To determine if a marker is a valid surrogate for progression to a particular clinical end point and whether it will be useful in studies investigating anti-HIV therapies, it is not sufficient to show that a pretreatment marker value is associated with such progression. Rather, we need to show that changes in the marker over time are related to that progression and that the effect of a particular drug on this progression can be explained by its effect on the marker [5].

Recent work by DeGruttola and colleagues [6] has indicated that the absolute number of CD4⁺ lymphocytes in the peripheral blood (the CD4⁺ cell count) is not a complete surrogate marker for death in persons with AIDS or advanced AIDS-related complex. Their analyses of data from the original phase II zidovudine (3'-azido-3'-deoxythymidine, AZT) trial [7] and of data from Protocol 002 of the AIDS Clinical Trials Group [8] showed that only a small portion of zidovudine's effect on survival was statistically related to its effect on the CD4⁺ cell count.

Despite the limitations of these findings, their implications, coupled with the perceived similarity between zidovudine and didanosine (2',3'-dideoxyinosine, ddI), appear to have played an important role in the recent Food and Drug Administration (FDA) provisional approval of didanosine for use in persons who cannot tolerate zidovudine or who are not benefiting from zidovudine therapy. Data were available to show that didanosine delayed the decrease in CD4⁺ cell counts during the first 24 weeks of therapy in persons with symptomatic HIV infection, but the data did not show a direct clinical benefit. This use of a surrogate marker for drug approval marks a precedent in AIDS research and may well pave the way for other drug-approval requests based on the effects on surrogate markers.

However, the usefulness of the CD4⁺ cell count as a surrogate marker for death in persons with AIDS may have little bearing on its merit as a surrogate marker for the development of AIDS in persons with early HIV infection. Surrogate markers are more valuable in early disease populations than in more advanced disease populations. Also, it is not clear whether measurements of CD4⁺ lymphocytes other than the absolute count correlate better with clinical progression. We assessed the extent to which zidovudine's effect on the development of AIDS is statistically related to its effect on CD4⁺ lymphocyte levels in persons with asymptomatic HIV infection.

Methods

Top Methods Results Discussion Author & Article Info References

Patients

We analyzed data from the stratum of Protocol 019 of the AIDS Clinical Trials Group (ACTG 019) that enrolled asymptomatic HIV-infected men and nonpregnant women (ages 18 years or older with 500 or fewer CD4⁺ cells/mm³) to receive placebo, a 500-mg total daily dose of zidovudine, or a 1500-mg total daily dose of zidovudine. The primary results of this stratum of ACTG 019 have been reported elsewhere [9]. For the purpose of this report, all analyses were based on data available through March 1989, before prophylaxis against Pneumocystis carinii pneumonia, a leading AIDS-defining event, was widely permitted. Furthermore, only patients in whom at least one post-baseline CD4⁺ cell count had been taken were included in this study. The data from the two zidovudine dose groups were pooled because they were similar with respect to pretreatment characteristics and rates of progression to AIDS [9] and because few such progressions occurred in either arm alone.

Laboratory Studies

Blood samples were collected at enrollment and at follow-up visits at weeks 8, 16, 32, 48, 64, and 80 by the institutions participating in the trial. The CD4⁺ cell count was calculated as the product of the leukocyte count, the percentage of lymphocytes in all leukocytes (percent lymphocytes), and the percentage of lymphocytes that were CD4⁺ cells (CD4⁺ percent). The leukocyte count and the differential count were measured by standard automated techniques. The CD4⁺ percent was obtained by flow cytometry. Interlaboratory results were monitored by ACTG quality-control procedures [10].

Statistical Analysis

The analyses in this report were based on an intention-to-treat approach. Estimates of distributions of the time-to-progression to AIDS were obtained by using the Kaplan-Meier method [11]. For assessing statistical associations among zidovudine, CD4⁺ lymphocyte levels, and progression to AIDS, the Cox regression [12] model was fitted with the current CD4⁺ lymphocyte values, the treatment assignment, or both. The current CD4⁺ cell count was the most recently measured preceding value. The amount of the zidovudine effect on progression to AIDS that could be explained by its effect on CD4⁺ lymphocyte levels was assessed by the proportional reduction in the regression coefficient for the treatment assignment after controlling for changes in the CD4⁺ lymphocyte levels. Separate analyses and analyses stratified by treatment were also made when appropriate. Missing marker values were imputed from trajectories of the most recent values by using a first-order autoregressive model [13] with time-varying autocorrelation coefficients. All reported P values are two sided.

