Reply to George and to Stapleton et al.
Department of HIV and Sexually Transmitted Disease Research, Cluster of Infectious Diseases, Municipal Health Service of Amsterdam, Sanquin Research at Landsteiner Laboratory, Academic Medical Center, National AIDS Therapy Evaluation Center
Department of Internal Medicine, Division of Infectious Diseases, Tropical Medicine and AIDS
Department of Human Retrovirology, Academic Medical Center, University of Amsterdam, Amsterdam, The Netherlands
We recently demonstrated that loss of GB virus C (GBV-C) RNA was associated with HIV-1 disease progression [1], confirming the results of other recent studies [2, 3]. Williams et al. [3] also observed an association between persistence of GBV-C RNA and slower HIV disease progression, relative to individuals who lack GBV-C RNA. We did not observe this in our study.
In the letters by George [4] and Stapleton et al. [5] written in response to our study, concerns are raised about the imputation method used to determine the date of HIV-1 seroconversion. We fully agree that a date of HIV-1 seroconversion cannot be determined on the basis of the CD4+ cell count at entry if the cofactor of interest is correlated with the CD4+ cell count. This is considered to be a problem of the marker-based approach [6]. Therefore, the imputed date of HIV-1 seroconversion was derived from a cohort-based estimate of the date of HIV-1 seroconversion and was not based on the CD4+ cell count [7]. Geskus [7] showed that, for the Amsterdam Cohort Study, a conditional mean imputation based on the cohort HIV-1 seroincidence curve gives unbiased results and that the uncertainty in the date of seroconversion hardly changes P values and confidence intervals (CIs). Moreover, in our study, similar results were obtained when the analysis was restricted to seroconverters (n = 123) [1]. For instance, the unadjusted hazard ratio (HR) for death in men who lost GBV-C RNA, compared with men who had no evidence of GBV-C infection, was 3.00 (95% CI, 1.605.61) in seroconverters and was 3.26 (95% CI, 2.314.59) in the total group. The HR decreased toward 1 when adjusted for time-updated CD4+ cell count in seroconverters (HR, 0.66 [95% CI, 0.281.51] and in the total group (HR, 1.21 [95% CI, 0.841.76]) [1].
We do agree with Stapleton et al. that the date of change in GBV-C status was imputed with a large degree of uncertainty. Therefore, we assessed the robustness of our findings by varying the time of GBV-C RNA loss. George raises concerns about the completeness of our data. However, CD4+ cell counts were available for all seroconverters at 1218 months after HIV-1 seroconversion and for most of the subjects entering the Amsterdam Cohort Study already infected with HIV-1 within 2 years after HIV-1 seroconversion, which is only 6 months after the maximum time at which Williams et al. measured baseline CD4+ cell counts [3]. During follow-up, CD4+ cell counts were measured every 3 months. Information on yearly HIV-1 load was available for all seroconverters and for 40% of the subjects already infected with HIV-1. In addition, information on HIV-1 load at baseline was available for almost all subjects already infected with HIV-1. If the HIV-1 load was unavailable, it was obtained from a random-effects model for the joint development of CD4+ cell count and HIV-1 load, which is a reliable method to estimate viral load [8].
George and Stapleton et al. also comment that, in our study, the CD4+ cell counts at baseline in subjects with GBV-C infection were significantly lower than those in subjects without GBV-C infection and raise concerns about whether subjects positive for GBV-C RNA and subjects negative for GBV-C RNA were matched for the duration of HIV-1 infection. However, in our cohort, subjects positive for GBV-C RNA and subjects negative for GBV-C RNA were homogeneous in their duration of HIV-1 infection at the time the baseline CD4+ cell count was obtained (table 1). In addition, in the seroconverters, CD4+ cell counts at baseline were also significantly lower in subjects positive for GBV-C RNA than in those negative for GBV-C RNA (P = .033).
