Population attributable risk of aﬂatoxin-related liver cancer: Systematic review and meta-analysis q

Background: Over 4 billion people worldwide are exposed to dietary aﬂatoxins, which cause liver cancer (hepatocellular carcinoma, HCC) in humans. However, the population attributable risk (PAR) of aﬂatoxin-related HCC remains unclear. Methods: In our systematic review and meta-analysis of epidemiological studies, summary odds ratios (ORs) of aﬂatoxin-related HCC with 95% conﬁdence intervals were calculated in HBV+ and HBV (cid:2) individuals, as well as the general population. We calculated the PAR of aﬂatoxin-related HCC for each study as well as the combined studies, accounting for HBV status. Results: Seventeen studies with 1680 HCC cases and 3052 controls were identiﬁed from 479 articles. All eligible studies were conducted in China, Taiwan, or sub-Saharan Africa. The PAR of aﬂatoxin-related HCC was estimated at 17% (14–19%) overall, and higher in HBV+ (21%) than HBV (cid:2) (8.8%) populations. If the one study that contributed most to heterogeneity in the analysis is excluded, the summarised OR of HCC with 95% CI is 73.0 (36.0– 148.3) from the combined effects of aﬂatoxin and HBV, 11.3 (6.75–18.9) from HBV only and 6.37 (3.74–10.86) from aﬂatoxin only. The PAR of aﬂatoxin-related HCC increases to 23% (21–24%). The PAR has decreased over time in certain Taiwanese and Chinese populations. Conclusions: In high exposure areas, aﬂatoxin multiplicatively interacts with HBV to induce HCC; reducing aﬂatoxin exposure to non-detectable levels could reduce HCC cases in high-risk areas by about 23%. The decreasing PAR of aﬂatoxin-related HCC reﬂects the beneﬁts of public health interventions to reduce aﬂatoxin and HBV.


Introduction
Aflatoxins are toxic and carcinogenic chemicals produced primarily by the fungi Aspergillus flavus and A. parasiticus, which infect food crops such as maize, peanuts, and tree nuts. About 4.5 billion people worldwide are exposed to dietary aflatoxins. 1 Exposures are highest in tropical and subtropical regions of the world, where maize and peanuts are dietary staples and food storage conditions are suboptimal. 1,2 Aflatoxins are amongst the most potent naturally occurring human hepatocarcinogens known. The International Agency for Research on Cancer (IARC) has classified "naturally occurring mixes of aflatoxins" as a Group 1 human carcinogen. 3 Abundant epidemiological evidence suggests that aflatoxin exposure synergises with chronic hepatitis B virus (HBV) infection to increase liver cancer (hepatocellular carcinoma, HCC) risk in populations with both risk factors. [4][5][6][7][8] More recently, toxicological models for the mechanism of the synergism of these two risk factors have emerged, [9][10][11] and are summarised in Wild and Gong. 12 Unfortunately, both high aflatoxin exposure and HBV are prevalent in many parts of the developing world, particularly Asia and Africa.
Previously, by compiling food consumption and aflatoxin contamination data in multiple countries and conducting a quantitative cancer risk assessment, we estimated that 25,200-155,000 (5-28%) annual HCC cases worldwide could be attributed to aflatoxin exposure. 13 This large range highlights the limitations in obtaining exposures solely from food surveys, uncertainties in the nature of the dose-response relationship, and uncertainties in HBV prevalence data in different nations.
In this context, systematically analysing human studies that relate biomarkers of aflatoxin exposure and HBV infection to HCC may provide a more precise and accurate measurement of burden of HCC caused by aflatoxin. Therefore, in this study, we systematically reviewed epidemiological studies on these associations in different world regions. By combining the relevant odds ratios (ORs) and relative risks (RRs) from these studies, we conducted meta-analyses to calculate population-attributable risk (PAR) of aflatoxin-related HCC in the population overall, as well as in HBV+ and HBVÀ populations. PAR is the proportion of disease cases that could be avoided if a particular risk factor was eliminated in a population. In the context of our study, PAR of aflatoxin-related HCC is the proportion of HCC cases that could be avoided in a chosen population by reducing aflatoxin exposures (as measured by biomarkers) from detectable to undetectable levels.

