Survival After AIDS Diagnosis in Adolescents and Adults During the Treatment Era, United States, 1984-1997

In 1987, 6 years after the first cases of acquired immunodeficiency syndrome (AIDS) were reported,¹ the US Food and Drug Administration approved the first antiretroviral drug for treatment of human immunodeficiency virus (HIV) infection. In 1988, the first randomized controlled trial of primary prophylaxis of Pneumocystis carinii pneumonia in persons with HIV appeared in the medical literature.² In the ensuing decade, several new antiretroviral agents were approved. In 1995, protease inhibitors were introduced and, in combination with reverse transcriptase inhibitors, became the standard of care by 1996.³ Several clinic-based studies and controlled trials have demonstrated increased survival times with use of antiretroviral therapies and opportunistic illness (OI) prophylaxis.^4-8

National death certificate data for 1996, 1997, and 1998 showed 29%, 48%, and 21% declines in the age-adjusted death rate for HIV infection, following an average annual increase of 16% from 1987 through 1995.^9,10 The annual numbers of AIDS cases diagnosed and deaths among persons with AIDS reported to the national HIV/AIDS surveillance system began to decline in 1996 and continued to fall during 1997 and 1998.^11,12 To examine the effect of diagnosis year on survival time, we analyzed trends in survival among persons diagnosed with AIDS from 1984 through 1997.

Using data from the national HIV/AIDS surveillance system of the Centers for Disease Control and Prevention (CDC), we analyzed adult and adolescent AIDS cases diagnosed in 1984-1997 and reported through December 1999. We used data on deaths that occurred through December 31, 1998, and were reported by December 31, 1999. All 50 states, the District of Columbia, and US dependencies and possessions report AIDS cases to the CDC using a uniform case definition and report form. Reporting areas receive federal funding for AIDS surveillance, which includes active AIDS case ascertainment and death ascertainment. Assessments of completeness of reporting have indicated that AIDS case reporting is between 80% and 97% complete.^13-16

Death ascertainment includes routine review of death certificates and death registry matching. In addition, 16 project areas (13 states and 3 cities) have conducted active death ascertainment using the National Death Index (NDI). The NDI and its matching algorithms have been described elsewhere.^17,18 Briefly, the NDI is a national computerized index to all death certificates on file in state vital statistics offices. Assessments of the performance of NDI matching algorithms indicate 83% to 98% sensitivity and 99% to 100% specificity.^17,19,20 To assess completeness of death ascertainment among persons reported as having AIDS, project area staff matched AIDS cases to the NDI. Project areas updated AIDS surveillance case reports if deaths were ascertained among persons previously assumed alive. Ninety-four percent of deaths identified in the NDI had been previously reported to AIDS surveillance.²¹ More than 60% of cases used in this analysis were from NDI project sites.

We examined changes in time from first AIDS-defining OI to death for persons reported during 1984-1997. In 1993, the AIDS case definition was expanded to include severe immunodeficiency (CD4⁺ cell count <200 × 10⁶/L or <14% of total lymphocytes), even without an OI.²² Because the new immunologic criteria may be met at an earlier stage of HIV disease, before diagnosis of an OI, survival time of persons with a diagnosis based on an immunologic criterion may be longer than survival of those with a diagnosis based on an OI. As a result, overall survival time will appear to increase. To account for this surveillance artifact, we focused our analysis on cases reported in 1993-1997, categorized by initial diagnosis (OI or immunologic criteria). Dates of diagnosis and death are reported as month and year. In all analyses, we excluded persons with AIDS diagnosis and death during the same month because they contributed no measurable survival time and an unknown proportion consists of previously diagnosed but unreported cases whose actual survival time is unknown. We censored all cases at 5 years of survival.

To estimate the probability of survival past some point in time, we calculated Kaplan-Meier survival curves for the initial descriptive analysis.²³ Because the assumption of proportional hazards did not hold,²³ we used direct standardization instead of Cox proportional hazards models to compare improvements in survival time for each year.²⁴ Direct standardization allows comparison of probability of survival for each diagnosis year while holding constant the effects of all other variables. Each comparison is adjusted for all factors not used in the primary stratification. Adjustment factors included race/ethnicity, initial AIDS diagnosis category (OI or immunologic), sex, age group at diagnosis, mode of HIV exposure, and region of residence. We used OI AIDS cases diagnosed in 1993-1997 as the standard population when comparing the survival effect of other factors among cases initially diagnosed with an OI. We used immunologic AIDS cases when comparing factors among cases initially diagnosed with immunologic AIDS. During standardization, we excluded the last 6 months of survival time in all subgroups, since small numbers of observations reduced stability of survival estimates. We used the log-rank test to calculate P values in stratified analyses. We did not use logistic regression models to estimate odds ratios. When an outcome is common, odds ratios no longer estimate relative risk. The 12- and 36-month survival probabilities (see "Results" section) were not close to 1 (mortality probabilities are not close to 0); thus, the odds ratios for these survival times were not interpretable as approximate relative risks. Standard errors for the individual estimates were small except for the 2 smallest racial/ethnic groups. The estimated survival probabilities calculated using standardization allow a direct estimate of survival probability ratios.