George interpreted a significant benefit of GBV-C acquisition on HIV-1 disease progression from our data. However, in table 3 in our study, 95% confidence intervals (CIs) for each category of GBV-C statusand not overall P values for GBV-C statusshould be used to evaluate the significance of GBV-C acquisition. Because 1 was always within the 95% CI of the hazard ratio, the data in our study do not support a significant effect of GBV-C acquisition on HIV-1 disease progression. Similarly, in model 1 of table 3, the data do not show a significant benefit of GBV-C RNA persistence on progression from AIDS to death (HR, 0.66 [95% CI, 0.401.11]). Indeed, we did observe that GBV-C RNA persistence was associated with a decreased risk of death in models 1, 2, and 3 of table 3 (not on the basis of P < .0001, which, again, is the overall P value) but not with any of the other end points. However, the beneficial effect of GBV-C RNA persistence on progression to death disappeared when adjusted for time-updated CD4+ cell counts during follow-upwhich George and Stapleton et al. considered to be invalid. However, this is exactly what confounding is about. Indeed, it is invalid to control for a variable that is an intermediate step in the causal pathway between exposure and disease if exposure is the variable of interest [9], but we adjusted for time-updated CD4+ cell counts and HIV-1 load to explore possible causal pathways. Because the effect of GBV-C RNA persistence and loss largely disappeared when adjusted for time-updated CD4+ cell counts, our study gives a possible explanation for the effects found: either the CD4+ cell count is an intermediate step in the causal pathway between GBV-C infection and HIV-1 disease progression or the effect of GBV-C infection can be explained by changes in CD4+ cell counts during follow-up, suggesting that GBV-C infection is associated with high CD4+ cell counts. The latter hypothesis seems biologically more plausible, because GBV-C can replicate in CD4+ cells [10]. The loss of CD4+ cells during the course of HIV-1 infection, therefore, implies a loss of target cells for GBV-C.
In conclusion and in agreement with the results of a study by Bjrkman et al. [2], our study provides evidence to support the hypothesis that GBV-C RNA loss is a consequence ofand not a cause ofCD4+ cell loss. We fully agree with Stapleton et al., however, that further studies, with frequent measurement of GBV-C load in each individual, are required to fully understand the relationship between GBV-C infection and HIV-1 disease progression.
References
1. Van der Bij AK, Kloosterboer N, Prins M, et al. GB virus C coinfection and HIV-1 disease progression: the Amsterdam Cohort Study. J Infect Dis 2005; 191:67885. First citation in article
2. Bjorkman P, Flamholc L, Naucler A, Molnegren V, Wallmark E, Widell A. GB virus C during the natural course of HIV-1 infection: viremia at diagnosis does not predict mortality. AIDS 2004; 18:87786. First citation in article
3. Williams CF, Klinzman D, Yamashita TE, et al. Persistent GB virus C infection and survival in HIV-infected men. N Engl J Med 2004; 350:98190. First citation in article
4. George SL. Persistent GB virus C infection is associated with decreased HIV-1 disease progression in the Amsterdam Cohort Study [letter]. J Infect Dis 2005; 191:21567 (in this issue). First citation in article
5. Stapleton JT, Chaloner K, Williams CF. GB virus C infection and survival in the Amsterdam Cohort Study [letter]. J Infect Dis 2005; 191:21578 (in this issue). First citation in article
6. Geskus RB. On the inclusion of prevalent cases in HIV/AIDS natural history studies through a marker-based estimate of time since seroconversion. Stat Med 2000; 19:175369. First citation in article
7. Geskus RB. Methods for estimating the AIDS incubation time distribution when date of seroconversion is censored. Stat Med 2001; 20:795812. First citation in article
8. Geskus RB, Meyer L, Hubert JB, et al. Causal pathways of the effects of age and the CCR5-32, CCR2-64I and SDF-1 3A alleles on AIDS development. JAIDS (in press). First citation in article
9. McNamee R. Confouding and confounders. Occup Environ Med 2003; 60:22734. First citation in article
10. Xiang J, Wunschmann S, Schmidt W, Shao J, Stapleton JT. Full-length GB virus C (hepatitis G virus) RNA transcripts are infectious in primary CD4-positive T cells. J Virol 2000; 74:912533. First citation in article, 百拇医药(Akke K. Van der Bij, Nico)
Department of Internal Medicine, Division of Infectious Diseases, Tropical Medicine and AIDS
Department of Human Retrovirology, Academic Medical Center, University of Amsterdam, Amsterdam, The Netherlands
We recently demonstrated that loss of GB virus C (GBV-C) RNA was associated with HIV-1 disease progression [1], confirming the results of other recent studies [2, 3]. Williams et al. [3] also observed an association between persistence of GBV-C RNA and slower HIV disease progression, relative to individuals who lack GBV-C RNA. We did not observe this in our study.
