- Record: found
- Abstract: found
- Article: found
Lives saved by Global Fund-supported HIV/AIDS, tuberculosis and malaria programs: estimation approach and results between 2003 and end-2007
research-article
Author(s):
Ryuichi Komatsu
1
,
,
Eline L Korenromp
1 ,
Daniel Low-Beer
1 ,
Catherine Watt
2 ,
Christopher Dye
2 ,
Richard W Steketee
3 ,
Bernard L Nahlen
4 ,
Rob Lyerla
5 ,
Jesus M Garcia-Calleja
6 ,
John Cutler
1
,
7 ,
Bernhard Schwartländer
8
,
1
Publication date (Electronic):
30 April 2010
Journal:
BMC Infectious Diseases
Publisher:
BioMed Central
Read this article at
There is no author summary for this article yet. Authors can add summaries to their articles on ScienceOpen to make them more accessible to a non-specialist audience.
Abstract
Background
Since 2003, the Global Fund has supported the scale-up of HIV/AIDS, tuberculosis and malaria control in low- and middle-income countries. This paper presents and discusses a methodology for estimating the lives saved through selected service deliveries reported to the Global Fund.
Methods
Global Fund-supported programs reported, by end-2007, 1.4 million HIV-infected persons on antiretroviral treatment (ARV), 3.3 million new smear-positive tuberculosis cases detected in DOTS (directly observed TB treatment, short course) programs, and 46 million insecticide-treated mosquito nets (ITNs) delivered. We estimated the corresponding lives saved using adaptations of existing epidemiological estimation models.
Results
By end-2007, an estimated 681,000 lives (95% uncertainty range 619,000-774,000) were saved and 1,097,000 (993,000-1,249,000) life-years gained by ARV. DOTS treatment would have saved 1.63 million lives (1.09 - 2.17 million) when compared against no treatment, or 408,000 lives (265,000-551,000) when compared against non-DOTS treatment. ITN distributions in countries with stable endemic falciparum malaria were estimated to have achieved protection from malaria for 26 million of child-years at risk cumulatively, resulting in 130,000 (27,000-232,000) under-5 deaths prevented.
Conclusions
These results illustrate the scale of mortality effects that supported programs may have achieved in recent years, despite margins of uncertainty and covering only selected intervention components. Evidence-based evaluation of disease impact of the programs supported by the Global Fund with international and in-country partners must be strengthened using population-level data on intervention coverage and demographic outcomes, information on quality of services, and trends in disease burdens recorded in national health information systems.
Related collections
Most cited references31
- Record: found
- Abstract: found
- Article: not found
Patient Retention in Antiretroviral Therapy Programs in Sub-Saharan Africa: A Systematic Review
Sydney Rosen, Matthew Fox, Christopher J Gill (2007)
Introduction In the half decade since the first large-scale antiretroviral treatment (ART) programs for HIV/AIDS were launched in sub-Saharan Africa, much attention has focused on patients' day-to-day adherence to antiretroviral (ARV) medications [1–3]. Long-term retention of patients in treatment programs, a prerequisite for achieving any adherence at all, has received far less attention. Perhaps because most large scale treatment providers have few resources available to track missing patients, most studies treat patient attrition as a side issue and focus solely on describing those patients who are retained. Moreover, adherence can be assessed over very short periods, whereas long-term retention requires, by definition, long-standing programs. Attrition from antiretroviral treatment programs is generally divided into four categories. The two most common are (1) the death of the patient—several studies have reported high rates of early mortality—and (2) “loss to follow-up,” a catch-all category for patients who miss scheduled clinic visits or medication pickups for a specified period of time. Some patients remain in care but stop taking ARV medications (3). Others transfer to other facilities and continue on ART (4). Treatment discontinuation raises some of the same concerns about drug resistance that incomplete adherence does and, even worse, negates much of the benefit sought by those implementing treatment programs. Patients with clinical AIDS who discontinue ART will likely die within a relatively short time [4]. High rates of attrition from treatment programs thus pose a serious challenge to program implementers and constitute an inefficient use of scarce treatment resources. In this study, we analyzed reported treatment program retention and attrition in sub-Saharan Africa in order to document the magnitude of the problem and help policy makers and program managers address the challenge of patient retention. Methods Definitions For this review, “retention” refers to patients known to be alive and receiving highly active ART at the end of a follow-up period. “Attrition” is defined as discontinuation of ART for any reason, including death, loss to follow-up, and stopping ARV medications while remaining in care. Transfer to another ART facility, where reported, is not regarded as attrition—patients who transfer are assumed to be retained. We accepted the varying definitions of loss to follow-up used by the respective studies. Many studies considered patients lost if they were more than 3 mo late for a scheduled consultation or medication pickup, but some studies used more or less stringent definitions ranging from 1 to 6 mo late for a scheduled consultation or medication pick-up. Inclusion and Exclusion Criteria Studies were included in the review if they reported the proportion of adult HIV-1 patients retained in highly active ART programs implemented in service delivery (nonresearch) settings in sub-Saharan Africa. All patients who initiated ART had to be included in the report, not just those still in care at the time of censoring (i.e., only intention-to-treat analyses were included). Clinical trials, including Phase 3 trials, were excluded, although some subjects of reviewed studies transferred into the treatment program from a clinical trial. A median follow-up period of at least six full months (26 weeks) was also required. Studies that reported mortality but not other categories of attrition and studies that reported only on-treatment analyses, or where we were unable to determine whether the study was intention-to-treat or not, were also excluded. A few of the reviewed studies did not differentiate between adult and pediatric patients; those that considered only pediatric patients were excluded. Search Strategy To identify eligible studies, we conducted a systematic search of the English-language published literature, gray literature (project reports available online), and conference abstracts between 2000 and 2007. The search included Ovid Medline (1996 to July 2007), EMBASE (inception to July 2007), ISI Web of Science (August 2002 to July 2007), the Cumulative Index to Nursing & Allied Health Literature (2002 to July 2007), and the Cochrane Database of Systematic Reviews (inception to second quarter 2007). We also searched the abstracts of the conferences of the International AIDS Society (inception to 2006), the Conference on Retroviruses and Opportunistic Infections (inception to 2007), the HIV Implementers' Meetings (2006–2007), and the South African AIDS Conference (2005–2007). The bibliographies of five recently published reviews of treatment outcomes, mortality, or ARV adherence in resource-constrained settings were also searched [1–3,5,6]. Our search strategy combined the terms “antiretroviral” and “Africa” or “developing countries” with each of retention/attrition/loss to