|Home | About | Journals | Submit | Contact Us | Français|
HIV-infected persons treated with highly active antiretroviral therapy (HAART) continue to have elevated risk for non-Hodgkin lymphoma (NHL). We conducted a retrospective cohort study of NHL among patients at an urban HIV clinic (N=3025). Proportional hazards models identified immunologic and virologic predictors of NHL. Sixty-five NHLs arose during 1989-2006. NHL incidence declined over time. Nonetheless, 51 NHLs (78%) occurred within the HAART era (1996-2006). NHL risk increased with declining CD4 count (p-trend<0.0001) and increasing HIV viral load (p-trend=0.005). In a multivariable model, NHL risk was independently associated with both current CD4 count (hazard ratios 7.7 and 3.8, respectively, for CD4 counts 0-99 and 100-249 vs. 250+ cells/mm3; p-trend<0.0001) and prior time spent with a viral load above 5.00 log10 copies/ml (hazard ratios of 3.4, 2.6, and 6.8, respectively, for 0.1-0.4, 0.5-1.4, and 1.5+ years vs. 0 years; p-trend=0.004). Although serum globulin levels were elevated compared to the general population, NHL risk was unrelated to this B-cell activation marker (p=0.39). Among HIV-infected individuals in the HAART era, NHLs are linked to immunosuppression and extended periods of uncontrolled HIV viremia. The association with high-level viremia could reflect detrimental effects on immune function related to incompletely effective HAART or direct effects on B-cells.
HIV-infected individuals have a markedly elevated risk for developing non-Hodgkin lymphoma (NHL) (1). Risk is especially increased for diffuse large B-cell lymphoma (DLBCL), Burkitt NHL, and central nervous system (CNS) NHL, and these NHL subtypes are considered AIDS-defining malignancies (2). Improvements in immune function attributable to highly active antiretroviral therapy (HAART), available in Western countries since 1996, have led to substantial declines in overall NHL risk in HIV-infected people (3).
The pathogenesis of NHL in the setting of HIV infection has not been fully elucidated. Transformation of B-cells by Epstein Barr virus (EBV) likely plays a role in CNS NHL and DLBCL (4). These NHLs presumably arise due to loss of cell-mediated immune control of latent herpesvirus infection attributable to progressive HIV disease (5). In contrast, EBV does not appear to be involved in the development of AIDS-related Burkitt NHL (4), and the incidence of this NHL subtype has not changed during the HAART era (3).
While progressive loss of CD4 positive T-cells is important in AIDS lymphomagenesis (6), other immune mechanisms may also be relevant. One possibility is that the development of AIDS-related NHL is determined not solely by immune deficiency at a particular point in time (reflected, for example, by the current CD4 count), but in addition by the depth of prior immunosuppression (i.e., the nadir CD4 count) or duration of immunosuppression (i.e., previous time spent with a low CD4 count). HIV viral load may be an independent marker of immunosuppression among people with advanced HIV disease (7), and one recent study demonstrated that cumulative duration of HIV viremia is predictive of NHL (8). These alternative measures of immune status could be especially relevant in the HAART era, when therapy can halt inexorably declining immune function and allow manifestation of outcomes determined by events that occurred much earlier, before initiation of therapy.
Of interest, in a case-control study nested within an HIV clinic cohort during the pre-HAART era, Grulich et al. reported that high levels of serum globulins (mostly reflecting elevated IgG immunoglobulin) preceded the diagnosis and were predictive of AIDS NHL in a dose-response manner (9). B-cell dysfunction in HIV-infected people is characterized by production of abnormally low levels of antibodies to specific pathogens and poor immune responses to vaccines. Simultaneously, total serum levels of IgG are actually elevated, reflecting a non-specific polyclonal activation of B-cells (10-12).
To a large extent, clinicians currently utilize the CD4 count to guide decisions about when to initiate HIV treatment (13), but additional markers might offer complementary information in capturing NHL risk and thus facilitate these decisions. In the present study, we evaluated immunologic and virologic predictors of NHL risk in a large urban clinic-based cohort of HIV-infected individuals. We sought to identify whether various markers of immune dysfunction (including current and nadir CD4 count, HIV viral load, serum globulin) were predictive of development of NHL. A sizeable fraction of the clinic follow-up occurred after 1996, so that our results reflect on the pathogenesis of AIDS-related NHL during the HAART era.
