People infected with human immunodeficiency virus (HIV) have an elevated risk of cancer.1–3 This increase is largely due to markedly elevated risk for 2 malignancies, Kaposi sarcoma (KS) and non-Hodgkin lymphoma (NHL), which are themselves considered sufficient to signify progression to the advanced stage of HIV infection, i.e., acquired immunodeficiency syndrome (AIDS).4 KS and NHL are caused by loss of immune control of latent infection with oncogenic viruses (human herpesvirus 8 for KS, Epstein Barr virus for certain NHL subtypes).5 Other cancers caused by viruses (e.g., cervical and anal cancers by human papillomavirus (HPV), liver cancer by hepatitis C and B viruses) also arise at increased frequency in this population, although the importance of immunosuppression is less clear.6, 7 Among HIV-infected people in the U.S., heavy use of tobacco and alcohol contribute to the development of additional cancers, especially lung and liver cancers.
Much of the existing data on cancer risk among HIV-infected people are derived from cohort studies and large registry-based studies of people with AIDS.2, 3, 8–13 Comparatively, few studies have provided information on HIV-infected individuals before their progression to AIDS.1, 14–16 This lack of data is noteworthy, because the widespread use of highly active antiretroviral therapy (HAART) since 1996 has led to dramatic improvements in immune status and prolonged survival among HIV-infected persons.17–21 An estimated 1.2 million people were living with HIV infection in the U.S. at the end of 2003; of these an estimated 42% had not yet progressed to AIDS, 34% had AIDS and 24% were undiagnosed.22 With HAART-related improvements in immune function, many HIV-infected people may never develop AIDS, and increasingly people with AIDS are less representative of the greater population with HIV infection. Persons only marginally immunosuppressed may still be at excess risk of developing AIDS-associated and other cancers, especially with extended survival.
The goal of the present study was to describe patterns in cancer incidence among HIV-infected people beginning at a point in time before development of AIDS. We sought to determine the importance of immunity in development of cancer by measuring associations with subsequent AIDS onset and with CD4 count, and to describe trends in cancer incidence over time with respect to the introduction of HAART in 1996. We used registry data from the HIV/AIDS Cancer Match Study, focusing on people first registered with HIV infection before the development of AIDS. Our study is the largest investigation yet of cancer risk in this moderately immunosuppressed population.
Materials and methods
Study population and general approach
The HIV/AIDS Cancer Match (HACM) Study links HIV/AIDS and cancer registries in multiple U.S. states and metropolitan areas to identify cancers arising in HIV-infected individuals.12 Registry linkage was conducted using a probabilistic matching algorithm, utilizing data on name, social security number, sex, dates of birth and death, and race. Registry personnel reviewed potential matches to exclude those judged unlikely to be a true match. Subsequently, identifying information was deleted from files retained for analysis. For the present study, we utilized HACM Study data from those HIV/AIDS registries that had implemented name-based registration of HIV infection and provided data for linkage. The study thus included data from three HACM sites: Colorado (subjects registered with HIV beginning 1991), Florida (1997) and New Jersey (1992). At the time of linkages in 2004–2005, cancer registry data were complete for 1988–2002 (Colorado), 1981–2002 (Florida) and 1979–2002 (New Jersey), and met standards for certification by the North American Association of Central Cancer Registries (www.naaccr.org). Institutional review boards at participating registries approved the study.
We used data in the HIV/AIDS registry to define a cohort of individuals who were prospectively reported with HIV infection before AIDS onset. Thus, individuals registered with HIV infection were included if either (i) they were never reported to develop AIDS,4 or (ii) they had an HIV diagnosis date and HIV report date that both preceded their AIDS diagnosis date by at least 3 months. We evaluated these individuals for incident cancers arising during a 4–60 month follow-up period after the date of HIV report to the registry (i.e., HIV registration); the median time from HIV diagnosis to HIV registration was 2 months (interquartile range 1–7 months). By definition, these subjects were AIDS-free at the start of the follow-up period, although some later developed AIDS. We identified 58,015 people registered with HIV infection who were covered by data in the corresponding cancer registry for at least some portion of the follow-up period. Because general population cancer rates were not available for persons of race/ethnicity outside the major categories, 665 such individuals were excluded, leaving 57,350 subjects (Table I).