follow-up/mortality/evaluation/efficacy. When more than one source reported on the same cohort of patients, the source containing the most detailed data about retention and attrition or the longest follow-up period was selected for the review. Although non-English databases were not searched, English-language abstracts of non-English papers identified in our search were included. Eligible studies were identified by the first author (SR) and eligibility confirmed by the other authors (MF and CG). It should be noted that the Antiretroviral Therapy in Low Income Countries (ART-LINC) collaboration has recently reported aggregate 1 y mortality and loss to follow-up rates for 13 cohorts in sub-Saharan Africa [5]. Some of the patients in these cohorts are included in the studies reviewed here. To avoid duplication, findings from the ART-LINC cohorts were not included in this analysis but are noted in the discussion. Data Analysis Most studies reported patient attrition at months 6, 12, and/or 24 after treatment initiation. We therefore used these same intervals in this analysis. For papers that reported on intervals other than 6, 12, or 24 mo, we classified the reported attrition rate using the nearest time point. If the report did not list attrition rates by time, but did list a median duration of observation, we estimated attrition at the 6, 12, or 24 mo interval closest to the reported observation period. In some cases, follow-up periods and/or retention rates were calculated by the authors using data provided in the article or extracted from figures (e.g. Kaplan-Meier survival curves). Where appropriate, we calculated weighted averages for demographic features of the cohort participants or other factors related to the studies. For proportions, averages were weighted by the inverse of their variances [1 ÷ (p × [1 − p] ÷ n), where p is the proportion and n is the sample size]. Because we did not have the individual patient data for continuous variables nor their standard deviations, we were unable to calculate variances for these variables. In these situations, we weighted by cohort size. In some instances, studies reported follow-up to 12 or 24 mo but did not report on intermediate retention rates. In plotting attrition for such studies over time we used extrapolated values, taking the midpoint between the known adjacent values. For example, if a study reported to 12 mo but did not report the 6 mo value, we defined the 6 mo value as the midpoint between 0 and 12 mo, with 100% at baseline representing all of those who initially started therapy. We calculated weighted average attrition rates at each interval (6, 12, and 24 mo) for the reported numbers of participants remaining when using reported values and for the estimated numbers of participants remaining when using extrapolated values. Selected demographic variables relating cohort or program characteristics to attrition rates were analyzed using linear regression. Because only a few studies reported beyond 24 mo, we were unable to calculate any meaningful summary statistics beyond the 24 mo mark. To estimate aggregate average attrition rates at 6, 12, and 24 mo we used several approaches. Attrition for each program was plotted separately and attrition rates calculated as the percentage of patients lost per month. We also plotted Kaplan-Meier survival curves using the 6, 12, and 24 mo intervals as the step-down points. Fewer studies presented 12 mo data than 6 mo data, however, and fewer still contained 24 mo data. Many of the studies with the highest attrition contributed data only for the shorter time intervals. Given the concern that shorter durations of reporting could be associated with lower rates of patient retention, we also conducted sensitivity analyses to model possible future retention. For the best-case scenario, we optimistically assumed that no further attrition would occur beyond the last reported observation and extrapolated the last reported retention value forward to 24 mo. In the worst-case scenario, we extrapolated the slope of attrition forward in time, assuming that each cohort's attrition would continue along the same slope from the last reported observation to month 24. We assigned a lower limit of 0% in those situations where the estimated future retention rate fell below 0%. Our midpoint scenario was the mean of the best- and worst-case scenarios. Analyses were conducted using Excel, SAS version 8.2, and SPSS version 11.0. Results We included 32 publications reporting on 33 patient cohorts totaling 74,289 patients in 13 countries in our analysis. These studies were selected from a total of 871 potentially relevant, unique citations identified in our search (Figure 1). Figure 1 Study Flow Chart Table 1 summarizes key features of the studies, including the sites at which they were conducted. Not all of the publications reported all the details we sought about program and patient characteristics and retention, but all provided at least one indicator of patient retention after a median follow-up period of at least 6 mo. The studies report on patients who initiated ART as early as 1996, though most enrolled their cohorts between 2001 and 2004. The studies were published or presented between 2002 and 2007, with the majority appearing as peer reviewed articles in 2006 or 2007. Most of the programs were implemented by the public sector (17 of 33, 52%). Of 33 cohorts, 15 (45%) fully subsidized the cost of ART; six (18%) were partially subsidized; and six (18%) required patients to pay fully for their care; the rest did not report their payment structure. Roughly half were single-site programs (15 of 33, 45%); multi-site programs contributed data from between two and 69 sites. Table 1 Characteristics of Antiretroviral Treatment Programs and Patient Cohorts Included in This Analysis (extended on next page) Table 1 Extended. Table 1 also provides the population characteristics of the cohorts studied. The weighted mean age of the cohort participants was 35.5 y, and 53.7% of all patients were female (range 6%–70%). All but one cohort had median starting CD4+ T cell counts at or well below 200 × 103 cells/mm3, with a weighted mean starting CD4+ T cell count of 132 × 103 cells/mm3 (range 43–204). Table 2 presents the proportion of patients from each cohort who remained alive and under treatment with antiretroviral medications, transferred to another treatment facility, died, were lost to follow-up, or discontinued treatment with ARVs but remained in care at the end of the median follow-up period. Bearing in mind that we excluded studies with less than 6 mo median follow-up, the weighted average follow-up was 9.9 mo, after which time overall retention of patients alive, in care, and on ART was 77.5%. Table 2 Median Follow-Up and Rates of Patient Attrition, as Reported, from Antiretroviral Treatment Programs Across all the cohorts, the largest contributor to attrition was loss to follow-up (56% of attrition), followed by death (40% of attrition). The widely varying definitions of loss to follow-up used by the studies are indicated in Table 2. A small fraction (4% of attrition) discontinued ART but remained under care at the same site. Table 3 reports overall retention at 6, 12, and 24 mo. SA 1 had the highest retention at 12 mo. While this program did not report for 6 mo, at 12 mo its retention of 90% was still higher than the highest reported value among the programs that reported their 6 mo outcomes. The programs with the lowest retention at each time point were Malawi 4 (55%) at 6 mo; Uganda 2 (49%) at 12 mo; and Uganda 1 (46%) at 24 mo. Malawi 4 did not report beyond 6 mo and Uganda 2 did not report beyond 12 mo, but were both on a trajectory toward even lower retention rates at the later time