Introduction The emergence and the dissemination of drug-resistant bacterial strains make treatment decisions challenging and may be associated with treatment failures. This phenomenon has become a major public health issue. Streptococcus pneumoniae is the most commonly identified bacterial cause of community-acquired invasive infections and pneumonia [1],[2]. Multidrug-resistant pneumococci (MRP) are now ubiquitous, despite the recent availability and now wide use of 7-valent pneumococcal conjugate vaccine (7-PCV) in developed countries, with more infections being caused by nonvaccine serotypes [3]. Very few new antimicrobial drugs are expected to become available in the near future, and several studies support the notion that antibiotic consumption is a key driving force in the rate (number of resistant isolates/total number of isolates) of beta-lactam–resistant pneumococci [4] and in the dissemination of MRP [5]–[7]. Furthermore, there is some evidence that decreasing antibiotic use can lower MRP rates [8]. For instance, the highest MRP rates are reported in southern Europe [9], southeastern Asia [10], and North America [11], where antibiotic consumption is generally higher than in northern Europe [12]. Thus, many countries have undertaken public health programs to optimize antibiotic prescriptions in the community [13]. In the early 2000s, France, a country with nearly 62,000,000 inhabitants and nearly 54,000 general practitioners, faced growing problems with MRP, with >50% of strains showing decreased penicillin G susceptibility. France was also identified as the country with the highest antibiotic consumption in Europe [12] and one of the highest antimicrobial users worldwide. Thus, the French government initiated a long-term nationwide campaign to reduce antibiotic overuse and control the dissemination of resistant bacteria in the community. The national program, named “Keep Antibiotics Working,” was launched in 2001, targeting both the general public and health care professionals, to encourage surveillance of antibiotic use and resistance and to promote better-targeted antibiotic use. Since 2002, a public service campaign entitled “Les antibiotiques c'est pas automatique” (“Antibiotics are not automatic”) is launched each winter with the primary goal of decreasing prescriptions, particularly during the viral respiratory infection (VRI) epidemic season and among children, for whom >40% of the prescriptions are written [14]–[16]. In this article we evaluate the effectiveness of these campaigns by analyzing the evolution of outpatient antibiotic use in France from 2000 to 2007, for each therapeutic class, as well as by geographic and age-group patterns. We performed a time-series analysis and accounted for flu-like syndrome (FLS) variations. Methods The Campaign (“Antibiotics Are Not Automatic”) In 2002 the French National Health Insurance (NHI) launched a long-term nationwide campaign in the community. From its conception, the aim of the campaign was to decrease total antibiotic use in the community by 25%, targeting predominantly VRIs in young children. An extensive information dissemination campaign was developed with the central theme “Antibiotics are not automatic.” Each winter (October–March) it is relaunched, as higher levels of infections and prescriptions occur during this period. A description of the campaign is detailed in Text S1. Definitions and Data Sources The NHI program covers all medical care provided by physicians in private practice, community clinics, and hospitals. Patients pay health service fees, which are refunded by the NHI. In France, everyone, even those with low or no income, are covered by the NHI program. We used anonymous computerized individual data from two main NHI agencies (General Scheme and Social Scheme), which cover salaried workers and the self-employed (≥91% of the French population), to access all antibiotics prescribed, dispensed by outpatient pharmacies, and reimbursed by NHI from 2000 to 2007 (beginning week 27 [July] 2000, ending week 13 [March] 2007). Each drug approved by the French Drug Agency is assigned a unique seven-digit code that enables identification of a particular product and its specific dosage, package size, formulation, and manufacturer. Each file contains this drug-related information, prescription date, patient's sex, year of birth, and region of residence. Weekly FLS incidence was provided by the French Sentinel Network [17] (see http://sentiweb.org/). According to the French Sentinel Network, FLS is defined as the combination of the following clinical symptoms: sudden onset fever ≥39°C, myalgia, and respiratory symptoms such as dyspnea and/or cough. Demographic data were