Incidence of HIV and hepatitis C virus among people who inject drugs, and associations with age and sex or gender: a global systematic review and meta-analysis

rather than being multi-city or nationwide. Estimates were measured over 1987–2021 for HIV and 1992–2021 for HCV. Pooled HIV incidence was 1·7 per 100 person-years (95% CI 1·3–2·3; I ²=98·4%) and pooled HCV incidence was 12·1 per 100 person-years (10·0–14·6; I ²=97·2%). Young PWID had a greater risk of HIV (RR 1·5, 95% CI 1·2–1·8; I ²=66·9%) and HCV (1·5, 1·3–1·8; I ²=70·6%) acquisition than older PWID. Women had a greater risk of HIV (RR 1·4, 95% CI 1·1–1·6; I ²=55·3%) and HCV (1·2, 1·1–1·3; I ²=43·3%) acquisition than men. For both HIV and HCV, the median risk-of-bias score was 6 (IQR 6–7), indicating moderate risk.


Introduction
Globally, around 18% of people who inject drugs (PWID) are living with HIV and more than 50% have been infected with hepatitis C virus (HCV). 1 Given that effective interventions are available, UNAIDS and WHO have recommended policies and targets for ending the HIV/AIDS epidemic and eliminating HCV as a public health threat by 2030. 2,3 Monitoring HIV and HCV incidence is key to understanding the scale of these epidemics, tracking progress towards achieving the 2030 UNAIDS and WHO targets, and evaluating the effect of interventions. 4,5 Over the past two decades, the incidence rates of HIV and HCV have declined among PWID in some high-income countries (HICs) 6-12 due to the scale up of harm reduction interventions and, more recently, treatment. Meanwhile, persistently high levels or outbreaks of HIV and HCV among PWID have been reported in other HICs and lowincome or middle-income countries (LMICs). 13-18 A better understanding of these shifting epidemiological patterns and the availability of data globally are needed to orient surveillance and programming efforts. However, no global study has summarised HIV and HCV incidence among PWID, except for a modelling study that estimated HCV incidence to be 8·6 per 100 person-years (95% credible interval 5·4-14·4) in 2015. 19 Equally important to informing prevention strategies is a better understanding of age and sex or gender differences in the risks of HIV and HCV acquisition among PWID. However, no study has synthesised data on the relationship between age and incident HIV or HCV infection, despite several studies 20-23 finding higher risk behaviours among younger versus older PWID. These higher risks are attributed to lower engagement in harm-reduction programmes, 22,24,25 a reliance on others to inject, 26,27 and greater exposure to structural determinants of harm (eg, homelessness or incarceration). 20,28 Studies have also highlighted differences in injection and sexual practices that potentially place women who inject drugs at higher HIV or HCV risk than men who inject drugs. 29,30 Women face greater stigma and criminalisation linked to sex work and gender-role stereotypes (eg, primary caregiver), which can reduce their ability to adopt safer practices. 31 Three systematic reviews [32][33][34] have explored the relationship between gender and prevalent HIV or HCV infection among PWID, and found either similar prevalence 32,33 or slightly higher prevalence in women than in men. 34 Only one review has evaluated the association of sex with incident HCV infection, finding a 1·36-times greater average risk in women than in men. 35 No similar study has been done for HIV incidence.
We aimed to summarise global HIV and HCV incidence among PWID, to characterise geographical and temporal differences in pooled HIV and HCV incidence, and to estimate the associations between age and sex or gender and the risks of HIV and HCV acquisition.

Research in context
Evidence before this study We did a global systematic review and meta-analysis of studies presenting data on HIV or primary hepatitis C virus (HCV) incidence among people who inject drugs (PWID) published between Jan 1, 2000, and Dec 12, 2022. We searched MEDLINE, Embase, and PsycINFO, without language restrictions, using terms related to HIV infection, HCV infection, injecting drug use, and study designs that could be used to evaluate HIV or HCV incidence (eg, cohort studies and longitudinal studies). We also contacted authors of identified studies to request unpublished or updated data. We included studies that estimated incidence by longitudinally re-testing people at risk of infection or by using assays for recent infection. No previous study has synthesised global incidence data for either of these two outcomes among PWID. One systematic review gathered data on HCV incidence among PWID in Europe from studies published between 2000 and 2012. This review found data from eight countries, where the incidence of HCV was high and variable, ranging from 2·7 per 100 person-years in one UK study to 66 per 100 personyears in an Irish study, with a median of 13 per 100 person-years; no meta-analysis was done. One systematic review and metaanalysis also synthesised data on the association between sex and HCV incidence among PWID and found that the incidence rate ratio for risk of infection in women who inject drugs compared with men who inject drugs was 1·36 (95% CI 1·13-1·64).

