Smit, Colette*; Geskus, Ronald*; Uitenbroek, Daan*; Mulder, Dieke*; van Den Hoek, Anneke*; Coutinho, Roel A.*†; Prins, Maria*
From 1995 onward, mortality resulting from AIDS has been declining in most industrialized countries,1,2 probably as a result of the success and widespread use of highly active antiretroviral therapy (HAART). However, mortality trends are also influenced by the HIV incidence pattern in the past, which must be taken into account when evaluating the effect of HAART. Such an analysis is not feasible for surveillance data, which is often used in mortality studies, because HIV incidence data are not available.
In The Netherlands, AIDS was first diagnosed in 1982. From then until 2000, 2305 AIDS cases (39% of all cases in the country) had been reported to the AIDS Surveillance System of the Municipal Health Service of Amsterdam. The HIV epidemic predominantly affected homosexual men. In 1990, AIDS became the leading cause of death for all men age 25–54 years in Amsterdam, a city with 750,000 inhabitants; mortality resulting from AIDS exceeded mortality caused by heart diseases and malignant neoplasms.3
A reconstruction of the yearly incidence of HIV infection among a cohort of homosexual men in Amsterdam shows a peak in 1984, followed by a decline from 1985 onward.4,5 Without effective treatment, the median time from HIV seroconversion to death is estimated to be 11 to 12 years.6,7 Consequently, more than 50% of those infected with HIV in the early 1980s died in the early 1990s.7 For this reason, a decline in deaths from AIDS is likely to have begun before the introduction of HAART in 1996.
In the present study, we differentiated the observed AIDS mortality with the estimated mortality based on the HIV incidence patterns without the benefits of HAART among homosexual men. This analysis was possible because HIV incidence data among homosexual men in the early years of the AIDS epidemic were available.
Multiple data sources were used. We illustrated the observed course of the AIDS epidemic among homosexual men in Amsterdam with Amsterdam AIDS surveillance data. We used the HIV incidence patterns among homosexual men participating in the Amsterdam Cohort Study and the Amsterdam Hepatitis B vaccine trial and those attending the Amsterdam sexually transmitted infections (STI) clinic, together with the time from seroconversion to AIDS and death in the pre-HAART era, to estimate the natural course of the AIDS epidemic if HAART had not been introduced.
Table 1 presents an overview of the different data sources.
Since the beginning of the AIDS epidemic, epidemiologic data about AIDS patients has been collected by the AIDS Surveillance System of the Municipal Health Service of Amsterdam (1982–2000). In this system, physicians from all hospitals in Amsterdam report newly diagnosed AIDS cases on a voluntary basis. As backup, the physicians are every year contacted by a public health nurse or medical doctor of the Municipal Health Service and reminded to report these cases. Only AIDS cases meeting the 1987 AIDS case definition criteria of the U.S. Centers for Disease Control and Prevention (CDC) are allowed in the AIDS Surveillance System.8 In The Netherlands as of January 1, 1994, 3 new indicator diseases (pulmonary tuberculosis, recurrent pneumonia, and invasive cervical carcinoma) were added to CDC criteria,9 and a CD4 count below 200 × 106 cells/L criteria is not included. For every newly diagnosed AIDS case, information on demographics and risk group was obtained. Dates of death were obtained from the municipal death records. For this study, we matched AIDS cases without a date of death against the register of the municipality of Amsterdam to make sure these cases were still alive and residing in Amsterdam. All AIDS cases who had died were checked to verify Amsterdam residence at the date of death. Only officially registered residents of Amsterdam were included in our analyses.
HIV Incidence Data
To estimate the course of the AIDS epidemic without the impact of HAART, we used HIV incidence data from 3 sources.
One of these sources was the Amsterdam Cohort Study, which is an ongoing prospective study that started in October 1984.10 Sexually active HIV-seropositive and HIV-seronegative homosexual men participate by visiting the Municipal Health Service every 3 to 6 months. At each visit, data are collected by standardized questionnaires and physical examination. Blood is taken for virologic testing, and, if HIV-positive, for further immunologic testing.
