The epidemiology of hepatitis C virus in Central Asia: Systematic review, meta-analyses, and meta-regression analyses

The objective was to delineate hepatitis C virus (HCV) epidemiology in countries of Central Asia (CA), specifically Kazakhstan, Kyrgyzstan, Tajikistan, Turkmenistan, and Uzbekistan. A systematic review was conducted guided by the Cochrane Collaboration Handbook, and reported using PRISMA guidelines. Meta-analyses were performed using DerSimonian-Laird random-effects models with inverse variance weighting. Random-effects meta-regression analyses were performed on general population studies. The systematic review identified a total of 208 HCV prevalence measures. No incidence or Turkmenistan studies were identified. Meta-analyses estimated HCV prevalence among the general population at 0.7% (95%CI: 0.7–0.8%) in Kazakhstan, 2.0% (95%CI: 1.7–2.4%) in Kyrgyzstan, 2.6% (95%CI: 1.7–3.6%) in Tajikistan, and 9.6 (95%CI: 5.8–14.2%) in Uzbekistan. Across CA, the pooled mean prevalence was 13.5% (95%CI: 10.9–16.4%) among non-specific clinical populations, 31.6% (95%CI: 25.8–37.7%) among populations with liver-related conditions, and 51.3% (95%CI: 46.9–55.6%) among people who inject drugs. Genotypes 1 (52.6%) and 3 (38.0%) were most frequent. Evidence was found for statistically-significant differences in prevalence by country, but not for a temporal decline in prevalence. CA is one of the most affected regions by HCV infection with Uzbekistan enduring one of the highest prevalence levels worldwide. Ongoing HCV transmission seems to be driven by injecting drug use and healthcare exposures.

extraction and analyses of data. Data from all reports considered relevant were extracted by WB and FZ. Data from all reports were subsequently double extracted by SM to ensure consistency and minimize errors in extracted information. Extracted information included study details (author, year of publication, title, and journal), location of study, year(s) of data collection, study design, sampling method, risk population, number of participants included in the study, number of participants tested, type and name of serological test used to test for HCV, and the primary outcome (HCV incidence or/and HCV prevalence). Rounding HCV prevalence measures to two decimal places was conducted if they were below 1%, while the remaining measures were rounded to only one decimal place. When available, HCV ribonucleic acid (RNA) data were also extracted. All studies identified in the secondary independent screening for genotype information were also extracted into a separate extraction file. Risk factors for HCV infection were extracted if they were statistically-significant through multivariable meta-regression. Extracted data were classified into population categories according to exposure risk to HCV infection, as presented in Fig. 1. The classification scheme was based on existing literature 10,24,25 , and earlier reviews of HCV prevalence 11-19 . Quantitative assessment. HCV prevalence reports with a minimum of 50 participants were categorized and reported in our reporting tables by risk population. Meta-analyses of HCV prevalence measures were performed by risk population and country for all studies with at least 25 participants. In reports where HCV prevalence was reported for mixed-country samples, the study was included only in meta-analyses for CA as a region. In reports that included prevalence measures but no reported sample size, a sample size of 300 was imputed and the study was included in the review and meta-analyses. This sample size was deemed reasonable and conservative, given that the median sample size of included studies with a reported sample size was 348.
HCV prevalence for the total sample size was replaced with stratified prevalence whenever a minimum of 25 participants were available for each stratum. Stratifications were included based on a predefined order, where nationality was prioritized, then sex, year, region, and finally age. To avoid duplication one final stratification for each study was included.
Freeman-Tukey type arcsine square-root transformation was used to stabilize the variance of HCV prevalence measures 26 . DerSimonian-Laird random-effects model was used to pool HCV prevalence (with inverse variance weighting). This model assumes a normal distribution of true effect sizes (that is HCV prevalence) across studies, and takes into account true heterogeneity as well as random chance effects across studies 27 . Heterogeneity measures were also assessed. All forest plots were visually assessed and the Cochran's Q test was performed, with a p-value of <0.10 indicating statistically strong evidence 27,28 . The I² measure and its confidence interval were assessed 27 . The prediction interval was also calculated to estimate the range in which HCV prevalence of 95% of future studies will fall 27,29 .
The number of HCV antibody-positive persons in each country was determined by multiplying the country-specific pooled mean HCV antibody prevalence estimate by the population size in each country. This was subsequently multiplied by the pooled mean fraction of HCV RNA positivity in antibody-positive persons (also commonly referred to as the "viremic rate" 30,31 ), to derive the number of HCV chronically-infected persons. The United Nations World Population Prospects database 32 was used to obtain the population size of each country.