Added value of this study
We identified 64 estimates for HIV incidence and 66 for HCV incidence. Estimates ranged from 0·1 per 100 person-years to 31·8 per 100 person-years for HIV (pooled estimate 1·7 per 100 person-years, 95% CI 1·3-2·3) and from 0·2 per 100 person-years to 72·5 per 100 person-years for HCV (pooled estimate 12·1 per 100 person-years, 10·0-14·6). There was considerable heterogeneity across geographical regions and study and participant characteristics, and there were substantial data gaps. Estimates were found for 31 countries overall, which were mostly high-income and middle-income countries for HIV and high-income countries for HCV. Estimates were available from 27 (14%) of 195 countries for HIV and from 24 (12%) countries for HCV, with 20 (10%) countries having estimates for both HIV and HCV. Approximately two-thirds of estimates were limited to small geographies (eg, a single city). Based on a subset of studies with HIV and HCV incidence data, the incidence of HCV was around 17-times higher than the incidence of HIV. Young PWID (generally defined as ≤25 years) had a greater risk of HIV and HCV acquisition than older PWID, and women who inject drugs had a greater risk of HIV and HCV acquisition than men who inject drugs.

Implications of all the available evidence
Our study highlights the sparse empirical data on HIV and HCV incidence available among PWID, particularly in low-income and middle-income countries. Given that HIV and HCV incidence rates are considered key for monitoring the trajectory of these epidemics, evaluating programmatic effect, and tracking progress towards elimination, our findings suggest that intensified efforts are needed to keep track of these outcomes among PWID. Efforts could include the use of indirect methods for estimating HIV and HCV incidence and alternative indicators, such as changes in HCV viraemic prevalence. In most settings, a range of data and indicators might have to be triangulated to adequately monitor the HIV and HCV epidemics among PWID. Given that young PWID and women who inject drugs have a higher risk of both HIV and HCV acquisition, age-appropriate and gender-appropriate prevention measures are urgently needed to reach and engage these higher risk subgroups.

Search strategy and selection criteria
This systematic review and meta-analysis is reported in accordance with PRISMA guidelines. 36 We updated an existing database produced for two previous systematic reviews, 37,38 which included studies published in any language reporting HIV incidence, HCV incidence, or both, among PWID published between Jan 1, 2000, and Sept 14, 2020. Using the same search strategy, AA and DS did a systematic literature search of MEDLINE, Embase, and PsycINFO for studies, including conference abstracts, published between Sept 14, 2020, and Dec 12, 2022. They searched these databases without language restrictions using terms related to HIV infection, HCV infection, injecting drug use, and study designs that could be used to evaluate HIV or HCV incidence (appendix pp 7-8). They also searched the reference lists of review articles published during the same period.
Eligible studies reported summary estimates of HIV or HCV incidence among people with a history of ever injecting drugs, measured either through repeat testing of people at risk or using assays of recent infection (eg, markers of antibody avidity or detection of HCV RNA positivity among anti-HCV antibody-negative participants). [39][40][41] Only primary HCV infection was considered. We excluded studies that estimated HIV or HCV incidence among PWID who were incarcerated.
In our two previous systematic reviews, 37,38 we contacted authors of published studies that reported HIV or HCV incidence estimates among PWID but did not report on the association with incarceration or housing (focus of those studies), and studies in which HIV or HCV incidence among PWID was measured, but not yet published (166 authors). We re-contacted authors who responded to previous requests (29 authors) to ask for up-to-date HIV or HCV incidence estimates and, if not published, the associations between HIV or HCV acquisition risks with age and sex or gender. We also contacted two additional authors because we needed further information 42 43 ). We use the term "unpublished" for estimates that were calculated for this study. However, most unpublished estimates were based on data from published studies, for which we cite the most recently published article.
We created an Endnote library to catalogue search results. The titles and abstracts of each record were first screened by one author (AA or DS), with 10% being checked by another (AA, DS, AT, JS, or ZW). If inconsistences arose between two authors, all their records were double-screened, with disagreements resolved by discussion. Full-text review of records deemed potentially eligible was completed in duplicate (AA, DS, HF, ZW, and JGW). We used Google Translate to read non-English language papers.
Contrary to the protocol, we included incidence data from the intervention arm of trials when data were unavailable for the control arm or when both arms received some form of intervention (appendix pp [18][19]; this decision was implemented at the data extraction stage. No other protocol deviations occurred. The study protocol can be found online.

