Association of Cancer With AIDS-Related Immunosuppression in Adults FREE

Immunosuppression associated with human immunodeficiency virus (HIV) infection and the acquired immunodeficiency syndrome (AIDS) markedly increases the risk of Kaposi sarcoma (KS) and non-Hodgkin lymphoma (NHL).¹ These malignancies and invasive cervical cancer are the only AIDS-defining cancers in HIV-infected individuals.² Other specific cancers also occur in excess, but the risk pattern depends on the geographic region and the HIV exposure group studied.^3- 4 In industrialized parts of the world, a high level of risk has been found for anal cancer⁵ and Hodgkin disease,^6- 8 and marginally significant increases were reported for seminoma, multiple myeloma, and brain cancer.^7- 8 We used data from the recently updated AIDS-Cancer Match Registry Study to investigate the cancer pattern in the years surrounding the time of AIDS diagnosis among 302 834 adult persons with HIV/AIDS (PWAs) in the United States. Specifically, we aimed at identifying cancers that are likely to be influenced by immunosuppression, distinguishing them from those occurring in excess among PWAs due to lifestyle-related exposures linked to cancer risk independently of immunosuppression.

We linked cancer data to 366 034 PWAs from 11 areas, including the states of Connecticut, Florida, Illinois, Massachusetts, New Jersey, and New York, and the metropolitan areas of Atlanta, Los Angeles, San Diego, San Francisco, and Seattle. Commercially available software (AutoMatch versions 3.0 and 4.1; MatchWare Technologies Inc, Burtonsville, Md) was used to implement a probabilistic matching algorithm that evaluated the likelihood of subject linkage on the basis of identical or near-identical information in AIDS and cancer registries. Before data linkage, we obtained institutional review board approval at all participating AIDS and cancer registries to ensure patient confidentiality. Details of the linkage procedure have been described elsewhere.⁹

This study describes the cancer profile of 302 834 adults (age 15-69 years) diagnosed with AIDS between 1978 and 1996. We excluded 56 669 persons whose AIDS diagnoses occurred outside the period when corresponding cancer registries were considered complete, 4954 children aged 0 to 14 years at AIDS diagnosis (dealt with in detail elsewhere¹⁰), and 1577 persons aged 70 years or older at AIDS diagnosis. The period covered varied by registry, but all regions included the 12-year period from January 1983 through December 1994, except San Diego, where cancer registration started in January 1988. For 2.3% of the study subjects, we backdated the onset of AIDS to the date of an AIDS-defining cancer (KS, n = 5155; NHL, n = 1731; invasive cervical cancer, n = 182) recorded in the cancer registry up to 5 years before the date of onset recorded in the AIDS registry. In examining the non–AIDS-defining cancers, we studied only invasive cancers and used the International Classification of Diseases for Oncology, second edition¹¹ to define cancer categories. Analyses were made for KS, NHL, and for specific sites (excluding KS and NHL). We studied all cancer types, except non-melanoma skin cancers. The latter were disregarded, because non-melanoma skin cancers are incompletely recorded in cancer registries and may include inadequately recorded cases of KS, the single most common cancer among PWAs.

