Cancers in the TREAT Asia HIV Observational Database (TAHOD): a retrospective analysis of risk factors

Background This retrospective survey describes types of cancers diagnosed in HIV-infected subjects in Asia, and assesses risk factors for cancer in HIV-infected subjects using contemporaneous HIV-infected controls without cancer. Methods TREAT Asia HIV Observational Database (TAHOD) sites retrospectively reviewed clinic medical records to determine cancer diagnoses since 2000. For each diagnosis, the following data were recorded: date, type, stage, method of diagnosis, demographic data, medical history, and HIV-related information. For risk factor analyses, two HIV-infected control subjects without cancer diagnoses were also selected. Cancers were grouped as AIDS-defining cancers (ADCs), and non-ADCs. Non-ADCs were further categorized as being infection related (NADC-IR) and unrelated (NADC-IUR). Results A total of 617 patients were included in this study: 215 cancer cases and 402 controls from 13 sites. The majority of cancer cases were male (71%). The mean age (SD) for cases was 39 (10.6), 46 (11.5) and 44 (13.7) for ADCs, NADC-IURs and NADCs-IR, respectively. The majority (66%) of cancers were ADCs (16% Kaposi sarcoma, 40% non-Hodgkin's lymphoma, and 9% cervical cancer). The most common NADCs were lung (6%), breast (5%) and hepatocellular carcinoma and Hodgkin's lymphoma (2% each). There were also three (1.4%) cases of leiomyosarcoma reported in this study. In multivariate analyses, individuals with CD4 counts above 200 cells/mm3 were approximately 80% less likely to be diagnosed with an ADC (p < 0.001). Older age (OR: 1.39, p = 0.001) and currently not receiving antiretroviral treatment (OR: 0.29, p = 0.006) were independent predictors of NADCs overall, and similarly for NADCs-IUR. Lower CD4 cell count and higher CDC stage (p = 0.041) were the only independent predictors of NADCs-IR. Conclusions The spectrum of cancer diagnoses in the Asia region currently does not appear dissimilar to that observed in non-Asian HIV populations. One interesting finding was the cases of leiomyosarcoma, a smooth-muscle tumour, usually seen in children and young adults with AIDS, yet overall quite rare. Further detailed studies are required to better describe the range of cancers in this region, and to help guide the development of screening programmes.

The widespread use of HAART has resulted in decreases in the incidence of KS and NHL [11,21], although a decline in incidence for other cancers is less evident [11]. Additionally, as patients with HIV are living longer, malignancy is becoming an increasingly prominent cause of death [12,[22][23][24][25]. Increasingly reported NADCs include lung cancer, liver cancer, anal cancer and leukaemia.
There are limited data on cancer occurrence in HIVinfected patients in Asia. Investigators at the Ramathibodi Hospital at Mahidol University, Bangkok, Thailand, a collaborating site of the Therapeutics Research, Education and AIDS Training in Asia (TREAT Asia) HIV Observational Database (TAHOD), have retrospectively reviewed pathological reports and medical records on malignancies and treatment outcome in Thai HIVinfected patients. Between 1999 and 2003, 3% of more than 1100 HIV-patients were diagnosed with malignancies. More than half (62%) were ADCs, NHL being the most common. NADCs included breast, colorectal and lung cancer.
In this study, treatment of the malignancy was the only significant factor associated with survival, while age, prior AIDS diagnosis and antiretroviral treatment history were not [26]. In India, among all cancers reported at the Tata Memorial Hospital in Mumbai from 2001 to 2005, 251 cases were identified to be in HIV-positive people, and more than half (56%) were NADCs. Among the ADCs, NHL was the most common, and there were no cases of KS. Among the NADCs, head and neck cancers were the most common [27].
Insight into the patterns of cancer occurrence in HIV/ AIDS can be inferred from studies of cancer-identifying risk factors in other immune-deficient populations. Such populations include organ transplant recipients who undergo iatrogenic immune suppression post-transplantation. A recent large study of cancer occurrence in Australian kidney transplant recipients found a marked increase in cancer risk at a wide variety of sites. After transplantation, 25 cancer sites occurred at significantly increased incidence, and risk increased three-fold at 18 of these sites. Most of these cancers were of known or suspected viral aetiology. These data suggest a broader than previously appreciated role of the interaction between the immune system and common viral infections in the aetiology of cancer [28,29].
Our objective was to undertake a retrospective survey of cancer diagnoses in HIV-infected subjects at the clinical sites in Asia that currently participate in the TREAT Asia HIV Observational Database (TAHOD). The specific aims of this study were to describe the range of cancers diagnosed in HIV-infected subjects in Asia, and to determine risk factors for cancer in HIV-infected subjects in Asia compared with contemporaneous HIVinfected subjects without cancer.

