", they often become empirical.
Yet, we should be mindful that the complexity of clinical decision-making in children is not the same as in adults. Although the analytical and clinical diagnostic accuracy of POCTs may be broadly similar in adults and children, the clinical effectiveness of using a POCT within a clinical pathway cannot be generalised from adults to children. The clinical needs are distinct. Factors such as parental concern and the potential for rapid clinical deterioration may alter the test's clinical effectiveness and diagnostic value in paediatric populations.
Currently, we do not know the existing evidence base for in-vitro POCTs in children and the clinical impact of this technology on patient outcomes and healthcare processes in paediatric ambulatory care. We therefore performed a systematic review to describe this.

Search strategy and inclusion criteria
The study protocol was published prospectively [23]. This review is a sub-study and evaluates the clinical impact of any in-vitro diagnostic point-of-care test (POCT) in paediatric populations in ambulatory care.
We systematically searched the six main electronic databases (MEDLINE, EMBASE, Pubmed, Cochrane Central Register of Controlled Trials, Cochrane Database of Systematic Reviews, and Web of Science) from database inception to 29 January 2020. With the help of an Information specialist, we used validated search filters for "primary care/ambulatory care", "point of care/rapid test", and "adolescent/child/infant" (example of search strategy S8 Appendix in S1 File).
We included randomised controlled trials (RCTs) and non-randomised studies of children presenting first to ambulatory care settings (general practice, hospital outpatient clinics, or emergency departments, walk-in centres, registered drug shops) where healthcare is delivered and/or POCTs are used. Children were defined by the authors of included studies.
Where studies involved both adults and children, we included studies where we were able to distinguish outcomes of children from adults. We only included studies that examined invitro POCTs that were defined as in-vitro i.e. tests involving blood or other bodily fluid or excreta that have been taken from the human body. Diagnostic POCTs that were not in-vitro (e.g. POC ultrasound) were excluded. Studies were eligible if they compared the POCT with usual care. Usual care could include no testing or central laboratory tests, but not another novel test or diagnostic strategy. We included and distinguished studies where POCTs were used in conjunction with another training or communication strategy.
We included all quantitative clinical outcome data across all conditions or infection syndromes reporting on the impact of POCTs on clinical care and healthcare processes. Outcome data could include: patient outcomes (e.g. mortality; morbidity); decision-making/clinical management decisions (e.g. hospital attendance/referral); medication prescribing (e.g. antibiotic prescribing); and additional diagnostic testing.
We compared studies according to similar condition as stated by authors, study design, and outcomes. Data had to be reported in sufficient detail to compare relevant outcomes between children with similar conditions/illnesses, and a POCT versus usual care.
We excluded health economic outcomes, qualitative studies, diagnostic accuracy studies, studies solely conducted in hospital inpatient settings and hospital-acquired infections. We excluded study designs that precluded comparisons between tested and non-tested participants (case studies, case series, and studies without a suitable control).

Analyses
Two reviewers (OVH, MR) independently screened articles in duplicate at title and abstract, and full-text levels. A third reviewer (GH) resolved any disagreement. The team (MR, GH, JJL, AvB, JV, PT, CRG) extracted data on the characteristics of included studies and assessed quality of included studies based on their respective risk-of bias tool. We used the Cochrane Risk of Bias tool for RCTs [24]. This was extended to accommodate non-randomised studies by including additional parameters such as reporting of baseline characteristics; whether intervention and control groups were similar; and whether there was a detailed description of the usual care pathway [7]. OVH checked data extraction and quality assessment. We contacted corresponding authors for clarification.
We used random effects meta-analyses (where possible) to generate pooled estimates with 95% confidence intervals (CI) for the same condition or infection syndrome. Heterogeneity was assessed using the χ 2 test and I 2 statistic. We calculated risk ratios (RR) for dichotomous outcomes and mean differences for continuous outcomes. Subgroup analyses were performed according to study design (RCTs vs non-randomised studies). We used sensitivity analyses, excluding studies to explore heterogeneity. Results were summarised narratively where data were not sufficient to perform meta-analysis. We used Covidence software [25] for citation management. Meta-analysis was performed with Revman [26] and STATA 14 SE [27].