Table I. Demographic and Immunologic Characteristics of 57,350 People Registered with HIV Infection in the United States (1991–2002)
Colorado (N = 8,285)
Florida (N = 26,896)
New Jersey (N = 22,169)
All subjects (N = 57,350)
MSM, men who have sex with men, IDU, injection drug user.
Entries in table are n, % unless otherwise stated.
Median CD4 count is specified for those subjects with non-missing CD4 counts.
Invasive cancers were identified through linkage with the cancer registries. Malignancies were coded according to the International Classification for Diseases for Oncology, third edition23 and analyzed by site using the Surveillance, Epidemiology and End Results (SEER) program's “Site recode with KS and mesothelioma,”24 except that, because of small numbers we collapsed some cancer subtypes and included some rare malignancies with cancers of unknown site in “other/unknown site.” We classified NHLs by histology (Burkitt NHL and diffuse large B cell lymphoma [DLBCL], which are AIDS-defining, and other histologic subtypes) and separately considered central nervous system (CNS) NHL, regardless of histology, which is also AIDS-defining.4 Cancers at any site with poorly specified histology (codes 8,000–8,005) were grouped separately, since they might not have been confirmed pathologically and could represent KS or NHL.
Subjects were followed for incident cancers from the beginning of month 4 after HIV registration or the start of cancer registry coverage, whichever was later, until the earliest of 60 months after HIV registration, death, or last date that cancer registry coverage was considered complete. Follow-up was truncated at 60 months, because migration out of the registry area after this time could reduce incidence estimates. Both first and subsequent cancers, if any, were included in analyses. We report cancer incidence and standardized incidence ratios (SIRs), which compare incidence to that in the general population. The SIR is defined as observed cancer incidence among HIV-infected subjects divided by that expected based on population rates. SIRs were calculated using general population rates derived from data provided by the cancer registries. For most malignancies, the expected rates were specific to sex, age, race/ethnicity, calendar year and cancer registry. However, because the great majority of KS and CNS NHL cases in the general population occur in HIV-infected person, SIRs calculated using contemporaneous rates yield biased estimates of risk related to HIV/AIDS.25 Therefore, for these 2 malignancies, SIRs were calculated using data collected by SEER registries before the AIDS epidemic (1973–1979). Exact confidence intervals (CIs) were calculated for SIRs.
We conducted additional analyses for malignancies for which SIRs were significantly elevated (p < 0.05). We divided each subject's person-time into approximate 1-year intervals (4–12, 13–24, 25–36, 37–48, 49–60 months after the HIV registration date). Each interval was classified according to whether an AIDS diagnosis was present before the start of the interval. Likewise, we classified the 1-year intervals according to whether the interval began in 1991–1995 or 1996–2002, as an indication of the availability of HAART. We then estimated incidence and SIRs for the malignancies of interest according to these strata. For KS, NHL and cervical cancer, these estimates for people who initially did not have AIDS approximate the risk of developing one of these malignancies as an AIDS-defining event. We also calculated incidence and SIRs according to the CD4 count measured at the start of follow-up (−6 to +3 months relative to HIV registration date). We used Poisson regression to derive relative risks (RRs) comparing incidence according to calendar period and prior AIDS onset, and to test for trends across categories of CD4 count. When there were at least 20 cancer events, these analyses were adjusted for sex and transmission category in 3 categories (men who have sex with men, other males, females), age at HIV registration, and race/ethnicity; otherwise, the analyses were unadjusted. Analyses in which we deleted subjects from Florida (which only provided HIV registration data beginning in 1997) yielded similar results (not shown).
Study subjects registered with HIV infection
The demographic characteristics of subjects registered with HIV infection are shown in Table I. Overall, the majority of the cohort was male, and the median age at HIV registration was 35.8 years. Other demographic characteristics varied across the 3 states (Table I). In Colorado, most subjects were non-Hispanic white, while in Florida and New Jersey the majority were non-Hispanic black. The Colorado subjects included a large proportion who were men who have sex with men, whereas in the other 2 states, there were large proportions of injection drug users or heterosexuals; the proportion without information on HIV transmission category also varied across states. Approximately one-third of the cohort, although none from Florida, were first registered in the pre-HAART era (1991–1995), and two-thirds were registered in the HAART era (1996–2002). Among the 15,697 subjects (27.4%) with CD4 count data, the median CD4 count at HIV registration was 491 cells/mm3 (Table I).