Data analysis
Eligible records were extracted into Microsoft Excel (version 16.7) by AA and double-checked by CA, HF, AGL, ALM, DS, AT, and ZW. We extracted HIV and HCV incidence rates; the associations between HIV and HCV acquisition risks and age and sex or gender; and incidence data disaggregated by age and sex or gender, where available. For some studies, particularly long-term studies and those for which several publications were available, several estimates of HIV and HCV incidence measured over different time periods were available. In these cases, we extracted the most recent HIV or HCV incidence estimate and, if available, the least recent without temporal overlap to explore temporal trends. Few studies had multiple non-overlapping incidence estimates, and, therefore, we extracted, at most, two estimates per study. Less recent HIV and HCV incidence estimates that overlapped temporally with estimates that were extracted were marked as duplicates.
For each record, we extracted the number of incident cases, the total person-years at-risk, incidence, and 95% CIs. Where available, we also extracted the incidence rate ratio (IRR), hazard ratio, or risk ratio (collectively referred to as the relative risk [RR]) comparing HIV and HCV acquisition risks among young and older PWID and among women and men who inject drugs. We only extracted unadjusted RR estimates, as our aim was to synthesise the total effects of age and sex or gender on HIV and HCV acquisition risks. Extraction of adjusted RRs was deemed unnecessary, as the individual-level factors included in adjusted estimates are likely to reflect mediators of these associations rather than confounders. Because studies used different age groupings, we defined young as the most frequently reported threshold (≤25 years) and allowed for variations in this definition across studies. We included sex or gender, as reported, when comparing women and men; other gender identities were not extracted as they were rarely reported. We also extracted several study (eg, design, sampling strategy, and recruitment sources) and participant characteristics (eg, mean or median age and whether ≥80% had injected within the past year [denoted as having injected recently]); all characteristics are listed in the appendix (p 9). For studies that reported both mean and median, we prioritised using the median as it is a better measure for summarising non-normally distributed data.
Records with duplicate data were excluded from the analysis. We assessed the risk of bias for each record using a modified Newcastle-Ottawa Scale, 44  See Online for appendix 9 points on criteria related to the selection of participants and assessment of the outcome (appendix p 10). We penalised all interventional studies in the risk-of-bias assessment on the sample representativeness criterion (ie, we did not allocate these studies a point for this criterion). Risk of bias was assessed by AA and reviewed by HF, AGL, ALM, DS, AT, and ZW. We classified risk of bias as high if a study scored 3 or less, moderate if they scored between 4 and 6, and low if they scored 7 or more.
We used random-effects meta-analysis (inverse-variance method) to estimate pooled HIV and HCV incidence rates (primary outcome) on the basis of the number of incident cases and duration at risk. Between-study variance was estimated by use of the DerSimonian-Laird method and 95% CIs were estimated by use of the formula by Rothman and Greenland. 45 For HIV or HCV incidence measured by use of assays for recent infection, total duration at risk was imputed from the number of incident cases and incidence rate. A fixed value (0·5) was applied to all cells of studies with no infection cases. Rates were log-transformed in analyses and back-transformed for reporting. We explored heterogeneity in HIV or HCV incidence through a-prioridefined subgroup analyses, including of study characteristics (eg, sampling strategy and recruitment venues) and participant characteristics (eg, whether ≥80% injected recently and mean or median age). Variables used in the subgroup analyses were defined in the protocol (appendix pp 5-6) and are outlined in the appendix (p 11). Heterogeneity was quantified by use of the I² statistic and differences between groups tested by use of the Q test. 46 We similarly synthesised the IRR of HCV to HIV incidence when both were available through the same study and explored heterogeneity in the pooled estimate by WHO region.
