Delayed Antibiotic Prescription by General Practitioners in the UK: A Stated-Choice Study
Round 1
Reviewer 1 Report
The manuscript is very interesting and sounding.
Please revise caption format (especially for fig. 3) to make them more readable.
Author Response
We thank the three reviewers who read our paper and kindly provided comments. Below we respond to the comments from Reviewer 1.
Please revise caption format (especially for fig. 3) to make them more readable.
Author response
The journal editors have now re-formatted the figure captions for us into their usual publication format. We have revised the figure caption texts by organising them more clearly, identifying the sections within the caption, and providing additional text to help with interpreting the figures (pages 10, 11, 17, 18).
Reviewer 2 Report
Article: Delayed antibiotic prescription by general 2 practitioners in the UK: a stated-choice study
General context: The invention of antibiotics saved the lives of millions of people. However, the overuse of antibiotics has become a serious problem in the modern world because it affects the formation of microbial resistance. The authors took up a very important problem, for which I thank them very much. Every effort should be made to limit the use of antibiotics and “Amongst the strategies aimed at reducing antibiotic consumption, one option in primary care is delayed (or ‘back-up’) prescriptions”.
The Materials and methods section is missing! Some information is included in the introduction, but the methodology section should be separated.
There is no information on the method of sampling, there is no description of the variables and the statistical methods used. When was the survey conducted? What was the survey return rate? Was the questionnaire validated?
Results:
„The sample was representative of UK GPs with respect to sex, age, country and practice size” - we know nothing about the method of recruiting the respondents; from the title, we can conclude that the choice was deliberate? if so, we cannot speak of the representativeness of the group; Only a random selection of respondents guarantees representativeness.
Next, if the questionnaire was difficult for every fifth doctor to complete, it is a lot; one may wonder about the reliability of the results. There was definitely no pilot study, which is good to do before the actual study to check, for example, if all the questions in the survey are well understood
Answers to the above comments should be found in the section Materials & Methods.
p.8-9 (Table 2). - There was no seniority among the surveyed variables. Work experience, and thus work experience, but also habits, can significantly affect the results of the model; it would also be good to take this variable into account in the analyzes!
Some descriptions of the methods used, e.g. logistic regression (page 11), should be included in the Materials and Methods section. However, the remaining results (from page 15) are described correctly.
I found the Materials and methods section at the end of the article - please skip the order and take into account the comments described earlier.
Author Response
We thank the three reviewers who read our paper and kindly provided comments. Below we respond to the comments from Reviewer 2.
The Materials and methods section is missing! Some information is included in the introduction, but the methodology section should be separated. There is no information on the method of sampling, there is no description of the variables and the statistical methods used. When was the survey conducted? What was the survey return rate? Was the questionnaire validated?
Author response
Thank you for this comment. Antibiotics as a journal asks for authors to place the Materials and Methods as the last section, and we followed this instruction. The information requested by the reviewer on sampling, variables, statistical methods, dates of fieldwork, and piloting, are included in the Methods section at the end of the paper (pages 22-26).
We have now added a comment in the Methods section on response rate, noting that we do not have a measure of response rate because the respondents were invited to participate via the provider’s website, and we did not have a way to measure the number of people who saw the invitation (page 24).
Results:
„The sample was representative of UK GPs with respect to sex, age, country and practice size” - we know nothing about the method of recruiting the respondents; from the title, we can conclude that the choice was deliberate? if so, we cannot speak of the representativeness of the group; Only a random selection of respondents guarantees representativeness.
Author response
In saying the sample was “representative”, we mean that the sample was recruited to match the population of UK GPs, in terms of age, sex, country within the UK, and practice size. This was done by use of quotas. We have revised the sentence, which hopefully makes this clearer (page 7, Results, section 2.1).
Because respondents opted in to the study voluntarily, however, it may be that our respondents have a particular interest in infectious disease or antibiotic resistance. We have added a discussion of this potential for bias to the Discussion (page 22) to help explain this potential issue.
Next, if the questionnaire was difficult for every fifth doctor to complete, it is a lot; one may wonder about the reliability of the results. There was definitely no pilot study, which is good to do before the actual study to check, for example, if all the questions in the survey are well understood
Author response
The scenarios in this study included eight attributes, which is more than we would typically use in studies among the general public, and 21% reporting the study as ‘difficult’ is higher than we would expect to see in general population studies. However, we did not receive negative feedback about the difficulty of the questions, either in the pilot nor the main survey. Furthermore, we are reassured by the consistency of the choices made by respondents (shown by the high level of variation explained by the regression model) that they were able to make rational responses, even if those decisions were difficult and required them to think carefully. We have expanded our discussion of the limitations of the study to include this point (page 21).
(The study was piloted among a sample of 23 GPs, as described in the Methods.)
p.8-9 (Table 2). - There was no seniority among the surveyed variables. Work experience, and thus work experience, but also habits, can significantly affect the results of the model; it would also be good to take this variable into account in the analyses
Author response
We agree that seniority might have been an interesting variable, and we will bear that in mind in future studies. We did, however, look at both the effect of age as a proxy for how long they had been practicing, and also of whether the respondent was a partner in their practice (another indicator of seniority). Neither of these variables showed a significant effect in the models. We have added a sentence to the section on the effect of respondent characteristics (page 16) to include this observation.
Some descriptions of the methods used, e.g. logistic regression (page 11), should be included in the Materials and Methods section. However, the remaining results (from page 15) are described correctly.
Author response
Logistic regression, and in particular the basis for using the mixed-effects logit, is described in the Methods section. We have added some explanation in the Results section (page 11), to reflect the fact that most readers will be reading the Results before they reach the details in the later Methods section.
Reviewer 3 Report
Some of the points in this study are not appropriate and cannot be accepted.
- In this study, all members completing the survey in full received 4000 eSR points. It is considered that the reliability of the content of the answer is reduced by answering in exchange for the reward because it includes answers for compensation purposes.
- Fig.2:Readers can not understand of each question because there are not explanation of question number.
- Fig.3: Why did several data have no difference between blue and red column.
Author Response
We thank the three reviewers who read our paper and kindly provided comments. Below we respond to comments from Reviewer 3.
1. In this study, all members completing the survey in full received 4000 eSR points. It is considered that the reliability of the content of the answer is reduced by answering in exchange for the reward because it includes answers for compensation purposes.
Author response
eSR points are the internal reward system used by our recruitment provider, and can be converted into vouchers by respondents. 4000 eSR point equates to £20, which is the standard level of reward used by the provider for this type and length of survey. We have added this information to the Methods section (4.4, Data Collection; page 24).
Rewards of this type are routinely used to encourage participation in surveys. Whilst there is potential for bias due to the presence of an incentive, the absence of an incentive could also produce a biased sample, due under-representation of certain subgroups of the target population. In this study we chose to follow standard practice of offering a small reward. We have noted the potential for bias due to the presence of an incentive, in the section on limitations in the Discussion (page 22).
2. Fig.2:Readers can not understand of each question because there are not explanation of question number.
Author response
We have now improved the explanation of what is shown in this graph (page 11)
3. Fig.3: Why did several data have no difference between blue and red column.
Author response
In the generalised ordered logit model shown in Figure 3, the two coefficients for each attribute are tested (Wald test) to determine whether there is a statistically significant difference between them. If not, the model includes them as being the same (ie having a consistent effect on the probability of choosing each of the outcomes). Where the two coefficients are significantly different (p<0.05), they are retained as different coefficients in the model, and those p-values are shown on the graph. We have added some further explanation to the text (page 16), which we hope that you will find is an improvement.
Round 2
Reviewer 3 Report
This manuscript is well revised.