obtained from the French National Institute for Statistics and Economic Studies (INSEE, http://www.insee.fr). We included the European part of France (henceforth referred to as France), which accounts for 83% of French territory and 96% of its population. For each of France's 22 administrative regions, reimbursement data were extrapolated to 100% of the regional population by dividing the number of prescriptions by the corresponding region's NHI coverage rate. The results are presented as the weekly rate of antibiotic prescriptions per 100 inhabitants. The study concerns antibiotics for systemic use (anatomical therapeutic class [ATC] code J01) in the community. Antibiotics were divided into six categories according to the ATC classification: penicillins, cephalosporins, macrolides, quinolones, cyclines, and “other.” Statistical Analysis Our analysis describes the crude number of antibiotic prescriptions and FLS incidence over time, overall, and by region, antibiotic class, and age group (0–5, 6–15, >15 y). We focused on the period targeted by the public service campaign (October–March); more specifically, we examined the truncated series corresponding to a 26-wk period starting at week 40 of year n and finishing at week 13 of year n+1. We conducted time-series analyses to quantitatively evaluate the impact of the successive, annual campaigns on antibiotic consumption. We used intervention models [18]–[20] that allowed us to estimate crude and FLS-adjusted effects of the campaign. The intervention models were built in two steps. For the first step, we chose the 2000–2002 period preceding the first campaign, i.e., starting at week 27 in 2000 and ending at week 40 in 2002, as the baseline and fitted a model to the series of antibiotic prescriptions accounting for seasonal variations and underlying autoregressive and moving average processes. As in other southern and eastern European countries [21], antibiotic consumption in France shows marked yearly seasonal fluctuations; therefore we first estimated a periodic trigonometric function and removed it from the truncated series so that the residual series was in stationary mode. We then fitted an autoregressive and moving average (ARMA) model [22] on the residual series, leaving white noise residual series. For the second step, we assumed that observations before and after the campaign were drawn from the same ARMA model with changes only in the mean. Following this assumption, we removed the previously estimated periodic trigonometric function from the whole antibiotic prescription series (2000–2007) and added five dummy variables, one for each annual campaign, to the ARMA model identified in step 1 so that only the mean could change. We quantified the impact of each campaign as the average relative change for the period (from winter 2002–2003 to winter 2006–2007) compared with baseline (2000–2002); that is, the ratio of the observed change over the expected change predicted from the model had there been no campaign. The same procedure was repeated after dividing the population into three age categories ( 15 y) as well as considering data of the summer period (April–September) and data of the entire year (October–September). Finally, we further adjusted the intervention model for the whole population for FLS incidence. We tested the hypothesis of a different link between antibiotic consumption and FLS incidence before and after the start of the 2002 campaign by introducing an interaction term in the model. Analyses were performed using SAS version 9.1 (SAS Institute, http://www.sas.com/). All statistical tests were two-tailed; a p-value of less than 0.05 was considered significant. Results Description of the Data Between July 2000 and March 2007, a total of 453,407,458 individual antibiotic prescriptions were reimbursed by the NHI; all files were processed for analysis. Because prescription numbers were similar for men and women (unpublished data), we combined their data. Antibiotic prescriptions and FLS fluctuated seasonally, with higher incidence rates during winter months (Figure 1). Maximal values differed from one year to another. The number of antibiotic prescriptions varied between 585,524 (week 33 of 2006) and 2,196,942 (week 3 of 2002) (amplitude ratio 3.8). FLS peaks varied between 396 (winter 2002–2003) and 939 (winter 2004–2005) cases per 100,000 inhabitants (amplitude ratio 2.3). The FLS epidemic began as early as week 45 (third week of November) in the 2003–2004 season and as late as week 4 (fourth week of January) in the 2001–2002 season. All epidemics occurred between week 40 (beginning of October) and week 13 (end of March). 