The novel findings of our study were trimodal incidence patterns for BL among males and bimodal among females for both Whites and Blacks. The bimodal incidence peaks were robust in APC models24
and in analyses excluding registries with many cases of HIV/AIDS, and were supported by significant BL rate slope contrasts near age 22 years and 58 years. The middle-aged peak was observed among both White and Black males, but not among females, suggesting that the HIV/AIDS epidemic among males 25
may explain the excess in the middle aged men compared with females.
Bimodality is not uncommon in other cancers,17, 30
but BL is currently not considered to be bimodal and, to our knowledge, trimodality has never been reported for BL. Tri/bimodality suggests heterogeneity in a disease due two or more distinct cancer populations for a condition coded a one disease.30
BL is a unique B-cell lymphoproliferative malignancy that is currently considered to be a single entity with isomorphic clinical-epidemiological variants.6
Tri/bimodality suggests that the underlying risk of BL does not increase monotonically log-linearly with age. Therefore, linear summary measures for the age effect on risk, such as the LAT, may obscure fluctuation in age-specific risk of BL and the insights, about exposure or the underlying biology, such fluctuation may provide. Confirmation of bi/trimodality of BL in the established BL clinical-epidemiological variants, i.e., endemic, sporadic, and AIDS-related,5, 6
will justify conducting hypothesis-driven studies to investigate the biological basis for peak risks of BL at different ages in different populations.
Trimodality among adult males was due to a third peak during the adult years and may be mostly explained by the U.S. HIV/AIDS epidemic, given the well-established association of BL with HIV/AIDS31
and that males are disproportionately affected by the HIV/AIDS epidemic25
. The lack of a peak among middle-aged females may be due to a lower prevalence of HIV/AIDS in this group25
and possibly to other factors that may be responsible for to gender disparity in the risk for BL in middle-aged adults as has been shown for childhood BL.32
In sensitivity analyses excluding registries with many AIDS cases, the age-incidence peak in male middle-aged adults was blunted, but not eliminated. The blunted adult peak may be due to residual HIV/AIDS-related BL cases or may imply that the adult peak is a mixture of AIDS- and non-AIDS-related BL. The attributable fraction (AF) of BL cases due to HIV/AIDS can be estimated mathematically using the formula AF=p
=prevalence of HIV/AIDS in the general population and RR
=relative risk for BL in persons with HIV/AIDS. We estimated that 71% (or 1099 of adult cases) of adult BL would be AIDS-related and 29% (448 of adult cases) non-AIDS related, assuming that the average HIV/AIDS prevalence was 0.05% 25
and the RR for BL given AIDS is 50 among U.S. adults with AIDS as compared to the general adult population 34
. Our results suggest that there are adult and geriatric onset BL cases that appear different from childhood and AIDS-related BL that need to be studied further. The possibilities include an AIDS-related adult BL peak that may be due to a confluence, i.e., mixture, of distinct late-onset pediatric-type and early-onset geriatric-type BL rates, which are shifted right and left, respectively, or due to an infectious agent other than HIV.
Geriatric BL among both males and females accounted for one-fifth of BL and rates increased rapidly in contrast to the stable childhood rates. Because HIV-related BL occurs relatively early during HIV disease and is often AIDS-defining, 28
and trends in geriatric BL were increasing before the mid-1980s when the impact of the HIV epidemic was minimal, geriatric BL is unlikely to be due to HIV/AIDS. However, it is possible that a small fraction of BL associated with HIV/AIDS may have a long latency. Trends in geriatric BL most likely have been influenced by improvements in diagnosis or access to care 35
or by improvements in treatment of more indolent forms of lymphoma in older persons thus allowing for increased transformation of indolent lymphomas to BL.36
Our finding of substantial differences in BL mortality by age group is consistent with heterogeneity in BL suggested by tri/bimodal incidence peaks, but differences in the treatments routinely used, or tolerance to them, at different ages may also contribute.
The major strengths of our study were its large size, population-based sample, and the use of APC models to simultaneously adjust for potential confounding by calendar-period and birth-cohort factors. APC models allowed us to confirm, as demonstrated in analysis of cross-sectional age-specific rates ( and ), that the underlying risk of BL is non-linear, consisting of a tri/bimodal pattern comprising a well established pediatric peak, a novel geriatric peak, and a peak in middle-aged adults, which is explained, in part, by the HIV/AIDS. Even so, large numbers and modeling cannot overcome some limitations. Lack of centralized, standardized expert pathology review of cases was a limitation, 8
but any resulting diagnostic misclassification should be random. SEER cancer registry data does not include important individual clinical data including HIV serology, EBV serology, or EBV or molecular phenotype.
To summarize, we found novel tri/bimodal incidence patterns for BL, which showed disparities by gender but not race. Future research should investigate factors contributing to occurrence of BL at different ages, which may involve different behavior of B-cells in immune function at different ages, different infectious organisms associated with risk at different ages, or variation in the interaction of genetic and infectious factors at different ages, particularly in relation to MYC translocation and/or dysfunction.