CBIIT/R-PIMixture: Prevalence Incidence Mixture Models

This package fits Prevalence Incidence Mixture models to data for which the time to event is interval censored or for which the event is prevalent at the time zero but is only partially observed. Such data often arises in medical screening for asymptomatic disease or disease precursors, such as precancerous lesions. In such data, 1) onset of incident disease occurs between screening visits (interval-censoring), 2) the disease may have already occurred before the initial screen (prevalent at time zero) but may be initially missed and found some time after the initial screen. These models estimates absolute and relative risks. Semi-parametric, weakly-parametric (integrated B-splines), and fully parametric members of the Prevalence-Incidence Mixture model family are supported. A non-parametric estimator is provided and is useful for checking parametric assumptions of fully parametric Prevalence Incidence Mixture models. Only weakly-parametric and semi-parametric models (no variance calculation) currently support stratified random samples in two frames: a superpopulation and a finite population (where cases and controls at time zero can be a stratum factor). A later version will add this functionality for the logistic-Weibull and logistic-exponential models. Semi-parametric, weakly-parametric models, logistic-Weibull, and logistic-exponential uses a logistic regression model as the prevalence model and a proportion hazard survival model as the incidence model. The semi-parametric model makes no assumptions regarding the baseline hazard function. However, it can be computationally expensive. The weakly-parametric model approximates the bazeline hazard function using integrated B-splines and is faster. When parametric assumptions can be made, the fully parametric models are fastest. The following parametric assumptions are supported: logistic-Weibull, logistic-exponential, logistic-lognormal, logistic-loglogistic, logistic-gengamma, and logistic-gamma. Variance estimates are available only for the weakly-parametric, logistic-Weibull, logistic-exponential, and non-parametric. For the non-parametric, this is achieved through boot-strapping by setting the conf.int parameter to "TRUE" and can be very computationally expensive. For identifiability of the mixture model, the data must contained observed prevalent disease and interval-censored incident disease.

install.packages("remotes")
remotes::install_github("CBIIT/R-PIMixture")