ERS/ESCMID state-of-the-science statement on the role of rapid diagnostic tests for the diagnosis and management of community-acquired lower respiratory tract infections

Abstract

Antimicrobial resistance is a significant threat to healthcare worldwide. The appropriate use of antimicrobials could decrease the excess morbidity of respiratory infections and contribute to reducing the emergence of multidrug resistance. For this purpose, the European Union has funded the VALUE-Dx project to assess the value of rapid diagnostic tests in combatting antimicrobial resistance by optimising antibiotic use in community-acquired lower respiratory tract infections (CA-LRTI). In this context, the current statement aims to review the literature and current practice regarding the role of rapid diagnostic tests in managing CA-LRTI for antibiotic stewardship purpose and combatting antimicrobial resistance. In particular, the present paper evaluates the accuracy of rapid diagnostic tests for the aetiological diagnosis of CA-LRTI, the usefulness of biomarkers and clinical algorithms in the management of CA-LRTI, the differences in aetiology according to the diagnostic protocol and tests applied for CA-LRTI after the introduction of rapid diagnostic tests for a complete microbiological investigation and, finally, the role of rapid diagnostic tests in antibiotic stewardship protocols in improving clinical outcomes.

Shareable abstract

The current literature on the use of RDTs for CA-LRTI management is scarce and heterogeneous. Nevertheless, the appropriate use of biomarkers seems to be useful, while the clinical impact of molecular tests and algorithms has to be better established. https://bit.ly/4m9NVp4

Introduction

Community-acquired lower respiratory tract infections (CA-LRTIs) are still a leading cause of death worldwide [1] and one of the reasons could be the increasing rate of antimicrobial resistance (AMR) among the most common bacterial pathogens. Currently, AMR is considered one of the top 10 global public health threats [2], and a well-known cause of AMR is the inappropriate use of antimicrobials.

VALUE-Dx is a European project raised in the context of the European framework of Horizon 2020, the European Union's research and innovation funding programme (IMI call IMI2-2017-13-03) to combat the emergence of AMR [3]. In particular, the VALUE-Dx project aims to assess the value of rapid diagnostic tests (RDTs) to combat AMR by optimising antibiotic use. Implementing rapid point-of-care diagnostic tests is expected to facilitate clinical management of CA-LRTIs, reduce the use of antibiotics by guiding antimicrobial stewardship, and promote better use of available resources (broad-spectrum antibiotics, etc.). Nevertheless, RDTs are still sparsely used owing to differences in availability, cost/efficiency issues, lack of integration of diagnostic tools in antimicrobial stewardship protocols and, finally, very heterogeneous and poor-quality scientific evidence. In fact, despite the large number of new diagnostic tests, the evidence supporting their implementation in clinical practice is still lacking. Moreover, studies are heterogeneous in terms of clinical settings, microbial targets, study populations, etc.

The European Respiratory Society (ERS) and the European Society of Clinical Microbiology and Infectious Diseases (ESCMID) partnered to create a specific Task Force within the VALUE-Dx Consortium to provide scientific evidence regarding the usefulness of RDTs in improving the diagnosis of bacterial LRTIs and therefore reducing inappropriate antibiotic use.

This statement covers the use of RDTs in the context of diagnosis and management of CA-LRTIs. The evidence assessment included the accuracy of RDTs, the usefulness of biomarkers and clinical algorithms, variation of LRTI aetiology according to the diagnostic procedures applied, and the role of RDTs in antibiotic stewardship protocols in improving clinical outcomes. After reviewing the literature on these topics, we also highlight potential gaps and future research topics.

Methods

This ERS/ESCMID statement combines an evidence-based approach, combining evidence from systematic literature searches with considerations stemming from the clinical, microbiological and research expertise of Task Force members. The statement summarises relevant literature and current practice by topic. It does not provide de novo recommendations for clinical practice but indicates where the Task Force members agree with published guidance.

The Task Force included different VALUE-Dx work package representatives (WP 1, 2, 4), experts in respiratory infections and AMR, representatives from both ERS and ESCMID, an ERS methodologist (TT), and representatives from the diagnostic companies involved in the VALUE-Dx Consortium (Abbott, BD, Biomerieux) as observers who were not allowed to make decisions about the scientific content of this manuscript or participate in its edition. Representatives from the other work packages (WP 3, 5, 6, 7) were not included in the Task Force because their contribution was considered unnecessary with regards to the scope of the manuscript. All members signed forms disclosing conflicts of interest.

A list of the five most relevant topics was selected by the Task Force members based on the objectives of the statement and the VALUE-Dx project. Relevant questions were formulated after consecutive discussions among panellists (table 1).

TABLE 1.

Research questions

Question number
1	Are the sensitivity and specificity of RDTs (molecular, etc.) better than classic microbiological testing in diagnosing CA-LRTI in a community-care setting? Working group members: PDN, EG, AG, TT, ET, CC
2	Is the use of biomarkers (procalcitonin, C-reactive protein) useful to improve clinical management of CA-LRTIs? Working group members: PDN, AH, HK, SKS, SMK, TT, IML, ET
3	Does the application of clinical algorithms for the diagnosis of CA-LRTIs improve the management of LRTIs in a community-care setting? Working group members: PDN, AG, TT, ET
4	Is the distribution of microbial aetiologies of CA-LRTI different when using a complete microbiological investigation including RDTs in comparison with standard cultures? Working group members: AL, JPRR, CPA, CC, TT, SMK
5	Does the addition of RDTs to usual microbiological testing improve outcomes of antibiotic stewardship? Working group members: EP, LT, TT, HG, MA

CA-LRTI: community-acquired lower respiratory tract infection; RDT: rapid diagnostic test.

The panel members were divided in five working groups to cover all topics according to their expertise. All work groups worked predominantly by e-mail or teleconference, but physical meetings were organised at the annual ESCMID and ERS congresses between 2021 and 2023. Reports from the different groups were integrated into a final manuscript and approved by the Task Force.

The search strategy is detailed in supplementary tables S1, S3, S5, S8 and S12. A systematic literature search of three electronic databases (MEDLINE, Embase (Elsevier) and Cochrane Central Register of Controlled Trials (CENTRAL)) was conducted by professional librarians with the input of Task Force members. Systematic reviews and meta-analyses were searched and, if the search strategy corresponded to one of the topic questions, data were extracted and updated. For topic 1, the Task Force used part of the results of a systematic review and meta-analysis that addressed the same question, and which had been completed by WP1 of VALUE-Dx and already published [4]. Therefore, there are some differences in methodology between topic 1 and the other topics. For instance, QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies-2) was used for risk of bias for topic 1. For the other questions, we followed the usual ERS rules for statements and did not conduct risk of bias assessment and meta-analysis but only reviewed the evidence narratively.