Methods
TAHOD commenced in 2003 and is a collaborative observational cohort study including 17 participating clinical sites in the Asia and Pacific region. A detailed description of this collaboration has been published previously [30]. TAHOD sites that maintained patient visit records from 2000 onwards were invited to participate in this retrospective case-control study. Individual TAHOD sites determined their capacity to review their entire clinic records for cases, or whether they restrict the review to TAHOD only patients.
In total, 13 of the 17 TAHOD sites were able to participate. Eight of the sites reviewed all clinic patient records regardless of whether patients were enrolled in the TAHOD study (n = 7 records from 2000; n = 1 record from 2004), totalling an estimated 12,000 patients. Four sites limited record reviews to patients within the TAHOD study (n = 3) or those participating in clinical trials (n = 1), approaching more than 800 patient records. In total, an estimated 13,000 patient records were reviewed to ascertain cancer cases. One site was not able to recruit controls.
Ethical approval for TAHOD was obtained from the University of New South Wales, Sydney, Australia, and for individual clinical sites from their local institutional review boards, as required. Unless required by a site's local ethics, written consent was not a requirement of sites in TAHOD because data are collected in an anonymous form. All TAHOD study procedures were developed in accordance with the revised 1975 Helsinki Declaration

Selection of cases
Contributing sites were required to review all medical records from 1 January 2000 (or later, if relevant) to 1 January 2008 to ascertain cases of cancer diagnosed. Only the first cancer diagnosed was considered for each case.
To standardize ascertainment and reporting across the sites, a half-day, face-to-face investigator training session was conducted. The facilitators of this training session were cancer epidemiologists from the National Centre in HIV Epidemiology and Clinical Research (NCHECR), and the Lowy Cancer Research Centre, University of New South Wales. The training included issues of determining morphology (histologocial classification of the cancer tissue, site and staging of the cancer, as well as establishing a date of diagnosis). The training also included describing the sources and hierarchy of information for a cancer diagnosis (e.g., pathology, biopsy, or cytology report, laboratory data, imaging, treatment details, autopsy report).
Participating sites were advised to report all pathological diagnoses of invasive/malignant, in situ or unknown/ uncertain neoplasms. In the absence of histologic or cytologic confirmation, sites were advised to report a case based on a clinical diagnosis made by a recognized medical practitioner.

Selection of controls
For each case, two contemporaneous cancer-free, HIVpositive controls were selected from a complete list of patients attending the respective clinic on the day or corresponding week that the cancer case was first diagnosed. If both daily and weekly patient lists were available, then sites randomly selected two controls from the daily patient list. Instructions were provided to the sites to ensure a standardized approach for the selection of cases and control. The selection of controls was determined using the RANDBETWEEN function in Excel.

Data collection
Data were entered into an Access database developed at the NCHECR. The data were then forwarded to the NCHECR for case review and confirmation. The following data were obtained from patient medical records and reported for both cases and controls: date of birth (or age); sex; mode of HIV exposure (patient selfreport); date of first positive HIV test; ethnicity; hepatitis B (HBV) and hepatitis C (HCV) status (defined as HBV surface antigen positive and HCV antibody positive, respectively); AIDS-defining illness diagnosed prior to case diagnosis; CDC stage; CD4 cell count at diagnosis; smoking and alcohol use (patient self-report); antiretroviral treatment history; and date of death (if known).
Case data included: date of diagnosis; site; morphology; method of diagnosis (e.g., pathology, cytology, radiology, laboratory data, clinical diagnosis, death certificate); stage; node; and class scheme. All measures (excluding death, patient demographics and cancer treatment) were recorded at the time of the cancer diagnosis for the case, or at the time of the corresponding clinic visit for the control.

