Study design

The three studies included were single‐centred studies and none received support from commercial interests. Two studies were conducted in a paediatric outpatient clinic in Sweden and both recruited children with over 10 days of cough (Darelid 1993; Gottfarb 1994). Marchant 2012 recruited children with chronic wet cough (greater than three weeks' duration) from a paediatric respiratory outpatient clinic in Australia. Although the inclusion criteria for duration of cough was short, the cough duration of the participants were longer in all the RCTs (described below under 'Participants'). A subgroup of children from one RCT had flexible bronchoscopy (FB) performed pretreatment and their BAL data were consistent with PBB (Marchant 2012).

One study was an open randomised study comparing erythromycin and a no treatment control group (Darelid 1993). Two studies compared amoxicillin/clavulanic acid and placebo in a double‐blind RCT (Gottfarb 1994; Marchant 2012). Two studies used a dosage regimen equal to current recommended treatment dosage (Darelid 1993; Marchant 2012). The dosage in Gottfarb 1994 was lower (dosage 20 mg/kg/day) than currently recommended doses (22.5 mg/kg/day to 45 mg/kg/day) (Kemp 1997). Although the studies used two different antibiotics, both would treat the respiratory pathogens found in the nasopharyngeal aspirates of these children and organisms in BAL in other studies on chronic cough in children (Chang 2006a; Gedik 2015; Zgherea 2012). The children in one study all received oxymetazoline nose drops and salbutamol mixture was allowed and recorded in both groups (Darelid 1993). Neither of these treatments should have influenced the outcome since they were available to both groups. The other two trials did not allow cointerventions (Gottfarb 1994; Marchant 2012). One trial measured compliance by the return of the medication bottles (Marchant 2012). Darelid 1993 reported 97% of participants completed the erythromycin medication; however, they failed to state how this was assessed.

Participants

All three studies reported different length of cough in their participants. Around 50% of children had cough of greater than three weeks' duration in Darelid 1993, the mean length of cough was three to four weeks in Gottfarb 1994, and median length of cough was 11 to 15 weeks in Marchant 2012. The median age of participants in Darelid 1993 was not provided but 63% of the participants were aged one to three years. Median age of children in Gottfarb 1994 was 2.6 years. Likewise, the children in Marchant 2012 were preschool age with a median of 1.75 to 2.8 years. Although two studies did not specifically mention the quality of cough, contact with the authors confirmed that at least 75% of children had wet cough (Darelid 1993; Gottfarb 1994). Also, as both studies included nasopharyngeal aspirates which grew a range of bacterial pathogens, it is highly likely that these children had wet cough.

Interventions

The antibiotics and their duration varied in the studies. Two studies compared amoxicillin/clavulanic acid and placebo (Gottfarb 1994; Marchant 2012); however, the children in Gottfarb 1994 took the antibiotics for seven days, whereas the children in Marchant 2012 took the antibiotics for 14 days. The intervention arm in Darelid 1993 took erythromycin for seven days and the control group did not take any treatment.

Outcomes

Two studies used paediatrician assessment of clinical recovery as primary outcome measure but the day of assessment was different. Outcomes were assessed at day eight (Darelid 1993) and day 12 to 14 (Gottfarb 1994), so the results described only short‐term effects. The primary outcome measure in both Swedish studies was paediatrician assessment of clinical outcome at follow‐up. Participants were classified as cured or treatment failure. Both studies also reported parental assessment and used individual symptom scores as secondary outcome measures.

The Australian study used 'cough resolution' as their primary outcome (Marchant 2012). They defined this as an improvement of more than 75% from baseline cough score, or three days of being cough free for the study duration. Their secondary outcome measures were absolute change in cough score or change in verbal category descriptive (VCD). Thirty‐seven of the 50 children underwent a bronchoscopy prior to commencing the intervention. The BAL microbiology in this subgroup identified typical respiratory pathogens including Haemophilus influenzae (14 children), Streptococcus pneumonia (nine children) and Moraxella catarrhalis (seven children). All these organisms were sensitive to amoxicillin/clavulanate acid.

Studies also documented adverse effects and progression of disease requiring further antibiotic therapy. Darelid 1993 also recorded and documented the elimination of nasopharyngeal pathogens in the intervention and control groups. Gottfarb 1994 excluded 15 children from the final analysis without indicating to which treatment group they were initially allocated. As contact with the authors was unsuccessful, we assumed that there were equal numbers of children in each group in the intention‐to‐treat analysis. Of the 15 children excluded, 12 had laboratory‐confirmed B pertussis after randomisation which is a potential weakness of the study. Ideally the diagnosis of B pertussis would have been made prior to randomisation and then these children would not have been eligible to enter the study.