Results

Top Methods Results Discussion Author & Article Info References

Effect of Zidovudine on Progression to AIDS and on CD4⁺ Lymphocytes

A total of 1075 patients were assessable: 350 in the placebo group and 725 in the zidovudine groups. After a maximum follow-up of 90 weeks (median, 55 weeks), 44 patients had progressed to AIDS: 24 in the placebo group and 20 in the zidovudine groups (log rank; P = 0.04; (Figure 1). The overall progression rates were 6.4 [placebo] and 3.2 (zidovudine) per 100 person-years.

View larger version (16K): [in this window] [in a new window]   Figure 1. AIDS-free probabilities for the placebo group and for the zidovudine groups over time. The numbers of at-risk patients given placebo/zidovudine are 350/725 (week 8), 322/640 (week 16), 271/445 (week 32), 232/320 (week 48), 166/226 (week 64), and 56/76 (week 80).

Lymphocyte values were not available for 3% of patients at week 8, 8% at week 16, 14% at week 32, and 23% to 25% thereafter. A summary of the effects of zidovudine on the CD4⁺ cell count is shown in (Figure 2). The effect of zidovudine on this marker appears to be transient, with an increase in the count for approximately 8 weeks followed by a slow decrease at approximately the same rate as that in patients given placebo. An opposite, adverse effect of zidovudine on the leukocyte count occurred that attenuated the drug's overall effects on the CD4⁺ cell count. In patients given zidovudine, a decrease occurred in the leukocyte count during the first 16 weeks (average loss of 60 cells/mm³ per week for the first 8 weeks and 30 cells/mm³ per week for the next 8 weeks); the leukocyte count was stable thereafter. The leukocyte count in patients given placebo was relatively stable throughout the observation period. Zidovudine's effect on percent lymphocytes and CD4⁺ percent was similar to its effect on the CD4⁺ cell count.

View larger version (17K): [in this window] [in a new window]   Figure 2. Cross-sectional medians of the CD4+ cell counts and the component markers in the placebo group and in the zidovudine groups.

Association between Markers and Progression to AIDS

Stratified (by treatment) Cox regression showed that the baseline values of the CD4⁺ cell count, leukocyte count, and CD4⁺ percent among the component markers and the net CD4⁺ percent (percentage of CD4⁺ lymphocytes among all leukocytes) were significantly correlated with the risk for clinical progression (Table 1). For example, a smaller baseline CD4⁺ cell count (50 cells/mm3) corresponded to a 36% higher risk for progression. When a patient's current markers were used in Cox regression with time-dependent covariates, the CD4⁺ cell count and all of its components were significantly correlated with the risk for progression (Table 1). Furthermore, after the current marker values had been accounted for, the baseline values were no longer significant.

CD8⁺ lymphocytes, either the absolute count or the percentage, had no significant value in predicting the progression to AIDS in this HIV-infected but asymptomatic patients (data not shown). Also, no significant effect of zidovudine on CD8⁺ lymphocyte levels was observed.

Effect of Zidovudine on Progression to AIDS that Is Explainable by Changes in CD4⁺ Lymphocyte Levels

If the CD4⁺ cell count were a complete surrogate marker for progression to AIDS, no residual association between treatment and progression would exist after we controlled for the current CD4⁺ cell count [5]. However, as seen in Table 2, zidovudine was still significantly correlated with progression after adjustment for the current CD4⁺ cell count. Additionally, the resulting zidovudine-placebo relative risk was virtually identical to the relative risk of 2.10 obtained if we did not account for the current CD4⁺ cell count. In contrast, the net CD4⁺ percent accounted for 37% of zidovudine's effect on progression. In either case, patients given zidovudine had a lower risk for clinical progression than did those given placebo, after we controlled for the current CD4⁺ cell count. This implies that the relation between the current CD4⁺ cell count and the risk for progression differed in the two treatment groups.

