Cancer risk factors among people living with HIV/AIDS in China: a systematic review and meta-analysis

Cancer is a leading cause of death among people living with HIV/AIDS (PLWHA). We conducted a systematic review and meta-analysis to evaluate prevalence of cancer risk factors among Chinese PLWHA based on 102 articles. Random effects meta-analysis was used to calculate the summary prevalence estimate (sPrev) and 95% confidence interval (CI) for each cancer risk factor by demographic group. Overall, the sPrev for each risk factor among Chinese PLWHA was: 41.1% (95% CI: 35.3–46.9%) for current smoking; 30.3% (95% CI: 23.3–37.4%) for current alcohol consumption; 24.4% (95% CI: 14.7–30.2%) for overweight and obesity; 12.5% (95% CI: 10.6–14.3%) for hepatitis B virus infection; 29.1% (95% CI: 23.6–34.5%) for hepatitis C virus infection; 33.9% (95% CI: 24.3–43.5%) for high-risk human papillomavirus infection from cervical samples and 78.6% (95% CI: 69.4–87.7%) from anal samples; 2.7% (95% CI: 0.7–4.7%) for Epstein-Barr virus (EBV) immunoglobulin M (IgM) positivity, 94.7% (95% CI: 90.7–98.8%) for EBV IgG positivity and 25.6% (95% CI: 12.4–38.8%) for EBV DNA positivity; 14.9% (95% CI: 12.4–17.4%) for human herpes virus 8 infection. The prevalence of major cancer risk factors was high among PLWHA in China, suggesting an urgent need for interventions to reduce cancer risk in this high-risk group.

Full-text review   (Table 3). All comparisons between PLWHA and HIV uninfected group reported significantly higher prevalence of HPV in PLWHA from the same studies (Table 4).

Epstein-Barr virus infection. Nine publications presented the prevalence of Epstein-Barr virus infection.
The number of studies for meta-analysis ranged from 1 to 4 when stratified by EBV testing method and demographic groups. Immunoglobulin M (IgM) was tested in 4 studies and the sPrev was 2.7% (95% CI: 0.7-4.7%,   Table 3). Significantly higher prevalence was consistently found in PLWHA than in the HIV-negative comparison groups from the same studies (Table 4).

Sensitivity analyses and publication bias.
We recalculated each sPrev omi for all studies except the "omitted" one in turn, with the results of full meta-analyses named as sPrev all . The influence was evaluated by  In our meta-analysis, substantial publication bias was significantly observed by the Egger test and/or the Begg test in estimating the overall sPrev of smoking (P-value for Egger test was 0.028, P-value for Begg test was 0.057), the overall sPrev of alcohol consumption (P Egger = 0.002, P Begg = 0.007), the overall sPrev of HBV infection (P Egger = 0.791, P Begg = 0.018), and the sPrev of HCV infection overall (P Egger = 0.141, P Begg = 0.014), in females (P Egger = 0.015, P Begg = 0.344), and in males (P Egger = 0.095, P Begg = 0.621).