points. Table 3 Retention of Patients at 6, 12, and 24 Months after Initiation of ART Using linear regression, we found no association between 6 mo attrition rates and cohort size (p = 0.32), attrition and baseline CD4+ cell counts (p = 0.72), proportion of women (p = 0.23), or year of program initiation (p = 0.40). Programs that required no payment had higher retention rates at 6 mo compared to those requiring partial or full payment (86.5% versus 76.7%, p = 0.01). Figures 2A–2C plot attrition rates for each cohort separately. The studies are clustered on the basis of duration of reporting. By 6 mo, 9 of 33 cohorts (27%) had 20% or greater attrition rates; by 12 mo this proportion had risen to 16 of 25 reporting cohorts (64%). Figure 2 Attrition Rates by Reporting Duration (A) Studies reporting to 6 mo median follow-up. (B) Studies reporting to 12 mo median follow-up. (C) Studies reporting to 24 mo median follow-up. (D) Weighted mean attrition rates by duration cohort. SA, South Africa. The weighted mean retention rates as reported in the studies were 79.8% at 6 mo, 75.1% at 12 mo, and 61.6% at 24 mo. As an alternative approach, we also plotted Kaplan-Meier survival curves at months 6, 12, and 24 for all the studies combined. The largest fall-off occurred between 6 and 12 mo; overall retention was approximately 89% by 6 mo, 70% by 1 year, and just under 60% by 2 years. Four of the eight studies in Figure 2A with attrition of at least 20% at 6 mo included data only to 6 mo. Similarly, of the cohorts with data at 12 mo and attrition of 25% or more, six of 11 did not extend beyond 12 mo. We therefore calculated the slopes for attrition rates for each group of cohorts in Figure 2A–2C separately, to determine if the average monthly attrition rates differed as the duration of reporting increased. As shown in Figure 2D, the weighted mean attrition rates were 3.3%/month, 1.9%/month, and 1.6%/month for studies reporting to 6 mo, 12 mo, and 24 mo, respectively, raising the possibility that shorter durations of reporting were associated with lower retention rates. Given this apparent reporting bias, we were concerned that reporting average retention rates using the simple aggregate weighted averages reported above would overestimate actual retention. We therefore conducted sensitivity analyses to model attrition rates under three different scenarios. As shown in Figure 3, all three scenarios are the same at 6 mo, with approximately 80% retention. Under the best-case scenario, further attrition would be negligible, with more than 76% still retained by the end of 2 y. Under the worst-case scenario, 76% of patients would be lost by 2 y. The midpoint scenario predicted patient retention of 50% by 2 y. Figure 3 Sensitivity Analysis for Attrition Discussion The analysis presented here suggests that ART programs in Africa are retaining, on average, roughly 80% of their patients after 6 mo on ART and between one-fourth and three-fourths of their patients by the end of 2 y, depending on the estimating method used. Prior to the availability of ART in Africa, the median interval from HIV infection to AIDS-related death was under 10 y; once a patient was diagnosed with AIDS, median survival was less than 1 y [7]. Since most patients in Africa initiate ART only after an AIDS diagnosis, most ART patients would have died within a year had antiretroviral therapy not been available. Each patient who is retained in care and on ART can thus be regarded as a life saved and a source of tremendous benefit to patients' families and communities. For those who have struggled to launch and expand treatment programs in resource-constrained settings, reaching a 60% patient retention—and thus survival—rate after two years of treatment, as estimated by the Kaplan-Meier survival analysis, in just a few years' time is an extraordinary accomplishment. It is also noteworthy in the global context: in developed countries, adherence to medication for chronic diseases in general averages only 50% [8]. Similarly, treatment completion rates for tuberculosis, which requires a temporary rather than permanent commitment to adherence and a less demanding dosing schedule, average 74% in the African region, with a range among countries from 22% to 94% [9]. Taken in the context of medication adherence in general, the record of African ART programs lies within the bounds of previous experience. At the same time, however, losing up to half of those who initiate ART within two years is cause for concern. From the data as reported, attrition averaged roughly 22% at 10 mo of follow-up. This average comprised mainly deaths (40% of attrition) and losses to follow-up (56%). In comparison, the ART-LINC Collaboration, which analyzed data from 18 cohorts across the developing world, reported loss to follow-up rates among the 13 sub-Saharan African cohorts averaging 15% (range 0%–44%) in the first year after initiation; mortality averaged 4.2% across all 18 cohorts (African regional rate not provided) [5]. On the basis of our survival and sensitivity analyses, we believe that actual attrition is higher than the 22% average we report, mainly because the programs with the highest attrition were least likely to provide data beyond the first 6 mo of ART. There are several plausible explanations for the higher attrition seen among programs with shorter durations of reporting. One possibility is that limited availability of resources to a given program could affect both its ability to retain patients and to conduct long-term surveillance of its outcomes. Another, less pessimistic explanation is that shorter durations of reporting reflect newer programs that are still in the process of developing optimal strategies for patient retention: had they reported at a later point in their implementation, retention rates might have been higher. The magnitude of the under-reporting bias is also uncertain, although our sensitivity analysis gives a plausible range between two implausible extremes (the best case being implausible because it assumes zero further attrition beyond the point of last reporting, and the worst case because it assumes that there will be constant attrition over time, rather than reaching a plateau or at least slowing substantially). The midpoint scenario suggests that approximately half of all patients started on ART were no longer on treatment at the end of two years. One of the principal challenges to this analysis is interpreting the large proportion of attrition from “loss to follow-up.” Some of these patients undoubtedly represent unrecorded deaths, but others may be patients who identified alternative sources for ART or had taken an extended “break” from therapy, to which they will return when their condition worsens again or they obtain the financial resources needed for transport or clinic fees. One study in Malawi discovered, for example, that 24% of patients originally recorded as lost to follow-up re-enrolled at the same site two years later when ART became free of charge [10]. For some of the studies included in this analysis, on the other hand, the unrecorded death explanation is more persuasive. For example, the Zambia 1 cohort of more than 16,000 patients reported 21% loss to follow-up after approximately 6 mo [11]. The scope and scale of this program means that it is the primary source of ART in Lusaka, making it unlikely that most of the estimated 3,300 lost patients could have found alternative sources of care. A recent attempt to trace lost-to-follow-up patients in Malawi determined that 50% had died, 27% could not be found, and most of the rest had stopped ART [12]. Because those reporting on these cohorts do not know what ultimately happened to patients categorized as lost to