Results
The searches resulted in 6,860 unique records, of which 163 full-text articles were eligible for inclusion after selection on title and abstract (Fig 1). We excluded 114 studies at the full text stage. The two most common reasons for exclusion after assessing the full text were that studies were conducted in adult populations (n = 30/114) or that clinical outcomes were not reported separately for adult and paediatric populations (n = 26/114). Finally, 49 studies satisfied our selection criteria. A systematic review on the clinical impact of influenza POCTs has been published in 2019 by our research group which included 11 studies (seven RCTs, four non-randomised studies) [6]. We found three further observational influenza studies [28][29][30] which do not change the overall findings of the original influenza review. As a result, we excluded the 14 influenza studies leaving 35 studies for this review.

PLOS ONE
In-vitro diagnostic point-of-care tests in paediatric ambulatory care There were six conditions or infection syndromes (number of studies): malaria (n = 14); non-specific acute fever 'illness' (n = 7); sore throat (n = 5); acute respiratory tract infections (n = 5); HIV (n = 3); and diabetes (n = 1). Data relating to at least one outcome were available for 89,439 children of whom 45,283 had a POCT. Two thirds of these data (66%, 58,987/ 89,439) related to suspected malaria.
We have summarised a brief description of usual care as comparator in Table 1 (extracted from the original text). This was often not clearly defined and, in most cases, this was taken to be a clinical diagnosis with no POCT used. For example, in malarial studies set in LMICs, usual care involved children prescribed antimalarials based on clinical symptoms in outpatient clinics and cared for at home. In the usual care arm, around 90% of children were prescribed antimalarials. In non-clinical settings e.g. registered drug shops, usual care was a presumptive diagnosis of malaria. Three studies used existing laboratory-based testing as comparator as part of usual care [33, 62,63]. In other studies, e.g. acute fever 'illness', antibiotic prescribing in usual care was around 28% and based on clinical assessment (see Fig 8) [31, 50,51,57,62]. In one study, usual care was a decision algorithm modelled on a set of important paediatric signs and symptoms [52].

Locations
Overall, 33 studies were conducted in ambulatory care settings and two malaria studies were conducted in registered drug shops where medicines are dispensed [34,38]. Twenty studies were in LMICs in Africa focussing mainly on malaria or HIV, and the remaining 15 studies in mainly high-income European countries (Table 1).
Of the 20 studies conducted in LMICs, 18 studies were conducted in primary healthcare clinics and two studies in registered drug shops.

Principle method and target analytes of POCTs
A summary and description of each POCT and target analyte is given in S1 Appendix in S1 File. All point-of-care malaria devices used an immunochromatographic assay to either detect histidine-rich protein (HRP-2) produced by Plasmodium falciparum, or parasite lactate dehydrogenase (pLDH, panmalarial antigen). The three HIV studies used the same nucleic acidbased HIV POCT [33, 58,63]. All studies using POCT-CRP employed a quantitative immunochemical assay for C-reactive protein. Five studies used a rapid antigen test for Strep A [42, 48,54,56,59].

POCTs and their intended role in clinical pathway and associated training
We examined the intended role of POCTs in the clinical pathway (S2 Appendix in S1 File) [65]. The majority of POCTs were defined as 'add-on' (n = 20) in which the new test is performed at the end of a clinical pathway to decide on appropriate treatment; nine studies used POCTs as 'replacement', in which the new test replaces an existing test, either as a faster equivalent test or to replace a non-point-of-care laboratory test, and six studies used POCTs as 'triage', in which the new test is used at the start of the clinical pathway excluding patients from further testing.
We tabulated any associated educational component in addition to POCT training offered (S2 Appendix in S1 File). Nine studies did not report any educational or training component, five studies focussed only on POC training for clinicians, and the remaining studies included a package of POC training, clinical training and guideline update, and/or interpretation of POCT.

