No change was made in the average number of unsafe injections per person per year in 2000 [1]–[3]. For 2010, in several regions there were reductions in unsafe injections, mostly through a lower proportion of re-use, and these figures were used for all three blood-borne viruses [5]. Given the lack of injections data for the three countries that constituted region SEAR B, extrapolations were made from India, Vietnam and Cambodia. Population figures were updated [6]. Changes in parameters specific for each virus are described below.

The probability of transmission of HIV, HCV and HBV per episode of unsafe healthcare injection cannot be measured directly, so that two proxies must be used: the risk of transmission during a needle stick injury in healthcare workers (HCW) and the risk of transmission per episode of needle/syringe sharing by injection drug users (IDU).

For HCV, the ‘p_t’ for HCW injuries previously used (1.8%), generally accepted at the time, was overestimated because early studies had used unreliable diagnostic assays. A review published in 2002 estimated this ‘p_t’ at 0.5% (59 infections after 11324 injuries) [7]. For HIV, the ‘p_t’ for HCW injuries is generally estimated to be 0.32%, based on follow-up after 6202 exposures [8]–[12]. A 2006 meta-analysis estimated this risk at 0.24% [13]. The risk of transmission of HIV per episode of sharing of needles and/or syringes was estimated at between 0.63% and 1.57% [14]–[16].

When comparing unsafe injections to needle stick injuries, competing factors must be considered. Actions associated with an injection (inserting a needle deep into a muscle, and pushing its content with the plunger) may enhance the risk compared to HCW injuries, which are generally superficial. But on the other hand, one third of HCW injuries occur after a needle had been placed in the patient's vein (to draw blood, to insert an intravenous line, etc.) [17]–[18]. Most healthcare injections being made intramuscularly or subcutaneously, the amount of blood from the index patient that ends up in the needle/syringe is lower than when a HCW manipulated a needle deliberately inserted into a patient's vein. Furthermore, the ‘p_t’ during unsafe injections must be lower than in IDU, among whom the potential transfer of viruses occurs from vein to vein.

Thus, the ‘p_t’ of HIV (1.2%) and HCV (1.8%) per episode of contaminated healthcare injection used for 2000 were presumably overestimated [1], [3]. It is more prudent to use, for each virus, a low estimate, corresponding to the probability of transmission during a needle stick injury to a HCW, and a high estimate which probably should be not more than double the low one. For HIV, this corresponds to 0.32% and 0.64%, nearly identical to the 0.24%–0.65% proposed elsewhere in a meta-analysis [13], our high value for medical injections being close to the lower estimates (0.63%) of the transmission risk among IDU. For HCV, the same approach yields values of 0.5% and 1.0%.

The probability of HBV transmission during an unsafe injection had been assumed to be 6% for HBeAg-negative source patients and 30% for HBeAg-positive patients, in line with other estimates, and we used the same values given that this has not been studied further in recent years [1], [3], [19], [20]. However, it had been arbitrarily assumed that in most regions 20% of the HBsAg-positive individuals were HBeAg-positive, yielding an overall p_t = 10.8%, while elsewhere 50% were HBeAg-positive, for an overall p_t = 18% [1], [3]. We rather calculated region-specific values of ‘p_t’, based on estimates of the proportion of HBeAg antigenemic individuals.

UNAIDS revised retrospectively its measures of national HIV prevalence when Demographic and Health Surveys (DHS), with HIV testing on capillary blood, revealed that in several countries the prevalence (previously measured through surveys of pregnant women) had been overestimated because of an under-sampling of rural populations and an overestimate of the prevalence in men [21], [22]. The regional estimates of HIV prevalence for 2000 were recalculated, and those for 2010 were calculated the same way. From UNAIDS data for 2001 (revised figures) and 2009 [21], we extrapolated to 2000 and 2010, based on the mean annual changes in prevalence. Assuming that in most primary care settings children and adults are treated with the same pool of needles and syringes, the overall prevalence was calculated (and not merely among those aged 15–49 years).

Previous calculations had used the HIV prevalence in the general population and assumed that prevalence among patients attending healthcare facilities was the same [1], [3]. However, HIV-infected patients develop symptoms for which they seek care and receive injections. Consequently, in certain healthcare settings, for instance patients hospitalized in a medical ward, the HIV prevalence is much higher than in the general population, as pointed out by Reid [23]. That effect, although less marked, is also present in primary care settings, even if a substantial fraction of their caseload corresponds to children, because HIV-infected children are also more likely to attend outpatient services than the seronegatives. Furthermore, in some primary care centers a substantial fraction of the caseload consists of patients with sexually transmitted infections, further enhancing HIV prevalence.