Case validation
Cancer cases were reviewed by a medical cancer epidemiologist at NCHECR, and clarification was sought from the sites as needed.

Data analysis
All cancers (excluding in situ neoplasms) were categorized into the following groups based on published reports (29)(30)(31): ADCs (KS, NHL and cervical cancer); NADCs infection-unrelated (NADCs-IUR); and NADCs infection-related (possible/probable) (NADCs-IR). NADCs-IR included: hepatocellular carcinoma, Hodgkin's lymphoma, leiomyosarcoma, and cancers of the anus, bladder, larynx, nasopharynx, oral cavity, penis, stomach, tongue and tonsils [29,31,32]. Key baseline demographic, HIV disease stage and health status were also summarized. Baseline was defined as the date on which the cases were first diagnosed with cancer, and for the controls, the date on which the control attended the clinic (on the day of, or within one week of the matched cases diagnosis date).

Statistical analysis
Conditional logistic regression methods were used to determine factors associated with ADCs and NADCs. The following baseline demographic and clinical factors were assessed as covariates: age; mode of HIV exposure; ethnicity; prior AIDS; CDC stage; CD4 cell count (within one year prior to case diagnosis); HBV and HCV status; antiretroviral treatment history; and smoking and alcohol use.
A sensitivity analysis for the NADC-IR endpoint was also conducted, excluding bladder, larynx and oral cavity cancers, less than 20% of which had been attributed to infections. All covariates with p < 0.100 in univariate analyses were assessed in the multivariate models. The final model included only covariates with p < 0.05. Forward stepwise methods were used. The site that was unable to identify controls was excluded from risk factor analyses. Analyses were conducted using Stata V10.0 (Texas, USA) and SAS V9.1 (Carey, NC, USA) statistical packages.

Results
A total of 617 patients were included in this study, including 215 cancer cases and 402 controls from 13 sites (nine cancer cases were from the site that did not recruit controls). The majority (65%) of cancer cases were ADCs (n = 141); of these, 62% were NHL, 24% were KS, and 14% were cervical cancers ( Table 1). The majority of KS cases were from Hong Kong (29%), Malaysia (18%), Philippines (15%), Bali (12%) and Taiwan (12%). One case was from Thailand, and there were no cases of KS reported from the India or Singapore sites (data not shown). Almost all KS cases were among men (32 of 34), of whom only 41% reported homosexual contact as mode of HIV exposure. Among NHL cases, 60% were among males, and 19% reported male homosexual contact as mode of HIV exposure.
Patient demographics are summarized in Table 2. The majority of cancer cases were male (71%), and the mean age (SD) for the cases was 39 years (10.6) for ADCs, 46 (11.5) for NADCs-IUR and 44 (13.7) for NADCs-IR. The rate of current smoking was greater among the NADC-IR (15%), compared with 9% and 10% among the ADC and NADC-IUR groups, and 12% among the controls. Homosexual contact as mode of HIV transmission was reported by 19% of those with ADCs, compared with 6% among those with NADCs-IUR, 15% among those with NADCs-IR, and 11% among the controls. When cases of KS were excluded from the ADC group, only 13% reported homosexual exposure as mode of HIV transmission.

Predictors of ADCs
In univariate analyses, homosexual contact as the mode of HIV exposure (p = 0.004), CDC stage C (p = 0.035)   3 were approximately 80% less likely to be diagnosed with an ADC (p < 0.001). After adjustment for these independent predictors, CDC stage was borderline significant (p = 0.058) ( Table 3). We also underwent a sensitivity analysis removing the KS cases and their controls from the logistic regression. Although the risks remained broadly similar for each of the exposure groups, overall, exposure category was no longer significant, and CD4 cell count remained the only independent risk factor for ADCs (data not shown).

