01.06.2013 Views

Statistical Methods in Medical Research 4ed

Statistical Methods in Medical Research 4ed

Statistical Methods in Medical Research 4ed

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

384 Further regression models<br />

where m ˆ 3 and …p1, p2, p3† ˆ…1, 1 2 ,0†. The parentheses around the powers of<br />

x are because x …0† is <strong>in</strong>terpreted as log x, as <strong>in</strong> the Box±Cox transformation (see<br />

§10.8).<br />

The form (12.5), with powers p1 < p2 < ... pm tak<strong>in</strong>g real, rather than just<br />

non-negative, <strong>in</strong>teger values, provides a wider selection of curves than conventional<br />

polynomials. However, the family can be widened further by allow<strong>in</strong>g<br />

equality between some of the powers. If pj ˆ pj‡1 then at first sight the<br />

term<br />

reduces to<br />

b jx …pj† ‡ bj‡1x …pj†<br />

…b j ‡ b j‡1†x …pj† ;<br />

<strong>in</strong> other words, the expression really has degree m 1. However, if pj‡1 ˆ pj ‡ d,<br />

then<br />

b jx …pj† ‡ bj‡1x …pj‡d†<br />

ˆ b j x …pj† ‡ bj‡1 x …pj† …x d<br />

1†=d ! b j x …pj† ‡ bj‡1 x …pj† log x<br />

as d ! 0 for suitable b j‡1 , b j . This rather heuristic argument suggests that the<br />

family can be extended to <strong>in</strong>clude products of the form<br />

x …p† log x<br />

and also (see Royston & Altman, 1994) to <strong>in</strong>clude<br />

x …p† …log x† j , j ˆ 2, ..., m 1:<br />

The full def<strong>in</strong>ition of a fractional polynomial of degree m, f m…x; b, p†, depends<br />

on a real-valued m-dimensional vector p with elements p1 p2 ... pm:<br />

where<br />

fm…x; b, p† ˆa ‡ Pm<br />

bjHj…x† Hj…x† ˆx …pj† if pj 6ˆ pj 1 and Hj…x† ˆHj 1…x† log x, ifpj ˆ pj 1:<br />

S<strong>in</strong>ce the coefficients bj enter fractional polynomials l<strong>in</strong>early they can be fitted <strong>in</strong><br />

exactly the same manner as conventional polynomialsÐnamely, us<strong>in</strong>g the<br />

methods of §11.7. The appearance of terms <strong>in</strong> log x and the possibility of terms<br />

such as x<br />

p mean that some fractional polynomials cannot be used directly if<br />

some x are negative or zero. A simple way out of this difficulty is to work with<br />

x ‡ z rather than x, where z is such that x ‡ z is positive for all cases. However,<br />

the choice of z can be a delicate matter.<br />

jˆ1

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!