Secondly, HIV incidence has been calculated based on retrospectively identified HIV-positive and HIV-negative samples of participants obtained in the Amsterdam Hepatitis B vaccine trial among homosexual men. This trial started in 1980 and was terminated in 1982. Subsequently, many of the participants in the hepatitis B vaccine trial consented to the retrospective testing of stored serum samples for HIV. Many of these men also became prospective participants in the Amsterdam Cohort Study.
Our third source of data was the cross-sectional HIV prevalence surveys that have been conducted at the sexually transmitted infection clinic of the Municipal Health Service of Amsterdam since 1991 (except for 1993 as a result of logistic problems). For this analysis, we used HIV incidence calculated for homosexual participants in the 1991–2000 HIV prevalence surveys.11 Recent infections were identified using a new testing strategy to detect early HIV infections among people who were HIV-positive.12 Homosexual men who tested HIV-positive with a sensitive test were then retested with a less sensitive test; men who were nonreactive on the less sensitive test were classified as recently infected. The incidence was calculated as the prevalence of persons with a recent infection among the susceptible population (HIV-negative men and recently infected men) divided by the mean time between seroconversion on the assays.
Effect of the Introduction of Treatment on the Risk of Death From AIDS
We used a Poisson regression analysis to test the effect of various treatments on the risk of death after an AIDS diagnosis at the population level. We categorized calendar time as multiyear periods, taking the years 1982–1985 as the reference period. In that earlier period, no therapy was available. The other time periods were defined as 1986–1990 (when Pneumocystis carinii pneumonia prophylaxis and monotherapy were available), 1991–1995 (dual therapy), and 1996–2000 (HAART). Incidence, relative risk (RR), and their 95% confidence intervals (CIs) were calculated. Subjects were considered to be at risk from the date of AIDS diagnosis until the date of death or the end of the study period, which was January 1, 2000. In addition to calendar time, the following covariates, which were associated with death from AIDS in univariate analyses, were considered in the multivariate analysis: age, nationality defined as Western (European, North American, Australian, and New Zealand combined) versus non-Western, sex, and risk group. Because sex and risk group are correlated, we used a combined variable with 11 categories: homosexual men; male and female intravenous drug users; male and female heterosexuals with multiple sexual partners; men and women originating from an AIDS endemic country; men and women with a high-risk partners (ie, bisexual, intravenous drug user, or from an AIDS endemic country). The remaining risk groups were merged as one but divided by sex. AIDS cases belonging to 2 risk groups were classified in the most important risk group, ie, homosexual or drug users.
Effect of HIV Incidence and HAART on AIDS Mortality Time Trends
To separate the course of the HIV epidemic from the effects of HAART on AIDS mortality, the incidence of AIDS cases and deaths without the influence of HAART was estimated. The number of new AIDS cases among homosexual men in a specific calendar year k was determined by multiplying the number of new HIV-infected individuals (λ) in the calendar year j before year k by the probability of developing AIDS k-j years after seroconversion. The numbers thus obtained for each calendar year before year k were added to obtain the total number of cases. The number of AIDS deaths could be determined by the same method. The number of AIDS cases and deaths Nd(k) is expressed by the formula:
We were not able to obtain accurate estimates of the absolute number of HIV seroconversions among the homosexual population of Amsterdam for each year. However, the shape of the AIDS and death curves can be reliably estimated for each year if we know the shape of the HIV seroconversion distribution over calendar time. Therefore, our estimated curves are based on the HIV incidence pattern.
The homosexual men in Amsterdam comprise a dynamic population, with new people entering the risk set over calendar time. If we assume that the total population of HIV-negative men remains fairly constant over time, it follows that the yearly HIV incidence in the Amsterdam Cohort Study represents the HIV dynamics of the Amsterdam population of homosexual men. The annual probability of seroconversion is a less accurate measurement because the probability does not take into account changes on the population size over time.
