Laboratory markers that predict HIV-1 disease progression include plasma viral burden, CD4+ T-cell count, and markers of immune activation such as CD38 expression on CD8 T cells. Some studies have suggested that a viral load set point is reached within the first few years of HIV-1 infection (1,2) and remains fairly stable until a few years before AIDS develops, when a gradual or rapid rise in plasma viral levels may occur (2-4). Although the set point theory is not supported by some studies (5,6), a lack of resolution to this ongoing debate has not precluded interest in the question of whether, or how, the prognostic value of viral load may differ in early- and late-stage disease (7). There is some appeal to the hypothesis that viral burden shortly after infection may be a stronger predictor of outcome, because the viral load predicts future CD4+ cell decline over the long term (8), although the CD4 cell count may more directly predict AIDS or death later in the course of infection (4,9).
An additional feature of HIV-1 disease is immune activation. Elevated CD38 expression on CD8+ T cells, measured as the median number of CD38 molecules found on the surface of cells, has been found to be a particularly strong prognostic marker during later stages of HIV-1 disease (10), even stronger than viral burden (9,11). In a more recent study among men with advanced HIV-1 disease (defined as having a CD4+ T-cell count of <50 cells/mm3) elevated CD38 expression on both CD4+ and CD8+ T cells was a strong predictor of poor outcome (12). Other immunologic markers, plasma viral burden, and virus coreceptor use did not predict length of survival in that study. To date, the predictive value of CD38 expression in early HIV-1 infection has not been investigated in detail.
Our goal in this study was to compare the relative predictive value for progression of HIV infection to AIDS of CD38 expression on CD8+ cells with that of CD4+ T-cell count and viral burden at early and late time points after HIV-1 infection. This is particularly important because not all HIV-1-infected people know with certainty how long they have been infected; thus, if marker measurements have a substantially different risk for AIDS early and late after infection, it would be difficult to interpret an individual's risk of AIDS from a single marker measurement alone. Although many people are now treated with antiretroviral therapies that effectively suppress viral burden and prolong life, it nevertheless remains important to be as accurate as possible in our estimation of the meaning of predictive marker measurements in the absence of treatment. These estimates can assist in determining when to start treatment and also provide the basis for judging a treated person's prognosis should treatment prove ineffective for that individual.
MATERIALS AND METHODS
Study Sample
Between March 1984 and April 1985, the Multicenter AIDS Cohort Study (MACS) enrolled a cohort of 4954 homosexual men who were 18 years of age or older and free of the diseases that were subsequently incorporated in the Centers for Disease Control and Prevention's 1987 definition of clinical AIDS. Details about the recruitment and characteristics of the MACS cohort have been reported elsewhere (13-15). The study was approved by the internal review board at each of four clinical centers (in Baltimore, Chicago, Los Angeles, and Pittsburgh), and all participants gave written informed consent. Participants in the MACS return for follow-up visits at 6-month intervals.
The baseline visit for the early infection sample occurred between April 1985 and November 1986. A sample of 191 men was randomly selected from among 1604 who were included in a published study of plasma viral load and CD4+ T-lymphocyte counts as prognostic markers in the MACS (8). An additional 49 men of the 1604 were selected because they had a low viral load, defined as <3000 copies in the branched DNA assay (Chiron Corporation, Emeryville, CA, U.S.A.). These 240 men were seropositive at enrollment in the MACS and remained AIDS-free (according to the Centers for Disease Control and Prevention's 1987 definition) at the early infection baseline time point, which occurred an estimated median of 2.8 years (10th-90th percentile: 2.1-4.4 years) after infection (16). For this study, measurements of CD38 expression of CD8+ T cells (see below) were made on all 240 men in the study sample to enable the current analysis.
The baseline visit for the late infection sample occurred between January and June 1992. This study sample consisted of the 252 men whose plasma viral load, CD4+ T-lymphocyte counts, and CD38 measurements were included in a previously published study (11). The men were AIDS-free at the late infection baseline time point, which occurred an estimated median of 8.7 years (10th-90th percentile: 5.0-10.0 years) after infection. Further analyses of these data are presented here so as to draw comparisons involving the predictive power of markers after long and short duration of HIV-1 infection. The ethnic characteristics of the participants were similar in both groups. At the early and late time points, respectively, 87% and 88% of the participants were white, 6% and 8% were Hispanic, 3% and 2% were black, and 4% and 2% were other.
