The reasons for recent declines in AIDS incidence and mortality may include advances in treatment, but these may be confounded by earlier declines in the incidence of human immunodeficiency virus (HIV) infection. To determine whether the declines in AIDS and mortality may, in part, stem from wider use of combination antiretroviral therapy, 622 HIV-positive men with well-characterized dates of seroconversion were followed. In this group, combination therapy came into widespread use in only 1996. In a Cox proportional hazards model, the 1996 calendar period was significantly associated with slower progression to AIDS (relative hazard [RH] = 0.19, 95% confidence interval [CI], 0.05–0.69, P = .01) and death (RH = 0.45, 95% CI, 0.21–0.95, P = .04). Declines in incidence of HIV infection, changes in HIV virulence, and end-point underreporting cannot fully explain the decline in AIDS and death in 1996. The introduction of combination antiretroviral therapy as the standard of care may already have had measurable effects.
Recent sharp declines in AIDS incidence and mortality have been reported for both the United States  and Canada . These trends may reflect declines in the incidence of human immunodeficiency virus (HIV) infection following the first epidemic wave in the early 1980s, as well as the recent introduction of more effective combination antiretroviral therapy. However, the surveillance data on which the reports are based do not make it possible to distinguish these two effects, since information on time of infection is lacking. This analysis takes advantage of data for 622 homosexual and bisexual men with well-characterized dates of seroconversion to determine whether the decline in risk of AIDS and AIDS-related mortality during 1996 compared with that in 1993–1995 is explained by earlier changes in HIV incidence. Then, after examining year of seroconversion and underreporting, we consider trends in the use of combination antiretroviral therapy as an alternative explanation for the 1996 decrease in risk.
Patients and Methods
Sample design and subject criteria
The sample of 622 is drawn from among 6704 high-risk homosexual and bisexual men screened at the San Francisco municipal sexually transmitted disease clinic in 1978–1980 for studies of hepatitis B and subsequently followed in the San Francisco City Clinic Cohort (SFCCC) study, a prospective study of the natural history of HIV infection and AIDS. Serum samples stored at screening and during a hepatitis B vaccine trial carried out in 1980–1983 or obtained in prospective follow-up made it possible to determine the timing of HIV seroconversions in this high-risk group with considerable precision. For 421 (68%) of 622 men, the earliest stored sample tested positive for antibodies to HIV and was collected before the beginning of 1981, shortly after the earliest evidence of the epidemic in San Francisco. Dates of seroconversion for this group were imputed using a model for HIV incidence. For the remainder, the first positive HIV antibody test followed an earlier negative test by at most 2 years, and dates of seroconversion were imputed at the midpoint between the two test dates.
Ascertainment of treatment history and end points
SFCCC participants are interviewed at 6- to 12-month intervals. Current use of combination therapy with at least two antiretrovirals and/or protease inhibitors was ascertained at each study visit by self-report. AIDS diagnoses were ascertained through matching with local and national AIDS case registries, with the most recent match in August 1998. Index diagnoses meeting the 1993 criterion of <200 CD4 cells were determined retrospectively for some study participants by the local registry, with the earliest such diagnoses occurring in 1988. Deaths were ascertained by matching with the National Death Index (NDI) and a San Francisco Department of Public Health registry of death certificates. The match with NDI is current through the end of 1996.
Unadjusted comparison of diagnoses and death rates for 1996 with the corresponding rates for earlier periods would confound calendar period with time since HIV seroconversion. Cox proportional hazards models for time from seroconversion to AIDS or death, with time-dependent covariates indicating calendar year at risk, avoid this difficulty . In these models, the effect of calendar period is assessed within risk sets matched for estimated duration of infection. The years 1993–1995 were chosen as the comparison period because, despite some year-to-year variation in this period, there was little overall change in diagnosis rates, the proportion of cases meeting only the 1993 case definition, and mortality, providing a stable baseline with which to compare data from the 1996 calendar year.