Using methods similar to those of a previous study,⁷ which assessed cancer risks in persons diagnosed with AIDS between 1978 and 1990, we studied the relative risk (RR) of cancer from 60 months before to 27 months after AIDS onset in the following intervals: from 60 to 25 months before AIDS (distant pre-AIDS period, during which most of the PWAs were likely to be HIV-infected but unlikely to have developed severe immunosuppression), from 24 to 7 months before AIDS (recent pre-AIDS period, during which most of the PWAs approached the time of opportunistic infections and cancers that define AIDS), from 6 months before to 3 months after AIDS (AIDS period), and from 4 to 27 months after AIDS (early post-AIDS period, which was truncated at 2 years after the AIDS period to avoid an overestimation of the number of PWAs under surveillance). To extract the maximum information available for rare cancers (ie, those with <25 cases), we also studied cancers occurring from 28 to 60 months after AIDS (late post-AIDS period, which was used only in the evaluation of pre- to post-AIDS trends for rare cancers). For each interval, the RR was the ratio of observed to expected cancers derived from contemporaneous population-based incidence rates. By definition, all PWAs who had cancer in the 5 years prior to AIDS survived long enough after their cancer diagnosis to develop AIDS. Failure to take this survival factor into account has a major impact on the RR estimates.⁸ Consequently, estimates of the expected cancers, to which observed cases were compared, had to take into account that some individuals developed cancer and died or were lost to follow-up before they developed AIDS. To do so, we calculated survival-conditioned cancer incidence rates covering each month back to 5 years before AIDS, as described elsewhere.⁹ Because reliable cancer survival data are not available for HIV-infected persons, we used cancer survival data from the Surveillance, Epidemiology, and End Results (SEER) program for the period 1981-1993 to adjust the pre-AIDS incidence rates. The resulting survival-conditioned cancer incidence rates were calculated for the 11 regions combined, in strata of sex, race (white, black, other/unknown), and age (5-year age groups). Stratum-specific monthly survival-conditioned incidence rates were multiplied by the appropriate person-months of observation among the PWAs and the products were summed to yield the expected number of pre-AIDS cancers. For the post-AIDS periods, we constructed 1 combined set of sex-, race-, age-, and period-specific incidence rates for the 11 regions under study. Expected numbers of cancers were calculated as the sum of stratum-specific products of background cancer incidence and person-months at risk among the PWAs.¹²

To provide an overall RR estimate, we used the sum of observed cancers over the sum of expected cancers for the 7.25-year period from 60 months before to 27 months after AIDS. We did not include cancers from the late post-AIDS period in this overall RR measure because the number of PWAs under observation in this interval may have been overestimated (and RRs underestimated) due to unrecorded late deaths and migration out of the catchment areas. For the AIDS-defining cancers, we only provide an estimate of the RR in the early post-AIDS period since, by definition, none of these cancers occurred before AIDS. We calculated 95% confidence intervals (CIs) assuming a Poisson distribution of the observed cancers.¹² When zero cancers occurred, 1-sided exact 95% CIs are provided with the upper confidence limit calculated as [ − ln(0.05)/expected number of cancers]. The RRs were calculated by sex, ethnic group (white, black, Hispanic, other/unknown), and HIV exposure group (homosexual contact [including homosexual intravenous drug users, men only]; heterosexual contact; non-homosexual intravenous drug users; patients infected during treatment for hemophilia or other bleeding disorders, or via transfusion or transplantation; and other/unknown HIV exposure), according to AIDS registry information. Sex-, ethnic group–, and age-specific background rates were used to calculate expecteds in all analyses. Since background cancer rates were not available for Hispanics, expected cancers in this group were based on cancer rates for whites.

Malignancies may occur in excess among PWAs for reasons unrelated to immunosuppression such as smoking and exposure to sexually transmitted human papillomavirus. To identify cancers that are likely to be influenced by the waning immune competence during the course of HIV infection and AIDS, we expanded previously defined criteria⁷ to also include the demand for a statistically significantly elevated overall RR. Specifically, a cancer was considered potentially influenced by immunosuppression when all of 3 criteria were met: (1) the overall RR for the period from 60 months before to 27 months after AIDS was significantly elevated; (2) the RR in the early post-AIDS period was significantly elevated; and (3) there was a statistically significant increasing trend in the RRs from before to after AIDS onset. To evaluate the trend, we modeled the RR in a Poisson regression model as:
ln(O_i/E_i) = α + βt_i, where O_i and E_i
were the observed and expected numbers of cancers, respectively, in the individual periods included in the trend test, and α was the sum of observed divided by the sum of expected cancers for periods contributing information to the trend test (with t_i, the midpoint of the time interval, measured from AIDS onset). In this model, a 2-sided score test¹³ of β = 0 evaluated whether RR changed over time. We attributed the observation of rather extreme RRs in the AIDS period for cancers at most sites to the generally increased diagnostic activity around the time of AIDS diagnosis (ascertainment bias). To avoid the influence of such ascertainment bias, we disregarded cancers occurring in the AIDS period in the trend test. Consequently, we used information from the distant and recent pre-AIDS periods and the early post-AIDS period in trend tests, using midpoints for these intervals (t = − 42.5, − 15.5, and 15.5 months, respectively) as values of t_i. However, for cancers with fewer than 25 observed cases, we also included data from the late post-AIDS period (t = 44 months) in the trend test, to use the maximum information available. Since underascertainment of deaths and migration out of the catchment areas renders the person-time at risk in the late post-AIDS period too high, and thus the corresponding RR too low, the trend test for these less common cancers is likely to be conservative.

