how to deal with missing value in exposure variable

I am examining the effect of A on B. A is exposure variable and B is outcome variable.There are almost 16777 records in the dataset. Out of which 21 records have missing value in A. A is a binary variable with 1 and 2, and my director ask me to treat missing value as a value. Now, A has three value:1,0 and missing. however, i think missing value is scare and random. Shall i remove the observation with missing value?