Author members of the pilot project teams for ABM and AMR surveillance were: Delhi: A. Kotwani, C. Wattal; Vellore: S. Chandy, E. Mathai, K. Thomas, M. Mathai, J. Mathews, I. Joseph; Mumbai: U. Thatte, R. Kulkarni, G. Koppikar; Brits: A. Gous, E. Pochee; Durban: A. Gray, S. Essack, A. Sturm.
Objective To investigate the feasibility of surveillance of antimicrobial use in the community in resource-constrained settings. Overuse and misuse of antimicrobial medicines is contributing to the development of resistance. The WHO Global Strategy for Containment of Antimicrobial Resistance recommends surveillance of use at all levels of the health sector but this is not done in most low and middle income countries.
Methods Pilot projects were established in three sites in India (Delhi, Mumbai and Vellore) and 2 in South Africa (Brits and Durban). Antimicrobial use data were collected monthly from both public and private facilities. Each pilot site sought to document 30 patient encounters where antimicrobials were provided from 7 to 30 facilities per month. Antimicrobial use was expressed as the percentage of patients receiving a specific antimicrobial and as the number of defined daily doses of each specific antibiotic per 100 patients attending the facility per month.
Results In all sites, there was extensive use of antimicrobials, with older agents being used more in the public sector and newer agents in the private sector. Although methodological differences limit the comparability of data, use appeared to be higher in India than in South Africa. Expressing antimicrobial use as the percentage of patients receiving a specific antimicrobial was more easily computed. Defined daily dose measure was useful in demonstrating differences in dosing and duration.
Conclusion All pilot sites provided data on antimicrobial use but also raised several issues related to methodology and logistics of long-term surveillance in community settings under resource constraints. Use measured as percentage of prescriptions is easier and more reliable in these settings.
Objectif: Investiguer la faisabilité de la surveillance de l’utilisation des antimicrobiens dans la communauté dans les endroits à ressources limitées.
Méthodes: Des projets pilotes ont été mis en place dans 3 sites en Inde (Delhi, Mumbai, Vellore) et 2 en Afrique du Sud (Brits, Durban). Les données sur l’utilisation des antimicrobiens ont été collectées chaque mois à la fois dans les établissements publics et privés. Chaque site pilote avait pour objectif de documenter 30 rencontres avec patients à qui des antimicrobiens ont été fournis dans 7 à 30 services par mois. L’utilisation d’antimicrobiens a été exprimée comme le pourcentage de patients recevant un antibiotique spécifique et comme le nombre de doses quotidiennes déterminées de chaque antibiotique spécifique par 100 patients fréquentant le service par mois.
Résultats: Dans tous les sites, il y avait une utilisation considérable des antimicrobiens, avec les plus vieux agents antibactériens plus utilisés dans le secteur public et les nouveaux dans le secteur privé. Bien que des différences méthodologiques limitent la comparabilité des données, l’utilisation semble être plus répandue en Inde qu’en Afrique du Sud. Exprimer l’utilisation des antimicrobiens comme le pourcentage de patients recevant un antibiotique spécifique a été plus facilement calculable. Définir la mesure de la dose quotidienne a été utile pour démontrer les différences dans le dosage et la durée.
Conclusion: Tous les sites pilotes ont fourni des données sur l’utilisation des antimicrobiens, mais ont aussi soulevé plusieurs problèmes liés à la méthodologie et à la logistique de la surveillance à long terme en milieu communautaire à ressources limitées. L’utilisation mesurée en pourcentage de prescriptions est plus facile et plus fiable dans ces milieux.
Objetivos: Investigar la viabilidad de la vigilancia del uso de antimicrobianos en la comunidad, en emplazamientos con recursos limitados.
Métodos: Se establecieron proyectos piloto en 3 emplazamientos en la India (Delhi, Mumbai, Vellore) y 2 en Sudáfrica (Brits, Durban). Los datos del uso de antimicrobianos se obtuvieron mensualmente de centros públicos y privados. Cada emplazamiento piloto documentó el encuentro con 30 pacientes a los que se les recetó antimicrobianos, en 7-30 centros por mes. El uso de antimicrobianos se expresó como un porcentaje de pacientes recibiendo un antimicrobiano específico, y como el número de dosis diarias de cada antibiótico específico por cada 100 pacientes que acudían al centro por mes.