All the studies relevant to the respective working groups were identified and screened for potential inclusion in two stages: title/abstract then full-text screening by two independent reviewers from each working group. In case of disagreement, the selected studies were re-discussed again with all members of the working group. The specific search strategies and time periods covered by the literature searches are described in detail in the specific topics’ sections and in the supplementary material. Data were independently extracted and analysed by each working group under the supervision of our methodologist (TT). The inclusion criteria and study selection strategy are presented in the supplementary material, as well as a Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) diagram for each research question. Special subgroups of the population, such as immunosuppressed patients, were not considered part of the scope of this review because they are not usually considered as CA-LRTI patients. Extracted data were then discussed and summarised between each working group. If the search identified an existing systematic review covering any of the topics included in the questions, this was used as base evidence that was complemented by publications posterior to the systematic review. All the evidence from the literature was narratively reviewed without conducting new analyses. Each working group drafted their part of the manuscript, which was first reviewed and agreed upon by all working group members and then all Task Force members at a later stage. All Task Force members approved the final draft.

Results

Topic 1. Are the sensitivity and specificity of RDTs (molecular, etc.) better than classic microbiological testing in diagnosing CA-LRTI in a community-care setting?

Overview

Our literature search identified 922 records included in the VALUE-Dx meta-analysis, from inception to May 2021. Among them, 518 assessed RDTs and 183 of these were included in the quantitative synthesis (meta-analysis) (supplementary tables S1 and S2, supplementary figures S1–S5) [4–7]. The VALUE-Dx meta-analysis showed that the accuracy of stand-alone signs and symptoms or biomarkers was poor and the diagnostic performance of molecular tests was higher than that of RDTs for all the pathogens studied [4]. Lung ultrasound showed high sensitivity and specificity for the diagnosis of bacterial pneumonia. Here, we review the evidence for the accuracy of RDTs for diagnosis of CA-LRTI to drive evidence-based statements. An economic assessment is not included (supplementary table S2, supplementary figure S1).

Streptococcus pneumoniae

For Streptococcus pneumoniae, 12 studies [8–19] (2826 individuals) assessing antigen-detection RDTs (Ag-RDTs) were included. All studies were performed at the emergency room and evaluated by an immunochromatographic assay (Alere BinaxNOW, Abbott) performed using a urine sample. These studies showed an overall sensitivity of 72% (95% CI 62–80%) and specificity of 83% (95% CI 65–93%).

Mycoplasma pneumoniae

For Mycoplasma pneumoniae, seven studies [20–26] (1970 individuals) were included. Overall, the reported sensitivity and specificity were 85% (95% CI 63–95%) and 90% (95% CI 75–97%). Of these, four studies (712 individuals) assessed antibody RDAs, i.e. immunoenzymatic assays (ImmunoCard Mycoplasma, Meridian Bioscience; Remel Mycoplasma Pneumoniae IgG/IgM antibody tests, Thermo Scientific), while three studies (1258 individuals) assessed Ag-RDTs, of which only one stated the brand used (Vircell Microbiologists). The meta-analysis of the four studies evaluating immunoenzymatic assays showed a sensitivity of 83% (95% CI 38–97%) and specificity of 88% (95% CI 79–94%).

Influenza virus

For influenza virus, 142 studies [27–167, 373] (69 699 individuals) assessing the accuracy of Ag-RDTs were included. Direct immunofluorescence technology (Sofia Influenza A+B FIA (fluorescent immunoassay), Quidel; D³ Influenza A/Influenza B DFA (direct fluorescent antibody), Quidel; Patho DX, Thermo Scientific; IMAGEN Influenza virus A and B Kit, Thermo Scientific) and automatised immunofluorescence detection technology (mariPOC, ArcDia International) showed higher sensitivity (78%, 95% CI 67–86% and 78%, 95% CI 61–89%, respectively) and specificity (95%, 95% CI 90–98% and 99%, 95% CI 97–100%, respectively) than immunochromatographic assays (BinaxNOW Influenza A and B, Abbott; Directigen EZ Flu A+B, Directigen Flu A and Directigen Flu A+B, BD; Espline Influenza A&B-N, Fujirebio; QuickVue Influenza A+B, Quidel; SD Bioline Influenza Ag, Abbott; Veritor Flu A+B, BD) (sensitivity 69%, 95% CI 64–74%; specificity 97%, 95% CI 96–98%). Among immunochromatographic assays, Espline Influenza A&B-N and Veritor Flu A+B provided the best sensitivities (85%, 95% CI 71–93% and 84%, 95% CI 70–92%, respectively) with similar specificity (supplementary figures S3 and S4).

Respiratory syncytial virus

For respiratory syncytial virus (RSV), 35 studies [37, 38, 40, 57, 70, 83, 86, 89, 98, 105, 118, 125, 136, 152, 153, 155, 156, 168–184, 374] (16 110 individuals) were included assessing five different types of Ag-RDTs. The overall sensitivity was 83% (95% CI 78–86%) and specificity was 97% (95% CI 94–98%) (supplementary figure S5).

Human metapneumovirus

For human metapneumovirus (hMPV), five studies [37, 86, 169, 185, 186] (1578 individuals) assessing Ag-RDTs were included: one considering two fluorescence immunochromatographic assays (Diagnostic Hybrids Inc., Quidel; IMAGEN, Thermo Scientific), one assessing an immunochromatographic assay (SAS hMPV test, SAS Scientific) and two evaluating an automatised immunofluorescence detection technology (mariPOC, ArcDia International; see dedicated paragraph “Multiple pathogen antigen-detection techniques”). The overall sensitivity was 59% (95% CI 36–78%) and specificity was 99% (95% CI 95–100%).

SARS-CoV-2

For SARS-CoV-2 we used the second update of the Cochrane systematic review and meta-analysis from the COVID-19 Open Access Project Living Evidence database from the University of Bern (https://zika.ispm.unibe.ch/assets/data/pub/search_beta/). This was searched in March 2021 and provided 155 study cohorts on rapid, point-of-care antigen tests, accounting for a total of 100 462 unique samples. Overall, the sensitivity was 69% (95% CI 66–72%) and the specificity was 99% (95% CI 99–99%) [187]. A more recent systematic review, updated in May 2022, included 135 studies for a total of 166 943 samples. The pooled sensitivity was 76% (95% CI 73–79%) and the pooled specificity was 100% (95% CI 100–100%) [188].

Multiple pathogen antigen-detection techniques

The mariPOC test is an automated immunofluorescence detection technology that allows the identification of multiple pathogens simultaneously. Five studies [37, 38, 86, 136, 153] (1231 individuals) assessing the mariPOC assay for influenza and RSV were included in the meta-analysis. This assay showed a better sensitivity for influenza than immunochromatographic assays (78% versus 69%) and an excellent specificity (99%). mariPOC showed the second best sensitivity (81% versus 84% for the Sofia RSV FIA) and high specificity (99%) for RSV. Our analysis did not allow for a comparison of the accuracy of mariPOC for other pathogens.

Summary statements

• Current evidence on the effectiveness of RDTs for the diagnosis of CA-LRTI has major limitations, including scant data from long-term care facilities (LTCFs) and outpatient clinics and the lack of control for confounders, e.g. specimen type and quality, duration of symptoms at the time of testing and seasonality, which could impact the diagnostic accuracy of pathogen-based tests.

• Overall, the evidence shows that the Ag-RDTs for diagnosing viral CA-LRTI (influenza, RSV and hMPV) have high specificity but suboptimal sensitivity.

• The assessment of evidence for diagnostic accuracy of the Ag-RDTs for diagnosing bacterial CA-LRTI is limited to Ag detection for S. pneumoniae and Ab and Ag for M. pneumoniae, with limited sensitivity and specificity.

