Communication training and the prescribing pattern of antibiotic prescription in primary health care.

BACKGROUND: The treatment of upper respiratory tract infections (URTIs) accounts for the majority of antibiotic prescriptions in primary care, although an antibiotic therapy is rarely indicated. Non-clinical factors, such as time pressure and the perceived patient expectations are considered to be reasons for prescribing antibiotics in cases where they are not indicated. The improper use of antibiotics, however, can promote resistance and cause serious side effects. The aim of the study was to clarify whether the antibiotic prescription rate for infections of the upper respiratory tract can be lowered by means of a short (2 x 2.25h) communication training based on the MAAS-Global-D for primary care physicians.
METHODS: In total, 1554 primary care physicians were invited to participate in the study. The control group was formed from observational data. To estimate intervention effects we applied a combination of difference-in-difference (DiD) and statistical matching based on entropy balancing. We estimated a corresponding multi-level logistic regression model for the antibiotic prescribing decision of German primary care physicians for URTIs.
RESULTS: Univariate estimates detected an 11-percentage-point reduction of prescriptions for the intervention group after the training. For the control group, a reduction of 4.7% was detected. The difference between both groups in the difference between the periods was -6.5% and statistically significant. The estimated effects were nearly identical to the effects estimated for the multi-level logistic regression model with applied matching. Furthermore, for the treatment of young women, the impact of the training on the reduction of antibiotic prescription was significantly stronger.
CONCLUSIONS: Our results suggest that communication skills, implemented through a short communication training with the MAAS-Global-D-training, lead to a more prudent prescribing behavior of antibiotics for URTIs. Thereby, the MAAS-Global-D-training could not only avoid unnecessary side effects but could also help reducing the emergence of drug resistant bacteria. As a consequence of our study we suggest that communication training based on the MAAS-Global-D should be applied in the postgraduate training scheme of primary care physicians.

Introduction

The widespread use of antibiotics and the lack of new drug development serve as the main causes for the emergence of drug resistant bacteria [1], limiting the effectiveness of antimicrobial therapy [2]. The rapid increase of resistant bacteria is regarded as one of the greatest threats to global health [3]. Infections with antibiotic-resistant bacteria may cause higher severity of illness, mortality rates, risk of complications, admissions to hospital, hospital length of stay and health care costs [2, 4–7].

Especially, not indicated antibiotic use is considered to be a primary cause of increasing risk of bacterial resistance [8]. Therefore, several initiatives address the improvement of prescribing practices of antibiotics worldwide [9-11]. A prominent example for an irrational use of antibiotics can be found in primary care, where primary care physicians (PCPs) often treat upper respiratory tract infections (URTIs) with antibiotics [12]. URTIs are one of the most common reasons for encounter in primary care and are mostly caused by viral infections, making antibiotic-therapy appropriate for only a small number of high risk patients [13]. However, the treatment of URTIs accounts for the majority of antibiotic prescriptions in primary care [12, 14, 15], although there is very limited evidence for their benefits [16-18]. Besides characteristics of the physicians (e.g., specialty, training, experience), patients (e.g., sex, age, insurance status, comorbidities) and environmental factors (e.g. access to and quality of care), patient knowledge and expectations, as well as the physicians’ assumptions regarding these expectations play a crucial role in the prescribing process [19-21]. Furthermore, evidence strongly suggests that antibiotic prescriptions are associated with a communication problem. Most patients seem to possess insufficient knowledge about the difference between viral and bacterial infections [22]. Due to the patients’ belief that a previously received antibiotic drug cured their infection, their expectations to receive antibiotic therapy when next presenting with URTI symptoms will increase [23]. Additionally, physicians may wrongly assume that the patient will demand antibiotics and preemptively prescribe the medicine [24-26]. Moreover, due to an overload of patients, physicians might not take the time to change the patient’s expectations by explaining the differences between viruses and bacteria in an understandable and effective way [27-29]. Therefore, patient expectations could strongly influence physicians, who are willing to prescribe an antibiotic to maintain a good relationship and to save time [20, 30, 31].

Communication trainings have been found to be effective in decreasing the antibiotic prescription rate [32-37]. Although the benefits of adequate communication skills are well known, they are not part of the postgraduate training scheme of any medical specialty in Germany [38]. In the Netherlands, a mandatory instrument for training and measuring physicians’ communication and medical skills is widely used in under- and postgraduate training [39]. This instrument, named Maastricht history taking and advice scoring list (MAAS-Global), has been recently translated and adapted for use in Germany (MAAS-Global-D) [40].

The aim of this study was to investigate whether a communication training based on the MAAS-Global-D can reduce the rate of antibiotic prescribing for URTIs. Since the expectations of the patients and their perceptions of the physicians are subjective and might differ between patients, we additionally evaluate the intervention effect by the patient’s age and sex to increase the insights of the communication effect.

Materials and methods

Data source

This study was based on the analysis of routine data of the years 2013 to 2016 from the Association of Statutory Health Insurance Physicians (ASHIP) of the federal state Schleswig-Holstein, located in Northern Germany. The ASHIP is in charge for the reimbursement of services that are provided to patients within the statutory health insurance system. The dataset covers 85% of the population and 83% of the PCPs of Schleswig-Holstein [41, 42]. The URTI cases were identified by the target-diagnoses of acute bronchitis, sinusitis and pharyngitis (classified by the International Classification of Diseases, version 10 (ICD-10) codes: J01.-; J02.-; J20.- [43]). We concentrate the analysis to these diagnoses, since only in some cases the use of antibiotics is suggested by respective guidelines within these diagnoses. For cases of acute bronchitis (J20) an antibiotic prescription is indicated for elderly patients as well as for those with a severe cardiac or respiratory disease or a congenital or acquired immunodeficiency [44]. In the case of acute pharyngitis (J02), the indications for an antibiotic therapy are: pharyngitis due to group A streptococcus bacterial infections (GAS pharyngitis), scarlet fever, peritonsillar abscess, a suspected serious illness or clinical worsening as well as consumptive diseases, immunosuppression and acute rheumatic fever in the personal or family history [45]. For acute sinusitis (J01), an antibiotic therapy should be considered for patients with specific risk factors, as well as complications such as severe headache, facial swelling, lethargy and acute exacerbation of recurrent sinusitis. Moreover, severe pain and an increased inflammation score complaints in the course of the disease and with fever above 38.5°C [46].

Since the antibiotic prescriptions have been inferred based on the visit diagnoses, we excluded cases with additional diagnoses. This includes the presence of diagnoses regarding puerperium/pregnancy (O00-O99), further (bacterial) infections (A00 to A37, A39 to A79, J15, J17, J18) or chronic diseases (I50, J44, J45, C00 to C75). If the diagnosis had been made several times or more than one diagnosis had been made from the three groups (J01, J02, J20), the corresponding cases were also excluded. To increase the comparability of the included cases and, thus, minimize a potential estimation bias of the communication training effect, only cases of patients that were older than 18 years are included in the analysis.

Recruitment and inclusion criteria

All primary care physicians in private practices, working in a contract with statutory public health insurance and with a work experience of at least five years, who have patients with at least one of the target-diagnoses between 2013 and 2015 were considered for the intervention. In total, 1554 (76%) primary care physicians of Schleswig-Holstein have been invited by letter to participate in a study named “Effects of communication training with the MAAS-Global-D on the prescription of antibiotics for respiratory infections”.

