如何将‘面板数据’结构转化为适合Cox回归的数据结构？

Question

我编写了一些虚拟数据，其结构与我们无法共享的内部数据非常相似。 我想知道如何为 Cox 回归做准备。

您能帮我将虚拟数据预处理成适合 Cox 回归的结构吗？生成虚拟数据的代码如下。 请考虑到一些“吸烟者”在死亡前几年才开始吸烟。此信息必须包含在您可以传递给 Cox 回归的数据中。

    data Cox_data;
    personID=1; *smoker who died at the age of 9;
    do age=1 to 10;
        if age<=8 then dead='N'; else dead='Y';
        if age<=6 then no_cigars=0; else no_cigars=age/2; * # of cigarettes smoked per day;
        if dead='Y' then no_cigars=.;
        output;
    end;
    personID=2;         *smoker who died at the age of 6;
    do age=1 to 10;
        if age<=5 then dead='N'; else dead='Y';
        if age<=3 then no_cigars=0; else no_cigars=age/2.2;
        if dead='Y' then no_cigars=.;
        output;
    end;
    personID=3;         *smoker who died at the age of 13;
    do age=1 to 15;
        if age<=12 then dead='N'; else dead='Y';
        if age<=7 then no_cigars=0; else no_cigars=age/3;
        if dead='Y' then no_cigars=.;
        output;
    end;
    personID=4;         *non-smoker died at the age of 17;
    do age=1 to 18;
        if age<=16 then dead='N'; else dead='Y';
        if age<=20 then no_cigars=0; else no_cigars=age/2;
        if dead='Y' then no_cigars=.;
        output;
    end;
    personID=5;         *smoker who did not die during the measured time;
    do age=1 to 13;
        if age<=15 then dead='N'; else dead='Y';
        if age<=6 then no_cigars=0; else no_cigars=1;
        if dead='Y' then no_cigars=.;
        output;
    end;
    personID=6;         *non-smoker who did not die during the measured time;
    do age=1 to 11;
        if age<=15 then dead='N'; else dead='Y';
        if age<=15 then no_cigars=0; else no_cigars=1;
        if dead='Y' then no_cigars=.;
        output;
    end;
    run;

Answer 1

展示您的模型和所需的数据会很有帮助。 我只是在这里猜测。

data tte;
   set cox_data;
   by personID dead notsorted;
   if first.dead and dead eq 'Y' then do;
      tte=age;
      censor=0;
      output;
      end;
   if last.PersonID and dead eq 'N' then do;
      tte=age;
      censor=1;
      output;
      end;
   drop no_cigars;
   run;
   
proc summary data=cox_data nway;
   class personID;
   output out=no_cigars(drop=_:) mean(no_Cigars)=;
   run;

data tte;
   merge tte no_cigars;
   by personid;
   run;

proc phreg data=tte;
   model tte*censor(1) = no_cigars;
   run;