Karon JM, Devine OJ; International Conference on AIDS.
Int Conf AIDS. 1989 Jun 4-9; 5: 154 (abstract no. Th.A.P.79).
AIDS Program, Centers for Disease Control, Atlanta, GA, USA
OBJECTIVE: To model time trends in the proportions of AIDS cases in groups, in order to make short-term projections of future proportions. METHODS: Let pi(t) be the proportion of cases in group i, i=1,2,..., G, at time t. Model the generalized logits: log(pi(t)/PG(t)) = fi(t) + error, where fi(t) is a polynomial in time (estimated using weighted regression) and group G is the reference group (fG(t) = 0). Predicted proportions are: Pi(t) = exp(fi(t)) / sigma exp (fi(t)). RESULTS: Predictions are nearly independent of the choice of the reference group. We modeled quarterly incidence of U.S. cases for Jan 1984 through June 1988 grouped geographically into metropolitan statistical areas: New York City, San Francisco, Houston, and others (reference group). Linear polynomials give good fits to observed proportions and predict decreases in each area, with the following observed and predicted percent of U.S. cases: TABULAR DATA, SEE ABSTRACT VOLUME. CONCLUSION: Modeling proportions can give useful short-term projections even with relatively few cases in some groups. Proportions in risk groups can be modeled after the effects of the new case definition on trends are understood. Estimates of uncertainty and prediction intervals can be obtained from a multinomial likelihood incorporating similar models.
Publication Types:
Keywords:
- Acquired Immunodeficiency Syndrome
- Forecasting
- Incidence
- Models, Biological
- New York City
- San Francisco
- Time
- United States
- statistics & numerical data
- therapy
Other ID:
UI: 102176485
From Meeting Abstracts