To quantify this difference, we fitted a time-dependent Cox regression in which the logarithm of the risk for clinical progression to AIDS was linearly related to the current CD4⁺ cell count but allowed a different linear relation in the two groups. The logarithm of the risk for progression was proportional to the current CD4⁺ cell count (the proportionality was –0.0095 ± 0.0019 for the placebo group and –0.0140 ± 0.0024 for the zidovudine groups). Put another way, the risk for progression of a patient in the zidovudine groups with a particular (current) CD4⁺ cell count was approximately the same as that for a patient in the placebo group with a 50% (that is, [0.014 – 0.0095]/0.0095) greater CD4⁺ cell count.

Of the other expressions of CD4⁺ lymphocyte levels we examined, the net CD4⁺ percent showed the least differential association with progression to AIDS, but it was still 25% greater in the zidovudine groups than in the placebo group. Separate or stratified analyses by treatment confirmed this relative homogeneity in the association between the net CD4⁺ percent and progression.

As noted above, the incompleteness of CD4⁺ lymphocyte levels as a surrogate marker was reflected by a residual treatment difference after we controlled for current values of the CD4⁺ lymphocyte levels. We therefore examined the time dependency of this residual effect. As shown in Table 3, the effect of zidovudine on progression to AIDS appeared to be substantially greater during the first 16 weeks of therapy. Of 322 patients in the placebo group and 640 in the zidovudine groups who had not progressed after 16 weeks in the study, 19 in each group subsequently progressed to AIDS. As shown in Figure 3, a zidovudine effect still remained after week 16.

View larger version (15K): [in this window] [in a new window]   Figure 3. AIDS-free probabilities of patients in the study for at least 16 weeks. Refer to Figure 1 for the numbers of at-risk patients.

By week 16, significant differences (P = 0.01) in markers had developed between the treatment groups; for example, the median CD4⁺ cell count was 341/mm³ in the placebo group and 376/mm³ in the zidovudine groups. Table 4 shows that 46% of zidovudine's subsequent effect was explainable by the higher CD4⁺ cell counts in the zidovudine groups at that time. Including CD4⁺ cell counts beyond week 16 did not make any notable improvement. A stronger relation was observed for net CD4⁺ percent, which explained 74% of zidovudine's effect on subsequent progression to AIDS. Figure 4 also shows the estimated extent of zidovudine's benefit on clinical progression that would be explained by its effect on net CD4⁺ percent. When patients with missing marker values at week 16 were excluded, virtually all of the effect of zidovudine on subsequent clinical progression was statistically associated with the net CD4⁺ percent at week 16.

View larger version (14K): [in this window] [in a new window]   Figure 4. Model-based AIDS-free probabilities after 16 weeks of treatment. The solid line and the short dashed line are the distributions predicted by the net CD4+ percent at week 16 and by treatment, corresponding to the nonparametric estimates in Figure3. The long dashed line represents the portion of zidovudine's effect on subsequent progression to AIDS that can be predicted solely by its effect on net CD4+ percent at week 16. (The AIDS-free time distribution for each patient was projected by evaluating the fitted Cox regression model with the patient's net CD4+ percent at week 16, and with or without the unexplained benefit of zidovudine. These estimates were then averaged by treatment.).

Discussion

Top Methods Results Discussion Author & Article Info References

CD4⁺ Lymphocytes as Surrogate Markers

We assessed the value of CD4⁺ lymphocyte levels as a surrogate marker for progression to AIDS in persons with asymptomatic HIV infection, using data from the ACTG 019 trial [9]. This trial found that treatment with zidovudine delays progression to AIDS and slows the decline in the CD4⁺ cell count and that the baseline CD4⁺ cell count is correlated with progression to AIDS.