The Johns Hopkins Hospital Moore Clinic provides primary and specialty care to HIV-infected individuals living in Baltimore, Maryland. Clinic patients receive a detailed baseline evaluation, with collection of demographic and clinical data. Patients are routinely seen in the clinic at three-month intervals, with frequent measurement of HIV disease markers, i.e., CD4 counts and (beginning in 1996) HIV viral loads. Decisions regarding antiretroviral therapy are made by individual clinic providers following accepted practice guidelines. Since 1989, clinic records and hospital sources have been abstracted into an observational database (14). For the present study, data were complete through 2006. Institutional review boards at Johns Hopkins Hospital and the National Cancer Institute approved this study.
NHL events were ascertained through review of clinic charts and hospital records, and were confirmed by pathology reports and clinical history. HAART use was defined as therapy with two nucleoside reverse transcriptase inhibitors and either a protease inhibitor or non-nucleoside reverse transcriptase inhibitor. The clinic database contained information on start and stop dates for each medication, although detailed information on treatment interruptions and adherence was not available. Serum globulin levels were calculated as the difference between serum total protein and albumin levels, which were routinely measured at clinic visits. Among HIV-infected people, IgG comprises the majority of serum globulins, and serum globulin levels are correlated with IgG levels (r=0.57) (9).
We evaluated NHL risk beginning with the latter of date of entry into the clinic cohort or first measurement of CD4 count, HIV viral load, or serum globulin. Evaluation continued until the earliest of NHL, death, or loss to follow-up. We calculated NHL incidence as a function of calendar year and compared that to the proportion of follow-up time in each calendar year for which subjects were receiving a HAART regimen. We also calculated NHL incidence in HAART users and non-users.
To assess predictors of NHL risk, we constructed Cox models (SAS PHREG) that utilized calendar year as the time metric, in order to adjust optimally for secular trends in NHL incidence. This approach allowed subjects to have delayed entry on the calendar date when their follow-up started. HIV viral loads were routinely measured beginning in 1996 and so were only evaluated as an NHL risk factor in this period. On average, for each year of follow-up there were 2.2 CD4 count measurements (corresponding to 5.3 months between measurements), 3.6 serum globulin measurements (3.3 months) and, beginning in 1996, 2.3 HIV viral load measurements (5.1 months). We considered the most recent laboratory value to be “current” until the next measurement. Laboratory markers (current and nadir CD4 count, current HIV viral load, current serum globulin) were evaluated as time-dependent covariates in the Cox models. Based on the observed associations with NHL risk, we selected thresholds defining a low CD4 count (below 250 cells/mm3) or a high viral load (at least 5.0 log10 copies/ml). We then evaluated the associations between NHL risk and cumulative duration with a low CD4 count or high viral load. These durations were measured since the start of follow-up and were modeled as time-dependent variables, calculated as the sums of separate time intervals considering the most recent laboratory values to be current until the next measurement was obtained. In univariate models, we assessed the proportional hazards assumption by incorporating an interaction between the variable of interest and calendar time, and by examining log-log survival plots. In additional models, we evaluated the effects of laboratory markers on NHL risk, adjusted for demographic factors that were significant predictors in univariate models.
As described below, we did not find HAART use to be protective against the development of NHL. One reason is that HAART use increased sharply after 1996, so that estimation of the effect of HAART on NHL risk was imprecise after we adjusted for calendar year trends. Also, we did not have detailed information on adherence and interruptions in treatment, so that some periods that we considered patients to be on HAART may not have corresponded to actual HAART use. In addition, observational studies can substantially underestimate the beneficial effect of HAART due to selection by clinicians of the most immunodeficient patients for its use (15). Although we thus could not reliably assess the effects of HAART on NHL risk, we included HAART use as a variable in adjusted and multivariable regression models to reduce the possibility of confounding of the other effects of interest.
The study included 3025 HIV-infected subjects (Table 1) who began follow-up in 1989-2006 (median baseline year 1998). The majority were male and non-Hispanic black, with a median baseline age of 38 years. The most common HIV risk factor was injection drug use (39.8%), while male-sex-with-males (MSM) activity was reported by 26.7%. At baseline, the median CD4 count was 285 cells/mm3 and median HIV viral load was 4.35 log10 copies/ml, and only 28.0% of subjects were receiving HAART. The median serum globulin level at baseline was 4.2 g/dL.
During follow-up, 65 subjects developed NHL (2.1%, incidence 4.1 per 1000 person-years). The NHLs comprised 28 DLBCLs, 2 Burkitt NHLs, 28 CNS NHLs, and 7 NHLs of other/unknown subtype. As shown in Figure 1, NHL incidence was highest in 1992, with a steep decline beginning well before 1996. NHL incidence continued to decline throughout the late 1990s and 2000s. The decline in NHL incidence thus began before the sharp increase in HAART use during 1996-1999 (Figure 1).
