Reporting methods in studies developing prognostic models in cancer: a review

Table 3 Variables in multivariable analysis

	% (n = 47) articles
Coding of variables in model
Coding explicit for all candidate variables	68 (32)
Coding explicit for all variables in final model‡	70 (33)
Coding of continuous candidate variables
All continuous	2 (1)
All categorised/dichotomised	51 (24)
Combination of continuous and categorised variables	19 (9)
No continuous candidate variables	0 (0)
Unclear/not reported	28 (13)
	Median (IQR) (range)
Number of variables
Candidates used to develop model* (n = 46)	11 (7 to 14) (2 to 27)
Included in final model** (n = 44)	4 (3 to 6) (2 to 12)

‡ Two papers could not be assessed as the final model was unclear.
* In four articles variables were excluded where it was unclear if these were candidate variables. In one article the number of candidate variables was completely unclear
** Three articles could not be included: two because the final model was not presented in the paper, one because it was not clear which of the alternative models in the paper was the final model.

ISSN: 1741-7015