The Clinical Utility of Point-of-Care Tests for Influenza in Ambulatory Care: A Systematic Review and Meta-analysis

We reviewed the evidence for the impact of point-of-care diagnostic tests for influenza. Testing reduced further investigation with chest radiography and full blood counts, and increased antiviral prescribing, but had no impact on antibiotic use, returning for care, or admissions.

Influenza is a major global disease. The World Health Organization estimates 1 billion infections and half a million deaths from respiratory complications each year [1,2]. Influenza affects healthcare, society, and the world economy, although often the impact is attributed to other infections such as pneumonia [3][4][5]. In the United Kingdom, influenza is responsible for more than half a million primary care consultations and more than 19 000 hospital admissions and deaths each year, though they are often not recognized as influenza [5].
Many respiratory infections cause the same syndrome as influenza; these are referred to as influenza-like illnesses (ILIs) [1]. Despite being unable to distinguish clinical features of influenza from other causes of ILI, clinical diagnosis is widespread [6,7]. Diagnostic uncertainty in ILI contributes to antibiotic prescribing [8], so diagnostics could improve antimicrobial stewardship. UK guidance from the National Institute for Health and Care Excellence recommends no antibiotic prescribing for patients with respiratory tract infections that are likely to be self-limiting, including influenza, unless patients are systemically unwell or at higher risk of unfavorable outcomes [9]. Nonetheless, 14%-40% of patients with influenza are prescribed antibiotics [10,11].
Influenza point-of-care tests (POCTs) are specific (>98%), but rapid antigen detection tests (RADTs) have low sensitivity compared to nucleic acid amplification tests (53%-54% vs 92%-95%) [12]. Even RADTs offer more accurate diagnoses than clinical evaluation and are fast enough to influence prescribing in ambulatory settings [6,13]. We cannot assume POCTs will automatically lead to beneficial outcomes [14]. This review aims to collate the available evidence on the impact of point-of-care influenza tests in ambulatory care. We sought to examine clinically relevant impacts, including hospital admissions, antibiotic and antiviral prescribing, and the use of other diagnostic tests.
in the 6 most important medical databases (Supplementary Materials). We updated the search, which included terms for POCTs for any condition, on 21 March 2017. We selected studies of influenza POCTs at the full text stage, and will publish findings for other POCTs elsewhere.

Inclusions and Exclusions
Participant demographics and preexisting conditions were not restricted. We included the following ambulatory care settings: primary care, emergency department, and clinic, but we did not include studies of hospitalized patients. We excluded tests sent to a different location for analysis, such as a laboratory. We included any POCTs for diagnosis of influenza, with or without other tests. Nondiagnostic biomarkers alone were ineligible. We compared POCTs with usual care. This could include no testing or laboratory tests for influenza, but not another novel test. We included all quantitative clinical outcomes, excluding health economic outcomes. When extracting data on further tests, we grouped routine blood tests with full blood counts and combined urinalysis techniques. We included randomized, controlled trials (RCTs) and nonrandomized studies for separate analysis. We excluded study designs that precluded comparisons between tested and untested participants (case studies, case series, and studies without controls).
We screened articles independently in duplicate at title, abstract, and full-text levels. Discussion or a third reviewer resolved conflicts. J. L. extracted data and assessed quality, and J. V. checked data extraction and quality assessment. We contacted corresponding authors for unpublished information.

Analyses
We used random effects meta-analyses to generate pooled estimates with 95% confidence intervals (CIs) and I 2 . We estimated risk ratios (RRs) for dichotomous outcomes and mean differences or standardized mean differences (where outcomes may have been measured differently) for continuous outcomes. We planned to calculate missing estimates using methods from the Cochrane handbook [16] (but were unable to do so) and used sensitivity analyses, omitting studies to explore heterogeneity. We used post hoc random effects metaregression to explore heterogeneity attributable to the prevalence of influenza and baseline outcomes where 10 or more studies reported an outcome using the log odds scale to allow linear regression [17]. We used Covidence software for citation management [18]. Metaanalysis was undertaken with Revman 5. 3 [19], metaregression with Stata 14 SE [20].

RESULTS
The searches resulted in 12 928 unique records ( Figure 1); 12 269 were excluded by title and abstract screening, and the remaining 659 underwent full-text review. A total of 225 full texts were eligible for inclusion in 1 or more review. Thirteen studies were of influenza POCTs ( Figure 1 and Table 1).
Randomized studies were of moderate risk of bias; nonrandomized studies had a higher risk ( Figure 2). None of the studies were able to blind participants and personnel to testing or test results. We found no study that blinded outcome assessors to test status.