Since potential issues have been identified with the Freeman-Tukey type arcsine square-root transformation 33 , a sensitivity analyses was performed to confirm the validity of our results in which the generalized linear mixed models (GLMM) method was used to perform meta-analyses.
A proportion of the general population data were on blood donors, a population typically including only healthy adults. Sensitivity analysis was performed to determine whether excluding blood donors could impact the pooled mean HCV prevalence estimate in the general population. This sensitivity analysis was done for each country separately, and for CA as a whole.
Based on established methodology 20 , univariable and multivariable random-effects meta-egressions were performed to assess country-level associations with HCV prevalence and the sources of between-study heterogeneity in the general population. Variables included in the univariable models included country, general population subpopulations, study site, sample size (<100 or ≥100), sampling method (probability-based or non-probability-based), year of publication, and year of data collection. Variables with a p-value of <0.1 were included in the multivariable model. Variables were deemed significant in the final multivariable meta-regression if they had a p-value of <0.05.
For each country and the whole CA, the frequency of each genotype was calculated. Individuals who were positive for mixed genotypes contributed separately to the number of each of the identified genotypes. The Shannon Diversity Index (H) was determined to assess the diversity of genotypes, with a higher score (out of 1.95) indicating more diversity 34 .
The meta package 35 on R version 3.4.3 36 was used to perform the meta-analyses. The metan command on STATA 13 37 was used to perform meta-regressions.
Qualitative analysis. Using the Cochrane approach to surmise risk of bias (ROB), the quality of HCV incidence and/or prevalence measures was evaluated. Based on three quality domains, studies were classified into either low or high ROB. These domains included HCV ascertainment (biological assay or otherwise), sampling method (probability-based or non-probability-based), and response rate (≥80% of the target sample size was reached or otherwise). Studies with information missing for any of the three domains were classified with unclear ROB for that specific domain. Studies in which the reported HCV measures were acquired from patients' medical records, or from individuals voluntarily visiting facilities where routine blood screening is performed, were considered as having low ROB on strictly the response rate domain.

Results
search results. Figure 2, adapting the PRISMA flow diagram 21 , shows the process by which studies were selected into this systematic review. A total of 771 citations were identified: 95 from PubMed, 129 from Embase, and 547 from the Scientific Electronic Library (eLibrary.ru). A total of 99 unique reports underwent full-text screening, after duplicates were removed and titles and abstracts were screened. From these, 69 reports were removed, the reasons for which are stated in Fig. 2. Eighteen reports were added to the systematic review from gray literature/unpublished data, and from screening of references of full-text articles and reviews. Finally, 47 reports qualified for inclusion in this systematic review, yielding no incidence measure and 208 prevalence measures. In the secondary systematic review, all 771 citations were screened for HCV genotype information. After duplicates were removed and titles and abstracts of all unique reports were screened, 35 reports underwent full-text screening. Finally, 6 reports qualified for inclusion in this secondary systematic review (Fig. S2).
HCV antibody prevalence overview. We present here a synthesis of HCV prevalence in each country of CA. The 208 HCV prevalence measures included 67 measures from Kazakhstan, 96 from Kyrgyzstan, 20 from Tajikistan, 23 from Uzbekistan, and 2 from mixed-country samples. No study was identified from Turkmenistan ( Fig. S8A).
HCV prevalence ranged from 4.0-40.3% among non-specific clinical populations, with a median of 8.5%. These included hospitalized populations with HCV prevalence ranging from 5.9-33.3%, with a median of 8.0%; and HIV patients with HCV prevalence ranging from 10.5-40.3%, with a median of 21.8% ( Table 2).
Country-level. In Kazakhstan, HCV prevalence ranged from 0.7-5.1% among the general population, with a median of 0.9%; and 2.0-50.0% among populations at intermediate risk, with a median of 29.0%. Only one study was identified among non-specific clinical populations, with an HCV prevalence of 40.3% in HIV patients 38 . HCV prevalence ranged from 23.8-40.4% in populations with liver-related conditions, with a median of 26.6%; and 43.3-90.2% among PWID, with a median of 60.3%.
In Kyrgyzstan, HCV prevalence ranged from 0.8-5.0% among the general population, with a median of 2.0%; 0.0-35.0% among populations at intermediate risk, with a median of 7.0%; and 4.0-33.3% among non-specific clinical populations, with a median of 8.0%. No studies were identified among populations with liver-related conditions. HCV prevalence ranged from 17.0-60.4% among PWID, with a median of 46.4%.
