AIDS-Related Cancer and Severity of Immunosuppression in Persons With AIDS

Transcription

1 ARTICLE AIDS-Related Cancer and Severity of Immunosuppression in Persons With AIDS Robert J. Biggar, Anil K. Chaturvedi, James J. Goedert, Eric A. Engels For the HIV/AIDS Cancer Match Study Background The incidence of Kaposi sarcoma, non-hodgkin lymphoma, and cervical cancer has been declining among persons with AIDS. We investigated the association between cancer risk and CD4 cell count among such persons. Methods Data from US AIDS registries were linked to local cancer registry data. Cancer incidence per 1 person-years was determined for the 4 27 months from the onset of AIDS from January 1, 199, through December 31, 1995 before highly active antiretroviral therapy (HAART) became available and from January 1, 1996, through December 31, 22. The relationships between CD4 count at AIDS onset and cancer incidence were assessed by proportional hazards models. Results Conclusions Among adults with AIDS, the incidence of Kaposi sarcoma was lower in (334.6 cases per 1 person-years) than in ( cases per 1 person-years), and the incidence of non-hodgkin lymphoma followed a similar pattern (i.e., 39.1 cases per 1 person-years in and cases per 1 person-years in ). In , for each decline in CD4 cell count of 5 cells per microliter of blood, increased risks were found for Kaposi sarcoma (hazard ratio [HR] = 1.4, 95% confidence interval [CI] = 1.33 to 1.5), for central nervous system non-hodgkin lymphoma subtypes (HR = 1.85, 95% CI = 1.58 to 2.16), and for non central nervous system diffuse large B-cell lymphoma (HR = 1.12, 95% CI = 1.4 to 1.2) but not for non central nervous system Burkitt lymphoma (HR =.93, 95% CI =.81 to 1.6). Cervical cancer incidence was higher in (86.5 per 1 person-years) than in (64.2 per 1 person-years), although not statistically significantly so (relative risk [RR] = 1.41, 95% CI =.81 to 2.46). After adjustment for age, race, and sex or mode of HIV exposure, the risks for Kaposi sarcoma (RR =.22, 95% CI =.2 to.24) and for non-hodgkin lymphoma (RR =.4, 95% CI =.36 to.44) were lower in the period of than in Similar relationships of these cancers to CD4 count were observed for Both before and after HAART was available, CD4 count was strongly associated with risks for Kaposi sarcoma and non-hodgkin lymphoma but not for cervical cancer and Burkitt lymphoma. The decreasing incidences of most AIDS-associated cancers in persons with AIDS during the 199s are consistent with improving CD4 counts after HAART introduction in J Natl Cancer Inst 27;99: Downloaded from at Pennsylvania State University on March 3, 214 Persons with AIDS are at high risk of three cancers: Kaposi sarcoma, non-hodgkin lymphoma, and cervical cancer ( 1 ). Indeed, these cancers are accepted as AIDS-defining cancers. Recent studies ( 2 5 ) have reported marked reductions in the risk of these cancers in persons with AIDS that were associated with the use of highly active antiretroviral therapy (HAART), a combination of drugs that effectively controls HIV replication. This therapy usually results in marked improvement in CD4 counts and consequent reductions in the risks of infectious diseases ( 6 9 ), even if normal immunity is not fully restored ( 1,11 ). However, data about the level of immunosuppression in persons with AIDS and cancer incidences in these persons are limited. In a previous analysis ( 12 ), we examined cancer risk by CD4 count for persons with AIDS diagnosed through December 31, 1995, before HAART was available. Information on this association in the HAART era has not yet been presented. Furthermore, estimates for less common cancers such as Burkitt lymphoma or Affiliations of authors: Viral Epidemiology Branch, Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institutes of Health, Department of Health and Human Services, Bethesda, MD (RJB, AKC, JJG, EAE); Department of Epidemiology Research, State Serum Institute, Copenhagen, Denmark (RJB). Correspondence to: Robert J. Biggar, MD, Viral Epidemiology Branch, Division of Cancer Epidemiology and Genetics, National Cancer Institute, 612 Executive Blvd, Room EPS 814, Bethesda, MD 2852 ( biggarb@mail.nih.gov ). See Notes following References. DOI: 1.193/jnci/djm1 The Author 27. Published by Oxford University Press. All rights reserved. For Permissions, please journals.permissions@oxfordjournals.org. 962 Articles JNCI Vol. 99, Issue 12 June 2, 27

2 cervical cancer given in the earlier study were unstable because of small number of cases. In a recent analysis ( 13 ), we reported that Hodgkin lymphoma incidence in persons with AIDS was statistically significantly lower in individuals with CD4 counts of less than 1 cells per microliter of blood than in individuals with higher CD4 counts, an unexpected finding that illustrates the importance of large studies to determine associations with confidence. In this study, we examined the relationship between CD4 count and the three AIDS-defining cancers Kaposi sarcoma, cervical cancer, and non-hodgkin lymphoma, including the subtypes of non-hodgkin lymphoma by use of a large dataset updated to include the HAART era ( ). Participants and Methods HIV/AIDS Cancer Match Study The HIV/AIDS Cancer Match Study links data from persons with HIV/AIDS and cancer registry data in many areas of the United States. This analysis used registry data from January 1, 22, through December 31, 25, in the states of Colorado, Michigan, New Jersey, Massachusetts, Connecticut, Georgia, and Florida and the metropolitan areas of San Francisco, Los Angeles, San Diego, Seattle, and New York City. The linkages between records in the HIV/AIDS Cancer Match Study and cancer registries were based on a probabilistic algorithm that included social security number, first and last name, sex, race, and birth and death dates obtained from the National Death Index ( 14 ). The linked records were reviewed by legally authorized personnel at the local registries, and any questionably linked records were rejected. The locally linked file was cleared of all personal identifying information and sent to the National Cancer Institute for analysis. These procedures, which have been presented in detail elsewhere ( 4, ), were approved by the ethical and legal review boards at participating registries. We evaluated cancer incidence in adults ( 15 years old) with AIDS diagnosed from