• A few Ag-RDTs can detect multiple pathogens, showing promising results in terms of accuracy, although, due to the low number of studies, the evidence still needs to be improved.

• The RDTs for SARS-CoV-2 infection show excellent specificity but still suboptimal sensitivity.

Recommendations for future research

• Studies are needed that assess feasibility, accuracy by type of patient and a cost–benefit balance of RDTs for the diagnosis of CA-LRTI in a community-care setting.

• Randomised clinical trials comparing the accuracy of different RDTs embedded in diagnostics algorithms, including laboratory and imaging tests, are strongly encouraged.

• Prospective studies should include assessment of follow-up to define the risk of adverse effects of RDTs in terms of delayed or inappropriate diagnosis.

• Randomised clinical trials are needed to assess antibiotic stewardship recommendations linked to the implementation of RDTs.

Topic 2. Is the use of biomarkers (procalcitonin, C-reactive protein) useful to improve clinical management of CA-LRTIs?

Overview

We examined literature published until May 2023 to assess whether the use of biomarkers is useful to improve the clinical management of patients presenting with symptoms of CA-LRTI. In particular, we assessed their role in distinguishing viral versus bacterial CA-LRTI, in guiding decisions on the site of care and antibiotic prescription or de-escalation, in assessing response to treatment, and in reducing length of hospitalisation and mortality rate (supplementary tables S3 and S4, supplementary figure S6).

Biomarkers in CA-LRTI diagnosis

Our searches identified nine relevant studies that investigated whether C-reactive protein (CRP) could be used in the diagnosis of LRTI, of which eight were observational [189–196] and one was case–control [197]. Five of the eight observational studies on CRP showed sensitivity/specificity of 50–97%, with cut-off values of 10–250 mg·L⁻¹ [189, 190, 192, 194, 196]. The case–control study was performed with 285 patients of all age groups, and CRP levels with a cut-off of ≥11 mg·L⁻¹ could distinguish community-acquired pneumonia (CAP) from healthy individuals with a sensitivity of 94% and specificity of 95%. The same study with a CRP cut-off of ≥33.13 mg·L⁻¹ could distinguish CAP from unconfirmed pneumonia with a sensitivity of 83% and specificity of 63% [197]. In addition, one study measured CRP in saliva in a paediatric population and showed a sensitivity of 97% and specificity of 90% with a cut-off of 3.2 μg·L⁻¹ [189].

Three studies investigated the use of procalcitonin (PCT) for the diagnosis of CAP, leading to a sensitivity/specificity range of 50–70% in two studies, which was considerably lower compared to CRP-led diagnosis in the same studies included as a comparator [190, 194]. In the third study, involving 319 adults, the reported sensitivity and specificity were high, 89% and 98%, respectively [196].

Biomarkers in distinguishing viral versus bacterial aetiologies, guiding decisions on the site of care and assessing response to treatment, including discontinuation of antibiotic use

We found 39 studies [191, 198–235] addressing the use of biomarkers in not only distinguishing viral versus bacterial aetiologies (n=37) [191, 198–228, 230–235] but also between different bacterial aetiologies (n=1) [203], typical versus atypical bacterial (n=4) [191, 201, 214, 228], mixed infections (n=2) [209, 234] and diagnosing specific pathogens (n=2) [229, 235].

C-reactive protein

CRP was used in 33 observational studies [191, 198–209, 211–213, 215–217, 219, 221, 224–235] to distinguish bacterial versus viral infections, 26 of which were in hospitalised patients [191, 198–203, 205–209, 211–213, 215–217, 219, 221, 224–235], five in both hospitalised and ambulatory patients [191, 204, 208, 228, 233] and two in ambulatory patients [202, 235]. Cut-off values in these studies ranged from 10 mg·L⁻¹ to 200 mg·L⁻¹. However, a high overlap of CRP concentrations was observed in bacterial and viral LRTIs with the same severity (e.g. [200]). This hinders setting an appropriate CRP cut-off.

Despite this limitation, 12 randomised controlled trials (RCTs) implemented point-of-care testing (POCT) for CRP to guide antibiotic stewardship in primary care (PC) settings to reduce antibiotic exposure in non-severe LRTIs [236–246]. Patients had acute cough/RTIs [236], COPD exacerbations [237] or other features with suspicion of LRTIs [238–240, 247]. Of these, six RCTs (50%) showed a significant reduction in antibiotic treatment (initiation or duration) in adult patients receiving CRP-guided treatment (compared to standard care) without compromising clinical recovery [236–240, 247]. Notably, four RCTs did not show a significant decrease in antibiotic use with CRP POCT in the emergency department (ED) compared to standard care [241–244] in both adults (n=3) [242–244] and children (n=1) [241] with suspected non-serious LRTI.

One trial showed that an internet-based training protocol in CRP use reduced the antibiotic prescription rate in PC patients with an upper or lower RTI [245] in both the short and long term (1 year) [246].

Procalcitonin

For the diagnosis of bacterial versus viral LRTI, PCT was used in 24 studies [198, 199, 201, 203, 204, 208–211, 214, 218, 220–228, 230, 231, 233, 234]. One of these was an RCT in 194 hospitalised children and adolescents [222] in which PCT alone (sensitivity 73%, specificity 88%) or in combination with clinical parameters such as lung ultrasound (sensitivity 91%, specificity 87%) and chest X-ray (sensitivity 98%, specificity 43%) was used to diagnose bacterial LRTI vis-à-vis nonbacterial LRTI including viral LRTI.

In 11 RCTs, the PCT level was successful in reducing total antibiotic exposure in adults with severe LRTI/CAP [248–251] or severe COPD exacerbations [252] without negatively affecting clinical resolution as a primary end-point [248–258]. One RCT investigated the use of PCT in a paediatric population and showed a shorter duration of treatment in the PCT-guided antibiotic prescription arm without affecting clinical recovery [253].

An additional four RCTs with PCT-guided antibiotic stewardship showed a significant reduction in antibiotic prescription rate as well as duration of therapy in low-risk adult outpatients with suspected non-severe LRTI/CAP in PC settings [254–257]. Three of the four RCTs showed no impact on the clinical recovery.

Nevertheless, despite the evidence for PCT reducing antibiotic duration in most studies, the average length of hospitalisation and antibiotic use in the placebo groups was well over a week, far longer than recommended in guidelines.

Reduction of antibiotic exposure was less apparent in RCTs in patients with suspected LRTI, showing a nonsignificant decreasing trend in the number of days of antibiotic treatment with PCT-guided treatment compared to standard care [258, 259].

Two studies assessed the role of an educational intervention in implementing PCT use in managing suspected LRTI [260, 261]. Both protocols significantly increased the use of PCT-guided antibiotic stewardship and reduced antibiotic exposure.

Biomarkers inducing a change in antibiotic prescription

Only one RCT investigated whether PCT POCT could be implemented to guide antibiotics, specifically choosing between levofloxacin and azithromycin in outpatients with suspected low-risk LRTI [262]. Patients were treated with oral azithromycin when PCT was <0.5 ng·mL⁻¹ or with levofloxacin when PCT was ≥0.5 ng·mL⁻¹. No difference in clinical cure rate, recurrence or mortality was observed between treatment arms, suggesting that a PCT-guided strategy could be safely implemented for selecting empirical narrow-spectrum antibiotics.