We performed 2 sensitivity analyses to assess the robustness of our findings. In one, we based the trend test only on cancers occurring in the 2 pre-AIDS periods and the AIDS period (t values of − 42.5, − 15.5, and − 1.5 months, respectively). This test is likely to be biased toward finding too many positive trends because of high RRs in the AIDS period due to ascertainment bias, as described above. In the second, we split the AIDS-relative time axis into 3 new pre-AIDS periods (60 to 41 months, 40 to 21 months, and 20 to 7 months before AIDS), the AIDS period (6 months before to 3 months after AIDS), and 2 new post-AIDS periods (4 to 24 months and 25 to 60 months after AIDS), using midpoints for the pre-AIDS periods (t = − 50.5, − 30.5, and − 13.5 months) and the early post-AIDS period (t = 14 months). When numbers of cancers were small (<25), we also used the late post-AIDS period (t = 42.5 months) in trend tests.

Demographic characteristics and route of HIV acquisition for the 302 834 adult PWAs are shown in Table 1. Nearly half were between 30 and 39 years old at AIDS onset and most were younger than 50 years. Whites were the largest group among men (47.8%), whereas blacks predominated among women (57.2%). Homosexual men (including homosexual intravenous drug users) constituted 62.8% of the men, and an additional 26.4% were non-homosexual intravenous drug users. Only 3.1% of the men were HIV-infected through heterosexual contact. Among women, 49.6% were infected through intravenous drug use, and 35.3% were infected heterosexually (Table 1).

Among men, KS was an AIDS-defining cancer or occurred within 27 months after another AIDS-defining event in 20 848 persons (9.9%), while NHL occurred in 8622 persons (3.4%). Among women, corresponding numbers were 307 KS cases (0.8%), 728 NHL cases (1.5%), and 348 invasive cervical cancer cases (0.7%). Among men, the post-AIDS RR for KS (n = 5823; RR,175.8) was highest among homosexual PWAs (n = 5410; RR, 267.2) while, among women, the post-AIDS RR for KS (n = 113; RR, 405.2) was highest among those who were HIV-infected by a bisexual man (n = 11; RR, 2044.2). The post-AIDS RR for NHL was 72.8, with RRs of 71.7 among men (n = 3073) and 88.4 among women (n = 271). The post-AIDS RR for cervical cancer was 5.2 (n = 46) (Table 2).

Overall, 4422 invasive cancers other than AIDS-defining cancers and non-melanoma skin cancers occurred vs 1611.3 expected (RR, 2.7; 95% CI, 2.7-2.8). The overall excess of non–AIDS-defining cancers was seen in both men (n = 3757; RR, 2.8; 95% CI, 2.7-2.9) and women (n = 665; RR, 2.4; 95% CI, 2.2-2.6). Significant overall excesses were seen for cancers at several sites (Table 2). Of the more than 60 individual cancer types studied, all 3 criteria for potential association with immunosuppression were met only for 6 cancers: Hodgkin disease, lip cancer, lung cancer, soft tissue malignancies, penile cancer, and testicular seminoma (Table 2, Figure 1). Other/unspecified lymphatic/hematopoietic cancers, unspecified types of cancer of brain and central nervous system, and cancers at unknown or ill-defined primary sites also met all 3 criteria. However, these latter groups of unspecified cancers likely comprise considerable proportions of the much more common AIDS-defining cancers, notably inaccurately classified cases of NHL, including brain lymphoma. It should be noted that for data assessed by trend test (Figure 1), the generally higher frequency of cancers occurring in the AIDS period is likely, in part, to be an artifact of ascertainment bias. Because it is impossible to know the extent of possible bias, cancer data from the AIDS period were not included in the trend test.