Despite the marked decline in NHL incidence over time, 51 of the NHLs (78%) occurred in 1996 or after; among these cases, 25 (49%) were receiving HAART prior to development of NHL. NHL incidence was lower in HAART users than non-HAART users (3.0 vs. 5.2 per 1000 person-years). However, this association with HAART was not apparent in the proportional hazards regression model (hazard ratio 1.0, 95%CI 0.6-1.9), which fully accounted for calendar trends in NHL risk.
As shown in Table 2, NHL risk was significantly higher in MSM than in other males and females, although this difference attenuated over time (p=0.05 for test of non-proportional hazards). NHL risk was lower in non-Hispanic blacks than individuals of other race/ethnicity (hazard ratio 0.5) but did not vary by age. NHL risk increased with declining CD4 count (p-trend<0.0001) and increasing HIV viral load (p-trend=0.005), and was especially elevated in persons with HIV viral load of 5.00 log10 copies/ml or higher (Table 2). Finally, NHL risk was unrelated to serum globulin level (p=0.39).
Table 3 presents associations between laboratory markers and NHL risk, adjusted for HAART use, sex and HIV risk group, race, and calendar year. Current CD4 count remained a strong predictor of NHL risk after adjustment (p<0.0001). Nadir CD4 count was also predictive, but appeared no better than current CD4 count, and only data from 1994 onward could be evaluated because of non-proportionality of hazards prior to this time point. NHL risk also increased with prior time spent with a CD4 count below 250 cells/mm3 (Table 3), but the amount of prior time was not a better predictor than the fact that any time was spent with a CD4 count below 250 cells/mm3 (p=0.39).
HIV viral load could be evaluated as an NHL risk factor only in 1996 and after. As shown in Table 3, NHL risk was significantly associated with elevated current HIV viral load (i.e., at least 5.00 log10 copies/ml, adjusted hazard ratio 3.8). Furthermore, NHL risk rose with increasing prior time spent with a viral load of at least 5.00 log10 copies/ml (p<0.0001), and this duration of time was a better predictor of NHL risk than was the mere presence of any prior viral load of at least 5.00 log10 copies/ml (p=0.05). Current serum globulin level still did not predict NHL risk in an adjusted model (p=0.70, Table 3).
We constructed two multivariable models for NHL risk that included CD4 count and HIV viral load (Table 3). These models were restricted to 1996 and after because of availability of viral load data. In the first multivariable model that included both current CD4 count and current HIV viral load (model 1, Table 3), only current CD4 count independently predicted NHL risk. In contrast, in another multivariable model (model 2, Table 3) that considered prior time spent with an HIV elevated viral load (at least 5.00 log10 copies/ml) instead of current viral load, both current CD4 count and prior time spent with an elevated HIV viral load were significant independent predictors of NHL risk (p-trend<0.0001 and p-trend=0.004, respectively).
In this study of over 3000 HIV-infected patients followed at an inner-city hospital clinic in Baltimore, Maryland, we observed a substantial decline in NHL incidence over an 18-year period. In the HAART era (1996 onward), NHL risk was strongly elevated in subjects who had a low CD4 count or had previously spent several months or more with an elevated HIV viral load. We did not see an association between serum globulin levels and NHL risk.
Our results showing that NHL risk rises sharply with declining CD4 count are in good accord with results from other studies (6,16,17). Although we had too few cases to analyze risk for separate NHL subtypes, prior work demonstrates that the relationship with CD4 count is apparent for the most frequent AIDS NHL subtypes, DLBCL and CNS NHL (6). EBV is commonly detected in tumor tissue from AIDS-related CNS NHLs (virtually 100% of cases) and DLBCLs (approximately 50%) but not in Burkitt NHL (4). As observed in our study, the current CD4 count (a marker of present immune status) appears to be as good a marker of NHL risk as nadir CD4 count (16), and the relationship between CD4 count and NHL risk continues to be important in the HAART era (6,16). Although we had no data on the EBV status of our NHL cases, the strong association with CD4 count supports the hypothesis that a major component of AIDS lymphomagenesis can be explained by severe HIV-induced immunosuppression, which allows for EBV reactivation and virus-driven transformation of lymphocytes (5).
The effect of HAART in reducing the risk of AIDS associated NHL has been dramatic, with declines of 40-80% comparing NHL incidence in the HAART era to that in the pre-HAART era (3,16,18,19). Prior studies have also described lower NHL risk in HAART users compared with non-users, with benefits apparent almost immediately and extending over years of HAART use (20). The mechanism underlying the effectiveness of HAART in reducing NHL risk is presumably related, at least in part, to its ability to lead to improved immune status (i.e., increases in CD4 count) and control of EBV.
