Biomarkers as point‐of‐care tests to guide prescription of antibiotics in patients with acute respiratory infections in primary care
Abstract
Background
Acute respiratory infections (ARIs) are by far the most common reason for prescribing an antibiotic in primary care, even though the majority of ARIs are of viral or non‐severe bacterial aetiology. Unnecessary antibiotic use will, in many cases, not be beneficial to the patients' recovery and expose them to potential side effects. Furthermore, as a causal link exists between antibiotic use and antibiotic resistance, reducing unnecessary antibiotic use is a key factor in controlling this important problem. Antibiotic resistance puts increasing burdens on healthcare services and renders patients at risk of future ineffective treatments, in turn increasing morbidity and mortality from infectious diseases. One strategy aiming to reduce antibiotic use in primary care is the guidance of antibiotic treatment by use of a point‐of‐care biomarker. A point‐of‐care biomarker of infection forms part of the acute phase response to acute tissue injury regardless of the aetiology (infection, trauma and inflammation) and may in the correct clinical context be used as a surrogate marker of infection, possibly assisting the doctor in the clinical management of ARIs.
Objectives
To assess the benefits and harms of point‐of‐care biomarker tests of infection to guide antibiotic treatment in patients presenting with symptoms of acute respiratory infections in primary care settings regardless of age.
Search methods
We searched CENTRAL (2013, Issue 12), MEDLINE (1946 to January 2014), EMBASE (2010 to January 2014), CINAHL (1981 to January 2014), Web of Science (1955 to January 2014) and LILACS (1982 to January 2014).
Selection criteria
We included randomised controlled trials (RCTs) in primary care patients with ARIs that compared use of point‐of‐care biomarkers with standard of care. We included trials that randomised individual patients as well as trials that randomised clusters of patients (cluster‐RCTs).
Data collection and analysis
Two review authors independently extracted data on the following outcomes: i) impact on antibiotic use; ii) duration of and recovery from infection; iii) complications including the number of re‐consultations, hospitalisations and mortality; iv) patient satisfaction. We assessed the risk of bias of all included trials and applied GRADE. We used random‐effects meta‐analyses when feasible. We further analysed results with a high level of heterogeneity in pre‐specified subgroups of individually and cluster‐RCTs.
Main results
The only point‐of‐care biomarker of infection currently available to primary care identified in this review was C‐reactive protein. We included six trials (3284 participants; 139 children) that evaluated a C‐reactive protein point‐of‐care test. The available information was from trials with a low to moderate risk of bias that address the main objectives of this review.
Overall a reduction in the use of antibiotic treatments was found in the C‐reactive protein group (631/1685) versus standard of care (785/1599). However, the high level of heterogeneity and the statistically significant test for subgroup differences between the three RCTs and three cluster‐RCTs suggest that the results of the meta‐analysis on antibiotic use should be interpreted with caution and the pooled effect estimate (risk ratio (RR) 0.78, 95% confidence interval (CI) 0.66 to 0.92; I2 statistic = 68%) may not be meaningful. The observed heterogeneity disappeared in our preplanned subgroup analysis based on study design: RR 0.90, 95% CI 0.80 to 1.02; I2 statistic = 5% for RCTs and RR 0.68, 95% CI 0.61 to 0.75; I2 statistic = 0% for cluster‐RCTs, suggesting that this was the cause of the observed heterogeneity.
There was no difference between using a C‐reactive protein point‐of‐care test and standard care in clinical recovery (defined as at least substantial improvement at day 7 and 28 or need for re‐consultations day 28). However, we noted an increase in hospitalisations in the C‐reactive protein group in one study, but this was based on few events and may be a chance finding. No deaths were reported in any of the included studies.
We classified the quality of the evidence as moderate according to GRADE due to imprecision of the main effect estimate.
Authors' conclusions
A point‐of‐care biomarker (e.g. C‐reactive protein) to guide antibiotic treatment of ARIs in primary care can reduce antibiotic use, although the degree of reduction remains uncertain. Used as an adjunct to a doctor's clinical examination this reduction in antibiotic use did not affect patient‐reported outcomes, including recovery from and duration of illness. However, a possible increase in hospitalisations is of concern. A more precise effect estimate is needed to assess the costs of the intervention and compare the use of a point‐of‐care biomarker to other antibiotic‐saving strategies.
PICOs
Plain language summary
Use of rapid point‐of‐care testing for infection to guide doctors prescribing antibiotics for acute respiratory infections in primary care settings
Review question
We reviewed the evidence of the effect and safety of a rapid test of infection at point‐of‐care for using antibiotics in people with acute respiratory infections (ARIs) (e.g. common colds) in primary care.
Background
Antibiotic treatment is common in ARIs despite the fact that the vast majority are caused by viruses, against which antibiotics are ineffective and unnecessary. The concern is that antibiotics may cause side effects and are directly associated with antibiotic resistance in common bacteria, causing treatment failure and complications, including death. Antibiotics have a modest, if any, effect against the majority of ARIs. Their use must be balanced against risking higher levels of antibiotic resistance, side effects and costs. Biomarkers of infection are proteins or components of the immune system that participate in the body's acute response to infection. No tests are currently able to provide perfect diagnostic accuracy for infections. This could lead to over‐ as well as under‐diagnosis. Some tests have been developed that assess the presence of infections by looking for certain of these biomarkers. These are rapid tests that may be used during the consultation by primary care doctors when people go to see them with symptoms of an ARI. In the correct clinical context these point‐of‐care tests could assist primary care doctors by identifying people with infections that are most likely to respond to antibiotics. We looked at the evidence for these tests to assess the possible harms and benefits of implementing such a strategy in primary health care.
Study characteristics
We included six studies with a total of 3284 participants with ARIs from primary care settings (point‐of care test: C‐reactive protein). Two of the included studies received direct financial support from manufacturers. The evidence is current to January 2014.
Key results
The only point‐of‐care biomarker of infection currently available to primary care identified in the review was C‐reactive protein. A reduction in antibiotic use is likely to be achieved by a C‐reactive protein point‐of‐care test but due to differences in the designs of the included studies, it was not possible to obtain a precise effect estimate of the reduction. There were no deaths in the studies and we did not find evidence suggesting that time to recovery from ARIs and their duration were longer, nor that levels of patient satisfaction or number of re‐consultations were affected in the C‐reactive protein group. However, a possible increase in the risk of hospital admission cannot be ruled out.
Quality of the evidence
We ranked the evidence as of moderate quality according to the GRADE levels due to an imprecise effect estimation.
Conclusion
Used as an adjunct to a doctor's clinical examination point‐of‐care tests (e.g. C‐reactive protein) can reduce antibiotic use in ARIs in general practice. The possibility of an increased risk of hospital admission suggests that care must be taken in how these tests are used. A more precise effect estimate is needed to assess the costs of the intervention and compare the use of a point‐of‐care biomarker to other antibiotic‐saving strategies.
Authors' conclusions
Summary of findings
| Point‐of‐care biomarker for infection compared with standard of care for guiding antibiotic therapy in acute respiratory infections | ||||||
| Patient or population: patients with acute respiratory infections Settings: primary care Intervention: point‐of‐care biomarker (C‐reactive protein) test Comparison: standard care | ||||||
| Outcomes | Illustrative comparative risks* (95% CI) | Effect | No. of participants | Quality of the evidence | Comments | |
| Assumed risk | Corresponding risk | |||||
| Standard care | C‐reactive protein | |||||
| Mortality (C‐reactive protein) Follow‐up: 28 days | ― | ― | ― | 3284 (6) | ⊕⊕⊕⊕ | No participants died in these studies |
| No. of antibiotic prescriptions (C‐reactive protein) Index consultation | Individual RCTs: study population | RR 0.90 (0.80 to 1.02) | 1309 | ⊕⊕⊕⊝ | I2 statistic = 5% | |
| 519 per 1000 | 467 per 1000 | |||||
| Cluster‐RCTs: study population | RR 0.68 (0.61 to 0.75) | 1975 (3) | I2 statistic = 0% | |||
| 525 per 1000 | 357 per 1000 | |||||
| No. of antibiotic prescriptions (C‐reactive protein) Follow‐up: 28 days | Individual RCTs: study population | RR 0.87 (0.75 to 1.02) | 497 | ⊕⊕⊕⊝ | I2 statistic = 7% | |
| 623 per 1000 | 542 per 1000 | |||||
| Cluster‐RCTs: study population | RR 0.68 (0.51 to 0.91) | 211 (2) | I2 statistic = 19% | |||
| 629 per 1000 | 428 per 1000 | |||||
| Clinical recovery. No. of participants with at least 'substantial improvement' Follow‐up: 7 days | Individual RCTs: study population | RR 1.03 (0.93 to 1.14) | 1264 | ⊕⊕⊕⊝ | I2 statistic = 0% | |
| 414 per 1000 | 426 per 1000 | |||||
| Clinical recovery Follow‐up: 28 days | Individual and cluster‐RCTs: study population | RR 0.94 (0.69 to 1.28) | 527 | ⊕⊕⊕⊝ | I2 statistic = 0% | |
| 758 per 1000 | 713 per 1000 | |||||
| *The assumed risk was calculated as the median control group risk across studies. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). | ||||||
| GRADE Working Group grades of evidence | ||||||
| 1We downgraded the GRADE judgement to moderate as the heterogeneity, albeit well explained, generates imprecision in the main effect estimate. | ||||||
Background
Description of the condition
Treating acute respiratory infections (ARIs) with antibiotics is common in primary care settings, despite their predominant (> 70%) viral aetiology (Gonzales 2001; Goossens 2005; Harnden 2007; Pavia 2011), and the fact that antibiotic treatment is of marginal benefit in uncomplicated cases (Arroll 2005; Butler 2009; Butler 2011; Little 2013b; Meropol 2013). Antibiotic use is associated with antibiotic resistance, which in turn leads to ineffective treatments and increased healthcare costs (Carlet 2011; Smith 2013). Limiting unnecessary antibiotic prescriptions in primary care settings is pivotal in reducing bacterial resistance to antibiotics at both societal (Gonzales 2001; Bronzwaer 2002; Sande‐Bruinsma 2008) and individual levels (Costelloe 2010), as well as reducing the risk of side effects. A reduction in antibiotic prescriptions in primary care settings will have a large impact on the total use of antibiotics, as the majority of antibiotic prescriptions are issued in primary care settings (Danmap 2010; Goossens 2005). Nevertheless, patient safety must be carefully assessed to minimise the risk of under‐treatment of serious bacterial infections.
