Background
Antibiotics abuse has been identified as one of the main problems involved in irrational drug use (Shankar, Reference Smolinski, Hamburg and Lederberg2009). An investigation shows that antibiotic use has remained sub-optimal in all regions of the world over the last 20 years and the situation does not appear to be improving (Lu et al., Reference Marshall, Hiscock and Sibbald2011). Antibiotic prescriptions for upper respiratory tract infections (URTIs) make up a great portion of overall antibiotic prescriptions (Schroeck et al., Reference Shankar2015). Although the routine use of antibiotics for URTIs has been proven unnecessary and wasteful (Kenealy and Arroll, Reference Li, Song, Yang, Chen, Gong, Yin and Lu2013), antibiotics are still commonly overused for URTIs worldwide (Hurley, Reference Jamtvedt, Young, Kristoffersen, O’brien and Oxman2014). Well-documented evidence indicates that over prescribing of antibiotics for URTIs in primary care settings does contribute to antimicrobial resistance (AMR) (Costelloe et al., Reference Costelloe, Metcalfe, Lovering, Mant and Hay2010), an alarming public health threat, which has cost the United States US$4000–5000 million and Europe €9000 million annually (Smolinski et al., 2003; Strategic Council on Resistance in Europe, 2004).
The excessive mortality and dramatic economic burden caused by AMR (Birnbaum, Reference Birnbaum2003; EJW Group, Reference Hamblin2009) has triggered a surge of research on interventions in antibiotic prescribing practices worldwide (World Health Organization, 2012), especially interventions targeted at physicians (van der Velden et al., Reference van der Velden, Pijpers, Kuyvenhoven, Tonkin-Crine, Little and Verheij2012). Audits and feedback on prescribing performance can result in a small to moderate change in the prescribing practices of physicians (ranging from a 16% decrease to 70% increase in compliance with prescription guidelines) (Jamtvedt et al., Reference Kenealy and Arroll2006). Although recent studies reported a relatively stronger effect of ‘audit and feedback’ when it was combined with ‘peer academic detailing’ (Gerber et al., Reference Gerber, Prasad, Fiks, Localio, Grundmeier, Bell, Wasserman, Keren and Zaoutis2013; Gjelstad et al., Reference Gjelstad, Høye, Straand, Brekke, Dalen and Lindæk2013), the enhanced effect can often be offset by the resources required and practical considerations (Naughton et al., Reference Nilsen, Myrhaug, Johansen, Oliver and Oxman2009). A review by the Cochrane Collaboration (Jamtvedt et al., Reference Kenealy and Arroll2006) concluded that intensive feedback may have a greater potential given that the tested ‘feedback’ interventions are usually confidential and contain only benchmarks on average.
In the recent decades, research interest in the role of public reporting on improving patient care is growing, especially in developed countries (Haustein et al., Reference Hibbard, Stockard and Tusler2011; Rechel et al., Reference Schroeck, Ruh, Sellick, Ott, Mattappallil and Mergenhagen2016). Public reporting usually involves three broad types of information: healthcare outcomes, provider performance and patient experience (Shahian et al., Reference Sigurøardøttir, Nielsen, Munck and Bjerrum2011; Burns et al., Reference Burns, Pettengell, Athanasiou and Darzi2016; Rechel et al., Reference Schroeck, Ruh, Sellick, Ott, Mattappallil and Mergenhagen2016). The rationale is anchored in both the citizens’ involvement in public affairs and the hypothesis that public reporting can be used to promote quality improvement (Nilsen et al., Reference Rechel, Mckee, Haas, Marchildon, Bousquet, Blumel, Geissler, van Ginneken, Ashton, Saunes, Anell, Quentin, Saltman, Culler, Barnes, Palm and Nolte2006). Extensive studies have been undertaken to evaluate the effectiveness of public reporting on patient outcomes, and the findings indicated that public reporting did trigger greater improvement than private disclosure of the same data (Hibbard et al., Reference Hurley2005).
The mechanism of public reporting to performance improvement is complicated. The expectations from the ‘selection pathway’ (Berwick et al., Reference Berwick, James and Coye2003), in specific, users modifying their choice of providers or other decisions based on available performance measures, may fail in practice (Fung et al., Reference Fung, Graham, Weil and Fagotto2004) given that consumers may have limited choice in some health systems (Fung et al., Reference Fung, Graham, Weil and Fagotto2004; Faber et al., Reference Faber, Bosch, Wollersheim, Leatherman and Grol2009). How physicians react to these publicly reported information is then the key to understanding the mechanisms (Contandriopoulos et al., Reference Contandriopoulos, Champagne and Denis2014a). Healthcare providers may face pressure from managerial interventions and social expectations to change their practices (Contandriopoulos et al., Reference Contandriopoulos, Champagne and Denis2014b).