Hodgkin Disease. The overall RR for Hodgkin disease (n = 612) was 11.5-fold increased, being higher in men (n = 558; RR, 12.0) than in women (n = 54; RR, 8.3) (Table 3). However, for each sex, differences were small among the major HIV exposure categories (men: homosexual contact [n = 386; RR, 12.5]; heterosexual contact [n = 14; RR, 11.0]; intravenous drug use [n = 121; RR, 10.7]; women: heterosexual contact [n = 17; RR, 7.0]; intravenous drug use [n = 25; RR, 7.7]). For men and women combined, the RR increased from 2.6 in the distant pre-AIDS period to 9.8 in the recent pre-AIDS period and 6.7 in the early post-AIDS period (P for trend, <.001) (Figure 1). Significant trends were seen in separate analyses for men (P<.001) and women (P = .04), and for subgroups of homosexual men (P<.001) and intravenous drug–using men (P = .004).

The expected distribution of histologic subtypes of Hodgkin disease was nodular sclerosis (64%), mixed cellularity (26%), lymphocytic predominance (8%), and lymphocytic depletion (2%). For PWAs with histologically specified Hodgkin disease (n = 403), the excess was much greater for mixed cellularity (54%; overall RR,18.3) and lymphocytic depletion (9%; overall RR, 35.3) than for nodular sclerosis (34%; overall RR, 4.7) or lymphocytic predominance (3%; overall RR, 3.0) (Table 2). In the distant pre-AIDS period, RRs for the mixed cellularity (n = 25; RR, 5.2) and lymphocytic depletion (n = 2; RR, 4.8) subtypes were already increased, but not for nodular sclerosis (n = 12; RR, 0.9) or lymphocytic predominance (0 observed vs 1.6 expected) subtypes. Even in the late post-AIDS period, the RR remained significantly high for the mixed cellularity subtype (n = 7; RR, 7.7; 95% CI, 3.1-15.9).

Lung Cancer. Lung cancer (n = 808) was the most frequently diagnosed non–AIDS-defining cancer. The RRs increased from 1.2 in the distant pre-AIDS period to 2.7 in the recent pre-AIDS period and 2.8 in the early post-AIDS period (P for trend<.001) (Figure 1). The crude lung cancer incidence in the early post-AIDS period was slightly higher in male than female PWAs (80.1 vs 68.0 per 100 000), but the overall RR was higher in women (RR, 7.1) than men (RR, 4.3) (Table 3). Although many women developed lung cancer (n = 110), the pre- to post-AIDS trend was significant only in men and in subgroups of whites (P = .002) and homosexuals (P<.001). All 3 statistical criteria for potential association with immunosuppression were met for squamous cell carcinoma, adenocarcinoma, and other and unspecified types of lung cancer (Table 2).

Penile Cancer. Anogenital cancers of the cervix, anus, vulva/vagina, and penis occurred in significant overall excess, but the trend test was significant only for penile cancer (n = 14) (Table 2, Figure 1). Blacks, Hispanics, and intravenous drug users were at particularly elevated risk (Table 3).

Soft Tissue Malignancies. Seventy-eight malignant tumors originated in soft tissues (overall RR, 3.3) (Table 2). Twenty-five (32%) were histologically unspecified, 14 (18%) were soft tissue sarcomas, and 12 (15%) were leiomyosarcomas. Including 6 leiomyosarcomas at sites other than soft tissues, the overall RR for leiomyosarcoma was 2.3 (Table 2). The excess risk for leiomyosarcoma was statistically significant only among the 15- to 29-year-old PWAs (n = 7; overall RR, 17.3; 95% CI, 6.9-35.7). Angiosarcomas occurred in approximately 13-fold excess, but the trend test was not significant (Table 2).

Lip Cancer. The overall RR for lip cancer was 3.1 (Table 2). All 20 cases of lip cancer occurred in males, but only 0.14 cases were expected in women. For men and women combined, RRs increased steadily from 1.6 in the distant pre-AIDS period to 5.8 in the late post-AIDS period (P for trend = .04).