From October 1984 to April 1985, both HIV-positive and HIV-negative men entered the Amsterdam Cohort Study. A seroconversion density has been estimated for this group,5 and we used this as the seroconversion pattern until 1985. From April 1985 until 1988, only HIV-seronegative men entered the Amsterdam Cohort Study. From 1988 until 1995, the emphasis was on recruiting HIV-positive people, and so very few seronegatives entered the study. In May 1995, a new cohort was started among young (under age 30 years) homosexual men. For 1991 onward, we have supplemented the Amsterdam Cohort Study (ACS) HIV incidence data with data from homosexual men attending the Amsterdam sexually transmitted infection clinic.11 The estimated HIV incidence pattern in the Amsterdam population of homosexual men was derived by visually combining the incidence of the ACS and the sexually transmitted infection (STI) clinic. Because homosexual men attending the STI clinic are assumed to have more risk behavior, more weight is given to the HIV incidence in the ACS.
The density of the distribution of time from seroconversion to AIDS diagnosis and death was based on the Kaplan-Meier estimate from the ACS data in the pre-HAART era using kernel smoothing.5 This density was summarized over yearly intervals to obtain the probability of AIDS diagnosis and death for each year after HIV seroconversion.
To evaluate the effects of HAART by comparing the number of observed deaths with the number of estimated deaths, we combined the 2 curves into one figure. Because the estimated curve does not reflect absolute numbers, its height was determined by matching the 2 curves based on the same cumulative incidence until 1996, when HAART became generally available in The Netherlands.
We are aware that the HIV incidence pattern that was used to estimate the AIDS incidence and death curves might not be representative. Therefore, we carried out a sensitivity analysis. If this pattern does not reflect the HIV incidence pattern in the Amsterdam population of homosexual men, the estimated curves will not be valid. In this analysis, the impact of changes in the HIV incidence on the shape of the AIDS and death curves was determined by moving the peak in HIV incidence to earlier and later times, and also to higher and lower incidences.
From 1982, when the first AIDS case was diagnosed in Amsterdam, until January 1, 2000, 2305 AIDS patients living in Amsterdam had been reported to the AIDS Surveillance System. Of these, 1611 had died by the beginning of 2000; 91% of the total number of AIDS patients were male, and the median age at diagnosis was 38 years (interquartile range: 33–44). Most AIDS cases were homosexual or bisexual men (75%), whereas 12% were intravenous drug users. Heterosexual groups were those with multiple sexual partners (2%), those from AIDS-endemic countries (2%), and those with a partner who is bisexual, injecting drugs, or originating from an AIDS-endemic country (2%).
The annual number of new AIDS cases rose from 1982 until 1992. It then stabilized and began to decline in 1996 (Table 2). The yearly number of AIDS deaths increased until 1992 and started to decline in 1994.
Effect of Changing Treatment Over Time on the Risk of Death From AIDS
The total number of person-years at risk, from AIDS diagnosis until death from AIDS or censor date, was 5666.8. The incidence of death from AIDS decreased over calendar time compared with the 1982–1985 reference period (Table 2). The risk of death from AIDS was lower in 1986–1996 when P. carinii pneumonia prophylaxis (RR = 0.40; 95% CI = 0.29–0.57) and then dual therapy (0.39; 0.28–0.55) became available. The likelihood of AIDS death was more than 10 times lower (0.09; 0.06–0.12) after the introduction of HAART in The Netherlands in 1996. In the univariate analysis, age, western nationality, and the combined variable of gender and risk group (female intravenous drug users, men from AIDS endemic countries, and women from remaining risk groups) were associated with the risk of death from AIDS. Adjustment for these variables and their interaction with calendar time did not substantially change the relative risk for any calendar period.