For the 5-year follow-up period 4–60 months after HIV registration, these subjects contributed 186,157 person-years of follow-up at risk for cancer (mean 3.2 years per subject). During follow-up, 14,135 (24.7%) subjects developed AIDS and 4,849 (8.5%) died. Most follow-up time occurred prior to an AIDS diagnosis (167,132 person-years, 89.8%). A total of 47,760 (25.7%) person-years occurred in the pre-HAART era, while 138,398 (74.3%) person-years occurred in the HAART era (1996–2002).
Cancer risk in persons registered with HIV infection
In the 5 years prior to the follow-up period, subjects had been diagnosed with 333 cancers. By definition, none of these prevalent cancers were AIDS-defining malignancies. The most common were cancers of the lung (N = 55), prostate (N = 38), and breast (N = 26), and Hodgkin lymphoma (N = 23).
Eight-hundred seventy-one incident cancers arose during the 5-year follow-up period after HIV registration (SIR 2.1, 95%CI 2.0–2.3, Table II). AIDS-defining cancers comprised 46.4% of all malignancies, while non-AIDS-defining cancers comprised 53.6%. The two most common malignancies were KS and NHL, and their incidence was greatly elevated compared with the general population (N = 173, SIR 1,300 and N = 203, SIR 7.3, respectively). Among NHL subtypes (Table II), DLBCL was most common (N = 93, SIR 9.6), but the elevation compared to the general population was especially high for CNS NHL (N = 28, SIR 250). Cervical cancer risk was significantly but less markedly elevated (N = 28, SIR 2.9).
Table II. Cancer Incidence and Standardized Incidence Ratios Among HIV-Infected Individuals in the Five-Year Period After HIV Registration
Incidence, per 100,000 person-years
SIR (95% CI)
SIR, standardized incidence ratio; CI, Confidence interval; NHL, non-Hodgkin lymphoma; DLBCL, diffuse large B celf lymphoma; CNS, central nervous system.
Standardized incidence ratios are presented to two significant digits.
All cancer types
Other histologicNHL subtypes
Colon and rectum
Bones and joints
Soft tissue including heart
Kidney and renal pelvis
Myeloid and monocytic leukemia
Poorly specified histology at any site
Among non-AIDS-defining cancers, elevated incidence was observed for lung cancer (N = 109, SIR 2.6), Hodgkin lymphoma (N = 36, SIR 5.6), and cancers of the oral cavity/pharynx (N = 26, SIR 1.7), anus (N = 18, SIR 9.2), larynx (N = 15, SIR 2.6), liver (N = 14, SIR 2.7), pancreas (N = 14, SIR 2.2) and penis (N = 3, SIR 5.4).
Cancer incidence in relation to AIDS and calendar year
Overall cancer incidence increased 3-fold after an AIDS diagnosis (RR 3.2, 95%CI 2.7–3.7, adjusted for sex, transmission category, age at HIV registration, and race/ethnicity; Table III). This was largely due to an increase in KS and overall NHL, with adjusted RRs (95%CIs) of 7.3 (5.4–9.9) and 4.3 (3.2–5.7), respectively, associated with prior AIDS diagnosis. Among NHLs, the increase in incidence following an AIDS diagnosis was stronger for the 3 AIDS-defining subtypes, especially CNS NHL, than for other histologic NHL subtypes (Table III). Cervical cancer incidence increased nonsignificantly after an AIDS diagnosis (unadjusted RR 2.2, 95%CI 0.9–5.5). Following an AIDS diagnosis, incidence also increased for Hodgkin lymphoma (adjusted RR 3.5, 95%CI 1.7–7.3) and lung cancer (adjusted RR 2.3, 1.4–3.5). Among the less common malignancies in Table III, only cancers of the oral cavity/pharynx increased significantly following an AIDS diagnosis. As shown in Table III, even before an AIDS diagnosis, risk was significantly elevated for the AIDS-defining cancers, especially KS (SIR 800) and CNS NHL (SIR 89). Risk was also elevated before AIDS diagnosis for some non-AIDS-defining cancers, notably cancers of the anus (SIR 8.1) and lung (2.3), and Hodgkin lymphoma (4.5).
