------------------------------------------------------------------------------------------------------------- name: log: C:\MyDocs\MPH\LectureNotes\ClassDoLogData\Survival\adjust.log log type: text opened on: 3 Nov 2009, 10:23:34 . * adjust.log . * . * Evaluate the adjust postestimation command after hazard regression . * and logistic regression models . * . * The adjust command tabulates the linear predictor, or the exponentiated linear . * predictor evaluated at one or more catigorical variables. It can be used to . * tabulate hazard ratios with their confidence intervals over one or two categorical . * variables, either with or without interaction terms. An example follows. . * . use "C:\WDDtext\2.20.Framingham.dta", clear . gen hypertension = dbp > 90 . stset followup, failure(chdfate) failure event: chdfate != 0 & chdfate < . obs. time interval: (0, followup] exit on or before: failure ------------------------------------------------------------------------------ 4699 total obs. 0 exclusions ------------------------------------------------------------------------------ 4699 obs. remaining, representing 1473 failures in single record/single failure data 3.79e+07 total analysis time at risk, at risk from t = 0 earliest observed entry t = 0 last observed exit t = 11688 . gen men = sex==1 . stcox men##hypertension, nohr failure _d: chdfate analysis time _t: followup Iteration 0: log likelihood = -11834.856 Iteration 1: log likelihood = -11699.217 Iteration 2: log likelihood = -11688.271 Iteration 3: log likelihood = -11688.204 Iteration 4: log likelihood = -11688.204 Refining estimates: Iteration 0: log likelihood = -11688.204 Cox regression -- Breslow method for ties No. of subjects = 4699 Number of obs = 4699 No. of failures = 1473 Time at risk = 37880111 LR chi2(3) = 293.31 Log likelihood = -11688.204 Prob > chi2 = 0.0000 ------------------------------------------------------------------------------ _t | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- 1.men | .7100704 .0631026 11.25 0.000 .5863915 .8337493 1.hyperten~n | .8758088 .0846962 10.34 0.000 .7098072 1.04181 | men#| hypertension | 1 1 | -.3199569 .1137809 -2.81 0.005 -.5429634 -.0969504 ------------------------------------------------------------------------------ . lincom 1.men+ 1.hypertens +1.men#1.hypertens ( 1) 1.men + 1.hypertension + 1.men#1.hypertension = 0 ------------------------------------------------------------------------------ _t | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- (1) | 1.265922 .0796777 15.89 0.000 1.109757 1.422088 ------------------------------------------------------------------------------ . * . * The identical linear predictors given above can be calculated by the adjust . * command as follows. . * . margins men##hypertension Predictive margins Number of obs = 4699 Model VCE : OIM Expression : Relative hazard, predict() ------------------------------------------------------------------------------ | Delta-method | Margin Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- men | 0 | 1.310332 .0450473 29.09 0.000 1.222041 1.398623 1 | 2.369148 .1397233 16.96 0.000 2.095296 2.643001 | hypertension | 0 | 1.450935 .0559711 25.92 0.000 1.341233 1.560636 1 | 2.900332 .1945861 14.91 0.000 2.51895 3.281714 | men#| hypertension | 0 0 | 1 . . . . . 0 1 | 2.400816 .2033401 11.81 0.000 2.002277 2.799356 1 0 | 2.034135 .1283592 15.85 0.000 1.782555 2.285714 1 1 | 3.546362 .2825658 12.55 0.000 2.992543 4.100181 ------------------------------------------------------------------------------ . * . * We now calculate the related hazard ratios from the same model as above . * . stcox men##hypertension failure _d: chdfate analysis time _t: followup Iteration 0: log likelihood = -11834.856 Iteration 1: log likelihood = -11699.217 Iteration 2: log likelihood = -11688.271 Iteration 3: log likelihood = -11688.204 Iteration 4: log likelihood = -11688.204 Refining estimates: Iteration 0: log likelihood = -11688.204 Cox regression -- Breslow method for ties No. of subjects = 4699 Number of obs = 4699 No. of failures = 1473 Time at risk = 37880111 LR chi2(3) = 293.31 Log likelihood = -11688.204 Prob > chi2 = 0.0000 ------------------------------------------------------------------------------ _t | Haz. Ratio Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- 1.men | 2.034135 .1283592 11.25 0.000 1.797491 2.301933 1.hyperten~n | 2.400816 .2033401 10.34 0.000 2.033599 2.834344 | men#| hypertension | 1 1 | .7261804 .0826255 -2.81 0.005 .5810239 .9076011 ------------------------------------------------------------------------------ . lincom 1.men+ 1.hypertens + 1.men#1.hypertens, hr ( 1) 1.men + 1.hypertension + 1.men#1.hypertension = 0 ------------------------------------------------------------------------------ _t | Haz. Ratio Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- (1) | 3.546362 .2825658 15.89 0.000 3.033621 4.145767 ------------------------------------------------------------------------------ . * . * Now repeat these calculations using the adjust command . * . margins men##hypertension // exp ci label(hazard ratio) Predictive margins Number of obs = 4699 Model VCE : OIM Expression : Relative hazard, predict() ------------------------------------------------------------------------------ | Delta-method | Margin Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- men | 0 | 1.310332 .0450473 29.09 0.000 1.222041 1.398623 1 | 2.369148 .1397233 16.96 0.000 2.095296 2.643001 | hypertension | 0 | 1.450935 .0559711 25.92 0.000 1.341233 1.560636 1 | 2.900332 .1945861 14.91 0.000 2.51895 3.281714 | men#| hypertension | 0 0 | 1 . . . . . 0 1 | 2.400816 .2033401 11.81 0.000 2.002277 2.799356 1 0 | 2.034135 .1283592 15.85 0.000 1.782555 2.285714 1 1 | 3.546362 .2825658 12.55 0.000 2.992543 4.100181 ------------------------------------------------------------------------------ . * . * Unfortunately this approach does not work in general with logistic . * regression. This is because in logistic regression the linear predictor . * rarely equals the log of an interesting odds ratio. For example . * . use "C:\WDDtext\5.5.EsophagealCa.dta", clear . gen smoke = tobacco >2 . logit cancer heavy##smoke [freq=patients] Iteration 0: log likelihood = -494.74421 Iteration 1: log likelihood = -445.20189 Iteration 2: log likelihood = -440.58031 Iteration 3: log likelihood = -440.55599 Iteration 4: log likelihood = -440.55599 Logistic regression Number of obs = 975 LR chi2(3) = 108.38 Prob > chi2 = 0.0000 Log likelihood = -440.55599 Pseudo R2 = 0.1095 ------------------------------------------------------------------------------ cancer | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- 1.heavy | 1.776528 .2071724 8.58 0.000 1.370477 2.182578 1.smoke | .7233938 .2311423 3.13 0.002 .2703631 1.176424 | heavy#smoke | 1 1 | -.192998 .3962547 -0.49 0.626 -.969643 .5836469 | _cons | -2.04122 .1270472 -16.07 0.000 -2.290228 -1.792212 ------------------------------------------------------------------------------ . lincom _cons + 1.heavy ( 1) [cancer]1.heavy + [cancer]_cons = 0 ------------------------------------------------------------------------------ cancer | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- (1) | -.2646926 .1636442 -1.62 0.106 -.5854293 .0560442 ------------------------------------------------------------------------------ . lincom _cons + 1.smoke ( 1) [cancer]1.smoke + [cancer]_cons = 0 ------------------------------------------------------------------------------ cancer | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- (1) | -1.317827 .1930953 -6.82 0.000 -1.696286 -.9393668 ------------------------------------------------------------------------------ . lincom _cons +1.heavy+ 1.smoke+ 1.heavy#1.smoke ( 1) [cancer]1.heavy + [cancer]1.smoke + [cancer]1.heavy#1.smoke + [cancer]_cons = 0 ------------------------------------------------------------------------------ cancer | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- (1) | .2657032 .277149 0.96 0.338 -.277499 .8089053 ------------------------------------------------------------------------------ . * . * The preceding results are replicated by the following adjust command. . * Alas, they are not interesting log odds ratios. . * . margins heavy##smoke Predictive margins Number of obs = 975 Model VCE : OIM Expression : Pr(cancer), predict() ------------------------------------------------------------------------------ | Delta-method | Margin Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- heavy | 0 | .1360654 .012313 11.05 0.000 .1119325 .1601984 1 | .4631449 .0347549 13.33 0.000 .3950266 .5312632 | smoke | 0 | .1820707 .0132528 13.74 0.000 .1560957 .2080457 1 | .2857912 .0291584 9.80 0.000 .2286418 .3429406 | heavy#smoke | 0 0 | .1149425 .0129246 8.89 0.000 .0896108 .1402743 0 1 | .2111801 .0321664 6.57 0.000 .1481351 .2742251 1 0 | .4342105 .0402028 10.80 0.000 .3554146 .5130065 1 1 | .5660377 .0680786 8.31 0.000 .4326061 .6994694 ------------------------------------------------------------------------------ . log close name: log: C:\MyDocs\MPH\LectureNotes\ClassDoLogData\Survival\adjust.log log type: text closed on: 3 Nov 2009, 10:23:43 -------------------------------------------------------------------------------------------------------------