Other types of interventions to reduce antibiotic use have been studied, for example, educational interventions (Arnold 2009), where use of multifaceted approaches and communication skills training have been effective (Butler 2012; Gjelstad 2013). A policy of delayed antibiotic prescription can also reduce antibiotic use (Spurling 2011).
The decision to prescribe antibiotics for an ARI in primary care settings is challenging and often based solely on clinical symptoms (Hopstaken 2005a), an approach known to have both low sensitivity and specificity (Hoare 2006; Metlay 1997) and high inter‐observer variability (Wipf 1999). In accordance with this, there is evidence of substantial between‐practitioner differences (Stocks 2002), and geographical variation in antibiotic prescribing patterns (Matthys 2007).
Description of the intervention
Biomarkers of infection, such as white blood cell levels, procalcitonin and C‐reactive protein, form part of the acute immune response and are activated by endogenous and exogenous stimuli following tissue injury due to infectious and non‐infectious conditions such as inflammation and trauma. Circulating levels are low in healthy people, but when stimulated synthesis and recruitment is rapid (less than 20 hours) levels remain high as long as the inflammation and tissue damage persists and then decline rapidly (Becker 2004; Volanakis 2001). Biomarkers of infection act as surrogate measures of the immune response to infection and may reflect the severity of the condition (i.e. degree of tissue damage and immune activation) (Aabenhus 2011; Kruger 2009; Schuetz 2012), but cannot determine aetiology or predict an infiltrate on chest X‐rays (Holm 2007; van der Meer 2005). No currently available test is able to provide perfect diagnostic accuracy, and false negative as well as false positive results may occur, leading to possible over‐ as well as under‐treatment of ARIs. However, in the correct clinical context biomarkers may guide appropriate antibiotic prescriptions in selected cases by ruling out a serious bacterial infection and identify patients in whom no benefit from antibiotic treatment can be anticipated (Melbye 2011; Schuetz 2012). A point‐of‐care test exists for some of these biomarkers to be performed at, or near, the site of patient care, delivering quick test results that can influence clinical decisions (Table 1).
| Biomarker | Status | Handling | Biochemistry |
| C‐reactive protein (CRP) | POC* test available | Droplet blood from finger prick. Results in approximately 3 minutes. Uninfected adult controls have levels < 10 mg/L | Inflammatory cytokines trigger C‐reactive protein release by the liver. Levels of C‐reactive protein increase within 6 to 18 hours, peaking at 48 to 72 hours |
| Leukocyte count | POC test available | Droplet blood from finger prick. Results in approximately 3 minutes. Uninfected adult controls have leukocyte levels < 9 x 109/L and neutrocyte levels < 7 x 109/L | Cells of the immune system activated by inflammatory cytokines and foreign antigens |
| Procalcitonin (PCT) | POC test not available** | Uninfected adult controls have levels < 0.05 nanogram/mL | Inflammatory cytokines and bacterial endotoxins trigger release of PCT from parenchymal tissues. Levels of PCT increase within 2 to 6 hours, peaking at 24 to 48 hours |
*POC: point‐of‐care
**No POC test in desired target range (0.05 to 0.50 nanogram/mL)
The decision to prescribe antibiotics for an ARI is guided by pre‐specified cut‐off values specific to the individual point‐of‐care test but the test cannot replace clinical skills and expertise, and test results may be overruled on clinical grounds.
How the intervention might work
Following a regular clinical examination that suggests presenting symptoms are indeed compatible with an ARI, a point‐of‐care biomarker may assist the clinician to assess the likelihood of a serious bacterial infection versus a less severe bacterial or viral infection, thus identifying those patients most likely to benefit from antibiotics (Aabenhus 2011; Hopstaken 2003; Melbye 2011; Schuetz 2012). If after the clinical examination the clinician is confident in the decision to initiate or withhold antibiotic treatment, there is no need for a point‐of‐care test. Possible detrimental effects of point‐of‐care biomarkers include suboptimal use of time, costs, handling errors, patient dissatisfaction and false negative values that can lead to lack of necessary antibiotic treatments or false positive values that may increase inappropriate antibiotic use. Studies indicate that the use of point‐of‐care tests during consultations is acceptable to both doctors and patients (Butler 2008; Wood 2011).
Why it is important to do this review
Avoiding both over‐ and under‐treatment with antibiotics in primary care settings is important to limit antibiotic resistance and exposure of patients to unnecessary risks. Debate concerning the effect of using point‐of‐care biomarkers is ongoing as published reviews have shown conflicting results (Engel 2011; Huang 2013; Schuetz 2012). However, additional potential relevant studies have been published since then and only the review of procalcitonin assessed patient safety outcomes in a systematic way (Schuetz 2012).
We included studies of all available point‐of‐care biomarkers of infection used for ARIs in our review. Updates of this review will include studies of additional point‐of‐care tests as they become available.
Objectives
To assess the benefits and harms of point‐of‐care biomarker tests of infection to guide antibiotic treatment in patients presenting with symptoms of acute respiratory infections in primary care settings regardless of age.
Methods
Criteria for considering studies for this review
Types of studies
Randomised clinical trials (RCTs) and cluster‐RCTs.
Types of participants
Primary care patients of all ages with symptoms from, or a diagnosis of, an ARI at study entry. Symptoms of ARI were defined as cough, discoloured/increased sputum, fever, runny nose, respiratory distress, feeling unwell, or combinations of focal and systemic symptoms having a duration of less than four weeks. Diagnoses included lower or upper respiratory tract infection, pneumonia, bronchitis, acute exacerbations of chronic obstructive pulmonary disease or asthma, pharyngitis, tonsillitis, laryngitis, rhinosinusitis, common cold, acute otitis media or influenza.
Types of interventions
Point‐of‐care biomarkers of infection to guide antibiotic treatment for ARI in primary care settings. We only included studies of biomarker point‐of‐care tests for infections available for general use. Specific diagnostic tests like the Strep A test or Monospot were not included in this review. The biomarkers we considered were C‐reactive protein, procalcitonin and white blood cell count. The comparator was standard care.
Types of outcome measures
Primary outcomes
-
Number of patients given an antibiotic prescription at the index consultation and at 28 days follow‐up.
-
Number of patients with substantial improvement (including full recovery) at day seven.
-
Total mortality at 28 days follow‐up.
Secondary outcomes
-
Number of patients in need of a re‐consultation at 28 days follow‐up.
-
Number of patients in need of a hospital admission at 28 days follow‐up.
-
Duration of the ARI (e.g. mean or median days with restrictions in daily activities due to the infection).
-
Number of satisfied patients.
-
Number of patients with substantial improvement (including full recovery) at 28 days follow‐up.
Search methods for identification of studies
Electronic searches
We searched the Cochrane Central Register of Controlled Trials (CENTRAL) (2013, Issue 12) (accessed 16 January 2014), MEDLINE (1946 to January week 2, 2014), EMBASE (2010 to January 2014), CINAHL (1981 to January 2014), Web of Science (1955 to January 2014) and LILACS (1982 to January 2014).
The search strategy used for CENTRAL and MEDLINE is described in Appendix 1. We combined the MEDLINE search with the Cochrane Highly Sensitive Search Strategy for identifying randomised trials in MEDLINE: sensitivity‐ and precision‐maximising version (2008 revision); Ovid format (Lefebvre 2011). We adapted the search strategy to search EMBASE (Appendix 2), CINAHL (Appendix 3), Web of Science (Appendix 4) and LILACS (Appendix 5). We applied no language or publication restrictions.
Searching other resources
Trials
We searched the trials registries of the US National Institutes of Health (www.clinicaltrials.gov) and the World Health Organization (www.who.int/ictrp) in March 2013 for completed and ongoing trials. We repeated the search in WHO ICTRP in January 2014.
Abstracts
We checked abstracts presented at the following conferences from 2000 onwards.
