To show how fractional polynomial methods can usefully replace the practice of arbitrarily categorizing data in epidemiology and health services research.
Methods
A health service setting is used to illustrate a structured and transparent way of representing nonlinear data without arbitrary grouping.
Results
When age is a regressor its effects on an outcome will be interpreted differently depending upon the placing of cutpoints or the use of a polynomial transformation.
Conclusions
Although it is common practice, categorization comes at a cost. Information is lost, and accuracy and statistical power reduced, leading to spurious statistical interpretation of the data. The fractional polynomial method is widely supported by statistical software programs, and deserves greater attention and use.
Get full access to this article
View all access options for this article.
References
1.
RoystonP, AltmanDG, SauerbreiW. Dichotomising continuous predictors in multiple regression: a bad idea. Statistics in Medicine2006; 25: 127–41
2.
AltmanDG, LausenB, SauerbreiW, SchumacherM. Dangers of using “optimal” cutpoints in the evaluation of prognostic factors. Commentary. J National Cancer Institute1994; 86: 829–35
3.
GreenlandS. Avoiding power loss associated with categorization and ordinal scores in dose-response and trend analysis. Epidemiology1995; 6: 450–4
4.
BrennerH, BlettnerM. Controlling for continuous confounders in epidemiological research. Epidemiology1997; 8: 429–34
5.
RoystonP, AmblerG, SauerbreiW. The use of fractional polynomials to model continuous risk variables in epidemiology. Int J Epidemiology1999; 28: 964–74
6.
RoystonP, SauerbreiW. Building multivariable regression models with continuous covariates in clinical epidemiology with an emphasis on fractional polynomials. Methods Inf Med2005; 44: 561–71
7.
ScottDW. On optimal and data-based histograms. Biometrika1979; 66: 3
8.
WandMP. Data-based choice of histogram bin width. American Statistician1997; 51: 59–64
9.
WainerH. Picturing the uncertain world. How to understand, communicate, and control uncertainty through graphical display.Princeton: Princeton University Press2009
10.
GreenlandS. Dose-response and trend analysis in Epidemiology: alternatives to categorical analysis. Epidemiology1995; 6: 356–65
11.
RoystonP, AltmanDG. Regression using fractional polynomials of continuous covariates: parsimonious parametric modelling. Applied Statistics1994; 43: 429–67
12.
MacCallumRC, ZhangS, PreacherKJ, RuckerDD. On the practice of dichotomization of quantitative variables. Psychological Methods2002; 7: 19–40
13.
OddyWH, SherriffJL, de KlerkNH, KendallGE, SlyPD, BeilinLJThe relation of breastfeeding and Body Mass Index to asthma and atopy in children: A prospective cohort study to age 6 Years. American J Public Health2004; 94: 1531–7
14.
HyndmanJCG, HolmanCDJ, P DV. Effect of distance and social disadvantage on the response to invitations to attend mammography screening. J Med Screen2000; 7: 141–5
15.
BrameldKJ, HolmanCD, LawrenceDM, HobbsMST. Improved methods for estimating incidence from linked hospital morbidity data. Int J Epidemiology2003; 32: 617–24
16.
HolmanCD, WisniewskiJB, SemmensJB, RouseIL, BassAJ. Mortality and prostate cancer risk in 19 598 men after surgery for benign prostatic hyperplasia. BJU International1999; 84: 37–42
17.
Stewart WilliamsJA. Using non linear decomposition to explain the discriminatory effects of male-female differentials in access to care. A cardiac rehabilitation case study. Soc Sci Med2009; 69: 1072–9
18.
Stewart WilliamsJA. Evidence and Equity in Healthcare.Newcastle: School of Medicine and Public Health, Faculty of Health, University of Newcastle, 2009
19.
Stewart WilliamsJA, BylesJE, InderK. Equity of access to cardiac rehabilitation: the role of system factors. Inter J Equity in Health2010; 9: 7
SauerbreiW, Meier-HirmerC, BennerA, RoystonP. Multivariable regression model building by using fractional polynomials: description of SAS, STATA and R programs. Computational Statistics & Data Analysis2006; 50: 3464–85
22.
RobertsonC, BoyleP, HsiehC, MacfarlaneGJ, MaisonneuveP. Some statistical considerations in the analysis of case-control studies when exposure variables are continuous measurements. Epidemiology1994; 5: 164–70
23.
WeinbergCR. How bad is categorisation? (Editorial). Epidemiology1995; 6: 345–7
24.
RothmanKJ, GreenlandS. Modern epidemiology.Second edition ed.Philadelphia: Lippincott Williams & Wilkins, 1998
25.
AhrensW, PigeotI, eds. Handbook of epidemiology.Bremen: Springer, 2005