In Tajikistan, HCV prevalence ranged from 0.5-7.3% among the general population, with a median of 3.9%. Only two studies were conducted among populations at intermediate risk 39 , with HCV prevalence of 4.2% among sex workers (male, female, unspecified) 40 , and 6.2% among HCW 41 . Only one study was conducted on non-specific clinical populations, with an HCV prevalence of 32.1% in HIV patients 41 . Only two studies were conducted on populations with liver-related conditions, reporting an HCV prevalence of 46.0% 42 and 36.0% 39 . HCV prevalence ranged from 24.9-67.1% among PWID, with a median of 32.6%.
Forest plots for the meta-analyses can be found in the Supplementary Material (Figs S3-S7). In the majority of meta-analyses, statistically significant heterogeneity was observed (Cochrane's Q statistic's p-value was always <0.0001; Table 4). Most of the variation across studies was due to variation in effect size (HCV prevalence) rather than chance (I 2 > 59.0%). The prediction intervals ranged from narrow to wide for the different meta-analyses. Collectively, the heterogeneity measures indicated high heterogeneity in HCV prevalence in each country and risk population category.
Too few studies reported HCV RNA viremic rate in the general population to warrant calculation of the pooled mean viremic rate for CA. Accordingly, the pooled mean viremic rate of 67.6% for MENA was used in calculating chronic-infection prevalence and the number of chronically-infected persons. This choice is justified by the fact that this measure is a biological measure that (in principle) should be largely independent of the region 31 , and given that CA and MENA countries are both developing countries. The highest number of chronically-infected persons was found in Uzbekistan at 2.1 million, followed by Tajikistan  In sensitivity analyses, the GLMM meta-analyses confirmed similar results for all risk populations (Table S3). Also in sensitivity analyses, after blood donor data were excluded, population-level HCV prevalence was overall similar across countries, and in CA as a whole (Table S4).

Meta-regressions and sources of heterogeneity.
The results of the meta-regression for the general population is presented in Table 5. In the univariable meta-regression analyses, country, study site, sample size, and year of data collection were significant predictors (p-value < 0.1), and therefore were included in the final multivariable analysis. Notably, sampling method (probability-based versus non-probability-based) had no effect on observed HCV prevalence.

Author, year (citation)
Year(s) of data collection Study site and year of data collection lost significance (p-value > 0.05) in the multivariable analysis-only country and sample size remained statistically significant. Relative to Kazakhstan, the prevalence in Kyrgyzstan, Tajikistan, and Uzbekistan was higher with an adjusted odds ratio (AOR) of 2.0 (95%CI: 1.1-3.4), 2.8 (95%CI: 1.4-5.6), and 10.0 (95%CI: 4.6-21.7), respectively. Sample size (>100) was associated with lower HCV prevalence, with an AOR of 0.4 (95%CI: 0.1-1.0). Notably, the AOR for year of data collection was 1.0 (95%CI: 1.0-1.1)-there was thus no evidence for declines in HCV prevalence with time. The model explained 51.4% of the variability in HCV prevalence.
HCV RNA prevalence. Our search identified only four HCV RNA measures, all of which were reported among HCV antibody-positive individuals: 39.2% in a study on a general population 43 , 100% in a study on HIV patients 38 , 100% in a study on chronic hepatitis patients 42 , and 70.5% in a study on liver cirrhosis patients 42 .
HCV genotypes. HCV genotype information was available in six studies with a total of 382 HCV RNA positive individuals (Table S5). Only 0.5% of individuals were infected with multiple genotypes, while the remaining majority were infected with a single genotype. No genotype information was available for Kyrgyzstan and Turkmenistan.
The highest proportions of infections for each HCV genotype in CA as a whole were for genotype 1 at 52.6% and genotype 3 at 38.0%, followed by genotype 2 at 9.4%. Genotypes 4, 5, 6, and 7 were not identified. Genotype diversity tended towards being low, but varied across CA, with the highest diversity observed in Kazakhstan (H = 1.04 out of 1. 95

HCV risk factors.
Only two studies reported statistically-significant risk factors for HCV infection after controlling for confounders. In Tajikistan, among PWID, daily injection, history of incarceration, and living/working outside of Tajikistan in the past 10 years, were associated with HCV infection 44 . In Kazakhstan, among a general population, tattooing and (unexpectedly) towel sharing were reported as associated with HCV infection 45 .
