Updates to Joinpoint Analyses in the SEER Cancer Statistics Review, 1975-2000
Since the first release of the SEER Cancer Statistics Review (CSR) on April 15, 2003, the tables with joinpoint analyses were updated on June 5, 2003 to include delay-adjusted joinpoint analyses for selected cancer sites. In addition, two slight modifications were made to the joinpoint software. The joinpoint software chooses the model that best describes the observed (or delay-adjusted) data in order to estimate where the trends change (the joinpoints) and the rate of increase or decrease (the Annual Percent Change (APC)) over each line segment. In addition, the program estimates whether the APC for each segment is significantly different from zero. The calculation for this significance test was modified after it was discovered that the original calculations did not match those proposed in the 1980 published paper by Lerman1. The effect on the CSR was to change a few non-significant APCs to significant increases or decreases (modified CSR on 7/30/2003). The second modification had the effect of changing back a few of these newly significant APCs to non-significant (modified CSR on 9/4/2003)2. This test of Lerman has been shown to work well for larger numbers of observations, but has not been fully studied for smaller numbers and should be interpreted with caution. In contrast, the permutation methods used to choose the best joinpoint model (which have not been modified) have been studied for smaller sample sizes, and in the chosen model all joinpoints may be interpreted as representing significant changes in trend.
1Lerman, P.M. (1980) "Fitting Segmented Regression Models by Grid Search" Applied Statistics 29:77-84.
2APCs that changed to non-significant:
Table II-1 and Figure I-21: All Sites/Male and female/All races delay-adjusted incidence (1995-2000) APC=0.6 is not significant.
Table III-1: Brain and Other Nervous System/White/Male mortality (1975-1977) APC=5.0. The p-value changed from 0.04970 to 0.05265, so it is not significant.