To investigate geographical and temporal differences in pooled HIV and HCV incidence, we did univariable and multivariable random-effects meta-regression (linear mixed-effects) analyses using the inverse-variance method and log-transformed incidence rates. Models assume that the random-error and random-effect terms are normally distributed and that the relationship between the continuous moderators and the outcome is linear. 47 Model assumptions were inspected by use of normal Q-Q plots, the Kolmogorov-Smirnov test, and residuals-versus-fitted plots. Geographical differences were explored by use of the World Bank income classification because data were sparse for some WHO regions. Temporal differences were explored by use of the midpoint of each study period as the time variable. Because only the most recent estimate per study was used in this analysis, we estimated temporal differences by comparing data between studies. We adjusted our meta-regression models for recruitment sources, study duration, and whether at least 80% of participants had injected recently. The multivariable model was fitted by use of a theory-led approach and the number of variables was minimised to avoid over-fitting and multicollinearity. Variables were selected a priori on the basis of their potential to confound the relationship between geographical or temporal differences and HIV and HCV incidence. Results are presented as unadjusted or adjusted IRRs with 95% CIs.
We used random-effects meta-analysis to pool RRs and their corresponding 95% CIs comparing HIV or HCV acquisition risk among young versus older PWID and among women versus men who inject drugs. Similar meta-analytical methods were used to quantify betweenstudy variance and to conduct subgroup analyses by several study characteristics, including WHO region, World Bank income classification, and publication status (appendix p 11). Where available, we also synthesised agedisaggregated and sex-disaggregated or genderdisaggregated absolute HIV and HCV incidence rates.
We did two sensitivity analyses to assess the robustness of the pooled HIV and HCV incidence estimates, excluding (1) studies that used assays for recent infection, for which we imputed the duration of follow-up, potentially leading to differences in the estimated 95% CIs relative to those reported in the original study and (2) intervention studies. Because there is some evidence that Poisson-normal models via generalised linear mixedeffects models could perform better than the conventional inverse-variance method in meta-analyses involving sparse data, 48 we did an additional sensitivity analysis to explore the robustness of our results to the use of different methods.
We also did several sensitivity analyses for our metaregression models. First, because use of the study period midpoint as the time variable can bias measures of temporal change in incidence when estimates are based on long follow-up periods, we excluded those estimates with follow-up exceeding 10 years. Second, we included non-recent incidence estimates derived from the same study, when available, and accounted for their nested structure using a multi-level meta-regression model. 49 This analysis estimated temporal differences in pooled HIV and HCV incidence rates by comparing data between and within studies. Finally, three additional sensitivity analyses were done to assess the robustness of results to the assumptions of the meta-regression models. These comprised log-transforming continuous moderators for which we observed potential departures from the linearity assumption; excluding estimates that appeared to be outliers and demonstrated potential departures from the normality assumptions; and using Poisson-normal models rather than the conventional inverse-variance method to synthesise estimates for the same reason as before (ie, potentially better suitability for sparse data). 48 Publication bias in measures of RR was explored by use of funnel plots and Egger's test. The effects of publication bias on measures of incidence rates and approaches to explore them are not well established, so we did not do such analyses. We did all analyses in R (version 4.0.5) using the "meta" and "metafor" packages. 50 This systematic For the World Bank income classification see https:// datahelpdesk.worldbank.org/ knowledgebase/articles/906519world-bank-country-andlending-groups review and meta-analysis is registered on PROSPERO, CRD42020220884.