Resultados: En todos los emplazamientos había un uso extensivo de antimicrobianos. En el sector público había un uso más extensivo de los antibióticos más antiguos, mientras que en el sector privado se utilizaban más los de últimas generaciones. Aunque las diferencias metodológicas limitan la posibilidad de comparar los datos, el uso parecía ser mayor en India que en Sudáfrica. Era más fácilmente computable el uso de los antibióticos expresado como el porcentaje de pacientes recibiendo un antibiótico específico. La dosis diaria definida era útil a la hora de demostrar diferencias en la dosificación y la duración.
Conclusión: Todos los emplazamientos piloto entregaron datos sobre el uso de antibióticos, pero también plantearon una serie de problemas relacionados con la metodología y la logística de una vigilancia a largo plazo en emplazamientos comunitarios con recursos limitados. El uso, medido como porcentaje de prescripciones, es el dato más fiable y de más fácil obtención en estos lugares.
Access to modern antimicrobials has enabled many health gains since the 1940s. However, the continued usefulness of these essential medicines is threatened by increasing resistance. As many as 70% of common bacterial infections are caused by organisms showing resistance to first-line antimicrobials WHO, 2005. The development of resistance is a natural response of bacteria to exposure to antimicrobials and is thus driven by use. Information on antimicrobial use in the community is of particular importance, as it has been estimated that fewer than 40% of primary-care patients in the public sector and fewer than 30% in the private sector are treated in compliance with guidelines WHO, 2009.
In 2001, WHO developed a Global Strategy for Containment of Antimicrobial Resistance. One of the strategies recommended was the establishment of effective, epidemiologically sound surveillance of antibacterial medicine (ABM) use and antimicrobial resistance (AMR) among common pathogens in the community, hospitals and other health-care facilities. This was subsequently confirmed in a World Health Assembly Resolution in 2005 (WHA58.27).
Reliable systems and methods to collect and analyse ABM use data already exist in several developed countries Bronzwaer et al., 2002; Ansari et al. 2009; Muller et al., 2007. However, the infrastructure, resources and feasible standard methods needed to implement such surveillance systems are unavailable in resource-constrained settings. Developing validated, reproducible and sustainable surveillance methodologies to quantify both ABM use and AMR in resource-constrained communities is therefore a priority. This report is based on experiences from five WHO-supported pilot projects designed to explore different methods for undertaking community-based surveillance of ABM use in resource-constrained settings. The full methods and results used in each site have been reported elsewhere Holloway et al. 2009.
Five pilot sites, three in India (Delhi, Mumbai and Vellore) and two in South Africa (Brits and Durban), were chosen, based on expression of interest and existing capacity to undertake microbiological and antibiotic use surveillance. The three Indian sites were selected and coordinated with the help of Delhi Society for Promotion of Rational Use of Drugs (DSPRUD). The two South African sites were selected with the help of the South African Drug Action Programme (SADAP). All sites were in urban areas and attached to large hospitals, although two (Vellore and Brits) were able to access material from more rural settings. A framework protocol was devised with each site being given the freedom to adapt this locally.
Sites used one of the following two basic approaches – prospective data collection by exit patient interviews conducted at health facilities or retrospective data collection from facility-retained records. The latter data were of two types: prescription data and/or bulk sales/purchase data. Where bulk purchase data were used, it was assumed that the same amount purchased by the facility was sold/dispensed to patients in the period under consideration.
Data collection was performed by pharmacists or by field workers specifically trained for this task. The following information was collected: the number of prescriptions with an ABM, the type of ABM used (allowing allocation to an anatomical therapeutic chemical (ATC) code) and dose and quantity of the ABM use (allowing for the calculation of the number of defined daily doses (DDDs) prescribed) World Health Organization, 2008. Data on the appropriateness of choice were collected where available.
All ABMs other than those used specifically for tuberculosis and leprosy were considered as ABM. Some sites considered metronidazole, as an ABM, while others did not. To understand overall community ABM use, data had to be collected from various facility types – public and private sector facilities such as primary health clinics, private general practices, outpatient clinics of public sector and private hospitals, and private pharmacies. A multistage sampling process was thus necessary. In the first stage, clusters (facility types prescribing or dispensing ABMs) were identified in the private and public sectors. In the next step, the facilities to be included from each cluster were decided upon. The final choice of facilities in most sites was based on feasibility and the ability to obtain consent. At each facility, the numbers of prescriptions containing ABM to be examined per month per facility were determined. Each site had to justify the type of facilities included, based on the type and numbers of facility that existed in their own communities.
