In haraldwf/nordpred: Prediction of cancer incidence (as used in the Nordpred project)

Background

The Nordpred software package is a package of R and S-PLUS functions for predicting trends in cancer incidence, using a special version of the Age-Period-Cohort model (APC). The package is originally written by Harald Fekjaer and Bjorn Moller at the Cancer Registry of Norway as part of a larger cancer prediction project with financial support from the Nordic Cancer Union (NCU).

Case study

library(nordpred)

Data sets

Assume we have thre data sets, all with the following properties:

• All cells are numeric (the data set could be a matrix)
• Each row represents an age group
• Each column represents a time interval

In this example, all data concerns Norwegian males. Cancer incidence concerns colon cancer.

1 Observed population

The example data set inpop1 gives observed population numbers for five year intervals from 1958 to 1997.

knitr::kable(inpop1)

2 Predicted population

The example data set inpop2 gives predicted population numbers for five year intervals from 1998 to 2022.

knitr::kable(inpop2)

3 Cancer incidence

The example data set indata gives observed number of cancer incidences from the same time period as inpop1.

knitr::kable(inpop2)

Population

Functions in the nordpred package expect one data set for population data (both observed and predicted values).

inpop <- cbind(inpop1, inpop2)

Run predictions

Use nordpred.estimate to estimate a model

est <- nordpred.estimate(cases=indata, pyr = inpop, noperiod = 4, startestage = 5)

Then, use the outcome (a nordpred object) for prediction by nordpred.prediction

res <- nordpred.prediction(est, startuseage = 6, cuttrend = c(0,.25,.5,.75,.75), recent=TRUE)

Or estimate and predict simultaneously using nordpred (same result)

res <- nordpred(cases = indata, pyr = inpop, startestage = 5, startuseage = 6, 
                noperiods = 4, cuttrend = c(0, .25, .5, .75, .75))

Choose number of periods

The "nordpred"-function can also choose number periods to base predictions on: This is done by listing candidate number of periods in "noperiods". If the goodness of fit test is rejected based on the widest base, the first period is excluded etc.

res <- nordpred(indata, inpop, startestage = 5, startuseage = 6, noperiods = 4:6,
                cuttrend=c(0, .25, .5, .75, .75))

Poisson link

We can also use a poisson link function (instead of the powerlink as used in the nordpred predictions).

est2 <- nordpred.estimate(indata, inpop, 4, 5, linkfunc = "poisson")
res2 <- nordpred.prediction(est2, startuseage = 6, cuttrend = c(0, .25, .5, .75, .75), recent = TRUE)

Results

print(res)
nordpred.getpred(res)
summary(res, printpred = FALSE)

Get results with standardisation:

## World population standard
wstand <- c(0.12, 0.1, 0.09, 0.09, 0.08, 0.08, 0.06, 0.06, 0.06, 0.06,0.05, 
            0.04, 0.04, 0.03, 0.02, 0.01, 0.005, 0.005)

round(nordpred.getpred(res, incidence = TRUE, standpop = NULL), 2)
round(nordpred.getpred(res, incidence = TRUE, standpop = wstand), 2)

Plot results:

plot(res, standpop = wstand)

Plot results with power5 and poisson links:

plot(res2, standpop = wstand)
plot(res, new = FALSE, lty = c(1, 2), standpop = wstand)

Different cut trend scenarios

using average drift (recent = FALSE)

plot(nordpred.prediction(est, startuseage = 6, cuttrend = c(0, 0, 0, 0, 0), 
    recent = FALSE), standpop = wstand, new = TRUE)
plot(nordpred.prediction(est, startuseage = 6, cuttrend = c(1, 1, 1, 1, 1), 
    recent = FALSE), standpop = wstand, new = FALSE, lty = c(1, 2))
plot(nordpred.prediction(est, startuseage = 6, cuttrend = c(0, .25, .5, .75, .75), 
    recent = FALSE), standpop = wstand, new = FALSE, lty = c(1, 4))

haraldwf/nordpred documentation built on May 17, 2019, 2:41 p.m.