Biomarkers to Guide the use of Antibiotics for Acute Exacerbations of COPD (AECOPD)

A Systematic Review and Meta-Analysis

George Hoult; David Gillespie; Tom M. A. Wilkinson; Mike Thomas; Nick A. Francis


BMC Pulm Med. 2022;22(194) 

In This Article


Following de-duplication and application of inclusion/exclusion criteria our search identified 509 papers. A review of the title and abstract of these 509 papers led to the exclusion of a further 469 papers. Reasons for exclusion were: the paper did not investigate patients with AECOPD (n = 108), did not assess the diagnostic accuracy of biomarkers (n = 264) or did not assess the properties of biomarkers in differentiating bacterial versus non-bacterial AECOPD (n = 97). A further one study was excluded during full text screening because it did not differentiate patients with acute exacerbations from patients with stable disease, leaving 39 studies which were included (Figure 1).

Figure 1.

PRISMA flow diagram of literature search results

Study publication dates ranged from 1998 to 2019. Four[10–13] were in an ICU setting, 24[14–37] were hospital inpatient (not ICU), one[38] was both inpatient and outpatient, nine[39–47] were just outpatient, and one[48] did not specify the setting. In several studies, multiple exacerbation events were recorded for single patients, leading to a maximum of 736[26] and minimum of 14[47] exacerbation events reported in the studies.

Risk of bias was assessed using the QUADAS-2 tool (Table 1); for patient selection, eight[11,16,19,24,32,37,42,48] and 22[15,17,18,21–23,25–29,31,35,36,38–41,43,45–47] studies were assessed as being at high and unclear risk of bias respectively. For the index test, one[19] and three[27,30,48] studies were assessed as being at high and unclear risk of bias respectively; and reference standard, five[19,27,28,37,45] and 18[14,15,16,17,22,24–26,30,32,33,35,36,38,41,43,47,48] studies were assessed as being at high and unclear risk of bias respectively. Only 14 (36%)[18–21,23,25,27,33,35,37–39,41] studies excluded patients that had been prescribed antibiotics and corticosteroids prior to sample collection. Five (13%) studies only excluded participants who had been prescribed antibiotics[16,17,31,36,42] and another two (5%) only participants who had been prescribed corticosteroids.[40,47] 17 (44%)[10–15,22,24,26,28–30,32,34,44–46] studies did not exclude patients who had taken either antibiotics and/or systemic steroids prior to sample collection.

COPD diagnosis was based on Global Initiative for Chronic Obstructive Pulmonary Disease (GOLD) criteria in 24 (62%) studies,[10,11,13,14,16,17,20–24,26–29,32–35,37–39,41,45] national thoracic society guidelines in six[12,30,31,36,43,47] studies, 'physician diagnosis' in three[17–19] studies, and was not described for six[15,40,42,44,46,48] studies.

Eight (21%) studies[10,13,18,26,28,30,33,39] used the GOLD definition of an exacerbation event ("acute worsening of respiratory symptoms that results in additional therapy"), and 26 (67%) studies[11,12,14–16,19–25,27,31,32,34–38,40,42–46] described exacerbations as 'worsening of respiratory symptoms', with no mention of need for additional therapy. Five[17,29,41,47,48] studies did not indicate how they defined exacerbation events.

Most studies used growth of pathogenic bacteria from respiratory specimens to define bacterial exacerbations, but many of these did not provide detailed descriptions. Three[22,46,47] studies defined bacterial growth as growth of a 'new strain', six[10,15,19,25,28,37] studies used a combination of clinical observations, sputum culture, X-ray images and lab results; and two[43,48] studies did not state how they defined bacterial exacerbation.

The 39 included studies evaluated 61 biomarkers (27 biomarkers that were evaluated in both serum and sputum samples, an additional 28 that were only evaluated in serum and an additional 6 that were only evaluated in sputum) giving a total of 55 serum and 33 sputum biomarkers (Table 2). For most biomarkers there was insufficient data to draw meaningful conclusions. Three serum biomarkers (CRP, PCT and white blood cell count) and six sputum biomarkers (IL-8, TNF-α, IL-1β, IL-6, MPO and NE) had data from more than two studies and the findings relating to these biomarkers are summarised below.

