1.RCS的优点
to characterize a dose-response association between a continuous exposure and an outcome
to visually and/or statistically check the assumption of linearity of the association
to minimize residual confounding when adjusting for a continuous exposure
2.RCS的目的
graphically characterize the dose-response association between a continuous exposure and an outcome
test the assumption of linearity of the association
quantify the association when the latter assumption is not valid
3.RCS的宏程序
By default, the SAS macro creates an RCS function with three knots, located at the 5th, 50th, and 95th percentiles
The macro provides in the SAS OUTPUT window two statistical tests for the main continuous exposure (specified by the user): (i) a test where the null hypothesis is_0=_1=···=_K−2=0 (test of an overall association between the main exposure and the outcome), and (ii) a test where the null hypothesis is_1=···=_K−2=0 (test of a non-linear association between the main exposure and the outcome).
Akaike Information Criteria (AIC) [33] may provide useful information for model selection. The model that better explains the observations, while requiring the lower number of parameters, is the one with the lower AIC.（评价指标）
4.RCS的宏程序使用
the location of knots has a little impact on the shape of the dose-response association. Durrleman and Simon proposed the 5th, 25th, 50th, 75th, and 95th percentiles for a 5-knot RCS function, the 5th, 25th, 75th, and 95th percentiles for a 4-knot RCS function, and the 5th, 50th, and 95th percentiles for a 3-knot RCS function [24] and these locations are the most widely used with RCS functions.（推荐使用节点数）
the number of knots has a non-negligible impact on the shape of the dose-response association.
1.For statistically testing the assumption of linearity of a dose-response association, a 3-knot RCS function, which is much smoother than a 5-knot RCS function, would be more powerful to detect departure from linearity.2.For adjustment for a continuous confounder by using an RCS function, a 3-knot RCS function would also be preferable to a 5-knot RCS function since it is more parsimonious and would remove most of the residual confounding（3节点优点）
A 5-knot RCS function would be desirable for explanatory analyses, when one wants to characterize the dose-response association more precisely (i.e. closer to the data), on thea prioriassumption that the association might be more complex than that accounted by a 3-knot RCS function（5节点优点，何时使用）
In this context, the value of AIC may help to determine if little ‘jumps’ that can be observed with a 5-knot RCS function would be spurious or relevant. If necessary, strategies to determine the optimal modeling can easily be implemented with serial comparison of AIC values, an objective criteria of fitting.（AIC使用）
具体参数解释见文献（Dose-response analyses using restricted cubic spline functions in public health research），程序使用见（http://bbs.pinggu.org/forum.php?mod=viewthread&tid=7139705）
5.RCS的宏程序示例（logistic）
%inc 'C:\sasbishe\RCS_Reg.sas';
%RCS_Reg( infile=all116,
Main_spline_var= ogtt_1, knots_msv= 5 25 50 75 95,
typ_reg=log, dep_var= delivery_mode,
adjust_var= pregage pre_bmi bpa_mean parity smoke drink family_income diabetes_history hpt_history height infant_sex ddif_ogtt_week,
Y_ref_line= 1,exp_beta=0,histogram=1,MAX_XAXIS=6.0);
网友评论