January 1, 199, through December 31, 22, who had CD4 counts available for their AIDS-onset period (i.e., from 6 months before to 3 months after their AIDS diagnosis). The cancer incidence evaluation period was the 2-year period from 4 through 27 months after AIDS diagnosis. Persons were followed from the fourth month after AIDS onset to the occurrence of an event (defined as a diagnosis of Kaposi sarcoma, any non-hodgkin lymphoma, or cervical cancer), death, the date at which cancer registration was considered to be incomplete, or 28. months after AIDS onset, whichever came first. The first 3 months after AIDS diagnosis were not evaluated for incidence to avoid inclusion of cancers that defined AIDS onset or that were detected during medical examinations at AIDS diagnosis. Data of cancers diagnosed after 27 months were excluded to minimize the impact of unknown follow-up losses, such as from migration outside the registry area or unlinked deaths. Persons who had developed the same cancer before entering the incidence evaluation period were excluded. Persons were censored at the first occurrence of any type of non-hodgkin lymphoma. AIDS-Related Cancers The AIDS-related cancers are usually defined as Kaposi sar coma, non-hodgkin lymphoma, and cervical cancer. However, non- CONTEXT AND CAVEATS Prior knowledge The incidence of Kaposi sarcoma, non-hodgkin lymphoma, and cervical cancer has been declining among persons with AIDS since the introduction of highly active antiretroviral therapy (HAART) in Study design Data from local cancer registries were linked to those from US AIDS registries to investigate the association between CD4 cell count at AIDS diagnosis and cancer incidence in periods before and after the introduction of HAART. Contribution After HAART became available, the incidence of both Kaposi sarcoma and non-hodgkin lymphoma decreased, but that of cervical cancer increased. CD4 count at AIDS onset was strongly associated with risks for Kaposi sarcoma and for non-hodgkin lymphoma but not for cervical cancer. Implications The decreased incidences of Kaposi sarcomas and non-hodgkin lymphomas are consistent with improving CD4 counts in persons with AIDS since the introduction of HAART. Limitations CD4 counts and cancer incidence and risk were assessed at different times. CD4 counts were measured at AIDS onset, but cancer risk was determined 4 27 months after AIDS onset, when CD4 counts would have been higher because of antiretroviral therapy. Hodgkin lymphoma includes histology types that are not AIDS related, as well as a large group of non-hodgkin lymphomas not otherwise specified by histology. Furthermore, all central nervous system non-hodgkin lymphomas are considered to be AIDS related regardless of histology. Because of these complexities, most reports of the AIDS non-hodgkin lymphoma include all non- Hodgkin lymphomas in persons with AIDS as AIDS non-hodgkin lymphoma, regardless of histology or site. However, our goal was to evaluate the relationship between immunity and non-hodgkin lymphoma risk by subtype. As AIDS-related cancers, therefore, we included all central nervous system non-hodgkin lymphomas and, at non central nervous system sites, diffuse large B-cell lymphoma (DLBCL), Burkitt lymphoma, and non-hodgkin lymphoma, not otherwise specified. The DLCBL subtype includes immunoblastic non-hodgkin lymphoma. Outside the central nervous system, we included non-hodgkin lymphomas, not otherwise specified, as AIDS related because 171 (89.4%) of 1192 patients with non central nervous system non-hodgkin lymphoma with known histology were AIDS-related subtypes. Overall, in this study from 199 through 22, 248 (95.2%) of the 2529 non-hodgkin lymphoma cases in persons with AIDS were thus considered to be AIDS related (i.e., they were of histology types considered to be AIDS related, including non-hodgkin lymphoma, not otherwise specified, or at central nervous system sites). Non AIDS-associated non-hodgkin lymphomas included all other non-hodgkin lymphomas with a specified histology. We used the International Classification of Diseases for Oncology, 3rd Edition ( 17 ) (ICD-O-3) to code site and histology Downloaded from at Pennsylvania State University on March 3, 214 jnci.oxfordjournals.org JNCI Articles 963

3 and updated records that used older coding classifications to conform to ICDO-3. Specifically, Kaposi sarcoma included histology code 914 regardless of site, non-hodgkin lymphomas included all lymphomas (codes 959 through 9729), except for Hodgkin lymphoma (codes ), and cervical cancer included all cancers with site code C53 and behavior code 3 (invasive cancer). Histology was often not available for central nervous system non-hodgkin lymphoma (site code C7 C72), and for non-hodgkin lymphoma at this site, all types were included regardless of histology. Of 231 central nervous system non-hodgkin lymphomas with known histology, 25 (89%) were DLBCL and six (3%) were Burkitt lymphoma. At non central nervous system sites, AIDS-associated non-hodgkin lymphomas included three groups: Burkitt lymphoma (code 9687), DLBCL (codes 968 and 9684), and non- Hodgkin lymphoma, not otherwise specified; non AIDS-associated non-hodgkin lymphoma included all specified B-cell non-hodgkin lymphomas that were not considered to be AIDS associated (codes , , and 9823 other than the AIDS-associated histology types listed above) and T-cell (codes and 9782) non-hodgkin lymphomas. Statistical Analysis For incidence evaluation, persons with AIDS were grouped by strata of the latest CD4 count in the AIDS-onset period by increments of 5 cells per microliter ( 49 to CD4 cells per microliter). Persons with AIDS with CD4 counts of 5 cells per microliter or higher (i.e., <1% in persons with AIDS) were excluded from all analyses. We also examined risk by excluding persons with CD4 counts more than 3 cells per microliter because we had relatively few persons with such high counts. HAART has been available since 1996, and its widespread use has led to higher CD4 counts than use of therapies available in ( 6 9 ). Persons with AIDS were, therefore, stratified into one of two groups by the year in which their AIDS was diagnosed (i.e., in or in ). As a test of robustness of the 1995/1996 cut point, we compared data for with those for in a sensitivity analysis; excluding the transition years (AIDS onset in ) did not change the