Search strategy
We performed a literature search until May 13th, 2011, using the following search terms on Medline/Pub-Med: (aflatoxin) and (hepatitis B) and (liver cancer); (aflatoxin) and (hepatitis B) and (hepatocellular carcinoma). Additionally, we searched reference lists from retrieved articles to identify further relevant studies. Our systematic review and meta-analyses were conducted in adherence to PRISMA standards for reporting meta-analyses. 14

Eligibility criteria
Studies were included in the systematic review if they met the following criteria: (1) case-control or cohort study design; (2) aflatoxin as the exposure of interest; (3) HBV as the infection of interest (hepatitis B virus surface antigen [HBsAg] as a marker of chronic HBV infection); (4) HCC as the outcome of interest; and (5) relative risk (RR) or odds ratio (OR) estimates with 95% confidence intervals (CIs) reported, or data to calculate these.    6 Taiwan, 7 or sub-Saharan Africa. 4 Fourteen studies reported biomarker measurements for aflatoxin exposure, while the other three studies relied on food consumption data.
Twelve studies included both HBsAg+ and HBsAgÀ individuals, with risk estimates that were adjusted for HBsAg positivity (nine studies). Five studies were conducted in HBsAg+ populations only. b Amongst the fourteen studies that utilised biomarkers, five measured urinary aflatoxin biomarkers, including AFM 1 and AFB 1 -N 7 -Guanine, six measured AFB 1 -albumin adducts, two measured AFB 1 -DNA adducts, and three measured TP53 249 ser mutations. Several studies included measures of more than one biomarker. c 15 out of 16 identified case-control studies provided matched ORs.

Data extraction
The following data were extracted from each study: authors, publication year, study design and sample size, study location, study period, participants' gender and age range, metric and range of aflatoxin exposure, estimated adjusted RRs/ORs and variables adjusted for analysis. Because all identified studies are case-control designs except one cohort study, and because RR and OR can be used interchangeably when the disease is relatively rare (<15%; HCC rates are lower than this in the populations studied), we combined the RR from this study with the ORs from the case-control studies to calculate a summary OR. If aflatoxin exposure was measured using different biomarkers in the same study, we selected the ones reflecting consistent biomarkers amongst different studies (one OR per study was used).

Statistical methods for meta-analysis
The ORs from the studies were first combined in the meta-analysis using a random-effects model, and then a fixed-effects model if heterogeneity in the study pool was insignificant. 15 The studies were categorised by the recruited population type: general populations, and HBV+ or HBVÀ populations. First, all the studies providing data for general populations (including both HBV+ and HBVÀ individuals) were combined, and ORs of aflatoxin-related HCC after HBsAg+ adjustment and ORs for combined (aflatoxin + HBV) effects were analysed. Then the studies with data from HBV+ populations (and studies that recruited from the general population but separately estimated ORs in HBV+ populations) were combined; and the ORs for HBV+ populations only were estimated. We also combined the studies that separately estimated the ORs in HBVÀ populations. If the study examined the association between aflatoxin exposure and HCC in various exposure categories, we chose the ORs reflecting highest and lowest levels of aflatoxin exposure for the meta-analysis.
Heterogeneity amongst the studies was evaluated using the Cochran's Q value calculated from the Mantel-Haenszel method and the I 2 statistic. 15 We performed sensitivity analyses in which each study was in turn removed and the rest analysed to evaluate if the results were significantly affected by one particular study. Publication bias was assessed by a funnel plot and associated statistical tests of asymmetry. All statistical analyses were performed with Comprehensive Meta-Analysis software Version 2.2.

Statistical methods for PAR calculations
We estimated the PAR for aflatoxin-related HCC in HBV+ and HBVÀ populations for each study if the data were available. To estimate the PAR for afla-toxin-related HCC using the adjusted ORs, we used the attributable fraction formula 16 : where AF POP is aflatoxin attributable risk fraction in the population including exposed and unexposed individuals, P i is the proportion of the population in stratum i that is exposed, and W i is the proportion of diseased individuals (cases) in stratum i. We use adjusted OR i in stratum i as an approximation of RR i .
If the study provided risk estimates adjusted by HBsAg positivity, we used the formula below 16 to estimate the PAR of aflatoxin-related HCC in the general population: where P c is the proportion of cases exposed in the combined population based on detection limits for aflatoxin biomarkers in the studies, and HBsAg positivity-adjusted OR is used as an approximation of RR. For each AF pop , we calculated 95% confidence intervals (CI) using the method described in Daly. 17