Flow Cytometric Measurements
At the early and late time points, CD4+ T-lymphocyte percentages were measured by flow cytometry and absolute numbers were calculated from hematologic measurements of lymphocyte numbers. These methods were described by Giorgi et al. (17) and Schenker et al. (18) relative to the early and late time periods, respectively. CD38 values were determined using fluorescein isothiocyanate-conjugated human leukocyte antigen-DR, phycoerythrin-conjugated CD38, and allophycocyanin-conjugated CD8 monoclonal antibodies. At the early time point, CD38 was measured on cryopreserved peripheral blood mononuclear cells that had been stored in the MACS repository (Biomedical Research Incorporated, Rockville, MD, U.S.A.), and at the late time point, CD38 expression was measured on fresh blood (10). We measured CD38 relative fluorescence intensity as the median relative fluorescence intensity channel number using single-parameter histograms with no cursors set (10). We then calculated the number of molecules of CD38 expressed per CD8+ T cell using a calibration factor for the number of phycoerythrin molecules detected per channel on our FACScan (early) or FACSCalibur (late) (Becton Dickinson Immunocytometry Systems, San Jose, CA, U.S.A.) flow cytometer (19). These values were 41 for the FACScan and 44 for the FACSCalibur.
Preliminary analyses adjusting for differences between the early and late measurements as a result of spurious technical variations in CD4 measurements and differences in CD38 expression on fresh versus frozen cells were conducted. Because the inferences were no different from those based on the unadjusted analyses reported here, the adjustments and results subsequent to them are not discussed further.
Plasma HIV-1 RNA
The Amplicor HIV-1 Monitor assay (Roche Molecular Systems, Branchburg, NJ, U.S.A.) was used to measure viral RNA at both visits. Specimens had been collected in heparin, and measurements were made in early and late infection specimens as described by Mellors et al. (8) and Liu et al. (11), respectively. For both sets of assays, we verified that the measurements made in heparinized plasma using the Roche assay were reliable and robust (20).
Statistical Methods
Descriptive statistics were compiled on demographics and marker values at both time points. Individuals were categorized based on marker values, and AIDS-free survival was assessed via Kaplan-Meier analysis and log-rank tests. Cox proportional hazards models (21) were used to evaluate increments in risk of AIDS over 4 years of follow-up across categories. Because early and late marker distributions were different, a common set of cutoff points was adopted as follows: CD4 count was divided at 200, 350, 500, and 750 cells/mm3; HIV RNA was divided at 3000, 10,000, 30,000, and 100,000 copies/mL; and CD38 was divided at 1250, 2500, 4000, and 7000 molecules per CD8+ T cell as recommended previously (10).
For greater resolution in the prognostic value of markers across time points, parametric survival models were then applied. Specifically, nested models using the formula:EQUATION
were fit to the data using commercial software (22), where T represents AIDS-free time and ε is assumed to be normally distributed with a mean of 0 and SD of σ. The lognormal model was selected because it leads to appealing interpretation of marker effects and because it has been shown to be a reasonable parametric model for the time to AIDS (23). Results were summarized in terms of likelihood values and χ2 statistics for the significance of marker values for prognosis at each time point.
Linear models were used to compare relations between HIV RNA, CD38, and CD4 cell counts early and late in the course of infection and according to therapy use. Simple random effects models (24) were applied to assess differences in average marker levels across the two MACS visits. Using the parametric survival models described previously, estimates of median AIDS-free time based on early and late marker values were summarized graphically for comparison across time points.
RESULTS
Descriptive Statistics
Table 1 summarizes the demographic and descriptive statistics in the cohort at the early and late time points. There were 191 men in the early time point group, and they were estimated to have been infected a median of 2.8 years. No therapy was available at the time the markers were measured, and 66.5% of men in this group remained treatment naïve during all 4 years of follow-up. Most who were treated received monotherapy with zidovudine. There were 252 men in the late time point group, and their median duration of infection was estimated at 8.7 years. At the time the markers were measured, 57.6% were untreated, whereas 28.0% reported monotherapy and 14.4% reported combination therapy of two nucleoside reverse-transcriptase inhibitors. During the 4 years of follow-up, 32.1% remained treatment naïve. Twenty-four participants contributed marker data at both the early and late time points. Estimated intraclass correlations were 0.50, 0.39, and 0.57 for CD4, CD38, and HIV RNA, respectively, indicating only moderate associations between marker values measured early and late in infection in the same individual.