Testicular Seminoma. Of 167 testicular cancers, 115 (69%) were seminomas. The RR for testicular seminoma was 0.7 in the distant pre-AIDS period, 2.0 in the recent pre-AIDS period, and 1.8 in the early post-AIDS period (P for trend = .003). The majority of seminomas (78%) occurred in homosexual men (RR, 2.2) (Table 3). A high risk was seen in hemophilia patients (RR, 14.5; 95% CI, 3.9-37.1), but numbers were small (n = 4).

Other cancers occurred in excess overall or in the early post-AIDS period (Table 2), including leukemia, multiple myeloma, and cancers of the oral cavity, salivary glands, pharynx, esophagus, stomach, anus, liver and intrahepatic bile ducts, pancreas, larynx, heart, mediastinum and pleura, vulva/vagina, kidney, and specified types of brain and central nervous system cancers, as well as melanoma, testicular non-seminoma, and conjunctival carcinoma (n = 7; overall RR, 14.0; 95% CI, 5.6-28.8). None of these met the criterion of increasing trend in RR in relation to AIDS onset.

The pattern for breast cancer was unusual. Overall, the number of breast cancers was close to the expected (overall RR, 1.1). However, among women (n = 134; overall RR, 1.0; 95% CI, 0.9-1.2) the pre- to post-AIDS trend decreased significantly (P for trend = .01), and the RR in the early post-AIDS period was significantly low (n = 14; RR, 0.4). In contrast, among men, there was a slight excess of breast cancer (n = 9; overall RR, 1.8; 95% CI, 0.8-3.3), particularly among intravenous drug users (n = 5; overall RR, 3.9; 95% CI, 1.3-9.2).

To assess robustness of our findings, we performed 2 alternative trend tests. The first, likely to be biased toward finding too many positive trends, was based on cancers occurring only in the 2 pre-AIDS periods and the AIDS period. For the majority of the more than 60 different cancer categories studied, trend results were reassuringly similar. In this alternative trend test, the only differences were that lip cancer lost formal statistical significance (P for trend = .19), while trends for anal squamous cell carcinoma (P for trend = .01) and tongue cancer (P for trend = .007) achieved statistical significance. In the second sensitivity analysis, we divided the time window around AIDS into 6 new intervals. Trend results were also essentially unchanged in this analysis. No new sites emerged as potentially influenced by immunosuppression, but the trend test for penile cancer, significant in the original trend analysis, now failed to reach statistical significance (P for trend = .08).

PWAs in the United States are at considerably increased cancer risk compared with the general population. However, of all cancers observed in the period from 60 months before to 27 months after AIDS onset, most (87%) were those already considered as AIDS-defining cancers, notably KS and NHL.²

Our RR estimates for KS (RR, 177.7) and NHL (RR, 72.8) in the early post-AIDS period are lower than estimates from the early HIV epidemic,¹⁴ partly because we used contemporaneous incidence rates (which include KS and NHL cases among PWAs) to calculate expected numbers. Also, the risk composition of the HIV/AIDS population has changed. The highest RRs for KS were seen in homosexual men and among women who were HIV-infected through sexual contact with a bisexual man, probably reflecting the concurrent epidemics of HIV and human herpesvirus 8, the viral cause of KS, among homosexual/bisexual men.¹⁵ The use of highly active antiretroviral therapy¹⁶ cannot have affected our findings for KS and NHL substantially, as the coverage period in our study ended in 1996.