Risk of bias assessment
Randomised trials were of moderate risk of bias (Fig 2); non-randomised studies had a higher risk (Fig 3). As we anticipated, none of the studies were able to blind participants and personnel to testing or test results. There were two cluster RCTs [32,46]. that were able to conduct blind outcome assessment (one reported in two separate papers) [31,32]. The non-randomised and before-after studies suffered from a high risk of selection, performance and detection bias and an unclear risk of reporting bias, as there was no protocol available.
We have categorised relevant outcome data according to four groups: As there were so few studies that evaluated the impact of POCT on patient outcomes, we have grouped these results together. The remaining outcomes are described per condition.
Patient outcomes. Across all conditions, six studies specifically reported mortality and morbidity measures such as illness course [31,35,41,44,49,52], of which one study for nonspecific acute fever illness in Belgium reported no deaths during the study using POCT-CRP [31], and one study for acute RTIs in Tanzania, found there were two fewer deaths in the POCT-CRP arm than usual care (0/865 vs 2/854) [52]. This latter study, using a two-step intervention (diagnostic algorithm followed by POCT-CRP), also found that the difference for patient recovery within one week was clinically negligible between intervention and control arms respectively, 97.1% (840/865) vs 95.2% (813/854).
The other four studies (two RCTs, two non-randomised studies) all relating to malaria studies in LMICs, found no difference in deaths between POCT and usual care (S3 Appendix in S1 File) [35,41,44,49]. There was no difference in patient recovery between malarial POCTs and usual care 3 to 7 days after treatment in four malaria studies (Fig 4) [39, 49,61,64]. The definition of patient recovery varied between studies, from self-reported full recovery [49], afebrile and negative blood smear [61], resolution of fever [39], or the absence of symptoms [64].

PLOS ONE
In-vitro diagnostic point-of-care tests in paediatric ambulatory care usual care) in which the two comparison arms were analysed separately in the meta-analysis [44]. Pooled estimates of RCTs showed that the use of malarial-POCTs better targeted antimalarial treatment by reducing over-treatment by a third in comparison to usual care (RR, 0.67; 95% CI [0.58 to 0.77], I 2 = 99%) (Fig 5).
There were two RCTs which had a marked effect on the pooled estimate [40,41]. In these two studies, almost all children in the usual care arm received antimalarials (97-99%). In addition, community health workers in the intervention arm of one of these two RCTs received additional refresher training six months after initial training [41]. Sensitivity analysis excluding these two studies showed that the effect remains significant (RR, 0.86; 95% CI [0.79 to 0.93], I 2 = 98%).
Based on two cluster RCTs [38,40], the proportion of children with malaria receiving prompt and targeted antimalarial treatment within 24 hours was significantly greater with rapid diagnostic tests compared to usual care, whether that care was delivered by community health workers or registered drug shops (RR 2.72, 95% CI [1.15 to 6.43], n = 11,304, S4 Appendix in S1 File). We also summarised safety aspects of malarial-POCT interpretation and antimalarial treatment in comparison to usual care (S5 Appendix in S1 File).
Antibiotic prescribing in suspected malaria. There was no significant difference in antibiotic prescribing in suspected malaria cases between POCTs and usual care, based on three cluster RCTs (RR 1.04, 95% CI [0.88 to 1.22], n = 8,403, Fig 6) [35, 36, 44]. In contrast, antibiotic prescribing was more likely in usual care in two non-randomised studies [55,60].

