Specify Factors in multi-state Coxph model

Question

I have an output using the following model:

> unadj3_factor
Call:
coxph(formula = Surv(Tstart, Tstop, status) ~ as.factor(EXPOSURE.2) + 
as.factor(EXPOSURE.3) + as.factor(EXPOSURE.4) + as.factor(EXPOSURE.5) + 
as.factor(EXPOSURE.6) + as.factor(EXPOSURE_2.4) + as.factor(EXPOSURE_2.5) + 
as.factor(EXPOSURE_2.6) + as.factor(EXPOSURE.4 * EXPOSURE_2.4) + 
as.factor(EXPOSURE.5 * EXPOSURE_2.5) + as.factor(EXPOSURE.6 * 
EXPOSURE_2.6) + strata(trans), data = filtercohort, control = coxph.control(iter.max = 5000), 
method = "breslow")

output

                                         coef exp(coef) se(coef)      z                    p
as.factor(EXPOSURE.2)1                    0.370     1.448    0.040    9.3 < 
0.0000000000000002
as.factor(EXPOSURE.3)1                    0.635     1.887    0.099    6.4   0.0000000001377649
as.factor(EXPOSURE.4)1                    0.202     1.224    0.052    3.9   0.0000916916640771
as.factor(EXPOSURE.5)1                    0.395     1.484    0.147    2.7                0.007
as.factor(EXPOSURE.6)1                   -0.477     0.621    0.148   -3.2                0.001
as.factor(EXPOSURE_2.4)1                  0.375     1.455    0.046    8.1   0.0000000000000005
as.factor(EXPOSURE_2.4)99                -4.579     0.010    0.005 -866.2 < 0.0000000000000002
as.factor(EXPOSURE_2.5)1                  0.640     1.897    0.126    5.1   0.0000003825579063
as.factor(EXPOSURE_2.5)99                    NA        NA    0.000     NA                   NA
as.factor(EXPOSURE_2.6)1                 -0.459     0.632    0.129   -3.6   0.0003579592922411
as.factor(EXPOSURE_2.6)99                    NA        NA    0.000     NA                   NA
as.factor(EXPOSURE.4 * EXPOSURE_2.4)1   0.141     1.152    0.215    0.7                0.513
as.factor(EXPOSURE.5 * EXPOSURE_2.5)1  -0.539     0.583    0.733   -0.7                0.462
as.factor(EXPOSURE.6 * EXPOSURE_2.6)1   1.479     4.390    0.426    3.5   0.0005097441097383
as.factor(EXPOSURE.6 * EXPOSURE_2.6)99  3.879    48.395    0.187   20.7 < 0.0000000000000002

Likelihood ratio test=1957129  on 13 df, p=<0.0000000000000002
n= 6996659, number of events= 865524 

The 99's were originally missing values but recoded as "99". The reason is that the coxph model deletes these observations with missing values and these observations need to be retained.

How can I re-run the models by removing these levels from the model ? Or can I use a procedure to include missing values instead of '99's but recode to NA?

as.factor(ABRUPT_ALL_2.4)99
as.factor(ABRUPT_ALL_2.5)99
as.factor(ABRUPT_ALL_2.6)99                        
as.factor(ABRUPT_ALL.6 * ABRUPT_ALL_2.6)99