The PWAs had an excess of a number of cancers without established links to immunosuppression. Some of these cancers have been linked etiologically to HIV-unrelated exposures that are common among PWAs, including cancers of lung and lip, which are considered smoking-related,^17- 18 and penile cancer, which may be related to infection with human papillomavirus.¹⁹

It is difficult to determine if, and to what extent, moderate excesses of non–AIDS-defining cancers are associated with immunosuppression. We used 3 statistical criteria to assist in this process, but none of the criteria, alone or in combination, can determine whether immunosuppression truly plays a causal role. Our findings must be evaluated in the context of other contributing or conflicting evidence. Causality criteria include strength of association, dose-response gradient, specificity, consistency with previous studies, and analogies to other states of immunosuppression.^20- 21 Our overall RR estimates measured the strength of the association between HIV/AIDS-related immunosuppression and cancer risk. The RR in the early post-AIDS period and the trend test served as indicators of a possible dose-response gradient, assuming parallelism between AIDS-relative time and level of immunosuppression. By demanding statistical significance for each of these 3 measures, the list of cancers studied was reduced to a handful deserving further attention.

The non–AIDS-defining cancer most likely to be genuinely associated with declining immunity is Hodgkin disease. This type of lymphoma was 11.5-fold elevated overall and met our 3 criteria for potential association with immunosuppression in all major strata of sex, race, and HIV exposure. Moreover, our findings are consistent with several smaller studies.^{1,6- 8,22- 25} As summarized previously,¹ the most frequent Hodgkin disease subtype among PWAs was mixed cellularity. Epstein-Barr virus DNA is detectable in high proportions of HIV/AIDS- and transplantation-associated Hodgkin disease.^26- 27 AIDS-associated NHLs are lso more often Epstein-Barr virus-positive than non–AIDS-associated NHLs,²⁸ and most cases of both Hodgkin disease and NHL appear to develop from transformed B-lymphocytes.²⁹ We hypothesize that the Epstein-Barr virus–associated subsets of Hodgkin disease may be associated with immune dysfunction through mechanisms similar to those underlying the established association with NHL.

Our findings for lip cancer are also characterized by consistency with other studies, dose-response relationship with level of immunity (as measured by AIDS-relative time), and analogies to other immunosuppressed states. Our observation of a steady increase in the RR in relation to onset of AIDS adds weight to the 5-fold risk observed by Grulich et al.⁸ The lip cancer excess was not restricted to any 1 HIV risk group in men, and the lack of an association in female PWAs could well be due to limited statistical power. Lip cancer risk has been found to be increased in organ transplant recipients^30- 31 and patients treated with immunosuppressive drugs.³² Squamous cell carcinoma of the skin might be similarly influenced by immunosuppression,^33- 34 but reliable non-melanoma skin cancer data are lacking in large-scale prospective studies and in registry studies such as ours.

The role of immunosuppression in testicular cancer remains unclear. Testicular seminoma met all our criteria for potential association with immunosuppression, and the lack of a similar pattern for non-seminoma testicular cancers adds specificity to the association. However, the strength of the association (overall RR, 2.0) does not suggest a major impact of declining immunity on seminoma risk. Corroborative evidence from the literature is sparse. Previous studies either failed to find an increased risk of testicular seminoma^24- 25,35 or did not have sufficient power to identify increased risks as statistically significant,^6,8 except by inclusion of non-testicular seminomas.⁷ We note that the excess of seminomas was seen predominantly in whites and homosexual men. An infectious etiology has been suggested for testicular seminoma,^36- 39 but the putative agent remains elusive.

All our statistical criteria were also met for soft tissue malignancies, lung cancer, and penile cancer, but these cancer excesses are unlikely to be related to immunosuppression. A significant excess of soft tissue malignancies was seen only in homosexual men, the group at highest risk for KS. This, combined with the large proportion of histologically unspecified soft tissue malignancies, suggests a major contribution by inaccurately coded cases of KS. None of the specified types of sarcoma met our trend criterion. The risk for leiomyosarcoma is markedly increased in children with AIDS,^10,40 but in this study of adult PWAs, the risk was only 2.3-fold increased and the pre- to post-AIDS trend was not statistically significant. Of note, however, the risk for leiomyosarcoma was 17.3-fold elevated among the 15- to 29-year-old PWAs, thus possibly representing a continuation of the even greater leiomyosarcoma risk in 0- to 14-year-old children with AIDS.¹⁰