-
British Thoracic Society (BTS) ‐ winter and summer meetings.
-
Primary Care Respiratory Society (PCRS) ‐ UK National Primary Care Conference.
-
Infectious Diseases Society of America (IDSA).
Correspondence
We contacted experts in the field to identify published, non‐published or ongoing studies eligible for inclusion. We also contacted companies that manufacture point‐of‐care biomarkers (Thermo‐Fisher, Hoffmann‐LaRoche, Orion Diagnostica, Axis‐Shield, Hemocue and Siemens Diagnostica).
Reference lists
We checked reference lists of included articles.
Data collection and analysis
Selection of studies
Two authors (RA and J‐U SJ) independently assessed titles and abstracts identified through the searches. We collected and assessed full‐text copies of potentially eligible articles. We resolved disagreements through discussion involving the remaining authors, when necessary.
Data extraction and management
Two authors (RA, J‐U SJ) independently extracted data and information on study design from the included trials and entered the information into a data extraction form. We contacted the authors if outcome data or trial characteristics were not complete. We extracted the following data.
-
Trial characteristics: unit of randomisation; allocation sequence generation; concealment of allocation; blinding; number of participants; number of intervention arms; length of follow‐up.
-
Patient characteristics: baseline characteristics (mean (or median) age; gender; co morbidities); number of patients randomised to each intervention arm; number of patients completing the trial; basis for inclusion in study; types of ARIs and duration; exclusion criteria.
-
Intervention characteristics: type of point‐of‐care biomarker and corresponding specified cut‐off values for guidance of antibiotic prescribing if any.
-
Outcome measures: all available primary and secondary outcome measures specified for this review.
We converted ranking scales on recovery and patient satisfaction to dichotomised outcomes by collapsing response categories when needed. For cluster‐RCTs we extracted intra‐cluster correlation coefficients.
Assessment of risk of bias in included studies
Two authors (RA, J‐U SJ) independently assessed the risk of bias of included studies using the 'Risk of bias' tool in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). This included assessment of sequence generation, allocation concealment, blinding (participants, personnel and outcome assessors), incomplete outcome data and selective outcome reporting bias, as well as other sources of bias. We searched for incomplete outcome data and selective outcome reporting by comparing the methods and results section with the trial protocols when available.
For cluster‐RCTs, we specifically checked for other sources of bias including selection bias, baseline imbalance between clusters, loss of clusters and incorrect analysis (Higgins 2011). We ranked the quality of evidence according to the four‐level Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach (Higgins 2011).
Measures of treatment effect
We reported the treatment effect as a risk ratio (RR) with 95% confidence intervals for each dichotomised outcome. We calculated the risk difference (RD) to estimate the number needed to test (NNT), indicating the NNT with a point‐of‐care test to save one antibiotic prescription. When we could not pool the results we presented them qualitatively.
Unit of analysis issues
The unit of analysis was the individual patient. For cluster‐RCTs we adjusted the unit of analysis by calculating the design effect to modify sample sizes and inflate confidence intervals (CIs) accordingly (Higgins 2011).
Dealing with missing data
We did a worst‐case scenario analysis where we considered missing outcome data as treatment failures in the intervention group and treatment successes in the control group.
Assessment of heterogeneity
We investigated heterogeneity using the I2 statistic with a cut‐off value of 40% to indicate important inconsistencies (Higgins 2011).
Data synthesis
We calculated a weighted estimate for the selected outcomes by means of a random‐effects meta‐analysis, using the Review Manager software (RevMan 2014), when possible.
Subgroup analysis and investigation of heterogeneity
We preplanned the following subgroup analyses.
-
Cluster‐RCTs versus individual RCTs.
-
Type of point‐of‐care test.
-
Trials with low risk of bias versus high risk of bias.
Sensitivity analysis
We planned a sensitivity analysis for our primary outcomes using a fixed‐effect model. However, this was not performed due to the substantial heterogeneity of data.
Results
Description of studies
See: Characteristics of included studies, Characteristics of excluded studies and Characteristics of ongoing studies.
Results of the search
The search flowchart is presented as Figure 1. We found six eligible studies, with a total of 3284 patients recruited from primary care settings. Diagnoses were predominately lower acute respiratory infections (75%) (Table 2). The only point‐of‐care biomarker included in the review was C‐reactive protein. We found no studies that compared different kinds of biomarkers.
Study flow diagram.
| Parameter | Studies | C‐reactive protein group | Control group |
| Age, mean (SD)a | 45.3 (16.8) | 46.0 (17.2) | |
| Gender (female) % (n/N) | All studies | 62.8 (2012/3203) | 64.3 (1916/2980) |
| Current smokers | 44.9 (1187/2639) | 45.0 (1079/2396) | |
| Co‐morbidityb | 21.2 (563/2652) | 19.6 (472/2403) | |
| Primary diagnosis | |||
| Unclassified upper ARIc | 21.5 (499/2325) | 21.1 (446/2118) | |
| Otitis media | 3.3 (13/394) | 4.5 (17/374) | |
| Common cold | 13.9 (15/108) | 16.8 (22/131) | |
| Rhinosinusitis | 27.3 (143/523) | 27.2 (137/502) | |
| Total upper ARId | Andreeva 2013; Cals 2010; Diederichsen 2000; Little 2013a; Melbye 1995 | 22.7 (670/2956) | 22.6 (622/2752) |
| Pneumonia | 7.7 (16/209) | 14.4 (30/209) | |
| LRTI/acute cough | All studies | 74.3 (2364/3183) | 73.5 (2173/2956) |
| Bronchitis | 37.9 (41/108) | 32.1 (42/131) | |
| Exacerbations of COPD or asthma | 14.8 (16/108) | 8.4 (11/131) | |
| Total lower ARIe | All studies | 76.8 (2446/3183) | 70.5 (2271/2956) |
| Influenza | 8.3 (9/108) | 9.2 (12/131) | |
| Other respiratory diseases | 13.3 (67/502) | 13.1 (66/505) |
*Crude numbers provided from all studies regardless of design.
aMelbye 1995 reported the median age: 50 (range 18 to 83) in the C‐reactive protein arm versus 44 (18 to 82) in the control arm.
bChronic obstructive pulmonary disease (COPD); asthma; heart disease; diabetes mellitus.
cAcute respiratory infection.
dAny upper acute respiratory infections.
eAny lower acute respiratory infections.
Included studies
The included studies were conducted between 1995 and 2013 in Europe and Russia. Three trials were cluster‐RCTs (Andreeva 2013; Cals 2009; Little 2013a) and three were individually RCTs (Cals 2010; Diederichsen 2000; Melbye 1995). Inclusion criteria differed among studies. Diederichsen 2000 and to a lesser extent Melbye 1995 used broad inclusion criteria, while the four newest studies used specific diagnostic criteria for lower and/or upper ARIs. We noted appreciable differences between the C‐reactive protein cut‐off values applied to guide antibiotic treatment, ranging from vague indications to specific recommendations for initiating and/or withholding antibiotic treatment (Table 3). Test results were made available to the doctors as part of the initial clinical assessment in the newest four studies, while Melbye 1995 only made results available to doctors after the initial clinical decision. The exact set‐up was left to the participating doctors to accommodate in the Diederichsen 2000 study. The doctor could overrule C‐reactive protein guidance in all trials. Outcome assessment was based on medical records regarding the number of antibiotic prescriptions, while secondary outcomes such as clinical recovery were patient‐reported, using diaries and questionnaires, or follow‐up visits at the clinics (Andreeva 2013; Melbye 1995).