(B) HIV
Three studies evaluated the use of HIV-POCT in children: one cluster RCT in Mozambique [33] and two observational studies [58,63] set in Malawi and multiple African countries respectively. Decision-making. Initiating antiretroviral (ARV) therapy within 60 days in newly-diagnosed HIV children was almost 3-fold higher in those children that had an HIV-POCT compared to usual care in three studies (RR, 3.11; 95% CI [1.55 to 6.25], p<0.001; n = 912) [33, 58,63].
HIV-positive children who initiated ARV therapy based on HIV-POCT were also more likely to be retained in care at 90 days follow-up compared to usual care (adjusted RR, 1.40; 95% CI [1.1-1.9], p<0.027; n = 213) [33].
All studies used POCT-CRP (S1 Appendix in S1 File). Guidance on the interpretation of CRP results was given/available in one study [51]; intentionally not provided to clinicians in one cluster RCT (citing that safe cut-off levels in primary care are unknown) [31, 32]; subdivided into groups greater or less than CRP 60mg/L with no threshold justification [62]; subdivided into two pre-defined CRP-POCT thresholds (CRP 20mg/L and CRP 40mg/L) before medical examination [5], and not reported in two studies [50,57]. In the one UK study, clinicians were informed that a CRP level <20 mg/L suggested a serious infection was less likely compared to a value of >80 mg/L where serious infection was more likely [51]. Additional information about CRP values can be found in S6 Appendix in S1 File.  [57,62] did not show a statistically significant effect on hospital attendance or admission rates (Fig 7). Immediate antibiotic prescribing. Six studies reported antibiotic prescribing [5,31,50,51,57,62]. using POCT-CRP. Neither RCTs (RR, 0.93; 95% CI [0.84 to 1.03], I 2 = 0%) nor non-randomised studies (OR, 0.95; 95% CI [0.83 to 1.10], I 2 = 0%) showed an effect of the use of POCT-CRP on antibiotic prescribing (Fig 8).
Further analysis of the Althaus study [5],where the researchers used two pre-defined POCT-CRP thresholds (CRP 20mg/L and CRP 40mg/L) before medical examination, showed that the effect on immediate prescribing was not significant for the POCT-CRP group (133/ 400) which used thresholds of 20mg/L (RR, 0.94; 95% CI [0.78 to 1.14]), but had an effect on immediate prescribing for the POCT-CRP group (114/399) which used thresholds of 40mg/L (RR, 0.73; 95% CI [0.54 to 0.99], p = 0.04) compared with the control group (142/402). The risk of additional antibiotic prescriptions between day 0 and day 14 in the POCT-CRP groups versus control groups was not statistically significant for either intervention group compared to the control group (CRP20 group RR, 0.94; 95% CI [0.46 to 1.92], n = 14; CRP40 group RR, 1.36; 95% CI [0.69 to 2.70], n = 20; control group n = 15).

