Statistical Methods for Analysis of NF Clinical Data.

摘要

This project describes research in statistical methods that has been usefully for statistical modelling and analysis of clinical data from NF1 and NF2 subjects. The statistical methods are classified into the areas: (a) estimation of familial correlation for different types of data, (b) assessment of multi-hit mutation models for incidence of tumors. Accomplishing the project required development of some new statistical theory, as well as non-trivial computer programming to implement the new and existing theory. One new statistical inference method for familial data is composite likelihood. This method is intended in general for situations in which a multivariate probability is too time consuming to compute because of numerical interactions; for the NF data, this situation occurred when the family size exceeded 4 to 6 (the lower bound on family size depends on the actual multivariate probability). Clinical data exist in many formats including binary, categorical, count, and continuous information. Also common are censored data (e.g., only a lower hound is known for some measurements). One goal of the project was the creation of a software package for familial data analysis for different types of data, such as binary, count, and right-censored survival data. The modules based on composite likelihood were the most recent additions to the package.