Patient Outcomes
No study reported mortality and morbidity measures, such as illness course. Most outcomes were measures of impact on management decisions and further investigation ( Table 2).
Of the 5 nonrandomized studies that reported on antibiotic prescribing, 4 [28,[31][32][33] reported significant reductions. Meta-analysis showed a strong association between POCTs and reduced antibiotic prescribing but with strong evidence of statistical heterogeneity (RR, 0.64; 95% CI, 0.48 to 0.86; I 2 = 81%). Random effects metaregression of all study types showed much of the heterogeneity in study log odds ratios could be attributed to the baseline proportion of patients with influenza and antibiotic prescribing (antibiotic prevalence in control arm × influenza prevalence; Supplementary Figure S5). The proportion of variation in antibiotic prescribing between studies (I 2 ) that the model attributed to this feature (adjusted R 2 ) was 79% (P = .003).
A single randomized study estimated the duration of antibiotic treatment. Esposito et al [23] found no evidence of a difference between groups (mean difference, 0.00; 95% CI, −0.35 to 0.35).
The composite outcome of any further testing was extractable from 2 studies (Supplementary Figure S7). A randomized trial estimated an RR of 0.83 (95% CI, 0.65 to 1.07) [27], and a nonrandomized study of decisions before and after test results were revealed to clinicians estimated an RR of 0.53 (95% CI, 0.43 to 0.64) [30].
Three RCTs reported blood cultures [21,22,26]. All 3 had point estimates in the direction of a reduction, but none were significant. The pooled estimate showed a significant reduction in blood cultures with point-of-care testing (RR, 0.82; 95% CI, 0.68 to 0.99; I 2 = 0%; Supplementary Figure S8). Chest radiography was reported in the 7 RCTs [21-27] and 3 nonrandomized studies [29][30][31]. Metaanalysis of the randomized trials (n = 4161) gave a pooled RR of 0.81 (95% CI, 0.68 to 0.96; I 2 = 32%; Figure 5). The results from the Cohen et al study were in the opposite direction of the other studies; a sensitivity analysis without the Cohen et al study removed all heterogeneity, and the result was robust (RR, 0.80; 95% CI, 0.70 to 0.91; I 2 = 0%). The 3 nonrandomized studies of 1009 participants had a pooled estimate that was similar to those of the randomized studies, but it was not significant (RR, 0.77; 95% CI, 0.57 to 1.05; I 2 = 65%). The largest reductions tended to be in studies with higher influenza and higher chest radiography. Random effects metaregression included randomized and nonrandomized studies (Supplementary Figure S9). We regressed log odds  Figure 3. Antibiotic prescribing. Abbreviations: CI, confidence interval; POCT, point-of-care test; RCT, randomized, controlled trial. ratios for chest radiography against the proportion of patients with influenza who might undergo chest radiography (influenza prevalence × radiography use in control arms). Up to 100% of between-study variance could be attributed to this combination of influenza prevalence and baseline requesting rates in studies of all study designs (adjusted R 2 = 100%; P = .03).   Urinalysis was not affected by influenza point-of-care testing, based on the meta-analysis of 5 randomized studies (RR, 0.91; 95% CI, 0.78 to 1.07; I 2 = 20%; Supplementary Figure S10) [21,22,24,26,27]. The only nonrandomized study looked at theoretical clinical decisions and found that urinalysis declined by approximately half in children with pyrexia of unknown origin (RR, 0.47; 95% CI, 0.37 to 0.61; Supplementary Figure S10) [30].