Predictors of NADCs
Increasing age, a history of smoking status and not receiving antiretroviral treatment were significantly associated with increased NADC risk overall in univariate analyses (p < 0.001; p = 0.024; p < 0.001, respectively). Declining CD4 cell count was borderline significant (p = 0.056). Older age (OR: 1.39, p = 0.001) and currently receiving antiretroviral treatment (OR: 0.29, p = 0.006) remained as the only independent predictors of NADCs overall in the multivariate model.
After adjustment for these predictors, CD4 cell count and smoking status were no longer significantly associated with NADCs (p = 0.419 and 0.296, respectively) ( Table 4). We also adjusted CD4 cell count for age and smoking covariates independently, and CD4 cell count remained non-significant (p = 0.398 and p = 0.090, respectively). In the regression analyses limited to the NADCs-IUR, increasing age and non-receipt of antiretroviral treatment were significant predictors in both univariate (p = 0.001 and p = 0.002) and multivariate analyses (p = 0.005 and p = 0.008) ( Table 5).
Due to the small numbers, covariates assessed for association with NADCs-IR were limited to factors that were significant in univariate analyses for either the ADC or NADC overall endpoints. These included age, CDC stage, CD4 cell count, smoking status and antiretroviral treatment. As HCC was one of the most common NADCs-IR, and all five cases were either HBV surface antigen (n = 4) or HCV core antibody positive (n = 3), we also assessed coinfection with HBV or HCV. In univariate analyses, increasing age (p = 0.023), increasing CDC stage (p = 0.032), CD4 category >200 cells/mm 3 (p = 0.041), and ever smoking (p = 0.020) were associated with NADCs-IR (Table 6). CD4 cell count and CDC stage remained significant in the multivariate analyses (p = 0.041 each). In the sensitivity analysis excluding bladder, larynx and oral cavity cancers (n = 3), the results remained largely unchanged (data not shown).

Discussion
In this retrospective case-control study, more than half (66%) the cancer cases identified were ADCs. The remaining cancers were either NADCs-IUR (22%) or NADCs-IR (12%). NHL was the most commonly reported ADC overall, as well as among men, while cervical cancer was the most common among women. Lung and breast cancers were the most commonly ART: antiretroviral treatment; IDU: Injecting drug user Missing reported NADCs overall, and hepatocellular carcinoma was the most common NADC-IR. Factors associated with ADCs were immunodeficiency and lower CD4 cell count, while among NADCs overall and for NADCs-IUR, factors were older age and not currently receiving antiretroviral treatment. CDC stage C and lower CD4 cell count were significantly associated with NADCs-IR. A novel finding in our study was the reporting of KS cases. KS has been thought to be relatively rare in some countries of Asia, largely attributed to the low prevalence of human herpes virus (HHV8) known to cause KS [33,34]. KS cases were largely reported from the Hong Kong site, and from Malaysia, the Philippines and Taiwan. In the Thai study, KS was reported in only 5% of ADCs [26], and no cases of KS were observed in the Mumbai study [27]. Almost all the KS cases in our study were among men (94%), of whom only 41% reported homosexual contact as the mode of HIV transmission. In western countries, KS occurs predominately among homosexual men [21]. We believe that our findings may reflect underreporting of male-to-male sex as a primary or concomitant risk factor for HIV infection in Asian countries.
The increased incidence of NADCs has been extensively reported in the literature [5,8,[35][36][37]. Specific NADCs that have been reported to be higher in HIVpositive people than in HIV-negative people include Hodgkin's lymphoma, lung cancer, hepatocellular carcinoma [38], anal, vaginal, oropharyngeal, colorectal cancers, melanoma and leukemia [39], and cancer of the lips and testis [7].
Even in developing countries, where HAART is largely unavailable, the incidence of NADCs has increased. In India, there has been an increase in anal cancer, Hodgkin's lymphoma, testicular and colon cancers, and head and neck cancers [27]. In our study, lung and breast cancers were the most commonly reported NADCs, as well as head and neck cancers. Although lung cancer has been identified as one of several NADCs at increased incidence in HIV-infected patients, for breast cancer, the evidence of increasing incidence is still equivocal [31,35,40,41]. In the Thai study, the prevalence of breast cancer was 3%, and 10% of all NADCs, a little lower than our 14% of NADCs [26]. Head and neck cancers have been reported as the most common NADC in one Indian study [27].
Among the NADCs-IR, HCC was the most frequently reported. HCC is a commonly reported NADC in the literature, and is likely to remain important in HIVinfected populations, particularly in the context of coinfection with HBV and HCV [22]. In our study, all five cases were either HCV or HBV positive. Although we did not find a statistically significant association of HCV or HBV coinfection and NADCs-IR, this may explained by low numbers, but may also likely be due to the inclusion of other cancers in this analysis, whose primary risk is not HBV or HCV infection.
Also of particular interest were the three cases of leiomyosarcoma reported, all in women. Leiomyosarcoma, smooth-muscle tumours, are usually seen in children and young adults with AIDS [42], yet overall are quite rare. Epstein-Barr virus has been associated with this tumour, and only sporadic case reports have been published [43].

