Intended for healthcare professionals

CCBYNC Open access

Rapid response to:

Research

Efficacy of anti-inflammatory or antibiotic treatment in patients with non-complicated acute bronchitis and discoloured sputum: randomised placebo controlled trial

BMJ 2013; 347 doi: https://doi.org/10.1136/bmj.f5762 (Published 04 October 2013) Cite this as: BMJ 2013;347:f5762

Rapid Response:

Re: Efficacy of anti-inflammatory or antibiotic treatment in patients with non-complicated acute bronchitis and discoloured sputum: randomised placebo controlled trial

Authors’ response

Dr Strauss addresses an interesting point with the observation of eight patients with non-asthmatic postviral non-productive cough lasting from 4 days to 3 months (six with a prior duration of at least 3 weeks) improving with the administration of oral prednisone at different doses.1 Previous studies have shown that 20-25% of patients diagnosed with acute bronchitis return within one month of the initial consultation and most might receive one or more courses of antibiotics.2 In our study, however, only patients with less than one week of cough and productive sputum were recruited, and therefore our results cannot be extrapolated to the population of patients with a longer duration of cough. In our trial the survival curves started diverging from day 10 onwards. We analysed the data of those patients still coughing at day ten, but the primary outcome analysis by Kaplan-Meier test showed no differences between groups in cough resolution (log-rank test=0.351); however, this subgroup analysis does not have enough statistical power to allow firm conclusions. Strauss’s proposal for the development of a randomised clinical trial in patients with long-standing cough without evidence of pertussis infection comparing the benefit of oral corticosteroids with placebo would definitely answer the question as to whether anti-inflammatory drugs have any benefit in this condition.

As Professor Van Oosten remarked, we agree that the null hypothesis cannot be accepted, we should have therefore concluded that there was weak evidence to reject the null hypothesis that the three survival curves were not different. We also agree that the use of the log-rank test is not the most appropriate when survival curves cross each other. However, in this study, it appears more likely that the survival curves cross each other at random points of time rather than at times signifying important systematic differences between groups. If this is indeed the case, the proportionality of the hazards assumption would be fulfilled and the log-rank test would be suitable. In order to strengthen the hypothesis that the survival curves cross each other randomly, we have performed a simulation study based on the assumption of identical underlying hazard rates in the two groups. The survival curves observed for the main outcome (number of days of frequent cough) for ibuprofen and placebo crossed each other five times, which were the two groups exhibiting the largest differences between each other. Consequently, we conducted a simulation study in which we randomly assigned 130 patients to one group and 136 to the other as in the real study. We counted the number of times the two curves crossed each other in 1,000 simulated datasets and found that this occurred five times or more in 42.1% of the datasets. If the two hazards are thus assumed to be identical, crossing of survival curves can be expected to be a very common phenomena, and given the very similar survival curves observed in the study, we consider the insignificant p-value of the log-rank test an adequate reflection of the study information here.

We have also performed Cox Proportional Hazards models for the main outcome of interest and the secondary outcome (number of days with any symptom) in which we included the treatment indicator as an explanatory variable as well as its interaction with time (categorized as 0-10 days; 10-20 days; or 20-30 days) using the Placebo group and the 0-10 days as the reference. None of the estimates in either of the two models were statistically significantly different from 0, which means that there is weak evidence to think that either of the two active treatments have different effects to those of placebo in any of the 3 periods of time, although the survival curves looked somewhat more different in the period between 10 and 20 days. On graphically checking the proportionality of the hazards plotting –ln(-ln(Survival)) vs. ln(time), the lines were not only parallel but also almost identical, which strengthens the null hypothesis of no difference between treatment groups. Although these analyses compare the two active treatments separately with placebo, we also performed the log-rank test to compare them. For the primary outcome none were significant, and only one was for the secondary outcome (ibuprofen vs. placebo, p=0.0435). This may well, however, be explained by multiple testing and, in any case, is only marginally significant.

Acknowledgment
We wish to acknowledge the contribution of Associate Professor Henrik Støvring for his assistance in this response.

References
1. Strauss RA. Treatment of postviral nonasthmatic cough with corticosteroids. J Allergy Clin Immunol: In Practice 2013;1:404-6.
2. Chalmers JD, Hill AT. Investigation of "non-responding" presumed lower respiratory tract infection in primary care. BMJ 2011;343:d5840.

Competing interests: No competing interests

17 February 2014
Oleguer Plana-Ripoll
PhD fellow
Carl Llor
Section of Epidemiology. Department of Public Health. Aarhus University, Denmark
Bartholins Allé, 2, 8000 Aarhus C, Denmark