0
$\begingroup$

'''

  1. coxph(formula = Surv(time, hyper) ~ age + drink + smoke + sodium * tchl * BMI, data = male)
  2. coxph(formula = Surv(time, hyper) ~ age + drink + smoke + sodium * tchl * BMI, data = female)
  • The 3-term variables are factors (=categorical)
  • male patient:1506, control: 162
  • female patient:1548, control: 363
  • time: 7 periods

'''

 
The result of 3 terms interaction of 1 (male, n=1668)) : 

                         coef  exp(coef)   se(coef)      z     Pr(>|z|)

 sodium:tchl:BMI        NA         NA    0.000e+00     NA       NA 


The result of 3 terms interaction of 2 (female, n=1971) : 

<&nbsp>                 coef     exp(coef)   se(coef)     z    Pr(>|z|)

 sodium:tchl:BMI       4.545e-01  1.575e+00  1.582e+00  0.287    0.774

Why is the result of the male NA?

Improve this question
$\endgroup$
4
  • $\begingroup$ Please edit the table to provide a cross-table of the sodium, tchl, and BMI categorical predictors for males. My guess is that one combination category will have either no members or no events. Please also edit to define "tchl"; I infer that to be "total cholesterol" but I don't think that's a standard abbreviation. Finally, consider why you are modeling these continuous variables as categorical. I recall a lot of confusion when someone I know was diagnosed with "low sodium" when the value was 134, just below the arbitrary low cutoff of 135. $\endgroup$
    – EdM
    Commented Apr 23, 2022 at 13:23
  • $\begingroup$ Thank you for your reply. I wondered why the different output came out for the same data types (males and females). And yes, tchl indicates total cholesterol. I abbreviated just for my convenience in my R code. The reason I divided the variables into categories is just for the group comparison, I understand your points, though. I don't think there are 0 but I'll check. thanks again. $\endgroup$
    – Nayeon
    Commented Apr 27, 2022 at 6:17
  • $\begingroup$ Make sure to check whether there are events in each combination of predictor values. Even if you have cases, without events the Cox model won't be able to calculate a finite coefficient estimate. Please show that breakdown by editing the question, otherwise there really isn't much more help to provide. $\endgroup$
    – EdM
    Commented Apr 27, 2022 at 11:58
  • $\begingroup$ You may mean the following, which I found in the document. I guess 0 might be generated in the process of interaction calculation. Thank you. If the model contains strata by covariate interactions, then the y matrix may contain structural zeros, i.e., deaths (rows) that had no role in estimation of a given coefficient (column). $\endgroup$
    – Nayeon
    Commented Apr 28, 2022 at 9:27

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.