10.1371/journal.pmed.1000084.g001 Figure 1 Antibiotic use and flu-like syndromes in France, from July 2000 to March 2007. Weekly totals of antibiotic prescriptions and FLS cases per 100 inhabitants plotted against time. The mean number of prescriptions per 100 inhabitants for the general population was 72.4 for the 2000–2001 and 2001–2002 winters; this figure gradually decreased to 56.6 during the 2006–2007 season (Table 1). 10.1371/journal.pmed.1000084.t001 Table 1 Mean number of prescriptions between October and March, per 100 inhabitants (percent change compared to 2000–2002). Antibiotic Class 2000–2002 2002–2003 2003–2004 2004–2005 2005–2006 2006–2007 Penicillins 27.0 21.7 (−19.6) 21.5 (−20.4) 20.7 (−23.6) 20.8 (−23.2) 20.2 (−25.3) Cephalosporins 16.3 13.6 (−17.0) 14.7 (−10.2) 14.6 (−10.7) 12.5 (−23.7) 12.3 (−24.6) Macrolides 16.4 14.2 (−13.7) 14.2 (−13.8) 13.9 (−15.3) 12.4 (−24.4) 11.5 (−30.1) Quinolones 4.2 4.3 (3.2) 4.3 (2.8) 4.8 (14.2) 4.6 (8.6) 4.7 (12.8) Cyclines 3.1 3.1 (1.0) 3.2 (3.7) 3.2 (3.2) 3.1 (1.1) 3.0 (−3.7) Other 5.3 8.1 (55.0) 8.5 (62.0) 7.4 (40.6) 5.1 (−3.8) 4.8 (−8.0) All 72.4 65.1 (−10.1) 66.4 (−8.3) 64.5 (−10.8) 58.4 (−19.3) 56.6 (−21.9) Variations in antibiotic use were observed among France's 22 administrative regions. Antibiotic consumption declined in all regions between 2001–2002 and 2006–2007 (Figure 2). In 2000–2001, 15/22 regions had >70 prescriptions per 100 inhabitants, but none exceeded this level in 2006–2007. The most important reduction was observed in the Centre Region (−28.4%). Among the six administrative regions with the highest average number of prescriptions per 100 inhabitants in 2001–2002 (Nord 91.8, Picardie 83.3, Haute Normandie 80.3, Lorraine 78.7, Champagne-Ardenne 78.7, and Poitou-Charentes 77.3), the decrease in antibiotic use was among the seven highest, with values of −25.2%, −26.1%, −25.6%, −25.2%, −27.1%, and −27.30%, respectively. Antibiotic prescription differences among regions persisted after the campaign but were less dramatic. 10.1371/journal.pmed.1000084.g002 Figure 2 Winter antibiotic prescriptions in France by region, from October 2000 to March 2007. The number of October–March prescriptions is divided by the number of regional inhabitants for the respective year in each of 22 France's regions: Al (Alsace), Aq (Aquitaine), Auv (Auvergne), BN (Basse Normandie), Bou (Bourgogne), Br (Bretagne), CA (Champagne-Ardenne), Ce (Centre), Co (Corse), HN (Haute Normandie), Li (Limousin), Lo (Lorraine), LR (Languedoc-Roussillon), IDF (Ile de France), FC (Franche-Conté), MP (Midi-Pyrénées), NPDC (Nord-Pas de Calais), PACA (Provence-Alpes-Cote d'Azur), PDL (Pays de Loire), PC (Poitou-Charente), Pi (Picardie), RA (Rhones Alpes). Penicillins, cephalosporins, and macrolides were the three most used antibiotic classes at baseline, with 27.0, 16.3, and 16.4 prescriptions per 100 inhabitants; their use also declined the most (changes of −25.3%, −24.6%, and −30.1%, respectively) among all antibiotic classes. Quinolones, which remain a less frequently prescribed antibiotic class (4.2 prescriptions per 100 inhabitants at baseline), was the only class whose use increased (+12.8%) (Table 1). For children 15 y old (−9.8% [95% CI −14.9% to −4.7%] and –12.5% [95% CI −16.8% to −8.1%], respectively). The evolution of this change differed according to age group. Nevertheless, by 2006–2007, the campaign had achieved significantly fewer antibiotic prescriptions: rates were −27.0% (95% CI −33.5% to −20.5%) versus baseline for the whole population, −30.1% (95% CI −40.7% to −19.6%) for the youngest children, −35.8% (95% CI −48.3% to −23.2%) for 6- to 15-y-olds, and −20.5% (95% CI −25.6% to −15.4%) for those >15 y (Table 2). 10.1371/journal.pmed.1000084.g004 Figure 4 One step ahead forecasts of the interrupted ARMA model and observed antibiotic prescriptions data. October–March horizontal lines indicate the estimated average level by the interrupted ARMA model (in red) and the observed average level without any campaign effect (in black) each winter. The percentages listed above the peaks denote the ratio of change. 