Biomarkers reducing relapse rate

No study that directly used biomarkers for reducing relapse rate as a primary end-point was identified. However, the use of CRP or PCT for guiding antibiotic treatment in the clinical trials described above did not lead to increased treatment failure and/or relapse rates, defined as hospital readmission or re-visiting PC with respiratory complaints, worsening or persistent symptoms, antibiotic use during 28-day follow-up, intensive care unit (ICU) admission or 30-day mortality.

Biomarkers in guiding risk stratification and in reducing length of hospitalisation and mortality rate

Different studies have used biomarkers for risk stratification (disease severity), enabling physicians to provide the proper treatment and site of care.

C-reactive protein

Regarding the decision for hospitalisation, one case–control and four observational studies showed that CRP in a PC setting for patients with clinical features suggestive of LRTI successfully identified patients requiring hospitalisation or ICU admission with a sensitivity and specificity of 81% [197, 208, 263–265]. Two of these four observational studies also predicted mortality based on CRP levels [264, 265]. Two out of these four studies reported a cut-off value for CRP to predict hospital admission, i.e. 106 mg·L⁻¹ (sensitivity 81%, specificity 81% [197]) and 125 mg·L⁻¹ (sensitivity 63%, specificity 70% [208]). A prospective cohort study with 245 children aged <3 years hospitalised with a viral LRTI reported an area under the receiver operating characteristic curve (AUROC) of 0.63 for CRP in predicting disease severity [266] and children with a mild course of disease had reduced length of hospitalisation.

Procalcitonin

The potential of PCT to predict hospital admission was investigated as a primary end-point in two observational studies [208, 263]. The first showed that PCT >0.15 ng·mL⁻¹ was better than CURB65 (Confusion, blood Urea nitrogen >7 mmol·L⁻¹, Respiratory rate ≥30 breaths·min⁻¹, Blood pressure: systolic <90 mmHg or diastolic ≤60 mmHg, age ≥65 years) at stratifying patients at risk of severe CAP and being hospitalised, with a sensitivity of 82% and specificity of 70% [208]. The second showed an AUROC of 0.619 for non-COVID viral pneumonia (an AUROC of 0.829 for COVID-19 pneumonia) and an AUROC of 0.672 for bacterial pneumonia [263].

A retrospective observational study in adults hospitalised with suspected LRTI showed that PCT testing (cut-off PCT >0.25 ng·mL⁻¹) was associated with reduced length of hospitalisation [267] while no reduction in length of hospitalisation was observed in two RCTs [250, 251] and a retrospective cohort study [268] with PCT-guided antibiotic therapy.

Lastly, the potential of PCT to predict mortality or ICU admission was addressed in two RCTs [269, 270] and 20 observational studies [203, 225, 263–265, 267, 271–284]. Despite high heterogeneity in patient populations and end-points, PCT alone had a relatively low predictive power for identifying disease severity or mortality (AUROC 0.55–0.83).

Summary statements

• Based on current evidence, CRP is particularly useful in diagnosing LRTI and CAP in PC, as well as for guiding decisions on the site of care for hospitalisation and risk stratification (disease severity) and in predicting mortality.

• High overlap was shown in CRP concentrations in bacterial and viral LRTI of the same severity, making it difficult to set an appropriate CRP cut-off for accurately distinguishing these aetiologies. Despite this limitation, CRP POCT successfully guided antibiotic stewardship in PC settings and reduced antibiotic exposure in non-severe LRTI.

• Current data on PCT show it could distinguish viral from bacterial CA-LRTI, guide antibiotic stewardship, and reduce the duration of treatment in both PC and hospital settings.

• In several clinical trials, neither CRP- nor PCT-guided antibiotic treatment led to increased treatment failure and/or relapse rates compared to standard care.

Recommendations for future research

• Research on the use of CRP POCT with consensus cut-offs should be further implemented in the context of antibiotic stewardship in adults with suspected LRTI, and possibly complemented by other clinical factors such as the standardised assessment of clinical stability.

• More research is needed to support host biomarker-based antibiotic stewardship in underrepresented patient populations such as children and in easily assessable respiratory samples such as saliva.

• RCTs comparing different bundles of molecular and antigen-based POCTs with imaging and diagnostic tests are needed to define the best cost-effective approach in different settings and by comorbidity/age/season to reduce inappropriate antibiotic usage.

• Cohort-based studies could be very useful to assess and compare homogenised data to sustain the outcomes associated with antimicrobial stewardship programmes.

Topic 3. Does the application of clinical algorithms for the diagnosis of CA-LRTIs improve the management of LRTIs in a community-care setting?

Overview

Our searches identified 34 studies carried out between 1986 and 2021, including different ages, settings (24 studies in the ED, six in a PC setting, one in a LTCF) and World Health Organization geographical regions (13 in the European Region, 10 in the Americas and eight in the Western Pacific [285]) (supplementary tables S5–S7, supplementary figures S7 and S8). These studies yielded 66 extractions, of which 24 validated previously published algorithms and 42 proposed novel algorithms. Overall, 47 unique clinical algorithms were retrieved (supplementary table S7).

Methodological aspects of the algorithms

Although the studies were quite diverse in choice of predictors, the vast majority of algorithms (n=42, 89%) were formulated as logistic regression-derived scores. All algorithms were trained to predict a binary outcome denoting diagnosis or aetiology.

Validation

Most algorithms were initially published without validation (n=26, 55%), 10 algorithms were validated externally (21%) and four algorithms (9%) used internal validation (cross-validation, bootstrap, train/test division). Clinical algorithms were either treated as a diagnostic test, with a predefined cut-off and reported the sensitivity and specificity (n=8, 24%), or as a prediction tool and reported the AUROC (n=11, 32%). In 14 studies (41%), all three measurements were provided.

Algorithms for patients presenting with acute RTIs

In total, 34 algorithms (72%) targeted patients presenting with an acute RTI or suspicion of CAP, and 18 (53%) were aimed at guiding X-ray diagnosis of CAP primarily in adults (n=7992), older adults (n=2334) and children (n=827). The studies were carried out in the ED (n=8), in PC (n=4) and in a LTCF (n=1). Overall, 36 features were used, with the majority (n=23, 72%) only by a unique algorithm. All algorithms included at least one respiratory symptom (fever in 16 studies (47%)), while eight used biochemical markers (44%). The algorithms showed diverse sensitivities (64–98%), specificities (8–70%) and AUROCs (49–93%) [195, 286–296].

The following seven algorithms for predicting CAP were evaluated by other studies:

1. Heckerling et al. [291]: algorithm used absence of asthma, decreased breathing sounds, fever, pulse and rales, and showed an AUROC of 76%. Validation studies conducted in multiple settings (ED and PC) and countries (China, USA and pan-European) showed AUROCs of 57–88% [195, 286, 289, 295].

2. Hopstaken et al. [293]: algorithm used dry cough, diarrhoea and fever symptoms, and an extension added CRP. The algorithms reached AUROCs of 70% and 90%, with 62% and 69% in the validation studies [195, 289].