| Study | Randomisation | Inclusion criteria | Algorithm used |
| Individual | Adults (> 18 years) with subjective complaint of i) pneumonia, bronchitis or asthma (no further description) or ii) 1 of the following symptoms: cough, shortness of breath, chest pain on deep inspiration or when coughing | Duration of illness < 24 hours and C‐reactive protein levels lower than 50 mg/L; no change in clinical decision. C‐reactive protein levels > 50 mg/L; immediate antibiotic prescribing was recommended Duration of illness 1 to 6 days and C‐reactive protein levels < 11 mg/L; no antibiotics recommended. Patients with C‐reactive protein levels between 11 and 49 mg/L; no change in clinical decision. C‐reactive protein levels > 50 mg/L; immediate antibiotic prescribing was recommended Duration of illness > 7 days and C‐reactive protein levels < 11 mg/L; no antibiotics recommended. Patients with C‐reactive protein levels between 11 and 24 mg/L; no change in clinical decision. C‐reactive protein levels > 25 mg/L; immediate antibiotic prescribing was recommended | |
| Individual | All patients with a respiratory infection (no further description) | Strict cut‐off values were not given, but information was provided that a normal C‐reactive protein level was < 10 mg/L and that C‐reactive protein levels < 50 mg/L were seldom the result of bacterial infection | |
| Cluster | Adults (> 18 years) with suspected LRTI (cough < 4 weeks AND 1 focal sign/symptom (shortness of breath, wheezing, chest pain, auscultation abnormalities) AND 1 systemic sign/symptom (fever > 38 °C, perspiring, headache, myalgia, feeling generally unwell) | C‐reactive protein levels < 20 mg/L: pneumonia extremely unlikely and antibiotic prescribing discouraged C‐reactive protein levels between 20 to 50 mg/L: pneumonia very unlikely C‐reactive protein levels between 50 to 100 mg/L: clear infection. Acute bronchitis most likely, possible pneumonia C‐reactive protein > 100 mg/L: severe infection. Pneumonia more likely. Immediate antibiotic prescribing was recommended C‐reactive protein levels between 20 and 99 mg/L: consider delayed prescribing | |
| Individual | Adults (> 18 years) with: i) LRTI (cough < 4 weeks) AND 1 focal sign/symptom (shortness of breath, wheezing, chest pain, auscultation abnormalities) AND 1 systemic sign/symptom (fever > 38 °C, perspiring, headache, myalgia, feeling generally unwell) ii) Rhinosinusitis < 4 weeks AND 1 symptom (history of rhinorrhoea, blocked nose) 1 symptom or sign (purulent rhinorrhoea, unilateral facial pain, headache, teeth pain, pain when chewing, maxillary/frontal pain when bending over, worsening of symptoms after initial improvement) | C‐reactive protein levels lower < 20 mg/L: bacterial infection was considered highly unlikely and antibiotic prescribing was discouraged C‐reactive protein levels > 100 mg/L: bacterial infection was considered likely and immediate antibiotic prescribing was recommended C‐reactive protein levels between 20 to 99 mg/L: consider delayed prescribing | |
| Cluster | Adults (> 18 years) with: i) LRTI/acute cough (up to 28 days duration) as the main symptom, or alternatively where cough was not the most prominent symptom (e.g. fever, malaise), but where the clinician considered acute LRTI was the main diagnosis. Pneumonia was not an exclusion criterion ii) URTI: as with LRTI, but judged by the physician to be another acute respiratory infection (sore throat, otitis media, sinusitis, influenza and/or coryzal illness) | C‐reactive protein ≤ 20 mg/L: self limiting ARI, withhold antibiotics C‐reactive protein 21 to 50 mg/L: majority of patients have self limiting ARI, withhold antibiotics, in most cases C‐reactive protein 51 to 99 mg/L: withhold antibiotics in the majority of cases and consider delayed antibiotics in the minority of cases C‐reactive protein ≥100 mg/L: severe infection, prescribe antibiotics | |
| Cluster | Adults (> 18 years) with LRTI/acute cough (including acute bronchitis, pneumonia and infectious exacerbations of COPD or asthma) for less than 28 days | C‐reactive protein < 20 mg/L antibiotics usually not needed C‐reactive protein > 50 mg/L antibiotic prescribing could be indicated taking into account the duration of illness |
All studies stated that physicians could deviate from the algorithm at any time.
ARI: acute respiratory infection
COPD: chronic obstructive pulmonary disease
LRTI: lower respiratory tract infection
URTI: upper respiratory tract infection
Melbye 1995 was terminated by the principal investigator after one year without reaching the target inclusion rate due to an interim analysis that showed no difference between groups and also due to lack of interest from the participating general practitioners.
Diederichsen 2000 was the only study to include all age groups, including 139 children. The remaining studies only included participants older than 18 years.
Two studies received economic funding from manufacturers of C‐reactive protein point‐of‐care tests (Cals 2010; Melbye 1995). Andreeva 2013 received test kits and/or reagents for the study. On‐site training in C‐reactive protein devices was performed by manufacturers in two studies (Diederichsen 2000; Little 2013a).
We successfully contacted a total of four study authors for additional details and in the case of Diederichsen 2000, we obtained raw data to calculate the number of patients with substantial improvement and to differentiate between children and adults.
Excluded studies
We excluded two RCTs using procalcitonin to guide antibiotic use in primary care because the analysis was not performed at the point‐of‐care (Briel 2008; Burkhardt 2010). Two RCTs were not conducted in a primary care setting (Gonzales 2011; Takemura 2005), and two studies used a before‐and‐after design (Kavanagh 2011; Llor 2012). We also excluded Dahler‐Eriksen 1999, as this study did not assess C‐reactive protein to guide antibiotic treatment decisions.
Risk of bias in included studies
We assessed the risk of bias for each study and this is presented graphically in Figure 2 and summarised in Figure 3. For further information on included studies see Characteristics of included studies.
'Risk of bias' graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.
'Risk of bias' summary: review authors' judgements about each risk of bias item for each included study.
Allocation
The cluster‐RCTs used computer randomisation programs to allocate practices to the intervention or control arms (Andreeva 2013; Cals 2009; Little 2013a). Cals 2010 used sequentially numbered, opaque, sealed envelopes prepared in different block sizes by an independent research team. Diederichsen 2000 provided no information on the sequence generation but state they used "pre‐randomised sealed envelopes in blocks of 34". Melbye 1995 did not specify the randomisation procedure but according to the principal investigator this was adequately done at study sponsor level. Allocation concealment of individual patients does not apply to cluster‐RCTs at practice level, so we graded this as 'unclear' risk of bias.
Blinding
This intervention did not lend itself to blinding at clinician level as the intervention was used in management decisions and all clinicians are considered non‐blinded. Assessment of antibiotic use was based on electronic or paper records. Clinical recovery was based on diaries and questionnaires completed by the patient and did not involve study personnel in the majority of studies. However, Melbye 1995 and Andreeva 2013 assessed clinical recovery non‐blinded at a follow‐up visit.
Incomplete outcome data
We successfully retrieved incomplete reported outcome data on the use of antibiotic prescriptions by contacting the individual study authors when needed. Data on clinical recovery rates ranged from 90% to 98% in completeness between studies. Information necessary for subgroup analyses of serious versus non‐serious infections could not be obtained as this was not reported and exact diagnoses not recorded. However, we were able to obtain data on the effect of C‐reactive protein on antibiotic use for lower versus upper ARIs.
Selective reporting
We did not suspect selective reporting but only newer studies had a published protocol. All outcomes were reported as intention‐to‐treat.
Other potential sources of bias
Selection (recruitment) bias is a risk in cluster‐RCTs as care providers assigned to the intervention group can select which patients to test (inclusion was at the discretion of the care provider). This means that patients with a higher than average likelihood that the test might change the clinical decision could preferentially be enrolled, e.g. those patients that the care provider perceived could be convinced that an intervention was not needed if a test was performed. This may exaggerate the estimated effect relative to more widespread use in clinical practice. However, measures to limit this 'active' recruitment by participating doctors were in place, e.g. by requirements for consecutive enrolment of the first eligible patients that presented in each practice. In the study by Cals 2009, significantly more patients in the control group had abnormalities on auscultation (60.3% versus 46.7%, P value = 0.005), a parameter closely linked to antibiotic prescription (Jakobsen 2010). However, in the larger study by Little 2013a symptom severity scores were balanced between groups.
Contamination bias is possible in individual RCTs as the general practitioner may gradually learn to foresee which patients have low C‐reactive protein levels and apply this acquired skill in the control group. As most patients will have low values of C‐reactive protein this would lead to decreased antibiotic prescription in the control group and underestimate the effect of the test.
Inclusion bias may occur in both trial designs as general practitioners may be reluctant to include patients with severe disease given the risk that antibiotic treatment is not recommended according to the test result. In individual RCTs, this potential bias would be non‐discriminative as opposed to the cluster‐RCTs, where this could be a discriminative bias. This may lead to a lower estimate of the effect of biomarkers in individual RCTs (a priori risk of antibiotic treatment is low in both groups) but may overestimate the effect in cluster‐RCTs (a priori risk of antibiotic treatment is different between intervention (low) and control groups (normal)).
Effects of interventions
See: Summary of findings for the main comparison
Primary outcomes
1. Number of patients given an antibiotic prescription at the index consultation and at 28 days follow‐up
See summary of findings Table for the main comparison.
All six studies, including 3284 patients (mean age 46, standard deviation (SD) 17), reported point estimates in favour of the C‐reactive protein test to reduce the number of antibiotic prescriptions in acute respiratory infections. The pooled result for all included trials showed a statistically significant effect of C‐reactive protein testing on the number of antibiotic prescriptions issued in primary care settings for acute respiratory infections (ARIs) (risk ratio (RR) 0.78, 95% confidence interval (CI) 0.66 to 0.92; I2 statistic = 68%) (Analysis 1.1; Figure 4), but with considerable heterogeneity. The heterogeneity disappeared in our pre‐planned subgroup analysis of cluster‐randomised controlled trials (RCTs) versus individual RCTs, suggesting that it may not be meaningful to pool all trials.
Forest plot of comparison: 1 C‐reactive protein ‐ antibiotic prescribing: all trials, outcome: 1.1 C‐reactive protein ‐ antibiotics prescribed at index consultation. All trials (cluster‐RCTs modified sample size):.
The individual RCTs (N = 1309) indicated a reduction in antibiotic use (RR 0.90, 95% CI 0.80 to 1.02; I2 statistic = 5%) (Analysis 1.1.1; Figure 4), although the result was not statistically significant, while the cluster‐RCTs (N = 1975) showed a more pronounced effect (RR 0.68, 95% CI 0.61 to 0.75) (Analysis 1.1.2; Figure 4).