Antibiotic abuse has been a severe problem in China. Over 80% URTI visits receive antibiotics as a recent study demonstrated (Li et al., Reference Liu, Zhang and Wan2016). Recent works by Yang et al. (Reference Yang, Liu, Wang, Yin and Zhang2014) and Liu et al. (Reference Lu, Hernandez, Abegunde and Edejer2015) have made initial attempts to applying public reporting as a prescription quality promoting strategy. The conclusions were promising: public reporting of prescription quality significantly caused the reduction in the antibiotic prescribing of physicians to URTI patients. However, which group, for instance, high, average or low antibiotic prescribers, accounted for the reduction in antibiotic prescription has not been fully understood. The major purpose of this study was to fill this information gap by examining the changes in antibiotic prescribing practices among different publicly reported physician performance groups in URTI visits, which would benefit us with a better understanding of the mechanisms behind the relationship. Therefore, we designed a cluster randomized trial of public reporting of antibiotic prescribing across a relatively large primary care network (20 primary healthcare institutions). Although the unit of observation was the prescription, we randomized at the institution level to avoid intra-practice contamination of the intervention.
Methods
We conducted a clustered randomized-controlled trial.
Study setting
This study was undertaken in Qianjiang city of Hubei province. Hubei is located in central China with a population of over 61 million. The average annual per capita income ranks in the middle of all provinces: 6898 Yuan for rural and 18 374 Yuan for urban residents (in 2012). Qianjiang has a population of around 950 400 and 47.5% of the population reside in the rural area. In 2012, Qianjiang produced a GDP of 49.3 billion (Yuan). The average annual per capita income reached 8785 Yuan for rural and 17 451 Yuan for urban residents.
Randomization
Qianjiang has 20 primary care institutions, on average 10 km away from the nearest counterpart. All of these primary care institutions participated in this study. We used matched-pair cluster randomization to assign the participating institutions into the intervention and control groups.
-
(1) A technique for order preference by similarity to ideal solution (TOPSIS) score was calculated for each institution based on six indicators: local population size, number of beds, number of physicians, annual outpatient visits, annual episodes of admissions and annual revenue from drug sales.
-
(2) The participating institutions were sorted in an ascending order according to the TOPSIS score and adjacent institutions were paired.
-
(3) For each pair, we flipped a coin to randomly assign one into the intervention group and another into the control group.
More details about the research setting, trial design and intervention strategies can be found in the study protocol published elsewhere (Du et al., Reference Du, Wang, Wang, Yang and Zhang2015). This study was approved by the Ethics Review Committee of Tongji Medical College, Huazhong University of Science and Technology (no. IORG0003571).
Interventions
The public reporting contained information regarding (1) percentage of prescriptions containing antibiotics; (2) percentage of prescriptions containing injections; and (3) average expenditure per prescription. The indicators were calculated by the research team using the computerized hospital information management systems of the participating institutions. They were ranked in ascending orders at the institution level and the prescriber level.
The ranking information was disseminated through a poster displayed in a conspicuous place at each institutions, and handout brochures together with a report submitted to the local health authority. The posters and brochures included a brief introduction of the purpose of the reporting. It was made clear that health risks are associated with excessive use of antibiotics and injections.
During the intervention period, four or five researchers disseminated the reporting information at the intervention sites on the ninth day of each month. To maximize compliance, the local health authority issued a policy to ensure the information dissemination activities. Meanwhile, the research team inspected the intervention sites irregularly. Damaged posters, if found, were replaced immediately.
Data collection
The data used in this study came from two sources. Prescription data were extracted from the electronic medical records, which included the name and work unit of the prescriber, time when the prescriptions were issued, demographic characteristics of recipients (age, sex and type of insurance), reason for prescription (only one diagnosis was recorded for each prescription) and details of medicines prescribed (drug name, administration route, dosage, frequency and costs). Data on the characteristics of prescribers were collected through a self-administered questionnaire including name, age, sex, level of education, professional title and income. The two data sets were linked by matching the name of each prescribers.
Physicians who provided services for URTI patients were included and analysed. The physicians who had lower than 10 URTI patient visits in each study months were excluded in our study for lack of sensitivity. In total, 60 physicians (27 in the intervention group and 33 in the control group) were then included and followed up from March 2014 to September 2014 (post-intervention period). In addition, the URTI prescriptions of the physicians in the months of 2013 (pre-intervention period) were also included in our analysis.