We examined the changing values of the patients' CD4⁺ lymphocyte levels over time and found that the current CD4⁺ cell count, as well as its components, was significantly correlated with the risk for progression to AIDS. In particular, the percentage of CD4⁺ cells among all leukocytes (or net CD4⁺ percent) was a potent predictor of progression to AIDS. Furthermore, after we had accounted for a patient's current marker values, the baseline values for the patient were no longer correlated with the risk for progression. These results reinforce the importance of CD4⁺ lymphocyte levels as a prognostic factor for clinical progression.

For CD4⁺ lymphocyte levels to satisfy the criteria for a surrogate marker, however, they must also explain zidovudine's effect on progression [5]. Our analyses showed that zidovudine's effect on CD4⁺ lymphocyte levels explains only a portion of its effect on progression: Even after we controlled for current CD4⁺ lymphocyte levels, patients given zidovudine had a lower risk for progression than those given placebo. In this sense, CD4⁺ lymphocyte levels were only a partial surrogate marker for clinical progression in this group of persons with early HIV infection.

These analyses also suggested that net CD4⁺ percent was a better partial surrogate marker than the absolute CD4⁺ cell count, even though, from a physiologic point of view, the latter would seem a better one. This may be due, in part, to the fact that the CD4⁺ cell count was obtained by multiplying the net CD4⁺ percent by the leukocyte count, and the leukocyte count was lowered by the administration of zidovudine. In addition, the leukocyte count was associated with considerable intrapatient, interpatient, and intralaboratory variation; hence, the possible measurement error associated with the net CD4⁺ percent was less than that from the CD4⁺ cell count.

The remaining, unexplained part of the effect of zidovudine on progression to AIDS could be attributed to several factors. First, zidovudine may also affect progression by other mechanisms, such as altering the functions of CD4⁺ lymphocytes or other cells that contribute to cell-mediated immune responses. For example, in patients with advanced disease, zidovudine was associated with a restoration of delayed cutaneous hypersensitivity reactions [7]; improvements in -interferon production, natural killer cell activity, and some virus-specific T-lymphocyte responses [14]; and greater antibody responses to pneumococcal vaccine [15]. If we had observed an effect of zidovudine on both CD4⁺ lymphocyte levels and these or some other factors, a larger portion of its effect on clinical progression might have been accounted for statistically. Also, CD4⁺ lymphocyte levels in the peripheral blood account for only a small portion of all the CD4⁺ lymphocytes in the body. Thus, zidovudine might affect cells in the peripheral blood differently than those in other compartments, such as the lymph nodes and the spleen. In addition, the inability to accurately measure the levels of CD4⁺ lymphocytes could also contribute to the finding that CD4⁺ lymphocyte levels are incomplete surrogate markers.

A large disparity between the two treatment groups was observed in the degree of association between the CD4⁺ cell count and the progression to AIDS. These results suggest that each CD4⁺ lymphocyte was 50% more active in delaying progression to AIDS in the presence of zidovudine. For example, the risk for progression of a zidovudine-treated patient with 200 CD4⁺ cells/mm³ was approximately the same as the risk for an untreated person with 300 CD4⁺ cells/mm³. The superiority of net CD4⁺ percent as a surrogate marker was partly due to the relative homogeneity in its association with progression between treatment groups; it is possible that net CD4⁺ percent reflected more of the zidovudine-induced improvement in cellular function.

Further analyses showed that the component of the zidovudine effect on progression that was not statistically associated with its effect on CD4⁺ lymphocyte levels seemed to occur during the first 16 weeks of therapy, suggesting that zidovudine might have a spontaneous effect on progression by delaying some imminent AIDS-defining events. Thus, the beneficial effect of zidovudine does not fully correlate with CD4⁺ lymphocyte levels because there is a large but transient positive effect that occurred before zidovudine altered the CD4⁺ lymphocyte levels. By focusing on the progression to AIDS after 16 weeks of therapy, we found that most of the beneficial effects of zidovudine could be explained statistically by the improved net CD4⁺ percent profile at week 16. The results suggest that zidovudine's overall beneficial effect on progression might be characterized by its observed clinical benefit during the first 16 weeks of therapy and by the improved net CD4⁺ percent at week 16.