Role of the funding source
The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report.

Results
Our new database searches identified 9493 potentially eligible records, of which 2224 were duplicates (figure 1). Initial screening of titles and abstracts resulted in 211 new records eligible for full-text review. We retrieved 377 additional records from our existing database of incidence studies, 37,38 resulting in 588 records for full-text    100 Anwar et al (2021) 54 Samo et al (2013) 92 Iversen et al (2021) 75 Roussos et al (2022) 89 Sypsa V, National and Kapodistrian University of Athens, personal communication Smyth et al (2003) 97 Sabbatini et al (2001) 91 van Santen and Prins (unpublished) 11 Kozlov et al (2006) 71 Hoffman et al (2013) 64 Niccolai et al (2011) 83 Kozlov et al (2016) 72 La Rosa (unpublished) 102 Hurtado Navarro et al (2008) 65 Blomé et al (2011) 57 Kaberg et al (2018) 70 Judd et al (2005) 69 Ompad et al (2017) 88 Hayashi (unpublished) 94 Oviedo-Joekes et al (2011) 86 Strathdee and Abramovitz (unpublished) 68 Mehta and Astemborski (unpublished) 96 Fuller et al (2003) 61 Ouellet et al (2000) 85 Murrill et al (2001) 82 Seage et al (2001) 95 Garfein et al (2007) 62 Batman et al (2018) 56 Des Jarlais et al (2003) 60 Des Jarlais et al (2016) 59 Skaathun et al (2022) 108 Kral et al (2003) 73 Mirzazadeh et al (2018) 81 Subgroup Heterogeneity: I 2 =89·3%; τ 2 =0·4; p<0·0001 South-East Asia region Azim (unpublished) 55 Solomon et al (2010) 98 Sarna et al (2014) 93 McFall et al (2017) 79 Patel et al (unpublished) 87 Clipman et al (2022) 67 Luo et al (2021) 52 Yang et al (2016) 53 Wei et al (2006) 103 Yang et al (2017) 18 Zhang et al (2007) 105 Ruan et al (2013) 90 Yen et al (2012) 104 Giang et al (2022) 107 Go et al (2015)   For both HIV and HCV, the median risk-of-bias score was 6 (IQR 6-7), indicating moderate risk, with most records being penalised on sample representativeness and adequacy of follow-up criteria (appendix pp [30][31][32][33]. 3246 incident cases of HIV were reported during 180 857·2 person-years of follow-up. HIV incidence ranged from 0·1 per 100 person-years in several settings, including Amsterdam (the Netherlands), Vancouver (BC, Canada), and New York (NY, USA), to 31·8 per 100 personyears in a multi-city study in Ukraine (figure 2). Pooled HIV incidence was 1·7 per 100 person-years (95% CI 1·3-2·3), with substantial heterogeneity (I²=98·4%). This pooled estimate varied by several study and participant characteristics (table). HIV incidence was highest in the Eastern Mediterranean and South-East Asia regions, and lowest in the Western Pacific and American regions (table). HICs had a significantly lower HIV incidence than LMICs. HIV incidence was highest where sampling occurred through peer referral and recruitment through participants' networks compared with other sampling strategies and recruitment sources (table). HIV incidence generally decreased as study duration and mean duration of follow-up per person increased. When stratified by participant characteristics, the incidence of HIV was higher in studies in which the mean or median age of participants and the proportion on opioid agonist treatment were lower and in studies in which HIV prevalence was higher (table; appendix pp [34][35]. 4233 incident cases of HCV were reported during 44 841·2 person-years of follow-up (figure 3). HCV incidence ranged from 0·2 per 100 person-years in a multicity study in the Netherlands to 72·5 per 100 person-years in a study in Madrid (Spain; figure 3). Pooled HCV incidence was 12·1 per 100 person-years (95% CI 10·0-14·6), with substantial heterogeneity (I²=97·2%). Pooled HCV incidence also varied by several characteristics (table; appendix pp [34][35]. When stratified by study design, sampling strategy, and recruitment sources, HCV incidence was lowest in retrospective cohorts, studies in which sampling was convenience based, and studies in which recruitment occurred through medical sources, and, conversely, HCV incidence was highest in linked repeated cross-sectional studies, if sampling was based on time-location, and if recruitment occurred in the community. The incidence of HCV decreased as study duration and mean follow-up duration per person increased. When stratified by participant characteristics, HCV incidence was significantly higher in studies in which at least 80% of participants had injected recently and in studies in which participants were younger or had a shorter duration of injection (table). HCV incidence did not vary by WHO region or World Bank income classification (table). Excluding intervention trials or studies that used assays for recent infection in our sensitivity analyses had little effect on pooled HIV and HCV incidence estimates (appendix p 36). Sensitivity analyses pooling estimates with the use of Poisson-normal models instead of the inverse-variance method produced similar results (appendix p 50).