3. Diehr et al. [296]: algorithm used fever, myalgia, night sweats, respiratory rate >25 breaths·min⁻¹, rhinitis, sore throat and sputum, and showed 74% sensitivity and 70% specificity. Validation studies reported 91% sensitivity and 34% specificity, with AUROCs of 67% and 72% [195, 286].

4. Singal et al. [287]: algorithm used cough, crackles and fever, and reported an AUROC of 73%. Validation studies reported AUROCs of 58% and 68% [195, 289].

5. Lynch et al. [292]: algorithm included fever, decreased breathing sounds, crackles and tachypnoea, showing a sensitivity of 98% and a specificity of 8%%. A validation study showed a sensitivity of 94% and specificity of 19% [288].

6. Melbye et al. [294]: the algorithm used duration of illness and coryza, sore throat, dyspnoea, chest pain and crackles in adults in a PC setting, with AUROCs of 49% and 65% in validation studies [195, 289].

7. González Ortiz et al. [297]: score used decreased breathing sounds, dyspnoea, neutrophil count and pleuritic pain, with an AUROC of 57% in the validation study [289].

Algorithms for CAP aetiology for patients presenting with X-ray-confirmed CAP

Nine algorithms (21.4%) were dedicated to patients with X-ray-confirmed CAP to diagnose the aetiology.

Moreno et al. [298] and Ruiz-González et al. [299] targeted a diverse population (103 mixed, 175 children) in the ED to predict bacterial aetiology. They used clinical features, biochemistry and X-ray, and showed a sensitivity between 89% and 100% and specificity ranging from 63% to 94% (AUROC 84–98%).

The Miyashita et al. [303] and Fiumefreddo et al. [301] algorithms were designed to diagnose Legionella, and were externally validated [300, 302]. The studies included a mixed population (1505 adults and adolescents, 450 older adults) and mixed settings (ED and PC). The two algorithms used cough and several biochemical measurements (CRP, lactate dehydrogenase and sodium) and achieved a sensitivity of 63–93%, specificity of 75–86% and an AUROC of 86–93%.

Fernández-Sabé et al. [304] evaluated the Community-Based Pneumonia Incidence Study (CBPIS) model [305]. The study included 298 adults diagnosed with CAP caused by either S. pneumoniae or L. pneumoniae in the ED and showed a sensitivity of 51% and specificity of 86% in detecting Legionella for a “highly probable” class.

Secondary outcomes

Included studies stated their work aims to be improving time to diagnosis, improving diagnosis in low-resource settings, opening other avenues of management (e.g. phone consultation) and saving unnecessary diagnostics (e.g. chest X-ray). Only five publications recorded secondary clinical outcomes, such as time to recovery, prescribed antibiotics or hospitalisation [148, 293, 295, 303, 304]. One concluded that their algorithm would have prevented unnecessary antibiotic prescription [293] and two observed that use of a clinical algorithm did not change the time to recovery [148, 295].

Clinical algorithms within respiratory societies’ recommendations

None of the respiratory societies recommended any clinical algorithm for the diagnosis of CA-LRTI [306–312]. A specific question that explored if the diagnostic algorithm plus clinical judgment was superior to clinical judgment alone to confirm the diagnosis of pneumonia in adult outpatients with acute cough was included in the CHEST guidance document [308]. After reviewing the evidence, the suggestion was to add chest radiography to improve the diagnostic accuracy of CA-LRTI [308].

Summary statements

• Studies assessing the diagnostic performance of clinical algorithms rarely include other clinical outcomes.

• The signs and symptoms in patients with CAP most frequently used to predict bacterial or viral aetiology through clinical algorithms were fever, cough, rhinitis, white blood cell count, gastrointestinal symptoms and patient's age. CRP, lactate dehydrogenase and sodium levels were used only in the bacterial aetiology algorithms, whereas white blood cell count was used for diagnosing both bacterial and viral aetiologies. Overall, there is high heterogeneity in the clinical, epidemiological and diagnostic variables selected to be included in clinical algorithms, with most of them included only in a single algorithm.

• Current evidence does not allow for a meta-analysis of studies reporting the development and/or validation of clinical algorithm for CAP owing to the high heterogeneity of studies by inclusion criteria, setting, population and outcomes assessed. Each clinical algorithm had been evaluated differently and reported various performance statistics, with a few being externally validated.

• None of the assessed clinical algorithms was included in any recent evidence-based recommendations by relevant scientific societies.

• The high heterogeneity of studies addressing clinical algorithms should be addressed in the future. No algorithm targeted more than one outcome or distinguished between two aetiologies, making it challenging to apply them to real-life situations.

Recommendations for future research

• Studies of clinical algorithms for diagnosing CA-LRTI should consider pathogen-specific outcomes (i.e. L. pneumonia) and viral versus bacterial aetiology. A clinical algorithm able to identify a bacterial aetiology with good diagnostic power could have a relevant impact on reducing the inappropriate use of antibiotics and may contribute to reducing AMR.

• The algorithms should consider more than one outcome at once, and include co-infection cases.

• Future studies should target other outcomes, such as antibiotic prescribing and time to recovery or hospitalisation, to better assess the impact of using clinical prediction rules on antibiotic stewardship.

• Diagnostic cohort studies should be preferred to case–control studies for external validation of clinical algorithms, because case–control studies are prone to spectrum bias that may affect pathogen prevalence and diagnostic performance.

• Studies should explore recent advancements in algorithmic, artificial intelligence and federated learning to leverage larger data sizes and increase the accuracy of results. Most clinical algorithms are formulated as clinical scores derived from multivariable logistic regression models. While these scores enable simple, bedside applications, they are limited in the number of variables they can include.

• RCTs should be developed to validate clinical algorithms and ensure the algorithms’ generalisability and transferability. The validation process should ensure the algorithm's performance throughout the seasons, in different geographical locations and clinical settings.

Topic 4. Is the distribution of microbial aetiologies of CA-LRTI different when using a complete microbiological investigation including RDTs in comparison with standard cultures?

Overview

Our searches identified 40 relevant studies spanning from 1999 to 2022 and were primarily prospective (87.5%, n=35) (supplementary table S8–S11, supplementary figure S9) [313–352]. Only 30% (n=12) of the studies were multicentre and included two to 17 centres [314, 315, 322, 326–328, 331, 337, 338, 343–345]. Regarding population, 15% (n=6) included only children [317, 337, 341, 342, 346, 347], 67.5% (n=27) included only adults [313, 314, 316, 318–322, 324, 327–329, 331–336, 338–340, 343–345, 348–350] and 17.5% (n=7) included patients of all ages [315, 323, 325, 326, 330, 351, 352]. Most studies were carried out in Europe (50%, n=20) [313, 314, 316, 318, 319, 321–325, 332, 333, 339, 343, 346, 348, 351, 352] and Asia (37.5%, n=15) [315, 317, 320, 326, 328, 330, 331, 334, 336–338, 340, 342, 349, 350], while only 7.5% (n=3) took place in the USA [329, 335, 347]. All the studies were conducted in hospitalised patients and only three included a subset of outpatients [314, 323, 352]. Supplementary table S10 shows the type of samples analysed in clinical studies. The heterogeneity in sampling and reporting results hinders the comparison of the reported detection rates. Additionally, not all studies reported results for all pathogens, because some focused only on typical bacteria (n=3) [316, 336, 350], typical and atypical bacteria (n=12) [315, 317, 320, 323, 326, 340, 342, 344, 345, 348, 349, 352], atypical bacteria and viruses (n=8) [314, 318, 321, 322, 324, 330, 334, 347] or only viruses (n=3) [327, 333, 341], depending on how extensive the syndromic panel was. Finally, the long timespan of the studies modifies the definition of conventional testing, i.e. conventional testing in the older studies included Gram staining, culture (bacterial and/or viral), serology (atypical bacteria and virus) and antigen testing, whereas more recent studies used singleplex molecular testing as the conventional method.

