The acyclic nucleoside acyclovir has been shown to be highly effective and well tolerated in the treatment of varicella and in the treatment and prophylaxis of herpes simplex virus infection [3, 4]. Although acyclovir has no clinical role in the treatment of cytomegalovirus infection, high oral doses (>3.2 g/d) decrease the development of symptomatic cytomegalovirus infections in renal and bone marrow transplant recipients who are positive for cytomegalovirus antibody [3]. Acyclovir alone has no activity against HIV-1, but in vitro studies have suggested that the drug acts synergistically when combined with zidovudine at concentrations between 2 and 200 µmol/L [5, 6]; however, this finding could not be confirmed by other investigators [7]. In early clinical trials, acyclovir was given in doses higher than 3.2 g/d in an attempt to achieve serum concentrations that were predicted to be potentially synergistic with zidovudine, but this combination therapy showed no evidence of increased antiviral, immunologic, or clinical benefit when compared with zidovudine alone [8-10]. No pharmacologic interactions between acyclovir and zidovudine or apparent increase in toxicity were shown [11]. However, the power of these trials to find a benefit for acyclovir was limited by their small sample size and short follow-up. Subsequently, using 12-month follow-up data from patients who had participated in their earlier randomized, double-blind, 6-month trial, Cooper and colleagues [12] reported that acyclovir (3.2 g/d) significantly interacted with zidovudine to prolong survival. However, after adjustment for baseline prognostic factors, the finding was no longer statistically significant. They found no effect of the combination therapy on progression to AIDS or on development of clinical cytomegalovirus disease.

Similar findings were reported from the Wellcome-United Kingdom trial [13-15], which was stopped in December 1991 because high-dose acyclovir was showing no effect on cytomegalovirus disease. However, patients receiving acyclovir plus nonblinded antiretroviral therapy had significantly prolonged survival when compared with those receiving nonblinded antiretroviral therapy alone (P = 0.035). In contrast to earlier trials, the preliminary findings of the Zidovudine/Acyclovir Collaborative Group study [16] indicated a trend toward a greater CD4 lymphocyte cell count response among patients receiving acyclovir plus zidovudine compared with those receiving zidovudine alone, but no clinical data were reported. These suggestive [12-15, 17] but inconclusive data on the possible survival benefit of acyclovir have led to such an increased use that acyclovir's manufacturer, Burroughs-Wellcome, has established a financial assistance program for recipients of the drug.

To determine whether acyclovir plus zidovudine has a beneficial effect on progression to AIDS, development of clinically apparent cytomegalovirus infection, or survival when compared with zidovudine alone, we analyzed data from the Multicenter AIDS Cohort Study, a large prospective epidemiologic study of homosexual men. Along with other information, clinical outcome and laboratory data on participants who used zidovudine with and without acyclovir have been gathered at regular 6-month intervals. Because assignment to treatment group is not randomized in such an observational study, care was taken to control for potential biases introduced by differential disease stage, use of other therapies, and temporal trends in survival between treatment groups.

Methods

Top Methods Results Discussion Author & Article Info References

Patients and Data Collection

The design of the Multicenter AIDS Cohort Study has been previously described [18]. According to the available data from the study, of the 2633 participants who were seropositive at entry or who later seroconverted, 969 began zidovudine therapy before a diagnosis of AIDS was made (using the 1987 Centers for Disease Control and Prevention [CDC] case definition [19]). Among this group, there were 786 participants for whom complete data were available and who had survived to at least the next visit. Of these men, 515 subsequently received acyclovir during the follow-up period. Patient-reported acyclovir use was classified on the study data forms based on the responses to two questions. Participants were asked if they had "used any medication for health reasons not related to AIDS or if they had taken any medication to help fight AIDS or the HIV virus." Acyclovir (Zovirax, Burroughs-Wellcome, Research Triangle Park, North Carolina) was one of the medications in the list under each question. The group that used acyclovir for any indication (n = 488) consisted of those who indicated use of acyclovir as a response to either or both questions, whereas the group that used acyclovir for AIDS or HIV infection (n = 242) indicated use of acyclovir only as a response to the latter question. It was hypothesized that the group using acyclovir for any indication represented a population with a broad range of doses, administration, and duration of use similar to what might be expected in overall clinical practice. The group using acyclovir for AIDS or HIV infection was hypothesized to represent a population that might better test the effects of acyclovir because they received possibly higher and less intermittent doses than did patients who used acyclovir for any indication. These hypotheses on dose, administration, and chronicity were subsequently formally investigated.