Discussion
In this systematic review and meta-analysis of HIV and HCV incidence among PWID, studies were highly variable in their methodology and participant characteristics and there were substantial data gaps. HIV incidence ranged from 0·1 per 100 person-years to 31·8 per 100 person-years and HCV incidence ranged from 0·2 per 100 person-years to 72·5 per 100 person-years, with pooled estimates of 1·7 per 100 person-years for HIV and 12·1 per 100 personyears for HCV. Based on a subset of studies with both HIV and HCV incidence data, the incidence of HCV was on average 17·4 times greater than the incidence of HIV. Young PWID had on average a 1·5-times greater risk of HIV and HCV acquisition than older PWID, and women had a 1·4-times greater risk of HIV and a 1·2-times greater risk of HCV than men.
We identified several sources of heterogeneity across pooled HIV and HCV incidence estimates. For example, HIV and HCV incidence estimates were higher in studies in which recruitment occurred through participants' networks rather than through medical sources, possibly reflecting differences in risk profiles. Studies in which participants had a longer mean duration of follow-up had lower HIV and HCV incidence rates, which could reflect reductions in risk behaviour following repeated riskreduction messaging or better retention of PWID with lower risk of acquisition. Higher HIV and HCV incidence rates were identified if most participants injected recently (albeit non-significant for HIV) and higher HIV incidence if a lower proportion of the sample received opioid agonist treatment. Our review also illustrates the scarcity of HIV and HCV incidence data among PWID globally, which were available in only 14% and 12% of countries, respectively. Moreover, less than half of estimates were from 2010 onwards and only eight (HIV) and five (HCV) estimates were initiated since 2015. We found a considerable geographical skew in the availability of estimates, with few from middle-income countries for HCV, and only one HIV and HCV incidence estimate from low-income countries. Two-thirds of estimates were from single cities and so might not be nationally representative.
This scarce and heterogeneous body of evidence made the evaluation of geographical and temporal differences in HIV and HCV incidence challenging. In several settings, HIV incidence has been shown to have decreased as a result of the implementation of combined prevention and care. 11,157 However, globally, temporal trend data were scarce and our assessment of the change in incidence with time was mainly based on comparing estimates from different studies. We found no evidence of a decline in HIV incidence with time, which could be due to methodological reasons such as little comparability between estimates, preferential measure ment in settings with extreme incidence (very low or very high), and inclusion of estimates averaged over long time periods. However, at least in LMICs, our finding might also reflect a real absence of decline in HIV incidence because of insufficient access to harm reduction interventions. 158 Supporting this theory, we found HIV incidence to be more than two-times higher in LMICs than in HICs. Conversely, there was no evidence that HCV incidence varied between geographical regions, and only some weak evidence that HCV incidence declined with time. As with HIV, the scale-up in HCV treatment in many HICs since 2015 might have led to lower HCV incidence in HICs compared with LMICs. However, data from this period are scarce, which might be why we found no difference by LMICs versus HICs.
Estimates of HIV and HCV incidence with time among PWID are important for monitoring the trajectory of epidemics, adapting the public health response, and assessing progress towards achieving the UNAIDS and WHO 2030 targets. 4,159 Although direct methods for measuring HIV and HCV incidence rates are preferred, these methods are not without limitations (eg, high cost and participant attrition) and, in some countries, might not be feasible to implement on a national scale. 4 In these contexts, indirect methods of estimating incidence from prevalence data or by the use of mathematical modelling could be preferred. 4,5 Alternatively, our group has shown that changes in chronic HCV prevalence can track changes in HCV incidence when resulting from increases in HCV treatment, and so HCV prevalence could also be reliably used instead. 4 Regardless of the options chosen, different data and indicators should be triangulated, including For four records (Balogun et al, 2009;110 Emanuel and Croxford, unpublished; 40 Hope et al, 2011; 122 and Palmateer and Hutchinson, unpublished 6 ), the follow-up duration was imputed from the number of HCV cases and the incidence rate, potentially leading to differences in the estimated 95% CIs relative to those reported in the original study (appendix pp [27][28][29]. Box sizes are proportional to the weight of the study in relation to the pooled incidence. HCV=hepatitis C virus.
measuring HCV reinfection to keep track of HCV incidence, to adequately monitor these epidemics among PWID.
We found the risks of HIV and HCV acquisition to vary by age and sex or gender. Although the magnitude of the associations varied in stratified analyses, HIV and HCV acquisition risks were consistently greater among young PWID compared with older PWID. These findings align with previous studies that report greater injecting and sexual risk behaviours among young compared with older PWID. [20][21][22][23] By contrast, the overall greater risks of HIV and HCV acquisition among women versus men were more