In Figure 1 the HIV incidence among homosexual men participating in the ACS and attending the STI clinic are shown. The incidence from the ACS peaked in 1984 and then declined. Among homosexual men participating in HIV surveys at the STI clinic, the HIV incidence is presented from 1991 until 2000 and shows a peak in 1997. When the 2 incidence curves are combined, the shape of the resulting curve represents the HIV incidence pattern of the homosexual population of Amsterdam.
Homosexual Men: Observed versus Estimated Curves of AIDS Cases and Deaths
Figure 2A shows the observed number of AIDS cases and deaths from AIDS among homosexual men reported to the AIDS Surveillance System from the beginning of the AIDS epidemic. After a strong increase in the number of AIDS cases, numbers began to stabilize in 1988 and peaked in 1992. Thereafter, the number of homosexual men diagnosed with AIDS began to decrease and continued to decrease until the end of the study period. The number of AIDS deaths increased from one in 1982 to 167 in 1992, remaining fairly stable through 1994 and then dropping to 13 in 1999. The decline in AIDS deaths was steeper and began earlier than the decline in AIDS diagnoses.
Figure 2B shows the estimated shape of the curve of new AIDS cases based on the HIV incidence data from ACS and the STI clinic, as well as the time from HIV seroconversion until AIDS diagnosis in the pre-HAART era. Comparable to the observed AIDS cases, the estimated curve of AIDS cases started to decline in 1992. The estimated curve of AIDS deaths decreased from 1993 onward, beginning 1 year earlier than for the observed cases. As expected, the decline in estimated AIDS cases and deaths was not as steep as the decline in observed cases and deaths.
We found that the sensitivity of the AIDS incidence and death curves was low. The shape of the curves were not affected by moving the peak in HIV incidence (Fig. 1) to higher and lower incidences. As expected, moving the 1984 peak in HIV incidence to earlier or later years simply shifted the curve to earlier or later.
The observed and estimated number of deaths from AIDS among homosexual men in Amsterdam are combined in Figure 2C. The observed and estimated curves cross in July 1994. By subtracting the number of observed deaths from the estimated numbers between 1996 and 2000, we estimated that 331 deaths have been prevented as a result of HAART.
We aimed to evaluate the impact of HAART on AIDS mortality, taking into account earlier HIV incidence patterns. The AIDS epidemic clearly had a major impact on mortality in Amsterdam, because AIDS was the leading cause of death among men age 25–54 years between 1989 and 1996 (data not shown).
We are not aware of any study that has taken into account HIV incidence patterns to determine whether the declines in AIDS mortality are also as a result of decreasing HIV incidence in the past. Other studies have attributed decreases in AIDS mortality to the introduction of HAART without taking into account HIV incidence patterns.1,2,13–15 These studies on mortality primary relied on surveillance data that do not include HIV incidence data. The ACS is one of the few epidemiologic studies worldwide that provide incidence data from the beginning of the AIDS epidemic.7
In the ACS, the yearly HIV incidence declined from 1984 onward among homosexual men and from 1986 among drug users.4,5,16 In the most affected group in Amsterdam, homosexual men, deaths from AIDS peaked in 1992 and then declined. Furthermore, the estimated curves of AIDS cases and deaths without the benefits of HAART in the Amsterdam population of homosexual men showed a decrease in mortality before the introduction of HAART. Therefore, we showed that even without HAART, the number of deaths would have declined because of a decreasing HIV incidence 10 years earlier.
The estimated curves of AIDS cases and deaths were based on an estimated pattern of the HIV incidence among the total population of homosexual men in Amsterdam. This pattern was derived from the HIV incidence data of the homosexual men participating in the ACS and attending the STI clinic. It could be questioned whether this HIV incidence is representative for the total population of homosexual men in Amsterdam. Sexual behavior, age, and history of syphilis have compared between a random sample of homosexual men living in Amsterdam, in 1989 and 1990, and homosexual men participating in the ACS during the same period.17 Their age was similar, but ACS participants reported a higher number of sexual partners in the preceding year, more receptive anogenital intercourse, more lifetime partners, and, more often, a history of syphilis—indicating that the participants of the ACS are more sexually active and might be engaged in risky sexual behavior. This finding suggests that HIV incidence among the general population of homosexual men might be lower. However, we found low sensitivity of AIDS incidence and death to changes in the HIV incidence. Results will change only if the pattern from the unseen population of homosexual men is very different from those excluded in our data sources and moreover if this group has a major contribution to HIV incidence. We consider both of these conditions to be unlikely.