In 1984-1997, 438 695 AIDS cases were diagnosed based on OI and reported to the national HIV/AIDS surveillance system by December 31, 1999. We excluded 2758 cases (0.6%) from the analysis because of incomplete information on age, race, residence, or date of death. Of the remaining 435 937 cases, we excluded 41 232 (9.5%) who were diagnosed as having AIDS and died in the same month. Kaplan-Meier estimates of survival for the remaining 394 705 cases showed that median survival time improved with each successive year of OI diagnosis, from 11 months for persons with AIDS diagnosed in 1984 to 46 months for persons with AIDS diagnosed in 1995 (Figure 1). The greatest 1-year increase in median survival time was from 21 months for persons with an OI diagnosed in 1994 to 46 months for those with an OI diagnosed in 1995. Median survival time could not be determined for persons receiving an AIDS diagnosis in 1996 and 1997 because fewer than half had died by December 31, 1998. Among persons with an OI diagnosed in 1996, 67% were alive at least 36 months after diagnosis. Among persons with an OI diagnosed in 1997, 77% were alive at least 24 months after diagnosis.

In 1993-1997, 311 758 AIDS cases were diagnosed using either OI or immunologic criteria and reported to the national HIV/AIDS surveillance system through December 31, 1999. We excluded 3737 cases (1.2%) from the analysis because of incomplete information. In 11 400 (3.7%) of the remaining 308 021 cases, AIDS diagnosis and death occurred in the same month. The proportion of these 0-month survivors was consistent each year, ranging between 3.2% and 3.9% of all cases reported in 1993-1997. Zero-month survivors were slightly more likely to be older, but their proportion varied little by race, sex, and mode of HIV exposure. We conducted a sensitivity analysis including the 0-month survivors and our results were substantively unchanged, indicating that their exclusion did not introduce bias.

On exclusion of the 0-month survivors, data from 296 621 cases were analyzed. Population characteristics and Kaplan-Meier estimates of the probability of survival for at least 24 months are shown in Table 1. As of December 31, 1998, 37% of these persons had died. Survival time improved with each successive year of diagnosis. The probability of survival for at least 24 months increased from 60% for those receiving diagnoses in 1993 to 87% in 1997. Survival for at least 24 months was greater for those who were younger when AIDS was diagnosed, ranging from 77% among persons aged 13 to 29 years to 60% among persons aged 50 years or older. Differences in 24-month survival by race/ethnicity, sex, mode of HIV exposure, population density, region of residence at diagnosis, and state's NDI project status were smaller. Differences in 24-month survival probabilities between categories of these variables ranged from 1% to 5%.

The probability of survival for at least 24 months was 77% for persons whose initial diagnosis was based on immunologic criteria and 60% for persons whose initial diagnosis was based on an OI (Table 1). For each successive diagnosis year, persons whose initial diagnosis was based on an immunologic criterion survived longer than persons whose initial diagnosis was based on an OI. For each year of diagnosis in 1993-1997, the probability of survival for at least 24 months among persons with an immunologic diagnosis was 67%, 71%, 80%, 87%, and 90%, respectively. The probability of survival for at least 48 months was 48%, 60%, and 73% among persons in whom immunologic AIDS was diagnosed in 1993-1995. Among persons with an initial OI diagnosis in 1993-1997, the probability of survival for at least 24 months was 49%, 52%, 63%, 74%, and 80%, respectively. The probability of survival for at least 48 months was 33%, 42%, and 56% among persons with an initial OI diagnosis in 1993-1995.

After standardization to adjust for significant confounders, improved survival time remained associated with later year of diagnosis. Survival time remained associated with diagnosis category (P<.001); in every month, persons with an OI diagnosis experienced lower probability of survival compared with those with an immunologic diagnosis. Table 2 shows 6-, 12-, and 36-month survival probabilities for persons with OI and immunologic diagnosis. Older age at diagnosis also remained significantly related to a lower probability of survival (P<.001) (Table 3). Sex was not associated with survival for persons receiving their diagnosis in 1994-1997. Among persons receiving their diagnosis in 1996, men had slightly higher survival probabilities; however, actual differences in survival probabilities between men and women were small, both for persons whose initial AIDS diagnosis was based on immunologic criteria and those whose diagnosis was based on OI. Among persons whose initial diagnosis was based on an OI in 1993, 34% of women survived at least 60 months compared with 31% of men. Among persons whose initial diagnosis was based on immunologic criteria in 1993, 45% of women compared with 46% of men survived at least 60 months.