At the later time point, CD4 values were lower (p < .001) and CD38 values were higher (p < .001) as expected, because the men had been infected for a longer time. Notably, the median viral loads at both time points were similar (p = .949). At the late time point, the correlation between each of the pairs of marker values was stronger than at the early point. This was especially marked for the correlation between CD4 count and CD38 expression on CD8+ T cells.
Relative Predictive Value of the Markers at the Early and Late Time Points
In Figure 1, Kaplan-Meier plots display time to AIDS development based on marker values at the early and late time points for CD4 count, plasma HIV RNA, and CD38 expression. Relative hazard estimates are displayed to indicate risk associated with subsequent marker categories over 4 years of follow-up; these can be compared according to infection duration, because a common set of cutoff points was used for each marker. As a comparison across time points of the discrimination provided by the various markers, we report χ2 statistics with corresponding p values in Table 2. Four years of follow-up were incorporated at both study time points (see Table 2), whereas in Figure 1, follow-up times were 10 and 4 years for the early and late time points, respectively. The χ2 values were determined from lognormal survival models that treat the markers as continuous variables. This was more attractive than separating the survival times in quartiles, because we also considered the predictive value of combinations of two or all three markers as well as the remaining predictive value of each marker when one or two others are known. For these models, age in years was adjusted for because it was a risk factor at the early time point. The χ2 values indicate that plasma viral load and CD38 expression were similarly predictive early on, whereas the predictive value of the CD4 count was markedly less than that of these other two markers. At the late time point, CD38 expression had a higher χ2 value than either the viral load or CD4 count when markers were analyzed individually, whereas the predictive value of the CD4 count was slightly greater than that of the viral load. Thus, the most notable difference between the time points was that the viral load has considerably greater predictive value than the CD4 count early but not later in infection as previously reported (4,9). Also, CD38 expression provided the greatest predictive value at both time points.
In further analyses, we quantified additional predictive values corresponding to adding variables to the model sequentially after one or two markers were already included; these results are summarized in the third and fifth columns of Table 2. At the early time point, including CD38 expression or viral load in a model with CD4 count as the initial predictor provided a substantial increase in information about the risk of AIDS. Conversely, the CD4 count added relatively little information when models already accounted for either viral load or CD38 expression and essentially no information when both of these other markers were included. CD38 expression and viral load also boosted prognostic value when added to a model that included the other, with or without controlling for CD4 count. At the late time point, CD38 expression and viral load each still contributed substantial information beyond that provided by CD4 count alone; however, a marked improvement in prognostic value was also now attributable to CD4 count when it was included in models already accounting for HIV RNA or CD38. When all three markers were considered in the same model at the late time point, each marker contributed a substantial and nearly equivalent amount of information on top of that provided by the other two.
Effect of Therapy on the Correlations Between Marker Values
In a preliminary analysis (see Table 1), we found that correlations between markers measured late in the course of infection were stronger than those between markers measured earlier. In addition, we noted that the viral load for a given CD4 count was lower at the late time point compared with the early time point (data not shown). Although these findings could stem from true differences in the pathobiology of the disease after a long and short duration, we also considered whether they might be a result of therapy use at the baseline visit by some of the subjects measured at the late time point.
Figure 2 suggests that therapy use did influence the correlations between the markers. In each panel, the solid dots represent observations made in the absence of therapy (i.e., combining the early data with late measurements under no therapy), whereas the open circles represent data obtained under therapy at the late time point. Regarding the association between HIV-1 RNA and CD4, coincidence of the three regression lines in Figure 2A is soundly rejected (p < .001). Further tests indicate that the lines are not parallel (p = .004) but that the hypothesis of equal intercepts is acceptable (p = .79). As is suggested in Figure 2A, the difference in slopes between the three lines is primarily a result of the steep rate of decline for group 3 (late, therapy); in fact, the lines for group 1 (early) and group 2 (late, no therapy) can be considered the same (p = .33 for test of coincidence). When tests were conducted to compare average HIV RNA levels at the twenty-fifth, fiftieth, and seventy-fifth percentiles (314, 460, and 620 cells/mm3, respectively) of the overall combined CD4 count distribution, all these tests confirmed that the late therapy group had lower levels of HIV-1 RNA than the no-therapy group or early group (p ≤ .001 in each case) for a given CD4 cell count.