Table III. Incidence and Relative Risks for Selected Cancers Among People Registered with HIV, According to AIDS Status and Calendar Year
AIDS, acquired immunodeficiency syndrome; RR, relative risk; CI, confidence interval; NHL, non-Hodgkin lymphoma; DLBCL, diffuse large B cell lymphoma; CNS, central nervous system.
Relative risks are presented to one decimal place. Standardized incidence ratios are presented to two significant digits.
Relative risks are adjusted for sex and HIV transmission category (men who have sex with men, other males, females), age at HIV registration, and race/ethnicity, except when the total number of cancer events was less than 20, in which case unadjusted relative risks are shown.—
Incidence estimates for “all cancer types” includes only first cancers.—
One-sided 95% confidence interval is (1.1-∞) and p-value is 0.03, based on an exact test.
A. Comparison of cancer incidence in relation to AIDS diagnosis
Overall cancer incidence declined slightly in the HAART era (adjusted RR 0.9, 95%CI 0.7–1.0, comparing 1996–2002 vs. 1991–1995; Table III). This reduction occurred because of substantial declines in KS and overall NHL (adjusted RR 0.4, 95%CI 0.3–0.6 and 0.7, 0.5–1.0, respectively). Among NHL subtypes, the decline in NHL incidence in 1996–2002 was strongest for CNS NHL and DLBCL (Table III). There was no evidence of a decline for Burkitt NHL, and the decline for other histologic NHL subtypes was not significant. Cervical cancer incidence did not change appreciably in 1996–2002 (Table III). In contrast, Hodgkin lymphoma incidence increased in 1996–2002 compared to 1991–1995 (adjusted RR 2.7, 95%CI 1.0–7.1). Lung cancer incidence appeared higher in 1996–2002 than in 1991–1995 (64 vs. 42 per 100,000 person-years), but this difference was not statistically significant (p = 0.08) and became markedly attenuated after adjustment for demographic factors (adjusted RR 1.1, 95%CI 0.7–1.7). Liver cancer incidence increased in 1996–2002 (p = 0.03).
Despite declines in the incidence of AIDS-defining cancers over time, the incidence remained substantially elevated during the HAART era, especially for KS (SIR 790), NHL (6.5), and among NHLs, particularly CNS NHL (170) (Table III). Overall, non-AIDS-defining cancers comprised the majority (58.0%) of all cancers in the HAART era (1996–2002), compared with only 31.4% of all cancers in the pre-HAART era (1991–1995). During the HAART era, risk was elevated for most non-AIDS-defining cancers in Table III, including cancers of the anus (SIR 9.1), liver (3.1), and lung (2.6), and Hodgkin lymphoma (6.7).
Temporal trends are shown in greater detail in Figure 1 for KS and NHL. For each cancer, there were very few cases in the first 2calendar years, in part because there were few subjects under follow-up (i.e., 9,010 person-years of follow-up in 1991–1992). We therefore do not present data for 1991–1992. KS incidence fell steeply over time for people who had AIDS, but the decline was less apparent for people without AIDS (Fig. 1a). A similar pattern was observed for overall NHL (Fig. 1b). For other cancers, there were too few cases to examine time trends in detail.
Cancer incidence in relation to CD4 count at HIV registration
Relationships between CD4 count at HIV registration and subsequent cancer incidence are presented in Table IV. Because CD4 counts were available for only a minority of subjects, sparse data limited analyses for most outcomes. Incidence increased significantly with declining CD4 count for overall cancer (p < 0.0001), largely due to trends present for KS (p < 0.0001), overall NHL (p = 0.01), and specifically for DLBCL (p = 0.03). Trends were nonsignificant for cervical cancer and the evaluated non-AIDS-defining cancers (Table IV).
Table IV. Incidence of Selected Cancers, According to CD4 Count at HIV Registration
NHL, non-Hodgkin lymphoma; DLBCL, diffuse large B cell lymphoma; CNS, central nervous system.