Statistical analysis
The International Network for Rational Use of Drugs developed a list of prescription indicators which have become widely accepted internationally (World Health Organization, 1993). We selected three prescription-level indicators for the purpose of this study, which covered the type (antibiotics) and administration route (injections) of medicines that are most frequently abused:
-
(1) percentage of prescriptions containing antibiotics
-
(2) percentage of prescriptions containing two or more antibiotics
-
(3) percentage of prescriptions containing antibiotic injections.
According to the antibiotic prescription rate in the previous month, the physicians included in both the intervention and control arms were divided into three equal groups with references to similar studies (Xu et al., Reference Xu, Wang, Li, Zhang, Pavlova, Liu, Yin and Lu2010; Tang et al., Reference Tang, Zhang, Yang, Yang, Wang and Zhang2013), namely, high, average and low prescribers. For instance, for the physician grouping strategy of March 2014, we referenced all antibiotic use rates of the physicians in February 2014 (this rates were publicly reported of physicians in the intervention arm on the first few days of March 2014, but not public reported in the control arm). Descriptive analysis were carried out for selected antibiotic prescribing indicators in earlier and later periods of both arms separately for three groups of physicians.
We adopted a difference-in-differences (DID) approach to test the effects of the intervention. Logit regression models were applied to these binary dependent variables (prescriptions containing antibiotics, prescriptions containing two or more antibiotics and prescriptions containing parenteral antibiotics, 1 for yes and 0 otherwise). The analysis unit were individual prescriptions. We randomized the study sample at the institution level. Considering the hierarchical structure of the data (prescription–prescriber-facility level), we used mixed effect logistic regression using the ‘xtmelogit’ command in Stata, which accounted for random intercepts of individual prescribers and facilities. The models used can be expressed as follows:
$$\eqalignno{{\rm Logit}\left( {P_{{ijk}} } \right){\,\equals\,}&\left( {\beta _{0} {\plus}\mu _{{0jk}} {\plus}\mu _{{0k}} } \right){\plus}\Sigma _{1} \beta _{1} X_{{1ij}} {\plus}\Sigma _{2} \beta _{2} X_{{2ij}} \cr {\plus}\beta _{m} M_{{ij}} {\plus}\beta _{t} T_{{ij}} {\plus}\beta _{p} P_{{ij}} {\plus}\beta _{{\rm e}} E_{{ij}} $$
where μ 0k is the random intercept of individual facility k and μ 0jk the random intercept of individual physician j nested in facility k; X 1 a set of prescription-level covariate including patient sex, age and insurance status; X 2 a set of physician level covariate including physician sex, age, education level, professional title and income level; T ij indicates whether prescription i from physician j was or was not in the intervention arm (0 for control and 1 for intervention); and P ij indicates whether prescription i from physician j was or was not from the post-intervention period (0 for pre and 1 for post). M ij is a month pair dummy variable that indicates the seven time pair (the same month in two different years were matched as a pair). E ij indicates interaction between the intervention–control status and pre–post-intervention periods and β e the DID estimators (effect size). The effect margins for β e were then calculated which can be explained as percentage change from baseline as recommended (Williams, Reference Williams2011). Statistical analysis was performed using Stata 12.1 (Stata Corp; College Station, Texas, USA).
The characteristic difference between the two arms were tested by χ 2 test for categorical variables and independent t test for continuous variables.
Results
Overall, 60 physicians were included in this study (27 in the intervention arm and 33 in the control arm). The physicians in the intervention arm had an average age of 39.89, younger than those in the control arm (mean age=46.39, P=0.007). The professional titles were also significantly different between the groups (χ 2=11.101, P=0.004). No significant difference was observed in the rest of physician characteristics.
In total, 61 843 URTI prescriptions from these physicians were included form both the pre- and post-intervention periods, with 31 952 prescriptions in the intervention arm (14 931 in pre- and 17 021 in post-intervention periods) and 29 882 in the control arm (14 945 in pre- and 14 937 in post-intervention periods) were collected from the health information system for analysis. The insurance status of the recipients was significantly different between the two groups in both the pre-intervention (χ 2=76.83, P<0.001) and post-intervention periods (χ 2=63.092, P<0.001). Age was only significantly different between the group in the pre-intervention period recipients (t=4.064, P<0.001). Sex was only significantly different between the groups in the post-intervention period recipients (χ 2=4.477, P=0.043). The detailed characteristics at both physician and prescription levels are shown in Table 1.