Literature search
The step-by-step process of our literature search is presented in Fig. 1. From 479 results, we excluded human cell line studies, animal studies, and review articles. Using the eligibility criteria described above, 27 studies were selected. Three more relevant studies were identified from the reference lists of the 27 selected studies. We then read the full texts of these 30 studies. Six studies were excluded because they were duplicated reports from the same population in the same time period, and seven more were excluded because quantitative measurements of association between aflatoxin exposure and HCC were not provided. Thus, 17 studies were included in this systematic review and PAR analysis. Table 1 provides an overview of the eligible studies. The 17 studies 5-8,18-30 on aflatoxin exposure and HCC risk -eight case-control studies, eight nested case-control studies, and one cohort study -were published between 1994 and 2009. There were 1680 HCC cases and 3052 controls in total.

Study characteristics
Four studies reported results for one Taiwanese cohort from four different time periods 7,18,25,30 from 1980s to 2000s. To determine if all these studies should be included in the meta-analysis, we first examined the heterogeneity between the risk estimates provided by these studies. Because of the significant heterogeneity a If there was a series of follow-up studies in the same cohort need to be combined, only the numbers of cases and controls from the largest follow-up study were counted, although different odds ratios from different follow-up studies were combined to assess the effect. All the cases and controls were only counted once, and as well as in calculations presented in Tables 4 and 5. b This row shows the summary odds ratio of combing three follow-up studies in a Taiwan cohort in different years. c The summary odds ratio obtained for the Taiwan cohort was used to represent the effect of all studies in this cohorts, and combine with other studies. d Seven studies 7,15,17,21,22,25,26 reported adjusted ORs on aflatoxin-related HCC risk in HBsAg+ individuals. Four studies 5,6,8,16 (including two studies conducted in Sub-Saharan Africa countries) did not provide adjusted ORs directly, but provided data to calculate the unadjusted ORs. We calculated the unadjusted ORs for each of these studies and combined them with ORs from other studies with eligible data, thus we can include the effects of studies in Sub-Saharan Africa population. In subgroup analysis, the large variation of summarised ORs of aflatoxin-related HCC in HBsAg+ individuals may be explained by combining the unadjusted ORs. The heterogeneity was significant when studies were combined to examine the association between aflatoxin exposure and HCC risk in the general population and in HBsAg+ individuals. e The summary odds ratio obtained from different follow-up studies for the Taiwan cohort was used to represent the effect of all studies in this cohorts, and combine with other studies.
of aflatoxin exposures and HCC risk estimates in this cohort between the follow-up studies through the years, we treated these as independent studies in the analysis. In analyses that included only the most recent of all studies in a particular cohort, the results were nearly identical to those obtained when including all studies ( the general population after adjustment of HBsAg+, and risk estimates for HBsAg+ or HBsAgÀ separately). 8,24 Likewise, two articles reported results from a study in the Gambia with risk estimates for the general population after adjustment of HBsAg+, and risk estimates for HBsAg+ or HBsAgÀ separately. 5,23

Aflatoxin exposure and HCC risk by HBsAg Status
The association between aflatoxin exposure and HCC, independently or in conjunction with HBV, was analysed by combining eligible studies by HBsAg+ status and calculating summary ORs ( Table 2). Meta-analyses were conducted by geographic region (China, Taiwan, and sub-Saharan Africa).
Aflatoxin exposure is significantly associated with HCC risk, regardless of HBsAg status, with a summarised OR of 4.75 (2.78-8.11) from nine studies in the general population adjusted by HBsAg positivity, 2.39 (1.50-3.82) from eleven studies in HBsAg+ populations and 5.91 (3.66-9.55) from six studies in HBsAgÀ populations.