Figure 2B presents the corresponding picture for CD38 versus CD4. In this case, the results of tests against coincidence, parallelism, and equal intercepts for the three regression lines were all highly significant (p < .001). The difference was minor between group 1 (early) and group 2 (late, no therapy;p = .053 for test of coincidence), however. Again, tests were done to compare CD38 levels at the various percentiles of the combined CD4+ distribution. At the twenty-fifth and seventy-fifth percentiles, the levels for group 1 and group 3 were significantly different (p = .008 and p < .001, respectively), although they were similar (p = .35) at the median CD4 count. Thus, as Figure 2B suggests, there was a tendency toward higher activation levels in the late therapy group relative to others given low CD4 cell counts, although the opposite was true given high CD4 cell counts.
Figure 2C displays HIV-1 RNA versus CD38 expression. The test for coincidence of the three lines was highly significant (p < .001), but parallelism was not rejected (p = .37). A test for equal intercepts assuming parallelism revealed that differences between lines were primarily the result of different levels as suggested visually (p < .001). Once again, little of this difference was caused by variation between group 1 and group 2 (p = .046 for coincidence), whereas most was caused by the lower average viral load for group 3 (late, therapy) at any given CD38 level. These correlation plots suggest that the similarity of the viral load measurements at the late time point, despite lower CD4+ counts and longer duration of infection, may be a result of the treatment of the group.
Analysis of AIDS-Free Time by Marker Value Taking Account of Therapy
Figure 3 presents a comparison of median AIDS-free times corresponding to ranges of marker values at the early and late visits, respectively. These estimates are based on parametric survival models of the type used to produce the results in Table 2, with marker values plotted on a logarithmic scale. Again, we have adjusted for age and have plotted the results corresponding to the overall (combined) median baseline age of 37.1 years. The plots illustrate that estimated median AIDS-free times associated with a given CD4 count or CD38 level tend to be similar at the early and late time points regardless of therapy. In contrast, the median AIDS-free time associated with a given HIV RNA level tends to be shorter at the late time point compared with the early time point, and this is exaggerated in the group that received therapy during the follow-up period subsequent to marker measurement at the late time point. Approximate 95% confidence interval bands for the curves in Figure 3 were constructed (data not shown), and these indicated no marked differences in median AIDS-free times across visits with the exception of a statistically significant difference given higher values (i.e., >30,000 copies/mL) of HIV RNA.
Predictive Markers in Individuals With Low Viral Load
To determine whether CD4 or CD38 marker levels could help to distinguish outcome in subjects with a low viral load, we performed a separate analysis of the 49 people at the early time point who had <3000 copies of HIV RNA in the Chiron branched DNA assay. Medians (and interquartile ranges) in these men were 636 (range: 405-810) CD4+ T cells/mm3, 3234 (range: 1824-5358) copies of HIV RNA by the Roche reverse-transcriptase-polymerase chain reaction assay, and 1452 (range: 876-1998) CD38 molecules per CD8+ T cell. After 10 years of follow-up, 33% of these men had developed clinical AIDS. Log-rank χ2 values for the markers, which were separated in three groups with group sizes dictated using the CD4 cutoff points of 500 and 750 CD4+ T cells, were 7.9 (p = .019) for CD4, 4.0 (p = .136) for HIV RNA, and 3.6 (p = .161) for CD38. Thus, among subjects with low levels of viral replication, the data indicated that CD4 cell counts provided more prognostic information than viral load or CD38 measurements.
DISCUSSION
Although studies involving immunologic, virologic, and/or immune activation markers have been performed in patients who are relatively newly infected with HIV and in those more advanced in the course of their disease, our investigation is unique in that it focuses on two such groups simultaneously. The data permit assessment of the relative prognostic power of HIV RNA, CD38 expression on CD8+ T cells, and CD4 cell counts both early and late in HIV infection, and thus a comparison of the findings according to infection duration.
The early data presented here are supportive of prior studies that found viral burden to be more predictive for development of AIDS than CD4 cell counts at early time points in the MACS (8), whereas the later resurgence of CD4 cell count predictive value is consistent with the findings of a previous study of markers and outcomes in long-term HIV-1 infection (9). In the latter study, CD4 cell counts as well as T-cell function measured in terms of proliferation in the presence of CD3 antibody were more predictive of AIDS development than viral burden in individuals in the Amsterdam cohort infected for at least 8 years. Further studies extended this conclusion, showing that both CD4 cell counts and soluble tumor necrosis factor receptor II, a serum marker of immune activation, were more predictive of AIDS development than viral burden in people with CD4 cell counts <200 cells/mm3 (25).