conclusions (data not presented). Therefore, we present data for a comparison of cancer incidence in with those in , after adjustment for age, race, and sex/mode of HIV exposure. To examine the association of CD4 count with cancer incidence, we used Cox proportional hazard models to calculate hazard ratio (HR) per incremental decrease of 5 CD4 cells per microliter, after adjustment for age (15 29, 3 39, 4 49, or 5 years), race (white, black, Hispanic, or other/unknown), and sex/mode of HIV exposure (men who have sex with men, other males, or female s). We included an interaction term between calendar period and CD4 count to allow the relationship between CD4 count and cancer risk to differ between the periods of and To determine whether the association of CD4 count at AIDS onset with subsequent cancer risk changed with increasing time after the baseline measurement, we tested the validity of the proportional hazard assumption by examining trends in hazard ratios within succes sive 6-month intervals. This analysis considered four successive 6-month intervals of follow-up time and, for each interval, estimated a separate hazard ratio associated with the CD4 count at AIDS onset. The proportional hazards assumption was then tested by evaluating the statistical significance in the Cox model of a 1-df interaction term between the CD4 count at AIDS onset and the interval of follow-up time (treated as an ordinal value). For many subjects, CD4 counts had been recorded before the onset of cancer in periods outside of their AIDS-onset period. Likewise, many cancers were diagnosed during the AIDS period or after the end of the incidence evaluation period for AIDS. These data were not included in the incidence evaluation, described above, for the following reasons: 1) Cancer in the AIDS-onset period precluded the analysis of subsequent cancer incidence because persons could not get the same cancer twice. 2) The longer time intervals between AIDS onset and cancers that occurred late in the course of AIDS made the CD4 count at AIDS onset increasingly irrelevant. 3) Finally, recording of CD4 counts outside of the AIDS-onset period was not systematic and was available for only a minority of persons with AIDS. Nevertheless, the CD4 data before cancer onset could be analyzed descriptively by examining median CD4 counts before cancer onset. To assure that the CD4 count was obtained before the diagnosis of cancer and not affected by cancer therapy, we used only CD4 results that were obtained 12 months before cancer onset. When multiple CD4 counts were recorded, we used the last recorded count in that period. Cancer onset was stratified by time from AIDS onset, to allow for changes in CD4 counts during the course of AIDS (i.e., 3, 4 14, 15 27, 28 59, or 6 months from AIDS onset). To assure that median CD4 counts were not influenced by changes in the availability of effective antiretroviral therapy, we restricted our analysis to cancers that were diagnosed in the HAART era, which began in Thus, we started this analysis in 1997 so that the 12 months before any cancer diagnosis were entirely within the HAART era. For trend analyses, we used only data from patients with cancer diagnosed at least 15 months after the onset of AIDS to ensure that all CD4 counts were obtained beyond the AIDS-onset period. For CD4 counts, median values were compared by the Wilcoxon rank-sum test, and trends in median values in cancer patients were evaluated by the nonparametric Jonckheere Terpstra test ( 18 ). All statistical tests used were two-sided. Differences were considered to be statistically significant at a P value of less than.5. Results The study included adults with AIDS diagnosed from January 1, 199, through December 31, 22. The distribution of subjects by sex, age-group, race, mode of HIV exposure, and CD4 count is presented by calendar-time period in Table 1. The proportions of women, older persons, nonwhites, and persons exposed to HIV heterosexually were higher in than in CD4 counts at AIDS onset were available for a higher proportion of those diagnosed between 1996 and 22 (85.8%) than between 199 and 1995 (63.2%). However, among those with known CD4 counts, distributions of proportions within demographic and mode of HIV exposure categories were similar in and ( Table 1 ). Also, the distributions of observed CD4 counts within different demographic and mode of exposure groups were similar (data not shown). In each period, Downloaded from at Pennsylvania State University on March 3, Articles JNCI Vol. 99, Issue 12 June 2, 27

4 Table 1. Demographic characteristics of persons with AIDS in the United States who had CD4 counts in the period from 199 through 22 Characteristic most individuals completed the full 24-month follow-up period, and there was only a modest difference in the mean overall followup duration between those with a CD4 count of 49 cells per microliter at AIDS onset and those with a CD4 count of cells per microliter (i.e., 17 and 21 months, respectively). The number of patients with Kaposi sarcoma, non-hodgkin lymphoma, and cervical cancer that occurred during the AIDS-onset period ( 3 months) is also listed in Table 1. Of all 2529 cases of non- Hodgkin lymphomas ( Table 2 ), 121 (4.8%) had non central nervous system specific histology types that were not ac cepted as AIDS associated (i.e., 71 B-cell and 5 T-cell non-hodgkin lymphomas), and these types were not included in the analysis. AIDS-Associated Cancer During the 4- to 27-month period after AIDS onset, 3753 incident cases of Kaposi sarcoma, 2529 incident cases of non-hodgkin lymphomas, and 55 incident cases of cervical cancers were reported ( Table 2 ). Among non central nervous system non-hodgkin lymphomas, 58 (39.3%) of the 1292 cases diagnosed in and 142 (33.1%) of the 429 cases diagnosed in were classified No. (% of column total in category) Total Overall (1.) (1.) (1.) Sex Male (81.4) 9424 (74.4) (78.6) Female 378 (18.6) (25.6) (21.4) Age at AIDS onset, y (15.5) (12.2) (14.2) (45.3) (4.9) (43.6) (28.) (32.) (29.6) (11.2) (14.9) 4193 (12.6) Race or ethnicity White 7969 (4.) (28.8) (35.7) Black (37.7) (47.3) (41.4) Hispanic 4216 (21.2) 2831 (22.4) 7416 (21.6) Other 2266 (1.1) 1955 (1.5) 4221 (1.3) Mode of HIV exposure * MSM 8849 (44.3) (33.5) (4.1) IDU 6524 (3.4) (22.1) (27.2) IDU + MSM 1151 (5.6) 4819 (3.8) 1587 (4.9) Heterosexual (1.9) (19.2) (14.1) Other/unknown (8.8) (21.4) 4473 (13.7) CD4 counts, cells per microliter Missing (36.8) (14.2) (28.) Available (63.2) (85.8) (72.) (32.7) 3314 (3.5) (31.7) (16.5) (15.8) (16.2) (15.9) (16.4) (16.1) (21.4) (22.2) 5956 (21.7) (13.6) (15.1) (14.3) Cancer type in AIDS onset period Kaposi sarcoma (5.9) 282 (2.2) (4.5) Non-Hodgkin lymphoma 4475 (2.3) 2433 (1.9) 698 (2.1) Cervical cancer 419 (.2) 35 (.2) 724 (.2) * MSM = men who have sex with men; IDU = intravenous drug users. Percent distribution of CD4 cell number was based on persons with known CD4 counts only. Cancers recorded in either the AIDS registry or the cancer registry files. These individuals were excluded from the analyses of incidence during the 4 27 months after AIDS onset. as non-hodgkin lymphoma, not otherwise specified. Among all 248 AIDS-associated non-hodgkin lymphomas, 687 (28.5%) were classified as central nervous system non-hodgkin lymphoma and 1721 (71.5%) were classified as non central nervous system non-hodgkin lymphomas (including 949 DLBCLs, 122 Burkitt lymphomas, and 65 non-hodgkin lymphoma, not otherwise specified). Among the adult persons with AIDS, the incidence of Kaposi sarcoma was lower in (334.6 cases per 1 person-years) than in ( cases per 1 person-years), and non-hodgkin lymphoma overall followed a similar pattern (i.e., 39.1 cases per 1 person-years in and cases per 1 person-years in ). After adjustment for age, race, and sex or mode of HIV exposure, the risks for Kaposi sarcoma (relative risk [RR] =.22, 95% confidence interval [CI] =.2 to.24) and for non-hodgkin lymphoma (RR =.4, 95% CI =.36 to.44) were lower in the period of than in The incidence of cervical cancer was higher in the later period (86.5 versus 64.2 per 1 person-years, respectively), although not statistically significantly so (RR = 1.41, 95% CI =.81 to 2.46) ( Table 2 ). The incidences Downloaded from at Pennsylvania State University on March 3, 214 jnci.oxfordjournals.org JNCI Articles 965

5 Table 2. Incidence of cancer during the period from 4 through 27 months after AIDS onset and relationship to a CD4 decline of 5 cells per microliter at AIDS onset * Tumor and period No. of cases Person-years of central nervous system non-hodgkin lymphoma were 77.4 and per 1 person-years, respectively, in the later and earlier periods, and those of non central nervous system non- Hodgkin lymphoma were and 71.1 per 1 personyears, respectively. The risk of central nervous system non-hodgkin lymphoma (RR =.27, 95% CI =.21 to.33) declined statistically significantly more between these two periods than the risk for non central nervous system non-hodgkin lymphoma (RR =.44, 95% CI =.39 to.49). At non central nervous system sites in both and , many non-hodgkin lymphomas were of unspecified histology types (with incidences of 94. and per 1 person-years, respectively). Among the specified non central nervous system non-hodgkin lymphomas, the incidences of DLBCL were and 39.7 per 1 person-years, respectively, in the two periods, whereas that of Burkitt lymphoma was 32.4 and 4.1 per 1 person-years, respectively. The risk for DLBCL (RR =.45, 95% CI =.38 to.52) declined more between these two periods than the risk of Burkitt lymphoma (RR =.92, 95% Incidence per 1 person-years RR (95% CI) in , compared with HR (95% CI) per 5 CD4 cell count decline Kaposi sarcoma (referent) 1.28 (1.24 to 1.31) (.2 to.24) 1.4 (1.33 to 1.5) Overall NHL (referent) 1.27 (1.23 to 1.31) (.36 to.44) 1.16 (1.11 to 1.22) NHL, AIDS-associated (referent) 1.28 (1.24 to 1.32) (.35 to.43) 1.18 (1.12 to 1.24) CNS (all types) (referent) 1.76 (1.64 to 1.88) (.21 to.33) 1.85 (1.58 to 2.16) Non-CNS (referent) 1.15 (1.11 to 1.19) (.39 to.49) 1.9 (1.4 to 1.15) DLBCL (referent) 1.15 (1.1 to 1.2) (.38 to.52) 1.12 (1.4 to 1.2) Burkitt lymphoma (referent).93 (.83 to 1.4) (.64 to 1.33).93 (.81 to 1.6) NOS (referent) 1.2 (1.13 to 1.26) (.3 to.44) 1.13 (1.3 to 1.24) NHL, non AIDS-associated (referent) 1.13 (.99 to 1.29) (.45 to.95).94 (.81 to 1.9) Cervical cancer (referent).88 (.72 to 1.8) (.81 to 2.46) 1.11 (.91 to 1.36) * Data were limited to person with fewer than 5 CD4 cells per microliter. HR = hazard ratio; RR = relative risk; CI = confidence interval; NHL = non-hodgkin lymphoma; CNS = central nervous system; DLBCL = diffuse large B-cell lymphoma, including immunoblastic lymphoma; NOS = not otherwise specified. Relative risk (95% confidence interval) in compared with values in ; adjusted for age, race, and sex/mode of HIV exposure. Results are from Cox proportional hazards regression models that were adjusted for age, race, and sex or mode of HIV exposure. Person-years for subtypes of NHL remain unchanged because only the first NHL was evaluated, regardless of whether another NHL was reported later. CI =.64 to 1.33), for which the risk was not statistically significantly different in the two periods. AIDS-Associated Cancers and CD4 Count at AIDS Onset The incidences of Kaposi sarcoma and of non-hodgkin lymphoma overall was strongly associated with declining CD4 counts in both calendar periods ( Fig. 1 ). As shown in Table 2, for the AIDS-associated cancers (i.e., Kaposi sarcoma, AIDS-associated non-hodgkin lymphoma, and cervical cancer), the trends of in - creasing risk with declining CD4 count were similar in magnitude in both periods (for each decline in CD4 cell count of 5 cells per microliter of blood, in , HR for Kaposi sarcoma = 1.28, 95% CI = 1.24 to 1.31; and, in , HR = 1.4, 95% CI = 1.33 to 1.5). The non-hodgkin lymphoma pattern was dominated by AIDS-associated types, but, within this group of AIDS-associated non-hodgkin lymphomas, the relationship between CD4 count and risk varied substantially by site (for example, in for central nervous system non-hodgkin lymphoma, HR for each decline of 5 CD4 cells per microliter = 1.85, Downloaded from at Pennsylvania State University on March 3, Articles JNCI Vol. 99, Issue 12 June 2, 27

