Using restricted splines and modelling differences in surgical experience

[Martin Imelda Borg]

Research question: Trying to plot a graph to show the difference in outcomes of surgeons in MESH surgery. The number of cases is valued as surgical experience.
I have found a nonlinear relationship so I have used - mkspline - restricted spline.

//Create spline 
local knots_list 4 10 
mkspline restrictedspline =yearlymesh, knot(`knots_list') cubic

//This creates 1 knot which is = yearlymesh
assert float(yearlymesh) == float(restrictedspline1) //confirms same variable 

//I can either categorise into groups 

gen volgrp = . 
replace volgrp = 1 if yearlymesh <4
replace volgrp = 2 if yearlymesh >=4 & yearlymesh <10 
replace volgrp = 3 if yearlymesh >=10 

//pERFORM a cox regression to see if hazard rate increases with increasing experience
stcox restrictedspline1 covar1 covar2 covar3

PROBLEM:

With backward selection Volgrp got eliminated - i wasn't keen on categorising anyway

I would like to plot a graph of hazards with different variable VolGrp

stcurve, hazard at1(volgrp=1) at 2(volgrp=2)

However, volgrp was eliminated with backward selection, what are my alternatives here?
Should I categorise restrictedspline1 with the same categories as volgrp?


//I would like to assess if there are differences in the type of patients surgeons within different volume groups operate on

Would you recommend using regress?

regress covar1 i.volgp

* Example generated by -dataex-. For more info, type help dataex
clear
input float(mesh primaryconsultant revised gender) str4 operationid float(OpDate revisiondate opyear yearlyrevised yearlymesh)
1 12 0 1 "2e"   16075     . 2004 1 2
1 12 1 1 "1a"   16071 16437 2004 1 2
1 13 1 2 "2a"   16834 16864 2006 1 2
1 13 0 1 "4a"   17086     . 2006 1 2
1 15 0 2 "6a"   17123     . 2006 0 1
1 16 1 2 "5a"   17118 16928 2006 2 2
1 16 1 1 "8a"   17154 18818 2006 2 2
1 16 1 1 "10a"  17186 17918 2007 2 2
1 16 1 1 "9a"   17185 20616 2007 2 2
1 17 0 2 "135d" 17192     . 2007 0 1
1 18 1 1 "143a" 17198 18322 2007 1 1
1 19 0 2 "11a"  17209     . 2007 1 4
1 19 0 2 "124e" 17211     . 2007 1 4
1 19 0 1 "56a"  17214     . 2007 1 4
1 19 1 2 "12e"  17212 21258 2007 1 4
1 20 1 1 "12r"  17240 21655 2007 1 1
1 21 0 2 "78r"  17272     . 2007 1 2
1 21 1 2 "24d"  17303 18034 2007 1 2
1 22 1 1 "980y" 17310 21328 2007 1 1
1 23 0 2 "879r" 17331     . 2007 2 4
end
format %td OpDate
format %td revisiondate