Sample sizes were based on the WHO/INRUD methodology for investigating drug use in health facilities World Health Organization, 1993. According to this method, 30 prescribing encounters per facility from 20 facilities are required to measure drug use indicators in a representative group of facilities. For this study, each site aimed to collect 30 ABM-containing prescriptions from each facility every month and to include at least 20 facilities, with at least seven facilities of any one type.
Antibacterial medicine use was expressed in two complementary ways. The first, drawing on the WHO/INRUD method, was the percentage of prescriptions containing a specific ABM. While this indicated the prevalence of ABM use, it did not provide information on the quantity of ABM used. This factor is considered to be important in determining the degree of ecological pressure driving AMR. A more specific measure of utilization, the number of DDDs prescribed per unit of population, was therefore used. The total number of patients seen in the time required to collect the ABM-containing prescriptions was used as the denominator (unit of population). Utilization was therefore expressed as the number of DDDs of a specific ABM prescribed per 100 patients seen, over a particular period of time. ATC and DDD data were obtained from the website of the WHO Collaborating Centre for Drug Statistics Methodology World Health Organization, 2008.
Ethical approval for the study in each site was obtained from the WHO Ethics Review Committee and from the ethics committee for each institution participating in the study.
Table 1 shows the extent and sources of prescription data collected from each site. In both sites in South Africa, retrospective data were collected from prescription or clinic records. While both succeeded in collecting data from public sector facilities, only Durban managed to collect data from private general practitioners (GPs) and private pharmacies. In Brits, data were only collected from public facilities and one non-governmental organization (NGO) facility, owing to lack of cooperation from the private-for-profit pharmacies. In all three sites in India, because of the lack of facility-retained records, data were collected prospectively by exit patient interviews. In Mumbai and Vellore, data were collected from public sector facilities, private GPs and private pharmacies. In Delhi, owing to resource constraints, data were collected only from private pharmacies. In addition, Delhi and Vellore collected bulk purchase data, one site expressing this as the number of DDDs of specific ABM purchased per 100 patients attending the facility, and the other expressing usage as the number of DDDs of specific ABM purchased per 1000 population.
Table 1. Data collection at the five pilot sites
*Two clinics were run by a non-governmental organization (NGO).
†One PHC facility was run by an NGO.
Patients or prescriptions assessed
Number with antibiotics
Types of facilities sampled
Private sector general practitioners (GPs)
Private sector hospitals
Public sector primary health-care (PHC) facilities
Public sector hospitals
While most sites chose facilities based on willingness to participate, Durban selected six public facilities and seven private pharmacies by random selection from sampling frames of 20 and 49 facilities, respectively. Mumbai selected a different random sample of 10 private pharmacies and 10 private GPs from a list of 55 pharmacies and 90 GPs, respectively, each month (excluding facilities visited in the previous 3 months) to gain cooperation from private sector providers over a long period of time.
In the Indian sites, where exit interviews were used, one data collector counted all the patients attending the facility and identified those receiving an ABM, while another data collector interviewed all those patients receiving an antibiotic. In the South African sites, numerator and denominator data were collected by a single fieldworker, from available records.
In Durban, the data from public sector clinics were extracted from daily statistical return sheets, or ‘tick registers’, in dispensing practitioner settings from non-computerized clinical records, and in pharmacies from computerized prescription records. All prescriptions or patient records for the 2 weeks prior to the visit were examined. In Brits, data were extracted from patient records. At all sites, data from facility-retained records or patient interview were captured on pre-designed forms.