Table 1 Characteristics of recipients (patients) and prescribers (physicians) of prescriptions
NCMS=new cooperative medical scheme; URBMI=urban resident basic medical insurance; SF=self-funded.
Antibiotic prescribing rate was high among the URTI patient visits at our investigated primary healthcare institutions. Overall, the percentage of prescriptions requiring antibiotics was 88.67%, percentage of prescriptions requiring two or more antibiotics was 17.65% and percentage of prescriptions requiring injection antibiotics was 75.24%. The prescription performances of all three physician groups before and after intervention are shown in Table 2.
Table 2 Prescribing indicators for groups in pre- and post-intervention periods
In DID logit analysis, the intervention showed positive significant effects on reducing the overall antibiotics prescribing rate (2.82% reduction, P<0.001) and combined antibiotic prescribing rate (3.81% reduction, P<0.001) of the physicians. However, the intervention showed no effect on reducing the overall prescribing rate of injection antibiotics.
Among the three-level prescribers, the effect size of the reduction in antibiotic prescriptions in the low antibiotic prescriber group was smallest (−1.41%, P=0.25) and combined antibiotic prescriptions (−2.42%, P=0.016), whereas largest in average antibiotic prescribers (−5.01 and −5.01% for reducing antibiotic prescriptions and combined antibiotic prescriptions, respectively, P<0.001) (Table 3).
Table 3 Estimates of effect sizes derived from the difference-in-difference analyses
OR=odds ratio; CI=confidence interval.
Parameter of interest is β e, which under our assumptions, indicates the effect of intervention. The effect margins for β e were calculated and reported using Stata software. Z and P values were derived from the result of regression analysis (Equation 1).
Discussion
Principle findings
To the best of our knowledge, this study is the first quantitative investigation that examined physicians’ prescribing performance after public reporting intervention with special concentration on the physicians’ different publicly reported performance status. The findings from this study demonstrate that a decrease in the combined antibiotic prescription after public reporting was experienced by all physicians, regardless of whether the baseline before the public release rates was high, average or low. However, the decrease in the antibiotic prescription was only observed among physicians with average and low baseline groups, which is publicly reported. The study findings also reveal that physicians who had average outcomes in the initial period showed the most significant improvement and the providers who had the best outcomes in the initial period showed the least significant improvement. No significant decrease in the injection antibiotic prescription was observed in all physician groups.
The antibiotic prescription rate for URTI patients was high even in the low physician group (77.40–85.44%). A recent systematic review has made a clearer view on antibiotic use for URTI patients in China, which indicated an 83.7% antibiotic prescribing rate out of the 45 individual eligible studies (Li et al., Reference Liu, Zhang and Wan2016). Together with the present study, we can conclude that the antibiotic prescribing rate was very high in China, far from a recent study in Denmark that revealed a 59.3% antibiotic use for URTI patient visits in general practice (Sigurøardøttir et al., Reference Reynolds and Mckee2015).
The antibiotics favouring the prescribing pattern were established by the interaction of both the health provider and patients. From a patient perspective, antibiotics are considered able to shorten the duration of URTI. Nevertheless, little was known about the bacterial resistance (Yu et al., Reference Yu, Zhao, Lundborg, Zhu, Zhao and Xu2014). The providers, who generally know that antibiotics should not be prescribed when encountering common colds, would still prescribe antibiotics to URTI visitors anyway (Sun et al., Reference Sun, Dyar, Zhao, Tomson, Nilsson, Grape, Song, Yan and Lundborg2015). Two main reasons are involved in such conflict in perception and behaviour as concluded by a qualitative study (Reynolds and McKee, Reference Shahian, Edwards, Jacobs, Prager, Normand, Shewan, O’brien, Peterson and Grover2009). Historically, the 15% drug sales mark-up policy encourages physicians to over prescribe, and antibiotics are definitely among them. The patient preferences worsened the situation as physicians attempt to satisfy the patients to retain them. Although the national essential medicines system was introduced to promote rational drug use, evidence to date reveals that such goal is hard to achieve.