Study design and estimation strategy

The intervention and the previously planned randomized controlled trial (RCT) has been described by Hammersen et al. [47] (Trial registration: DRKS00009566). The study was originally designed to consist of two interventional study arms. In addition to the communication training, the second intervention group received an educational introduction into the use of and online-access to EbMG online (Evidence–based Medicine Guidelines) [48]. This point-of-care online tool provides further information material on the prescribing of antibiotics for uncomplicated respiratory infections. Since the inclusion rate was lower than initially expected, both intervention groups have been consolidated. Furthermore, a comparison between the pooled intervention group and the control group did not yield significant results due to a lack of power because of the small sample size. Instead, we formed a control group from observational data and applied a combination of difference-in-difference (DiD) estimation and matching approach that is considered to reproduce the results of RCTs very well under certain assumptions [49]. For instance, under the assumption that the average outcomes for the intervention and control group would have followed parallel trends over time before intervention, the DiD estimator identifies causal effects by contrasting the change for the intervention and control groups in pre- and post-intervention outcomes [50]. However, the assumption of parallel trends might be implausible in our setting. For instance, if physicians recognized a too high antibiotic prescription rate for URTIs, they presumably tried reducing it. Therefore, they might have been more likely to respond to the training offer that advertised a reduction of the prescription rate through improved communication skills. Consequently, the evolutions of the prescription rates were suspected to differ between the intervention and control group if the control group, as in this case, had not been built upon a controlled randomization. An alternative identifying assumption is that the potential outcomes are independent of intervention status, conditional on past outcomes and covariates [51]. By means of balancing the intervention and control group according to pre-intervention outcomes and covariates all potential outcome trends are perfectly aligned and the DiD estimates can be interpreted as causal effects [49, 52]. However, recent studies showed that the combination of DiD and matching might also deliver biased estimates [53]. In order to enhance the robustness of our findings and minimize the risk of estimation bias we compared DiD estimates from both unmatched and matched (on pre- intervention outcomes) data [54].

In the search for relevant variables determining the decision to prescribe an antibiotic for a specific URTI case (our dependent variable) we first estimated a multi-level random effects logistic regression model based on case-, patient- and physician-level data of the pre-intervention period. A logistic regression model was chosen to account for the binary nature of the dependent variable. Moreover, the logit model showed computational merits and, unlike the probit model, it did not suffer from any convergence failures. Random effects were specified on the physician level to account for intra-physician variability [55]. In a second step, we aggregated the data on the physician level and matched the intervention and control groups according to aggregated pre- intervention outcomes and covariates by means of entropy balancing [56]. Based on the balanced data, in the third step, we estimated a multi-level random effects logistic DiD regression model using the weights of the physician-level from entropy balancing. Alternatively, we also specified fixed physician effects in the pre-intervention analysis and the DiD regression models. For all models the results between fixed and random effects models are very similar and we conclude there is no correlation between the explanatory variables and the individual effects. The physicians, who had previously been selected to the control group were excluded from the third step of the analysis, since we could not rule out that their prescribing behavior might have been affected by the cancellation of participation in the communication training.

The study was approved by the ethics committee of Luebeck University before the recruitment of participants on 9 June 2015 (number of approval: 15–139). Statistical analyses were performed with STATA 15 (StataCorp LLC, College Station, TX, USA).

Intervention

The intervention group received a communication training with an interactive workshop character (two times 2.25 hours), which was held at the Institute of Family Medicine in February and March of 2016. It was delivered face-to-face by members of the research team, including an expert in physician-patient communication. The curriculum of the training was derived from the German version (MAAS-Global-D [40]) of the Dutch instrument MAAS-Global [39]. After establishing the relevance and success of physician-patient communication, the participant were provided with information concerning the associated evidence base regarding treatment of URTIs. Furthermore, they learned about the different communicative phases of a consultation, corresponding communication skills as well as general communication skills for the whole consultation (e.g. adequate provision of information, structuring and empathy, shared decision-making).

Measurements

As the outcome variable we considered the binary choice whether an antibiotic was prescribed for a URTI case. The selection of potential determinants serving as control variables in both the pre- intervention and the DiD regression analyses was based on related previous literature [25, 57]. They can be classified into three categories: (i) case related (year, quarter, diagnosis and its certainty, emergency service), (ii) patient specific (insurance status, age, sex) and (iii) physician characteristics (age, sex, number of URTI-patients in that quarter). Seasonal effects and a general trend in the prescribing pattern were considered by respective dummy variables identifying the quarter and the year of the consultation, respectively. As the prescription rate might differ between the considered diagnoses, we introduced dummy variables for sinusitis and pharyngitis with bronchitis serving as reference. According to the German coding policy, primary care physicians are required to designate their diagnoses as validated (certain) or suspected (cases without an established definite diagnosis). We controlled for the cases with a certain diagnosis by including a respective dummy variable. Further, we distinguish whether the patient visited an emergency care center during the out-of-hours care (emergency service). Demographic variables of the patient were comprised of the sex (sex = 1: female), the age and the insurance status (normal, family or retired). The age was grouped by respective dummy variables for patients aged <35, 35–65, 65+ to allow for nonlinear age effects. In Germany, the insurance status signifies whether the patient is ordinary insured, retired or coinsured. Children and grandchildren aged below 25 as well as spouses that are unemployed, not self-employed and are not exceeding an income of EUR 450 per month are coinsured with an ordinary insurance member. The considered age and insurance status based clusters reflect different stages of life that might go along with different expectations about the treatment. At the physician level, we controlled for the specialty, since primary care physician workforce in Germany consists of general practitioners, physicians in general internal medicine and a declining number of practitioners without special training in primary care (12%). Previous studies have shown substantial differences in prescribing behavior between general internists and general practitioners [58]. Further, we considered the age and the sex of the physician. Finally, to approximate the workload of the physician’s practice we included the number of total URTI patients in the respective quarter. The logarithmic function to this variable accounts for unequal variation.

In the intervention analysis, the DiD dummy variable identifies observations of the intervention group for the post-intervention period. To control for any other time-invariant differences between both groups a dummy variable trained is additionally included.

Results

In the first part of the analysis (pre-intervention), the sample of the pre-intervention analysis (2013 to 2015) consisted of 315,752 adult patients with 476,260 cases from 2,189 PCPs. For the second part of the study, we invited 1,554 PCPs in SH to participate in the training. The group of interested participants has been divided randomly in a control and an intervention group with each 17 PCPs. Due to a lack of power, we alternatively form the control group from observational data. Do to so, we excluded the prior control group physicians (n = 17) and practitioners without special training in primary care, since they are lacking in the intervention group (n = 198). Moreover, 492 PCPs are not considered because they are not treating URTIs in each of the considered years, for instance since they are entering or leaving the ASHIP payment system during the study period. Finally, the intervention/control group in the intervention analysis consisted of 17/1,460 PCPs with 1,807/170,683 patients with 2,284/235,355 cases in the pre-intervention period (2013:q1 to 2015:q4) and 585/61,755 patients with 698/75,167 cases after the intervention (2016:q2 to 2016:q4) (Fig 1).

Fig 1

Flow chart.

Pre-intervention analysis

The mean values of the considered variables in the pre-intervention analysis and the regression results are shown in Table 1. An antibiotic was prescribed in half of the considered cases (49%).

Table 1

Multilevel logistic regression analysis of the pre-intervention period of prescribing an antibiotic.