Incompleteness in reporting of 1996 AIDS diagnoses could account for some of the declines in AIDS incidence. We estimated underreporting of 1996 compared with 1993–1995 cases and then repeatedly imputed extra 1996 cases among those at risk, approximately equalizing the degree of underreporting in the two periods being compared. Proportional hazards analysis was carried out on each completed set of outcomes, and summary relative hazards (RHs), confidence intervals (CIs), and P values were computed, taking into account the multiple imputation .
The NDI database for 1996 is >99% complete. However, because of potential errors in matching variables, the expected incompleteness of matches is ∼10% . Since this would affect 1993–1995 and 1996 deaths equally, there is no need to impute extra 1996 deaths.
Seroconversion times, AIDS diagnoses, and mortality
In this sample of 622 men, HIV seroconversions were highest in 1979 and fell rapidly after 1980 (table 1). By the end of 1996, 505 (81%) were known to have been diagnosed with AIDS, and 450 (72%) had died. Table 2 shows AIDS incidence and mortality per 100 person-years by calendar year for 1990–1996. The incidence of AIDS peaked in 1995 at 20.3 per 100 person-years, averaged 15.8 per 100 person-years in 1993–1995, and in 1996, dropped to 2.6 per 100 person-years. The proportions of index AIDS diagnoses meeting only the 1993 case definition were 72%, 69%, and 85% in 1993, 1994, and 1995, respectively, and all 3 1996 cases met this criterion. Mortality closely tracked AIDS incidence over the entire period, but with a smaller decrease, both relative and absolute, in 1996.
Calendar period effects
In the proportional hazards model for AIDS diagnosis controlling for duration of infection, and thus for the HIV incidence, the RH for 1996 compared with 1993–1995 was 0.19 (95% CI, 0.05–0.69, P = .01). In the model for survival time, the RH for 1996 compared with 1993–1995 was 0.45 (95% CI, 0.21–0.95, P = .04).
Adjustment for underreporting of AIDS cases
Only 3 AIDS cases were reported for 1996. Estimated underreporting for these events was 5%. The multiple imputation analysis was carried out using 1 and then 2 imputed cases, equivalent to underreporting of 25% and 40%. With 1 imputed case, the multiple imputation RH was 0.25 (95% CI, 0.08–0.80, P = .02), and with 2 it was 0.31 (95% CI, 0.10–0.92, P = .03). One is the maximum number of extra cases that is statistically consistent with 3 reported cases, in the sense that with ⩾5 actual cases and an expected reporting rate of 95%, the binomial probability of ⩽3 reported cases is <5%.
Effect of year of seroconversion
The decrease in 1996 AIDS incidence might in part reflect temporal decreases in the virulence of the transmitted viral strain or access to therapy relatively soon after infection, either of which would be reflected in decreasing incidence rates by calendar year. However, in a supplementary model for time to AIDS there was significantly increasing risk with calendar year of seroconversion (RH = 2.01 per 10 years, 95% CI, 1.10–3.70, P = .02). In this multivariate model, which included age and calendar period as well as year of seroconversion, the 1996 calendar period remained significantly associated with AIDS (RH = 0.16, 95% CI, 0.04–0.60, P = .01). The increasing trend in risk of death with year of seroconversion was weaker and not significant (RH = 1.30 per 10 years, 95% CI, 0.64–2.64, P = .48). Again, the RH for the 1996 calendar period remained significant (RH = 0.42, 95% CI, 0.20–0.91, P = .03).