From 1988 until 1995, when the ACS emphasis was on recruiting HIV-positive persons, a small number of HIV-negative homosexual men entered the study. Nonetheless, we believe that the estimated curves of AIDS and death in the first years after 1988 remain valid because a sufficient number of HIV-negative persons entered the ACS before 1988. Afterward, the validity could have decreased as a result of saturation of the participants at risk and increasing age of the participants in the ACS. However, it more likely increased as a result of the 1995 start of a new cohort among young homosexual men and the use of the STI clinic data to estimate the HIV pattern.
We compared the number of AIDS deaths reported to the AIDS Surveillance System of the Amsterdam Municipal Health Service with the mortality data of Statistics Netherlands, a government bureau, that collects this data from death certificates supplied by physicians or coroners (data not shown). The annual numbers of AIDS deaths were compared between the 2 sources and only small differences were seen, except for the first and last years of study. For 1985, Statistics Netherlands showed fewer AIDS deaths, probably reflecting retrospective coding of AIDS deaths. For 1999, Statistics Netherlands shows 19 more AIDS deaths than did the AIDS Surveillance System database—which we learned is incomplete for 1999 because one of the hospitals ceased reporting AIDS cases (an indication that physicians could be less concerned about such reporting in the era of HAART). The 1999 discrepancy suggests that our mortality rate in the most recent years might be underestimated. The incomplete data of the AIDS Surveillance System is unlikely to result from reporting delay, because such delay would mostly influence the most recent years and our calculation in 2003 used data through 1999. Furthermore, there were no changes in the procedure for obtaining data, because the AIDS Surveillance System regularly contacted physicians asking them to report AIDS cases.
The decrease in the observed number of AIDS deaths among homosexual men was not preceded by a decline in the number of new AIDS cases. Also seen in New York City,1 this finding could be explained by the availability of P. carinii pneumonia prophylaxis and dual therapy. In a previous study, we showed that after the introduction of P. carinii pneumonia prophylaxis in Amsterdam in 1985, the 1-year survival for AIDS cases diagnosed with this disease pneumonia improved in 1986 and continued to rise in the following years.18
This study did not encompass the total impact of the HIV epidemic on the mortality in Amsterdam, because it did not take into account the HIV-infected persons who died before an AIDS diagnosis. However, this number is relatively small,19 and among HIV-infected individuals, living in The Netherlands and known to be using HAART, HIV-related mortality decreased and non-HIV related mortality did not change over time.20
In conclusion, our study shows that when evaluating the effect of HAART on AIDS mortality, changes in the HIV incidence pattern should also be taken into account. When we did so, we estimated that 331 deaths have been prevented by HAART in Amsterdam between 1996 and 2000. Although the 1996 introduction of HAART greatly contributed to the decline in AIDS mortality, mortality rates had already begun to drop as a result of a decline in the HIV incidence in the 1980s before the introduction of HAART.
We thank Joke Bax, Maja Totte, and Ans Snuverink for matching the AIDS surveillance data against the register of the municipality of Amsterdam. We thank the physicians in the hospitals of Amsterdam and general practitioners for providing information on AIDS cases and deaths to the AIDS Surveillance System of Amsterdam. We also thank Lucy D. Phillips for editing the manuscript.
1. Chiasson MA, Berenson L, Li W, et al. Declining HIV/AIDS mortality in New York City. J Acquir Immune Defic Syndr. 1999;21:59–64.
2. Hogg RS, O'Shaughnessy MV, Gataric N, et al. Decline in deaths from AIDS due to new antiretrovirals. Lancet. 1997;349:1294.