We calculated the number needed to test (NNT) to save one antibiotic prescription at the index consultation as 20 for individual RCTs and six for cluster‐RCTs (Table 4).
| NNT | 95% CI | |
| All trials | 9 | 6 to 20 |
| Individually RCT | 20 | ‐100 to 9 |
| Cluster‐RCT | 6 | 5 to 8 |
Cluster‐randomised trials with modified sample size
CI: confidence interval
NNT: number needed to test
RCT: randomised controlled trial
The effect found at index consultation on the reduction in antibiotic use was maintained at day 28 and no evidence was found that patients in the C‐reactive protein group needed additional antibiotic treatment between the index consultation and 28 days of follow‐up compared to standard of care (Analysis 1.2; Figure 5).
Forest plot of comparison: 1 C‐reactive protein ‐ antibiotic prescribing: all trials, outcome: 1.2 C‐reactive protein ‐ antibiotics prescribed within 28 days (cluster‐RCT with modified sample size).
2. Number of patients with substantial improvement (including full recovery) at day seven
We found no differences in clinical recovery (defined as at least substantial improvement) at day seven between groups (Analysis 2.1; Figure 6).
Forest plot of comparison: 4 C‐reactive protein ‐ Patient recovery day 7: Individually randomised trials, outcome: 2.1 Substantial improvement day 7.
3. Total mortality at 28 days follow‐up
No deaths or serious complications were reported in any of the studies.
Secondary outcomes
1. Number of patients in need of a re‐consultation at 28 days follow‐up
There were no significant differences in re‐consultation rates (Analysis 3.1).
2. Number of patients in need of a hospital admission at 28 days follow‐up
Five of the six studies reported that there had been no hospitalisations in the follow‐up period. Little 2013a reported a total of 30 hospitalisations in 4264 patients, 22 in the C‐reactive protein group versus eight in the control group (crude RR 2.53, 95% CI 1.13 to 5.66). However, when adjusting for the design effect of cluster‐RCTs by modifying sample sizes, this difference ceased to be statistically significant (RR 2.45, 95% CI 0.65 to 9.19). Information on hospital admissions was obtained through a medical history review in 15 cases. The reasons were cardiac (two); respiratory (eight), generally unwell/fever (two); gastrointestinal symptoms (two); sinusitis (one). All hospitalisations may not have been directly related to the intervention. However, an increase in the risk of hospitalisation in the C‐reactive protein group cannot be ruled out, although the absolute event rate is low. Data were not available to determine the number of hospitalised patients who were initially withheld from receiving antibiotic treatment, nor the C‐reactive protein level at index consultation.
3. Duration of the acute respiratory infection (e.g. mean or median days with restrictions in daily activities due to the infection)
Three studies reported on this outcome but a pooled analysis could not be performed due to differences in assessing duration of symptoms (Table 5). Cals 2009 reported no differences in the median symptom duration to full recovery, while Cals 2010 also provided this measure as a mean number of days. Little 2013a reported the time to resolution of symptoms rated moderately bad or worse. No differences were observed in any of these patient‐reported measures.
| Study | Mean (SD) | Median (IQR) | |||
| C‐reactive protein | Control | C‐reactive protein | Control | ||
| Cals 2009a | ‐ | ‐ | 22 (14 to 28) | 22 (14 to 28) | |
| Cals 2010a | LRTI | 17.5 (9.2) | 19.8 (9.5) | 15.5 (9.5 to 28) | 20 (13.5 to > 28) |
| Rhinosinusitis | 17.3 (9.3) | 16.6 (9.9) | 14 (10 to 28) | 14 (7 to > 28) | |
| Little 2013b | LRTI | ‐ | ‐ | 6 (3 to 9) | 5 (3 to 9) |
| URTI | ‐ | ‐ | 5 (3 to 7) | 4 (3 to 8) | |
| ARI | ‐ | ‐ | 5 (3 to 9) | 5 (3 to 9) | |
aReported as time to full recovery.
bReported as resolution of moderately bad or worse symptoms.
ARI: acute respiratory tract infection (LRTI + URTI)
IQR: interquartile range
LRTI: lower respiratory tract infection
SD: standard deviation
URTI: upper respiratory tract infection
4. Number of satisfied patients
We detected no differences. However, the substantial heterogeneity (I2 statistic = 45%) detected and the fact that only two studies reported on this outcome does not allow us to draw clear conclusions (Analysis 4.1).
5. Number of patients with substantial improvement (including full recovery) at 28 days of follow‐up
We found no differences in clinical recovery (defined as at least substantial improvement) at day 28 between groups (Analysis 5.1).
Sensitivity and subgroup analyses
As substantial heterogeneity related to trial designs was present, we omitted the pre‐planned sensitivity analysis using a fixed‐effect meta‐analysis.
Trials with low versus high risk of bias: as the intervention did not lend itself to blinding, we chose to omit this component when selecting studies with a low risk of bias. Accordingly, Cals 2010 was the only trial with a low risk of bias (Figure 3). The result of this trial was RR 0.77, 95% CI 0.60 to 0.98.
Only one study, Diederichsen 2000, reported specifically on the effect in children (N = 139) and found no significant effect (RR 1.09, 95% CI 0.70 to 1.71) (Analysis 6.1). Individual disease labelling in severe versus less severe diseases as planned in the protocol was not possible due to lack of data. However, the effect of the C‐reactive protein test on antibiotic prescriptions was similar in upper and lower ARIs (Analysis 7.1).
To assess the substantial heterogeneity and the subgroup differences detected, we performed a post hoc analysis of the newer studies (Andreeva 2013; Cals 2009; Cals 2010; Little 2013a), with specific guidance on antibiotic prescription if C‐reactive protein levels were < 20 mg/L. This analysis showed a significant reduction in antibiotic use (RR 0.69, 95% CI 0.62 to 0.76; I2 statistic = 0%) (Analysis 8.1).
Sensitivity analyses assuming a worst‐case scenario, where all patients in the C‐reactive protein group lost to follow‐up did not improve and all patients in the control group lost to follow‐up did substantially improve, also showed no significant differences (day seven; RR 1.10, 95% CI 0.99 to 1.21; I2 statistic = 0% and day 28; RR 1.11, 95% CI 0.84 to 1.48; I2 statistic = 0%) (Analysis 9.1; Analysis 9.2).
A summary of the secondary outcomes is presented in Table 6.
| Outcome | Studies | Patients | Pooled results* RR (95% CI); I2 | Individually randomised RR (95% CI); I2 | Cluster‐randomised RR (95% CI); I2 | Analysis |
| Antibiotic use day 28 | 4 | 715 | 0.80 (0.67 to 0.96); 40% | 0.87 (0.75 to 1.02); 7% | 0.68 (0.51 to 0.91); 19% | 1.2 |
| Recovery day 28a | 3 | 527 | 0.94 (0.69 to 1.28); 0% | ― | ― | 5.1 to 5.3 |
| Hospital admissions | 6b | 1764 | ― | ― | 2.45 (0.65 to 9.19) | ― |
| Re‐consultations | 4 | 2228 | 1.08 (0.93 to 1.27); 0% | ― | ― | 3.1 |
| Patient satisfaction | 2 | 674 | 0.79 (0.57 to 1.08); 45% | ― | ― | 4.1 |
*When I2 > 40%, separate analyses of individually and cluster‐randomised trials are presented.
aDefined as at least substantial improvement.
bLittle 2013a was the only trial with any cases of hospital admission. The calculation is done for this trial alone.
Discussion
Summary of main results
We identified and analysed six randomised trials with 3284 participants. Our results indicate that C‐reactive protein point‐of‐care tests to guide antibiotic prescription in lower as well as upper acute respiratory infections (ARIs) in general practice can reduce antibiotic use and it is unlikely that the intervention increases morbidity. No studies reported that deaths had occurred in either the intervention or control groups.
A precise estimate of the reduction in antibiotic use was not obtained due to substantial heterogeneity between trials that were likely related to differences in design. Individual RCTs showed a statistically non‐significant relative reduction of antibiotic prescriptions (risk ratio (RR) 0.90, 95% confidence interval (CI) 0.80 to 1.02; I2 statistic = 5%), while cluster‐RCTs at the general practice level reported a statistically significant reduction (RR 0.68, 95% CI 0.61 to 0.75; I2 statistic = 0%). We note that individual point estimates from all trials indicated a reduction in antibiotic use.
In this context and despite being prone to bias, the cluster‐RCT method may be considered a more pragmatic design that more closely reflects everyday practice, where C‐reactive protein testing is either available or not in a given general practice.
The observed heterogeneity may also in part be explained by the different inclusion criteria and C‐reactive protein algorithms applied, where restrictive recommendations on antibiotic use generally showed a more pronounced effect (Table 3). Newer studies (published in the last five years) provide guidance on when to withhold or initiate antibiotic treatment using specific cut‐off values.
No differences were found regarding patient‐reported outcomes. No deaths or serious adverse events were reported, but one trial reported an increase in hospitalisations in the C‐reactive protein group (Little 2013a). However, the absolute numbers of events were low (22 versus 8 events in 4264 patients) and the finding was non‐significant when adjusting for the design effect. Nevertheless, this suggests that the suspected benefits of reducing antibiotic use in ARIs (de‐medicalisation, containing development of antibiotic resistance, costs and fewer side effects) must be balanced against the potential safety concerns of a small increased risk of hospitalisation. C‐reactive protein may be an adjunct to the physical examination but cannot replace clinical skills and appropriate safety‐netting must be applied.