PLOS ONE
In-vitro diagnostic point-of-care tests in paediatric ambulatory care clinics in Tanzania [52]. One study used a viral panel POCT in a Canadian emergency department [47].
Decision-making. The viral panel POCT showed no effect on re-consulting within a 7-10 day time period, and found no effect (RR, 0.88; 95% CI [0.61 to 1.27]) [47]. The duration of patient visits was not found to be different when using the viral panel POCT in the emergency department (POCT 105.7min vs usual care 156.1min; mean difference -50.4min, 95% CI [-104.6 to 3.7] [47].
In the Vietnam study [46], where CRP data for children only were available (n = 81), a third of children younger than 6 years old (n = 28) received immediate antibiotic prescription when the CRP value at enrolment was 10 mg/L or less. However, this is substantially less than in the control group of children of all ages receiving an immediate antibiotic prescription (333/518, 64�3%). In the Danish study [45], where there was no significant effect [45], there was a small sample size (n = 139), and the baseline antibiotic prescribing was almost half of that of the Vietnamese study (34%). The authors infer that the clinicians may have ignored low CRP values for prescribing antibiotics. For example, at CRP values of less than 11 mg/l, antibiotics were prescribed to 25% of patients, and at values of between 11 mg/l and 25 mg/l they were prescribed to 51% of patients. The hybrid study in Tanzania [52] children with a POCT-CRP <80 mg/L were prescribed salbutamol as a home treatment in 17% (136/780) of patients in the ePOCT arm (based on a respiratory rate decrease after a salbutamol trial) and for 2% (17/769) of patients in the usual care arm. In the Dutch study [53], GPs were not provided with strict decision rules based on POCT-CRP levels, but were given guidance (Table 3). However, a relatively sample size (n = 309), protocol violations in the control group, and risk that clinicians were unblinded to the CRP level before noting a final diagnosis, which may have influenced their diagnostic labelling, limit the conclusions of this study. There is substantial heterogeneity between the four studies described. When the Danish and Dutch studies are excluded, the findings suggest that well conducted RCTs in LMICs which include guidance on interpretation of POCT-CRP, specific training or employ a diagnostic algorithm prior to POCT-CRP testing, may reduce antibiotic prescribing by around a third (risk difference, -0.29 [-0.47, -0.11], n = 2,747) [46,52].
There was no effect on the frequency of subsequent antibiotic prescriptions at re-consultation (day 3-5) when POCT-CRP was compared with usual care in the Vietnam study (RR, 1.16; 95% CI [0.83 to 1.61]) [46] or in the Dutch study (specified as same illness period), (RR, 0.92; 95% CI [0.33 to 2.53]) [53]. The Dutch study also assessed the effect on future consultations within the next three months, and found that 16% (13/81) of children in the POCT-CRP group consulted their GP for a new respiratory tract illness, compared to 29% (29/99) in the control group (OR 0.61; 95% CI = 0.32-1.17) [53].
In the context of the hybrid Tanzanian study, POCT-CRP underpinned by a diagnostic algorithm, led to less subsequent antibiotic prescriptions at day 7 than usual care employing another decision algorithm (RR, 0.16; 95% CI [0.12 to 0.20]) [52]. A viral panel POCT in Canada did not influence the immediate prescription of antibiotics when compared to usual care (RR, 0.86; 95% CI [0.48 to 1.53], p<0.61), but did find that fewer antibiotic prescriptions were prescribed at re-consultation within 1 week (RR, 0.36; 95% CI [0.14 to 0.95], p<0.04) [47]. Table 3. POCT-CRP guidance and interpretation for antibiotic prescribing (acute RTIs).

Do et al. (2016) [46]
Clinicians trained to use specific CRP cut-offs: no antibiotics when the CRP level was �20 mg/L for patients aged �6 years old, and � 10 mg/L for patients aged 1-5 years; referral or antibiotics when the CRP level was �50 mg/L. Between these thresholds no specific recommendation was given and clinicians were advised to use their clinical discretion.

Diederichsen et al. (2000) [45]
Clinicians informed of the normal value of CRP (<10 mg/l) and that CRP values <50 mg/l were seldom the result of bacterial infection. No strict guidelines for the use of antibiotics in relation to the CRP value were given.

Schot et al. (2018) [53]
GPs were given the following guidance: POCT-CRP levels should be interpreted in combination with symptoms and signs; POCT-CRP levels <10mg/L make pneumonia less likely, but should not be used to exclude pneumonia when the GP finds the child ill, or when duration of symptoms is <6 hours; POCT-CRP levels >100mg/L make pneumonia much more likely, however, such levels can also be caused by viral infections; between 10mg/L and 100mg/L, the likelihood of pneumonia increases with increasing CRP levels. https://doi.org/10.1371/journal.pone.0235605.t003 Test use. One study (n = 199) evaluated the effect of POCTs to detect multiple viral pathogens in acute RTIs and showed no statistically significant effect on the frequency of other test investigations (Table 4) [47] (E) Sore throat Five studies focussed on POCT in paediatric sore throat: one RCT [48]; one quasi-randomised trial [42]; one pre-/post-implementation observational study [59]; and two retrospective chart review studies [54,56]. Three studies were set in primary care and two studies in a paediatric emergency department [54,59]. All studies used a rapid Strep A POCT.
Decision-making. One RCT evaluated the effects of a Strep A test POCTs on re-consultation events [48] and found a statistically significant effect for decreasing subsequent visits when compared to usual care (RR, 4.70; 95% CI [2.94 to 7.51]; n = 1307). However, the time interval between visits was not reported.