Variable		means	(1)	(2)
Dependent variable
antibiotic prescription (= 1)		0.49
Case characteristics
quarter
2nd quarter		0.21	-0.01	-0.01
3rd quarter		0.17	-0.07**	-0.07**
4th quarter		0.26	-0.08**	-0.08**
(Reference: 1st quarter)
year
2014		0.32	-0.03**	-0.03**
2015		0.33	-0.14**	-0.14**
(Reference: 2013)
diagnosis
sinusitis (J01)		0.21	-0.19**	-0.19**
pharyngitis (J02)		0.29	-0.18**	-0.18**
(Reference: bronchitis (J20))
certainty
certain diagnosis		0.99	0.38**	0.38**
type of service
emergency service		0.03	0.40**	0.40**
Patient characteristics
Insurance status
Family insured		0.12	0.11**	0.11**
Pensioners insured		0.16	0.16**	0.17**
(Reference: ordinary insured)
Patient demographics
Patient aged 35–65		0.52	0.33**	0.45**
Patient aged 65+		0.14	0.28**	0.39**
(Reference: < 35)
Female patient		0.59	0.09**
(Reference: male)
sex-age interactions
Female patient aged <35		0.19		0.22**
(Reference: male<35)
Female patient aged 35–65		0.31		0.02
(Reference: male aged 35–65)
Female patient aged 65+		0.09		0.04
(Reference: male aged 65+)
Physician characteristics
PCP specialty
PCP without special training		0.13	0.16	0.16
General Internist		0.18	0.00	0.00
(Reference: GP)
PCP demographics
Physician age		55.11	0.00	0.00
Female physician		0.31	-0.04	-0.04
(Reference: male)
PCP workload
log(#URTI-patients)		3.60	0.15**	0.15**
intercept			-1.19**	-1.27**
σRE2 (Variance of random effects on physician level)			0.90**	0.91**
Intra-class correlation (in %)			21.57	21.58
Log-Like			-288,587	-288,480
Akaike Info Criterion (AIC)			577,216	577,006
R2-MacFadden (in %)			12.56	12.60

The first column presents sample means. The other columns display the estimated regression coefficients. Based on 476,260 observations (315,752 patients from 2,189 primary care physicians). Estimated by means of Maximum Likelihood. Significance levels: * 5%, ** 1%.

The results of two logistic regression models with specified random effects on the physician level are shown in the third and fourth column of Table 1. In both models, the estimated intra-class coefficients (21.6%) suggest that conditional on the covariates, almost one quarter of total variation in antibiotic prescription could be explained by the individual physician’s practice style. The estimated regression coefficients indicated that patients aged over 35 years were significantly more likely to receive an antibiotic prescription than younger patients. The strongest effect was achieved for patients aged between 35–65 years. Female patients were also more likely to receive an antibiotic. The interaction effects between the patients’ gender and age groups in Model (2) signified that the gender difference only exists for patients younger than 35 years. As indicated by the smaller Akaike Information Criterion (AIC), the fit of the model was significantly improved, leading to our final model that was considered for the intervention analysis.

Matching

The entropy balancing was applied to match physicians of the intervention group with physicians of the control group in the pre-intervention period. In addition to the control variables, the pre-intervention prescription rates served as conditional variables used in the matching. Case- and patient-level variables were aggregated on the physician level. Table 2 shows the means of the variables for the intervention as well as the matched and non-matched control group. Further, the differences between intervention group and unmatched controls as well as the share of missing observations are shown for each variable.

Table 2

Means of aggregated variables before intervention.

variables		Intervention group	Control group		Difference between (a) and (b)	Share of missing observations (in %)
			Un matched	matched
		(a)	(b)	(c)
Outcome: Prescription rate (in %)
2013		51.5	46.5	51.5	5.0	3.74
2014		48.3	44.7	48.3	3.6	2.01
2015		47.6	43.2	47.6	4.4	2.64
Number of URTI-patients
2013		66.5	88.4	66.4	-21.9	3.74
2014		65.1	79.0	65.0	-13.9	2.01
2015		69.3	82.3	69.3	-13.0	2.64
Share of cases (in %)
quarter
2nd quarter		22.3	21.8	22.3	0.5	0
3rd quarter		18.6	18.1	18.6	0.5	0
4th quarter		27.6	26.2	27.6	1.4	0
(Reference: 1st quarter)
diagnosis
sinusitis (J01)		14.8	19.7	14.8	-4.9	0
pharyngitis (J02)		47.2	41.3	47.2	5.9	0
(Reference: bronchitis (J20))
certainty
certain diagnosis		99.3	97.9	99.3	1.4**	0
service-type
Emergency services		6.6	5.5	6.6	1.1	0
Patient demographics
Patients aged 35–65		55.0	51.5	55.0	3.5	0
Patients aged >65		13.6	14.0	13.6	-0.4	0
Female patients		59.4	60.6	59.4	-1.2	0.02
sex-age interactions
Female patients aged 35–65		32.9	31.3	32.9	1.6	0.02
Female patients aged >65		7.9	8.5	7.9	-0.6	0.02
Insurance status
Patients family insurance		8.7	12.4	8.7	-3.7*	0
Patients pensioners insurance		15.6	16.1	15.6	-0.5	0
Physician characteristics
PCP specialty
General Internists (in %)		23.5	26.4	23.5	-2.9	0
(Reference: GP)
Female physician (in %)		23.5	37.9	23.5	-14.4	0
(Reference: male)
Physician age		54.3	53.6	54.3	0.7	0.00
Number of PCPs		17	1,460	17

The first three columns present means of selected variables used for the matching before intervention for trained controls and matched controls, respectively. The last column displays the differences between intervention and control group before matching. Significance levels: * 5%, ** 1%. Patient variables are aggregated on physician level by summing up (Number of URTI-patients) or computing as shares of cases.

The intervention group is characterized by higher average prescriptions per physician in comparison with the control group. This hints for a selection of the participants in the intervention group due to their pre-intervention outcome. Furthermore, the change over time differed between both groups, underlining that the assumption of parallel trends might not hold. The average number of patients was higher in the control group. However, none of the differences were significant, except for the fraction of patients with a certain diagnosis and family insurance. This might have been due to the low number of observations at the physician level in the intervention group (n = 17). Nevertheless, after applying the reweighting approach based on entropy balancing the means in the control group equaled the means in the intervention group.

Univariate DiD analysis

To assess the sensitivity of the DiD analysis due to model specifications and the balancing we started presenting univariate DiD estimates (simple mean comparison) based on unmatched and matched sample data in Table 3. Neglecting physician-specific effects and other covariates, the reduction in the overall prescription rate of the intervention group between the pre-intervention and post-intervention period was 11.2%. For the control group a reduction of 4.7% could be detected. The difference between both groups in the difference between the periods is the DiD estimator, which is -6.5% [95% CI: (-10.7%; -2.3%)], and significant. Reweighting the observations of the case-level by the entropy weights on the physician-level increased the prescription rate of the matched control group to 52.9%. This was also slightly smaller than the rate of the intervention group (55.4%), which might be, because the matching was done at the physician level and not at the case level. However, the DiD estimate for the matched sample was rather similar (-6.1% [95% CI: (-12.0%; -0.2%)],) and also significant. Concluding, both univariate DiD estimates suggest a significant reduction of antibiotic prescriptions after the communication training.

Table 3

Univariate difference-in-difference analysis of the communication training on the antibiotic prescribing behavior.

	Prescribing rate (in %)
	Intervention Group	Control Group		Difference-in-Difference
		unmatched	matched	unmatched	matched
Before	55.43	47.27	52.86
(2014–2015)	(n = 2284)	(n = 235355)	(∑wi≈2282)
After	44.27	42.61	47.80
(2016)	(n = 698)	(n = 75167)	(∑wi≈736)
Difference	-11.16**	-4.65**	-5.07*	-6.51**	-6.10*
p-value	<0.001	<0.001	0.017	0.003	0.043

313,504 observations (234,723 patients from 1,477 general practitioners). Significance levels: * 5%, ** 1%. Matching is based on Entropy balancing using the variables listed in Table 2 and w denotes the Entropy balancing weights.