Percentage of microbiological diagnosis

From the studies reporting the detection of any pathogen with conventional methods and molecular testing, molecular methods provided a clear increase in the detection rates (1–69% to 9–95%). This increase declined when molecular methods became the conventional method, although the addition of true syndromic testing increased the rates of pathogen co-detection (supplementary tables S10 and S11).

In comparison with conventional testing, molecular methods showed increased detection of atypical bacteria and of those requiring special culture methods (supplementary table S10), but could miss untargeted pathogens. By contrast, comprehensive conventional testing provided high aetiological yields, but with a longer turnaround time than a single syndromic approach.

The detection of a pathogen, mainly bacteria, in clinical specimens does not always constitute a microbiological diagnosis, especially when detected in nonsterile specimens. This becomes accentuated with the use of metagenomic techniques, which detect all organisms present in the sample [336, 340, 344, 347]. The establishment of a cut-off to determine the aetiological agent, as well as its clinical role, is very complex, which was highlighted in most of the studies, and some aimed to provide at least standardised clinical definitions (supplementary table S10). Moreover, it is worth noting the huge variability of the gold standard used for the diagnosis according to different studies, which increases the complexity of achieving a standard microbiological diagnosis.

Prevalence of bacterial and viral infections

For the studies that explored bacterial infections, conventional methods included Gram staining, culture and antigen detection (S. pneumoniae) for typical bacteria and antigen detection (L. pneumophila) and serology (acute and convalescent antibody titres) for atypical bacteria. The introduction of molecular testing increased the percentage of detection for every pathogen, especially in the case of atypical bacteria (supplementary table S10).

For typical bacteria, the main drawback continues to be the lack of a gold standard that allows the distinction between colonisation and infection when using a respiratory sample, although some studies attempted to correlate quantitative detection with clinical significance [316]. In cases of systemic infections, molecular detection in serum or whole blood samples also increased diagnostic yield, and some studies used the detection of S. pneumoniae in urine as an aetiological determination of CAP [321, 322, 332, 333]. When applied to respiratory samples, studies highlighted the use of lower respiratory tract samples to determine that the detected bacteria were the aetiological agents of the infection. However, these studies focused on hospitalised patients, who represent only a small fraction of CA-LRTI patients, because these patients receive microbiological testing. Finally, the performance of different types of samples was compared in some studies, including sputum versus nasopharyngeal aspirates or oropharyngeal swabs, with results usually favouring the use of lower respiratory tract samples [322, 348].

For the studies that included viral testing (n=21), older conventional methods included serology in some cases, and only a few of them performed viral culture. In more recent studies, singleplex molecular testing became the conventional method. RDTs, including multiplex nucleic acid amplification tests and syndromic panels, increased the percentage of detection in comparison to serology/singleplex (supplementary table S11).

Prevalence of polymicrobial detection

The ability of conventional methods to detect polymicrobial infections might be limited. For viral detection, antigen testing is quite limited in terms of targets, and singleplex molecular testing is labour intensive and not cost-effective. Viral culture is only available in some settings, offering a confirmatory diagnosis with long turnaround times. The advent of multiplexed molecular testing circumvented these limitations, especially the development of true syndromic panels, in which bacteria and viruses are detected simultaneously. Rates of co-detection increased from 0–11% with conventional to 0–57%, mainly due to co-detection of viruses and bacteria in the same sample (supplementary table S11). Sample type can hinder the determination of detection, co-detection and aetiology, e.g. detection of S. pneumoniae in the nasopharynx is not indicative of its role as an aetiological agent.

Impact on treatment/prevalence of antibiotic resistance

Detection of AMR was only performed in one study [344]; AMR testing has been included recently in some syndromic panels, although covering a low number of targets.

Only eight studies analysed the impact of testing in antimicrobial treatment [329, 333, 334, 336, 337, 340, 343, 345]. Syndromic testing results led to a change in empirical therapy or guided the initial treatment in five of these studies [333, 336, 337, 340, 345] and a microbiological diagnosis by syndromic testing led or could have led to a decrease in the cost of therapy, either by shorter duration of antimicrobial treatment or by more targeted initial treatment, when compared to conventional testing (supplementary table S11).

Summary statements

• Literature in the field was quite limited because most studies were single centre and often only evaluated hospitalised patients, while the impact on clinical decisions was rarely assessed.

• In general, we observed an increased microbial detection-diagnostic yield when adding molecular methods, and a clear added value was found for detecting atypical bacteria and respiratory viruses. Similarly, the addition of RDTs increased the frequency of polymicrobial detection, although this does not necessarily indicate infection, and an appropriate clinical interpretation is required in these cases.

• Some studies defined criteria for establishing a microbial aetiology and differentiating from just detection of genetic/antigenic material.

• Accurate microbiological diagnosis of RTIs requires good-quality specimens. In general, samples are collected by taking medical conditions into account, either from the upper respiratory tract (nasal/throat specimens) or lower respiratory tract (bronchoalveolar lavage fluid, tracheal aspirate). Additionally, new detection methods are making it possible to use new types of specimens, and even saliva and oral mucosa may have a role to play in some situations.

• The studies including extensive sampling (blood cultures, serology, antigen detection, respiratory sampling and culture for both typical and atypical pathogens) for both conventional testing and RDTs showed an increased detection rate over those including only respiratory samples.

Recommendations for future research

Given the current literature analysis, we suggest including the following in future research:

• proper definitions of the clinical syndrome evaluated, and extensive versus pathogen-directed detection/diagnosis;

• comparability of clinical settings by laboratory testing level, including complexity of local healthcare infrastructure, availability of sentinel testing centres for general surveillance, accessibility to specialised laboratories and follow-up of the patient;

• “real-time” studies including whether the syndromic testing changed the diagnosis and the associated costs/savings derived from its introduction, i.e. not only comparing whether a pathogen was detected but giving insights on its added value to guide prescriptions.

Topic 5. Does the addition of RDTs to usual microbiological testing improve outcomes of antibiotic stewardship?

Overview

Our searches identified 17 studies and one meta-analysis eligible for data extraction published between 2014 and 2021. Most were performed in the USA [335, 353–360], two each in China [334, 361] and UK [362, 363], and one each in Europe [364], Argentina [365], Australia [366] and New Zealand [367]. Five of them were RCTs [334, 353, 363–365] and six were prospective studies [334, 335, 353, 363–365]. Only five studies out of the 17 were multicentric [355, 356, 358–360] (supplementary tables S12 and S13, supplementary figure S10).