The use of acyclovir for AIDS or HIV infection was first reported at visit 11 (1989), whereas the use of acyclovir for any indication was first recorded at visit 6 (1986). Only acyclovir used in combination with antiretroviral therapy has been analyzed in this study. Participants had their HIV antibody status determined by standard HIV enzyme-linked immunosorbent assay antibody testing with Western blot confirmation. The CD4⁺ T-lymphocyte enumeration by flow cytometry was done according to standard methods and was quality-controlled [20]. Essentially all Multicenter AIDS Cohort Study participants are positive for cytomegalovirus antibody at entry, as determined by fluorescent antibody testing. Limited data are available on the seroprevalence of herpes simplex type 1 or 2 infection in Multicenter AIDS Cohort Study participants from a sample at the Pittsburgh Multicenter AIDS Cohort Study site (seroprevalence, 90%, n = 184) (Kingsley L. Personal Communication). Data from other cohorts of homosexual men would suggest a similar seroprevalence [21-24].

Statistical Analysis

Baseline data were compared for differences using the t-test [25]. Separate multivariate Cox proportional-hazard models were fit for outcomes of AIDS, cytomegalovirus, or survival using SAS software (SAS Institute, Cary, North Carolina [25]). Baseline risk factors for disease progression (covariates) ascertained at the last visit before first reported use of zidovudine were platelet count (continuous, per 25 x 10⁹/L increment) and hemoglobin concentration (continuous, per 10 g/L increment). Initial models also included neopterin concentrations (continuous, per 1 nmol/L increment) and ß₂-µglobulin concentrations (continuous, per 1 mg/L increment) as baseline covariates, but they did not confound the relation between acyclovir and survival and were therefore excluded from the final parsimonious models. Time-dependent covariates ascertained at 6-month intervals from (and including) the time of first reported use of zidovudine were the following: CD4 lymphocyte count (continuous, 100 cells/µL increment), presence of HIV-associated symptoms (dichotomous, yes or no), Pneumocystis carinii prophylaxis (dichotomous, yes or no), reported clinical herpes episodes (dichotomous, yes or no), use of other antiretroviral therapy (didanosine or zalcitabine; dichotomous, yes or no), and acyclovir use (dichotomous, yes or no). Symptoms of HIV were defined as development of herpes zoster, unintentional weight loss of 4.5 kg or more, or any of the following if unexplained and present for 2 weeks or longer: diarrhea, temperature of 37.8 °C or greater, fatigue, and oral candidiasis. Continuous covariates were compared from higher to lower count increments. Dichotomous covariates were compared as yes or present as opposed to no or absent. The assumption of proportional hazards in the models was formally tested and was found to be valid. No significant interaction was found between acyclovir use and herpes episodes in the models. For development of the outcome being studied, a relative hazard calculated by the model in comparison of the groups would be 1.0 if there was no difference in risk, greater than 1.0 for an increase in risk, and less than 1.0 for a decrease in risk.

Acyclovir use was modeled as a time-dependent covariate for two reasons. First, within the cohort that received zidovudine, acyclovir therapy was started at varying calendar times and disease stages. The time-dependent model therefore allowed us to compare men who received acyclovir plus zidovudine with men who received zidovudine alone and who survived to the same time point, thus eliminating potential survival biases (also called onset confounding). Second, time-dependent modeling of CD4 count, HIV-related symptoms, and use of other drugs (such as antiretroviral and prophylactic agents) allowed adjustment for disease stage at the same time as the treatment groups were compared.

Use of acyclovir and zidovudine was modeled as "intent to treat," that is, once a participant reported use of the drug, subsequent use of such therapy was presumed until death or the last visit for which data were available. This type of analysis is conservative because it ignores changes in therapy that occur after initiation of therapy, which may bias estimates of effectiveness, and because it attenuates the ability to discriminate outcome differences between groups.

Three analyses of the constancy of acyclovir use were done to determine if differences in chronicity of use were associated with differential survival. Patients who reported any use of acyclovir for either three or more or four or more consecutive visits were compared over time with participants receiving zidovudine who did not meet these criteria in two separate time-dependent Cox models. In the third Cox model, the cumulative number of visits at which a participant reported any use of acyclovir was modeled as a time-dependent covariate. The three Cox models were adjusted for the previously identified covariates. These analyses are potentially biased against finding a difference between the groups because, over time, the same participants can be part of either of the two groups being compared.

