Two-part survival models applied to administrative data for determining rate of and predictors for maternal-child transmission of HIV

Walter W. Hauck, Linda J. McKee, Barbara J. Turner

Research output: Contribution to journalArticlepeer-review

12 Scopus citations

Abstract

In analysing maternal-child HIV transmission from Medicaid claims data, we must deal with follow-up that is sometimes so short that we cannot claim that an apparently uninfected infant is actually uninfected as opposed to not yet exhibiting HIV-associated symptoms. To overcome this, we have been using analyses of 'time-to-diagnosis' of HIV infection to estimate transmission rates and predictors of transmission. Such analyses mix the event of transmission with that of our ability to diagnose HIV infection from coded claims data. We would like to separate these two pieces. Also, due to incomplete follow-up, Kaplan-Meier analyses will underestimate transmission rates. In econometrics and biostatistics there are two-part (mixture) models that can serve the goal of separating transmission from the process of diagnosing HIV infection in the newborn. Farewell describes a model that combines a logistic regression for the yes/no event (in our case, HIV transmission) and a Weibull regression model for the survival analysis portion (in our case, time-to-diagnosis). We use this approach to fit models that have potentially separate covariates for transmission and for time-to-diagnosis. The results allow us to identify predictors of transmission and estimate transmission rates with reduced concern for adequacy of follow-up.

Original languageEnglish (US)
Pages (from-to)1683-1694
Number of pages12
JournalStatistics in Medicine
Volume16
Issue number15
DOIs
StatePublished - Aug 15 1997

ASJC Scopus subject areas

  • Epidemiology
  • Statistics and Probability

Fingerprint Dive into the research topics of 'Two-part survival models applied to administrative data for determining rate of and predictors for maternal-child transmission of HIV'. Together they form a unique fingerprint.

Cite this