Multilevel DiD regression analysis

To take into account the control variables and the random effects on the physician-level, we estimated the specification of Model (2) based on the extended data set, as well as the DiD and training dummy variable. Table 4 shows the estimated DiD effects and the moderation effects. To ease the interpretation of the estimated DiD coefficient, the marginal effect on the prescription rate was also shown for the direct effects.

Table 4

Multilevel logistic regression analysis of the difference-in-difference effect of the communication training on prescribing an antibiotic.

	(3)	(4)	(5)	(6)
matching	no	yes	no	yes
trained	0.15	-0.08	0.14	-0.08
DiD	-0.31**	-0.28*	0.19	0.14
95%-CIa of DiD	[-0.50, -0.12]	[-0.50, -0.05]
MEb of DiD (in %)	-6.34**	-6.44*
95%-CIa of ME (in %)	[-10.31–2.37]	[-11.67, -1.22]
Odds Ratio DiD	0.73**	0.76*
95%-CI of OR	[0.61, 0.89]	[0.60, 0.95]
Interaction effects
DiD*Pat age (35–65)			-0.52	-0.49*
DiD*Pat age (65+)			-0.33	-0.26
DiD*Fem pat (<35)			-0.73*	-0.65**
DiD*Fem pat (35–65)			-0.13	-0.03
DiD*Fem pat (65+)			-0.00	0.07

313,504 observations (234,723 patients from 1,477 general practitioners). Significance levels: ** 5%, *** 1%.

a Confidence interval

b Marginal Effect. Estimated coefficients of the control variables and the variance of the random effects are not shown. Matching is based on Entropy balancing using the variables listed in Table 2.

All specifications obtained a significant reduction of the prescription rate due to the intervention. There were no substantial differences between the estimates of the matched and unmatched sample. The marginal effects were close to the estimated univariate DiD effects.

The results of a moderation effect of the DiD effect by the age and sex of the patients are also shown in Table 4. They suggest that the intervention had a significantly stronger effect on the treatment of female patients aged below 35.

Discussion

In this study, we estimated the effect and its moderations of a communication training based on the MAAS-Global-D instrument on the antibiotic prescription rate of primary care physicians for the treatment of upper respiratory tract infections. Since the control group was formed from observational data, we applied a combination of difference-in-difference estimation and statistical matching based on entropy balancing to estimate the intervention effect. Relevant variables for the matching were selected after estimating a multi-level logistic regression model for the antibiotic prescribing decision, based on case-, patient- and physician-level data of the pre-intervention period in the first stage. In the second stage, the same model was estimated, based on matched data and extended by the intervention period and DiD specification.

During the pre-intervention period, an antibiotic was prescribed in almost half of the considered cases. This relatively high number of antibiotic prescriptions is also observed in related studies [57, 59]. In both groups (intervention and control), the prescription rate slightly decreased over time. This is similar to the declining trend of general antibiotic use in other countries [60] and might be explained by an increased awareness of antimicrobial resistance [61], e.g. due to successful antibiotic stewardship programs as the German Strategy against Antibiotics Resistance [62]. The estimated intra-class coefficient of the multilevel regression model shows that the individual physician’s practice style explains about 22% of the total variance in antibiotic prescription and is similar to the results of related studies [25, 35, 63]. It suggests the prospect of a successful reduction in the prescription rate by changing the individual physician’s prescribing behavior. Most of the observable characteristics of the physician do not explain the variance in prescribing behavior. Only the number of URTI patients (serving as a proxy of the physician’s workload) increases the probability of antibiotic prescription. This effect underlines the hypothesis that insufficient communication determines the antibiotic prescription. It is more complicated for physicians lacking sufficient time for the consultation due to an overload of patients, to change the patients’ expectations [28, 64]. A similar mechanism might explain the positive association of emergency service and the antibiotic prescription probability. In Germany, PCPs face an overload of patients, especially when providing out-of-hours care in emergency service [65]. Patients visiting the emergency service for respiratory complaints might be more severe and therefore might have a strong expectation of receiving an antibiotic [66]. The expectations might also differ between patients, as suggested by the estimated effects of the patient characteristics.

Patients above the age of 35 years receive a significantly higher number of antibiotic prescriptions than younger patients do. For patients belonging to a higher-risk group (e.g., elderly patients) respective guidelines suggest the use of antibiotics in some cases [44-46]. Therefore, the application of guidelines cannot explain the lower prescription rates for patients aged above 65 in comparison with patients aged between 35 and 65 (Model (1): 0.28 vs. 0.33). Differences between the patients’ expectations in the age-groups below and above 35 respectively, might be more likely to serve as an explanation. Work pressure and other related stress cause patients to desire rapid relieve from symptoms and cure of their sickness [67]. The perceived importance of the patient’s job promotes the decision to prescribe an antibiotic [30, 68]. This might explain that the antibiotic prescription rate considerably exceeds the clinically justified amount for young and middle-aged adults with respiratory infections in the UK [69].

Our results further suggest that women receive more frequent an antibiotic prescription. Women are more likely to visit a physician for URTI than men [57] and are found to be more skeptical towards the physician’s suggestions [70]. This patient group might combine higher expectations and wariness that might lead to additional communication requirements. This hypothesis is in line with the higher antibiotic prescription rate that is observed for female patients in our data. However, the underlying mechanisms were not aim of our research focus and is, therefore, subjected to future research. Similar to other studies, the gender gap vanishes with the increasing patient-age [71]. Only female patients below the age of 35 receive a significantly higher number of antibiotic prescriptions. This result might indicate that the communication problem is mostly pronounced for the treatment of this group of patients. In the following, we discuss the effects of the communication training on the antibiotic prescription probability.

Intervention analysis

We applied different approaches and specifications to robustly estimate the effect of the communication training on the antibiotic prescribing behavior. The univariate approach estimates an 11-percentage-point reduction of prescriptions for the intervention group after the training. This result is very similar to a related study [72]. All our approaches (univariate and multivariate) estimate a decrease of around 6.5 percentage-points in the prescription probability of the trained physicians. These robust estimates are in line with the findings of other related studies applying RCT methodology [32, 35, 73]. The moderation analysis confirms that the effect of the communication training is stronger for the treatment of patients marked by a larger communication problem. The impact of the training on the reduction of antibiotic prescription is significantly stronger for the treatment of young women. Thus, physicians with improved communication skills might be able to better address the potentially higher expectations of young female patients to receive an antibiotic therapy and their wariness towards the physician’s suggestions [70].

As argued by Fritz and Holton [74], the lack of trust in the patient-doctor relationship enhances the likelihood of overprescribing. A patient trusting in the physician’s clinical judgment, can be reassured to accept non-prescribing [75]. Furthermore, secured trust between the patient and physician could reduce the probability of the physician to misperceive the patient’s expectation to receive antibiotic treatment. To establish a trustful relationship it is important for the patient to recognize the physician’s trust in them and believe that the physician acts in their best interest [76]. Signals of trustworthiness are given by verbal and nonverbal communication and serve to establish patients’ trust, and, thus, influence the doctor-patient relationship [77]. For this purpose, the MAAS-Global-D might be a promising tool to improve effective communication since both verbal and nonverbal communication skills are part of the training. To comprehend emotions as well as feelings and to react adequately, the MAAS-global-D-manual proposes the physician to render the feelings expressed by the patient during the consultation either in words or nonverbally [39]. Trust is considered for most patients to be an integral part of an ongoing relationship with a physician [78]. An increased continuity of care enables, on the one hand, physicians to better evaluate the patient’s expectations of receiving an antibiotic by the more intimate knowledge of their living conditions. On the other hand, patients can build up a deep understanding of appropriate antibiotic use and will change their expectations permanently. The findings of Robert et al. [79] suggest that receiving information about antibiotics from family physicians is usually not associated with an increased knowledge of the patients. A trustful and continued relationship might be helpful for physicians to provide information about the use of antibiotics, and to improve knowledge about antibiotics especially among target groups [79, 80]. As we found in our pre-intervention data analysis, one specific target group consists of young female patients.

