观测可能有十几万条，您确定数据集可以有十几万个变量吗？

转置完了，就相当于有多少个变量就有多少个观测，然后问题就转化为只输出含有缺失值的观测了

这个得看你电脑限制了，一般来说几万变量没问题，我刚刚用我电脑试了一下跑100万的变量数据，还是可以运行的，只是时间稍微有点长，花了一分钟左右才跑完。

Yes. It can be done easily. You don't even need to have the variable names in advance.

See example below. All b's are missing of all values.

data t1;
   
   retain b1-b100 .;
   retain a1-a5 1;
   retain c 'c';
   do i=1 to 10;
      output;
        end;
run;

proc means data=t1 noprint;
var _numeric_;
output out=check(drop=_:) nmiss= ;
run;

data _null_;
  if 0 then set t1 nobs=nobs;
  set check;
  array var_names(*) _numeric_;
  do i=1 to dim(var_names);
  if var_names(i)=nobs then do;
     variable_name=vname(var_names(i));
         put "variable_name=" variable_name;
  end;
  end;
run;

If the data set is huge. Then you can run first 10000 or randomly select 10000 obs. The likelihood of All 10000 values being missing is very small.