It is to be expected that PWAs, who smoke more than the general population,⁴¹ have an excess of lung cancers.¹⁷ Small case series have suggested a particular excess of lung adenocarcinomas.^42- 43 In accordance with some^6,8 but not all studies,^22,25,35 we observed an excess lung cancer risk in PWAs. However, this excess was neither restricted to nor particularly pronounced for adenocarcinomas. Although we observed a statistically significant pre- to post-AIDS trend for lung cancer, this trend was only significant in men (despite large numbers of lung cancer in women). The inconspicuous RR in the distant pre-AIDS period (RR, 1.2) may be too low and, consequently, the significant trend for lung cancer may be an artifact. Lung cancers in PWAs carry an especially poor prognosis, with few patients surviving more than a few months.^42- 43 The lung cancer survival estimates we obtained from SEER data and used to derive expected cancers in the pre-AIDS period might have been too optimistic when applied to an HIV-infected population.

As reported previously, cervical cancer^6- 7 and anal cancer⁵ occurred frequently, and vulvar/vaginal and penile cancers also were in significant excess. Of these, only penile cancer met our trend criterion, but our sensitivity analyses suggested that this was not a robust finding. Human papillomavirus–associated malignancies in PWAs are considered in detail elsewhere.⁹

Breast cancer was the only cancer to exhibit a statistically significant pattern of decreasing RRs with AIDS-relative time. This finding is compatible with the suggestion⁴⁴ that physiological immune activity may somehow facilitate breast carcinogenesis and, consequently, that immunosuppression might reduce breast cancer risk.⁴⁵ However, the overall risk for breast cancer was not reduced. It appears that female PWAs have a normal, but redistributed, breast cancer incidence, perhaps resulting from heightened medical attention around the time of AIDS diagnosis. If HIV-mediated immunosuppression were truly protective, we would have expected an overall reduction in breast cancer risk. Although rare, male breast cancer risk was not reduced but, indeed, almost 4-fold increased in non-homosexual intravenous drug–using men. Further detracting from the idea that immunosuppression protects against breast cancer,⁴⁵ a study of 5692 renal transplant patients in Nordic countries found that breast cancer incidence was not reduced.³⁰

Our study confirms prior findings of elevated risks for multiple myeloma, anal cancer, brain cancer, and conjunctival carcinoma in PWAs in the United States.^5,7 Grulich et al⁸ also reported an elevated risk of multiple myeloma in a separate linkage analysis. These cancer excesses do not seem associated with HIV-mediated immunosuppression. With our current, substantially larger database, these cancers exhibited no consistent post-AIDS excess, and none showed statistically significant pre- to post-AIDS trends.

This is the first population-based study with sufficient power to examine the general cancer risk among PWAs by sex, race, and HIV risk group. We had considerable statistical power to detect even moderate increases in the risk for the major cancers. However, despite these advantages, certain limitations need consideration. As discussed for lung cancer, we believe that RRs in the distant pre-AIDS period, particularly for cancers with poor survival rates, may be underestimated due to the use of too optimistic population-based survival rates. Also, if cancer treatment in HIV-infected individuals leads to hastened progression from HIV infection to clinical AIDS, this acceleration would tend to move observed cancers from the distant pre-AIDS period to the recent pre-AIDS period and thus produce too low RRs in the distant pre-AIDS period. Such artificial lowering of the starting point for the trend test would increase the likelihood of detecting spurious increasing trends. Reassuringly, however, this potential bias seems not to have had a major impact on our results, since only a handful of cancers met our criteria for potential association with immunosuppression in the original analysis and in the 2 sensitivity analyses.

In summary, PWAs are at increased risk of cancers other than KS, NHL, and cervical cancer. For most cancers, excesses are probably attributable to lifestyle factors among PWAs or to inaccurate recording in cancer registries of the more common AIDS-defining cancers. However, 3 specific types of cancer, notably Hodgkin disease, but possibly also lip cancer and testicular seminoma, may be genuinely influenced by immunosuppression. Our finding of clearly increased risks of Hodgkin disease in all groups of PWAs, and a statistically significant pre- to post-AIDS trend, show that the risk of Hodgkin disease increases with advancing immunosuppression and suggest it be considered as an AIDS-defining condition.