adjustrcspline error

[parneet sethi]

I'm trying to fit an stcox model with cubic splines to see the association between mortality and ratiopasit where ratiopasit is the ratio of PA (mins) and sitting time (hours).When I use the adjustrcspline command I get an error.

mkspline2 ssratiopasit =ratiopasit, cubic nknots (5) displayknots
mat ratiopasitknots=r(knots)
matrix list ratiopasitknots

ratiopasitknots[1,5]
knot1 knot2 knot3 knot4 knot5
ratiopasit 0 1.5155344 4.496788 10.675714 35.928143

stcox ssratiopasit1 ssratiopasit2 ssratiopasit3 ssratiopasit4 drage_05 i.sex_05 i.gq5_edu_05_recoded i.smokstat_05recoded i.gq3_marr__05_recoded i.no_comorbid_recoded waist_05

-------------------------------------------------------------------------------------------
_t | Haz. ratio Std. err. z P>|z| [95% conf. interval]
--------------------------+----------------------------------------------------------------
ssratiopasit1 | .8297227 .0597145 -2.59 0.009 .7205641 .9554178
ssratiopasit2 | 1609080 1.18e+07 1.94 0.052 .8802037 2.94e+12
ssratiopasit3 | 1.96e-10 2.35e-09 -1.86 0.062 1.23e-20 3.144156
ssratiopasit4 | 4733.167 24257.47 1.65 0.099 .2054684 1.09e+08
drage_05 | 1.128093 .0040805 33.32 0.000 1.120124 1.136119
|
sex_05 |
female | .6823007 .0531652 -4.91 0.000 .5856657 .7948805
|
gq5_edu_05_recoded |
high school or more | .9878815 .0885629 -0.14 0.892 .8286959 1.177645
|
smokstat_05recoded |
ex-smoker | .4832277 .0583604 -6.02 0.000 .3813734 .6122845
non-smoker | .4035807 .0482261 -7.59 0.000 .3193122 .5100882
unknown category/missing | .4148576 .0792575 -4.61 0.000 .2852851 .6032801
|
2.gq3_marr__05_recoded | 1.289657 .0951568 3.45 0.001 1.116011 1.490321
|
no_comorbid_recoded |
1 | 1.227457 .1227486 2.05 0.040 1.008984 1.493236
2 | 1.45518 .1514774 3.60 0.000 1.186617 1.784526
3 | 1.782304 .1993574 5.17 0.000 1.431435 2.219176
|
waist_05 | 1.005195 .0028721 1.81 0.070 .9995814 1.01084
-------------------------------------------------------------------------------------------

adjustrcspline

all variables created in the last call to mkspline2 must be
independent variables in the last estimation command
r(198);

I've looked for this error on Statalist. Since the function (adjustrcspline) is calling values from e() estimated from the equation, these matrices should not be empty.
. matlist e(b)

| 1b. 2.
| ssratio~1 ssratio~2 ssratio~3 ssratio~4 drage_05 sex_05 sex_05
-------------+-----------------------------------------------------------------------------
y1 | -.1866637 14.29117 -22.35132 8.46235 .1205287 0 -.3822848

| 1b. 2. 1b. 2. 3. 4. 1b.
| gq5_edu~d gq5_edu~d smoksta~d smoksta~d smoksta~d smoksta~d gq3_mar~d
-------------+-----------------------------------------------------------------------------
y1 | 0 -.0121925 0 -.7272673 -.9073788 -.8798199 0

| 2. 0b. 1. 2. 3.
| gq3_mar~d no_com~ed no_com~ed no_com~ed no_com~ed waist_05
-------------+------------------------------------------------------------------
y1 | .2543762 0 .2049448 .3751297 .5779068 .0051814


Can someone shed light on why do I still get this error. Thanks in advance.

[Maarten Buis]

You should have gotten a different error message:

last estimation command (cox) was not recognized
specify the link() or custominvlink() option to manually specify the link function
r(198);

Regardless, there is a bigger problem here: what is your prediction going to mean. Remember that Cox regression is semi-parametric in the sense that it does not estimate the baseline hazard function. For predictions to have a meaning like the "hazard of getting the event", you need the baseline hazard function. This is the reason why Cox regression is not supported by adjustrcspline.