Most studies [334, 335, 353–357, 361, 363, 368] compared single or multiplex PCR in sputum, nasopharyngeal swab or bronchoalveolar lavage samples with other conventional methods (culture, conventional PCR, antigens or clinical diagnosis only); the remaining studies explored the use of rapid pneumococcal urine antigen test (PUAT) [358–360]. In one case, PUAT and PCR were done simultaneously [362].

Regarding study population, most studies were carried out in adult subjects [334, 335, 353, 355, 356, 358, 360, 363, 364, 366], three in paediatric patients [357, 361, 367] and four in both categories [354, 359, 362, 365]. A total of 11 736 patients were included (median per study n=588, minimum–maximum n=45–2837), with 4604 in the RDT group and 5948 in the comparator group. In two studies [335, 358] with a total of 1184 patients, the same group of patients was evaluated to compare the results of different tests.

As expected, having excluded upper RTIs, disease presentation was mainly CAP [335, 353, 354, 356, 358–360, 362, 364] or unspecified acute respiratory infection/LRTI [334, 355, 357, 361, 363, 365, 367]; only one study meeting inclusion criteria considered patients with influenza-like illness [366]. All studies were conducted within hospitals, seven in ED patients [334, 335, 358, 359, 364, 365, 367], seven in hospitalised patients [353, 356, 357, 360–363] and three in both settings [354, 355, 366].

Description of outcomes: general overview

The presence of antibiotic stewardship was only reported in six studies [353, 354, 356, 358, 362, 366], while in the remaining 11 studies, there was no mention of a specific antimicrobial local strategy. Therefore, we extended the concept of our topic, considering the general impact of RDTs in guiding therapeutic choices against respiratory infection.

Viral testing was present in most studies [334, 335, 355, 357, 361–368], while detection of specific pathogens was realised only in a few studies: methicillin-resistant Staphylococcus aureus (MRSA) in four studies [335, 353, 354, 358], Pseudomonas aeruginosa in two studies [335, 358] and Haemophilus spp. in one study [335].

As for RDTs characteristics, variability in the outcomes considered in each study makes it hard to formulate evidence-based statements. However, we tried to describe the most important clinical changes based on the published data that entered our selection.

Antibiotic stewardship and changes in antibiotic treatment

A significant variability of outcomes related to antibiotic stewardship was found (table 2).

TABLE 2.

Summary of main outcomes from research question 5

Outcome	Studies (n/N)	Results		References
Antibiotic treatment
Antibiotic prescription	17/17	Antibiotics were initiated in all papers
Differences in antibiotic prescription in RDT versus comparator group	5/17	Reduced with RDT	3 studies (24–56% versus 37–88%)	[355], [366]*, [367]*
		Increased with RDT	1 study (53.9% versus 46%)	[357]*
		No difference	1 study (84% versus 83%)	[363]
Antibiotic change after RDT	9/17
Antibiotic de-escalation	5/17	More frequent in RDT*	3 studies (7.8–38.4% versus 1.1–7.8%)	[360]*, [361]*, [365]
		No difference	1 study (82.9% versus 81.2%)	[358]
		Control rate not reported	1 study	[359]
No change	1/17			[358]
Antibiotic escalation	3/17	Lower in RDT*	1 study (7.8% versus 14.16%)	[361]*
		No difference	1 study (1.1% versus 1% in adults; 1.8% versus 2.3% in children)	[365]
		Control rate not reported	1 study	[367]
Antibiotic discontinuation	2/17	Significant early discontinuation rate in RDT group compared to control*		[363]*, [367]*
Clinical and cost outcomes
Length of stay	7/17	No significant changes		[354, 356, 361–365]
Median duration of treatment	5/17	Shorter in RDT*	3 studies	[353]*, [354]*, [355]*
		Longer in RDT	1 study	[364]
		No difference	1 study	[363]
Economic impact	3/17	Lower cost in RDT	2 studies	[335], [361]*
		Higher cost in RDT	1 study	[364]

The different outcomes were reported only by a limited number of papers out of those selected for data extraction, detailed in the table. RDT: rapid diagnostic test. *: significant results (p<0.05).

Only five studies reported differences in antibiotic initiation between patients that underwent RDTs and controls, with equivocal results: in three trials [355, 366, 367], the RDT group had limited antibiotic prescription (average rate 45.50% versus 64.75%), ranging between 24% and 56% for the RDT group and between 37% and 88% in the control group. In the two remaining studies, antibiotic prescription was slightly increased in the RDT group [357] and comparable between the two groups (84% versus 83%, p=0.83) [363].

Eight studies [354, 359–361, 363, 365–367] reported a change in antibiotic treatment secondary to the use of a RDT, while in one case [358] no difference was noted. The remaining eight studies did not mention changes in treatment.

De-escalation was the most reported treatment modification in patients undergoing RDTs, ranging from 8% to 83% of cases; this huge variability seems to be related to the kind of disease treated, with more frequent de-escalation in studies considering pneumonia only than in those in which other LTRIs were included [358–361, 364]. However, the de-escalation rate in controls was only reported in half of the studies to allow group comparison: Shen et al. [361], Schimmel et al. [360] and Echavarría et al. [365] found significant differences between the two groups, suggesting a role of RDTs in guiding antibiotic therapy, while Greenfield et al. [358] found similar de-escalation rates in the control group (82.90% versus 81.20%).

Escalation was mentioned only in three studies: Walls et al. [367] described an escalation rate of 9.6% in the RDT group, without comparison with the control group. Shen et al. [361] found significantly higher percentages of escalation in the control group (7.80% versus 14.16%, p=0.002), while Echavarría et al. [365] reported similar changes between the RDT group and comparator, both in adult and paediatric age groups (adult: 1.1% versus 1%; paediatric: 1.8% versus 2.3%).

Discontinuation was reported in two studies only, both reporting a significantly increased early discontinuation rate in the RDT group (Walls et al. [367]: 11.6% versus 3.30%, p<0.25; Brendish et al. [363]: 17% versus 9%, p=0.0047).

Aside from antibiotic use, six studies explored antiviral use [355, 357, 361, 363, 365, 366]: all reported augmented or unchanged antiviral prescription when multiplex rapid PCR was performed. However, the data distribution was too wide and heterogeneous to support any indication.

Clinical outcomes

A variety of clinical outcomes were reported (table 2). Length of stay, reported in seven of 17 studies [355, 357, 362–366], did not differ between patients in the RDT group and control group. Nevertheless, Cresswell et al. [362] demonstrated a significant reduction in length of stay associated with the use of RDTs in patients with mild CAP (2.8 versus 4.4 days, p<0.001) and moderately severe CAP (4.3 versus 7.6 days, p<0.001). In severe CAP, no differences were reported.

The median duration of treatment was significantly reduced in three of five studies that reported this outcome [353–355]; in one study [364], duration of treatment in the RDT group was incremented, even if without reaching significance. The remaining study found no differences [363].

The economic impact of the use of RDTs was evaluated in only three studies: Shen et al. [361] reported a reduction of overall costs in patients in the RDT group (USD 1413.51±1438.01 versus USD 1759.37±1929.22; p=0.008), associated with a decrease in duration of treatment, early de-escalation of antibiotics and reduced escalation after 72 h. Gilbert et al. [335] demonstrated that the use of multiplex PCR eliminated the need for other conventional tests, reducing laboratory costs. Interestingly, Saarela et al. [364] obtained opposite results: while overall care costs and clinical outcomes in the two groups were comparable, laboratory costs were significantly higher in the RDT group.

