joinpoint.relaxProp: Relaxing the proportionality assumption
In JPSurv: Joinpoint Model for Relative and Cause-Specific Survival

joinpoint.relaxProp

R Documentation

Relaxing the proportionality assumption

Description

Fitting a joinpoint survival model by relaxing the proportionality assumption

Usage

joinpoint.relaxProp(data, subset, max.cutpoint=5,
       year="Year", interval="Interval", number.event="Died", 
       number.alive="Alive_at_Start", number.loss="Lost_to_Followup",
       expected.rate="Expected_Survival_Interval", 
       observed.rate="Observed_Survival_Interval", 
       model.form=NULL, maxnum.jp=0, 
       proj.year.num=5, op=list(), delLastIntvl=FALSE, add.data.cols=NULL)

Arguments

`data`	Data frame containing all variables in the model.
`subset`	A logical vector of length `nrow(data)`, a character string or NULL to include a particular subset of `data` in the analysis. See Details and Examples.
`max.cutpoint`	A positive integer or NULL giving the number of cutpoints to consider. If NULL, then it will be set to the number of intervals minus one. The default is 5.
`year`	Column name of `data` giving the year or year code. This column must be numeric. The default is 'Year'.
`interval`	Column name of `data` giving the time interval elapsed from the starting time to the event time. This column must be numeric. The default is 'Interval'.
`number.event`	Column name of `data` giving the number of events or deaths. This column must be numeric. The default is 'Died'.
`number.alive`	Column name of `data` giving the number of subjects alive. This column must be numeric. The default is 'Alive_at_Start'.
`number.loss`	Column name of `data` giving the number of subjects lost to followup. This column must be numeric. The default is 'Lost_to_Followup'.
`expected.rate`	Column name of `data` giving the expected interval survival. This column must be numeric. The default is 'Expected_Survival_Interval'.
`observed.rate`	Column name of `data` giving the observed interval survival. This column must be numeric. The default is 'Observed_Survival_Interval'.
`model.form`	an object of class "formula": a symbolic description of covariates. Example: ~-1+age+as.factor(stage)
`maxnum.jp`	The maximum number of join points allowed. The default is zero, which is equivalent to a proportional hazard relative survival model.
`proj.year.num`	The number of projection years for use in the prediction step. Default value is 5 years, with a valid range of 0 to 30 years.
`op`	List of more options. Details — numbetwn: integer value, number of skipped obs between joinpoints exclusive (not count for the joinpoints). Default is 2. numfromstart: integer value, number of skipped obs from the first obs to joinpoints exclusive (not count for the joinpoint). Default is 3. numtoend: integer value, number of skipped obs from the first obs to joinpoints exclusive (not count for the joinpoint). Default is 4.
`delLastIntvl`	an logical value indicating whether or not deleting records of last intervals of all years. The default is FALSE.
`add.data.cols`	Character vector of column names in `data` to add onto the returned data frames of results. Use "_ALL_" to add all columns and use NULL to not add any columns. The default is NULL.

Details

This function finds the optimal clustering of intervals (1, ..., I), where I is the number of intervals, such that there are at most two ordered clusters of the form (1, ..., j) and (j+1, ..., I). For each ordered cluster, a model is fit and the BIC is computed. The algorithm is as follows:
1. Fit the (unconditional) joinpoint survival model on intervals (1, ..., I) and compute the BIC and call it BIC-0.
2. For each cutpoint j, j = 1, ..., max.cutpoint, fit the (unconditional) joinpoint survival model on intervals (1, ..., j) and fit the conditional joinpoint survival model on intervals (j+1, ..., I). Compute the BIC and label it BIC-j.
3. The optimal clustering is the one with minimum BIC = min(BIC-0, BIC-1, ...)

Value

A list of class "jp.relaxProp" with the following objects:

`fit.info`	A data frame containing fitting information from the joinpoint and conditional joinpoint models at each step of the algorithm. The data frame contains the joinpoints and number of joinpoints.
`predicted`	The fitted relative survival rates
`fullpredicted`	The full output matrix, with all year/interval combinations and projections
`fit.uncond`	The fitted model from the (unconditional) joinpoint model corresponding to the best fit.
`fit.cond`	The fitted model from the conditional joinpoint model corresponding to the best fit.
`all.results`	A list containing all the results at each cutpoint. Each element of `all.results` is a list containing `cutpoint`, `fit.cond`, and `fit.uncond`.

References

Yu, B., Huang, L., Tiwari, R. C., Feuer, E. J. and Johnson, K. A. (2009), Modeling population-based cancer survival trends by using join point models for grouped survival data. Journal of the Royal Statistical Society: Series A, 172, 405-425.

Examples

#Load the provided SEER 18 breast cancer example data.
data("breast.example", package="JPSurv")
 
# Subset of observations to use
subset <- "Age_groups == '00-49' & Breast_stage == 'Localized'"
fit    <- joinpoint.relaxProp(breast.example, subset, max.cutpoint=2,
                              year="Year_of_diagnosis_1975")

JPSurv documentation built on June 8, 2025, 12:11 p.m.