Quality assessment. Table S6 summarizes the results of the ROB assessment performed on HCV prevalence measures. The majority of measures were of high precision (94.7%), with a sample size ≥100. Most measures were of low risk of bias in the HCV ascertainment domain, with 99% being based on biological assays, and 1% being based on self-reporting. Though most of the studies reported the name of the biological assay used to assess HCV antibody prevalence, the majority of studies (90%) did not explicitly report the generation of the assay. Among studies reporting the generation of the used assay, all used the more sensitive and specific 3 rd generation enzyme-linked immunosorbent assays (ELISA) tests. The majority of studies employed non-probability-based sampling, and were characterized by a high response rate. To summarize, 100% of studies had low ROB based on at least one ROB domain, 65.0% of studies had low ROB based on at least two ROB domains, and 13.4% of studies had low ROB based on all three ROB domains. No study had high ROB based on two or three ROB domains. In all, the quality assessment indicates reasonable though not optimal study quality.

Discussion
We presented, to our knowledge, the first systematic review and synthesis of HCV epidemiology in CA, a region perceived to be heavily affected by this infection 9,10 . Our results indicated that HCV antibody prevalence varies across countries of CA, ranging from 0.7% in Kazakhstan to 9.6% in Uzbekistan (Table 4 and Fig. S8). Accordingly, HCV prevalence in Uzbekistan is considerably higher than global levels, and one of the highest worldwide 9,10 . This finding is of concern considering that Uzbekistan is also the most populous country in CA, with 32 million inhabitants 46 , and a country struggling with a weakened healthcare system since the collapse of the Soviet Union 6 . With an estimated 2.1 million chronically-infected persons, >80% of all chronically-infected persons in CA reside in Uzbekistan. Notably, Uzbekistan has also the highest rate of HIV among all countries in this region 47 , and a main mode of transmission appears to be injecting drug use, a shared mode of transmission with HCV.
Remarkably, HCV prevalence does not appear to be decreasing with time in CA (Table 5), contrary to global trend 48,49 . This may in part be reflective of the majority of studies from this region being reported more recently, with approximately 85% of all studies included in this review being from the last decade.
High HCV antibody prevalence was observed across all risk populations (Tables 1-3 and S2), and more so for PWID, HIV patients, and prisoners, suggesting a major role for injecting drug use in infection transmission. HCV antibody prevalence was also high in populations with liver-related conditions, suggesting a major role that HCV plays in liver disease burden in CA.
Strikingly, no studies were identified among high risk clinical populations such as haemodialysis, haemophilia, and thalassemia patients-the role of healthcare in transmission remains uncertain. However, the relatively high HCV antibody prevalence in non-specific clinical populations (Table 2), and HCV epidemiology in other soviet-era-related countries 9,10,49,50 , suggest that healthcare could be a major mode of exposure, at least in earlier decades.
Subregional disparities in quality of healthcare services may have also contributed to the heterogeneity in HCV prevalence across CA 50 . For example, in Uzbekistan, it appears (anecdotally) that there is an excessive practice of medical and non-medical invasive procedures, such as blood transfusions and bloodletting, in addition to poor infection control 51 , inadequate blood screening 43,51,52 , and use of unsafe medical injections 50,51 , all of which are probable causes for the high HCV prevalence in this country, as has been observed in other developing  In randomized controlled trials the extracted HCV prevalence measure was the cross-sectional baseline HCV prevalence measure. ǂ Study did not report sample size. The included sample size was imputed based on the median sample size of all studies that reported a sample size.  Table 4. Meta-analyses for hepatitis C virus (HCV) prevalence in Central Asia (CA) by risk population. Abbreviations: CI, confidence interval ªQ: Cochran Q statistic assesses if heterogeneity is present in HCV prevalence estimates. b I²: Assesses the percentage of between-study variation that is due to true differences in HCV prevalence estimates across studies rather than chance. c Prediction interval: Estimates the 95% interval in which the true HCV prevalence in a new HCV study will lie. ¥ This range is for all studies included in the metaanalyses database and covers the range of HCV prevalence across not only main HCV prevalence measures, but also across all strata. countries [53][54][55] . Furthermore, the challenging political climate in Uzbekistan has prevented the introduction of up-to-date healthcare guidelines and effective approaches to reduce HCV transmission in healthcare settings 51,52 . While no genotype information was available for Kyrgyzstan and Turkmenistan, pooled analysis on data from Kazakhstan, Tajikistan, and Uzbekistan suggest that HCV genotype 1 (53% of infections) and genotype 3 (38%) are the major circulating strains, but with strong presence of genotype 2 (9% ; Table S5). While genotype 1 is common globally 10,56 , its major presence may reflect healthcare-related exposures, given the frequency of identifying this genotype in clinical populations in CA 42,43,57,58 , as well as the global association between this genotype and healthcare exposures 56 . The major presence of genotype 3 may be due to injecting drug use being a major driver of incidence, given the global association between this genotype and injecting drug use 56 , or may just reflect a sub-regional pattern-genotype 3 is the main circulating strain in each of Afghanistan 13,59 and Pakistan 59 , both of which are neighbouring countries of CA.