African region
Kurth and Walker (unpublished) 76

South-East Asia region
Clipman and Solomon (unpublished) 106 Patel modest compared with the age analysis, particularly for HCV. The pooled estimates for sex or gender differences in risk could mask regional differences. We found the magnitude of the association for HIV risk between women and men to be highest in the African and South-East Asia regions and lowest in the Western Pacific region, consistent  Hayashi (unpublished) 149 Jacka et al (2019) 153 Macphail and Coffin (unpublished) 42 Hagan et al (2010) 119 Havens (unpublished) 121 Mehta and Astemborski (unpublished) 96 Morris et al (2020) 131

South-East Asia region
Clipman and Solomon (unpublished) 106

Western Pacific region
IIversen et al (2013) 124 Maher et al (2006)  with findings from a 2019 systematic review focused on prevalence. 32 For HCV, there were no estimates for the African and South-East Asia regions to enable a similar comparison. The only other systematic review to have explored the association between gender or sex and incident HCV infection found a 1·36-times greater risk of acquisition in women than in men, 35 slightly higher than our pooled estimate, possibly due to methodological differences (eg, including prison samples and adjusted estimates).
( Figure 5 continues on next page)

African region
Kurth and Walker (unpublished) 76

European region
Mravcik (unpublished) 51 Lucidarme et al (2004)  Hayashi (unpublished) 149 Jacka et al (2019) 153 Roy et al (2012) 135 Spittal et al (2012) 141 Hagan et al (2004) 120 Hagan et al (2010) 119 Havens (unpublished) 121 Mehta and Astemborski (unpublished) 96 Morris et al (2020) 131 Thorpe et al (2002)   Our efforts to include unpublished data increased the number and recency of estimates for all outcomes. Encouragingly, we found no evidence of a difference between published and unpublished estimates. Owing to a preponderance of studies that were more than 10 years old and based in HICs, one of the most important limitations of our review is that our findings provide an incomplete depiction of current global HIV and HCV incidence rates among PWID. The inclusion of data derived through intervention trials and studies that used assays for recent infection might have biased pooled estimates, although results remained largely consistent when these estimates were excluded. Additionally, our assessment of temporal and geographical differences in HIV and HCV incidence rates was exploratory due to sparse data from most global settings, including trend data. Finally, sparse data bias in some studies, owing to a small number of events, small sample sizes, or both, can carry over to the pooled estimates, leading to potentially biased summary estimates and 95% CIs. 160,161 However, pooled estimates were similar in sensitivity analyses that used meta-analytical approaches that are likely to be more suited for sparse data, which is reassuring.
In conclusion, this systematic review and meta-analysis produced, to our knowledge, the first pooled estimates of HIV and HCV incidence rates derived by use of direct methods among PWID. The low availability of incidence estimates globally, particularly in LMICs, suggests that intensified efforts are urgently needed to keep track of the HIV and HCV epidemics among PWID in these countries. Because PWID often face stigma and discrimination, there is a risk that they will be overlooked in elimination efforts if national data on HIV and HCV incidence remain absent. This gap needs to be addressed to achieve the global goals of eliminating HIV and HCV. Given the higher risks of HIV and HCV acquisition in young versus older PWID and in women versus men who inject drugs, ageappropriate and gender-appropriate prevention and harm reduction measures are also urgently needed to serve these subgroups at high risk. Factors other than age and sex or gender are also likely to be influencing HIV and HCV acquisition risks (eg, types of drugs injected and the context of injection, identifying as men who have sex with men, and engaging in sex work), and research is also needed to synthesise the role of these other factors to better strengthen HIV and HCV prevention responses. We plan to explore some of these factors in future studies. Collaborative Group contributed unpublished data for the study. JS, HF, AGL, ALM, DS, AT, and ZW accessed and verified the data reported in this study. All authors contributed to the interpretation of data and the critical revision of the Article. All authors had full access to all the data in the study and had final responsibility for the decision to submit for publication. For contributions of the study group members, see the appendix (p 51).