Figure 1 shows the use of specific ABMs across the sites, expressed either as the percentage of patients/prescriptions or as the number of DDDs per 100 patients. In all sites, the older antimicrobials (e.g. cotrimoxazole, amoxicillin and tetracyclines) appeared to be used more in the public sector, whereas more recently marketed antimicrobials (e.g. fluoroquinolones and cephalosporins) were used more extensively in the private sector. Figure 2 shows a sample of monthly use data for selected ABM in two Indian sites (Vellore and Mumbai). In Vellore, the data were obtained from 13% (4/30) public and 87% (26/30) private facilities. If the ‘not for profit private’ facilities were considered to be ‘public’, then the data would ‘represent’ 20% (6/30) public and 80% (24/30) private use. In Mumbai, data were obtained from nine public PHC facilities, 1 public hospital, 10 private GPs and 10 private pharmacies, ‘representing’ 33% (10/30) public and 67% (20/30) private use. A degree of monthly variation was seen, with some indication of higher use of ABM in the winter months. The apparently higher use of quinolones and lower use of co-trimoxazole in Vellore when compared to Mumbai, seen in Figure 2, is likely to be due to the greater proportion of private facilities in the Vellore sample than in Mumbai’s.
The sample numbers decided in the protocols were often not obtained in most sites. There were wide variations in the numbers of prescriptions obtained between months and facility types in some sites. The data quality also varied, especially in relation to the more detailed data required for calculating the number of DDDs per 100 patients. These data were more prone to error than the data for calculating percentage of ABM-containing prescriptions. All sites complained that facilities, particularly private ones, became fatigued with monthly data collection over 1–2 years. The Mumbai site, which randomly selected different private providers each month, did not report less fatigue than the other Indian sites. Details of results from each site have been presented elsewhere Holloway et al. 2009.
This attempt, probably the first of its kind to establish pilot sites to collect data on ABM use in the community, allowed recognition of several practical issues related to community-based surveillance in resource-poor settings. Although the study did not set out to compare ABM use between sites, it was noted that the three Indian sites appeared to have much higher ABM use compared with the two South African sites in all types of facilities. However, there were substantial differences in data collection methods, so direct comparison between sites cannot be made, nor was this an objective of the project.
Because large databases like those used in some ABM surveillance in developed countries are generally unavailable in resource-poor settings, sampling was necessary. Deciding on a representative sample for community use and also obtaining the desired numbers was challenging. A multistage approach, as was carried out in this project, is included in the current recommendations on assessing medicine use in communities Hardon et al. 2004.
To decide on the number of facilities (second step) per cluster, it would seem appropriate to consider the care-seeking behaviour of the population in the area. For example, if 80% of the population seeks health-care in the public sector, 80% of the facilities selected might need to be in the public sector. However, establishing this figure with some amount of certainty might be complex. For example, while more than 80% of the South African population is uninsured and might therefore be considered to be ‘public sector dependent’, they may also resort to out-of-pocket purchase of health services and medicines in the private sector. However, even if the proportion of public vs. private sector facilities sampled does not reflect the true care-seeking behaviour of the community, trends in use can be followed and the impact of interventions evaluated, if data are collected the same way over time. In the pilot study, random sampling of facilities was possible in only two sites (Durban and Mumbai), but this should not detract from the quality of data obtained from convenience samples over time.
The method in which a sample of 30 ABM-containing prescriptions was to be examined per facility per unit of time was initially designed for cross-sectional surveys. A more precise sample number could be derived for time series analyses using the data generated in this study. The variations observed in the numbers actually collected per month at some sites could also be minimized by regular supervision.
Both prospective data collection from exit patient interview and retrospective data collection from record review proved acceptable. Exit interviews were resource intensive but could capture over-the-counter (OTC) sales of ABM although the presence of interviewers probably influenced provider behaviour through the Hawthorn effect. While this effect may diminish over time, it probably did reduce OTC sales (which were much lower than expected and thus not reported here) during the study period. In South Africa, accessing of facility-retained records may have been less resource intensive, but more cooperation was needed from the facilities, and it was not possible to capture OTC sales of ABM. In some cases, the records proved incomplete, which probably affected the results at least in one site (Brits). Changing sites every month (Mumbai) did not appear to reduce fatigue and may have contributed to greater variability in results.
Both measures of use – percentage of prescriptions containing an ABM and DDDs of an ABM prescribed per 100 patients – were useful in understanding different aspects of ABM use. Neither measure takes account, however, of the fact that individuals may not always purchase the total amount prescribed.
