SAS软件分析数据的交互作用

数据是1:2匹配的病例对照数据，已经运用了条件Logistic回归分析了因变量(fpg3)和两个自变量(WHR1和BMI2)的关系，最后想分析它们的交互作用，我在找到了一篇名为《实现logistic与Cox回归相乘相加交互作用的临床实践宏程序》[网址：http://www.doc88.com/p-7087671227853.html]，可是我对于宏程序方面还是小白一个，不是特别明白，老师催着我赶快把数据分析完，我只能通过模仿的方式编写了一个宏程序，但感觉好像有挺多问题的，想请教一下大神，看看哪里有什么错误，感激不尽。
因变量(fpg3)和和两个自变量(WHR1和BMI2)均为二分类变量,id1是对子编号。
其中变量time的得来如下：
data pd1;
set pd;
if FPG3=1 then time=2-FPG3;
if FPG3=0 then time=2-FPG3;
run;

1. %macro additive2dum_LOGISTIC (dat, a, b, statues, cov1, cov2, cov3, cov4, cov5, cov6, cov7, cov8);
   
2. Data p1;
   
3. Set &dat;
   
4.    xx=&a*&b;
   
5. proc phreg outset=est data=p1 covout;
   
6. class &a(ref=first) &b(ref=first)/pram=ref;
   
7. model time*&statues(0)=&a &b xx &cov1 &cov2 &cov3 &cov4 &cov5 &cov6 &cov7/risklimits COVB include=3 selection=stepwise ties=discrete; BETA3=t(BETAP[,2:4]); V3=COV[1:3,2:4];
   
8. strata id1;
   
9. Constrast “OR10” &a 1 &b 0 xx 0/estimate=exp;
   
10. Constrast “OR01” &a 0 &b 1 xx 0/estimate=exp;
    
11. Constrast “OR11” &a 1 &b 1 xx 1/estimate=exp;
    
12. Run;
    
13. 
    
14. Proc export data=estoutfile= “D:\&b” DBMS=excel replace;
    
15. Run;
    
16. %mend additive2dum_LOGISTIC;
    
17. 
    
18. %let ICR_LOGISTIC=%str(
    
19.      Proc print data=est;
    
20.      Run;
    
21.      Data parmscovb;
    
22.         Set est;
    
23.         Drop _Inlike_;
    
24.         If_type_ =‘PARMS’ then output parms; elseif_type_=’COV’ then output covb;
    
25. run;
    
26. proc IML;
    
27. Use parms; Read ALL var _all_ INTO BETAP;
    
28. Close parms;
    
29. print betap;
    
30. use covb; read all var _all_ into cov;
    
31. close covb;
    
32. print cov;
    