Sensitivity analysis
For the meta-analysis of aflatoxin-related HCC risk in the general population, our sensitivity analyses revealed that Wu et al. 30 was the most influential study in determining the summarised OR. After excluding this particular study, heterogeneity was significantly reduced (Q = 8.40, P = 0.30, I 2 = 16.66), and the summarised OR was 5.57 (3.78-7.79).
For the meta-analysis of aflatoxin exposure and HCC in HBsAg+ populations, our sensitivity analyses showed that two studies, Szymanska et al. 29 and Wu et al., 30 substantially influenced the summarised OR. After excluding the two studies, heterogeneity was significantly reduced (Q = 11.16, P = 0.19, I 2 = 28.29), and the summarised OR of HCC risk for detectable vs. non-detectable aflatoxin exposure in HBsAg+ individuals was 2.90 (2.09-4.01). These results suggest that the two studies that measured the association between HCC and aflatoxin exposure in the most recent years 29,30 appear to have significantly different results from relatively earlier studies.
For the 10 studies 6,7,18,20,[22][23][24]27,28,30 associating aflatoxin and liver cancer in the general population, we assessed publication or other forms of selection bias by a funnel plot (Fig. 2) and associated statistical tests of funnel plot asymmetry. 31 Seven studies are not included in this plot; five studied the association in HBsAg+ individuals only, and two are duplicate studies included in meta-analysis for different data extraction purposes, as explained in the methods. The funnel plot provides little evidence of an important departure from symmetry, indicating that publication or other forms of selection bias were not a serious limitation in our meta-analysis. This visual impression of symmetry was corroborated by the statistical tests of funnel plot asymmetry.

Multiplicative model of effects between aflatoxin exposure and chronic HBV infection
The meta-analysis allowed us to quantitatively evaluate the model of effects between the two risk factors aflatoxin and HBV in liver cancer. The summary OR of six studies [5][6][7][8]20,30 reporting ORs of HCC risk from both aflatoxin exposure and HBV is 54.1 (21.3-137.7) with significant heterogeneity (Q = 13.65, P = 0.02, I 2 = 63.36). The summary OR of the same group of studies for HCC from aflatoxin exposure alone is 5.91 (3.66-9.55), while the summary OR on HCC risk from chronic HBV alone is 11.2 (7.48-16.7), both with no significant heterogeneity. When we excluded Wu et al. 30 which contributes most to the heterogeneity, the summarised OR for combined effects increased to 73.0 (36.0-148.3), 6.37 (3.74-10.86) for aflatoxin exposure alone, and 11.3 (6.75-18.9) for chronic HBV infection alone (Fig. 3). These estimates indicate an almost perfectly multiplicative model of effects between aflatoxin exposure and chronic HBV in HCC risk.

PAR of HCC from aflatoxin exposure in each study population
The PAR of aflatoxin-related HCC was calculated for each study population (Table 3). PAR is the proportion of the HCC cases that could be prevented by reducing aflatoxin exposures to "control" levels in each study. For example, HCC in the Chen et al. 18 Taiwanese study population could be reduced by about 10% (2.5-12%) if dietary aflatoxin exposures in this population were reduced such that aflatoxin-albumin adduct levels were below 0.01 fmol/lg (detection limit in this study), or if dietary aflatoxin exposures could be decreased to below 4.3 ng/kg bw/day (biomarker detection limit extrapolated to dietary exposure). HCC in the study population of Shanghai males in Qian et al. 6 could be reduced by about 9.0% (5.9-10.4%) if aflatoxin exposures in this population were reduced to below 6 ng/kg bw/day: the average aflatoxin exposure level in the control group.
Our results showed that the PAR of HCC caused by aflatoxin is higher in HBV+ populations than in HBVÀ populations.
In HBV+ populations in a Taiwanese cohort, the PAR for aflatoxin-related HCC is consistently decreasing, as indicated by a series of follow-up studies: 31% in 1980s 7 , 12% in 1990s, 25 and 3% in 2000s. 30 Overall, the PAR of aflatoxin-related HCC is decreasing in Taiwan in both HBV+ and HBVÀ individuals, from as high as 44% in 1990s 20 to 2% in 2000s. 30 We combined all aflatoxin-exposed cases, HBV+ and HBVÀ individuals, and controls from all the eligible studies to calculate the PAR of aflatoxin-related HCC by HBsAg status and world region (Tables 4 and 5). The PAR of aflatoxin-related HCC in the general population after HBV adjustment is 17% (14-19%). Because the earlier sensitivity analysis demonstrated that the remaining studies after exclusion of Wu et al. 30 do not have statistically significant heterogeneity, we also calculated the PAR of aflatoxin-related HCC after exclusion of. 30 The PAR increased to 23% (21-24%).
The PAR of aflatoxin-related HCC in the HBV+ population is 21% (10-29%). A separate calculation was performed excluding Szymanska et al. 29 and Wu et al., 30 the most influential studies indicated by the sensitivity analysis. The new PAR of aflatoxin-related HCC in the HBV+ population was 25% (18-30%). The PAR of aflatoxin-related HCC in HBVÀ populations is 8.8% (6.7-10%).