follow-up, high loss to follow-up rates can have varied interpretations. A good deal of research on barriers to adherence and reasons for treatment discontinuation has been published [13]. Important barriers to adherence include cost of drugs and/or transport, fear of disclosure or stigma, and side effects [14,15]. Some of these barriers can be addressed relatively easily, for example by providing transport vouchers to ensure that patients can attend the clinic; others, such as stigma, require more profound changes. In any case, high reported rates of loss to follow-up are a strong call to improve patient tracing procedures, to minimize the number of patients who fall into the difficult-to-address category of “lost, reason unknown.” Given that the long-term prognosis of ART patients is inversely related to starting CD4+ T cell counts [16,17], an additional issue to consider is the low median starting CD4+ cell counts reported by every one of the studies in this analysis. This problem has been identified previously, particularly in South Africa [18–20]. The analysis here makes clear that the problem is nearly universal in Africa and cuts across all types of treatment programs. It is evident in the high death rates reported by some studies after only a few months of follow-up, such as Malawi 3. There is a high degree of heterogeneity in retention rates between the different cohorts in our analysis and among categories of attrition. Some programs appear to have been highly successful in retaining patients, while others clearly struggled to do so. Some programs have suffered high mortality rates but low loss to follow-up, others the opposite. Early mortality, which may be largely due to the late stage at which many patients present for treatment, requires interventions different from those needed to address later loss to follow-up, about which very little is known. Interventions to address the various types of attrition must thus be tailored to local circumstances. The success of some programs with very high retention may provide examples that others can follow. The findings here can thus be seen as a part of an ongoing process to identify and solve problems within existing treatment programs, even as we expand their scope and launch new ones. Our analysis has a number of limitations, chiefly that incomplete reporting forced us to extrapolate some values. Extrapolating backward assumes that attrition rates are distributed linearly over time, which is unlikely to be the case. Evidence from this and other studies suggest that the highest attrition occurs during the first 6 mo. However, this limitation only pertains to the shape of the attrition curves, not to their final end points. Extrapolating forward, which we used only in the sensitivity analysis to establish the hypothetical “worst case” scenario, also suffers from this limitation, compounded by the fact that our confidence in the forward boundary is limited. In addition, our analysis is necessarily limited to publicly available reports and thus potentially subject to publication bias. Researchers may be less inclined to publish long-term outcomes from cohorts that have experienced very high early attrition. It is also likely that programs with better access to resources, both financial and human, are also better able to monitor, analyze, and publish their results. Our aggregate findings may thus represent the better-resourced programs in Africa. In conclusion, African ART programs are retaining about 60% of their patients in the first two years. This average masks a great deal of heterogeneity, however. At one end of the spectrum represented by the reviewed studies, two-year retention neared 90%; at the other end, attrition reached 50%. Better information on those who are lost to follow-up is urgently needed. Since losses to follow-up account for the majority of all attrition in more than half of the studies reviewed, the problem of attrition cannot be addressed effectively without better means to track patients. Only then can we address the pressing question of why patients drop out and what conditions, assistance, or incentives will be needed to retain them. Supporting Information Protocol S1 Search Protocol (85 KB DOC) Click here for additional data file.
- Record: found
- Abstract: found
- Article: not found
The Limits and Intensity of Plasmodium falciparum Transmission: Implications for Malaria Control and Elimination Worldwide
Introduction The magnitude of the public health burden posed by malaria worldwide [1] and its connection to poverty [2] has galvanized the international donor community to put malaria control high on the development agenda and helped leverage unprecedented additional financing for malaria endemic countries [3]. Progress toward agreed targets of intervention coverage has been slow [4–6], but recent evidence indicates a precipitous increase in access to effective drugs and prevention strategies in several countries [7–10]. In part, this renaissance in malaria control has served as a catalyst to revisit the possibility of malaria elimination in many regions and countries [11–14]. A changing malaria landscape requires an accurate spatial and dynamic description of malaria risk that maps the spatial extent and need for control and elimination over the coming decades. Such a map is conspicuous by its absence [15]. Here, we present the first detailed description of the global distribution of P. falciparum risk in 40 y [16,17] by using geopositioned assemblies of national surveillance of malaria risk, medical intelligence, biological models of transmission suitability, and surveys of parasite prevalence. The paper focuses on detailing the data sources and their adaptation for the malaria cartography necessary to guide current disease control, with an emphasis on how we define the spatial limits of stable and unstable P. falciparum risk worldwide. Methods Using Medical Intelligence to Define the Limits of P. falciparum Risk Many countries have information assembled from medical intelligence on the distribution of malaria risk within their national borders. This information is documented primarily in reports from national health information systems that define the annual numbers of confirmed parasite-specific local malaria infections by geographic unit, referred to classically as the annual parasite incidence (API) [18–21]. The API is generated from various combinations of active (fever surveys in communities where every person presenting with a fever is tested for parasite infection) and passive (reports from febrile patients attending the local health services) case detection, and usually expresses the combined results as the number infected per 1,000 people per annum (pa) [18–21]. The precision of these estimates of malaria incidence are highly variable, and with the exception of some countries where case identification is a primary control tool [22], these data cannot be used confidently to derive the public health burden posed by malaria [1,23–26]. They can, however, be a useful indicator of where local parasite species-specific malaria risk is likely or absent, and are particularly plausible when triangulated with other sources of medical intelligence, reported in international travel health guidelines or by national malaria control programmes. Malaria coordinating officers in the regional offices of the World Health Organization (WHO), responsible for the collation of national API data from member countries were contacted to obtain data reported nationally to the highest possible geographic administrative unit level on populations at risk and numbers of confirmed P. falciparum cases, for as many years as were available between 2002 and 2006. Among the