Declaration of interests
AA acknowledges support through postdoctoral fellowships from the Canadian Institute of Health Research (CIHR), Fonds de recherche du Québec -Santé (FRQ-S), and the Canadian Network on Hepatitis C. MA reports grants paid to his institution from the Public Health Agency of Canada and grants from the Ministère de la santé et des services sociaux du Québec, during the conduct of this study, and grants from the CIHR, outside the submitted work. JA reports grants from the National Institutes of Health (NIH), during the conduct of this study. JB reports consulting fees from Gilead Sciences, AbbVie, and Cepheid; payment or honoraria from Gilead Sciences for an educational event; and receipt of equipment, materials, drugs, medical writing, gifts, or other services from Gilead Sciences, as part of an NIH R01 grant, outside of the submitted work. CSC reports grants from Gilead Sciences, Janssen Pharmaceuticals, and GSK (paid to the University of Calgary); consulting fees from Roche Pharmaceuticals, Gilead Sciences, Janssen Pharmaceuticals (paid to the University of Calgary, on behalf of the Canadian Hepatitis B Virus [HBV] Network), and Altimmune (paid to the University of Calgary, on behalf of the Canadian HBV Network); payment or honoraria for lectures, presentations, speakers bureaus, manuscript writing, or educational events through Gilead Sciences; and patents planned, issued, or pending (PCT/ CA2021/050234; international patent application title: polypeptides directed against viral infection and uses thereof; compositions and methodologies for the detection and treatment of HBV infection; USA; 62/982,474; Feb 28, 2020). SC reports payments for chairing or presenting at a European HIV Testing Week webinar from the Centre of Excellence for Health, Immunity and Infections, outside the submitted work. KH reports financial support for the present manuscript paid to her institution from the US National Institute of Drug Abuse (NIDA; Vancouver Drug Users Study: the impacts of evolving drug use patterns on HIV/AIDS; U01DA038886), the Michael Smith Foundation for Health Research Scholar Award, and St Paul's Foundation, and grants or contracts from the CIHR and the William and Ada Isabelle Steel Fund through Simon Fraser University (paid to institution) and the British Columbia Centre on Substance Use (paid to KH). GM reports grants or contracts from AbbVie (clinic support, advisory board, and speaker fees), the Canadian Network on Hepatitis C (research support), Coverdale (clinic support), Gilead Sciences (clinic support and speakers fees), International Network on Health and Hepatitis in Substance Users (INHSU; speaker fees), and TD Bank (clinic support); payment or honoraria from the Canadian Association for the Study of the Liver (scientific planning committee for web-based learning, conjoint with AbbVie; payment to institution), AbbVie (speakers fees for educational podcasts and brochure; payment partly to institution and partly to GM), Gilead Sciences (speaker fees; payment partly to institution and partly to GM), and ECHO plus (honorarium for speaking); and support for attending meetings or travel, or both, from INHSU and being on the advisory board for AbbVie and Gilead. JGW reports grants and contracts from Gilead Sciences and FIND, The Global Alliance for Diagnostics. MH reports having a leadership or fiduciary role as trustee of the Society for the Study of Addiction. All other authors declare no competing interests. Declaration of interests for the members of the HIV and HCV Incidence Review Collaborative Group are provided in the appendix (p 52).

Data sharing
Extracted data sheets will be shared with researchers who provide a methodologically sound proposal approved by AA and PV. Proposals should be directed to adelina.artenie@bristol.ac.uk and peter.vickerman@bristol.ac.uk; requesters will need to sign a data access agreement. Data can be made available starting with the date of publication of this Article and up to 5 years thereafter.