Summary of main findings
The range of conditions or illnesses for which in-vitro diagnostic POCTs have been evaluated in paediatric ambulatory care is very limited. Of the 35 studies we identified, 14 studies focused on malarial-POCTs. Only three studies focused on POCTs in other acute paediatric illness in LMICs [5,46,52]. Most outcomes centred around decision-making such as hospital referral or prescription of medications; mortality data and other safety data were generally not reported. Pooled estimates from eight RCTs showed that the use of malarial-POCTs better targeted antimalarial treatment by reducing over-treatment by a third. This is not surprising as almost all children (90%) in the usual care arm were prescribed antimalarials. However, there was no significant difference in antibiotic prescribing between children who had a malarial-POCT and those that did not for malaria cases with a suspected bacterial co-infection [35,36,44]. HIV-POCTs helped initiate ARV therapy early in HIV-positive children and kept them in care. This suggests that POCT can also indirectly improve access to healthcare.
POCT-CRP in undifferentiated acute fever illness did not reduce hospital attendance or admission or immediate antibiotic prescribing. Likewise, pooled estimates for Strep A POCT in sore throat did not reduce immediate antibiotic prescribing. In acute RTIs, there is some evidence that POCT-CRP may reduce immediate antibiotic prescribing in LMICs, but only in well conducted RCTs which include guidance on interpretation of POCT-CRP, specific training or employ a diagnostic algorithm prior to POC testing [46,52].

Interpretation of results
There are a number of factors to consider when interpreting our findings. Diagnostics are complex interventions where clinical context, patient flow, and timing affect their impact. In addition, study variability (different setting, participants, intervention design) also needs to be considered when interpreting study findings and assessing the value of a POCT.
For example, POCT-CRP in acute fever illness did not affect hospital attendance or admission. However, most studies did not offer clear guidance on the interpretation of POCT-CRP in children leaving room for variation in practice and subsequent adherence to established practice.
In acute RTI, POCT-CRP studies were conducted in very different clinical settings (Vietnam; Denmark, Tanzania, The Netherlands), using different methodology. For example, the Danish study [45], found no effect on antibiotic prescribing likely because the baseline prescription rate was so low (half that of the Vietnamese study [46]). Most outcomes did not meet the accepted threshold of statistical significance. Many studies were underpowered to detect clinically relevant effects, or focussed on a selected population at low risk of serious infection. Other studies employed POCT-CRP in different roles in the clinical pathway in acute fever illness (e.g. triage or add-on).
These are problems that have been widely recognised as being major hurdles for diagnostic randomised controlled trials [67]. There are also likely to be important social determinants of prescribing that may override POCT-driven prescribing e.g. parental concern, the potential of rapid deterioration, and especially in LMICs, access to care [31, 68,69]. Our data were not able to evaluate if and how parents or carers of children might influence prescribing decisions despite a 'normal' POC result.

Comparison with existing literature
Existing literature on POCTs has focused on adult populations or mixed populations of adults and children. Our findings concur with a 2011 Cochrane meta-analysis evaluating POCTs versus clinical diagnosis of malaria in febrile mixed populations in African malaria endemic regions, where malarial-POCTs reduced antimalarial prescribing by over 50% based on four RCTs [70]. This reduction in antimalarial prescribing was more modest in our systematic review in children only (33%). One reason for this difference, might be that clinicians are more risk averse in children in LMICs where the prevalence of infectious disease, malnutrition and risk of death are greater [68]. Other reasons include perceptions that the risk of taking antimalarials is negligible for individual patients or that in high prevalence areas of malaria transmission, there is a significant false-positive malarial-POCT rate (i.e. slide negative) influencing a clinician's trust in the POCT result (due to persistent antigenaemia in individuals recently infected by malaria in hyperendemic areas [71]). Although other systematic reviews have found a reduction in antibiotic prescribing when POCT-CRP was used in adult-only and mixed populations [7,72], we found that when POCT-CRP are evaluated in children only, there is limited evidence of benefit for their use in undifferentiated acute illness.