Limitations and strengths

The study estimated the effects of a communication training for primary care physicians on the antibiotic prescription rate for infections of the upper respiratory tract and its moderation by age and gender of the patients.

The study has strengths as well as limitations. In contrast to the previously planned randomized controlled trial [47], in this study we formed a control group from observational data. In contrast to the control group physicians the members of the intervention group did know that their prescription data would be analyzed for the periods before and after the training. Therefore, we cannot exclude that behavior change in the intervention group is due to the physicians’ awareness of being under observation rather than solely due to the intervention (Hawthorne effect) [81]. However, since we considered the data of the physicians up to one year after the training, we do not believe that this effect is responsible for persistent behavior changes. Further, the approach that has been applied to estimate the intervention effect is more sophisticated and is, thus, more susceptible to misspecification than an RCT [82]. To minimize the risk of biased estimates we applied several alternative approaches (univariate, multivariate, matching, no-matching) and specifications (fixed and random effects) as robustness checks. All estimated effects of the intervention are very similar. Therefore, we believe that misspecification is not a big issue here.

A strength of this study is that it relies on routine data collected from all primary care physicians in a specific region of Germany. The relatively large number of physicians of the (matched) control group (n = 1,460) might ensure a higher external validity of our findings than the rather small sample sizes of other related studies applying an RCT [32, 33, 35, 36]. However, the small number of the intervention group highlights the problem to convince PCPs to participate in intervention studies [83, 84]. Another reason for the low response rate might have been rooted in the PCPs’ (who already faced an overload of patients) concerns that improved communication skills would prolong the consultation, although so far, there is no evidence to support this claim [85].

While on the one hand, the focus on the federal state of Schleswig-Holstein constrained the representativeness of the findings, it on the other hand also reduced practice variations based on regional differences and state-specific regulations [86]. The analyzed moderation of the patient’s age and gender on the communication training effect further increased the insights of antibiotic prescribing behavior. In line with the findings of our pre-intervention data analysis, our results suggest that improved communication skills are mostly effective in cases where the underlying communication problem is particularly pronounced due to high expectations of the patient to receive an antibiotic or due to the physicians’ perceptions. To clarify the moderating role of expectation and its perception for the communication training effect on antibiotic usage future research should include direct measures of these variables [25].

Conclusion

In this study, we estimated the effect and its moderations of a communication training on the antibiotic prescription rate of primary care physicians for the treatment of upper respiratory tract infections, i.e. acute bronchitis, sinusitis and pharyngitis. The short communication training has been based on the MAAS-Global-D [40], the German version of the Dutch instrument MAAS-Global [39]. Since the control group has been formed from observational data, a combination of difference-in-difference (DiD) and matching has been applied to estimate the intervention effect. To minimize the risk of biased estimates we applied several alternative approaches and specifications as robustness checks that all reveal similar intervention effects. The results show that the communication training decreases the prescribing probability by around 6.5-percentage-points for the physicians of the intervention group. For the treatment of female patients aged below 35, the intervention has a stronger impact.

Our results suggest that communication skills implemented via MAAS-Global-D-training lead to more prudent prescribing of antibiotics for URTIs. Therefore, the MAAS-Global-D-training could not only avoid unnecessary side effects but could also help to reduce the emergence of drug resistant bacteria. The instrument MAAS-Global-D has been proven to provide a valid tool for a training of physicians that encourages an effective communication with the patient. In the Netherlands, communication training is an integral part in the postgraduate-training program of general practitioners. A similar communication training based on the MAAS-Global-D could also be applied in Germany, as well as in other countries, where postgraduate training schemes of PCPs lack in training of communication skills. The instrument and the explanatory manual in German language are available for free download [87].

Trial registration

The intervention and the previously planned randomized controlled trial (RCT) has been registered in the German Clinical Trial Register (DRKS00009566).

STROBE statement—checklist of items that should be included in reports of observational studies.

(DOCX)

Click here for additional data file.

STATA software codes.

(ZIP)

Click here for additional data file.

20 Dec 2019

Submitted filename: Reply to Editor and Reviewers.docx

Click here for additional data file.

25 Feb 2020

PONE-D-19-35281

Communication training and the prescribing pattern of antibiotic prescription in primary health care

PLOS ONE

Dear Dr. Strumann,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

==============================

One of the previous reviewer is happy with the answers provided.

The second previous reviewer was not available for this re-submission and a third reviewer has raised some additional minor concerns. Please address them.

==============================

We would appreciate receiving your revised manuscript by Apr 10 2020 11:59PM. When you are ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter.

To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'.

Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.

We look forward to receiving your revised manuscript.

Kind regards,

Martin Chalumeau, MD-PhD

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements:

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at http://www.plosone.org/attachments/PLOSOne_formatting_sample_main_body.pdf and http://www.plosone.org/attachments/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. Thank you for stating the following in the Competing Interests section:

"CS, KG, TE and JS: the authors declare that they have no competing interests.

I have read the journal's policy and the author AS of this manuscript have the following competing interests: AS is an editor of the German version of EbMG. The online access is provided free of charge to the study PCPs by Verlagshaus der Aerzte, Vienna."

Please confirm that this does not alter your adherence to all PLOS ONE policies on sharing data and materials, by including the following statement: "This does not alter our adherence to  PLOS ONE policies on sharing data and materials.” (as detailed online in our guide for authors http://journals.plos.org/plosone/s/competing-interests).  If there are restrictions on sharing of data and/or materials, please state these. Please note that we cannot proceed with consideration of your article until this information has been declared.

Please include your updated Competing Interests statement in your cover letter; we will change the online submission form on your behalf.

Please know it is PLOS ONE policy for corresponding authors to declare, on behalf of all authors, all potential competing interests for the purposes of transparency. PLOS defines a competing interest as anything that interferes with, or could reasonably be perceived as interfering with, the full and objective presentation, peer review, editorial decision-making, or publication of research or non-research articles submitted to one of the journals. Competing interests can be financial or non-financial, professional, or personal. Competing interests can arise in relationship to an organization or another person. Please follow this link to our website for more details on competing interests: http://journals.plos.org/plosone/s/competing-interests

3. We note that you have indicated that data from this study are available upon request. PLOS only allows data to be available upon request if there are legal or ethical restrictions on sharing data publicly. For information on unacceptable data access restrictions, please see http://journals.plos.org/plosone/s/data-availability#loc-unacceptable-data-access-restrictions.

In your revised cover letter, please address the following prompts:

a) If there are ethical or legal restrictions on sharing a de-identified data set, please explain them in detail (e.g., data contain potentially identifying or sensitive patient information) and who has imposed them (e.g., an ethics committee). Please also provide contact information for a data access committee, ethics committee, or other institutional body to which data requests may be sent.

b) If there are no restrictions, please upload the minimal anonymized data set necessary to replicate your study findings as either Supporting Information files or to a stable, public repository and provide us with the relevant URLs, DOIs, or accession numbers. Please see http://www.bmj.com/content/340/bmj.c181.long for guidelines on how to de-identify and prepare clinical data for publication. For a list of acceptable repositories, please see http://journals.plos.org/plosone/s/data-availability#loc-recommended-repositories.