6 A 3 Kaposi sarcoma B 3 Overall NHL C 2 Cervical cancer D 3 AIDS NHL E 1 CNS NHL F 1 Non-CNS AIDS NHL % CI = 1.58 to 2.16; for non central nervous system non- Hodgkin lymphomas, HR = 1.9, 95% CI = 1.4 to 1.15). No association between CD4 count and cervical cancer incidence was observed in either period. In contrast, the relationship between CD4 count and incidence varied by type among non central nervous system non-hodgkin lymphomas ( Table 2 and Fig. 2 ). Because the proportional hazards assumption was violated (in that the association of CD4 counts at AIDS onset and subsequent cancer risk varied statistically sig - nificantly during the quartiles of follow-up), the hazard ratios presented in Table 2 should be considered average effects for the 2-year incidence evaluation period. In , the incidences of both DLBCLs and non- Hodgkin lymphoma, not otherwise specified, were associated with CD4 count (for DLBCL, HR for each decline of 5 CD4 cells per microliter = 1.12, 95% CI= 1.4 to 1.2; and for non- Hodgkin lymphoma, not otherwise specified, HR = 1.13, 95% CI = 1.3 to 1.24), whereas the incidence of Burkitt lymphoma was not associated with CD4 count ( Table 2 ). Central nervous system non-hodgkin lymphoma represented a smaller proportion of all AIDS-associated lymphomas in (117 of the 546 AIDSassociated lymphomas or 21.4%) than in (57 of the 1862 AIDS-associated lymphomas or 3.6%) (exact odds ratio =.62, 95% CI =.49 to.78). Among the non central nervous system AIDS-associated non-hodgkin lymphomas with specified histology, Burkitt lymphoma represented a larger proportion of all AIDS-associated lymphomas in (49 of the Fig. 1. Incidence of major AIDS-associated cancers by in persons with AIDS, by calendar-year period = solid circles ; = open squares. A) Kaposi sarcoma. B) Overall non-hodgkin lymphoma. C) Cervical cancer. D) AIDS-associated non-hodgkin lymphoma. E) Central nervous system non-hodgkin lymphoma. F) Non central nervous system non-hodgkin lymphoma. Note that vertical scales vary between cancer types. NHL = non-hodgkin lymphoma; CNS = central nervous system AIDS-associated lymphomas or 17.1%) than in (73 of the 784 AIDS-associated lymphomas or 9.3%) (exact odds ratio = 1.62, 95% CI = 1.28 to 2.5). The incidences of non-aids B-cell and T-cell non-hodgkin lymphoma types were not statistically significantly related to CD4 count. As a test of robustness, when we further restricted our analyses to persons with AIDS who had a CD4 count of less than 3 cells per microliter (because few cancers occurred in CD4 strata above this level), the conclusions were not appreciably different (data not presented). We also evaluated the relationship between CD4 count at AIDS onset and incidence in each successive 6-month interval within the incidence evaluation period of 4 27 months after the AIDS-onset period. Except for Burkitt lymphoma, the incidence of which was not related to CD4 count ( Table 3 ), associations between CD4 count and incidence in were strongest in the first 6-month period after AIDS was diagnosed and weakened statistically significantly thereafter. This finding was also present for Kaposi sarcoma diagnosed in the period but not for non-hodgkin lymphoma diagnosed in this period. For persons with a CD4 count available within 1 year before cancer diagnosis, the median CD4 count at AIDS onset for Burkitt lymphoma was higher than that for DLBCL (13 CD4 cells per microliter for Burkitt lymphoma versus 11 for DLBCL in , Wilcoxon P =.2; 134 CD4 cells per microliter for Burkitt lymphoma versus 1.5 for DLBCL in , Wilcoxon P =.2). The median CD4 counts in the period from through 12 months before cancer onset were Downloaded from at Pennsylvania State University on March 3, 214 jnci.oxfordjournals.org JNCI Articles 967

7 A 1 Diffuse large B-cell B 1 NHL NOS C 2 Burkitt lymphoma D 2 Non-AIDS NHL E 2 Non-AIDS B-cell NHL F 2 Non-AIDS T-cell NHL higher when CD4 counts included those obtained at AIDS onset than when they did not (i.e., cancers arising 15 months or more after the AIDS onset) ( Fig. 3 ). Even so, both at and long after AIDS onset, the relationship between median CD4 count and cancer type remained almost unchanged, with patients with central nervous system non-hodgkin lymphomas having the lowest median CD4 counts and those with cervical cancer or Burkitt lymphoma having the highest median CD4 counts. For the periods of cancer diagnosis for which the CD4 count excluded counts at AIDS onset (i.e., cancers arising 15 27, 28 6, and >6 months after AIDS onset), the CD4 counts were essentially flat (trends not statistically significantly different from each other) for all cancer types examined. Discussion Our results illustrate the variable impact of immunity on the risk of AIDS-associated cancers within the range of CD4 counts in persons with AIDS. In both and , CD4 counts at AIDS onset were strongly associated with the risk of both Kaposi sarcoma and non-hodgkin lymphoma, and the associations were of similar magnitude, although incidence was much lower in than in However, the incidence of cervical cancer was not associated with CD4 count in either period. Within the non-hodgkin lymphoma subgroups, the risk of Fig. 2. Incidence of AIDS-associated and non AIDS-associated non- Hodgkin lymphoma (NHL) outside the central nervous system, by CD4 count in persons with AIDS, by calendar-year period = solid circles ; = open squares. A) Diffuse large B-cell lymphoma (includes immunoblastic lymphoma). B) Non-Hodgkin lymphoma, not otherwise specified. C) Burkitt lymphoma. D) Non AIDS-associated non-hodgkin lymphoma. E) Non AIDS-associated B-cell non-hodgkin lymphoma. F) Non AIDS-associated T-cell non- Hodgkin lymphoma. Note that vertical scales vary between cancer types. NOS = not otherwise specified. central nervous system non-hodgkin lymphoma increased strongly with lower CD4 count, whereas non central nervous system non- Hodgkin lymphoma risk increased moderately, even when the analysis was restricted to those subtypes of non-hodgkin lymphomas that are accepted as being AIDS associated. Risk of DLBCL increased with CD4 count decline, but risks of Burkitt lymphoma and of the non AIDS-associated non-hodgkin lymphomas were not related to CD4 count. CD4 counts at AIDS onset were similar in the two calendar periods, reflecting the fact that persons with AIDS are immunosuppressed. However, among cancers associated with reduced CD4 counts, for any CD4 count at AIDS onset, the risk was lower in than in The lower risk in than in likely reflected the higher CD4 count after AIDS onset in patients treated with HAART. For this study, we could not measure the change in the CD4 count in patients after the onset of AIDS, but greater increases in the CD4 counts after the introduction of HAART than that of other antiretroviral therapies have been well documented ( 6 9 ). The risk of some AIDSassociated cancers was not associated with the CD4 count, and the incidences of these cancers were, therefore, not altered by HAART. The variable impact of HAART on cancer incidence changed the proportions of tumors being diagnosed over time. We found that, in AIDS patients diagnosed during the HAART era, the incidence of Kaposi sarcoma declined more than that of Downloaded from at Pennsylvania State University on March 3, Articles JNCI Vol. 99, Issue 12 June 2, 27