Ongoing studies

We found no ongoing studies.

Children not cured or not substantially improved at follow‐up (measured as proportion of participants) (using intention‐to‐treat analysis)

We combined data from all included studies for children not cured or not substantially improved at follow‐up. The number of children not cured at follow‐up was 110 using an intention‐to‐treat analysis. The control event rate was 64% (Darelid 1993), 88% (Gottfarb 1994), and 84% (Marchant 2012) in the three studies. Treatment with antibiotics significantly reduced the proportion of children not cured at follow‐up; the pooled OR effect estimate was 0.15 (95% CI 0.07 to 0.31; P < 0.00001; participants = 190, studies = 3; Analysis 1.1; Figure 3). These studies suggested that one child will be cured for every three children treated (NNTB 3, 95% CI 2 to 4) (Figure 4). Using GRADE, we assessed the strength of evidence to be moderate for this effect.

Figure 3

---

---

Forest plot of comparison: 1 Antibiotics versus no antibiotics/placebo for wet cough in children, outcome: 1.1 Children not cured or not substantially improved at follow‐up (using intention‐to‐treat analysis).

Figure 4

---

---

In the control group, 75/100 people had not been cured over two weeks, compared to 31/100 (95% CI 17 to 48) for the active treatment group. Number needed to treat for an additional beneficial outcome 3 (95% CI 2 to 4).

Children not cured or not substantially improved at follow‐up (measured as proportion of participants) (excluding those know to have B pertussis)

Two studies reported B pertussis data which could be combined for this outcome, hence we made the decision post‐hoc to do an analysis excluding these children (Darelid 1993; Gottfarb 1994). The number of children not cured (excluding those with known B pertussis) was 17 in the antibiotics group versus 47 in the placebo group, which was statistically significant (OR 0.13, 95% CI 0.05 to 0.30; P < 0.00001; participants = 128; studies = 2; Analysis 1.4).

The data from Marchant 2012 could not be included in the meta‐analysis as microbiological assessment for pertussis was not undertaken. However, they reported that 27/50 children had clinically important BAL microbiology (organism density of 10⁴ or more CFU/mL BAL) with typical respiratory flora including Haemophilus influenzae (38%), Streptococcus pneumoniae (24%) and Moraxella catarrhalis (19%). They reported that all these organisms were sensitive to amoxicillin/clavulanate acid.

Change in clinical state (mean difference)

All studies reported overall improvement in the antibiotic group but data could not be combined for this outcome.

Difference in sputum, bronchoalveolar lavage or blood indices of inflammation or infection (mean improvement in markers of infection)

Darelid 1993 examined clearance of potential pathogens from the nasopharynx using bacteriology. It found a significant higher elimination of Moraxella catarrhalis from the nasopharynx in treated (68%) versus untreated participants (20%; P < 0.001). There were similar results for Streptococcus pneumoniae (treated elimination 100%; untreated 23%; P < 0.0001). They also found a correlation with growth of Moraxella catarrhalis and persistent cough after the trial. In the untreated group, 80% of participants with untreated Moraxella catarrhalis continued to cough versus 33% of participants with no growth of this organism (P < 0.01). None of the studies provided data on the change in the analysis of sputum, BAL or blood indices of infection before and after treatment.

Improvement in cough indices (cough diary, cough frequency, cough scores, quality of life) (measured as mean difference)

All studies reported an improvement in a type of score that involved cough. These measures could not be combined due to the differing nature of the indices. Darelid 1993 used a score which combined cough frequency, morning temperature and level of daily activity, but provided no numerical value. Gottfarb 1994 looked at the number of coughing attacks in a 24‐hour period but a single numerical value could not be obtained as data were expressed as proportion of children in each category. They reported a greater reduction in the coughing attacks in the antibiotic group compared to the placebo group. Marchant 2012 used a validated cough diary and reported a significant difference between the treatment group and the placebo group and found that the median improvement in cough score was significantly higher in the antibiotic group (1.5, IQR 0 to 2.5) compared to the control group (0.5, IQR –0.4 to 1.0; P = 0.02).

Children with progression of the disease resulting in additional medical therapy (complications) (proportion of participants)

Overall 125 children contributed data on this outcome (excluding the 15 dropouts in Gottfarb 1994). Marchant 2012 did not record further antibiotic use and therefore could not be included in this analysis. The overall control event rate for complications was 36% versus a treatment event rate of 5%. Progression of illness, as defined by requirement for further antibiotics in both studies (Darelid 1993; Gottfarb 1994), was significantly lower in the treatment group (OR 0.10, 95% CI 0.03 to 0.34; P = 0.0002; Analysis 1.2; Figure 5). The NNTB to avoid progression of disease was 4 (95% CI 3 to 5). Using GRADE, we assessed the quality the of evidence to be moderate for this effect.