The issue of timing the initiation of acyclovir therapy either for any indication or for HIV infection was analyzed in two separate Cox models adjusted as described previously. The participants who began receiving acyclovir before the diagnosis of an AIDS-defining condition were compared with participants who began receiving acyclovir after clinical AIDS was diagnosed.

Time-dependent multivariate Cox models provided an unbiased estimate of the relative survival risk (or hazard) associated with acyclovir use but do not allow an estimate of the absolute difference in the survival time. Therefore, landmark analyses were done to address this question. In these analyses, participants must have started zidovudine and survived to one of the two landmark time points (1 year after initiation of zidovudine therapy; development of a CD4 count less than 50 cells/µL or AIDS) that were chosen because they represented an early and a late event after starting antiretroviral therapy. Because the populations differ over time and because of selection biases, we adjusted for the same prognostic risk factors (covariates) as in the earlier time-dependent Cox models. However, no adjustment was made at subsequent time points to plot estimated survival curves. To adjust for these prognostic risk factors, the mean value of the continuous covariates at the time of the landmark were used and the dichotomous covariates were set to discrete values. The participants who used acyclovir at or before the landmark were compared with two groups: those who either never initiated acyclovir therapy or started it after the landmark or those who never initiated acyclovir therapy. The former group represents a conservative comparison group because it contains men who received acyclovir after the landmark and therefore has a tendency toward better survival, whereas the latter group is less conservative because only men who never started acyclovir therapy are included. The true or unbiased estimate of the absolute survival time in men not receiving acyclovir at the landmark is probably somewhere between the estimates of these comparison groups.

Results

Top Methods Results Discussion Author & Article Info References

Patients reporting use of acyclovir at any semi-annual study visit were statistically more likely to be receiving zidovudine, to have lower CD4 counts, and to have more symptoms at each visit than those not receiving acyclovir (data not shown). However, at baseline (before the initiation of zidovudine therapy), patients who used both acyclovir and zidovudine did not significantly differ from those who continued to receive zidovudine alone except for the higher incidence of a history of herpetic episodes (P < 0.001 for either indication) and the presence of a higher percentage reporting HIV-related symptoms (P = 0.01 for acyclovir use for any indication; P = 0.17 for acyclovir use for HIV infection). Baseline CD4 lymphocyte cell counts differed between those reporting acyclovir use for HIV infection and those receiving zidovudine alone (P < 0.001) (Table 1).

Models that included neopterin and ß₂-µglobulin levels as baseline covariates for the end points of AIDS, cytomegalovirus disease, and death yielded results similar to those that did not but resulted in a smaller study population secondary to missing baseline values. Therefore, only the results of the larger, more parsimonious, and potentially less selective model are shown.

Progression to AIDS and Cytomegalovirus Disease

Multivariate Cox modeling of time to progression to AIDS (1987 CDC case definition) did not show that the use of acyclovir either as medication for HIV infection or as therapy for any indication significantly affects the time to progression to AIDS (Figure 1). In these models, the CD4 cell count (relative hazard, 0.45; P < 0.001), presence of HIV-associated symptoms (relative hazard, 1.93; P < 0.001), baseline hemoglobin concentration (relative hazard, 0.87; P = 0.004), and platelet count (relative hazard, 0.92; P = 0.003) all predicted progression to AIDS, which is consistent with data from other studies [26]. Finally, a Cox model was run using diagnosis of cytomegalovirus disease as the outcome variable. This analysis, which adjusted for the same baseline and time-dependent covariates as described above, did not show a protective effect of acyclovir when used for either indication on the development of clinically apparent cytomegalovirus infection (P > 0.5) (Figure 1). However, CD4 cell count (relative hazard, 0.37; P < 0.001), presence of HIV-associated symptoms (relative hazard, 1.91; P = 0.01), and baseline hemoglobin concentration (relative hazard, 0.84; P = 0.04) all predicted the development of cytomegalovirus disease.