Limitations

As with other secondary analyses, this investigation had various limitations. Because the original study was terminated early due to significant evidence of zidovudine's benefit, our analyses were based on rather small numbers of events and thus lack statistical confidence. Furthermore, although our approach for assessment of whether CD4⁺ lymphocyte levels are a complete surrogate marker was known in advance, our consideration of the value of CD4⁺ lymphocyte levels after week 16 and the choice of net CD4⁺ percent as a marker were suggested by the data. Thus, our findings are speculative until they can be further investigated. Also, the superiority of the net CD4⁺ percent compared with the CD4⁺ cell count as a surrogate marker may not be accurate. In any case, a marker that is a variation of the CD4⁺ cell count is not likely to incorporate all of the beneficial effect of zidovudine on preventing clinical progression to AIDS.

Numerous other models and methods could have been used for imputing missing marker values, possibly leading to different conclusions. We used observed values rather than model-based fitted values as covariates; this simplifies the analysis. However, it may have diluted the link between markers and progression to AIDS if excessive measurement errors occurred in the observed marker values. We did not study the influence from imprecise imputation; however, because only 8% of the data at week 16 were imputed, it is unlikely that our method for imputation substantially affected these findings. Similarly, differences between the participating institutions in ACTG 019 with respect to the evaluation of CD4⁺ lymphocyte level subsets could have diluted the effect of CD4⁺ lymphocyte levels on progression to AIDS.

We were also unable to assess simultaneously the potential value of other laboratory markers, such as neopterin and ß₂-µglobulin, which have been found to correlate with disease progression [16, 17]. Such markers individually, or with CD4⁺ lymphocyte levels, could have been used as surrogate markers for immune system function. These other markers may have value as surrogate markers, as recently reported [16, 18], but their relation to the risk for clinical progression in both untreated and treated persons has not been reported in a controlled trial.

Considering the complex nature of HIV and its effect on the immune system, it is not surprising that zidovudine's effect on clinical progression to AIDS could not be explained by its effect on a single biological marker. Our results suggest that CD4⁺ lymphocyte levels should not be abandoned as a surrogate marker, but rather that additional markers should be identified that, when combined with CD4⁺ lymphocyte levels, can explain all or most of zidovudine's benefit on progression.

Finally, all of our analyses were based on a single antiretroviral agent, zidovudine, whose effect on progression to AIDS is well known. The value of studies validating surrogate markers is in using those markers to assess the clinical efficacy of other drugs. Yet, even if the results reported here could be reproduced in a similar group of patients treated with zidovudine, concluding that the net CD4⁺ percent at week 16 would fully characterize the subsequent clinical benefit of any other drug would be a leap of faith. It is clear that our results might not apply to drugs whose mechanisms of action are fundamentally different from those of zidovudine. They may not apply to similar drugs, such as other nucleoside analogs. Unfortunately, this will always be the case in validation studies of candidate surrogate markers. We are not implying that these markers should not be used for predicting long-term efficacy based on short-term results but are cautioning that such predictions may not be accurate.

Appendix

The participating institutions included Albert Einstein/Montefiore Medical Center, Queens, New York; Baylor College of Medicine, Houston, Texas; University of California, Los Angeles; University of California, San Diego; University of California, San Francisco; Case Western Reserve University; University of Cincinnati; Cornell University, New York, New York; Duke University; George Washington University; Harvard University and Massachusetts General Hospital, Boston, Massachusetts; Indiana University; Johns Hopkins University, Baltimore, Maryland; University of Massachusetts; Memorial Sloan-Kettering Cancer Center, New York, New York; University of Miami, Miami, Florida; University of Minnesota; Mount Sinai Hospital, New York, New York; New York University; State University of New York, Stony Brook; University of North Carolina, Chapel Hill; Northwestern University; Ohio State University; Pennsylvania State University and Hershey Medical Center; University of Pittsburgh, Pittsburgh, Pennsylvania; Robert Wood Johnson Medical School; University of Rochester, Rochester, New York; Saint Luke's-Roosevelt Hospital; University of Southern California; Stanford University; Tulane University; University of Washington, Seattle; Washington University.