Discussion
Aflatoxin exposure is significantly associated with HCC risk regardless of HBV status. Our meta-analyses show that in areas of high aflatoxin exposure and  chronic HBV infection, aflatoxin exposure and HBV have a nearly perfectly multiplicative relationship in increasing HCC risk. In populations including both HBV+ and HBVÀ individuals in the geographic regions studied, the PAR of aflatoxin-related HCC was estimated at 17% (14-19%). This implies that if it were possible to reduce aflatoxin to below detectable limits in these regions, HCC incidence could be reduced by 14-19%. There are roughly 520,000 new HCC cases in China, southeastern Asia and sub-Saharan Africa each year. 32 If the PARs are generalised to these areas, the implication is that, by reducing aflatoxin in human diets to below detectable levels, 72,800 to 98,800 new HCC cases could be prevented every year. If this PAR was generalised to regions of the world beyond Africa and Asia, the overall number of HCC cases (749,000 new cases per year 32 ) that could be prevented by aflatoxin control would reach 105,000-142,000.
The PAR of aflatoxin-related HCC increases to 23% (21-24%), and heterogeneity amongst the studies decreases significantly, if one study 30 is excluded from the meta-analysis. However, this study is important because it suggests that aflatoxin exposure is decreasing over time in the Taiwanese (Penghu) population studied. Our PAR estimates for individual studies showed a decrease in PAR of aflatoxin-related HCC in the Penghu cohort in the last three decades. It is worth noting that in a 1970s food survey, over one-third of peanuts in Penghu were heavily contaminated by aflatoxins, with an average aflatoxin content of 167 lg/kg. 33 Mean urinary aflatoxin in HCC patients in this cohort form was 219 lg/ml in 1991/1992, 7,18 and decreased to 0.017 lg/ ml in HCC patients in the same cohort in 2004. 30 Also, the HBV vaccination programme in Taiwan has successfully reduced HBV prevalence, further reducing HCC risk. 34 In some parts of the world such as Taiwan, aflatoxin exposure is decreasing. In other parts of the world such as Africa, rural China, and Southeast Asia, there is little evidence that aflatoxin exposure is decreasing; in fact, two recent Kenyan events of extremely high aflatoxin levels in maize (in 2004-2005, and again in 2010) suggest the opposite. With climate change, aflatoxin contamination in food crops may become exacerbated due to the conditions favoring proliferation of Aspergilli. 35 Hence, further efforts to reduce aflatoxin-related disease are needed in high-risk areas of the world.
There are several limitations in this analysis. First, the epidemiological studies included were conducted in areas of the world with both high aflatoxin and HBV (Asia and sub-Saharan Africa). Thus, although these regions account for most of the aflatoxin-induced HCC cases worldwide, 13 the estimated PAR is not necessarily applicable in areas with much lower aflatoxin exposures. Second, odds ratios from studies employing food surveys, exposure biomarkers and biological effect biomarkers were combined. This decreases the precision of the analysis, as different biomarkers have different detection limits and measure different endpoints, and food surveys are less precise than biomarkers for exposure estimation. Third, the PAR is meant to represent the proportion by which the disease could be reduced if the risk factor in question was removed. It is not possible to instantaneously reduce aflatoxin to below detectable limits worldwide -rather, the PAR calculated is meant to estimate the burden of HCC caused by one risk factor (aflatoxin) and to project the extent to which the problem could be reduced in future generations if aflatoxin control strategies were widespread.
In summary, this study is the first to quantitatively evaluate the model of effects between aflatoxin and HBV in inducing liver cancer by combining results from multiple epidemiological studies. The range of PARs calculated in this analysis, 14-19% (21-24% excluding one study contributing to heterogeneity), is consistent with our previous report of 5-28% using a different methodology (quantitative cancer risk assessment). 13 The PAR of aflatoxin-related HCC is higher in HBsAg + populations than HBsAg-populations. In recent years, the PAR of aflatoxin-related HCC has shown a decreasing trend in areas such as Taiwan, indicating the benefits of reduced aflatoxin exposure and HBV prevalence by public health interventions.