3. Bindels PJE, Reijneveld SA, Mulder-Folkerts DK, et al. Impact of AIDS on premature mortality in Amsterdam, 1982–1992. AIDS. 1994;8:233–237.
4. van Griensven GJP, de Vroome EMM, Goudsmit J, et al. Changes in sexual behavior and the fall in incidence of HIV infection among homosexual men. BMJ. 1989;298:218–221.
5. Geskus RB. On the inclusion of prevalent cases in HIV/AIDS natural history studies through a marker-based estimate of time since seroconversion. Stat Med. 2000;19:1753–1769.
6. Collaborative Group on AIDS incubation and HIV survival including the CASCADE EU Concerted Action. Time from HIV-1 seroconversion to AIDS and death before widespread use of highly-active antiretroviral therapy: a collaborative re-analysis. Lancet. 2000;355:1131–1137.
7. Koblin BA, van Benthem BHB, Buchbinder SP, et al. Long-term survival after infection with Human Immunodeficiency virus type 1 (HIV-1) among homosexual men in Hepatitis B vaccine trial cohorts in Amsterdam, New York City and San Francisco, 1978–1995. Am J Epidemiol. 1999;150:1026–1030.
8. Centers for Disease Control and Prevention. Revision of the CDC surveillance case definition for Acquired immunodeficiency syndrome. MMWR Morb Mortal Wkly Rep. 1987;36:1S–15S.
9. Ancelle-Park R. Expanded European AIDS case definition. Lancet. 1993;341:441
10. van Griensven GJP, Tielman RAP, Goudsmit J, et al. Risk factors and prevalence of HIV antibodies in homosexual men in The Netherlands. Am J Epidemiol. 1987;125:1048–1057.
11. Dukers NH, Spaargaren J, Geskus RB, et al. HIV incidence on the increase among homosexual men attending an Amsterdam sexually transmitted disease clinic: using a novel approach for detecting recent infections. AIDS. 2002;16:F19–F24.
12. Janssen RS, Satten GA, Stramer SL, et al. New testing strategy to detect early HIV-1 infection for use in incidence estimates and for clinical and prevention purposes. JAMA. 1998;280:42–48.
13. Crocetti E, Giovannetti L. Decreasing AIDS mortality rates among young adults in the city of Florence, 1987–1999. J Epidemiol Community Health. 2002;56:399–400.
14. Karon JM, Fleming PL, Steketee RW, et al. HIV in the United States at the turn of the century: an epidemic in transition. Am J Public Health. 2001;91:1060–1068.
15. Klevens RM, Neal JJ. Update: AIDS-United States, 2000. MMWR Morb Mortal Wkly Rep. 2002;51:592–595.
16. van Ameijden EJC, van den Hoek JAR, Mientjes GHC, et al. A longitudinal study on the incidence and transmission patterns of HIV, HBV and HCV infection among drug users in Amsterdam. Eur J Epidemiol. 1993;9:255–262.
17. Veugelers PJ, van Zessen G, Hendriks JC, et al. Estimation of the magnitude of the HIV epidemic among homosexual men: utilization of survey data in predictive models. Eur J Epidemiol. 1993;9:436–441.
18. Bindels PJE, Poos RMJ, Jong JT, et al. Trends in mortality among AIDS patients in Amsterdam, 1982–1988. AIDS. 1991;5:853–858.
19. Prins M, Sabin CA, Lee CA, et al. Pre-AIDS mortality and its association with HIV disease progression in haemophilic men, injecting drug users and homosexual men. AIDS. 2000;14:1829–1837.
20. Sighem AI, van de Wiel MA, Ghani AC, et al. Mortality and progression to AIDS after starting highly active antiretroviral therapy. AIDS. 2003;17:22–27.
Everything you've always wanted to know about John Snow but were afraid to ask
Log on to The John Snow Archive and Research Companion at www.epi.msu.edu/johnsnow
© 2004 Lippincott Williams & Wilkins, Inc.