The data described here were collected after a longer duration of infection and provide additional support for the concept that immune activation plays an important role as a contributor to outcome in late HIV-1 disease (11,12). In a prior study in individuals who attained a CD4 cell count of <50 cells/mm3, T-lymphocyte activation (measured as CD38 expression on both CD4+ and CD8+ T cells) but not viral burden was strongly predictive of progression to death (12). These results are indicative of the importance of understanding the role of immune activation in HIV-1 disease.
A further aspect of the current study is its incorporation of therapy information in the cohort setting. In a prior report involving the same group of men at the late time point, it was found that antiretroviral drug use did not have an effect on the predictive value of marker measurements (11). Here, Figure 2 suggests that the primary effect of therapy was to lower HIV RNA for a given CD4 cell count and that the associations among markers (HIV RNA, CD4, and CD38) remained fairly consistent regardless of infection duration in the absence of therapy. Figure 3 indicates that median AIDS-free times were quite consistent early and late for given marker values under no therapy but that the prognosis was worse for a given value of HIV RNA among those late in infection who received therapy.
The similarity between the early group and the late group with no therapy in Figures 2 and 3 suggests potentially little in the way of a long-term survivor effect on either marker correlations or prognosis for a given marker level. Confounding by indication may influence the discrepancy indicated for HIV RNA in Figure 3, however, because sicker patients are more likely to seek and receive therapy aimed at decreasing the viral burden. It is interesting to note that little discrepancy in prognosis was seen between the treated and untreated groups when the other markers (CD4 cell count or CD38 expression) were fixed. The prognosis in the treated group may also suffer because of a relative lack of therapy effect in terms of improving nonvirologic marker values (e.g., CD4 count, CD38 expression), which have demonstrated strong prognostic significance even given the HIV RNA level late in the course of the disease.
In summary, our investigation supports prior studies that have found the HIV RNA level to be highly prognostic early in infection and a relative resurgence of the CD4 cell count to be a predictor of AIDS later in the course of infection. Importantly, we also found that CD38 expression on CD8+ T cells was an exceptionally strong predictive marker after both short and long duration of HIV infection. Further quantification of the extent to which this marker of immune activation contributes information above and beyond HIV RNA level and CD4 cell count is ongoing in the MACS.
Acknowledgments:
The MACS is funded by the National Institute of Allergy and Infectious Diseases, with additional supplemental funding from the National Cancer Institute: UO1-AI-35042, 5-MO1-RR-00722 (GCRC), UO1-AI-35043, UO1-AI-37984, UO1-AI-35039, UO1-AI-35040, UO1-AI-37613, and UO1-AI-35041. The authors gratefully acknowledge the work of Holly Bazmi in Pittsburgh and Andrew Kaplan in Los Angeles, who performed the plasma RNA assays. Data cited in this report were collected by the MACS with centers (Principal Investigators) at The Johns Hopkins University Bloomberg School of Public Health (Joseph Margolick, Alvaro Muñoz); Howard Brown Health Center and Northwestern University Medical School (John Phair); University of California, Los Angeles (Roger Detels, Beth Jamieson); and University of Pittsburgh (Charles Rinaldo). The MACS web site is located at http://www.statepi.jhsph.edu/macs/macs.html.
REFERENCES
1. Henrard DR, Phillips JF, Muenz LR, et al. Natural history of HIV-1 cell-free viremia. JAMA 1995; 274:554-8.
2. Katzenstein TL, Pedersen C, Nielsen C, et al. Longitudinal serum HIV RNA quantification: correlation to viral phenotype at seroconversion and clinical outcome. AIDS 1996; 10:167-73.
3. Pantaleo G, Graziosi C, Fauci AS. The immunopathogenesis of human immunodeficiency virus infection. N Engl J Med 1993; 328:327-35.
4. Bruisten SM, Frissen PH, Van Swieten P, et al. Prospective longitudinal analysis of viral load and surrogate markers in relation to clinical progression in HIV type 1-infected persons. AIDS Res Hum Retroviruses 1997; 13:327-35.
5. Vidal C, Miró JM, Romeu J, et al. Lack of evidence of a stable viral load set-point in early stage asymptomatic patients with chronic HIV-1 infection. AIDS 1998; 12:1285-9.
6. Lyles RH, Muñoz A, Yamashita TE, et al. Natural history of human immunodeficiency virus type 1 viremia after seroconversion and proximal to AIDS in a large cohort of homosexual men. Multicenter AIDS Cohort Study. J Infect Dis 2000; 181:872-80.
