Code:
webuse stan3 stset stptime stptime, by(surgery) stir surgery
-
Login or Register
- Log in with
- You are not logged in. You can browse but not post. Login or Register by clicking 'Login or Register' at the top-right of this page. For more information on Statalist, see the FAQ.
-
My objective is to estimate incidence rate ratios for surgery vs non surgery but in a setting when I have individual-level sample weights to weight up to a population of interest. How can I estimate the incidence rate ratios while incorporating these sample weights? -
Specify weights when you declare data to be survival-time data,
Code:stset timevar [pw=weight] ,id(idvar) failure(failvar)
-
Thank you very much But when I try this I get an error:: stir does not allow pweightsOriginally posted by Øyvind Snilsberg View Post
]Code:set seed 98034 generate WEIGHT = trunc(runiform()*1000) stset stime[pweight = WEIGHT], id stptime stptime, by(surgery) stir surgery
Comment
-
Leon, you should know by now our policy on cross-posting as mentioned in the FAQ. This question is cross-posted to Stack Exchange.
It's not something I do, but you can either use -stcox- to work with hazards and their ratios (additionally, I suppose the you can use the baseline hazards and time information to get incidence rates but that's going to require you to do the math). I believe this was the suggestion you received from Stack Exchange.
Alternatively, I think the manual for -stcox- has an example of using Poisson regression as a way to approximate a Cox regression model (and under some conditions will be identical). That might be a way to get reasonable incidence rates, using -margins- after -poisson-, for example.Comment
-
Leonardo Guizzetti Thank you but what about for the incidence rate ratios? I can calculate the WEIGHTED incidence rates using the stcox command with stset and I can get the incidence rate ratio by division but what about the confidence interval for the incidence rate ratio?Originally posted by Leonardo Guizzetti View PostComment
-
They can be read directly as the exponentiated coefficient from Poisson regression (e.g., -poisson ..., ... irr-) and similarly for the hazard ratio after stcox. In either case, you need a variable for surgery indicating yes(=1) or no (=0).Comment
-
Leonardo Guizzetti Is it possible by any chance that you can provide sample syntax for this? To get weighted incidence rate ratios with confidence intervals?Originally posted by Leonardo Guizzetti View PostComment
-
I am out of the office for the next week but will revisit this when I return.Comment
-
Just writing this from the top of my head and not tested, so errors may be present, and you will need to substitute in your own variable names.
Code:* The reported IRR is the incidence rate ratio, weighted by your smoking weights, for the surgery group versus non-surgical group. poisson outcome i.surgery [pw=wtvar], irr exposure(timevar) vce(rob) * To check incidence rates in each group: margins i.surgery
Comment
-
Here's an example using an example dataset from the -stcox- manual, example #9.
The -predict(ir)- option to -margins- is needed to request incidence rates. These calls to -margin- will generate incidence rates in each drug group. The incidence rate ratio (here about 0.2) and 95% CI is taken directly from the coefficient for -1.drug-. The incorporation of -pweights- in the survival date is not needed since we already know -stir- can't use them, but added to show how it would work in general. Finally, the pweights are generated to yield very similar results to the unweighted case to show that they are similar.Code:clear * cls // example data from [ST] stcox manual, Example 9. Data have already been stset. use https://www.stata-press.com/data/r17/drugtr, clear // incidence rates in drug groups stir drug // using poisson regression to estimate incidence rate and their ratios poisson died i.drug , exposure(studytime) vce(robust) irr nolog margins i.drug, predict(ir) // generate a fake pweight gen wt = 1 + rnormal(0, 0.01) // Start again, adding weights stset, clear stset studytime [pw=wt], failure(died) // note: -stir- won't work with pweights so this step isn't necessary poisson died i.drug [pw=wt], exposure(studytime) vce(robust) irr nolog margins i.drug, predict(ir)
Selected results:
Code:. stir drug Failure _d: died Analysis time _t: studytime Incidence-rate comparison Exposed: drug = 1 Unexposed: drug = 0 | Drug type [0=placebo] | | Exposed Unexposed | Total -----------------+------------------------+----------- Failures | 12 19 | 31 Time | 564 180 | 744 -----------------+------------------------+----------- | | Incidence rate | .0212766 .1055556 | .0416667 | | | Point estimate | [95% conf. interval] |------------------------+------------------------ Inc. rate diff. | -.084279 | -.1332445 -.0353134 Inc. rate ratio | .2015677 | .0892308 .4373619 (exact) Prev. frac. ex. | .7984323 | .5626381 .9107692 (exact) Prev. frac. pop | .6052632 | +------------------------------------------------- . poisson died i.drug , exposure(studytime) vce(robust) irr nolog Poisson regression Number of obs = 48 Wald chi2(1) = 27.59 Prob > chi2 = 0.0000 Log pseudolikelihood = -51.746927 Pseudo R2 = 0.1564 ------------------------------------------------------------------------------ | Robust died | IRR std. err. z P>|z| [95% conf. interval] -------------+---------------------------------------------------------------- 1.drug | .2015677 .0614619 -5.25 0.000 .1108851 .3664114 _cons | .1055556 .0173419 -13.69 0.000 .0764954 .1456554 ln(studyt~e) | 1 (exposure) ------------------------------------------------------------------------------ . margins i.drug, predict(ir) Adjusted predictions Number of obs = 48 Model VCE: Robust Expression: Predicted incidence rate, predict(ir) ------------------------------------------------------------------------------ | Delta-method | Margin std. err. z P>|z| [95% conf. interval] -------------+---------------------------------------------------------------- drug | 0 | .1055556 .0173419 6.09 0.000 .071566 .1395451 1 | .0212766 .0054654 3.89 0.000 .0105646 .0319886 ------------------------------------------------------------------------------ . poisson died i.drug [pw=wt], exposure(studytime) vce(robust) irr nolog Poisson regression Number of obs = 48 Wald chi2(1) = 27.59 Log pseudolikelihood = -51.69405 Prob > chi2 = 0.0000 ------------------------------------------------------------------------------ | Robust died | IRR std. err. z P>|z| [95% conf. interval] -------------+---------------------------------------------------------------- 1.drug | .20152 .0614614 -5.25 0.000 .1108437 .3663744 _cons | .1053375 .0172558 -13.74 0.000 .0764089 .1452185 ln(studyt~e) | 1 (exposure) ------------------------------------------------------------------------------ . margins i.drug, predict(ir) Adjusted predictions Number of obs = 48 Model VCE: Robust Expression: Predicted incidence rate, predict(ir) ------------------------------------------------------------------------------ | Delta-method | Margin std. err. z P>|z| [95% conf. interval] -------------+---------------------------------------------------------------- drug | 0 | .1053375 .0172558 6.10 0.000 .0715167 .1391582 1 | .0212276 .005461 3.89 0.000 .0105242 .031931 ------------------------------------------------------------------------------Comment
Comment