Biostatistics III in R

In biostat3: Utility Functions, Datasets and Extended Examples for Survival Analysis

Exercise 8. Diet data: Using Poisson regression to study the effect of energy intake adjusting for confounders on two different timescales

Use Poisson regression to study the association between energy intake (hieng) and CHD adjusted for potential confounders (job, BMI). We know that people who expend a lot of energy (i.e., are physically active) require a higher energy intake. We do not have data on physical activity but we are hoping that occupation (job) will serve as a surrogate measure of work-time physical activity (conductors on London double-decker busses expend energy walking up and down the stairs all day).

Fit models both without adjusting for ‘time’ and by adjusting for attained age (you will need to split the data) and time-since-entry and compare the results.

---

Load the diet data using time-on-study as the timescale.

library('knitr')
read_chunk('../q8.R')
opts_chunk$set(cache=FALSE)

You may have to install the required packages the first time you use them. You can install a package by install.packages("package_of_interest") for each package you require.

Load diet data and explore it.

The rates plotted on timescale attained age show a clear increasing trend as age increases, which is to be expected (older persons are more likely to suffer from CHD). The rates plotted on timescale time-since-entry are almost constant (if you have some imagination you can see that the rates are flat).

(a)

(b) ##

Patients with high energy intake have 48% less CHD rate. The underlying shape of the rates is assumed to be constant (i.e. the baseline is flat) over time.

(c) ##

The effect of high energy intake is slightly confounded by bmi and job, since the point estimate changes a little.

(d) ##

The y variable is not correct since it is kept for all splitted records, and contains the complete follow-up rather than the risktime in that specific timeband. The risktime variable contains the correct amount of risktime for each timeband. The event variable chd is not correct since it is kept constant for all splitted records, while it should only be 1 for the last record (if the person has the event). For all other records (timebands) for that person it should be 0.

The effect of high energy intake is somewhat confounded by age, but also confounded by job and bmi.

Fitting the model for CHD, with adjustment for job and bmi.

(e)

There seems to be no confounding by time-since-entry, but there is confounding by BMI and job.

biostat3 documentation built on Oct. 29, 2024, 5:07 p.m.