Serum Markers

Serum C-reactive Protein (CRP). We identified 28 studies that included an assessment of serum C-reactive protein (CRP) to determine bacterial aetiology in AECOPD (Table 3). Most studies were small, with 17 having fewer than 85 exacerbation events. 18 studies provided quantitative data, of which 15 (83%) reported higher levels of serum CRP in bacterial versus non-bacterial exacerbations, and the difference was statistically significant in 12.[11,13,20,23,27,31,35–37,39–41] Of the 10 papers that did not provide accessible numerical data, two reported a significant association between CRP level and bacterial AECOPD,[19,21] and another two reported a significant association between CRP level and purulent sputum, a proxy for bacterial AECOPD.[25,42] A further three[14,38,45] did not find any significant association between CRP and bacterial AECOPD. One paper found that isolation of new pathogenic strains in sputum was associated with a greater increase in CRP compared to pre-exacerbation levels than for non-bacterial AECOPD, pre-existing strain isolation or other strain isolation.[46]

Only 10[12,13,17,20,22,23,31,35–37] studies provided a mean and standard deviation CRP for each group, and therefore could be included in a meta-analysis. The meta-analysis found that bacterial exacerbations were associated with significantly higher CRP values, with a weighted mean difference of 29.44 mg/L (Figure 2). However, high heterogeneity with I2 = 96.93% was observed. Subgroup analysis by setting, use of antibiotics or steroids, or definitions of COPD, exacerbation or bacterial infection did not reduce heterogeneity.

Figure 2.

Forest plot of the difference in mean CRP values in AECOPD patients with and without a bacterial infection

Several authors have suggested cut-points for use in clinical practice to best identify bacterial vs non-bacterial AECOPD (Table 4). The cut-points range from 5 to 110 mg/L and tend to vary by setting. The best overall test characteristics were reported by Hassan et al.[19] who found that a cut point of 15 mg/L resulted in a sensitivity and specificity of 95.5% and 100% respectively for identifying bacterial pathogens in a study of 30 patients with AECOPD in a hospital inpatient setting.

Serum Procalcitonin (PCT). 17 papers described the association between serum Procalcitonin (PCT) and evidence of a bacterial infection in AECOPD (Table 5). Of the 15 that provided numerical data, 11 (65%) found higher PCT concentrations in patients with bacterial AECOPD compared to non-bacterial AECOPD, and six of these reported a statistically significant difference.[13,31,35–38] Of the two papers that did not include numerical data, one reported that PCT concentrations > 0.5 ng/mL and "positive CRP" (which was not defined) were both associated with positive sputum culture, and that PCT < 0.5 ng/mL was strongly associated with non-bacterial AECOPD,[24] whilst the other reported no significant difference in PCT concentrations between bacterial and non-bacterial AECOPD.[14]

We were able to extract mean and standard deviation PCT levels for those with and without evidence of a bacterial infection from nine[12,13,18,22,31,35–38] studies. Combining these data using meta-analysis we found higher mean PCT in those with a bacterial exacerbation, with a weighted mean difference of 0.76 ng/mL (95% CI: 0.16, 1.36 ng/mL; Figure 3). High heterogeneity (I2 = 97.95%) was also observed. Again, subgroup analysis by setting, use of antibiotics or steroids, or definitions of COPD, exacerbation or bacterial infection did not reduce heterogeneity.

Figure 3.

Forest plot of the difference in mean PCT values in AECOPD patients with and without a bacterial infection

The authors of these papers describe cut-points ranging from 0.03–1.03 ng/mL to identify bacterial vs non-bacterial AECOPD (Table 6). The study with the greatest area under receiver operating characteristic curve (ROC) used a cut-point of 0.76 ng/mL resulting in a specificity of 92.5% and sensitivity of 78.95%, with area under the curve in diagnosing bacterial AECOPD of 0.941.[35]

Serum WBC. Five studies, ranging in size from 60 to 150 participants and 26 to 82 bacterial exacerbations, examined associations between serum WBC count and bacterial (versus non-bacterial) AECOPD.[14,29,35,36,38] Three were in hospital inpatient settings and two were in emergency departments. None of the studies demonstrated a statistically significant association.