Figure 5

---

---

Forest plot of comparison: 1 Antibiotics versus no antibiotics/placebo for wet cough in children, outcome: 1.4 Illness progression: participants with progression of disease resulting in additional medical therapy required.

Children experiencing adverse effects of antibiotics (e.g. diarrhoea, nausea, skin rash, allergic reactions) (proportion of participants)

In Darelid 1993, children in the erythromycin group had a reported adverse effect rate of 2% (Darelid 1993). These were diarrhoea and vomiting. It can be argued that these results should not been pooled as Darelid 1993 did not report any adverse effects for the control group. The rate of adverse effects in the study comparing amoxicillin/clavulanic acid and placebo was 12% for both treatment and control groups (Gottfarb 1994). Marchant 2012 also used amoxicillin/clavulanic acid at a significantly higher dosage than Gottfarb 1994, and as such the reported adverse effect rate in their study was higher at 20% in the treatment group and 8% in the control group. Pooling the results, assuming that no one in the control (no treatment) group in Darelid 1993 had any adverse events gave an OR of 1.88 (95% CI 0.62 to 5.69; P = 0.26; participants = 190; studies = 3; Analysis 1.3). Given the wide CIs, this finding remains uncertain and one can neither be certain of harm, or otherwise, from adverse effects of intervention.

Children under seven years of age

Two studies recruited children aged under seven years (Darelid 1993; Gottfarb 1994), and the mean age for the remaining study was under three years (SD 0.9) in both control and placebo group (Marchant 2012); hence, subgroup analysis was unnecessary.

Control type (placebo versus no placebo/usual treatment)

Two studies used a placebo (Gottfarb 1994; Marchant 2012). When compared to the one study that used a no‐treatment control group (Darelid 1993), both subgroups continued to favour antibiotics over control and in terms of children not cured or substantially improved. The test indicated no significant difference between subgroups (Chi² = 1.30, df = 1 (P = 0.25), I² = 23.1%; Analysis 1.5). However, the test was an underpowered due to low number of studies included.

Variation in the inclusion criteria (cough duration longer than four weeks)

Although the two earlier studies had about 50% of children with prolonged wet cough for over four weeks, we could not extract disaggregated data for numbers of children not cured or substantially improved (Darelid 1993; Gottfarb 1994). Marchant 2012 was the sole study where children had been coughing for longer than four weeks (although inclusion criterion was over three weeks). Thus, for this subgroup analysis, we compared data from Marchant 2012, with data from the remaining two studies. Again, both subgroups favoured antibiotics over control and the test for subgroup difference was negative (Chi² = 0.28, df = 1 (P = 0.59), I² = 0%; Analysis 1.6). However, the test was underpowered due to low number of studies included.

Antibiotic types (differences in the medications used in the intervention group)

Two studies used amoxicillin/clavulanic acid (Gottfarb 1994; Marchant 2012). We compared these studies to the single study using erythromycin (Darelid 1993). Both subgroups favoured antibiotics over control in terms of the number of children not cured or substantially improved and the test for subgroup difference was negative (Chi² = 1.30, df = 1 (P = 0.25), I² = 23.1%). However, the test was underpowered due to low number of studies included.

Sensitivity analysis by intention to treat but only including participants not lost to follow‐up (available data analysis) for primary outcome

Combining the data from the three included studies using data that were available (as opposed to intention to treat), the proportion of children randomised to the antibiotics arm were significantly less likely to be 'not cured' compared to controls (OR 0.14, 95% CI 0.07 to 0.29; P < 0.00001; participants = 171; studies = 3; Analysis 1.8). The effect size was slightly reduced compared to the intention‐to‐treat analysis. The NNTB for benefit at the end of study was 3 (95% CI 2 to 4). Using GRADE, we assessed the strength of evidence to be moderate for this outcome.

Sensitivity analysis for primary outcome model used: fixed effect versus random effects, risk ratio versus odds ratio

There was no difference when a random‐effects model was used versus a fixed‐effect model (OR for primary outcome 0.15 (fixed) versus 0.15 (random)). Both the Chi² and I² statistical tools show low level of heterogeneity between the studies, with I² = 0% and Chi² = 1.33 for the primary outcome measure.

In the primary outcome (clinical failure), the overall estimate of effect and degree of statistical heterogeneity were sensitive to the model used for meta‐analysis although they were still significant (pooled RR 0.51, 95% CI 0.39 to 0.66 (fixed‐effect model, I² = 81%); pooled RR 0.51, 95% CI 0.28 to 0.95 (random‐effects model, I² = 81%). Both results significantly favoured use of antibiotics to prevent progression of disease.