View larger version (17K): [in this window] [in a new window]   Figure 1. Relative hazards with 95% CIs from time-dependent multivariate Cox models. The relative hazard for cumulative acyclovir use (\#9671;) reflects the modeling of acyclovir use as a continuous variable in which the relative hazard refers to the per visit increase (6-month intervals on drug). All other relative hazards reflect the modeling of acyclovir use as a dichotomous variable. A relative hazard of 1.0 suggests no difference; a relative hazard greater than 1.0, an increased risk; and a relative hazard less than 1.0, a decreased risk for the group using acyclovir and the outcome listed. The three analyses of constancy of acyclovir use were based on patients who had used acyclovir for any indication. For a description of the time-dependent model, the covariates used, and the populations, see Methods. CMV = cytomegalovirus disease; AIDS = the acquired immunodeficiency syndrome; AIDS death = the time from the development of an AIDS-defining condition to death.

Survival

We examined the effect of acyclovir on survival in men who were receiving zidovudine Table 2 and Figure 1. When compared with zidovudine alone, the use of acyclovir with zidovudine was associated with a 26% decreased probability of death when acyclovir was used for any indication (relative hazard, 0.74; 95% CI, 0.53 to 1.03; P = 0.07) and with a 36% decreased probability when used for HIV infection (relative hazard, 0.64; CI, 0.45 to 0.91; P = 0.01). In the same model, CD4 cell count, presence of HIV-related symptoms, and baseline hemoglobin concentration all predicted survival. The use of P. carinii prophylaxis and the use of other antiretroviral therapy showed a trend toward being predictive of survival. Reported symptomatic herpes episodes in the two models were associated with a trend toward shortened survival (relative hazard, 1.46; CI, 1.04 to 2.05; P = 0.03; relative hazard, 1.35; CI, 0.98 to 1.88; P = 0.07, respectively).

Survival after Development of AIDS

Our results indicate that the prolongation of survival appeared to occur during the time from development of AIDS to death because time to development of AIDS was not affected by the use of acyclovir plus zidovudine. As shown in Table 3, initiation of acyclovir therapy after a diagnosis of AIDS was significantly associated with a 44% decreased probability of death (use of acyclovir for any indication: relative hazard, 0.56; CI, 0.37 to 0.86; P = 0.007; use of acyclovir for HIV infection: relative hazard, 0.57; CI, 0.38 to 0.84; P = 0.005), whereas initiation of acyclovir therapy before a diagnosis of AIDS was not associated with better survival.

Landmark Analyses of Comparative Survival

Models that include time-dependent covariates cannot be used to calculate the absolute differences in survival between treatment groups; landmark analyses were therefore done to further investigate the prolongation of survival. The two landmark time points were 1 year after initiation of zidovudine therapy and the development of either a CD4 lymphocyte count of less than 50 cells/µL or AIDS. As shown in Figure 2, multivariate Cox models, based on data at the time of the landmark, are used to estimate survival curves for three groups: patients receiving acyclovir at or before the landmark, those who either never started acyclovir therapy or started it after the landmark, and those never receiving acyclovir. In the analysis that used a landmark of 1 year after starting zidovudine therapy, the estimated 2-year survival proportions after the landmark were 93.9%, 89.8%, and 86.2% for the group that used acyclovir at or before the landmark (n = 131), the group that never used acyclovir or used it after the landmark (n = 502), and the group that never used acyclovir (n = 351), respectively. The corresponding 90% survival times for the three groups were 1325 days, 1059 days (relative risk, 0.58; P = 0.06), and 982 days (relative risk, 0.46; P = 0.009). Based on the landmark of developing either a CD4 count less than 50 cells/µL or AIDS, the estimated 1- and 2-year survival proportions after the landmark were 90.7%, 84%, and 71.5% and 69.8%, 52.6%, and 31.4% for the group that used acyclovir at or before the landmark (n = 85), the group that never used acyclovir or used it after the landmark (n = 203), and the group that never used acyclovir (n = 154), respectively. The corresponding 90% survival times were 398 days, 261 days (relative risk, 0.56; P = 0.01), and 176 days (relative risk, 0.51; P = 0.004). Median survival times for the landmark of developing a CD4 count of less than 50 cells/µL or AIDS for the three treatment groups were 1018, 745, and 544 days, respectively.