Overall completeness and applicability of evidence
The included trials were mainly from European countries with considerable differences in antibiotic use and organisation of primary care. The studies had high levels of completeness for both primary (100%) and secondary outcomes (90% to 98%) and all results were reported as intention‐to‐treat. All studies provided a measure of clinical recovery and four studies used very similar case report forms and C‐reactive protein algorithms (Andreeva 2013; Cals 2009; Cals 2010; Little 2013a). The algorithms obviously affect both patient safety and the potential reduction in antibiotic use, as does the a priori likelihood of antibiotic use in any given patient population. By including different algorithms in this review, we regard the findings as a 'proof of concept', but identification of an optimal algorithm was not possible.
This review encompassed different respiratory infections with varying anatomical localisation, but C‐reactive point‐of‐care testing was associated with a similar reduction in antibiotic use for both upper and lower ARIs.
Intra‐cluster coefficients to inflate the confidence intervals of cluster‐RCTs were provided (Cals 2009; Little 2013a), to allow inclusion in our meta‐analysis. We had pre‐specified a random‐effects model to account for an expected moderate heterogeneity among included trials regarding study design (cluster‐RCTs versus individual RCTs), and differences in the C‐reactive protein algorithms and inclusion criteria (Table 3).
Only the two oldest trials did not show a significant effect on antibiotic prescriptions (Diederichsen 2000; Melbye 1995), which could partly be because the use of antibiotics in primary care in Europe has increased overall during the last decade (Adriaenssens 2011). If this increase in antibiotic use mainly reflects excessive prescriptions, the net effect of C‐reactive protein guidance may have increased.
Many patients express worries about their symptoms and seek medical re‐assurance of the benign course of their illness. However, general practitioners are often faced with varying degrees of uncertainty in their management decision and a point‐of‐care test to rule out serious infection may increase confidence and acceptance of the decision not to use antibiotic treatment (Stanton 2010). E‐learning or short seminars can be used to achieve the necessary skills for interpretation of the test results in clinical practice (Cals 2009; Yardley 2013).
The results of this review should not be generalised to include children or patients with severe co‐morbidities and/or immunocompromised patients.
Quality of the evidence
The available information was from trials with a low to moderate risk of bias that address the main objectives of this review. Included studies provided data on antibiotic use at index consultation and reported at least one measure of patient safety or recovery. One of the primary outcomes (antibiotic use) was directly observed and not assessor‐dependent. This intervention did not lend itself to blinding of the provider as its purpose was to influence clinical decisions, but studies were otherwise well reported and appeared to be of moderate to good quality. The studies included patients relevant to a European primary care context.
The relatively small number of individual RCTs, adding a total of 40% of the cases, is of some concern. However, we accounted for the included cluster‐RCTs by inflating confidence intervals accordingly and assessing the increased risk of bias from selection and allocation concealment. We preplanned subgroups of analysis based on study design.
However, due to the considerable heterogeneity in the pooled analysis of all trials we have presented subgroup results. The observed heterogeneity may well be explained by differences in study design (individual RCTs versus cluster‐RCTs) and the different C‐reactive protein algorithms applied (Table 3). Our decision to downgrade the quality of the evidence was primarily driven by imprecision of the estimated effect of the pooled analysis on antibiotic prescribing (summary of findings Table for the main comparison).
Potential biases in the review process
To the best of our knowledge, no bias was introduced in the review process.
Agreements and disagreements with other studies or reviews
To our knowledge only two other studies have systematically reviewed the evidence for C‐reactive protein point‐of‐care tests to guide antibiotic prescription in primary care. Engel 2011 concluded that current evidence did not support the use of C‐reactive protein in primary care for this purpose. However, no meta‐analyses were performed without a stated reason. Huang 2013, on the other hand, reported a reduction in antibiotic use for ARIs (RR 0.75, 95% CI 0.67 to 0.83) but with considerable heterogeneity (I2 statistic = 76%). However, the main meta‐analysis included both RCTs as well as observational studies. Also, both reviews did not include the two latest trials (Andreeva 2013; Little 2013a), adding a total of 1851 patients to the analysis with a weight of 9.2% and 24.6%, respectively.
Studies have reported that the C‐reactive protein test may not be sufficiently sensitive and specific to be of diagnostic value in primary care where the incidence of serious bacterial infection is low (Falk 2009; van der Meer 2005), but it forms part of a number of prediction rules for pneumonia (Steurer 2011; van Vugt 2011), and its use is advocated in the most recent European guidelines on the management of lower ARIs (Woodhead 2011). Of note, C‐reactive protein is no perfect test and a risk exists for over‐ as well as under‐treatment with antibiotics. This highlights the importance of limiting the use of this tool to a correct clinical context: a doctor stating that the symptoms presented are caused by an acute respiratory tract infection and uncertainty exists regarding the potential benefit of antibiotic therapy.
A Cochrane review suggests that the biomarker procalcitonin (currently unavailable as point‐of‐care test for primary care) could be a safe and effective tool to guide decisions about antibiotic treatment of ARIs (Schuetz 2012). The results of this review are in line with these recommendations.
Of note, studies comparing communication training to C‐reactive protein point‐of‐care testing showed similar potential to reduce antibiotic use, while an additive effect was observed when both C‐reactive protein tests and training sessions in communication skills were combined (Cals 2009; Little 2013a).
'Risk of bias' graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.
'Risk of bias' summary: review authors' judgements about each risk of bias item for each included study.
Forest plot of comparison: 1 C‐reactive protein ‐ antibiotic prescribing: all trials, outcome: 1.1 C‐reactive protein ‐ antibiotics prescribed at index consultation. All trials (cluster‐RCTs modified sample size):.
Forest plot of comparison: 1 C‐reactive protein ‐ antibiotic prescribing: all trials, outcome: 1.2 C‐reactive protein ‐ antibiotics prescribed within 28 days (cluster‐RCT with modified sample size).
Forest plot of comparison: 4 C‐reactive protein ‐ Patient recovery day 7: Individually randomised trials, outcome: 2.1 Substantial improvement day 7.
Comparison 1 CRP ‐ Antibiotic prescribing: all trials, Outcome 1 CRP ‐ Antibiotics prescribed at index consultation. All trials (cluster‐randomised with modified sample size).
Comparison 1 CRP ‐ Antibiotic prescribing: all trials, Outcome 2 CRP ‐ Antibiotics prescribed within 28 days (cluster‐randomised trials with modified sample size).
Comparison 2 CRP ‐ No. of patients substantially improved day 7: individually randomised trials, Outcome 1 Clinical recovery day 7.
Comparison 3 CRP ‐ Number of re‐consultations within 28 days, Outcome 1 CRP ‐ Number of re‐consultations at follow‐up within 28 days.
Comparison 4 CRP ‐ Patient satisfaction, Outcome 1 CRP ‐ Patient satisfaction.
Comparison 5 CRP ‐ No. of patients substantially improved at follow‐up within 28 days, Outcome 1 Clinical recovery day 28 (cluster‐randomised trials with modified sample size; ICC0.06).
Comparison 5 CRP ‐ No. of patients substantially improved at follow‐up within 28 days, Outcome 2 Clinical recovery day 28: sensitivity analysis (ICC 0.01).
Comparison 5 CRP ‐ No. of patients substantially improved at follow‐up within 28 days, Outcome 3 Clinical recovery day 28: sensitivity analysis (ICC 0.12).
Comparison 6 CRP ‐ Subgroup analysis: Children versus adults. Antibiotic prescribing at index consultation, Outcome 1 Children.
Comparison 6 CRP ‐ Subgroup analysis: Children versus adults. Antibiotic prescribing at index consultation, Outcome 2 Adults (cluster‐randomised trials with modified sample size).
Comparison 7 CRP ‐ Subgroup analysis: upper respiratory tract infections versus lower respiratory tract infections, Outcome 1 Antibiotics prescribed at index consultation: Cluster‐randomised with modified sample size.
Comparison 8 CRP ‐ Algorithms with specific cut‐offs to rule out serious disease (< 20 mg/L) (sensitivity analysis), Outcome 1 CRP ‐ Antibiotic prescribing when algorithms provide clear cut‐offs to rule out (< 20 mg/L).
Comparison 8 CRP ‐ Algorithms with specific cut‐offs to rule out serious disease (< 20 mg/L) (sensitivity analysis), Outcome 2 CRP ‐ Recovery at day 7.
Comparison 8 CRP ‐ Algorithms with specific cut‐offs to rule out serious disease (< 20 mg/L) (sensitivity analysis), Outcome 3 CRP ‐ Recovery at follow‐up (max 28 days).
Comparison 9 CRP ‐ Sensitivity analysis: missing data. Patient recovery (worst case), Outcome 1 CRP ‐ Patient recovery day 7: missing data in CRP = not recovered.