We will update your Data Availability statement on your behalf to reflect the information you provide.

4. Your ethics statement must appear in the Methods section of your manuscript. If your ethics statement is written in any section besides the Methods, please move it to the Methods section and delete it from any other section. Please also ensure that your ethics statement is included in your manuscript, as the ethics section of your online submission will not be published alongside your manuscript.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: I Don't Know

Reviewer #2: Yes

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: No

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: This is a well-written and clear manuscript that discusses an important antibiotic stewardship intervention with potential implications for both public health and clinicians which would be a helpful addition to the literature. Despite the small intervention sample size, I do think it is worth publication, and so have some suggestions for the authors to improve this manuscript. Additionally, I have very little experience with entropy balancing and using observational data as a control and, therefore, would recommend the journal consider a statistical review to ensure the approaches used were appropriate as I am unable to provide a full evaluation.

Suggestions:

• Throughout the paper, the authors reference inappropriate antibiotic prescribing. However, with the diagnoses listed here, there are some cases where antibiotic prescribing is appropriate (at least based on the information provided), so it is too speculative to call all prescribing for these conditions inappropriate. The authors either need to present more data about what proportion of these diagnoses might warrant antibiotic therapy (I am not sure if this data is available in the literature) or be very cautious and explicit in their interpretations.

• Introduction: Overall, a clear and concise summary of the evidence. At times, statements summarizing the literature seem over-generalized or too broad. This is especially pronounced on pages 3-4, starting with the sentence that begins with “Most patients possess insufficient knowledge…”. I believe all that is needed to ameliorate this is softening the language, for example, consider changing “Additionally, physicians wrongly assume to “Additionally, physicians may wrongly assume….”.

• Introduction: I found the wording in the last paragraph of the introduction to be slightly confusing. Consider rephrasing and more succinctly stating the aims of the study.

• Methods: How were antibiotic prescriptions linked to diagnoses? Are indications listed for the prescription or was it inferred based on the visit diagnoses? If the latter, were visits with diagnoses for conditions where antibiotics are appropriate excluded (e.g., UTI, pneumonia)?

• Methods: I saw in the response to previous reviewer comments that tried to address the certainty of diagnosis variable, but I am still confused by this. It might be helpful to include more information about how this is recorded in the claim/medical record.

• Methods: were there any collinearity issues in your model when using both age and insurance states as these reflect similar “stage of life” clusters?

• Tables: I found the tables header in table 1 and 4 difficult to follow. I think it would be clearer to put a more descriptive title than just the model number, or if that is not feasible, include in a footnote the model specifics. This information is in the methods, but the tables need to be interpretable on their own.

• Discussion: I think there might be other reasons for a declining trend in antibiotic use besides awareness of AMR. Consider other secular changes such as visit and disease incidence, antibiotic stewardship practices, and immunizations.

• Discussion, page 16: The points about emergency medicine visits are fair, but the authors should also consider that these visits may be more severe than scheduled primary care visits.

• I found the discussion of why younger adults and women might receive higher antibiotics speculative and unconvincing. Consider revising.

• Limitations: regarding the Hawthorne effect, did intervention clinicians know their data was being tracked? Were they aware of how long their data would be tracked? These are important considerations for understanding the extent of possible bias due to the Hawthorne effect.

Reviewer #2: Manuscript ID PONE-D-19-15846: Communication training and the prescribing pattern of antibiotic prescription in primary health care: A case-control study.

This manuscript reports the effect of a communication training intervention on the antibiotic prescription rate of primary care physicians (PCP) for the treatment of upper respiratory tract infections in adults.

The information reported in this manuscript is interesting and the statistical approach seem appropriate, on a subject which remains a major public health concern worldwide. However several points should be considered.

A) Major comments:

1) You state that your methodology (Difference-in-Difference estimation and matching approach) allows the demonstration of a causal relationship between the intervention and the outcomes. Such quasi-experimental designs seem to be a valuable option in numerous situations where the classic controlled randomized design is not feasible, and is clearly stronger than observational designs such as before-after studies. However, demonstrating a causal relationship remains particularly challenging, and statistical analysis comparing results from some RCTs to such DID finding similar results are probably not a sufficient proof (your reference 52). Even such RCTs often provide discordant conclusions, and sometimes only meta-analysis of well conducted RCTs allow providing the demonstration of a causal effect of an intervention. Moreover, to my knowledge, it is not recognized by international guidelines of level of evidence gradation, such as the GRADE system, that DID with matching is comparable to RCTs in term of level of evidence. Could you please provide clear support of this? If not, I strongly support removing this affirmation from all the manuscript (actually you used 8 times the term “causal” and 2 times “demonstrate”), including the abstract.

Author Response: Thank you very much for your hint. You are right; we stated too loosely

that we demonstrated a causal relationship. We applied the combination of DID

with matching and several robustness checks to minimize the risk of biased

estimates. However, of course, we cannot exclude that the results are biased with

some accepted statistical error or even examine the risk that might be left. We

therefore revised the manuscript accordingly to your suggestion by deleting the

term “causal” and “demonstrate” when we are referencing to our estimated effects.

Reviewer response: The authors adequately took into account the reviewer suggestion.

2) Your first hypothesis H1 is that your intervention could reduce inappropriate antibiotic prescriptions. However, all you analyses are focused to the reduction of overall antibiotic prescription rates, without detailing which prescriptions are appropriate or not. Moreover, you did not detail German guidelines on antibiotic treatments for URTIs in adults, to allow distinguishing appropriate and inappropriate antibiotic prescriptions. And finally in the discussion you stated that this hypothesis H1 is supported by your results.

I suggest to provide detailed data and analysis on which prescriptions were appropriate or not, in term of indication, class of molecule, duration of treatment, etc; or to change this hypothesis by “reducing overall prescription rate for URTIs”, which was the objective that you previously published in your protocol (your reference 50).

Author Response: Thank you very much for this important comment. In the previous

version of the manuscript, we have formulated our hypothesis too broadly.

Unfortunately, our data set only provides the information about the diagnosis and

whether an antibiotic has been prescribed. Therefore, we cannot distinguish

between appropriate and inappropriate antibiotic prescriptions. We have implicitly

argued that in the case of treating upper respiratory tract infections (URTIs) any

antibiotic prescription is considered to be inappropriate. However, this is not true,

as we have written in the manuscript “For patients belonging to a higher-risk group

(e.g., elderly patients) respective guidelines suggest the use of antibiotics in some

cases.” Moreover, in the revised version of our manuscript we describe in more

detail the cases, where respective guidelines suggest the use of antibiotics (as

suggested by reviewer 2). Therefore, we very welcome your suggestion to change

our hypothesis to “reducing overall prescription rate for URTIs”. Since we have

transferred and adapted the parts with the development of the hypotheses to the

discussion, as suggested by the second reviewer, we have excluded the

hypotheses from the text. However, we scrutiny adapted any misleading wording

in regard to your suggestion that we are aiming to reduce the overall prescription

rate for URTIs by means of the communication training.

Reviewer response: The authors adequately took into account the reviewer suggestion., no additional change needed.

3) Your initial aim was to conduct a randomized trial, but you stated that, due to lower inclusion rate than expected, and a lack of power, the analysis based on this trial was not statistically significant. Then, you reported that you “formed a control group from observational data” to apply a DID method and your matching.