Discussion
To our knowledge, this is the first comprehensive meta-analysis to report the summary prevalence of cancer risk factors among PLWHA in China. We found that Chinese PLWHA had higher prevalence of smoking, alcohol consumption, and virus infections including HBV, HCV, HPV, EBV and HHV8, but had lower prevalence of overweight and obesity when compared with the HIV-negative participants from the same studies and the Chinese general population.  For such two modifiable behavioral risk factors as tobacco and alcohol use, Chinese PLWHA showed high prevalence of tobacco smoking and alcohol drinking overall and especially among males. For example, the smoking prevalence is 53.3% for general Chinese men 117 , while the sPrev is even higher, reaching 63.2% for male PLWHA. Such elevated smoking prevalence in HIV infected people was also found in independent Chinese studies comparing with HIV-negative participants 24,35 , and in a meta-analysis in western countries with lower smoking prevalence in general population 118 . The sPrevs also varied by group-specific analyses, with certain subgroups at higher risk in tobacco and alcohol use. The overall sPrev of alcohol drinking is close to the prevalence from Chinese general population (30.3% vs. 28.8%) 119 while MSM showed slightly higher sPrev (35.7%) and IDUs showed significantly lower estimate (13.1%). As group I carcinogens defined by IARC, tobacco smoking and alcohol consumption are not only associated with a series of human cancers, but also with other important chronic diseases. Based on our summarized estimates, the high prevalence of tobacco and alcohol use especially among male PLWHA suggested the need for an effective and more targeted intervention in smoking and drinking behaviors as well as a close observation of their roles in HIV and cancer disease progression to reduce morbidity and to improve quality of life in this aging population with prolonged life expectancy.
The prevalence of cancer-related viral infections, including HBV, HCV, HPV, EBV and HHV8 infection, are significantly higher in PLWHA than in general Chinese population [120][121][122][123][124][125][126] , shown by the results from both meta-analyses and independent studies. By demographic group, 10.4-13.4% of PLWHA were HBsAg positive, and 29.1-47.1% were HCV antibody positive. Moreover, the highest prevalence of HBV (32.8%) and HCV (83.4%) infection were both found in IDUs, probably due to the shared transmission route with HIV. For female PLWHA, the sPrev of both any type of HPV (42.0%) and high-risk type of HPV (33.9%) from cervical samples were higher than the prevalence in Chinese general population aged 15-60 years 122,123 . Moreover, we found that in anal samples collected from male PLWHA, the HPV prevalence was very high no matter whether they were MSM or not. These observations call attention since HPV infection is also a risk factor for anal cancer. The high seropositivity of EBV-IgG was close to the prevalence reported earlier in the Chinese general population (97.6%) 125 . Although the latter statistic might be limited to experimental conditions and in need of update, it is believed that EBV infection is very common among adults 127 . Moreover, several studies suggested higher detection rate of EBV DNA among HIV-infected subjects compared to HIV-negatives, suggesting potentially more active viral application in immune-suppressed population 110,128,129 . Last but not least, the overall seroprevalence of HHV8 infection was 14.9%, higher than the prevalence in the Chinese general population 126 , with the highest prevalence found in MSM (32.3%, 95% CI: 24.2-40.4%).
The overall prevalence of overweight and obesity among PLWHA was 22.4% (95% CI: 14.7-30.2%), lower than the prevalence of 42.6% in Chinese general population 119 . This is similar with the observations in western countries that the obesity prevalence was lower in HIV population 118 . We need to further observe the change in weights in PLWHA to have a better understanding and balance in control of both the HIV wasting syndrome and risk for obesity-related cancers and chronic diseases.
A meta-analysis was recently published on prevalence of risk factors for NADCs among PLWHA in western high-income countries 118 , which, consistent with our analysis, also observed higher prevalence of cancer risk factors among PLWHA than the general population. However, the prevalence level seems to be different between PLWHA in western countries and in China. Compared with PLWHA in western countries, PLWHA in China seemed to have higher prevalence in alcohol consumption (30.3% versus 24.0%), HCV (29.1-47.1% versus 23.0-28.0%) and HBV (10.4-13.4% versus 4.0-5.0%) infection, but lower prevalence in smoking (41.1% versus 54.0%), overweight and obesity (22.4% versus 53.0%) and HPV infection (cervical high-risk types: 33.9% versus 46.0%, anal high-risk types: 52.1% versus 66.0%). Thus, the spectrum of cancers among PLWHA in China might be different from that in western countries and further investigations and interventions might be targeted to different high-risk groups.
Substantial heterogeneity across subgroups was significantly observed with I 2 values larger than 90%. This was expected since heterogeneity could be found in almost all 'overall' groups, because we could not know the proportion of genders or behavioral characteristics in these study populations across publications. So was the case in 'male' groups, which could have different proportions of heterosexual male, MSM and IDU. Meanwhile, extensive heterogeneity still existed in specific demographic groups such as female, MSMs and IDUs. This could be attributable to many potential factors including study design, sampling method, geographical regions, study period, age, ethnicity, the definition or method of measurement for risk factors, differences in CD4 + and the length of time in ART, etc 118,130 . Insufficient number of studies has limited our further stratification analysis. However, random effects models were used to minimize the impact of heterogeneity on precision and provided rational approximations of pooled prevalence estimates in our meta-analyses. Sources of heterogeneity should be further explored in the future with more research for specific subgroups.
Several limitations in our study should be noted. First, there might be some information bias since the methods for measuring these risk factors might vary across studies. Most (69.2%) studies did not give a clear definition of smoking habits (e.g., ever smoking: have smoked at least 100 cigarettes in life; Current smoking: smoked within the past 30 days) or did not have a precise measurement for smoking dose or frequency, while only 50% of studies presented the range of time (past 30 days or a day) for current alcohol drinking. Although we tried to extract information from reference studies using standardized definitions, misclassification might still exist since not every study gave a clear definition for each risk factor. Also, the testable HPV subtypes varied across studies, and therefore, underestimation of HPV prevalence might have occurred in studies with fewer testable HPV subtypes. Second, the representativeness of our meta-analyses is compromised since it covered 24 but not all 31 provinces or municipalities in China, underscoring the need for expanded research and surveillance efforts on cancer risk factors among PLWHA all over the country. Third, comparability was not considered across different sources of data due to time lags among studies. Some risk factors may present a temporal trend, some may not. For those with an increasing or decreasing trend, it may have compromised the comparison. Fourth, as prevalence estimates of cancer risk factors can vary with characteristics including age, lack of matching for age and other factors also limited the validity of the comparisons between PLWHA and the general population. However, consistent results were still observed when stratified roughly by age where possible (Supplemental Table S1). Fifth, substantial publication bias was observed according to rigorous publication bias tests, and might be attributable to several factors, including the presence of extensive heterogeneity, which we have discussed above 131,132 . The association between behavioral changes such as smoking and alcohol consumption and severity of HIV patients might lead to underestimation of prevalence of cancer risk factors among PLWHA.
Despite these limitations, the present study provides a broad overview of prevalence of cancer risk factors among PLWHA in China. Both in comparison with the Chinese general population and with the uninfected comparison groups, the studies reach a wide consensus on higher prevalence of cancer risk factors among PLWHA. Along with the management of HIV individuals and ART program, interventions to reduce cancer risk factors should also be implemented in this population. Cancer prevention measures in PLWHA should include smoking cessation, HBV and HCV treatment, vaccination against HBV and HPV, annual cervical and anal Pap tests and cancer screening according to relevant guidelines 4,[133][134][135][136][137][138] . Furthermore, a better understanding of cancer risk attributable to specific factors in different HIV infected groups is needed to develop strategies in preventing cancer. Study selection. One author (Jin Z-Y) identified relevant articles, excluding those with unrelated titles and abstracts before obtaining full-text references. Two independent authors (Jin Z-Y and Liu X) conducted full-text review. Cross-sectional, case-control, prospective or retrospective cohort and intervention trial studies were included. Studies with sample size greater than 100 were considered to have a sufficiently stable prevalence estimate to be qualified for inclusion in the study. Studies lacking original data for the numerator or denominator of the prevalence estimate were excluded. If information from one study had been published for more than once, we chose the publication with more detailed information such as prevalence estimates stratified by specific demographic groups, studies with a larger sample size, or those most recently updated. Data extraction. Data were extracted and organized using Microsoft Excel (Version 2010, Microsoft Corp., Redmond, WA, USA). For each risk factor, the number of PLWHA with known status was recorded as the denominator and the number of positive subjects was recorded as the numerator. The participants with unknown status were excluded from the calculation. If an article contained more than one of the mentioned risk factors, we kept all of them. If the PLWHA were stratified into different groups such as females, males, men who have sex with men (MSM) or injection drug users (IDU), we included data from each subgroup for prevalence estimate. The definitions of smoking and drinking status including "ever/never", "former/current" were based on the descriptions in the original studies when a clear definition is not available. We also extracted prevalence estimates for HIV uninfected comparison groups if data were available from the same study. Other information including the first author's name, year of publication, study site, original study design, sampling frame, calendar years of research, major information for participants including the mean or median age, percentage of males, MSM, IDU, treatment with antiretroviral therapy (ART), and the measurement of risk factors were also extracted.