It was easier to collect the data required to calculate the percentage of prescriptions with ABM. DDDs required much more data collection, thus allowing more room for error, which may have accounted for the greater fluctuation in use over time using this outcome measure (Figure 2). Data on the percentage of patients receiving an ABM can be used to understand preferences and pattern of ABM use and to make comparisons between time points. However, to estimate the ABM exposure in an area, the quantity of ABMs used should be calculated as DDDs per unit population per unit time. Ideally, the total population served by the facilities should be the denominator. Because there was no defined catchment population for facilities at the sites used in the pilot projects, this number was difficult to obtain. As can be seen in Figure 2, ABM use measured by the two methods was similar over time. It may be therefore more practical and also sufficient to express ABM use as percentage in these settings, provided doses and durations of specific ABMs do not vary considerably over time. Furthermore, DDDs per patient per day do not indicate whether patients are treated in compliance with guidelines. It may be more useful to undertake ad hoc surveys on prescribing habits, in addition to surveillance, to develop interventions to improve the use of antibiotics by practitioners and patients.
In general, it appeared that, by both outcomes measures and in all sites, older antibiotics (cotrimoxazole, amoxicillin and tetracyclines) were used more in the public sector and newer antibiotics (fluoroquinolones and cephalosporins) more in the private sector. Both in India and in South Africa, the higher fluoroquinolone and cephalosporin use in the private sector was more evident when use was expressed as DDD per 100 patients. In the South African public sector, this was particularly true of the fluoroquinolones, where single doses of ciprofloxacin were prescribed for syndromic management of sexually transmissible infections. However, it must be emphasized that any differences seen might have been affected by the varying methods employed.
Only a limited number of studies have reported on ABM use. The results of 35 country studies from 1988 to 2002, which followed the standard WHO/INRUD methodology, were reported in the World Medicines Situation WHO, 2004. In general, surveys performed in African countries showed lower ABM use than did those performed in parts of Asia as was seen in this study as well. The results obtained in this study in terms of the percentage of patients receiving ABMs are similar to other studies in India, carried out both in public sector and in private sector facilities WHO, 2009, 2004; Chaudhury et al. 2005; Rishi et al. 2003; Indira 2004; Karande et al. 2005; Mhetre et al. 2003;. Similarly, our results for public sector facilities in South Africa were similar to other studies conducted in the past Fairall et al. 2005; Directorate, Department of Health; Moller et al. 1997;. Concern has also been expressed about the appropriateness of ABM use in South Africa’s private sector Katende-Kyenda et al. 2006; Haupt et al. 2002.
The collection of bulk sales or distribution data would appear to hold many advantages. However, access to such data requires the full cooperation of commercial and governmental establishments. The choice of a suitable denominator for such data is also more problematic. Where the number of sampled facilities is far smaller than the total in the geographical area, use of the total population (for example, in a municipal ward) may not be accurate, although it might be useful for depicting trends over time in the same setting. However, where data from a centralized warehouse that services all public sector in a large geographical area (such as a province), where the total number of public sector–dependent persons in that area is known with some accuracy, then that number can be used with more confidence. The alternative method, in which the number of patients seen in a particular time period is extrapolated to the full operating time of the facility in a month, is highly dependent on the generalizability of the sampled time period.
Fatigue in data collection may result in deterioration of data quality over time. Therefore, data management is an important aspect of long-term surveillance and requires careful planning and implementation. Attention to training and supervision of data collectors with well-designed data entry formats and committed local supervision are essential to ensure good-quality data. Commitment at the political level and by health authorities is critical for such projects.
Community-based surveillance in five resource-constrained settings resulted in useful ABM data. Long-term surveillance of ABM use is difficult, even in well-resourced settings, and is extremely challenging in resource-poor settings. Lessons learnt from these pilot sites could prove to be extremely valuable in developing a robust methodology for undertaking the surveillance necessary to inform actions to improve ABM use.
We would like to acknowledge the following persons who contributed to the work: A.Sharma, R. Raveendran, S. Bhandari, S. Khanna, R. Chaudhury, N. Shinkre, B. Antonisami, M. Leteka, E. Crous, J. Tumbo, M. Tumbo, B. Milanzi, P. Mtshali, L. Combrink, P. Sein, F. Deedat, T. Pillay and J. van Maasdorp. The pilot projects described in this publication were financed by the US Agency for International Development (USAID), USA. The views expressed herein are those of the authors and can therefore in no way be taken to reflect the official opinion of the US Agency for International Development.