8 Table 3. Associations between cancer risk and CD4 count at AIDS onset for 6-month intervals within the incidence evaluation period between AIDS onset and cancer onset * HR (95% CI) Tumor and period 4 9 mo 1 15 mo mo mo P value Kaposi sarcoma (1.4 to 1.53) 1.34 (1.28 to 1.4) 1.21 (1.16 to 1.27) 1.12 (1.6 to 1.17) < (1.48 to 1.81) 1.21 (1.9 to 1.35) 1.28 (1.12 to 1.46) 1.19 (1.4 to 1.37) <.1 Overall NHL (1.27 to 1.42) 1.3 (1.23 to 1.38) 1.28 (1.2 to 1.35) 1.21 (1.14 to 1.28) (1.7 to 1.24) 1.11 (1.2 to 1.22) 1.2 (1.7 to 1.35) 1.9 (.96 to 1.24).79 AIDS NHL (1.28 to 1.44) 1.31 (1.23 to 1.38) 1.28 (1.21 to 1.36) 1.21 (1.14 to 1.28) (1.1 to 1.29) 1.13 (1.3 to 1.23) 1.22 (1.8 to 1.39) 1.11 (.98 to 1.27).649 CNS NHL (1.95 to 2.83) 1.99 (1.72 to 2.31) 1.64 (1.45 to 1.85) 1.51 (1.34 to 1.7) < (1.48 to 2.5) 1.76 (1.29 to 2.39) 2.18 (1.46 to 3.25) 1.34 (.97 to 1.83).229 Non-CNS NHL (1.15 to 1.3) 1.16 (1.8 to 1.23) 1.15 (18 to 1.24) 1.1 (1.3 to 1.18) (1.1 to 1.19) 1.5 (.95 to 1.16) 1.9 (.96 to 1.25) 1.6 (.92 to 1.23).765 DLBCL (1.13 to 1.34) 1.14 (1.5 to 1.24) 1.16 (1.6 to 1.28) 1.1 (1. to 1.2) (.97 to 1.22) 1.11 (.97 to 1.28) 1.1 (.93 to 1.31) 1.13 (.93 to 1.36).785 NHL NOS (1.9 to 1.32) 1.25 (1.12 to 1.39) 1.22 (1.8 to 1.37) 1.15 (1.3 to 1.29) (1.6 to 1.47) 1.3 (.88 to 1.19) 1.9 (.86 to 1.38).99 (.75 to 1.31).128 * Hazard ratios were based on Cox proportional hazards models that were adjusted for age, race, and sex/mode of HIV exposure. HR = hazard ratio; CI = confidence interval; NHL = non-hodgkin lymphoma; CNS = central nervous system; DLBCL = diffuse large B-cell lymphoma; NOS = not otherwise specified. From a 1-df interaction term (CD4 count at AIDS onset follow-up interval) included in the Cox proportional hazards model. All statistical tests of significance were two-sided. non-hodgkin lymphoma, so that the incidence of non-hodgkin lymphoma became more common than that of Kaposi sarcoma among persons with AIDS. Among the non-hodgkin lymphomas, the incidence of central nervous system non-hodgkin lymphoma declined most sharply from the pre-haart era and to the HAART era. The incidence of non central nervous system non- Hodgkin lymphoma continued to be dominated by DLBCL; however, because the incidence of Burkitt lymphoma did not change substantially over time, the proportion of non central nervous system non-hodgkin lymphoma cases that were classified as Burkitt lymphoma increased. Similarly, another study ( 19 ) found that the incidences of central nervous system and non central nervous system lymphomas were sharply lower in AIDS patients diagnosed during the HAART era than those diagnosed earlier. In that study, the incidence was also highest in those with the lowest CD4 counts, with the CD4 count being more important for the incidence of central nervous system non-hodgkin lymphoma than for the incidence of non central nervous system non-hodgkin lymphoma. For cancers whose incidence was associated with CD4 count, the increase in risk with each decline in CD4 count of 5 cells per microliter appeared modest only because we used narrow CD4 strata. When translated across the CD4 count range usually observed at AIDS onset, they are more impressive. For example, for persons with AIDS diagnosed in , those with a CD4 count of cells per microliter had a fourfold higher Downloaded from at Pennsylvania State University on March 3, 214 Fig. 3. Median CD4 counts in the period from through 12 months before the onset of selected cancers in Data were stratified by cancer onset in months after AIDS onset. Median CD4 cell counts are presented as the number of CD4 cells per microliter. DLBCL = diffuse large B-cell lymphoma (includes immunoblastic lymphoma); NHL = non-hodgkin lymphoma; CNS = central nervous system. Asterisk indicates that trend analyses used the Jonckheere Terpstra test (18) and were limited to time period of 15 months or more after AIDS onset to exclude CD4 counts that were collected during the period from through 3 months after AIDS onset. All statistical tests were two-sided. Median CD4 cell count/ul I Trend evaluation period* Month of cancer onset after AIDS onset I Burkitt lymphoma, P =.71 Cervical cancer, P =.74 DLBCL, P =.21 AIDS NHL, P =.38 Kaposi sarcoma, P =.61 CNS NHL, P =.6 jnci.oxfordjournals.org JNCI Articles 969

9 incidence of central nervous system non-hodgkin lymphoma than those who had a CD4 count of less than 5 cells per microliter and twice the incidence of Kaposi sarcoma and non central nervous system DLCBL. In contrast, the incidences of cervical cancer and Burkitt lymphoma were both unchanged as the CD4 count declined. The CD4 count near the time of cancer onset appeared to be most important for the cancers associated with the CD4 count. In , the associations with CD4 count were strongest in the time intervals closest to AIDS onset ( Table 3 ). The likely explanation for this attenuation of hazard ratios is that CD4 counts change over time, whereas we considered associations with cancer incidence that was based only on CD4 count measured at AIDS onset. This attenuation of hazard ratios over time was not observed for non-hodgkin lymphoma in Patterns of the median CD4 counts in the period of 12 months before cancer onset were similar to those for cancer type and CD4 count both at and long after AIDS onset, indicating that the relationships between CD4 count and cancer risk could be generalized to all periods. Finally, after excluding the CD4 counts that might have been measured in the AIDS-onset period (i.e., excluding all counts with 15 months of AIDS onset), the median CD4 count for each cancer remained fairly constant regardless of the time from AIDS onset, indicating that cancer risk had not been dominated by CD4 levels for a substantial time before cancer onset. With regard to specific cancers, the risk of Kaposi sarcoma was strongly and inversely associated with CD4 count, indicating a role for immunity in preventing the expression of