Your initial number of eligible PCP was 1554, but only 17 received the intervention. I don’t understand how many were allocated in the control group initially planned? None of them? The initial planned RCT had 3 arms, how much were in the third arm? Finally in the results section, you talk about 2189 eligible CPC? It is very difficult to follow it without any clear flow chart. If there was also 17 CPC in the initial control group, I don’t understand your matching, which was 1:1 if I well understood (It is not specifically mentioned) could increase your sample size? Finally, the total number of patients involved in the study is not clearly provided. A flow chart with the number of CPC and the number of patients at each stage would be very helpful.

In the same way in the abstract, you said that 1554 PCP were invited to participate, but you neither report the number of PCP finally recruited, nor the number of patients.

Author Response: Our study consists of two different analyses. In the first part, we conduct a preintervention

analysis based on data from 2013 to 2015 of 2,189 PCPs. This is the number of enrolled physicians from 2013 to 2015 with non-missing data. In the second part of the

study, we carried out the intervention analysis, which included data from 1,477 PCPs.

In the study protocol, the sample size has been computed to be 31 per study arm. In total,

we have been able to recruited 34 physicians to participate in the study. Since the inclusion

rate was much lower, both intervention groups have been consolidated. Splitting the group

of 34 physicians resulted in 17 physicians that have been treated and 17 physicians that

have assigned to the control group. However, due to the small number of recruited

physicians even a comparison between the pooled intervention group and the control group

had not resulted in statistical significant effects. That is the reason why we alternatively

formed the control group from observational data consisting of 1,460 PCPs.

Since, we expect that the evolutions of the prescription rates are suspected to differ

between the intervention and control group we apply a matching approach that is based

on covariates and pre-intervention outcomes. The covariates are the same that have been

used in the pre-intervention analysis. The additional conditioning on pre-intervention

outcomes enables that all potential outcome trends are aligned between the intervention

and control group.

In the matching approach, we do not apply a 1:1 matching. Instead, we extracted weights

for each physician of the control group such that the control and the intervention group are

balanced accordingly to pre-intervention out-comes and covariates. Physicians that are

more similar to the trained physicians of the intervention group have higher weights.

Therefore, we use all of the 1,460 PCPs of the control group; however, the specific PCPs

are weighted accordingly to their entropy weights. The estimation procedure takes into

account the weighting of the physicians and we did not just blow up the sample size.

We are aware about showing the results of both analyses (pre- and intervention) is much

material for one manuscript. However, we prefer to show also the results of the preintervention

analysis to increase the transparency. To ease the understanding we provide

in the new version of the manuscript a flow chart with the number of PCP and the number

of treated URT cases at each stage.

Reviewer comment: The number of PCP and patients at eauch stage of the study appear clearer thanks to the flow chart. No additional change needed.

Other comments:

1) Could you define more precisely the study periods for each group? In the published protocol of the initial RCT, you planned to only include months from April to June, from 2013 to 2016. Was it the case in this study? If it is, could explain this choice? It is well known that most of the URTI diagnoses and prescriptions occur in the winter period.

Author Response: Since in the first quarter of 2016 the training has been conducted, the

periods initially planned to include in the study has been restricted to April to June.

We could increase the sample length until the end of 2016. However, the first period

of 2016 is still excluded from the study due to the training period.

Reviewer response: we understand the author choice regarding the inclusion period, not additional change required.

2) Your main analysis report a significant reduction of antibiotic prescriptions, without detailing for which pathology. Do you have some precision about it? One concern when reducing antibiotic prescriptions is to increase the risk of treatment failure/complications. Do you have any data about hospitalization rate before and after the intervention, or the number of readmission?

Author Response: No, unfortunately the data do not include any further information, e.g.

about hospitalization. We only know the target-diagnoses of acute bronchitis,

sinusitis and pharyngitis of the specific patient treated.

Reviewer response : we understand the author response, however the first part of the question has no answer. It would be interesting to know which pathology beneficiate from the higher reduction of ATB use thanks to the intervention.

3) Another concern of such intervention is the long-term benefice. If I well understood, you assessed the effect of the intervention only few months after. Could you justify this choice? An important cluster randomized trials on antimicrobial stewardship showed that such intervention could have a transient effect (doi: 10.1001/jama.2014.14042), and without any continued feedback to clinician, the durability of the intervention could be jeopardized. Could you discuss it in the limits? Did you plan to provide feedback, which is an important component of antimicrobial stewardship?

Author Response: This is a very good idea and of course an important concern for the

limitations that we have now included in the new version of the manuscript. Thank

you very much! Unfortunately, our choice is based on our data set. We only have

data until the end of 2016. For future research, this is an interesting question.

Reviewer response: no additional comment.

4) Could you detail why among 1554 invited PCP, only 17 received the intervention? This is less than 1.5%, and could represent a substantial risk of selection bias, since the PCP who accepted to participate may be particularly aware about the need to reduce antibiotic use, while the remaining PCP, which constituted your control observational data, may be less interested in this. Could you discuss it in the limits? Again, this should lead you to be very careful in using the “causality” term, because this characteristic cannot be recorded in the baseline variables, and thus is very difficult, if not impossible, to account for in the matching analysis.

Aurhor Response: The risk of selection bias is exactly the reason why we applied the matching

approach that is based on pre-intervention outcomes. After weighting the observations of

the control group by the physician specific entropy weights, the outcome trends between

the intervention and control group are similar as shown in Table 2. This means that

physicians that have a similar prescribing behavior over time before the intervention have

higher weights. Of course, we cannot guarantee that the applied matching takes into

account the distinct awareness between the PCPs who accepted to participate and the

ones who do not accepted. Therefore, we deleted the term “causal” and “demonstrate”

when we are referencing to our estimated effects and add this issue to the limitations.

Reviewer response: we agree with the author response and changes in the manuscript.

5) Could you explain why you choose only 3 CIM10 codes for URTIs, is there no other relevant code?

Author Response: We concentrate the analysis to these diagnoses, since in the primary care

setting for these cases an antibiotic is often prescribed, although only in some cases the

use of antibiotics is suggested by respective guidelines within these diagnoses.

Reviewer response: We understand the author respons, but could you at least provide the proportion represented by these three codes among all the ARTIs ? This would be helpful to be confortable to extrapolate your findings to any ARTIs

6) Your main analysis report a significant reduction of 6.5% of antibiotic prescriptions compared to the control group. Due to the major burden of antibiotic resistance,; even this small reduction may have an impact, but could you discuss, from an economic point of view, why you believe that this intervention may be more cost effective than another to reduce antibiotic use in the community?

Author Response: We have trained primary care physicians to improve their communication skills,

since literature suggests, that antibiotic prescriptions can be associated with a

communication problem. Our results underline this hypothesis. Other studies have

implemented different interventions like providing information material about antibiotic use

for the patients or integrated a warning button in the physicians’ software. To assess the

cost effectiveness of these interventions one needed to compare the effects and the

underlying costs that are hardly available. However, we believe our intervention is cost

effective because of two reasons. First, our intervention consists of a short (2 x 2.25h)

communication training that can be easily integrated in advanced clinical education, as it

is already part of under- and postgraduate training in the federal state of Schleswig-

Holstein. Second, we believe that the effects of the communication training are not limited

to the antibiotic prescriptions behavior. In future research, we plan to analyze the training

effect on other behavioral aspects of the physicians, for instance withdrawals of protonpump

inhibitors prescriptions.

Revier response: no additional comment.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: Yes: Naïm Ouldali

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at figures@plos.org. Please note that Supporting Information files do not need this step.

7 Apr 2020

Response to Reviewers, PLOS ONE, Submission PONE-D-19-35281

Communication training and the prescribing pattern of antibiotic prescription in primary health care

Thank you very much for the review of our paper. Again, it was very helpful indeed for improving the previous version of the manuscript. We incorporated (almost) all suggestions. Below you will find the specific responses to the reviewer’s comments.