View larger version (19K): [in this window] [in a new window]   Figure 2. Landmark analyses of estimated survival. Multivariate Cox models that used the mean value of the continuous covariates and set the dichotomous covariates to discrete values at the time of the landmark point were constructed to give estimated survival curves. Left. Estimated survival curves using a landmark of 1 year after initiation of zidovudine therapy. The curves are estimated for a patient with symptomatic HIV infection who is receiving zidovudine but had no other antiretroviral use, herpetic episodes, or antipneumocystis prophylaxis for before the landmark and who has a CD4 count of 300 cells/µL, a hemoglobin concentration of 135 g/L, and a platelet count of 212.5 x 109/L. Group 1 started acyclovir therapy at or before this landmark (n = 131; 15 deaths); group 2 either never used acyclovir or started it after this landmark (n = 502; 145 deaths); and group 3 never used acyclovir (n = 351; 108 deaths). Right. Estimated survival curves using a landmark of a CD4 count of 50 cells/µL or a clinical AIDS diagnosis. The curves are estimated for a patient with symptomatic HIV infection who is receiving zidovudine but had no other antiretroviral use, herpetic episodes, or antipneumocystis prophylaxis before the landmark, and has a CD4 count of 75 cells/µL, a hemoglobin concentration of 122 g/L, and a platelet count of 187.5 x 109/L. Group 1 started acyclovir therapy at or before this landmark (n = 85; 28 deaths); group 2, either never used acyclovir or started it after this landmark (n = 203; 142 deaths); and group 3 never used acyclovir (n = 154; 112 deaths). AZT = zidovudine; ACV = acyclovir.

Drug Administration

The only available data on acyclovir dosage are from visits 13 to 17 (April to September 1990 to April to September 1992), in which the median dosage (from reported use as a medication for HIV infection or AIDS) was between 600 to 800 mg/d at each visit. A Cox model comparing patients not receiving acyclovir with those receiving a dosage of 600 mg/d or greater, a dosage of less than 600 mg/d, and unknown dosage for HIV infection did not show a dose effect for either progression to AIDS or survival. The low median dosage reported for acyclovir when reported for use as an HIV medication and the absence of a dose interaction in the Cox model led us to examine whether more uninterrupted use was associated with better survival.

Constancy of acyclovir administration was examined in three ways. Two Cox models compared survival between participants who reported any use of acyclovir for three or more or four or more consecutive visits with those who did not meet these criteria. In the third model, the cumulative number of visits of reported acyclovir use was used as a covariate. Acyclovir taken for any indication was used because in previous analyses it had shown a borderline effect on prolonging survival and because it was a potentially less restricted population than the group that used acyclovir only for HIV. The three models showed a consistent trend in favor of more constant use of acyclovir in an 18% decrease in the risk for death (relative hazard, 0.82; CI, 0.59 to 1.13; P = 0.23), a 28% decrease (relative hazard, 0.72; CI, 0.50 to 1.05; P = 0.09), and a 7% per visit decrease (relative hazard, 0.93 per visit; CI, 0.87 to 0.99; P = 0.029) for three or more consecutive visits, four or more consecutive visits, and cumulative visits of use models, respectively (Figure 1). Consistent with our other models, no effect on time to progression to AIDS was found for constancy of acyclovir dosing.

Discussion

Top Methods Results Discussion Author & Article Info References

The main findings of our analysis, summarized in Figure 1, were a clinically significant prolongation of survival associated with the concurrent use of acyclovir with zidovudine and no effect on progression to AIDS or development of cytomegalovirus disease. Using the mean of the two comparison curves in the landmark analyses Figure 2, the estimated length of the additional time to 90% survival is between 154 and 210 days, depending on the landmark used. Median survival was only reached for the late landmark (the development of either a CD4 count <50 cells/µL or AIDS). Using the same assumptions, the increase in survival associated with acyclovir use would be approximately 374 days. Our analyses of drug administration indicate that the prolongation of survival by acyclovir in combination with other HIV therapy was associated with more uninterrupted use, but it was not associated with a discernible dose effect. The benefit appeared to be largest if concomitant acyclovir was given after a diagnosis of AIDS had been made. However, an important caveat is that this could also be secondary to the higher mortality in more advanced disease, which allows a difference to be more easily detected. The landmark analyses, although subject to a greater possibility of survival bias than the time-dependent models, showed an estimated survival that was greater for the groups that started acyclovir therapy at or before the landmark or the group that either never received acyclovir or received it after the landmark as compared with the group that never received acyclovir. Together, these analyses support the hypothesis that the chronic use of acyclovir therapy was probably more important than the time at which therapy was initiated.

