There is a great need for good intervention studies that evaluate the factors which are significant determinants for injudicious antibiotic prescribing, from different hospital settings and for a variety of patient populations. Farinas et al. present a very thorough and quite large study from three wards in one Spanish tertiary hospital where one has evaluated the impact of infectious diseases specialist’s recommendations to change antibiotic prescriptions by treating physicians. The study has an interesting design but raises several methodological concerns which need clarification. The article should be tightened; more specifically could the Results section be improved in readability and tables be used more sparingly. Overall, the text should be subject to some improvement in language.

Main comments 1. When an antibiotic regimen in the intervention group was deemed adequate in the first place, what is then the rationale to analyse adherence to infectious diseases consultants (IDCs) recommendations? Or put another way: to what extent could any corrective recommendations be given, for already appropriately prescribed antibiotics?

1. Throughout the article, the antibiotic substances that were prescribed during the study period are not being accounted for. Were they not recorded? Antibiotics may be presented as proportions in the groups that were subject to analysis, and subclasses of antibiotics may even represent important independent variables in the multivariate analyses. The authors may consider tabulating antibiotic subclasses to show the proportions subject to correction, which may appear as interesting as repeated presentations of e.g. co-morbid conditions.

Methods Population Why were the General Surgery, Pulmonology and Endocrinology departments chosen? Was it because of any historically high consumption of (undesirable) antibiotics, or any other indication that these were units of deviating prescription practices? How many had the respective specialities, and what was the average length of stay?

Randomization Although no statistician, I find it questionable to state that for comparison between proportions, a much lower p-value of < 0.001 was defined “due to multiple comparisons…” (line 135-136). The authors may want to elaborate more on the rationale for this. The implication is that for several comparisons of strata, no significant differences are being reported between the intervention and non-intervention group. Applying the conventional p> 0.05, in Table 1 the “Coming from healthcare facility (p=0.005), “Previous hospitalization <90 days” (p=0.002) and not least “Surgical ward of admission” (p=0.001) would have to be reported significantly different in addition to the two variables mentioned in the text (lines 157-158). The intervention and non-intervention group thus seem surprisingly dissimilar on several baseline clinical characteristics, which could raise a question about the randomization procedure. Especially puzzling is the significant (p= 0.001, RR 0,87, CI95 0.80-0.95) difference in “Ward admission (surgical)” (Table 1), when the randomization was stratified by units and one was clearly general surgical while the other two were medical. Please comment.

Classifications and definitions used

1. It is not clear to me if any local antibiotic guideline was available in the hospital at the time of the study? There are references (No. 10-12) to three American (IDSA) guidelines for intraabdominal, SSTI, and community-acquired pneumonia, but it is not stated whether or not these recommendations were translated and implemented, e.g. if a local antibiotic guideline existed for use by all treating physicians. The authors only state (lines 101-102) that the IDCs used these (IDSA) treatment guidelines “adapted to local data on antimicrobial susceptibility” when making their recommendations.

2. No mention is made of the qualifications and clinical experience of the IDCs making the recommendations in this study. Although, as stated by the authors (lines 277-279), a limitation is the potential lesser impact of recommendations inserted as notes in patients records, one should in my opinion describe the IDCs in the Methods section. Furthermore, it would be interesting to know how many physicians actually responded to the offered oral consultation (line 105). The reference No. 9 “Why don’t they listen?” by Lo et al is accompanied by an editorial in CID where Tenenbaum reflects upon the importance of direct personal communication on adherence, which the authors may find it worth to elaborate on.

3. Appropriate use of antibiotics and adherence to recommendations are the main focus of the article, and there are two issues I would like to rise.

A. It would be highly desirable if the authors could elaborate more on the classification they have used in their evaluation of appropriate antibiotic use. In lines 118-119, two references are mentioned; no. 13 by Kunin et al (Ann Int Med 1979) give the classic criteria, but treatment duration is not mentioned as a criterion, and the emergence of multiresistant bacteria was a lesser concern at the time than an increase in Gram-negatives and escalating drug costs. The second cited the article (14. Erbay et al) expands the list of criteria from Kunin’s five to twelve, a quite complex classification with several subsets of inadequate antibiotic choices and dosing errors. As for the duration of the antibiotic regimens, these authors emphasise “prolonged antibiotic prophylaxis”.