These contradictory results could depend on test pricing differences between distinct countries (USA and Finland) or the use of different test bundles as a comparator.

Regarding the remaining clinical outcomes (mortality, relapse/rehospitalisation and antibiotic side-effects), the scarcity and heterogeneity of data prevented our drawing any conclusions.

Summary statements

• Most studies assessed the use of multiplex PCR guiding the management of hospitalised patients in the ED.

• Only a limited number of centres reported the presence of antibiotic stewardship, possibly indicating that this measure of antibiotic control is still underused in clinical practice.

• Our literature search did not find enough data to demonstrate that RDT can guide antibiotic treatment; however, most of the included articles described some kind of treatment optimisation secondary to their use (i.e. antibiotic de-escalation). Although we excluded studies that considered RDTs together with non-microbiological biomarkers, it is interesting to note that studies assessing some biomarkers, e.g. PCT, in combination with RDTs (composite intervention) reported more considerable changes in antibiotic use [258, 369, 370]. The same trend was noted when RDTs were accompanied by non-pharmacological strategies, such as antibiotic surveillance teams or automatic reports of results and/or treatment indications [371].

• Clinical evaluation is still the fundamental criterion guiding antibiotic treatment, and even if RDTs can confirm clinical suspicion, other tools such as biomarkers and antibiotic stewardship protocols have an important role in guiding physician decisions.

• Our literature review has not found enough data to establish the usefulness of RDTs in optimising antibiotic treatment in respiratory infection.

Recommendations for future research

• Appropriate randomised clinical trials are urgently needed to evaluate the role of RDTs for optimising antibiotic treatment.

• These studies should also investigate the interactions between the use of RDTs, (mis)use of antimicrobials and emerging AMR.

• More investigation is needed to assess the potential role of RDTs in antimicrobial stewardship protocols for high-risk patients, e.g. immunosuppressed or multimorbid individuals.

• Specific research is needed to assess the potential role of RDTs in antiviral stewardship protocols in the context of seasonal peaks of influenza and SARS-COV-2.

• More research is needed on the health-economic aspects of the use of RDTs to improve outcomes of antimicrobial stewardship.

Discussion

The literature review of the common RDTs used for the management of CA-LRTI has highlighted numerous limitations related in part to the quality of studies and in part to the heterogeneity of tests and pathogens.

Overall, RDTs based on Ag or Ab detection of viral pathogens (influenza, RSV, hMPV and SARS-CoV-2) have shown high specificity but suboptimal sensitivity. Lower sensitivity and specificity values have been described for diagnosis of bacterial CA-LRTI (S. pneumoniae and M. pneumoniae).

According to current literature, the addition of molecular tests to conventional microbiology has increased the microbial detection-diagnostic yield, particularly in cases of atypical bacteria and respiratory viruses and, in general, of polymicrobial infection. However, not all identified microorganisms are necessarily disease-causing; specific knowledge and experience in this field is required to properly interpret these results and use them in clinical practice. In some cases, referral to experts in the field is usually considered to avoid overtreating patients.

Biomarkers are useful tools in improving clinical management of patients presenting with symptoms of CA-LRTI, both in PC and hospital settings.

More specifically, CRP is particularly useful in diagnosing LRTI and CAP in PC settings, as well as guiding decisions on the site of care, risk stratification and predicting mortality.

PCT can successfully differentiate bacterial versus viral CA-LRTI, successfully guiding antibiotic stewardship, assessing response to treatment and reducing duration of treatment. The use of both CRP and PCT in several clinical trials did not lead to increased treatment failure and/or relapse rates compared to standard care.

Unfortunately, studies assessing the utility of clinical algorithms in CA-LRTI have shown major variability in definitions, methods and outcomes. These variations and the lack of clear validation standards limit the applicability of these algorithms in real life and their inclusion in clinical recommendations by medical societies.

Finally, the clinical impact of RDTs in the management of CA-LRTI has still not been sufficiently investigated. While antibiotic de-escalation or discontinuation and appropriate antiviral prescription seem to be positively associated with the use of RDTs, the duration of treatment, costs and other clinical outcomes have shown contrasting results. Thus, we currently do not have enough data to demonstrate the usefulness of RDTs in optimising antibiotic treatment in CA-LRTI.

Although it is out of the scope of this review, studies assessing composite interventions to support antimicrobial stewardship, such as the combination of RDTs and biomarkers (i.e. PCT) or RDTs and non-pharmacological strategies, have shown promising results and should be further assessed.

Limitations

The major limitation of this review is the extreme heterogeneity of studies, which reduces our ability to draw conclusions that would be potentially useful for clinical practice.

For instance, the studies were heterogeneous in the definitions used for conventional testing (including or not molecular tests) depending on timeframe; in the sensitivity and specificity values, depending on pathogen, method and sample type [372]; in clinical diagnosis settings; in study population (adults and/or children); and in comparisons.

Moreover, while studies assessing the accuracy of RDTs have been mostly performed in long-term care and outpatient settings, most studies on microbial aetiologies and biomarkers were performed in hospitalised patients and in all cases the impact on clinical outcomes such as cost–benefit balance were poorly investigated.

Assessing RDTs and biomarkers in clinical practice by implementing specific algorithms is extremely difficult. In fact, every clinical algorithm had been evaluated differently and reported various performance statistics, making it impossible to achieve clear recommendations beyond the RCTs.

Overall, there are numerous limitations to consider even beyond the already mentioned heterogeneity of studies. For instance, publication bias should be considered with regards to the potential overrepresentation of small studies (e.g. Topics 2 and 5). In addition, geographic limitations are also evident with regards to the use of RDTs or related algorithms, because their cost is a barrier for low-income countries that is not sufficiently represented in the publications. Lastly, the fast progress in technological tools could make the diagnostic tests used in the current literature out of date in the short term. All these potential limitations can affect the generalisability of the statements in this document with regards to different healthcare systems or settings in which RDTs might be too expensive or not easily available.

However, the gap analysis is crucial for the definition of research priorities in this field, and we hope this report can inspire future research.

General recommendations for future research

RDTs are a potentially valid resource for the diagnosis of CA-LRTI in a community-care setting. However, evidence on feasibility, accuracy by type of patient and cost–benefit is still limited.

Randomised clinical trials comparing the accuracy of different RDTs and/or biomarkers embedded in diagnostics algorithms including laboratory and imaging tests are strongly encouraged. The studies should include assessment of follow-up to define the risk of adverse effect of RDTs in terms of delayed or inappropriate diagnosis. It is also essential to include antibiotic treatment strategies based on antibiotic stewardship recommendations.

Clinical algorithms including RDTs and biomarkers could be beneficial in reducing antibiotic misuse. Still, more consensus is needed regarding the clinical, epidemiological and diagnostic data to be included and the validation methods to be used.

Composite interventions combining RDTs with biomarkers or other non-pharmacological interventions should be further investigated to provide better support for antimicrobial stewardship.

Supplementary material

Please note: supplementary material is not edited by the Editorial Office, and is uploaded as it has been supplied by the author.

Supplementary material

ERJ-01601-2024.Supplement

Executive summary

ERJ-01601-2024.Executive_Summary