These findings are consistent with data from recent clinical trials [12-15]. The Wellcome-United Kingdom trial [13-15] enrolled 302 participants randomly assigned to 3.2 g/d of acyclovir or placebo plus nonblinded, uncontrolled antiretroviral therapy and reported a 1-year mortality rate of 23% in the acyclovir arm compared with a 35% rate in the placebo arm (P = 0.035). This preliminary report [15] of a 50% reduction in relative risk is similar to our findings. In their trial, Cooper and colleagues [12] also found significant prolongation of survival in both the group with AIDS (1-year estimated survival probability, 0.54 compared with 0.73 for the zidovudine and zidovudine plus acyclovir arms, respectively; P = 0.014) and the group with AIDS-related complex (1-year estimated survival probability, 0.81 compared with 0.97 for the zidovudine and zidovudine plus acyclovir arms, respectively; P = 0.045). In a Cox model that adjusted for imbalances in baseline CD4 lymphocyte counts and disease symptoms, the observed difference in survival between the zidovudine and combination therapy arms became nonsignificant. However, the reduction in risk was still approximately 50%, consistent with both our findings and those of Youle and coworkers [15]. These and other trials [8, 9, 12, 15] have failed to show an apparent difference in progression to AIDS or the development of cytomegalovirus disease, although some studies had insufficient power to detect such an effect.

Our results are not based on a randomized controlled trial but rather on an analysis of observational data; it is therefore subject to potential confounding biases. As we have shown previously, patients who start therapy for HIV infection in the Multicenter AIDS Cohort Study are more ill than those who do not [27], which leads to a potential selection bias. This can be controlled if careful attention is given to adjusting analyses for imbalances in disease staging when comparing treated and untreated groups [28, 29]. An opposing or nonconservative bias can also be introduced if men who survive longer are more likely to receive therapy than those who have shorter survival. This leads to a survival bias that may overestimate a treatment effect. This can be minimized by excluding historical controls and by adjusting for the timing of therapy in the analysis [28-30]. However, the most direct method for correcting this bias is to compare the use and non-use of treatment among participants who have survived to the same point in time while simultaneously adjusting for disease stage. It was for these reasons that time-dependent covariates (risk factors) were used in our modeling of acyclovir use because this would minimize potential survival and selection biases by allowing comparison of participants who survived to the same time point and would correct for changes in care that occur both over time and at various disease stages.

Our use of intent-to-treat analysis, which assumes that once use of zidovudine or acyclovir is reported the participant continues to receive it, is conservative because it ignores changes in therapy that occur after initiation, which may bias estimates of effectiveness. In addition, the inclusion of patients in the treated group who are not actually receiving the stated therapy would tend to attenuate the ability to detect differences between the groups. The Multicenter AIDS Cohort Study includes a highly motivated homosexual male cohort in whom access to health care is not a confounding factor [31].

Other potential biases include a systematic difference between participants receiving acyclovir for different indications and its possible link to other care. The presence of HIV-associated symptoms and symptomatic episodes of herpes were both more common at baseline in the participants who reported receiving acyclovir. Although adjustment for these prognostic factors was done in the time-dependent Cox models, any potential bias in these factors would tend to predict an increased mortality for the acyclovir groups, as we found that both of these covariates were associated with a tendency to shorten survival. Both groupings of acyclovir use had similar effects on survival, indicating that the patient's reported reason for receiving acyclovir did not appear to bias the outcome. The results of the dosage analysis, in which the participants receiving acyclovir for HIV might be predicted to be receiving a higher dose for longer-term use showed a median dose of 600 to 800 mg at the visits for which information was available and no dose effect in the Cox modeling. It does not indicate a systematic difference between the groups. Consistent with the dosage analysis, the analysis of constancy of dosing done on all participants reporting acyclovir use indicated that greater uninterrupted use of acyclovir was associated with the greatest effect on survival.