Standardized incidence ratios are presented to two significant digits. CD4 count is measured in cells/mm3.
p-values for trend are adjusted for sex and HIV transmission category (men who have sex with men, other males, females), age at HIV registration, and race/ethnicity, except when the total number of cancer events was less than 20, in which case unadjusted p-values are shown.—
Incidence estimates for “all cancer types” include only first cancers.
In Figure 2, the relationship with CD4 count at HIV registration is depicted for KS and overall NHL, separately for the 2 calendar periods. For KS, incidence was lower in 1996–2002 than 1991–1995, independent of the CD4 count at HIV registration (Fig. 2a). For overall NHL, incidence was lower in 1996–2002 than 1991–1995 only for individuals whose CD4 count at HIV registration was below 500 cells/mm3, whereas for individuals with higher CD4 counts, the incidence was similar in 1996–2002 and 1991–1995 (Fig. 2b). As a result, the association between overall NHL incidence and CD4 count at HIV registration was weak in 1996–2002 (Fig. 2b).
Our study documents substantial changes in the burden of cancer affecting HIV-infected individuals in the U.S. during a 12-year period spanning the introduction of HAART. During this time, the incidence of the 2 major AIDS-defining malignancies, KS and NHL, declined markedly. Nonetheless, the incidence of both cancers remained quite high during 1996–2002. Several non-AIDS-defining malignancies arose at elevated incidence even in people who had not yet developed AIDS, and Hodgkin lymphoma and liver cancer actually increased in incidence over time. As a result of these temporal changes, non-AIDS-defining malignancies have come to comprise the majority of cancers in HIV-infected persons during the HAART era.
Our results regarding KS and NHL strongly support the paradigm that these cancers arise in large part due to immunosuppression. Incidence of both malignancies increased following an AIDS diagnosis and with declining CD4 counts, and incidence decreased in temporal association with introduction of HAART in 1996. Others have also described declines in the incidence of these cancers with HAART use.1 Notably, the trends over time were strongest for people who had already developed AIDS (Figs. 1a and 1b), which we hypothesize could reflect patterns in the utilization of HAART. For example, by the end of 1996, approximately half of those with AIDS in the U.S. who were under medical care had started HAART, but the fraction was lower (∼15–25%) among those who had not progressed to AIDS.18, 26 During this time, physicians often waited to start HAART and preferentially provided antiretroviral therapies to the most immunocompromised HIV-infected patients, a practice still supported by current clinical guidelines.27, 28 Such patterns in HAART use would have blunted the time trends in KS and NHL risk among HIV-infected individuals who had not already developed AIDS. A similar explanation may pertain to the trends depicted in Figures 2a and 2b: during 1996–2002, clinicians frequently prescribed HAART to patients with low CD4 counts but delayed prescription for those with relatively preserved CD4 counts, which would attenuate the relationship between CD4 count and cancer risk.
HIV-infected individuals continue to be at increased risk for for NHL subtypes considered AIDS-defining (DLBCL, Burkitt NHL, CNS NHL). Of these, the associations with immunosuppression were most apparent for DLBCL, for which we saw an increase in incidence with declining CD4 count (Table IV). For Burkitt NHL incidence increased after an AIDS diagnosis but was unrelated to calendar year. There were too few subjects with CD4 data for us to examine trends in Burkitt NHL incidence related to this marker.
The heightened risks of cervical and anal cancers are consistent with the known high prevalence of coinfection with oncogenic HPV subtypes among HIV-infected women and men (especially men who have sex with men).29–31 HIV-induced immunosuppression may also play a role by impairing clearance of HPV and facilitating frequent progression to early-stage neoplastic lesions.31–33 However, the importance of HIV-induced immunosuppression in the later progression to HPV-related cancers has remained less certain.7, 34 Evidence supporting a role for immunosuppression derives from a recent meta-analysis reporting elevated risk among solid organ transplant recipients.35 We observed nonsignificant increases in the incidence of cervical and anal cancers following an AIDS diagnosis, but their incidence did not change over time with the introduction of HAART, nor based on limited data was it demonstrably related to CD4 count (Tables III and IV). The incidence of penile cancer, another HPV-related malignancy, was also elevated among HIV-infected individuals, but this cancer was too uncommon to evaluate further.