Furthermore, the propensity of HIV-infected patients to attend a health facility increases as the disease progresses and so does their viremia, hence their infectiousness. We assumed that this latter phenomenon was intrinsically tailored in within the estimates of the efficacy of transmission to HCW, and no further adjustments were made.

We reasoned that, worldwide, most injections (and most unsafe injections) are given to outpatients in primary care facilities: for-profit clinics (operated by physicians, nurses or unqualified personnel), governmental health centers, facilities run by non-profit organizations, outpatient departments of hospitals, etc. We assumed that, with regard to the syringes/needles used, patients treated in such facilities represent a single population (a mix of children and adults), rather than two distinct compartments each with their own pool of syringes/needles.

To identify relevant studies, Medline searches were performed (Appendix S1) and the US Census Bureau database was searched [24], seeking reports about patients in healthcare settings in developing/transitional countries published since 1995. The goal being to obtain measures of HIV prevalence among unselected, consecutive patients attending healthcare facilities, studies that represented obvious biases one way or the other were excluded, for instance measures among: i) inpatients, in which the HIV-infected would be much over-represented compared to outpatient settings; ii) patients presenting with conditions strongly associated with HIV infection (tuberculosis, pneumonia, etc.); iii) patients attending sexually transmitted diseases clinics or facilities for voluntary testing where the HIV-infected are over-represented; iv) antenatal clinic attendees and blood donors, since these visits are not prompted by ongoing symptoms. Furthermore, were excluded studies where the HIV status had been determined by a single test, studies with fewer than 200 participants, or with unavailable full text.

A total of 4052 titles and abstracts were scanned for full-text review and potential inclusion. Ultimately, 16 studies fulfilled all of the inclusion criteria and presented no exclusion criterion [25]–[41]. These measures of healthcare prevalence were compared with measures of HIV prevalence in the population of the same city or region. In some locations, this was possible through a DHS measure [22]. The prevalence in men and women combined was used, except for an all-women study in Uganda [34] for which the female prevalence was used as comparator. When the study population had been limited to some age groups the prevalence in age groups as close as possible was used as comparator. The comparator prevalence was adjusted for the interval that had elapsed between the study and the corresponding DHS, based on estimates for the variation in HIV prevalence between 2001 and 2009 [19]. For the paediatric studies, regional estimates made by the South African Department of Health for children aged 2–14 years were used as comparator [41]. For the other studies, our comparator prevalence was based on surveys of HIV among antenatal clinic attendees of the same location [24], generally available for the same year as the study itself. To translate this into a prevalence for the whole adult population, an adjustment took into consideration differences in prevalence between men and women, based on UNAIDS estimates in that particular country [21].

Researchers recently estimated the prevalence of HCV seropositivity and HBsAg antigenemia in various regions in 1990 and 2005, for each sex and age stratum, based on a review of respectively 232 and 396 scientific papers and mathematical modelling [42], [43]. These estimates seemed more evidence-based than the empirical ones previously used [1]–[3]. We calculated the annual variation in the prevalence of HCV seropositivity and HBsAg antigenemia between 1990 and 2005, to extrapolate the prevalence in 2000 and 2010. To calculate the overall regional prevalence, the populations of each age stratum were used as weights [6]. As the data for 1990 and 2005 were presented along a revised classification of countries (GBD 2010), the latter was converted into prevalence for GBD 2000 regions, according to the proportions of each 2000 region that came from each 2010 region. No adjustment was made for a potentially higher prevalence in healthcare settings, which seems unlikely given that only a minority of HCV-infected and HBV-infected persons develop cirrhosis.

A similar exercise measured, in each region, the proportion of HBsAg-positive individuals who are HBeAg-positive, based on fewer publications [44]. Prevalence of HBeAg antigenemia among HBsAg-positive individuals was much higher in young children and decreased steadily with older age; geographic variations were modest.

In many low-income countries, HBV vaccine was introduced into the immunisation programme during the 2000–2010 decade, and the fraction of recipients of unsafe injections susceptible to iatrogenic HBV infection decreased progressively among children and adolescents. The proportions of the population of various age groups deemed non-susceptible through vaccination or natural infection were not altered for 2000, but had to be corrected for 2010. WHO collates data provided by member states concerning the proportion of infants who have received the third dose of HBV vaccine by the age of 12 months [45]. For each country, immunisation rates were calculated for two age strata, 0-4 years and 5–14 years, and translated into proportions of susceptibility (‘p_s’) to HBV infection for each region, allowing for natural infections as well. For individuals older than 14 years, the same ‘p_s’ as in the 2000 model [1], [3] were used.