DISCUSSION
POCTs reduced the risk of routine blood tests by 20%, blood cultures by 18%, and chest radiography by 19% in RCTs. These results suggest POCTs have a role in reducing diagnostic uncertainty for children with ILI, but the impact on patient outcomes remains unclear. Antibiotic prescribing was not affected by testing, but prescriptions for antiviral medications more than doubled. POCTs did not affect time in the emergency department or numbers of patients returning for care. Most evidence came from RADTs, which are known to be specific but have low sensitivity [6]. Newer tests have higher sensitivity [12], which may increase their impact.
Nonrandomized studies had different results compared to RCTs that included reduced antibiotic prescribing, overall testing, and urinalysis. Routine blood tests were reduced in trials but not in nonrandomized studies. We attribute the differences to baseline prescribing rates and influenza prevalence but also to higher risk of bias. Diagnostics are complex interventions [35]; clinical context, flow, and timing are important components that affect impact. A POCT cannot reduce further testing unless the POCT result is available and considered before further tests are requested, which may not happen outside of trials. The nonrandomized result was driven by a large cohort study that included adults and children before and after a Korean emergency department introduced a POCT [29]. In that study, POCTs had become routine, so clinicians may have requested them at the same time as blood tests. Overall, tests and urinalysis were examined in only 1 nonrandomized study of questionable risk of bias [30]. Investigators asked clinicians for their decisions before and after having a result revealed to them. Asking in this way likely focused the clinicians' attention on what they can do differently. The impact of POCTs may be less without this interaction, although a carefully designed implementation might replicate it.
The diagnostic accuracy of POCTs for influenza has been examined extensively in individual studies and systematic reviews [6,12,36], but we looked at direct evidence of clinical outcomes. We believe this is the first systematic review of the impact of influenza POCTs on clinical outcomes and includes all relevant primary studies. Individual studies had insufficient power to show effects. Pooling results from nonsignificant studies allowed us to reveal previously unknown effects on the outcomes of chest radiography, antiviral prescribing, blood cultures, and routine blood tests.
This review used a comprehensive search strategy and included a variety of study types. It is unlikely that we missed a large body of work that would change the interpretation of the results. Our inclusion of nonrandomized studies has advantages-a priori it was unlikely trials would be powered to address rarer and more serious complications of influenza, and this is what we found. Unfortunately, no observational evidence for these outcomes exists.
The available evidence limits this review. There is little evidence from primary care settings. Most of the evidence comes from low-sensitivity RADT tests. Higher-sensitivity tests would detect more influenza and increase prevalence estimates. In addition to the direct impact of fewer false negatives, better tests might increase clinicians' confidence to act on results.
RCTs had lower risk of bias than nonrandomized studies. RCTs were at moderate risk, but future studies are unlikely to be much lower risk. The Cochrane risk-of-bias tool gives harsh results for trials of diagnostic tests as interventions. Allocation concealment is impossible because effects work through knowledge of the test result. Blinding of outcome assessment is also difficult for self-reported outcomes.
Our examination of between-study heterogeneity underlines the importance of the prevalence of both influenza and outcomes of interest to clinicians, future studies, and policy makers.
Studies of POCTs in both adults and children are needed; there is an evidence gap in primary care settings. Most patients are seen in primary care settings, but the low prevalence of serious outcomes would require large studies [37]. Studies will need to be even larger to account for influenza's variable and generally low prevalence, even during epidemics. Consequently, there is a space for well-conducted observational studies.
Future studies should examine clinical course, mortality, and morbidity measures. They should report outcomes by POCT results as well as status because effects may differ by result. Studies should examine the results of additional tests as a proxy for appropriateness of further investigation [38]. Reducing negative tests implies efficiency, but reducing positive tests implies missed bacterial infections.
Future research should explore appropriate contexts for POCT use and implementation. Combinations of newer influenza tests with other POCTs, C-reactive protein, for example, may help better identify patients with bacterial coinfections and give clinicians confidence to conserve antibiotics. Bias assessment for randomized trials of tests as interventions and therefore idealized study designs and reporting guidelines should be a research priority.
POCTs for influenza have a role for children with ILIs, particularly in emergency departments and during influenza epidemics. There is little evidence for or against implementation in primary care. Clinicians should consider local practice before implementation. Influenza POCTs reduce blood tests and chest radiography, but the reduction is greatest in settings with high levels of additional tests.
Tests are not a substitute for clinical assessment. We have not addressed the appropriateness of reducing chest radiography, blood cultures, or routine blood tests. However, the vast majority of childhood infections are self-limiting illnesses, so reductions are likely to be appropriate. The benefit of antiviral prescribing is debatable. Recent reviews have suggested benefit [39], but a Cochrane review was derisive about the evidence for effectiveness [40].

CONCLUSIONS
There is evidence from randomized trials that influenza POCTs influence clinical decisions in ambulatory care, resulting in fewer blood tests and chest radiographs. The evidence is mostly for rapid antigen tests in children in emergency department settings.

Supplementary Data
Supplementary materials are available at Clinical Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.