this tumor. The etiologic agent underlying Kaposi sarcoma development is human herpesvirus 8, a virus that is closely related to Epstein Barr virus (EBV) ( 2 ). Perhaps, in immunosuppressed persons, human herpesvirus 8 transformed spindle cells escape surveillance and progress to Kaposi sarcoma. Consistent with this potential mode of pathogenesis, localized Kaposi sarcoma tumors can regress when immunity is improved e.g., by HAART use in persons with AIDS ( 21 ) or by removal of immunosuppressive regimens in organ-transplant patients being treated with such regimens to suppress rejection ( 22 ). The relationships of CD4 count with Burkitt lymphoma and DLCBL indicate that the reasons for their increased incidences in persons with AIDS differ by non-hodgkin lymphoma type. Although the incidence of Burkitt lymphoma is greatly increased in persons with AIDS ( 5, 15, 16 ), Burkitt lymphoma incidence was not related to a decline in the CD4 count. In persons with AIDS, Burkitt lymphoma is not strongly associated with EBV, with the proportion of EBV-positive patients with Burkitt lymphoma (3%) being only slightly higher than the proportion of sporadic cases of Burkitt lymphoma in persons without AIDS (15% 2%) ( ). Thus, the incidence of Burkitt lymphoma in persons with AIDS is not driven by lack of EBV control in such persons. We speculate that, for Burkitt lymphoma, antigenic stimulation, certainly from HIV but perhaps also from other antigens, may be driving lymphocyte division to amplify lymphocyte clones that produce relevant antibodies. Errors occurring during mitosis could result in chromosomal rearrangements that place immunoglobulin genes upstream of the MYC gene and thereby provide a factor that increases the risk of Burkitt lymphoma. Because both AIDS and non-aids Burkitt lymphomas have MYC gene translocations ( ), a similar mechanism may contribute to Burkitt lymphoma occurrence in non-aids settings, where lymphocyte proliferation may be driven by antigens other than HIV antigens. For example, the antigenic stimulation of malaria and other infections may drive lymphocyte replication, which increases the likelihood of an error that places the MYC gene upstream of the immunoglobulin gene, which could explain the increased risk of Burkitt lymphoma in areas of Africa in which malaria is endemic ( 25 ). The incidence of non central nervous system DLBCLs in persons with AIDS increased steadily with higher CD4 levels. EBV is found in most patients with AIDS-associated DLBCLs ( 23, 24 ), indicating that EBV may play an important role in the genesis of DLBCL. Perhaps, EBV-activated B cells are not recognized or not eliminated in persons with low CD4 counts or with severely dysfunctional immune systems. However, not all DLBCL tumors have integrated EBV. Thus, although EBV may be a powerful oncogenic agent, other viruses, known or unknown, may have the potential to malignantly transform lymphocytes, especially under conditions of immunosuppression. Notably, human herpesvirus 8, which is strongly associated with primary effusion lymphoma ( 22, 23 ), and possibly a subset of related noneffusion DLBCLs ( 26 ), may be another example of a transforming virus whose action is enhanced by immunosuppression. The incidence of central nervous system non-hodgkin lymphoma increased strongly with decreases in the CD4 count. However, non central nervous system immunosuppression should not directly impact events within the central nervous system. Thus, events external to the central nervous system presumably give rise to central nervous system non-hodgkin lymphoma. Most central nervous system non-hodgkin lymphomas are DLBCLs, and virtually all are positive for EBV ( 23, 24 ). Possibly, the lack of immune defenses in the central nervous system allows EBVtransformed lymphocytes to emerge as central nervous system lymphoma cells before appearing at non central nervous system sites. If so, events occurring peripheral to the central nervous system, such as high EBV levels in peripheral blood, may predict the development of this tumor. Women with AIDS clearly have an increased risk of cervical cancer, but whether this risk is related to immunosuppression or simply reflects a higher risk of acquiring human papillomavirus in the same population is unclear. An etiologic relationship between cervical cancer and HIV-related immunosuppression is supported by the observation that women with AIDS have a higher risk of developing low- and high-grade dysplasia ( 27, 28 ), but the lack of an association between the time relative to AIDS onset and the risk of cervical cancer ( 29 ) argues against a causal relationship. The potential impact of screening for cervical cancer complicates interpretation of these findings; however, among high-risk African women who were not screened for cervical cancer, no increase was observed in the incidence of cervical cancer associated with the HIV/AIDS epidemic ( 3 ). In this study, we found no relationship between incidence of cervical cancer and immunosuppression. Immunodeficiency could promote early-stage cervical dysplasia through poorly controlled human papillomavirus infection, but events that permit subsequent development of malignancy appear Downloaded from at Pennsylvania State University on March 3, Articles JNCI Vol. 99, Issue 12 June 2, 27

10 not to be related to immunity, at least in the time scale needed for the progression to cervical cancer. This study has several limitations. The most important limitation is that we measured CD4 counts at AIDS onset but assessed cancer risk 4 27 months after AIDS onset, when CD4 counts would have been higher because of antiretroviral therapy. The lower incidence of cancer in the HAART era (from 1996 through 22) is consistent with higher CD4 counts in the period of 4 27 months after AIDS onset than in the period from 199 through Despite this limitation, we observed strong associations between CD4 count and cancer incidence even in the HAART era. The descriptive data on median CD4 counts 12 months before cancer onset indicate that the same relationships were present whenever the cancer occurred during the course of AIDS. Another limitation is that we lacked individual data about antiretroviral therapy use and, therefore, used calendar period of HAART availability. The sharp declines in cancer incidence indicate that most people were on effective therapy but