Related Experiment Videos
An exploratory analysis of survival with AIDS using a nonparametric tree-structured approach
J D Piette1, O Intrator, S Zierler
1Center for Gerontology and Health Care Research, Brown University, Providence, RI.
Epidemiology (Cambridge, Mass.)
|July 1, 1992
Summary
Classification and regression trees identified key factors influencing survival in people with human immunodeficiency virus (HIV) disease. Kaposi's sarcoma, central nervous system opportunistic diseases, advanced age, race, and illicit drug use significantly impacted mortality rates.
Area of Science:
- Epidemiology
- Biostatistics
- Medical Informatics
Background:
- Human immunodeficiency virus (HIV) disease presents complex survival challenges.
- Understanding factors influencing HIV/AIDS survival is crucial for public health interventions.
- Opportunistic diseases and sociodemographic characteristics play a significant role in patient outcomes.
Purpose of the Study:
- To analyze survival data in people with HIV disease using classification and regression trees.
- To identify key opportunistic diseases and sociodemographic factors affecting HIV/AIDS survival.
- To compare the utility of tree-based methods with proportional hazards models in epidemiologic research.
Main Methods:
- Application of classification and regression trees to survival data from 43,795 HIV cases (1984-1987).
- Utilized vital status data up to December 31, 1989, to estimate mortality rates.
- Fitted proportional hazards models for comparative analysis.
Main Results:
- Kaposi's sarcoma and CNS opportunistic diseases (cryptococcosis, primary brain lymphoma, CMV disease, PML) were identified as major predictors of death.
- Advanced age (50+), white/other race, and a history of illicit drug use were significant determinants of mortality.
- Tree-structured analysis effectively illustrated survival probability variations across subgroups based on these determinants.
Conclusions:
- Classification and regression trees offer valuable insights into complex survival data in HIV/AIDS research.
- Identified factors like specific cancers, opportunistic infections, age, race, and drug use are critical for understanding HIV mortality.
- The study highlights the utility and limitations of tree-based methods alongside traditional proportional hazards models for epidemiologic analysis.