Sputum Mrkers

Sputum IL-8. Seven studies compared sputum IL-8 levels in bacterial and non-bacterial AECOPD. The studies varied in size from 14[47] to 84[46] participants, with between 14 and 158 exacerbation events. Four[16,23,39,40] of these studies found significantly higher IL-8 levels associated with bacterial AECOPD (Table 7). One study[43] reported a positive association between the rise in airway bacterial load and the rise in sputum IL-8 concentrations during exacerbation, although this is not the same as differentiating bacterial from non-bacterial exacerbations. Another study[46] found no significant difference in sputum IL-8 concentrations between those with a new strain, and those with a pre-existing strain, other strain or non-bacterial AECOPD.

Importantly, there was large variation in the detected levels of sputum IL-8 between studies. One[40] reported a mean IL-8 of 7.78 μg/mL in bacterial exacerbations, while another[23] reported a mean concentration of 468 pg/mL in those with bacterial exacerbations. Five[16,23,39,43,46] studies had mean sputum IL-8 concentrations between 0.468 and 16.6 ng/mL, whilst one[47] had mean concentration of 100 ng/mL, the other[40] 7.78 μg/mL. The variation in concentrations did not appear to be explained by setting or testing procedure, as there was no important difference between methods of sputum induction nor the concentration of Il-8 in studies that used ELISA (n = 3) and studies that did not (n = 4).

Sputum TNF-α. The association between sputum TNF-α concentrations and evidence of bacterial exacerbation was studied in five papers (Table 8). The studies were performed in an outpatient setting,[39,40,46,47] or only included patients with stage II AECOPD according to GOLD classification,[16] and so included patients with mild-moderate exacerbations.

Average sputum TNF-α was significantly higher in bacterial exacerbations than non-bacterial in four papers.[16,39,40,46] The other study only identified one patient with bacterial AECOPD and therefore could not draw meaningful conclusions.[47] One of these studies further classified causes of exacerbation into common bacteria, Pseudomonas aeruginosa, viral and non-infective. They found mean TNF-α to be significantly higher in patients with Pseudomonas aeruginosa infections than the other three groups.[16]

Sputum IL-1B. Three studies investigated the utility of sputum IL-1β as a marker for bacterial AECOPD.[16,39,48] One study found that sputum IL-1β had an area under ROC of 0.89 for detecting bacterial exacerbations, and that a cut-point of 125 pg/mL had sensitivity and specificity of 90% and 80% respectively.[39] Another found a significant association between ≥ twofold increase (compared to stable state) in sputum IL-1β concentrations (IL-1β+ event) and bacterial exacerbations.[48] The third reported an area under receiver operating curve of 0.87 from use of a combination of sputum IL-8 and IL-1β for determining common bacterial from viral and non-infectious AECOPD, but only after first excluding P. aeruginosa infections.[16]

Sputum IL-6. Sputum IL-6 was investigated in three studies. One study involving 45 exacerbations reported a difference in IL-6 concentrations that was not statistically significant (680 pg/mL vs. 325 pg/mL; p > 0.05), but a difference in percentage change that did reach statistical significance (116% vs. −16%; p < 0.05).[40] Another study of 182 exacerbation events from 96 patients found that the area under ROC was 0.7 in determining bacterial from non-bacterial AECOPD.[39] In contrast, Wilkinson found that there was no significant change between stable state and AECOPD, and that there was no association between sputum IL-6 concentrations and airway bacterial load.[43]

Sputum MPO. Three small studies investigated sputum MPO as a marker of bacterial AECOPD. Two of these were too small to draw any meaningful conclusions (29 exacerbation events[45] and 14 exacerbation events).[47] The third found significantly higher MPO concentrations in patients with bacterial versus non-bacterial AECOPD (57.7 vs. 12.6 μg/mL; p < 0.05) in a study involving 45 exacerbation events.[40]

Sputum NE. Sputum NE was investigated in three studies. One study of 30 exacerbations with 13 of bacterial aetiology found an association between sputum NE concentrations and bacterial AECOPD (log difference 3.873; p = 0.011). In addition, NE was positively correlated with CFU load (r = 0.506; p = 0.005).[34] Neither of the other two (N = 150 and 29) studies found significant associations with bacterial AECOPD, but one reported a significant association with detecting a new bacterial strain, not present in stable state, at AECOPD (new strain; p < 0.001).[45,46]