Comparison 9 CRP ‐ Sensitivity analysis: missing data. Patient recovery (worst case), Outcome 2 Patient recovery day 28: missing data in CRP = not recovered. Cluster‐randomised trials with modified sample size.
| Point‐of‐care biomarker for infection compared with standard of care for guiding antibiotic therapy in acute respiratory infections | ||||||
| Patient or population: patients with acute respiratory infections Settings: primary care Intervention: point‐of‐care biomarker (C‐reactive protein) test Comparison: standard care | ||||||
| Outcomes | Illustrative comparative risks* (95% CI) | Effect | No. of participants | Quality of the evidence | Comments | |
| Assumed risk | Corresponding risk | |||||
| Standard care | C‐reactive protein | |||||
| Mortality (C‐reactive protein) Follow‐up: 28 days | ― | ― | ― | 3284 (6) | ⊕⊕⊕⊕ | No participants died in these studies |
| No. of antibiotic prescriptions (C‐reactive protein) Index consultation | Individual RCTs: study population | RR 0.90 (0.80 to 1.02) | 1309 | ⊕⊕⊕⊝ | I2 statistic = 5% | |
| 519 per 1000 | 467 per 1000 | |||||
| Cluster‐RCTs: study population | RR 0.68 (0.61 to 0.75) | 1975 (3) | I2 statistic = 0% | |||
| 525 per 1000 | 357 per 1000 | |||||
| No. of antibiotic prescriptions (C‐reactive protein) Follow‐up: 28 days | Individual RCTs: study population | RR 0.87 (0.75 to 1.02) | 497 | ⊕⊕⊕⊝ | I2 statistic = 7% | |
| 623 per 1000 | 542 per 1000 | |||||
| Cluster‐RCTs: study population | RR 0.68 (0.51 to 0.91) | 211 (2) | I2 statistic = 19% | |||
| 629 per 1000 | 428 per 1000 | |||||
| Clinical recovery. No. of participants with at least 'substantial improvement' Follow‐up: 7 days | Individual RCTs: study population | RR 1.03 (0.93 to 1.14) | 1264 | ⊕⊕⊕⊝ | I2 statistic = 0% | |
| 414 per 1000 | 426 per 1000 | |||||
| Clinical recovery Follow‐up: 28 days | Individual and cluster‐RCTs: study population | RR 0.94 (0.69 to 1.28) | 527 | ⊕⊕⊕⊝ | I2 statistic = 0% | |
| 758 per 1000 | 713 per 1000 | |||||
| *The assumed risk was calculated as the median control group risk across studies. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). | ||||||
| GRADE Working Group grades of evidence | ||||||
| 1We downgraded the GRADE judgement to moderate as the heterogeneity, albeit well explained, generates imprecision in the main effect estimate. | ||||||
| Biomarker | Status | Handling | Biochemistry |
| C‐reactive protein (CRP) | POC* test available | Droplet blood from finger prick. Results in approximately 3 minutes. Uninfected adult controls have levels < 10 mg/L | Inflammatory cytokines trigger C‐reactive protein release by the liver. Levels of C‐reactive protein increase within 6 to 18 hours, peaking at 48 to 72 hours |
| Leukocyte count | POC test available | Droplet blood from finger prick. Results in approximately 3 minutes. Uninfected adult controls have leukocyte levels < 9 x 109/L and neutrocyte levels < 7 x 109/L | Cells of the immune system activated by inflammatory cytokines and foreign antigens |
| Procalcitonin (PCT) | POC test not available** | Uninfected adult controls have levels < 0.05 nanogram/mL | Inflammatory cytokines and bacterial endotoxins trigger release of PCT from parenchymal tissues. Levels of PCT increase within 2 to 6 hours, peaking at 24 to 48 hours |
| *POC: point‐of‐care **No POC test in desired target range (0.05 to 0.50 nanogram/mL) | |||
| Parameter | Studies | C‐reactive protein group | Control group |
| Age, mean (SD)a | 45.3 (16.8) | 46.0 (17.2) | |
| Gender (female) % (n/N) | All studies | 62.8 (2012/3203) | 64.3 (1916/2980) |
| Current smokers | 44.9 (1187/2639) | 45.0 (1079/2396) | |
| Co‐morbidityb | 21.2 (563/2652) | 19.6 (472/2403) | |
| Primary diagnosis | |||
| Unclassified upper ARIc | 21.5 (499/2325) | 21.1 (446/2118) | |
| Otitis media | 3.3 (13/394) | 4.5 (17/374) | |
| Common cold | 13.9 (15/108) | 16.8 (22/131) | |
| Rhinosinusitis | 27.3 (143/523) | 27.2 (137/502) | |
| Total upper ARId | Andreeva 2013; Cals 2010; Diederichsen 2000; Little 2013a; Melbye 1995 | 22.7 (670/2956) | 22.6 (622/2752) |
| Pneumonia | 7.7 (16/209) | 14.4 (30/209) | |
| LRTI/acute cough | All studies | 74.3 (2364/3183) | 73.5 (2173/2956) |
| Bronchitis | 37.9 (41/108) | 32.1 (42/131) | |
| Exacerbations of COPD or asthma | 14.8 (16/108) | 8.4 (11/131) | |
| Total lower ARIe | All studies | 76.8 (2446/3183) | 70.5 (2271/2956) |
| Influenza | 8.3 (9/108) | 9.2 (12/131) | |
| Other respiratory diseases | 13.3 (67/502) | 13.1 (66/505) | |
| *Crude numbers provided from all studies regardless of design. aMelbye 1995 reported the median age: 50 (range 18 to 83) in the C‐reactive protein arm versus 44 (18 to 82) in the control arm. bChronic obstructive pulmonary disease (COPD); asthma; heart disease; diabetes mellitus. cAcute respiratory infection. dAny upper acute respiratory infections. eAny lower acute respiratory infections. | |||
| Study | Randomisation | Inclusion criteria | Algorithm used |
| Individual | Adults (> 18 years) with subjective complaint of i) pneumonia, bronchitis or asthma (no further description) or ii) 1 of the following symptoms: cough, shortness of breath, chest pain on deep inspiration or when coughing | Duration of illness < 24 hours and C‐reactive protein levels lower than 50 mg/L; no change in clinical decision. C‐reactive protein levels > 50 mg/L; immediate antibiotic prescribing was recommended Duration of illness 1 to 6 days and C‐reactive protein levels < 11 mg/L; no antibiotics recommended. Patients with C‐reactive protein levels between 11 and 49 mg/L; no change in clinical decision. C‐reactive protein levels > 50 mg/L; immediate antibiotic prescribing was recommended Duration of illness > 7 days and C‐reactive protein levels < 11 mg/L; no antibiotics recommended. Patients with C‐reactive protein levels between 11 and 24 mg/L; no change in clinical decision. C‐reactive protein levels > 25 mg/L; immediate antibiotic prescribing was recommended | |
| Individual | All patients with a respiratory infection (no further description) | Strict cut‐off values were not given, but information was provided that a normal C‐reactive protein level was < 10 mg/L and that C‐reactive protein levels < 50 mg/L were seldom the result of bacterial infection | |
| Cluster | Adults (> 18 years) with suspected LRTI (cough < 4 weeks AND 1 focal sign/symptom (shortness of breath, wheezing, chest pain, auscultation abnormalities) AND 1 systemic sign/symptom (fever > 38 °C, perspiring, headache, myalgia, feeling generally unwell) | C‐reactive protein levels < 20 mg/L: pneumonia extremely unlikely and antibiotic prescribing discouraged C‐reactive protein levels between 20 to 50 mg/L: pneumonia very unlikely C‐reactive protein levels between 50 to 100 mg/L: clear infection. Acute bronchitis most likely, possible pneumonia C‐reactive protein > 100 mg/L: severe infection. Pneumonia more likely. Immediate antibiotic prescribing was recommended C‐reactive protein levels between 20 and 99 mg/L: consider delayed prescribing | |
| Individual | Adults (> 18 years) with: i) LRTI (cough < 4 weeks) AND 1 focal sign/symptom (shortness of breath, wheezing, chest pain, auscultation abnormalities) AND 1 systemic sign/symptom (fever > 38 °C, perspiring, headache, myalgia, feeling generally unwell) ii) Rhinosinusitis < 4 weeks AND 1 symptom (history of rhinorrhoea, blocked nose) 1 symptom or sign (purulent rhinorrhoea, unilateral facial pain, headache, teeth pain, pain when chewing, maxillary/frontal pain when bending over, worsening of symptoms after initial improvement) | C‐reactive protein levels lower < 20 mg/L: bacterial infection was considered highly unlikely and antibiotic prescribing was discouraged C‐reactive protein levels > 100 mg/L: bacterial infection was considered likely and immediate antibiotic prescribing was recommended C‐reactive protein levels between 20 to 99 mg/L: consider delayed prescribing | |