Prevalence of combination antiretroviral therapy
Information from study visits on antiretroviral treatment is available for 316 men (51%), including 1996 data for 115 (63%) of 183 men still alive at the beginning of that year and 62 (53%) of the 118 remaining AIDS-free. Among men with <500 CD4 cells, use of at least one antiretroviral has been reported by 32%–57% between 1993 and 1996; among those with <200 CD4 cells, prophylaxis against Pneumocystis carinii pneumonia has been reported by 83%–91% over this period. Current use of combination therapy with at least two drugs was reported by 22%, 13%, and 23% of participants with <500 CD4 cells at 1993, 1994, and 1995 study visits, but by 49% and 79% at their 1996 and 1997 visits. Use of protease inhibitors was first reported by only 4% of participants in 1995, and then by 30% in 1996 and 61% in 1997. The number of men treated with combination therapy among those at-risk in 1996 substantially exceeded the expected number of “averted” 1996 AIDS diagnoses. Among men in follow-up for whom CD4 counts and treatment information were available, 40 with <500 CD4 cells at their 1996 study visits were AIDS-free at the beginning of the year. If 1993–1995 rates had continued in 1996, 12 AIDS diagnoses would have been expected in this group, compared with 1 actual case. Furthermore, 16 men in this group (40%), including the person with AIDS, reported receiving combination therapy in 1996.
The prevalence of combination therapy and protease inhibitor use in 1996 was almost surely higher than our estimates. Use was clearly increasing in this period, and the majority of study visits occurred early in the year, so that treatment initiated later in the year, sometimes in response to decreased CD4 counts at the study visit, would be missed. In addition, use among men continuing to make study visits may be lower than among others remaining at risk.
In our cohort of 622 homosexual and bisexual SFCCC study participants with well-characterized dates of seroconversion, both AIDS incidence and mortality fell dramatically in 1996. Several factors were examined that fail to explain this decline completely. First, seroconversion declined sharply in this cohort after 1982, when only 8% of the group of 622 remained uninfected. Seroincidence in the wider community of men who have sex with men in San Francisco followed similar patterns, although the peak was 2–3 years later . However, the 1996 calendar period was significantly associated with increasing AIDS latency and survival after controlling for earlier declines in HIV incidence.
In addition, multiple imputation of extra 1996 cases showed that underreporting of 1996 diagnoses does not account for the decline in AIDS incidence. After imputing 2 extra 1996 cases, 1996 remains significantly associated with lower risk of AIDS diagnosis. Furthermore, there was no evidence for confounding of the effect of the 1996 calendar period at risk by year of seroconversion. The increasing risk of diagnosis and perhaps death with year of seroconversion may reflect increasing virulence or resistance in transmitted strains or underrepresentation of the fastest progressors among study participants. Finally, the 1993 changes in the AIDS case definition do not appear to account for the decrease in 1996 AIDS incidence, since the proportion of index diagnoses meeting only the new definition did not significantly change over the entire period of 1993–1996.
A possible explanation for the 1996 decline in AIDS incidence and mortality is the increase in use of highly effective combination antiretroviral therapy. Use of prophylaxis against Pneumocystis carinii pneumonia and reverse transcriptase inhibitors has been frequently reported in the cohort since the early 1990s. In contrast, use of combination therapy including protease inhibitors was uncommon until the end of 1995 and then increased rapidly in 1996 and 1997. Actual use in 1996 was almost surely higher that our self-reports indicate. The less dramatic increase in survival time may reflect lower efficacy among men who begin combination therapy only after diagnosis of AIDS, or those with a history of receiving sequential monotherapy.
The SFCCC was a convenience sample of high-risk men who have sex with men. Although median times to AIDS and death, cofactors for HIV progression, and the proportion of long-term nonprogressors in this cohort are similar to those reported for other cohorts of MSM [7, 8], our group may be unrepresentative in having been infected relatively early. Thus generalizability would be limited if combination therapy were differentially effective among slower progressors. However, in light of other evidence [1, 2, 9], the decrease in risk of progression to AIDS and death recently observed in our sample of 622 homosexual and bisexual men with well-characterized dates of seroconversion suggests that combination therapy is having measurable effects at the population level and provides additional support for its adoption as the standard of care.
We thank Steven Bryzman, Sandra Schwarcz, and Ling-Chin Hsu of the HIV Research Section, San Francisco Department of Public Health, and Robert Bilgrad of the National Center for Health Statistics for help in preparation of the manuscript.