to the extent that some subjects in the HAART era were not on HAART, we would have underestimated the risk relationships. Finally, because persons with AIDS usually have low CD4 counts at AIDS onset, we cannot provide definitive information about cancer risk in persons with only moderate immunosuppression. In summary, we found that the relationships between CD4 count and the incidence of AIDS-associated cancers varied by cancer type. The incidences of central nervous system non-hodgkin lymphoma and Kaposi sarcoma were the most strongly associated with severe CD4 depletion, whereas those of Burkitt lymphoma and cervical cancer were not associated with CD4 counts, at least in the ranges observed in persons with AIDS. These associations persisted in the HAART era, although the risk of Kaposi sarcoma and most non-hodgkin lymphomas has declined substantially. With the availability of HAART, CD4 counts have improved greatly ( 6 9 ), and the higher CD4 counts have probably resulted in a lowered incidence of cancers associated with the CD4 count, especially central nervous system non-hodgkin lymphoma and Kaposi sarcoma. In contrast, the incidences of Burkitt lymphoma and cervical cancer, which were not associated with CD4 counts, did not change greatly between the two periods studied, despite HAART availability, thereby increasing their proportional frequency. These relationships between immunity and cancer risk also have etiologic implications. Although the incidence of Kaposi sarcoma and of DLCBL non-hodgkin lymphoma increased steadily with immunosuppression with no apparent threshold effect, the incidence of Burkitt lymphoma was stable at all CD4 levels. We suggest that antigen stimulation from HIV or other infections drives lymphocyte proliferation and thereby increases the incidence of Burkitt lymphoma. References (1) Centers for Disease Control revised classification system for HIV infection and expanded surveillance case definition for AIDS among adolescents and adults. MMWR Morb Mortal Wkly Rep 1992 ; 41 : (2) International Collaboration on HIV and Cancer. Highly active antiretroviral therapy and incidence of cancer in human immunodeficiency virusinfected adults. J Natl Cancer Inst 2 ; 92 : (3) Herida A, Mary-Krause M, Kaphan R, Cadranel J, Poizot-Martin I, Rabaud C, et al. Incidence of non-aids-defining cancers before and during the highly active antiretroviral therapy era in a cohort of human immunodeficiency virus-infected patients. J Clin Oncol 23 ; 21 : (4) Clifford GM, Polesel J, Rickenbach M, Dal Maso L, Keiser O, Kofler A, et al. Cancer risk in the Swiss HIV Cohort Study: associations with immunodeficiency, smoking, and highly active antiretroviral therapy. J Natl Cancer Inst 25 ; 97 : (5) Engels EA, Pfeiffer RM, Goedert JJ, Virgo P, McNeel TS, Scoppa SM, et al. Trends in cancer risk among persons with AIDS in the United States AIDS 26 ; 2 : (6) Mocroft A, Vella S, Benfield TL, Chiesi A, Miller V, Gargalianos P, et al. Changing patterns of mortality across Europe in patients infected with HIV-1. Lancet 1998 ; 352 : (7) Centers for Disease Control and Prevention. Update : AIDS: United States, 2. MMWR Morb Mortal Wkly Rep 22 : 51 : (8) Detels R, Tarwater P, Phair JP, Margolick J, Riddler SA, Munoz A. Effectiveness of potent antiretroviral therapies on the incidence of opportunistic infections before and after AIDS diagnosis. AIDS 21 ; 15 : (9) Sterne JAC, Hernan MA, Ledergerber B, Tilling K, Weber R, Sendi P ; for the Swiss HIV Cohort Study. Long-term effectiveness of potent antiretroviral therapy in preventing AIDS and death: a prospective cohort study. Lancet 25 ; 366 : (1) Kaufmann GR, Perrin L, Pantaleo G, Opravil M, Furrer H, Telenti A, et al. CD4 T-lymphocyte recovery in individuals with advanced HIV-1 infection receiving potent antiretroviral therapy for 4 years. The Swiss HIV Cohort Study. Arch Intern Med 23 ; 163 : (11) Tarwater PM, Gallant JE, Mellors JW, Gore ME, Phair JP, Detels R, et al. Prognostic value of plasma HIV RNA among highly active antiretroviral therapy users. AIDS 24 ; 18 : (12) Mbulaiteye SM, Biggar RJ, Goedert JJ, Engels EA. Immune deficiency and risk for malignancy among persons with AIDS. J Acquir Immune Defic Syndr 23 ; 32 : (13) Biggar RJ, Jaffe ES, Goedert JJ, Chaturvedi A, Pfeiffer R, Engels EA ; for the HIV/AIDS Cancer Match Study. Hodgkin lymphoma and immunodeficiency in persons with HIV/AIDS. Blood 26 ; 18 : (14) Borges HT, Watkins J, Stafford R, Biggar RJ. Linkage of selected AIDS and cancer registries in the United States. J Regist Manage 21 ; 28 : (15) Goedert JJ, Cote TR, Virgo P, Scoppa SM, Kingma DW, Gail MH, et al. Spectrum of AIDS-associated malignant disorders. Lancet 1998 ; 351 : (16) Frisch M, Biggar RJ, Engels EA, Goedert JJ. Association of cancer with AIDS-related immunosuppression in adults. JAMA 21 ; 285 : (17) Fritz A, Percy C, Jack A, Shanmugaratnam K, Sobin L, Parkin DM, et al., editors. International Classification of Diseases for Oncology, Third Edition (U.S. Interim Version 2). Geneva (Switzerland) : World Health Organization ; 2. (18) Hollander M, Wolfe DA. Nonparametric statistical methods. 2nd ed. New York (NY) : John Wiley & Sons ; (19) Besson C, Goubar A, Gabarre J, Rosenbaum W, Pialoux G, Chatelet FP, et al. Changes in AIDS-related lymphoma since the era of highly active antiretroviral therapy. Blood 21 ; 98 : (2) Chang Y, Cesarman E, Pessin MS, Lee F, Culpepper J, Knowles DM, et al. Identification of herpesvirus-like DNA sequences in AIDSassociated Kaposi s sarcoma. Science 1994 ; 266 : (21) Krischer J, Rutschmann O, Hirschel B, Vollenweider-Roten S, Saurat JH, Pechere M. Regression of Kaposi s sarcoma during therapy with HIV-1 protease inhibitors: a prospective pilot study. J Am Acad Dermatol 1998 ; 38 : (22) Moray G, Basaran O, Yagmurdur MC, Emiroglu R, Bilgin N, Haberal M. Immunosuppressive therapy and Kaposi s sarcoma after kidney transplantation. Transplant Proc 24 ; 36 : (23) Jaffe ES, Harris NL, Stein H, Vardiman JW, editors. World Health Organization Classification of Tumours. Pathology and genetics of tumours of haematopoietic and lymphoid tissues. Lyon (France): IARC Press ; 21. (24) Knowles DM. Molecular pathology of acquired immunodeficiency syndrome-related non-hodgkin s lymphoma. Semin Diagn Pathol 1997 ; 14 : Downloaded from at Pennsylvania State University on March 3, 214 jnci.oxfordjournals.org JNCI Articles 971