Our results for Hodgkin lymphoma highlight its complex relationship with immunosuppression and HAART. We found that Hodgkin lymphoma incidence increased substantially following an AIDS diagnosis but also, somewhat paradoxically, increased over time in association with the introduction of HAART (Table III). These observations confirm results from prior studies describing an increasing risk over time subsequent to an AIDS diagnosis,12 and an elevated risk associated with use of HAART.1, 36 We previously showed that, among people with AIDS, the relationship between Hodgkin lymphoma risk and CD4 count is nonlinear. Specifically, Hodgkin lymphoma incidence rises as the CD4 count falls to 225–249 cells/mm3 but then declines as the CD4 count falls further.37 Unfortunately, we had insufficient data on CD4 counts for individuals with Hodgkin lymphoma to extend these results in the present study. Taken together, findings from the various studies raise the possibility that partial immune reconstitution with HAART is causing an increase in Hodgkin lymphoma risk, by shifting HIV-infected individuals to an intermediate level of immunosuppression associated with highest risk.12, 37
Several malignancies related to use of tobacco (i.e., cancers of the lung, oral cavity/pharynx, larynx, and pancreas) or alcohol (i.e., cancers of the liver and oral cavity/pharynx) also occurred at elevated incidence in this cohort. Given that 60–80% of HIV-infected individuals in the U.S. smoke cigarettes, the high incidence of lung cancer is not surprising.38 Nonetheless, tobacco use alone appears insufficient to entirely explain the marked excess of lung cancer.38, 39 HIV-induced immunosuppression may be important, as there is some evidence for an increase in lung cancer risk with advancing time relative to AIDS diagnosis,38 and lung cancer risk is likewise elevated among transplant recipients.35 Other processes, such as repeated lung infections or chronic pulmonary inflammation, may also play a role. In the present study, we observed that liver cancer incidence rose over time, probably reflecting the effects of prolonged alcohol abuse or co-infection with hepatitis C or B viruses.6, 40
It is useful to review several strengths and limitations of our study. A major strength was its population-based design, including all people registered with HIV in the covered geographic regions. Although our study included data from only 3 states, the demographic characteristics of subjects indicate that all major subgroups of the U.S. epidemic were well-represented. In addition, our study was large enough to characterize patterns in cancer incidence for common cancers responsible for the greatest morbidity. Another strength is that, by including individuals beginning at HIV registration rather than at AIDS onset, we were able to evaluate HIV-infected individuals at an earlier stage of immunosuppression than possible in studies of people with AIDS. However, the availability of data on CD4 counts was more limited for our subjects than for people registered with AIDS, for whom a low CD4 count can be the AIDS-defining condition,4 and CD4 counts would have changed over time, especially with introduction of HAART. Also, the proportion with missing CD4 count varied somewhat across demographic groups (data not shown). Thus, the lack of complete CD4 count data reduced our ability to assess the associations with this marker and may have affected the generalizability of the results. Another limitation is that we had no individual data on actual HAART use. We used calendar year as a measure of the availability of HAART, but the trends in cancer risk that we observed also reflect patterns in healthcare access and utilization of HAART to an uncertain degree. Nonetheless, the calendar trends in cancer risk reflect the overall effects of HAART use at the population level. We had no data on the prevalence of other cancer risk factors (e.g., infections with oncogenic viruses, use of tobacco or alcohol) or cancer screening, which could have affected the observed incidence of cancer. Finally, although we truncated follow-up 60 months after HIV registration, it is still possible that migration out of the study regions during the follow-up period could have led us to somewhat underestimate cancer incidence.
In conclusion, since 1996 there have been substantial declines in KS and NHL among HIV-infected individuals in the U.S. Nonetheless, the risk for these AIDS-defining cancers remains markedly elevated compared with the general population. At the same time, we document a shift in the burden of cancers occurring in HIV-infected persons to non-AIDS-defining malignancies. Efforts to better understand the etiology of these cancers and prevent their occurrence are warranted. As HIV-infected individuals live longer and age with improved antiretroviral therapies, continued monitoring may reveal further changes in cancer incidence over time and prompt new strategies for cancer screening and early intervention.
The authors thank the staff at the HIV/AIDS and cancer registries for collecting and maintaining these valuable data resources.