The revisions in HIV prevalence in the overall population had only a modest impact on the estimated regional prevalence for 2000 [1], [3]. For the prevalence in healthcare settings, thirteen studies contained data about adults, two presented paediatric data, and one had included both children and adults. Most studies had been performed in Africa, two in India and one in Haiti. HIV prevalence among study populations varied widely. The ratio between the prevalence among patients attending a healthcare facility and that in the comparator in each of 16 studies is shown in Supporting Information, Table S1. The means of these ratios was 2.48 for studies with adults, 2.69 for the paediatric studies, and 2.52 overall. This latter figure was multiplied by the prevalence in the overall population to derive the prevalence in healthcare settings, for each region, in 2000 and 2010 (Table 2). In 2010, this prevalence decreased in AFR E, and increased in EMR D and EUR C.

The revised estimates of regional prevalence of HCV seropositivity for 2000 are displayed in Table 3 along with the data for 2010. Compared to the previously used data [1], [3], there was relatively little change in regional HCV prevalence for 2000. In 2010, prevalence increased in six regions.

The revised estimates of regional prevalence of HBsAg and HBeAg for 2000 are shown in Table 4. For the high-prevalence regions, the revised estimates of prevalence of HBsAg antigenemia were lower than before [1]. The revised proportions of the HBsAg-positives who were HBeAg antigenemic in 2000 were generally higher than in the original model, except for the two regions where this prevalence had been arbitrarily estimated at 50%. Since HBeAg antigenemia has a profound influence on the ‘p_t’ for HBV, the regional ‘p_t’ varied accordingly. Table 4 also displays the data for 2010. In all but one region, the prevalence of HBsAg antigenemia decreased. There was a modest reduction in the proportion of HBsAg-positive individuals who were HBeAg antigenemic.

Table 5 shows the revised estimates of HIV infections transmitted through unsafe injections in 2000, based on the same model but with HIV prevalence in health care settings as ‘p_v’ and with the two revised ‘p_t’ values. Our higher estimate for 2000, based on p_t = 0.64%, yielded estimates similar to those presented initially [1], [3], with roughly a quarter of a million HIV infections acquired through unsafe injections. Naturally, the estimates with p_t = 0.32% yielded figures that were half the other measure. Based on previously mentioned assumptions, between 133,328 and 266,405 HIV infections were acquired worldwide through unsterile injections in 2000. Region SEAR D (mostly India) had represented more than half of the global number of injections-related cases of HIV, to a large extent because it was estimated that 75% of injections in SEAR D were made with re-used needles and syringes, based on a survey in India [1]–[3]. Despite a much higher ‘p_v’, the contribution of sub-Saharan Africa was lower than SEAR D, because of fewer injections and less frequent re-use. Using UNAIDS revised data as denominators, in developing and transitional economies, between 4.6% and 9.1% of all new HIV infections in 2000 were caused by unsafe injections [21].

For 2010, the main changes in parameters were in the number of unsafe injections per person per year [5] and the HIV prevalence in healthcare settings. The same two values of ‘p_t’ were used. Between 16,939 and 33,877 HIV infections were acquired through unsafe injections worldwide (Table 5). Sub-Saharan Africa represented 48% of such cases, while the contribution of SEAR D decreased to 18%. Compared to 2000, the number of injections-related HIV infections acquired worldwide decreased by 87% in 2010, when between 0.7% and 1.3% of all new HIV infections were so acquired [21].

Table 6 shows the revised estimates of HCV infections transmitted through unsafe injections in 2000, based on the revised measures of prevalence and the two values of ‘p_t’. We estimated that in 2000 between 952,111 and 1,867,904 HCV infections were injections-related. Again, the higher estimate for all ten regions was similar to the one generated previously [1], [3], even if the regional distribution varied. Table 6 also displays the results for 2010, using the same ‘p_t’ values, the updated regional prevalence and the updated numbers of unsafe injections [5]. In 2010, between 157,592 and 315,120 HCV infections were acquired from unsafe injections, about one third of which occurred in EMR D and another third in WPR B.

Table 7 displays the estimates of HBV infections transmitted through unsafe injections in 2000, using the same HBeAg-specific values of ‘p_t’ applied on the revised estimates of the prevalence of HBsAg and HBeAg antigenemia. Although the regional figures varied along with modifications in the prevalence of antigenemia, the total for all ten regions was again similar to the one calculated previously for 2000, with 19,710,444 HBV infections acquired from injections. Table 7 also shows the results for 2010, based on the same values of ‘p_t’ and the updated estimates of the number of unsafe injections and of the prevalence of HBsAg and HBeAg antigenemia in 2010 [5]. Compared to 2000, there was a 91% reduction in the number of injections-related HBV infections, to 1,679,745 new infections.