Methods
We also extracted prevalence estimates for cancer risk factors among the general population from the most representative surveys. For tobacco smoking, alcohol consumption, overweight and obesity, prevalence estimates were extracted from a national survey, which was carried out in Chinese adults (aged 18 years and over with the median age in the range from 45 to 49 years) at 162 disease surveillance points (DSPs) from 31 provinces in 2010 117,119 . For HBV and HCV infections, prevalence estimates were obtained from a national serosurvey in 2006, of which participants were local residents aged 1-59 years (the median age was in the range from 15 to 59 years for HBV and in the range from 15 to 20 years for HCV) living in 31 provinces with 160 DSPs 120,121 . For HPV infection, prevalence estimates were acquired from a nationwide investigation of female subjects aged 15-60 years in 37 Chinese cities in 2012, a review (age ranging from 15 to 59 years) and a population-based study among men aged 25-65 years (the median age was 43 years) [122][123][124] . For EBV infection, the prevalence estimate was extracted from a 1989 study by Liu 125 . For HHV8 infection, the prevalence estimate was from a systematic review and meta-analysis published in 2012 126 .
Statistical analysis. PLWHA were stratified into overall, female, male, MSM, and IDU demographic groups when such subgroups were available. The 'Overall' group represents PLWHA not restricted by gender or HIV transmission category (i.e., did not include a research on specific group by gender or HIV transmission category). Similarly, the 'male' group implies that PLWHA were not restricted by HIV transmission category. A meta-analysis of prevalence was performed for each risk factor by demographic group if there was more than one study included. Heterogeneity tests across subgroups were examined by Q-test (P < 0.1 refers to statistically significant heterogeneity) and I 2 values 139,140 . Random effects models were used because of significant heterogeneity across subgroups 141,142 . DerSimonian and Laird method (D-L) was applied to estimate the summary prevalence estimate (sPrev) and its 95% confidence interval (CI) for each risk factor 143 . If only one study was presented for the specific subgroup, we showed the single prevalence and its 95% CI. In sensitivity analyses, each sPrev was recalculated excluding studies evaluated as having higher potential for bias. Lastly, we performed a publication bias test when the meta-analysis included more than 10 studies 144,145 . Publication bias was assessed through the Egger weighted regression and Begg rank correlation method (P < 0.1 represents statistically significant publication bias) 146,147 . All the analyses were conducted by Stata 12.0 (Stata Corp LP).