• Specific questions:

i) Which “parts” of the two cited classifications were used by the authors, more specifically? Did the authors ever consider using the classification published by Inge Gyssens ( JAC 1992,30(5):724-7? (It is widely used and seems to incorporate all evaluation aspects with a sort of pragmatic simplicity).

ii) In evaluating treatment duration, from which guideline(s) were the authors able to draw any conclusion of appropriateness? The cited IDSA guidelines are mostly vague on the number of days of treatment, and rightly so: very little high-quality documentation exists on duration for most infections and the physician in charge must use her clinical judgment. Furthermore, since assessments of adherence were done at days 7 and 10, how were the authors able to conclude on appropriateness when quite a number of serious infections in hospitals are being treated for longer than this?

iii) The variable “antibiotic prophylaxis …administered for more than 24 hours” (lines 111-112) seems incorrectly termed since inclusion in the study required that IV antibiotics had been administered for three days, i.e. several “prophylactic antibiotics > 24 hours” regimens may have been discontinued between 24 and 74 hours. Furthermore, it is not clear to me if (surgical) antibiotic prophylaxis in this context is a real prolongation of antibiotics preoperatively administered, according to written guidelines for surgical prophylaxis, or just another term for any (therapeutic) antibiotics given to a patient after surgery where no “definite infection” could be identified. Which would equal a classification of Please clarify.

B. The designated adherence classification is somewhat complex, especially when it comes to applying limits in the percentage of the “correct” treatment duration (see above). However, my main concern is that the classification for “partial adherence” (lines 123-125) seems very wide: the treating physician should comply with the recommendation within 24 to 48 hours, but if this is not fulfilled (“and/or”) the deviation from recommended dose (>20%) and duration (> ±30%) may in fact be unlimited and implemented at any time during the evaluation period. If this is correctly perceived, I would think that “partial adherence” is closer to “non-adherence” and should thus be less emphasised, perhaps even considered discarded as a separate group. Please comment.

1. The comorbidity index by Charlton and severity classification by McCabe and Jackson are used in the article but no references are made to these methods and they are seldom part of routinely collected clinical parameters (at least in any hospital I know of). Specific questions: i) How were these variables collected, did you prospectively classify the patients at the time of inclusion (and in that case, by whom and how); or were they retrospectively determined from i.e. the discharge ICD codes or some other routine hospital system? ii) Since the Charlson index has undergone several revisions since the original work published in 1987, it would be appropriate to specify the version used in the Methods section. iii) I am a bit annoyed by an apparent mismatch between the quite large proportions of patients with underlying comorbidities, both in the intervention and non-intervention group (Table 2) and the relatively low mean Charlson indices presented in Table 1 (2.43 and 2.09, respectively). Largely speaking, Table 2 state that about 4 out of 10 patients had diabetes, malignancies, heart disease, chronic respiratory or renal insufficiency and consequently, many patients would have to have several of these conditions simultaneously. Each of these score at least one point in any version of Carlson index; furthermore, age 65 years alone (the average) would assign 2 points to a patient. Please explain.

Results, general comments I find the Results section somewhat difficult to read and it should at least be reorganised with appropriate subheadings. Attention should also be paid to duplication of information in the text that may be read from the numerous tables. The tables would perhaps be more informative if compressed into fewer numbers, where only key (significant) findings are presented. A separation in two table designs could perhaps also be made for comparison of intervention/non-intervention and the comparison of adherence groups, respectively. See also my remark about significance levels, above.

Discussion

Language In general, the text needs to be improved and made more concise and precise. Some examples of English language that need to be improved upon (italicised by me): Line 103: “Recommendations in written (in specially designed study forms) were included…” Line 173: “- infection was definitely diagnosed…” Line 245: “..are within the frame of 53% to 90% reported in the literature9 or nearly… (for complete adherence, incidentally, the percentage was 34.9 which seem far from “nearly”)