As shown in Figure 1, NHL incidence in this HIV clinic cohort decreased dramatically well before HAART use climbed after 1996. The reason for a decline in NHL incidence before the introduction of HAART is unclear, but it has been noted previously in U.S. national trends among people with AIDS (3). The temporal decline could partly relate to increasing use by clinicians of moderately effective HIV therapy (i.e., dual nucleoside therapy). After we controlled for the calendar trend in NHL incidence in our proportional hazards model, we were unable to demonstrate a protective effect of HAART on the risk of developing NHL. Nonetheless, the close correlation between calendar year and HAART use led to wide confidence limits for the estimated HAART effect, including the possibility of substantial benefit (i.e., lower limit of 0.6 for the hazard ratio, corresponding to a 40% reduction in NHL risk). As noted above, we lacked detailed information on interruptions in HAART and subjects’ adherence to their regimens. Also, in clinical practice, HAART was prescribed preferentially to patients with more severe immunodeficiency, as indicated by contemporaneous guidelines (e.g., reference (13)). This pattern of use could have attenuated the apparent benefits attributable to HAART (15).
Of note, we observed an increase in NHL risk with increasing prior duration of high-level viremia (i.e., HIV viral load of at least 5.00 log10 copies/ml). NHL risk rose steeply even after only a few months of high-level viremia (multivariable hazard ratios = 2.6-3.4) and increased further with high-level viremia for 1.5 years or more (multivariable hazard ratio = 6.8). These results corroborate those recently reported by Zoufaly et al., who described increasing NHL risk to be independently associated with cumulative HIV viremia among individuals treated with HAART (8). In that study, the investigators measured cumulative viremia as a product of the log-transformed HIV viral load (regardless of level) and duration. In contrast, we considered only time spent with an HIV viral load of at least 5.00 log10 copies/ml, because our initial analyses indicated that only viremia of this magnitude was associated with elevated NHL risk (Table 2). The relationship between viremia and NHL risk could reflect detrimental effects on immune status associated with absence of treatment or incompletely effective HAART. Our analyses showed that this association between duration of high-level viremia and NHL risk was independent of CD4 count, and in general, the effect of HIV replication on AIDS risk is only partly explained by its association with future declines in CD4 count (7,21). The association of NHL risk with prolonged uncontrolled viremia may therefore reflect other disturbances of immune function, such as chronic immune activation or direct effects of HIV on B-cells (7,10,12).
An elevated level of total serum globulin is thought to reflect hypergammaglobulinemia and therefore generalized B-cell activation (11). In accord with expectation, serum globulin levels in our HIV-infected subjects were higher than observed in healthy individuals (i.e., median 4.2 g/dL vs. 2.0-3.5 g/dL in the general population) (22). It is difficult to reconcile our null findings relating serum globulin results and NHL risk with those reported by Grulich et al. (9). However, supporting our negative results, a case-control study found no association between the risk of AIDS NHL and levels of IgG, IgM, IgA, and IgE measured in pre-diagnostic serum specimens (23). While the inconsistency in association between globulin levels and NHL risk might suggest that B-cell activation is irrelevant in the pathway to NHL, another explanation could be that B-cell activation is important only for some NHL subtypes, or that alternative markers of B-cell activation are required. Along these lines, serum levels of soluble CD30 (a lymphocyte surface molecule) or free immunoglobulin light chains may provide more sensitive methods for assessing B-cell activation (23,24).
We found additional associations between demographic characteristics and NHL risk. Among HIV-infected individuals, higher NHL risks are generally observed in males compared to females, whites compared to non-whites, and with increasing age (16,20,25-27). Some studies (16,20,26), but not all (8,28), have reported an especially elevated risk for NHL among HIV infected MSM. These associations could reflect accumulation of genetic damage with age, differences in exposures to environmental cofactors or socioeconomic status, or the variable presence of an additional causative infectious agent. For example, human herpesvirus 8 is implicated in a minority of DLBCLs (29), and coinfection with this virus is especially common in HIV-infected MSM. NHL risk could also have been affected by duration of HIV infection or polymorphisms in host immune genes, but we did not have data on these factors. The availability of longitudinal data for a large clinic population, extending into the HAART era, was a strength of our study. We also had access to repeated laboratory measurements over time, which allowed us to thoroughly examine associations between relevant laboratory markers and NHL risk.
In conclusion, our results highlight the continued importance of immunosuppression in the development of NHL among HIV-infected individuals during the HAART era. The increased NHL risk observed with prolonged HIV viremia may reflect the effects of chronic immune activation and could support earlier initiation of HAART to prevent development of this malignancy. Studies utilizing additional markers of B-cell activation would be informative.
This study was supported by the Intramural Research Program of the National Cancer Institute, and by National Institutes of Health grants R01 DA11602, R01 AA16893, and K24 DA00432.