Standardized analyses stratified by race/ethnicity revealed improvements in survival probabilities with successive year of diagnosis. Table 3 displays 6-, 12-, and 36-month survival probabilities and percentage change from the previous year by diagnosis year for each racial/ethnic group. While differences in survival probabilities between racial/ethnic groups were statistically significant (P<.001), the proportion surviving at each point was similar between groups.

Annual improvements in survival time were not constant across the observation period (Table 2 and Table 3). Figure 2 shows improved survival time with later year of diagnosis for white (non-Hispanic), black (non-Hispanic), and Hispanic cases. Figure 3 shows the percentage improvement from the year before in the standardized 1-year survival probabilities after AIDS diagnosis by race/ethnicity. Relative improvement in survival time among persons with an AIDS diagnosis in 1996 compared with 1995 was greater than the improvement in survival time among persons with an AIDS diagnosis in any other year. Diagnosis year had the greatest effect on survival; the effect and magnitude were similar across racial/ethnic groups.

Survival time after AIDS diagnosis improved with each subsequent year of diagnosis, whether analysis start time was based on an OI or immunologic diagnosis. We observed similar improvements at the population level among each racial/ethnic group. Among all racial/ethnic groups, survival time improvements were not constant across the observation period. We observed the greatest relative gain in survival time for persons with AIDS diagnosed in 1996 compared with those whose diagnosis was in 1995. Although we did not have individual-level data on treatment, the survival time improvements were likely due to improved HIV and OI therapies as well as the increasing proportion of persons receiving these therapies.

Improvements in survival time by year of diagnosis were reported early in the epidemic, before the introduction of combination antiretroviral therapy and protease inhibitors. Using US AIDS surveillance data, Harris²⁵ found improvements in 1-year survival probabilities for persons in whom P carinii pneumonia was diagnosed in 1986 and 1987 compared with persons with similar diagnoses in 1984 and 1985. Harris attributed improvements to better diagnosis and treatment of pneumocystosis and introduction of zidovudine in 1986. With an additional decade allowing for delayed reporting of deaths, our analysis, using the same data, found similar improvements in 1-year survival probabilities among persons with an OI diagnosed during those years (data can be estimated by visual inspection of Figure 1). Lemp et al²⁶ suggest that survival time improvements among persons in whom AIDS was diagnosed in 1981-1987 in San Francisco, Calif, were due in part to zidovudine treatment. They reported median survival estimates for persons with diagnoses during 1984-1987 as 11.0, 10.8, 12.2, and 15.6 months, respectively. Our data show similar median survival estimates of 11, 12, 13, and 17 months for the respective years (Figure 1).

The effect of diagnosis year on survival time in our analysis is likely a reflection of improvements in OI prophylaxis, OI treatment, and antiretroviral therapy. Our data did not contain individual-level information on therapies; however, several studies of patients in care have found an association between OI prophylaxis or combination antiretroviral treatment and increased survival time.^4-6 Additionally, the percentage of patients receiving combination antiretroviral therapy with protease inhibitors increased from 0% in 1993 to more than 40% by 1997 in one study²⁷ and to more than 65% in another.⁸

Racial/ethnic differences in survival times have been reported since early in the AIDS epidemic. However, in a synthesis of 5 survival analyses reporting results by race and published between 1987 and 1991, race differentials disappeared when analyses accounted for socioeconomic differences and other characteristics, such as HIV exposure risk.²⁸ Other studies have found that when racial differences in disease severity at time of entry into care or insurance coverage are accounted for, survival differences by race are attenuated.^4,29 Evidence suggests that HIV-infected blacks may use fewer medical services compared with whites,³⁰ and the patterns of care received by blacks, Medicaid recipients, and uninsured persons may be less desirable than patterns of care received by whites and insured persons.³¹ In our population-level data, we observed small but statistically significant differences in survival probabilities between groups. A smaller proportion of blacks survived in each month compared with whites and Hispanics (Figure 2). However, we observed a similar pattern of improvement in survival among all racial/ethnic groups. White, black, and Hispanic improvement was more likely than Asian or American Indian (Table 3), but the numbers in the latter groups were small. Figure 3 indicates that the pattern of improvement in survival was similar across racial/ethnic groups (white, black, and Hispanic). While we were unable to quantify by race the proportion of persons in care, it seems reasonable that differences in survival times by race may reflect differences in disease severity at entry into care; access to care, including insurance coverage; and utilization patterns.