Factors associated with ADCs and NADCs
Immune suppression and exposure category were independently associated with ADCs in our study, similar to what has been reported in the literature [3,39]. The risk of an ADC decreased by approximately 80% with CD4 cell counts >200 cells/mm 3 , while patients who reported homosexual contract as the mode of HIV infection had an almost three-fold increased risk of an ADC compared with those who reported heterosexual contact, and sixfold increased risk compared with those who reported injecting drug use. However, the exposure category effect we observed is largely explained by a greater proportion of KS cases reporting homosexual contact. In the sensitivity analyses where KS cases were excluded, the homosexual exposure effect was removed, and immunodeficiency remained the only significant factor. A substantial proportion of patients in this study were diagnosed with HIV at the time of the cancer diagnosis, with more than a quarter of cancers diagnosed within one month of HIV diagnosis (28% and 26% of ADCs and NADCs, respectively). Late diagnosis may reflect particular marginalized groups, such as men who have sex with men, and their reluctance to seek medical care.
Older age and having never received antiretroviral treatment, but not CD4 count, were associated with being diagnosed with any NADC, both overall and for NADC-IUR, similar to other studies [3,37]. Longer  [35]. In our study, immune suppression (CDC stage and lower CD4 cell count) was associated with NADCs-IR. Several studies examining CD4 and NADCs during the HAART era have not found an association with CD4 [3,7,37]. However, these studies did not restrict their analysis to NADCs that are currently thought to be infection related. More recent data have shown several NADCs in HIV-positive patients to be increasing, similar to the risk seen in transplantation recipients, suggesting a link between immune suppression and increased risk of a specific NADC [29].
Smoking was not associated with NADCs-IUR, despite lung cancer being the most commonly cancer reported in this group, and yet a mild association was observed for NADCs-IR. We believe the lack of association with the NADC-IUR group is due largely to the inclusion of cancers where smoking is not a primary risk factor. The association with smoking and NADCs-IR on the other hand may reflect risk behaviour, where people who may engage in unprotected sex or drug use, which lead to viral coinfections, may also be more likely to smoke.
There are some limitations to this study. First, this is a retrospective survey of cancer cases and controls, and subject to selection bias and concerns regarding complete ascertainment of cases. Most sites reviewed all patient records for case ascertainment since 2000, while some sites reviewed more recent time periods or restricted review to TAHOD patients. Furthermore, there may be variation between TAHOD sites in the level of cancer screening, despite sites being primarily urban referral centres. Care must be taken in generalizing our results to Asian patients in other types of clinical centres or geographic locales.
Second, our division between NADCs-IUR and NADCs-IR may be debated. Included in our list of NADCs-IR were cancers where only a small proportion (≤20%) were attributed to infections (including bladder, larynx and oral cavity). However, we did conduct a predictor analysis for NADCs overall, as well as dividing cases by relationship to infection, and our results were consistent with what has been shown previously in the literature. Finally, we did not have enough NADC cases to examine patient characteristics and factors associated with individual cancers potentially masking risk factors for specific cancers, such as that demonstrated by smoking and lung cancer. Nor did we collect any other behavioural data beyond smoking and alcohol use.

Conclusions
In conclusion, the spectrum of cancer diagnoses in the Asia region currently does not appear dissimilar to that observed in non-Asian HIV populations. Key findings in our study include the much more commonly diagnosed KS than expected, as KS has been widely thought to be rare in Asia. This may reflect a common misreporting of HIV exposure among men in Asia. Second, we identified three cases of leiomyosarcoma, a rare and uncommonly reported malignancy, with links to Epstein-Barr virus.
Third, this is the first study to our knowledge to have examined both infection-related and infectionunrelated NADCs in the Asia region. Given the diversity in prevalence and infectious agents between countries within this region, further detailed studies are required to better describe the range of cancers in this region. To this end, new diagnoses of cancers have been prospectively reported in TAHOD since 2008. These studies will help guide where screening programmes are needed.