Review Comments to the Author

Reviewer #1

This is a well-written and clear manuscript that discusses an important antibiotic stewardship intervention with potential implications for both public health and clinicians which would be a helpful addition to the literature. Despite the small intervention sample size, I do think it is worth publication, and so have some suggestions for the authors to improve this manuscript. Additionally, I have very little experience with entropy balancing and using observational data as a control and, therefore, would recommend the journal consider a statistical review to ensure the approaches used were appropriate as I am unable to provide a full evaluation.

Suggestions:

• Throughout the paper, the authors reference inappropriate antibiotic prescribing. However, with the diagnoses listed here, there are some cases where antibiotic prescribing is appropriate (at least based on the information provided), so it is too speculative to call all prescribing for these conditions inappropriate. The authors either need to present more data about what proportion of these diagnoses might warrant antibiotic therapy (I am not sure if this data is available in the literature) or be very cautious and explicit in their interpretations.

Response: Thank you very much for this important comment. Of course, you are right; there are some cases where antibiotic prescribing is appropriate in cases of the listed diagnoses. We have described these cases in the Data source subsection. Unfortunately, our data set do not provide additional information that would enable us to assess the appropriateness of an antibiotic prescription. To make this point more explicit in the revised version of our manuscript, we indicate the aim of the study as to reduce the overall prescription rate for URTIs by means of the communication training. The term “inappropriate prescription” has been deleted to avoid any misleading wording.

• Introduction: Overall, a clear and concise summary of the evidence. At times, statements summarizing the literature seem over-generalized or too broad. This is especially pronounced on pages 3-4, starting with the sentence that begins with “Most patients possess insufficient knowledge…”. I believe all that is needed to ameliorate this is softening the language, for example, consider changing “Additionally, physicians wrongly assume to “Additionally, physicians may wrongly assume….”.

Response: At several stages, we have softened the language by inserting words as “may”, “might” etc.

• Introduction: I found the wording in the last paragraph of the introduction to be slightly confusing. Consider rephrasing and more succinctly stating the aims of the study.

Response: In the revised version of the manuscript, the aim of the study is stated more compactly.

• Methods: How were antibiotic prescriptions linked to diagnoses? Are indications listed for the prescription or was it inferred based on the visit diagnoses? If the latter, were visits with diagnoses for conditions where antibiotics are appropriate excluded (e.g., UTI, pneumonia)?

Response: Thank you very much for this important hint that we had not addressed in the previous version of the manuscript. We now have added the following paragraph in the Data source subsection. “Since the antibiotic prescriptions have been inferred based on the visit diagnoses, we excluded cases with additional diagnoses. This includes the presence of diagnoses regarding puerperium/pregnancy (O00-O99), further (bacterial) infections (A00 to A37, A39 to A79, J15, J17, J18) or chronic diseases (I50, J44, J45, C00 to C75). If the diagnosis had been made several times or more than one diagnosis had been made from the three groups (J01, J02, J20), the corresponding cases were also excluded.”

• Methods: I saw in the response to previous reviewer comments that tried to address the certainty of diagnosis variable, but I am still confused by this. It might be helpful to include more information about how this is recorded in the claim/medical record.

Response: We included further information about the German coding policy for primary care physicians regarding the certainty of the diagnosis in the Measurements subsection.

• Methods: were there any collinearity issues in your model when using both age and insurance states as these reflect similar “stage of life” clusters?

Response: Yes, there is an issue regarding collinearity, especially for the variables Patient aged 65+ and Pensioners insured. If we neglect the variable Pensioners insured, the estimated effects of Patient aged 65+, Female patient aged 35-65 and Female patient aged 65+ are slightly stronger. All other effects are not affected. However, in our point of view, this collinearity problem is not critical, since it does not affect our main results. Moreover, we decided to include the variable Pensioners insured, since 3% of the observations are indicated as Pensioners insured, but are below the age of 65 and 1.5% of the observations are not indicated as Pensioners insured that are aged above 65. To summarize, we think that Pensioners insured serves as an important control variable and the collinearity issues can be neglected.

• Tables: I found the tables header in table 1 and 4 difficult to follow. I think it would be clearer to put a more descriptive title than just the model number, or if that is not feasible, include in a footnote the model specifics. This information is in the methods, but the tables need to be interpretable on their own.

Response: We extended the header and notes of the tables.

• Discussion: I think there might be other reasons for a declining trend in antibiotic use besides awareness of AMR. Consider other secular changes such as visit and disease incidence, antibiotic stewardship practices, and immunizations.

Response: Thank you very much for this hint. In the revised version of the manuscript, we extended the discussion by mentioning antibiotic stewardship programs. Since we included the number of URTI patients, the visit and disease incidence might be controlled for and do not serve as an explanation here.

• Discussion, page 16: The points about emergency medicine visits are fair, but the authors should also consider that these visits may be more severe than scheduled primary care visits.

Response: We integrated this important point in the discussion.

• I found the discussion of why younger adults and women might receive higher antibiotics speculative and unconvincing. Consider revising.

Response: We agree; this part of the discussion is somewhat speculative. However, our aim of the study was to gain further insights about the antibiotic prescribing process. Our hypothesis was that patients’ expectations are a key factor driving inappropriate antibiotic prescription due to a larger underlying communication problem. As we argue by means of the findings of other studies, the expectations might vary with age and gender. The underlying mechanisms behind these factors are far beyond our research focus. Therefore, the discussion remains here speculative. However, using these structural differences in the antibiotic prescribing behavior - potentially reflecting differences in the patients’ expectations - allows us to gain insights about differences in the effects of the communication training in the form that the higher the communication problem the more effective is an communication improvement. This result seems to be trivial, but it obtains insights of the underlying mechanism that is mostly not considered by other related studies.

However, to reduce the extent of the speculation, we revised the respective paragraph accordingly.

• Limitations: regarding the Hawthorne effect, did intervention clinicians know their data was being tracked? Were they aware of how long their data would be tracked? These are important considerations for understanding the extent of possible bias due to the Hawthorne effect.

Response: Yes, the participants did know that their data would be tracked before and after the intervention. Therefore, of course, we cannot control for the Hawthorne effect. This is an important limitation of our study that we discuss in the limitation section.

Reviewer #2

Manuscript ID PONE-D-19-15846: Communication training and the prescribing pattern of antibiotic prescription in primary health care: A case-control study.

Reviewer response: no additional comment.

Submitted filename: Response to Reviewers.docx

Click here for additional data file.

5 May 2020

Communication training and the prescribing pattern of antibiotic prescription in primary health care

PONE-D-19-35281R1

Dear Dr. Strumann,

We are pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it complies with all outstanding technical requirements.

Within one week, you will receive an e-mail containing information on the amendments required prior to publication. When all required modifications have been addressed, you will receive a formal acceptance letter and your manuscript will proceed to our production department and be scheduled for publication.

Shortly after the formal acceptance letter is sent, an invoice for payment will follow. To ensure an efficient production and billing process, please log into Editorial Manager at https://www.editorialmanager.com/pone/, click the "Update My Information" link at the top of the page, and update your user information. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, you must inform our press team as soon as possible and no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

With kind regards,

Martin Chalumeau, MD-PhD

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

8 May 2020

PONE-D-19-35281R1

Communication training and the prescribing pattern of antibiotic prescription in primary health care

Dear Dr. Strumann:

I am pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please notify them about your upcoming paper at this point, to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

For any other questions or concerns, please email plosone@plos.org.

Thank you for submitting your work to PLOS ONE.

With kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Martin Chalumeau

Academic Editor

PLOS ONE