The pooled mean HCV prevalence in PWID indicated that over half of this population is already exposed to HCV (Table 4), similar to global trends [60][61][62] . Notably, CA is geographically located along drug trafficking routes originating from Afghanistan 7,63 , and is believed to have one of the highest rates of injecting drug use in the world 64 . These factors further corroborate a major role for injecting drug use in transmission. Furthermore, with the high HCV prevalence found in prisoners (Table S2), incarceration could be influential in HCV transmission dynamics, just as in other regions 62,65 . The high prevalence observed among sex workers (male, female, unspecified; Table S2) may suggest also high rates of injecting drug use in these populations, as supported by HIV biobehavioral surveillance data-sexual and injecting networks could be overlapping hotspots of both HCV and HIV transmission 7,64 .
Despite progress in characterizing HCV epidemiology in CA, our study highlights key challenges and limitations to establishing a satisfactory understanding. Evidence varied by country, with no data identified from Turkmenistan (Fig. S8A). No data were available for high risk clinical populations, though healthcare could be a major mode of exposure, as it is in other soviet-era-related countries 9,10,50 , and in countries with similar stage of development, e.g. in MENA [11][12][13][14][15][16][17][18][19]66 . No data was identified for community-related exposures, e.g. informal healthcare, but such exposures could play a role as seen in other regions 67 . There was an insufficient number of studies reporting HCV RNA prevalence in CA, a measure that informs assessment of chronic-infection prevalence, as antibody prevalence reflects both current infection as well as past infection (that is persons who spontaneously cleared the infection or were treated) 68 .
Most available studies were descriptive-few had analytic epidemiologic designs where risk factors and modes of exposure could be ascertained. Most studies employed non-probability-based sampling, however, results of the meta-regressions indicated this had no effect on HCV prevalence in the general population, and therefore may not have limited the representativeness of reviewed data in our study. There was high heterogeneity in HCV prevalence measures (Table 4), but most heterogeneity (for the general population) was subsequently explaineddifferences by country were the main driver of prevalence variation (Table 5). A small-study effect was observed, with studies with a smaller sample size reporting higher HCV prevalence (Table 5), thereby potentially limiting the representativeness of reviewed data. HCV genotype data was relatively sparse, with no studies identified from Kyrgyzstan and Turkmenistan.
In spite of these limitations, a key strength of our study is that we identified a substantial number of studies, including a volume of unpublished data, in a significantly affected, but poorly understood region, thereby facilitating a synthesis of evidence and identification of knowledge gaps. A priority in addressing these gaps is to carry   out nationally-representative probability-based and population-based surveys in each of these countries. Such surveys can yield a precise estimate of HCV prevalence, delineation of spatial variability in infection exposure, identification of modes of transmission, and assessment of HCV knowledge and attitudes, as has been done in recent years in other countries, e.g. in Egypt 15,69-74 and Pakistan 11,75-77 .

Conclusion
In context of inadequate and underfunded healthcare systems 8,52 , CA is one of the most affected regions by HCV infection. Uzbekistan, in particular, appears to be enduring one of the highest prevalence levels worldwide. HCV transmission appears to be driven by injecting drug use and healthcare exposures, with no evidence for declines in prevalence in recent years. Genotypes 1 and 3 are the most frequently-circulating strains, with some presence for genotype 2.
Our findings inform HCV response for public health planning, health service provision, development of HCV policy guidelines, and implementation of HCV programming to reduce transmission and associated disease burden. Achieving HCV elimination in CA by 2030 can only be accomplished by aggressive action and commitment, given the extent of challenges. There is an urgent need for expansion of affordable HCV testing and treatment for key populations, and targeted control based on settings of exposure. In context of this region being heavily affected by injecting drug use and the global opioid epidemic, harm reduction services must incorporate HCV services and be accessible to all PWID, by being expanded to all relevant settings, such as prisons. Nationally-representative probability-based population-based surveys must be conducted to precisely delineate HCV epidemiology in these countries and address the knowledge gaps, as identified in this study. Improving infection control in healthcare facilities is also warranted, such as through updating (otherwise outdated) clinical guidelines for healthcare workers 52 , and adopting safety-engineered syringes as recommended by WHO 28,29 .