The clinically significant difference in survival without an effect on progression to AIDS is an apparent incongruity. Potential explanations of this difference include prophylaxis for various opportunistic infections and other changes in care that have resulted in a progressive prolongation of time to development of an AIDS-defining illness; changes over time in the types of events occurring; trends over time to progressively lower CD4 lymphocyte counts at the time of the first AIDS event [28, 30, 32]; and a trend toward an increasing proportion of patients who die without experiencing a 1987 CDC case-definition AIDS illness [33]. These trends would tend to increase the difficulty in distinguishing a difference between groups. The specific explanation, however, remains unknown.

In our analysis and previous clinical trials, no apparent effect on development of cytomegalovirus disease or progression to AIDS was found. As previously noted, acyclovir does not have a direct antiviral effect on HIV, and the data on its potentiation of the effect of zidovudine is unconfirmed [7]. However, other mechanisms exist by which acyclovir could potentially affect HIV disease without a direct antiviral effect on HIV-1 or potentiation of the action of zidovudine. Several studies have shown an in vitro interaction between herpesviruses such as herpes simplex virus or cytomegalovirus and HIV-1 replication. The early regulatory proteins of herpes simplex virus (for example, ICP0 and ICP4) and cytomegalovirus (for example, the IE gene) appear to lead to an activation of the HIV-1 long terminal repeat and increased transcription [34, 35]. Whether such interactions are clinically relevant is unclear, but two case reports of acute coinfection of both HIV-1 and cytomegalovirus suggested additional morbidity over what would be expected from either infection separately [36]. In addition, a comparison of skin biopsy specimens from patients with both HIV-1 and herpes simplex virus infection compared with either those with only HIV-1 infection or those with only herpes simplex virus infection suggested a significant increase in the viral load of both viruses in the coinfected patients' keratinocytes and dermal macrophages [37]. A constant suppressing concentration of acyclovir could result in less reactivation of subclinical herpes replication and therefore less HIV replication. As indicated in our analysis, higher concentrations would not have additional benefit.

Our analysis would suggest that use of acyclovir consistently in a dose sufficient to suppress herpetic recurrences (600 to 800 mg/d) has a significant effect on prolonging survival in a well-characterized cohort with extensive previous exposure to herpes group virus infections. Regardless of the exact mechanism of the effect of acyclovir on patients treated with antiretroviral agents, our analysis is consistent with other data and needs confirmation by a randomized controlled trial before this becomes standard care. Several ongoing trials should provide additional information on the effect of acyclovir on disease progression and survival.

Appendix

The following are the investigators in the Multicenter AIDS Cohort Study: The Johns Hopkins School of Hygiene and Public Health: Alfred J. Saah, MD; Ellen Taylor, BS; Haroutune Armenian, MD, DrPH; Homayoon Farzadegan, PhD; N.M.H. Graham, MBBS, MD; Joseph Margolick, MD, PhD; Justin McArthur, MBBS. Data Coordinating Center, Johns Hopkins School of Hygiene and Public Health: Alvaro Munoz, PhD; Leonardo Epstein, PhD; Noya Galai, PhD; Donald R. Hoover, PhD; Lisa P. Jacobson, ScM; Curtis Mienert, PhD; Kenrad Nelson, MD; Lawrence P. Park, MSE; Steven Piantadosi, MD, PhD; Sol Su, ScD. Howard Brown Health Center and Northwestern University Medical School: John P. Phair, MD; Joan S. Chmiel, PhD; Bruce Cohen, MD; Maurice O'Gorman, PhD; Diane Variakojis, MD; Jerry Wesch, PhD; Steven Wolinsky, MD. University of California, Los Angeles, Schools of Public Health and Medicine: Roger Detels, MD, MS; Barbara Visscher, MD, DrPH; Irvin Chen, PhD; Jan Dudley, MPH; John Fahey, MD, MS; Janis Giorgi, PhD; Moon Lee, PhD; Otto Martinez-Maza, PhD; Eric Miller, PhD; Hal Morgenstern, PhD; Paruang Nishanian, PhD; Jeremy M.G. Taylor, PhD; Jerome Zack, PhD. University of Pittsburgh Graduate School of Public Health: Charles Rinaldo, Jr, PhD; Lawrence Kingsley, DrPH; James Becker, PhD; Phalguni Gupta, PhD; Monto Ho, MD; Sharon L. Zucconi, PhD. National Institute of Allergy and Infectious Diseases: Lewis Schrager, MD; Sten Vermund, MD; Richard Kaslow, MD; Mark Van Raden, MA. National Cancer Institute: Daniela Seminara, PhD.