The mechanism from pubic reporting to quality improvement of healthcare is complicated. The earliest and most-cited theory was put forward by Berwick et al. in (Reference Berwick, James and Coye2003), the ‘selection pathway’ and ‘change pathway’. Selection pathway relies on the information users modifying their behaviour towards the ‘high performer’ and eventually improving the overall performance, and ‘change pathway’ is based on the effect of providers’ efforts to use performance measures to improve their performance. Subsequently, an empirical study by Hibbard et al. indicated that a reputation pathway exists, where the providers are concerned about their public image when performance is made public (Hibbard et al., Reference Hurley2005). Based on these theories, a recent study conducted an analytical review and built a typology of four complementary causal pathways, which subdivided the ‘change pathway’ into three more detailed pathways, namely, change through managerial interventions, change through social structuring and change through internal pressures (Contandriopoulos et al., Reference Contandriopoulos, Champagne and Denis2014a). Although all these analyses target the organization as a unit, they still shed great light on how individual physicians may act on public reporting, especially the internal pressure pathway. In a qualitative study, the investigated primary practitioners expressed their concerns that public reporting, which encourages the ‘name and shame’ culture, would exert stress on them (Marshall et al., Reference Naughton, Feely and Bennett2002). This pressure, both from the political concerns and reputation, triggers change in behaviour and prescription patterns in this study, especially in low performers (Hannan et al., Reference Haustein, Gastmeier, Holmes, Lucet, Shannon, Pittet and Harbarth1994; Baker et al., Reference Baker, Einstadter, Thomas, Husak, Gordon and Cebul2003). This study shows that when physicians perceive that their antibiotic prescribing rates were not in the lowest range, they tend to lower their antibiotic prescribing practice in the following month. From the theoretic analysis above, we can assume that when physicians’ public reporting performances are not in the best range, they bear greater pressure to change that situation, and when the physicians’ public reporting performance were in the best range, they bore less pressure and their motivation to be better was not that strong. Few empirical studies analysed the pressure level of physicians when their publicly reported performances were poor or good, and such analysis would make a clearer view of the intrinsic mechanism from public reporting to performance improvement.
Comparison with other studies
The previous study by Yang et al. indicated that, after four months of publicly reporting the prescribing indicators of physicians, significant reductions in oral antibiotics prescription and combined prescription practice were observed, but not in the case of injection prescription practice. The result was further explained by this study that the reductions in antibiotic and combined antibiotic prescription practices were mainly observed in the physicians when their public reporting performances were not in the best range.
Very few studies have analysed the individual physician’s performance change when the public reported poor or good performance, but several studies analysed the organization, as an observation unit, and performance change when the public reported poor or good performance. Both results indicated that the groups that showed the highest initial mortalities manifested the most improvement. The result of this study differs from the results above, indicating that the publicly reported average range was the most change-inspired group. This implies that the personality of physicians also influence these changes as pointed out by researchers (Hamblin, Reference Hannan, Kumar, Racz, Siu and Chassin2007), which would be interesting to explore further.
Limitations
Although this study explored the change in the prescribing patterns of physicians after public reporting, the influence of such changes on the AMR is beyond the scope of our study because of constrained condition, such relationship would be of great practical significance. The depth of clinical coding was not sufficient to allow us to determine the cause (eg, bacterial, virus or other) of URTIs, which would provide a better understanding of excessive antibiotic prescription in China. Ideally, the whole sample of prescriptions should be included in the data analysis with proper risk adjustments. Unfortunately, we were not able to do so because of the unavailability of relevant data. However, the selection of patients with URTI in this study, which is the most common cause of visits in primary health institutions, provided us with some unique insights into the effect of public reporting interventions on different physician groups.
Conclusions
Public reporting can affect the antibiotic prescribing patterns of physicians when dealing with URTI patient visits, including decreased antibiotic and combined antibiotic prescribing rates. When the physicians’ publicly reported antibiotic prescribing rates are in the optimal range, the possibilities that they would change their prescribing patterns the following month are much less. By contrast, the possibilities that they will lower their antibiotic prescribing practices are obviously larger in the average or highest ranges. However, antibiotic use remained high even after intervention, especially the use of injections, which may in themselves be hazardous.
Acknowledgements
The authors thank Qianjiang Municipal Health Bureau for their support in carrying out this experiment. The authors also thank all research participants including Lianping Yang, Xin Du, Lijun Wang, Xiaopeng Zhang and Shiru Yang for their hard work in data collection. The authors also thank Ruoxi Wang for her hard work in language editing.
Financial Support
This work was supported by the National Natural Science Foundation of China (No.71373092).
Conflicts of Interest
None.
Ethical Standards
The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional guidelines on human experimentation [Ethics Review Committee of Tongji Medical College, Huazhong University of Science and Technology (no. IORG0003571)] and with the Helsinki Declaration of 1975, as revised in 2008.
