 |
Methodology for Characterizing Trends |
Appendices |
In order to obtain a consistent characterization of population trends in factors related to the prevention, early detection, or treatment of cancer, the joinpoint statistical methodology was used in this report. This methodology characterizes a trend using joined linear segments on a logarithmic scale; the point where two segments meet is called a "joinpoint." The methodology has previously proven useful in characterizing trends in cancer incidence and mortality rates (e.g., in the Annual Report to the Nation on the Status of Cancer, 1975–2004, Featuring Cancer in American Indians and Alaska Natives).
The joinpoint software (Joinpoint Version 3.0) uses statistical criteria to determine:
- The fewest number of segments necessary to characterize a trend
- Where the segments begin and end
- The annual percent change (APC) for each segment. (A linear trend on a log scale implies a constant annual percent change.)
In addition, a 95-percent confidence interval around the APC was used to determine if the APC for each segment differed significantly from zero. Whenever possible, weighted regression lines (utilizing standard errors) were calculated using the joinpoint software. Using a log response variable, the weight (motivated by the delta method) equals the square of the response variable divided by the square of the standard error. If the standard errors were unavailable, an unweighted regression was used.
Using the results of these analyses, we characterize trends in this report with respect to both their public health importance and statistical significance. If a trend was:
- Changing less than or equal to 0.5% per year (-0.5 ≤ APC ≤ 0.5), and the APC was not statistically significant, we characterized it as STABLE
- Changing more than 0.5% per year (APC < -0.5 or APC > 0.5), and the APC was not statistically significant, we characterized it as NON-SIGNIFICANT CHANGE
- Changing with a statistically significant APC > 0, we characterized it as RISING
- Changing with a statistically significant APC < 0, we characterized it as FALLING
While these categorizations are somewhat arbitrary, they do provide a consistent method to characterize the trends across disparate measures. However, statistical significance in addition to the absolute value of change for incidence and mortality trends were used to ensure consistency with all major publications on national cancer trends.
To avoid statistical anomalies, segments had to contain at least three observed data points, and no segment could begin or end closer than three data points from the beginning or end of the data series. The maximum number of segments was limited to four (i.e., three joinpoints), because for most practical situations this has been shown to be sufficient, and the calculations become computer intensive when searching for all possible model-fits with many segments.
However, because we constrained the joinpoint models to those in which no segment could begin or end closer than three data points from the beginning or end of the data series, if there were four data points or fewer, only one segment could be fit; from five to seven data points, up to two segments could be fit; and from eight to 10 data points, up to three segments could be fit. To avoid some of these limitations, for two to six data points we connected the data points to determine the APC for each time period, and then employed a two-sample test using the standard errors derived from the survey to determine the statistical significance of the change across periods.
Age adjustment (to a standard population) for measures was done using the direct method of standardization. Whenever possible, age adjustment for measures was done using the age adjustment groups specified for Healthy People 2010 age-adjusted measures (http://wonder.cdc.gov/data2010/aagroups.htm). The year 2000 standard population for specific age groups is available in Klein and Shoenborn (2001). For cancer incidence, 19 age groups were used with the 2000 standard population as specified in http://seer.cancer.gov/stdpopulations).
References:
- Kim HJ, Fay MP, Feuer EJ, Midthune DN. Permutation tests for joinpoint regression with applications to cancer rates. Stat Med 2000;19:335-351.
- Espey DK, Wu X, Swan J, Wiggins C, Jim M, Ward E, Wingo PA, Howe HL, Ries LAG, Miller BA, Jemal A, Ahmed F, Cobb N, Kaur JS, Edwards BK. Annual Report to the Nation on the Status of Cancer, 1975–2004, Featuring Cancer in American Indians and Alaska Natives. Cancer; Published online, October 15, 2007 (DOI: 10.1002/cncr. 23044).
http://seer.cancer.gov/report_to_nation/
- Joinpoint Regression Program, Version 3.0, April 2005, National Cancer Institute. http://srab.cancer.gov/joinpoint.
- Klein RJ, Schoenborn CA. Age adjustment using the 2000 projected U.S. population. Healthy People 2010 Statistical Notes, No. 20. Hyattsville, Md.: National Center for Health Statistics. January 2001. (http://www.cdc.gov/nchs/data/statnt/statnt20.pdf)