Strengths and limitations
Our search strategy was comprehensive using validated search filters, and we included both RCTs and non-randomised studies conducted in ambulatory healthcare settings. We focussed on paediatric populations, an under-researched group and specifically focused on the impact of in-vitro POCTs in clinical care as opposed to diagnostic accuracy studies.
There are also important limitations. We accept that many studies showed high risk-ofbias. We had to exclude some mixed population studies where data for adults and children were inseparable and not suitable for meta-analysis. We also recognise that the distinction between acute fever illness, acute RTIs and sore throat is somewhat arbitrary, and does not necessarily reflect routine practice where infection syndromes are not always clear-cut. The data available did not allow us to sufficiently compare studies in terms of POCT-CRP thresholds. Although consultations in ambulatory care have a dual purpose-to rule out serious infections and make antibiotic prescribing decisions-the interpretation of POCT results also needs to be seen in this heterogenous context as explained above.

Implications for clinical practice and future research
Children represent a significant proportion of consultations in ambulatory care. Yet, unlike the growing evidence in adult populations, there is a clear evidence gap for the use of POCTs to improve clinical outcomes in children worldwide. There is some evidence for POCTs in a few well-defined areas in specific settings e.g. HIV in LMICs. Yet for many other areas, mainly in HICs, the evidence for POCTs is scarce and often at high risk of bias. Therefore, because the impact of POCTs is so context-specific, we would recommend that any implementation of POCT be closely monitored to investigate their clinical effectiveness including monitoring of any unintended consequences of testing. Failure to heed these caveats, will mean that many new tests are not routinely taken up into routine care, or are implemented despite skipping essential stages such as clinical effectiveness, and waste resources [73].
Secondly, fit-for-purpose POCTs need to be accompanied by clear guidelines on their interpretation e.g. POCT-CRP cut-off values for children. Strategies are needed to help clinicians deal with inconclusive or dubious results e.g. to aid decisions in malarial POCT-negative children who are prescribed antimalarials, or where children with acute RTIs receive antibiotics when the CRP value � 10 mg/L. Likewise, the role of POCTs in paediatric ambulatory care will differ between LMICs and HICs in helping to guide treatment decisions in acute illness and chronic disease monitoring. For example, in LMICs, where the prevalence of serious infections is high, the role of POCTs will be to exclude serious infection. In HICs, this role may be to make antibiotic prescribing decisions based on prognosis of common (self-limiting) infections. For LMICs in particular, POCTs ought to be incorporated into existing clinical pathways e.g. the World Health Organization (WHO) Integrated Management of Childhood Illnesses (IMCI) guidelines, to ensure that there is a seamless transition [52].
Finally, this review is important to provide direction and design of future studies. Studies should expand their remit beyond malaria and HIV in LMICs and incorporate POCTs for common infection syndromes. The impact of POCTs requires careful evaluation in welldesigned RCTs or other controlled study designs, taking into account that the introduction of a new diagnostic test is a complex intervention. This will require mapping the patient pathway to understand all steps from patient presentation, selection for testing, interpretation of the test result, to integration of the result in clinical decision-making. For this to be possible, qualitative and quantitative contextual information needs to be embedded in to future clinical trials. Producing studies that are too small do not guide clinicians in their interpretation and clinical decision-making, or are at high risk of bias because of methodological shortcomings, or may even lead to wrongly rejecting a valuable tool for clinical practice.

Conclusion
There are clear evidence gaps for the use of POCTs in paediatric ambulatory care. Research has focussed on malaria-and HIV-POCTs in LMICs where they have shown benefits. There is emerging evidence that POCT-CRP may better target antibiotic prescribing for children with acute RTIs in LMICs but not in HICs. More paediatric-focussed research is urgently needed to understand where POCTs are likely to improve clinical outcomes in paediatric ambulatory settings worldwide.