dpfr_gam: DPFR using gam

In MatthewGrigsby/growthmetrics: Tools for applying statistic modelling and comparison techniques developed by researchers at HBGDKi and the Johns Hopkins University Center for Global NCD Research and Training

Description

DPFR uses penalized functional regression to incorporate all the history of Y up to time t* for subject i in a scalar-on-function regression. At each t* + j the response for subject i is the scalar Y[i,t*+j] and the functional predictor data consists of Y[i,1:t*]. We show how to obtain DPFR dynamic prediction using the function pfr from the refund R package and the gam function from the mgcv R package.

Usage

dpfr_gam(data = HAZ, time = 0:15, hist.lngth = 7)

Arguments

data	A matrix of values with each row representing an individual and each column representing measurements at each time point (e.g. column 1 is time point 1, etc.). Measurements are assumed to be taken at the same time intervals for every participant.
time	A vector of times. This equal the number of columts in the matrix provided (e.g. 0 through 15 months for the example dataset).
hist.lngth	The length of known history for the observed process. We use leave one-curve out cross validation for prediction (hist.lgth is 7 for the example dataset).

Details

Another option for fitting DPFR instead of using pfr, is using the gam function from the mgcv R package. To use gam, there is a step that involves arranging the data in some specific form prior to calling the gam function for DPFR model fitting.

Value

A matrix of predictions with rows representing each individual and columts representing predictions for each time point.

References

Ivanescu AE, Crainiceanu CM, Checkley W. Dynamic child growth prediction: A comparative methods approach. Statistical Modelling. 2017 Dec;17(6):468-93.

MatthewGrigsby/growthmetrics documentation built on May 25, 2019, 8:29 p.m.