10.1371/journal.pmed.1000084.t002 Table 2 Estimated mean percent reduction of antibiotic use [95% CI] and associated p-values, compared to 2000–2002 baseline values. Period Age Group 2002–2003 2003–2004 2004–2005 2005–2006 2006–2007 October–March (winter) 0–5 y (FLS unadjusted) −1.2 [−7.7 to 5.3] −2.4 [−10.8 to 5.9] −14.9 [−24.2 to −5.6] −26.9 [−36.9 to −17.0] −30.1 [−40.7 to −19.6] p-value 0.71 0.87 0.002 15 y (FLS unadjusted) −12.5 [−16.8 to −8.1] −11.3 [−16.2 to −6.5] −13.4 [−18.4 to −8.5] −19.0 [−23.9 to −14.0] −20.5 [−25.6 to −15.4] p-value 15% of antimicrobial drug use for penicillins, cephalosporins, and macrolides (similarly, no reduction was noted for quinolones) during the 2000–2004 period [4]. The overall decline we observed can be compared to those observed in three other national studies for which comparable data are available: Belgium, Sweden, and Australia. In Belgium, a significant decrease (after controlling for FLS variation) was obtained after the first year but not after the second [37]; a longer follow-up might show a more pronounced and long-term decrease [43]. In Sweden, a country that had one of the lowest antibiotic consumption rates in Europe [23], an intervention was launched in 1995 [44]. This intervention did not involve a public campaign and was reported as having led to a −20% change in overall antibiotic sales between 1995 and 2004 [38]. The Australian investigation focused mainly on consumer awareness and physician behavior changes [39]. In France, comprehensive coverage of the population by the NHI and drug reimbursement for outpatient care offer a unique opportunity for in-depth analysis of drug use data, in particular use of antibiotics among outpatients for almost the entire population. To the best of our knowledge, this is the largest body of data—with nearly half a billion data entries and the longest time-series of individual and weekly data on antibiotic prescriptions—to evaluate antibiotic use in the community ever analyzed. Our data show that the primary objective of the French national campaign was largely achieved, with a 30.1% decrease in antibiotic use in children <6 y old. This result is very encouraging, because a substantial proportion of antibiotic prescriptions for young children are unnecessary because of the viral origins of their infection [26]. The most important decrease in antibiotic use for children <5 y was noted after the second campaign. There is no clear explanation for this change, but it suggests that repeated campaigns may be necessary to overcome initial resistance by parents and physicians to reduced antibiotic prescription. In addition, children aged 6–15 y also had significantly lower antibiotic use (a −35.8% change over the study period). This change may be due to the use of rapid tests to diagnose group A streptococci tonsillitis, which was promoted by the campaign in this age group, which is at a higher risk of group A streptococci throat infections. Antibiotic consumption changed by −24.1% among young adults (26–35 y), who were initially the biggest antibiotic consumers among adults. Demographic characteristics of this population suggest that most parents of young children fall within this age category; as a result, it is likely that young adults have been specifically affected by the campaign. Finally, prescriptions in the 21–25 y age group remained stable from 2000 to 2007 (as compared to decreases observed in older and younger age groups). Two hypotheses may explain this observation. First, 21- to 25-y-old adults are less likely to live in or interact with collective institutions (daycares, schools, etc.) or be in close contact with children than other age groups; as a result, they are less likely to be exposed to VRIs, which are typically spread by young children. Second, it may be difficult to further decrease antibiotic use in this age group, as they already represent the age group with the lowest antibiotic prescription rate before the campaign. While crude results did not differ markedly from those adjusted for FLS fluctuations, accounting for the latter was crucial to the interpretation of the changes recorded. Indeed, in many countries, VRIs account for a high share of unnecessary antibiotic prescriptions. Our intervention model highlighted a significant weakening of the association between FLS incidence rates and antibiotic prescriptions after the second yearly campaign. This observation suggests that the fraction of antibiotics prescribed for viral illnesses has significantly decreased, which is a highly encouraging result, as one of the campaign's main objectives was to reduce VRI-associated antibiotic overuse. However, the association between FLS and antibiotic prescriptions persists, at a low level, even after the campaign. Several limitations of this study should be noted. First, due to the quasiexperimental design (i.e., absence of a control group) and limited preintervention data, a cause–effect relationship between the campaign “Antibiotics are not automatic” and decreased antibiotic use cannot be proved. For example, the influence of the antibiotic campaigns in other geographically proximate European countries such as Belgium cannot be excluded. Some southern European countries also conducted campaigns during the 2002–2007 period (Greece: mass media campaign in 2001–2003; Spain: mass media campaign since 