countries in the American Regional Office, P. falciparum–specific API (PfAPI) data from national surveillance systems in Brazil, Colombia, Peru, and Honduras were obtained directly from personal communication with malaria specialists. The reported cases of confirmed P. falciparum malaria per 1,000 resident population were computed for each year by administrative level and averaged over the number of reporting years. Summary data were categorized as no autochthonous P. falciparum cases reported, <0.1 autochthonous P. falciparum cases per 1,000 people pa, and ≥0.1 autochthonous P. falciparum cases per 1,000 people pa. The threshold around 0.1 cases per thousand pa was used to provide some indication of unstable conditions versus more stable transmission. This threshold is consistent with previous uses of PfAPI during the Global Malaria Eradication Programme [27] and balanced against the confidence in the precision of reported PfAPI values (Protocol S1). Each PfAPI summary estimate was mapped by matching it to its corresponding first-, second-, or third-level administrative unit in a geographic information system (GIS; ArcView GIS 3.2, ESRI, 1999). Mapped PfAPI data were then compared to other sources of medical intelligence, notably national malaria control presentations at regional malaria meetings obtained from regional WHO malaria coordinators and from Web sites, published sources that described national malaria epidemiology, and international travel and health guidelines [28,29]. These combined approaches were particularly useful to identify mapped descriptions of risk defined at higher spatial resolution than those described by the PfAPI reported across large first-level administrative units. Details of all sources used are provided in Protocol S1. Defining the Biological Limits of P. falciparum Transmission Within the limits of risk described through PfAPI, environmental conditions suitable for transmission vary enormously. These variations can be captured at much higher spatial resolution than it is possible to define by stratifying risk at administrative unit levels. Climate-based determinants of parasite and vector development and survival were developed that impose biological constraints on the geographical limits of P. falciparum transmission. First, we used a combination of the temperature-dependant relationship between P. falciparum sporogony and the longevity of the main dominant vectors to estimate the proportion of vectors surviving parasite development (Protocol S2). Using mean monthly temperature records from a 30-arcsec (∼1 km) spatial resolution climate surface [30], the duration of P. falciparum sporogony was estimated for each synoptic calendar month, and those pixels where the duration of sporogony was 31 d or less were identified. The exception was small areas that potentially support the longer-lived Anopheles sergentii and A. superpictus, where 62 d were considered more appropriate biologically (Protocol S2). This resulted in 12 images with a binary outcome: P. falciparum sporogony could or could not be completed in the month. These images were then combined to identify the number of suitable months for P. falciparum transmission in a synoptic year. All pixels where the duration of sporogony exceeded 1 mo, or 2 mo for areas within the range of A. sergentii and A. superpictus, were masked since it was highly unlikely that transmission would occur. Second, there are areas within several malaria endemic countries where, despite temperature being suitable for sporogony, arid conditions restrict Anopheles development and survival [31]. Limited surface water reduces the availability of water bodies for oviposition. Moreover, low ambient humidity in arid environments further affects egg and adult survival through the process of desiccation [32]. The ability of adult vectors to survive long enough to contribute to parasite transmission and of preadult stages to ensure minimum population abundance is, therefore, dependent on the levels of aridity and species-specific resilience to arid conditions. To capture the influence of aridity on transmission we used the enhanced vegetation index (EVI) derived from the bidirectional reflectance-corrected MODerate-resolution Imaging Spectroradiometer (MODIS) sensor imagery, available at approximately 1-km spatial resolution [33,34] (Protocol S2). Temporal Fourier–processed, monthly EVI images were used to develop 12 monthly surfaces that reclassified EVI ≤ 0.1, assuming this corresponded to a good proxy for arid conditions [35,36]. Pixels were classified as suitable for transmission if their EVI values were higher than 0.1 for at least two consecutive months in an average year. This definition was based on the biological requirement, at optimum temperatures, of at least 12 d to complete vector development from egg to adult [37] and on the assumption that a second month is required for a sufficient vector population to establish and transmit malaria [38]. These reclassified aridity images were then overlaid in a GIS to produce 12 paired images. The 12 pairs were then combined to define pixels where conditions were suitable for transmission. The aridity mask was treated differently from the temperature-limiting mask to allow for the possibility, in arid environments, of highly over-dispersed transmission due to man-made water collection points and nomadic human populations transporting vectors and parasites [39–41]. A more conservative approach was taken, therefore, which down-regulated PfAPI risk by one class. In other words, extremely arid areas defined originally as at stable risk were stepped down to unstable risk and those classified initially as unstable to malaria free. Estimating Populations at P. falciparum Transmission Risk in 2007 The Global Rural Urban Mapping Project alpha version provides gridded population counts and population density estimates for the years 1990, 1995, and 2000, both adjusted and unadjusted to the United Nations' national population estimates [42]. We used the adjusted population counts for the year 2000 and projected them to 2007 by applying national, medium variant, intercensal growth rates by country [43], using methods previously described [44]. This resulted in a contemporary population density surface of approximately 1-km spatial resolution, which was combined with the climate-adjusted PfAPI risk surface to extract population at risk estimates using ArcView GIS 3.2 (ESRI, 1999). Describing Global Patterns of Parasite Prevalence We have described previously the rigorous process of identifying, assembling, and geolocating community-based survey estimates of parasite prevalence undertaken since 1985 [45]. These data were used here to define the ranges of P. falciparum parasite prevalence rates (PfPR) in areas of stable and unstable malaria risk by WHO region. We acknowledge that these geopolitical boundaries do not necessarily conform to ecological or biological spatial representations of malaria [46,47]. They do, however, represent coherent regions of collective planning and cooperation for malaria control. In an attempt to minimize epidemiologically unrealistic divides for summary purposes, we have combined the Southeast Asian (SEARO) and Western Pacific (WPRO), as well as the Eastern Mediterranean (EMRO) and European (EURO) regions. The American WHO region (AMRO) and the African WHO region (AFRO) were considered separately. PfPR estimates were reported in various age groupings. To standardize to a single, representative age range of 2–10 y, we applied an algorithm based on catalytic conversion models first adapted for malaria by Pull and Grab [48] and described in detail elsewhere [49]. The geolocated and age-standardized prevalence data (PfPR2−10) [45] were overlaid on the PfAPI risk surface to extract a corresponding PfAPI value. Results PfAPI Data and Medical Intelligence to Define Spatial Limits of Transmission The PfAPI data identified 87 countries at risk of P. falciparum transmission between 2002 and 2006, which we now consider as P. falciparum endemic countries (PfMEC) in 2007 (Protocol S1). PfAPI data were mapped to first, second, or third administrative level units across 41 PfMECs covering a total of 8,789 unique polygons. These data incorporate complete years between 2002 and 2006, including summaries of three consecutive years for 16 countries, two consecutive years for eight countries, and the most recent complete year for 17 countries (Protocol S1). No information was available for 46 countries; mostly those in Africa. The spatial representation of no risk, unstable (PfAPI < 0.1 per 1,000 people pa), and stable risk (PfAPI ≥ 0.1 per 1,000 people pa) of P. falciparum transmission globally is shown in Figure 1, top panel. Figure 1 P. falciparum Malaria Risk Defined by Annual Parasite Incidence (top), Temperature, and Aridity (bottom) Areas were defined as stable (dark-red areas, where PfAPI ≥ 0.1 per thousand pa), unstable (pink areas, where PfAPI < 0.1 per thousand pa), or no risk (light grey). The few areas for which no PfAPI data could be obtained, mainly found in India, are coloured in dark grey. The borders of the 87 countries defined as P. falciparum endemic are shown. Highland areas where risk was excluded due to temperature appear in light grey. The aridity mask excluded risk in a step-wise fashion, reflected mainly in the larger extents of unstable (pink) areas compared to the top panel, particularly in the Sahel and southwest Asia (southern Iran and Pakistan). Temperature and Aridity Masks to Constrain Limits of Transmission Within the PfAPI limits of stable transmission (PfAPI ≥ 0.1 per 1,000 pa) on the African continent, the areas with no temperature-suitable months for transmission were congruent with the high altitude areas in Ethiopia, Eritrea, western Kenya, eastern Tanzania, Rwanda, Burundi, eastern Democratic Republic of the Congo, the Malagasy highlands, Mount Cameroon, and the eastern highland ranges in Zimbabwe (Figure 1, bottom panel). Outside of Africa, there was a close correspondence between the areas masked by the absence of reported autochthonous cases and areas classified as unsuitable for transmission based on low temperature in Andean and Himalayan areas (Figure 1, bottom panel). The application of the temperature mask provided a finer spatial resolution constraint to PfAPI data, particularly for the island of New Guinea and the highlands neighbouring the city of Sana'a, Yemen. Important reductions in the spatial areas of risk were also evident in some administrative units in Afghanistan, Bhutan, China, India, and Kyrgyzstan. The aridity mask constrained the mapped P. falciparum transmission risk to small pockets in large administrative boundaries from southern areas of Hilmand and Kandahar, in Afghanistan, the municipality of Djibouti, in Djibouti, and the south-eastern provinces of Iran. The risk areas along the Red Sea coast of Saudi Arabia were also reduced further using the aridity mask. Additional areas constrained within their spatial margins to no risk using the aridity mask included administrative units in India (n = 4), Pakistan (n = 9), Peru (n = 3), Kyrgyzstan (n = 2), Tajikistan (n = 1), and the low risk areas of Namibia bordering the Namib desert. Large areas covered by the aridity mask were reduced from stable (PfAPI ≥ 0.1 per 1,000 pa) to unstable risk (PfAPI < 0.1 per 1,000 pa) in the Sahel. The transmission reducing effects of aridity were also evidenced in Djibouti, Eritrea, northwest Kenya, northeast Ethiopia, northern Somalia, central and coastal areas of Yemen, and southern Pakistan. Importantly, these areas retained small pockets of higher, more-suitable transmission conditions, corresponding to river tributaries and irrigated land where higher transmission risk is supported [50]. Populations at Risk Table 1 provides a summary of the spatial extents and the projected 2007 populations at risk (PAR) within areas of assumed unstable (PfAPI < 0.1 per 1,000 pa) and stable P. falciparum transmission (PfAPI ≥ 0.1 per 1,000 pa) globally and by WHO region. Country PAR estimations are also provided (Table S1). We estimate that there are 2.37 billion people at risk of P. falciparum transmission worldwide, 26% located in the AFRO region and 62% in the combined SEARO-WPRO regions (Table 1). The definition of unstable risk outlined here is the predominant feature of exposure to transmission in the EMRO-EURO region (Table 1). Low-risk areas in AFRO were also coincident with arid, low population density areas. Globally, 42% of the population exposed to some risk of P. falciparum was classified as inhabiting areas of unstable transmission; the total population in these areas was 0.98 billion people. Table 1 Area and Population at Risk of P. falciparum Malaria in 2007 Global and Regional Summary of P. falciparum Parasite Prevalence The summary data on age-corrected PfPR are presented without adjustments for biological and climatic covariates, urbanization, congruence with dominant Anopheles vector species, or any sampling issues inherent in an opportunistic sample of this kind. This is the subject of ongoing work. The summarized data, however, do provide important new insights into the ranges of infection prevalence reported between regions of the world within the P. falciparum spatial limits of stable and unstable transmission. A total of 4,278 spatially unique cross-sectional survey estimates of PfPR were assembled as part of the activities of the Malaria Atlas Project (MAP) by 01 September 2007. These included 186 (4.4%) surveys that were not possible to geolocate and are not considered further in the analysis. Of the positioned survey data, 3,700 (90.4%) were derived from individual communities (about 10 km2 or less), 131 from wide areas (more than about 10 km2 and about 25 km2 or less), 145 from small polygons (more than about 25 km2 and about 100 km2 or less), and 116 from large polygons (more than about 100 km2) [45]. A total of 406 surveys were undertaken outside the defined spatial limits of P. falciparum transmission, of which 46 reported presence of P. falciparum infection in the populations surveyed and 360 reported zero prevalence after allowing for a 10-km buffer around the limits. Thus, the overall sensitivity adjusting for plausible positioning errors [51] was 98.5%. There were 611 surveys falling inside the limits that reported zero prevalence. Even using the 10-km buffer the specificity of the limits was low (37.1%). This reflects the difficulties in estimating zero prevalence without large sample sizes [52], as well as the over-dispersed nature of infection risks between communities within small spatial scales [53]. The global diversity of the age-corrected PfPR2–10 estimates within the limits of transmission is shown in Figures 2–5. A total of 253 surveys reported zero prevalence among 2,121 surveys undertaken in AFRO (Figure 2). Outside of Africa, 358 surveys reported zero prevalence among 1,565 surveys undertaken within the defined limits of transmission. Over 92% and 95% of surveys reporting PfPR2–10 ≥ 50% and ≥ 75%, respectively, were located in AFRO and concentrated mostly between 15° latitude north and south, areas inhabited by Anopheles gambiae s.s. [54] (Figure 2). Conversely lower estimates of PfPR2–10 were described among those surveys conducted in areas occupying the A. arabiensis–dominant regions along the Sahel, horn, and southern areas of Africa [54] (Figure 2). In AMRO (Figure 3) and EMRO-EURO (Figure 4), 87% and 65% of surveys reported PfPR2–10 below 10%, respectively, referred to classically as hypoendemic. Over 65% of PfPR2–10 survey estimates in the combined SEARO-WPRO region reported infection prevalence below 10% (Figure 5), including 218 surveys reporting zero prevalence. Figure 2 Community Surveys of P. falciparum Prevalence Conducted between 1985 and 2007 in AFRO Other regions are shown in Figures 3–5. Of the 4,278 surveys reported globally, 4,092 could be geopositioned of which 3,686, shown in these figures, fell within the predicted limits of P. falciparum malaria risk. A total of 406 records, not shown in the figures, were found outside the limits, of which 46 reported presence of P. falciparum. Data shown are age-standardized (PfPR2–10) and represented as a continuum from zero to 100%. Table 2 and Figure 6 present detailed descriptive statistics for these data. Figure 3 Community Surveys of P. falciparum Prevalence Conducted between 1985 and 2007 in AMRO Figure 4 Community Surveys of P. falciparum Prevalence Conducted between 1985 and 2007 in EMRO-EURO Figure 5 Community Surveys of P. falciparum Prevalence Conducted between 1985 and 2007 in SEARO-WPRO Table 2 Summaries of the P. falciparum Parasite Rate Data Reported between 1985 and 2007 and Mapped within the Spatial Limits of P. falciparum Malaria Figure 6 Box and Whisker Plots of PfPR2–10 by Period and WHO Regions Thick black lines are the medians, and the light-blue boxes represent interquartile ranges; whiskers show extreme, non-outlier observations. Empty circles represent mild and/or extreme outliers. Sample sizes correspond to those shown in Table 2. Despite notable gaps in the coverage of PfPR2–10 data in many areas (Figures 2–5), a summary of the ranges of prevalence survey estimates is provided in Table 2 and Figure 6. These data are presented for the whole time period (Figure 6, top panel) and stratified by time (Figure 6, middle and bottom panels). We stress that these data are not spatially congruent and therefore should not be viewed as representing secular changes in PfPR2–10 estimates by WHO region. The data used for the bottom panel of Figure 6 are potentially of greater value, however, when describing the endemicity characteristics of malaria within the spatial limits shown in Figure 1, as they represent the most contemporary summary of malaria endemicity judged by PfPR2–10. Discussion We have triangulated as much information as we could assemble from exhaustive searches to provide an improved evidence-based description of the limits of P. falciparum transmission globally. The spatial referencing of health statistics, medical intelligence, and national expert opinion represents, to our knowledge, the most complete, current framework to understand the global distribution of P. falciparum risk in 2007. The use of plausible biological constraints upon transmission, based on long-term temperature data and remotely sensed correlates of vegetation cover, improved the spatial precision of the limits and categories of risk. We estimate that there were 2.37 billion people at risk of P. falciparum worldwide in 2007, and 40.1 million km2 of the world might be able to support P. falciparum transmission. Assembling geographic information on disease risk is an iterative process, building on new data and identifying gaps in our knowledge. We have presented previously the distribution of P. falciparum using historical descriptions of risk [1,16] and through the reconciliation of information in multiple travel advisories [55,56]. None have been perfect representations of contemporary malaria distributions worldwide, but such work has initiated a dialogue on the importance of providing an evidence base to malaria cartography and in the sharing of this information [15]. We have not considered the spatial distribution of P. vivax in this paper for a number of methodological reasons. First, the accuracy of health reporting systems for P. vivax clinical cases in areas of coincidental P. falciparum risk is notoriously poor [57]. Second, the climatic constraints on parasite–vector survival are less well defined and thus harder to predict using standardized regional-specific vector bionomics [58]. Third, the combined effects of a prolonged liver stage and the consequences upon natural and drug-resistant recrudescence make the interpretation of prevalence data considerably harder for P. vivax compared to P. falciparum [59]. We are acutely aware that the spatial extent and disease burden of P. vivax merits more attention than it has received, but to achieve an informed evidence-based map similar to that of P. falciparum demands a more fundamental construction of the basic biology of transmission and clinical epidemiology before this can be attempted effectively. We have been cautious in the use of the PfAPI data reported at national levels, recognizing the inadequacies and incompleteness of malaria surveillance [1,23–26]. The intention has been to identify administrative reporting areas that had not detected cases of P. falciparum malaria between 2002 and 2006. It was also recognized that there existed a wide range of reported PfAPI estimates, from one case per 100,000 people pa to reports of confirmed cases in almost 50% of the population every year, which presents a problem for the classification of risk. We therefore applied threshold criteria that would distinguish areas of low clinical risk (i.e., those areas reporting few cases and likely to support unstable transmission conditions) from areas with higher reported case incidence and probably more stable in their P. falciparum transmission characteristics. Our use of a distinction between unstable and stable transmission at 0.1 per thousand pa, while conservative is not without precedent. During the era of the Global Malaria Eradication Programme, epidemiologists proposed a variety of criteria to describe malaria risk in concert with preparatory, active, consolidation, and maintenance phases of elimination and ultimate “eradication” [60–63]. Parasite prevalence was the metric of choice for defining baseline endemicity in the preparatory phase and was useful as an indicator of control progress in the attack phase [52,64], until it became impossible to measure with cost-efficient sampling at very low levels of endemicity. At this juncture, it was proposed that malaria risk be measured through incidence metrics such as the PfAPI [65]. We identified very few PfPR surveys (n = 233) undertaken in areas where reported PfAPI was below 0.1 per thousand pa, 70 (30%) of which reported zero prevalence (Figures 2–5); and the median parasite prevalence was 1.4% (Table 2). It seems appropriate, practical, and feasible to consider multiple metrics during the assembly of malaria risk maps, and we have combined two common malariometric measures of risk: the PfAPI and PfPR. The mathematical relationship between these measures and other traditional epidemiological measures, such as the basic reproduction rate of infection and the entomological inoculation rate, is the subject of ongoing research [61]. Stratification of these risk areas by dominant vector species to enable a more informed assessment of the appropriate suites of intervention measures is also being pursued actively [15]. The PfPR data have been assembled from peer-reviewed literature, unpublished ministry of health sources, postgraduate theses and provision of raw data from malaria scientists in all malaria endemic regions [45]. They do not derive from nationally representative, random-sample surveys. Their coverage might, therefore, be subject to bias toward areas thought to be more malarious. The inclusion of 971 geopositioned surveys reporting zero prevalence (including 