7. Engels EA, Rosenberg PS, O'Brien TR, et al. Plasma HIV viral load in patients with hemophilia and late-stage HIV disease: a measure of current immune suppression. Multicenter Hemophilia Cohort Study. Ann Intern Med 1999; 131:256-64.
8. Mellors JW, Muñoz A, Giorgi JV, et al. Plasma viral load and CD4+ lymphocytes as prognostic markers of HIV-1 infection. Ann Intern Med 1997; 126:946-54.
9. Spijkerman IJB, Prins M, Goudsmit J, et al. Early and late HIV-1 RNA level and its association with other markers and disease progression in long-term AIDS-free homosexual men. AIDS 1997; 11:1383-8.
10. Liu Z, Cumberland WG, Hultin LE, et al. Elevated CD38 antigen expression on CD8+ T cells is a stronger marker for the risk of chronic HIV disease progression to AIDS and death in the multicenter AIDS Cohort Study than CD4+ cell count, soluble immune activation markers or combinations of HLA-DR and CD38 expression. J Acquir Immune Defic Syndr Hum Retrovirol 1997; 16:83-92.
11. Liu Z, Cumberland WG, Hultin LE, et al. CD8+ T-lymphocyte activation in HIV-1 disease reflects an aspect of pathogenesis distinct from viral burden and immunodeficiency. J Acquir Immune Defic Syndr Hum Retrovirol 1998; 18:332-40.
12. Giorgi JV, Hultin LE, McKeating JA, et al. Shorter survival in advanced human immunodeficiency virus type 1 infection is more closely associated with T lymphocyte activation than with plasma virus burden or virus chemokine coreceptor usage. J Infect Dis 1999; 179:859-70.
13. Kaslow RA, Ostrow DG, Detels R, et al. The Multicenter AIDS Cohort Study: rationale, organization, and selected characteristics of the participants. Am J Epidemiol 1987; 126:310-8.
14. Detels R, Phair JP, Saah AJ, et al. Recent scientific contributions to understanding HIV/AIDS from the Multicenter AIDS Cohort Study. J Epidemiol 1992; 2(Suppl):S11-9.
15. Chmiel JS, Detels R, Kaslow RA, et al. Factors associated with prevalent human immunodeficiency virus (HIV) infection in the Multicenter AIDS Cohort Study. Am J Epidemiol 1987; 126:568-77.
16. Muñoz A, Taylor JMG, Chmiel JS, et al. Estimation of time since exposure for a prevalent cohort. Stat Med 1992; 11:939-52.
17. Giorgi JV, Cheng H-L, Margolick JB, et al. Quality control in the flow cytometric measurement of T-lymphocyte subsets: the Multicenter AIDS Cohort Study experience. Clin Immunol Immunopathol 1990; 55:173-86.
18. Schenker EL, Hultin LE, Bauer KD, et al. Evaluation of a dual-color flow cytometry immunophenotyping panel in a multicenter quality assurance program. Cytometry 1993; 14:307-17.
19. Hultin LE, Matud JL, Giorgi JV. Quantitation of CD38 activation antigen expression on CD8+ T cells in HIV-1 infection using CD4 expression on CD4+ T lymphocytes as a biologic calibrator. Cytometry 1998; 33:123-32.
20. Kirstein LM, Mellors JW, Rinaldo Jr, CR, et al. Effects of anticoagulant, processing delay, and assay method (branched DNA, reverse transcriptase PCR) on measurement of human immunodeficiency virus type 1 RNA levels in plasma. J Clin Microbiol 1999; 37:2428-33.
21. Cox DR. Regression models and life-tables. J R Stat Soc 1972; 34:187-220.
22. Allison PD. Survival analysis using the SAS system: a practical guide. Cary, NC: SAS Institute, 1995.
23. Muñoz A, Xu J. Models for the incubation of AIDS and variations according to age and period. Stat Med 1996; 15:2459-73.
24. Littell RC, Milliken GA, Stroup WW, et al. SAS system for mixed models. Cary, NC: SAS Institute, 1996.
25. Fahey JL, Taylor JMG, Manna B, et al. Prognostic significance of plasma markers of immune activation, HIV viral load and CD4 T-cell measurements. AIDS 1998; 12:1581-90.
© 2002 Lippincott Williams & Wilkins, Inc.