2006; Portugal: radio campaign in 2004–2007) and observed evolution of outpatient antibiotic use [23]. Thus, effects of targeted campaigns versus the spontaneous decrease of antibiotic use should be evaluated in Europe. Very few new antibiotics have been launched in the past decade; as a result, the promotion of antibiotic prescription by the pharmaceutical industry has probably decreased. This might be a confounding factor for the observed reduction of antibiotic use. Second, we did not have access to information about the pathology for which antibiotics had been prescribed. In France, no information system exists that provides easy access to data linking drug use to a clinical condition. Third, we did not account for the introduction of the 7-PCV in our analysis. The 7-PCV initially received marketing authorization in France in 2002, specifically for children presenting specific risk factors. It became widely used only at the end of 2006, once it had been recommended for all children <2 y old (unpublished data). It is thus unlikely that its market introduction could explain much of the decreased antibiotic use observed over the 5-year investigation period. Fourth, FLS surveillance data do not account for other VRIs such as infections due to respiratory syncytial virus, which generally occur several weeks before a flu outbreak. It was not feasible to account for these viral infections, as surveillance data of these infections are not available in France. As a result, it is likely that we underestimated the association between community viral infection and antibiotic prescription. Fifth, other local initiatives were promoted in France since 2000, such as the campaign “Antibiotics Only When Necessary” promoted in a county in southeastern France (see http://www.gepie.org/). The added value of such initiatives was not specifically investigated. Sixth, some authors have reported that antibiotics may substantially reduce the risk of pneumonia after chest infection [45]. Therefore, adverse effects of reduced antibiotic use (e.g., increase in certain severe infections) is questionable. We do not address this question here, which remains to be investigated. Reasons for reduced antibiotic prescriptions, e.g., fewer consultations or improved prescribing, were not evaluated here, the data did not provide the necessary information to address this question. It has been reported that mass media campaigns play a role in influencing antimicrobial prescription practice in the UK [46]. Due to the multifaceted approach and targeting of the general public and physicians in parallel, the individual effect of each approach could not be evaluated. We believe that the success of the intervention was in fact a result of the combined approach, e.g., face-to-face peer education and widespread public campaigns, allowing both the practitioner not to prescribe and the patient not to ask for antibiotic therapy. The results of the present report are highly promising in terms of bacterial resistance control. A recent European study confirms the ecological relationship between antibiotic consumption and rate of MRP at the national level [4]. This study strongly underlines the responsibility of countries with higher levels of antibiotic use and recommends that they urgently undertake campaigns devoted to the control of MRP, including the promotion of prudent antibiotic use. In France from 2001 to 2006, a decreasing trend was observed in the rate of pneumococci resistant to penicillin (47% to 32% of isolates) and the rate of pneumococci resistant to macrolides (49% to 36%) in France (see http://www.rivm.nl/earss). Because our campaign did not target any specific therapeutic class, it may have prevented interclass switching. Nevertheless, a slight increase of quinolone prescriptions occurred. Although this was a moderate increase compared to large decreases recorded for all other classes, this trends points to the need for careful monitoring of quinolone-resistant bacteria in the community [47]. Despite the sharp reduction of antibiotic prescriptions observed, France remains a high user of antibiotics [23]. Nevertheless, the impact of the decrease in antibiotic use on the prevalence of infections caused by antimicrobial-susceptible and antimicrobial-resistant strains must be investigated. Future studies should combine the assessment of 7-PCV vaccination and antibiotic-reduction policies, and evaluate their respective role in the evolution of S. pneumoniae invasive infections, according to strain susceptibility. Supporting Information Text S1 Description of the public health campaign “Antibiotics are not automatic” between 2002 and 2007. (0.04 MB DOC) Click here for additional data file.