In the letters by George [4] and Stapleton et al. [5] written in response to our study, concerns are raised about the imputation method used to determine the date of HIV-1 seroconversion. We fully agree that a date of HIV-1 seroconversion cannot be determined on the basis of the CD4+ cell count at entry if the cofactor of interest is correlated with the CD4+ cell count. This is considered to be a problem of the marker-based approach [6]. Therefore, the imputed date of HIV-1 seroconversion was derived from a cohort-based estimate of the date of HIV-1 seroconversion and was not based on the CD4+ cell count [7]. Geskus [7] showed that, for the Amsterdam Cohort Study, a conditional mean imputation based on the cohort HIV-1 seroincidence curve gives unbiased results and that the uncertainty in the date of seroconversion hardly changes P values and confidence intervals (CIs). Moreover, in our study, similar results were obtained when the analysis was restricted to seroconverters (n = 123) [1]. For instance, the unadjusted hazard ratio (HR) for death in men who lost GBV-C RNA, compared with men who had no evidence of GBV-C infection, was 3.00 (95% CI, 1.605.61) in seroconverters and was 3.26 (95% CI, 2.314.59) in the total group. The HR decreased toward 1 when adjusted for time-updated CD4+ cell count in seroconverters (HR, 0.66 [95% CI, 0.281.51] and in the total group (HR, 1.21 [95% CI, 0.841.76]) [1].
We do agree with Stapleton et al. that the date of change in GBV-C status was imputed with a large degree of uncertainty. Therefore, we assessed the robustness of our findings by varying the time of GBV-C RNA loss. George raises concerns about the completeness of our data. However, CD4+ cell counts were available for all seroconverters at 1218 months after HIV-1 seroconversion and for most of the subjects entering the Amsterdam Cohort Study already infected with HIV-1 within 2 years after HIV-1 seroconversion, which is only 6 months after the maximum time at which Williams et al. measured baseline CD4+ cell counts [3]. During follow-up, CD4+ cell counts were measured every 3 months. Information on yearly HIV-1 load was available for all seroconverters and for 40% of the subjects already infected with HIV-1. In addition, information on HIV-1 load at baseline was available for almost all subjects already infected with HIV-1. If the HIV-1 load was unavailable, it was obtained from a random-effects model for the joint development of CD4+ cell count and HIV-1 load, which is a reliable method to estimate viral load [8].
George and Stapleton et al. also comment that, in our study, the CD4+ cell counts at baseline in subjects with GBV-C infection were significantly lower than those in subjects without GBV-C infection and raise concerns about whether subjects positive for GBV-C RNA and subjects negative for GBV-C RNA were matched for the duration of HIV-1 infection. However, in our cohort, subjects positive for GBV-C RNA and subjects negative for GBV-C RNA were homogeneous in their duration of HIV-1 infection at the time the baseline CD4+ cell count was obtained (table 1). In addition, in the seroconverters, CD4+ cell counts at baseline were also significantly lower in subjects positive for GBV-C RNA than in those negative for GBV-C RNA (P = .033).