33. BETA3=t(BETAP[,2:4]); V3=COV[2:4,2:4];
    
34. Print BETA3 V3;
    
35. C=BETA3[1,]+BETA3[2,]+BETA3[3,];
    
36. D=EXP(C)-1;
    
37. I=EXP(BETA3[1,])+EXP(BETA3[2,])-1;
    
38. F=EXP(BETA3[1,])+EXP(BETA3[2,])-2;
    
39. H=EXP(C)/D;
    
40. ICR=EXP(C)-EXP(BETA3[1,])-EXP(BETA3[2,])+1;
    
41. AP=ICR/EXP(C);
    
42. SI=D/F;
    
43. A=J(3,1,0);
    
44. A[1,]=EXP(C)-EXP(BETA3[1,]);
    
45. A[2,]=EXP(C)-EXP(BETA3[2,]);
    
46.      A[3,]=EXP(C);
    
47.      VARICR=T(A)*V3*A;
    
48.      Z=ICR/SQRT(VARICR);
    
49.      PVAL=2*(1-PROBNORM(ABS(Z)));
    
50.      L_ICR=ICR-1.96*SQRT(VARICR);
    
51.      H_ICR=ICR+1.96*SQRT(VARICR);
    
52.      B=J(3,1,0);
    
53.      B[1,]=(EXP(BETA3[2,]-1))/EXP(C);
    
54.      B[2,]=(EXP(BETA3[1,]-1))/EXP(C);
    
55.      B[3,]=I/exp(C);
    
56.      VARAP=t(B)*V3*B;
    
57.      L_AP=AP-1.96*SQRT(VARAP);
    
58.      H_AP=AP+1.96*SQRT(VARAP);
    
59.      S=J(3,1,0);
    
60.      S[1,]=H-EXP(BETA3[1,])/F;
    
61.      S[2,]=H-EXP(BETA3[2,])F;
    
62.      S[3,]=H;
    
63.      VRS=t(S)*V3*S;
    
64.      L_LSI=log(SI)-1.96*SQRT(VARS);
    
65.      H_LSI=log(SI)+1.96*SQRT(VARS);
    
66.      L_SI=exp(L_LSI);
    
67.      H_SI=exp(H_LSI);
    
68.      Print ICR L_ICR H_ICR AP L_AP H_AP SI L_SI H_SI Z PVAL;
    
69. RUN;
    
70. QUIT;
    
71. Option mcompilenote=all;
    
72. 
    
73. %macro addit_sign(dat, a, b, y, beta0, beta1, beta2,beta3);
    
74. Proc nlmixed data=&dat DF=10000;
    
75.   PAEMS b0=&beta0 b1=&beta2 b3=&beta3;
    
76.   Odds=exp(b0+b1*&a+b2*&b+b3*&a*&b);
    
77.   pi=odds/(1+odds);
    
78.   model &y ~ Binary(pi);
    
79.   estimate ‘ICR=ICR using OR’ exp(b1+b2+b3)-exp(b1)-exp(b2)+1;
    
80.   estimate ‘AP using OR’ (exp(b1+b2+b3)-exp(b1)-exp(b2)+1)/(exp(b1+b2+b3));
    
81.   estimate ‘SI=ICR using OR’ (exp(b1+b2+b3)-1)/(exp(b1)+exp(b2)-2);
    
82. run;
    
83. %mend addit_sign;
    
84. Option mcompilenote=all;
    
85. 
    
86. %macro addit_sign1(dat, a, b, y, beta0, beta1, beta2, WaldICR);
    
87. Proc NLP data=%dat VARDEF=N technique=dbldogpcov;
    
88.    PARMS b0=&beta0, b1=&beta1, b2=&bata2, ICR=&WaldICR;
    
89.    b_interaction=log((ICR+exp(b1)+exp(b2)-1)/(exp(b1)*exp(b2)));
    
90. if &y=1 then p=exp(b0+(b1*&a)+(b2*&b)+(b_ interaction*&a*&b))/(1+exp(b0+(b1*&a)+(b2*&b)+(b_ interaction*&a*&b)));
    
91.        else p=1/(1+exp(b0+(b1*&a)+(b2*&b)+(b_ interaction*&a*&b)));
    
92.        loglike=log(p);
    
93.        max loglike;
    
94.        PROFILE ICR/ALPHA=0.05;
    
95.    RUN;
    
96. %mend;

复制代码

最后的调用宏程序是：

1. %additive2dum_LOGISTIC(pd1, BMI2, whr1, FPG3, culture1, smoke, gaoya1, diya1, LDL_C1, TG1, HDL_C1, TC1);
   
2.    &ICR_LOGISTIC;
   
3. %addit_sign(pd1, BMI2, whr1, FPG3, -0.1882, 0.3241, 0.2377, 0.3312);
   
4. %addit_sign1(pd1, BMI2, whr1, FPG3, -0.1882, 0.3241, 0.2377, 0.7917);

复制代码

请问问题解决了，我也想学一下sas交互作用分析。

您好，请问问题解决了吗？可以分享下程序吗

为什么要用宏程序呀？logistic和cox 不可以做吗？