523 [53.8%] from Africa), however, does not support this view. Future investigation of the ecological and climatic covariates of PfPR2–10 will need to move from the categorical descriptions of over-dispersed endemicity data presented here, to geostatistically robust estimates of risk that are cognisant of the many potential biases in these data across the entire limits of stable transmission shown in Figure 1. We note, however, that as infection prevalence responds to increased intervention coverage and access to effective medicines, the use of traditional biological covariates might prove less effective in predicting the distribution of P. falciparum transmission intensity. Spatial models of PfPR distribution are being developed and tested as part of MAP's ongoing research to more accurately reflect the ranges of malaria transmission intensity within the margins of stable endemicity. Moreover, the PfAPI and PfPR data described in the present paper will change with time, and future data assemblies need to be maintained in a world with a rapidly changing malaria epidemiology. The supporting geostatistical models used to predict the spatial distribution of endemicity must also therefore facilitate rapid updates. The annual revision of the spatial limits of stable and unstable malaria, based upon new medical intelligence, PfAPI summaries, and the increasingly available contemporary PfPR information will iteratively redefine the cartography of malaria and be hosted on the MAP website (http://www.map.ox.ac.uk) as a public domain resource [15]. Assuming some degree of fidelity in the descriptions of unstable malaria used here, we estimate that one quarter (∼26%) of the malaria-endemic areas of the world are exposed to some degree of unstable P. falciparum transmission and home to approximately one (0.98) billion people. Even within the regions with more stable transmission, the available empirical evidence from contemporary PfPR2–10 survey data is that outside of AFRO, the intensity of transmission is best described as hypoendemic [66] (Figure 6). This observation has important implications for how we view malaria control and broader development goals at a global scale over the next decade. The provisional categorical descriptions of global P. falciparum malaria risk are shown in Figure 1 and suggest that, at a global scale, an aggressive approach to P. falciparum elimination might be reconsidered as a more ambitious and achievable objective in many areas. Regional initiatives aimed at elimination have begun [11–14]. In the Americas, elimination is considered in the most recent 5-y regional strategic plan [12]. In the European region, the two PfMECs (Tajikistan and Kyrgyzstan) are targeted for P. falciparum elimination within the next 5 y [11,13]. Detailed plans have been developed in the Eastern Mediterranean region to consider targeted P. falciparum elimination strategies in Iran and Saudi Arabia, while strengthening maintenance phases of elimination in currently P. falciparum–free countries [14]. With the exception of EURO, detailed maps of the spatial extents of risk in these various regions are not available. Where elimination is considered a viable strategy, resource requirements, targets, and maps become a regional and sub-regional public good and are no longer solely national concerns. Saudi Arabia is providing substantial financial support for the elimination of malaria in its neighbour, Yemen [67]. If this plan is successful, the reportedly high rates of population inflow from Somalia [68] will pose a continued concern due to the potential reintroduction of the parasite. This situation further highlights the need for a reproducible and evidence-based global map of malaria risk that is maintained as a dynamic platform to estimate and predict cross-border risk. Maintaining the detail necessary to map the spatial extent of malaria risk is paramount to the future of malaria control outside of Africa over the next 5 y. We would also argue, however, that Africa has been labelled inappropriately as a vast expanse of holoendemic transmission, intractable to control. Less than a third of all surveys retrieved from AFRO (29%) reported parasite prevalence above 50%, and, as has been described, these results followed closely the distribution of A. gambiae s.s. [54]. The conditions of hypoendemic and mesoendemic transmission were common in surveys conducted outside of this belt (which are not subject to the ravages of this most efficient vector) and are likely to benefit from approaches to prevention and control specific to the underlying ecologic and epidemiologic conditions [15,69,70]. The descriptions of transmission intensity are dynamic and the PfPR2–10 estimates in Africa (Figure 2) do not correspond to levels of endemicity described four decades ago [17]. In the AFRO region, there has been a recent expansion of insecticide-treated net coverage and provision of effective medicines. These programmatic successes are showing tangible impacts on mortality [8,9,71] and morbidity [8,9,72], and it would seem entirely plausible that similar effects will be operating at the level of transmission. If Africa is undergoing a malaria epidemiological transition, capturing this dynamic through mapped information on infection prevalence, and planning accordingly, should be high on the control agenda. The current focus of the Roll Back Malaria movement is, appropriately, in Africa, as this continent bears the brunt of malaria morbidity and mortality [73,74] and the descriptions presented here reinforce this view. P. falciparum transmission is a global problem, however, requiring a global strategy with regional targets and approaches tailored to what can be achieved within defined intervention periods [61]. This strategic planning demands an epidemiologically consistent map that is constantly updated. The assembly of risk data presented here represents the first attempt to combine disparate sources of malariometric data that should serve as a dynamic platform to define a global strategy and map its progress over the coming decades. The maps and national levels of population at unstable and stable risk are released in the public domain, with the publication of this paper, to further that global effort (MAP, http://www.map.ox.ac.uk). Supporting Information Protocol S1 Sources and Descriptions of Medical Intelligence Used to Describe the PfAPI (346 KB DOC) Click here for additional data file. Protocol S2 Developing Global Biological Limits for P. falciparum Transmission (1.3 MB DOC) Click here for additional data file. Table S1 National Estimates of Population at Risk of P. falciparum Malaria in 2007 (231 KB DOC) Click here for additional data file.
Author and article information
Journal
Journal ID (nlm-ta): BMC Infect Dis
Title:
BMC Infectious Diseases
Publisher:
BioMed Central
ISSN
(Electronic):
1471-2334
Publication date Collection: 2010
Publication date
(Electronic):
30
April
2010
Volume: 10
Page: 109
Affiliations
[1
]The Global Fund to Fight AIDS, Tuberculosis and Malaria, Chemin Blandonnet 8, 1214
Vernier, Geneva, Switzerland
Article
Publisher ID: 1471-2334-10-109DOI: 10.1186/1471-2334-10-109
PMC ID: 2876166
PubMed ID: 20433714
SO-VID: 0c2b190e-b228-457d-ac9a-679cd5f368a1
Copyright © Copyright ©2010 Komatsu et al; licensee BioMed Central Ltd.
License:
This is an Open Access article distributed under the terms of the Creative Commons Attribution License ( http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
History
Date
received
: 28
May
2009
Date
accepted
: 30
April
2010
Categories
Subject:
Research Article
ScienceOpen disciplines: Infectious disease & Microbiology
Data availability:
ScienceOpen disciplines: Infectious disease & Microbiology