Survival time differences by other demographic characteristics observed in this analysis are consistent with those observed in other studies. We observed no differences in survival by sex. Two early studies of AIDS survival times among persons living in New York, NY, showed an increased risk of death among women compared with men,^32,33 but subsequent analyses found no differences by sex when controlling for CD4⁺ cell count, treatment, mode of HIV exposure, OI, and year of diagnosis.^6,34,35 An early review of HIV disease among women concluded that much of the reported sex difference in survival times is accounted for by differences in access to and use of therapy.³⁶ Moncroft et al³⁷ also concluded that sex differences reported early in the epidemic were likely not biological but may have reflected differences in race/ethnicity and socioeconomic status. We also found an association between older age and shorter survival times, as has been observed in other investigations.^32,34,35

The annual percentage reduction in adjusted risk of death following AIDS diagnosis was not constant over the period. We saw the greatest percentage reductions in risk of death among persons with AIDS diagnosed in 1995 and 1996. Persons among whom AIDS was diagnosed in these years likely had survived long enough to take advantage of new multidrug regimens, including those with protease inhibitors. We saw a smaller reduction in risk of death for persons in whom AIDS was diagnosed in 1997, which mirrors the slowing rate of decline in AIDS deaths reported in 1998.¹² This may reflect maximization of benefits of new therapies. Suboptimal adherence and viral resistance may be limiting factors. Further study is required to determine the impact of adherence and resistance on survival at the population level.

It is important to continue to monitor AIDS diagnoses and deaths as sentinel events. As current clinical management strategies effectively prevent morbidity and mortality among persons with HIV, AIDS cases increasingly represent persons in whom HIV was first diagnosed at the time of AIDS diagnosis, persons with no or late access to medical care, and persons among whom new treatments have failed. In addition, because persons who respond well to therapy are less likely to have an AIDS diagnosis, using AIDS diagnosis date as the start time for estimating survival may select for sicker persons. Without continued improvements in access to care and therapies, length of survival time after AIDS diagnosis could decrease in coming years. However, it should be noted that in the future, survival time after AIDS diagnosis may not reflect the overall disease experience, including benefits of current management strategies, in that the drugs may successfully prevent the onset of AIDS until close to the time of death.

A potential limitation to this analysis is the underascertainment of deaths among persons reported as having AIDS, resulting in a bias toward increased survival times. Tu et al³⁸ state that there is a sizable fraction of AIDS cases in the national AIDS surveillance system for whom death will never be reported. Reporting areas are provided federal funds to conduct active ascertainment of deaths among persons with AIDS, including ongoing death certificate review and death registry matching. To ascertain deaths among residents whose death occurred out of state, 16 project areas matched AIDS cases to the NDI and found that 94% of deaths identified through the NDI had previously been reported to AIDS surveillance.²¹ We compared survival times among persons diagnosed in states implementing typical death ascertainment (death certificate review and death registry matching) with those implementing the NDI match. The differences were negligible. Kaplan-Meier probabilities of survival beyond 24 and 60 months were 71.8% and 55.9%, respectively, in states that used conventional methods and 70.8% and 55.9% in states that supplemented conventional methods with the NDI. Additionally, our primary analysis included cases diagnosed after the 1993 change in the AIDS surveillance case definition, which was estimated to capture up to 98% of HIV-related deaths.³⁹ Finally, a small proportion of immunologic AIDS cases may be misclassified because of failure to detect an earlier OI,⁴⁰ resulting in underestimation of survival among immunologic cases. However, misclassification likely only affected estimates in the 1993-1995 period, as widespread use of highly active antiretroviral therapy in 1996 reduced risk of OI.²⁷ Additionally, an increasing number of states have implemented laboratory-based CD4⁺ cell count reporting, which identifies cases earlier because immunodeficiency precedes most OIs.

This population-based analysis of AIDS survival times indicates improvements among all demographic groups. Although differences in treatment and disease severity at the individual level have not been taken into consideration, this surveillance-based analysis can inform prevention and intervention services planning. Improved survival times at the population level present a growing public health challenge as the number of persons living with AIDS increases. Between 1996 and 1999, estimated AIDS prevalence increased 33%, from 240 184 to 320 282.¹² Increasing prevalence resulting from improved survival times implies that more infected persons will need HIV-related services for a longer time. The longer need for treatment may make it more difficult for patients to adhere to complex treatment regimens. Enhanced prevention efforts to decrease subsequent transmission will be necessary to help sustain safer sex and needle-sharing behaviors for a longer period.