| Cluster | Adults (> 18 years) with: i) LRTI/acute cough (up to 28 days duration) as the main symptom, or alternatively where cough was not the most prominent symptom (e.g. fever, malaise), but where the clinician considered acute LRTI was the main diagnosis. Pneumonia was not an exclusion criterion ii) URTI: as with LRTI, but judged by the physician to be another acute respiratory infection (sore throat, otitis media, sinusitis, influenza and/or coryzal illness) | C‐reactive protein ≤ 20 mg/L: self limiting ARI, withhold antibiotics C‐reactive protein 21 to 50 mg/L: majority of patients have self limiting ARI, withhold antibiotics, in most cases C‐reactive protein 51 to 99 mg/L: withhold antibiotics in the majority of cases and consider delayed antibiotics in the minority of cases C‐reactive protein ≥100 mg/L: severe infection, prescribe antibiotics | |
| Cluster | Adults (> 18 years) with LRTI/acute cough (including acute bronchitis, pneumonia and infectious exacerbations of COPD or asthma) for less than 28 days | C‐reactive protein < 20 mg/L antibiotics usually not needed C‐reactive protein > 50 mg/L antibiotic prescribing could be indicated taking into account the duration of illness | |
| All studies stated that physicians could deviate from the algorithm at any time. | |||
| NNT | 95% CI | |
| All trials | 9 | 6 to 20 |
| Individually RCT | 20 | ‐100 to 9 |
| Cluster‐RCT | 6 | 5 to 8 |
| Cluster‐randomised trials with modified sample size | ||
| Study | Mean (SD) | Median (IQR) | |||
| C‐reactive protein | Control | C‐reactive protein | Control | ||
| Cals 2009a | ‐ | ‐ | 22 (14 to 28) | 22 (14 to 28) | |
| Cals 2010a | LRTI | 17.5 (9.2) | 19.8 (9.5) | 15.5 (9.5 to 28) | 20 (13.5 to > 28) |
| Rhinosinusitis | 17.3 (9.3) | 16.6 (9.9) | 14 (10 to 28) | 14 (7 to > 28) | |
| Little 2013b | LRTI | ‐ | ‐ | 6 (3 to 9) | 5 (3 to 9) |
| URTI | ‐ | ‐ | 5 (3 to 7) | 4 (3 to 8) | |
| ARI | ‐ | ‐ | 5 (3 to 9) | 5 (3 to 9) | |
| aReported as time to full recovery. | |||||
| Outcome | Studies | Patients | Pooled results* RR (95% CI); I2 | Individually randomised RR (95% CI); I2 | Cluster‐randomised RR (95% CI); I2 | Analysis |
| Antibiotic use day 28 | 4 | 715 | 0.80 (0.67 to 0.96); 40% | 0.87 (0.75 to 1.02); 7% | 0.68 (0.51 to 0.91); 19% | 1.2 |
| Recovery day 28a | 3 | 527 | 0.94 (0.69 to 1.28); 0% | ― | ― | 5.1 to 5.3 |
| Hospital admissions | 6b | 1764 | ― | ― | 2.45 (0.65 to 9.19) | ― |
| Re‐consultations | 4 | 2228 | 1.08 (0.93 to 1.27); 0% | ― | ― | 3.1 |
| Patient satisfaction | 2 | 674 | 0.79 (0.57 to 1.08); 45% | ― | ― | 4.1 |
| *When I2 > 40%, separate analyses of individually and cluster‐randomised trials are presented. | ||||||
| Outcome or subgroup title | No. of studies | No. of participants | Statistical method | Effect size |
| 1 CRP ‐ Antibiotics prescribed at index consultation. All trials (cluster‐randomised with modified sample size) Show forest plot | 6 | 3284 | Risk Ratio (M‐H, Random, 95% CI) | 0.78 [0.66, 0.92] |
| 1.1 Individually randomised trials | 3 | 1309 | Risk Ratio (M‐H, Random, 95% CI) | 0.90 [0.80, 1.02] |
| 1.2 Cluster‐randomised trials (modified sample size) | 3 | 1975 | Risk Ratio (M‐H, Random, 95% CI) | 0.68 [0.61, 0.75] |
| 2 CRP ‐ Antibiotics prescribed within 28 days (cluster‐randomised trials with modified sample size) Show forest plot | 4 | 708 | Risk Ratio (M‐H, Random, 95% CI) | 0.80 [0.67, 0.96] |
| 2.1 Individually randomised trials | 2 | 497 | Risk Ratio (M‐H, Random, 95% CI) | 0.87 [0.75, 1.02] |
| 2.2 Cluster‐randomised trials (modified sample size) | 2 | 211 | Risk Ratio (M‐H, Random, 95% CI) | 0.68 [0.51, 0.91] |
| Outcome or subgroup title | No. of studies | No. of participants | Statistical method | Effect size |
| 1 Clinical recovery day 7 Show forest plot | 3 | 1264 | Risk Ratio (M‐H, Random, 95% CI) | 1.03 [0.93, 1.14] |
| Outcome or subgroup title | No. of studies | No. of participants | Statistical method | Effect size |
| 1 CRP ‐ Number of re‐consultations at follow‐up within 28 days Show forest plot | 4 | 2486 | Risk Ratio (M‐H, Random, 95% CI) | 1.08 [0.93, 1.27] |
| 1.1 Individually randomised trials | 1 | 258 | Risk Ratio (M‐H, Random, 95% CI) | 1.43 [0.89, 2.30] |
| 1.2 Cluster‐randomised trials (modified sample size) | 3 | 2228 | Risk Ratio (M‐H, Random, 95% CI) | 1.05 [0.89, 1.24] |
| Outcome or subgroup title | No. of studies | No. of participants | Statistical method | Effect size |
| 1 CRP ‐ Patient satisfaction Show forest plot | 2 | 674 | Risk Ratio (M‐H, Random, 95% CI) | 0.79 [0.57, 1.08] |
| Outcome or subgroup title | No. of studies | No. of participants | Statistical method | Effect size |
| 1 Clinical recovery day 28 (cluster‐randomised trials with modified sample size; ICC0.06) Show forest plot | 3 | 527 | Risk Ratio (M‐H, Random, 95% CI) | 0.94 [0.69, 1.28] |
| 2 Clinical recovery day 28: sensitivity analysis (ICC 0.01) Show forest plot | 3 | 739 | Risk Ratio (M‐H, Random, 95% CI) | 0.96 [0.73, 1.25] |
| 3 Clinical recovery day 28: sensitivity analysis (ICC 0.12) Show forest plot | 3 | 429 | Risk Ratio (M‐H, Random, 95% CI) | 0.91 [0.65, 1.27] |
| Outcome or subgroup title | No. of studies | No. of participants | Statistical method | Effect size |
| 1 Children Show forest plot | 1 | 139 | Risk Ratio (M‐H, Fixed, 95% CI) | 1.09 [0.70, 1.71] |
| 2 Adults (cluster‐randomised trials with modified sample size) Show forest plot | 6 | 3145 | Risk Ratio (M‐H, Random, 95% CI) | 0.77 [0.66, 0.90] |
| 2.1 Individually randomised trials | 3 | 1170 | Risk Ratio (M‐H, Random, 95% CI) | 0.89 [0.79, 1.00] |
| 2.2 Cluster‐randomised trials (modified sample size) | 3 | 1975 | Risk Ratio (M‐H, Random, 95% CI) | 0.67 [0.60, 0.75] |
| Outcome or subgroup title | No. of studies | No. of participants | Statistical method | Effect size |
| 1 Antibiotics prescribed at index consultation: Cluster‐randomised with modified sample size Show forest plot | 2 | 2024 | Risk Ratio (M‐H, Random, 95% CI) | 0.70 [0.63, 0.78] |
| 1.1 Upper respiratory tract infections | 2 | 510 | Risk Ratio (M‐H, Random, 95% CI) | 0.72 [0.58, 0.90] |
| 1.2 Lower respiratory tract infections | 2 | 1514 | Risk Ratio (M‐H, Random, 95% CI) | 0.69 [0.62, 0.78] |
| Outcome or subgroup title | No. of studies | No. of participants | Statistical method | Effect size |
| 1 CRP ‐ Antibiotic prescribing when algorithms provide clear cut‐offs to rule out (< 20 mg/L) Show forest plot | 4 | 2233 | Risk Ratio (M‐H, Random, 95% CI) | 0.69 [0.62, 0.76] |
| 1.1 Individual trials | 1 | 258 | Risk Ratio (M‐H, Random, 95% CI) | 0.77 [0.60, 0.98] |
| 1.2 Cluster‐randomised trials. Modified sample size | 3 | 1975 | Risk Ratio (M‐H, Random, 95% CI) | 0.67 [0.60, 0.75] |
| 2 CRP ‐ Recovery at day 7 Show forest plot | 1 | 243 | Risk Ratio (M‐H, Random, 95% CI) | 1.03 [0.89, 1.18] |
| 3 CRP ‐ Recovery at follow‐up (max 28 days) Show forest plot | 2 | 608 | Risk Ratio (M‐H, Random, 95% CI) | 1.03 [0.75, 1.41] |
| Outcome or subgroup title | No. of studies | No. of participants | Statistical method | Effect size |
| 1 CRP ‐ Patient recovery day 7: missing data in CRP = not recovered Show forest plot | 3 | 1309 | Risk Ratio (M‐H, Random, 95% CI) | 1.10 [0.99, 1.21] |
| 2 Patient recovery day 28: missing data in CRP = not recovered. Cluster‐randomised trials with modified sample size Show forest plot | 3 | 549 | Risk Ratio (M‐H, Random, 95% CI) | 1.11 [0.84, 1.48] |