George interpreted a significant benefit of GBV-C acquisition on HIV-1 disease progression from our data. However, in table 3 in our study, 95% confidence intervals (CIs) for each category of GBV-C statusand not overall P values for GBV-C statusshould be used to evaluate the significance of GBV-C acquisition. Because 1 was always within the 95% CI of the hazard ratio, the data in our study do not support a significant effect of GBV-C acquisition on HIV-1 disease progression. Similarly, in model 1 of table 3, the data do not show a significant benefit of GBV-C RNA persistence on progression from AIDS to death (HR, 0.66 [95% CI, 0.401.11]). Indeed, we did observe that GBV-C RNA persistence was associated with a decreased risk of death in models 1, 2, and 3 of table 3 (not on the basis of P < .0001, which, again, is the overall P value) but not with any of the other end points. However, the beneficial effect of GBV-C RNA persistence on progression to death disappeared when adjusted for time-updated CD4+ cell counts during follow-upwhich George and Stapleton et al. considered to be invalid. However, this is exactly what confounding is about. Indeed, it is invalid to control for a variable that is an intermediate step in the causal pathway between exposure and disease if exposure is the variable of interest [9], but we adjusted for time-updated CD4+ cell counts and HIV-1 load to explore possible causal pathways. Because the effect of GBV-C RNA persistence and loss largely disappeared when adjusted for time-updated CD4+ cell counts, our study gives a possible explanation for the effects found: either the CD4+ cell count is an intermediate step in the causal pathway between GBV-C infection and HIV-1 disease progression or the effect of GBV-C infection can be explained by changes in CD4+ cell counts during follow-up, suggesting that GBV-C infection is associated with high CD4+ cell counts. The latter hypothesis seems biologically more plausible, because GBV-C can replicate in CD4+ cells [10]. The loss of CD4+ cells during the course of HIV-1 infection, therefore, implies a loss of target cells for GBV-C.
In conclusion and in agreement with the results of a study by Bjrkman et al. [2], our study provides evidence to support the hypothesis that GBV-C RNA loss is a consequence ofand not a cause ofCD4+ cell loss. We fully agree with Stapleton et al., however, that further studies, with frequent measurement of GBV-C load in each individual, are required to fully understand the relationship between GBV-C infection and HIV-1 disease progression.
References
1. Van der Bij AK, Kloosterboer N, Prins M, et al. GB virus C coinfection and HIV-1 disease progression: the Amsterdam Cohort Study. J Infect Dis 2005; 191:67885. First citation in article
2. Bjorkman P, Flamholc L, Naucler A, Molnegren V, Wallmark E, Widell A. GB virus C during the natural course of HIV-1 infection: viremia at diagnosis does not predict mortality. AIDS 2004; 18:87786. First citation in article
3. Williams CF, Klinzman D, Yamashita TE, et al. Persistent GB virus C infection and survival in HIV-infected men. N Engl J Med 2004; 350:98190. First citation in article
4. George SL. Persistent GB virus C infection is associated with decreased HIV-1 disease progression in the Amsterdam Cohort Study [letter]. J Infect Dis 2005; 191:21567 (in this issue). First citation in article
5. Stapleton JT, Chaloner K, Williams CF. GB virus C infection and survival in the Amsterdam Cohort Study [letter]. J Infect Dis 2005; 191:21578 (in this issue). First citation in article
6. Geskus RB. On the inclusion of prevalent cases in HIV/AIDS natural history studies through a marker-based estimate of time since seroconversion. Stat Med 2000; 19:175369. First citation in article
7. Geskus RB. Methods for estimating the AIDS incubation time distribution when date of seroconversion is censored. Stat Med 2001; 20:795812. First citation in article
8. Geskus RB, Meyer L, Hubert JB, et al. Causal pathways of the effects of age and the CCR5-32, CCR2-64I and SDF-1 3A alleles on AIDS development. JAIDS (in press). First citation in article
9. McNamee R. Confouding and confounders. Occup Environ Med 2003; 60:22734. First citation in article
10. Xiang J, Wunschmann S, Schmidt W, Shao J, Stapleton JT. Full-length GB virus C (hepatitis G virus) RNA transcripts are infectious in primary CD4-positive T cells. J Virol 2000; 74:912533. First citation in article, 百拇医药(Akke K. Van der Bij, Nico)