cases. In 1986 New York City predicted that 8,279 new cases of AIDS would be diagnosed during 1991.12 But, in 1987, with new data, the City had to increase ...
PREDICTING THE FUTURE OF THE AIDS EPIDEMIC AND ITS CONSEQUENCES FOR THE HEALTH CARE SYSTEM OF NEW YORK CITY MICHAEL H. ALDERMAN, M.D.* ERNEST E. DRUCKER, PH.D.,* ALAN ROSENFIELD, M.D., * AND CHERYL HEALTON, M.P.A.** T HE AIDS epidemic and its consequences are a public health challenge l unprecedented in modern history. Despite considerable progress in basic HIV research, several areas remain-medical care, prevention and social policy-where there is inadequate experience or data upon which to base effective responses. This is also true of the problem of predicting the probable course and magnitude of the epidemic.1 An accurate and complete picture of the prevalence of HIV infection and the incidence of HIV related disease are not yet available, nor is the timing or risk for development of AIDS among those infected known. The full-blown AIDS syndrome appears to emerge within five to eight years in 30 to 50% of those infected,2,3,4 but some investigators believe that the median latency may prove to be as long as 10 to 12 years.5 Moreover, the fraction of HIV infected persons who will ultimately develop AIDS remains unknown. Finally, there is no evidence, so far, that the risk of transmitting HIV ever disappears after HIV infection, but neither the duration nor pattern of contagiousness are well understood. LIMITATIONS OF PREDICTIVE MODELS BASED ON CASE REPORTS A statistical approach to the prediction of any epidemic says, in essence, that the future course will follow a pattern which can be discerned from past occurrences. In the past, statistical models of this sort have served well for epidemics in which infection was characteristically followed by a rather short and uniform incubation period, e.g., measles. For AIDS, however, there is neither uniformity of incubation nor, among those infected, consistency of disease incidence over time. In these circumstances, mathematical models may prove less reliable because some vital pieces of information (e.g., transmission efficiency) are not yet available.1 Nonetheless, a model based upon *Department of Epidemiology and Social Medicine, Montefiore Medical Center, Albert Einstein College of Medicine, Bronx, New York **Columbia University School of Public Health, New York, New York
Vol. 64, No. 2, March 1988
176
M.H. ALDERMAN AND OTHERS
the application of a polynomial quadratic formula best fitted to adjusted case counts and transformed by the Box-Cox method has been utilized by the Centers for Disease Control for its national estimates.6,7 Because of the unknown latency that separates infection from expression as clinical disease, as well as uncertainty about cofactors that influence the natural history of HIV infection, it is possible that different groups of patients may experience different courses. For example, those AIDS cases being seen now may reflect only the characteristics of the early stage of the epidemic -a period during which a few "rapid progressors" account for most cases. But many more infected persons may be "slow progressors" who nevertheless develop the disease later. Using the first cases to predict the future course may underestimate the final cumulative total of disease incidence.8.9 Unfortunately, no sufficiently representative cohorts have been followed prospectively in which the number of persons infected, when they became infected, and their subsequent annual incidence of AIDS is known. But there are some preliminary studies of small patient groups whose time of seroconversion is roughly known.1011I These data reveal, in contrast to measles or mumps, that there is no fixed incubation period for HIV disease, but that AIDS incidence increases over time, particularly after five years. It is this dissociation between HIV spread and disease occurrence that limits the long-term validity of any regression line constructed during the early phase of the epidemic. Nonetheless, most public health agencies, including the New York City Department of Health, apply this type of statistical model to estimate future cases. In 1986 New York City predicted that 8,279 new cases of AIDS would be diagnosed during 1991.12 But, in 1987, with new data, the City had to increase its projection by 7% to predict 8,833 incident cases during 1991. Such updates are necessary to keep the future in line with the past but reflect the inherent limitations of any statistical approach in a rapidly changing situation. AN EPIDEMIOLOGIC APPROACH
Ultimately, of course, the number of future AIDS cases depends upon biological relationships that might be expressed in mathematical terms but are not determined by them. As noted above, there is now some preliminary evidence about the growth of HIV prevalence and the incidence of AIDS cases in small but highly selected populations. These cohorts were assembled early in the epidemic, before significant numbers of cases had occurred. It is, therefore, reasonable to believe that the subsequent experience of these groups genuinely reflects the true incidence of the cohort and suffers less Bull. N.Y. Acad. Med.
THE AIDS EPIDEMIC
177
from the potential bias to which a later "prevalence cohort" might be subject.'O Using these data, it is possible to produce plausible estimates of the course of the epidemic by approximating the number of people in any population harboring the HIV virus, ascribing probable time of acquisition of the infection, and then, by applying the annual incidence rates, to predict the number of cases of AIDS in future years. This method of projecting AIDS cases differs from statistical regression lines based on observed AIDS cases because it incorporates available biological and epidemiological knowledge into the process of predicting the future course of the epidemic. This alternate approach, however, does depend upon the acceptance of some assumptions. The first of these relates to the size of the infected population. A widely quoted estimate of the number of infected individuals in New York City is between 400,000 and 500,000. The New York City Department of Health believes that among the City's approximately 200,000 to 250,000 intravenous drug users, more than 50% are infected -about 110,000 men and 30,000 women. Moreover, they estimate that an additional 15,000 female sex partners of these intravenous drug using men also harbor HIV. To these can be added 250,000 infected homosexual and bisexual men, for a total of 405,000.13 Even such a large and reasonably well supported estimate of HIV prevalence does not include hemophiliacs, recipients of infected blood, Haitians, children, and a hopefully still small group of others infected strictly through heterosexual contact. A further critical unknown is the date of acquisition of infection. There is general agreement that infection in the American population began during the mid to late 1970s. The Centers for Disease Control cohort study of homosexual men in San Francisco found a pattern of rapid HIV penetration into that population during the years 1978 to 1984. In New York City, with its larger and more diverse population, the most aggressive spread of HIV infection into the high risk groups (intravenous drug users and homosexual men) may have occurred in a slightly different time frame. In any case, it is reasonable to assume it had occurred prior to 1987. The final element needed to construct the future course of AIDS occurrence would be the annual incidence of AIDS diagnoses among each cohort of HIV carriers. The San Francisco cohort study provides a template that might, with appropriate caveats, be applied to New York. In a group of 155 homosexual men recruited from attendees at sexually transmitted disease clinics, whose frozen serum samples had been stored, a date of seroconversion could be assigned."I For this group, the cumulative incidence of AIDS Vol. 64, No. 2, March 1988
178
M.H. ALDERMAN AND OTHERS
diagnosis at 88 months after seroconversion was 36% (c. i. = 26-46%) with an additional 26% having ARC, and only 38% remaining healthy (later findings indicate 20% healthy at 96 months). 14 It is crucial to note that in this study the 36% cumulative AIDS incidence did not occur in a linear fashion. Instead, few cases (
