Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Hierarchical mixed-effects logistic regression

    Hello everyone,

    I am working on the National Inpatient Sample (NIS) database. There are certain types of discharges that we are interested in. We selected them based on the ICD codes for the procedure of interest. We have a variable that shows how many procedures have been done by each operator/physician and we call it simply "operator volume". Now we want to see if there is a significant correlation between the complication rates of that procedure (variable: complications) with operator's experience (variable: OPERATOR_VOLUME). I found a similar paper that used the same database and has a very similar methodology with our project. So we want to duplicate their method – hierarchical mixed-effects logistic regression – to find out about the aforementioned correlation. A very simplified version of another population just as an example to work with in here as a .dta file. Below is the methodology that we want to use:
    • Hierarchical mixed-effects logistic regression models were generated to identify the independent multivariate predictors of postprocedural complications
    • Two level hierarchical models (with patient level factors nested within hospital level factors) were created with the unique hospital identification number incorporated as random effects within the model
    • In all multivariate models, we included hospital level variables such as hospital bed size, hospital region (Northeast, South, Midwest with West as referent), teaching versus nonteaching hospital, and patient level variables such as age, sex, comorbidity index, median household income, and primary payer (with Medicare/ Medicaid considered as referent) in addition to hospital procedure volume or operator procedure volume or both
    • The effects of hospital volume and operator volume were studied separately by creating separating models incorporating each without the other
    • Subsequently a third model was created incorporating both hospital and operator volume with a term to adjust for the interaction effect between hospital and operator volume
    • Hospital identification was incorporated as a random effect in the model to account for the effect of hospital clustering (meaning that patients treated at the same hospital may experience similar outcomes as a result of other processes of care)
    • Because operator identification did not remain constant across the years, we could not incorporate it as a random effect in the model
    these are the description of the variables:
    • YEAR is year of discharge
    • HOSPID is the hospital identification number which is unique to each one
    • TRENDWT is the frequency weight which is used to estimate the total national estimate from these (so each discharge represents more than one discharge)
    • AGE is age
    • gender is gender
    • NIS_STRATUM is the stratum used to sample hospital
    • primary_payer is the primary payer
    • comorbidity_index is a continuos index for measuring comorbidities
    • HOSPITAL_VOLUME is the hospital procedure volume which is devided in 1 (hospitals with less than 20 procedures per year) , 2 (20-40 procedures/year), and 3 (>40 procedures/year)
    • complications is whether a complication occured or not
    • HOSP_BEDSIZE is the bedsize
    • HOSP_REGION is the region of the hospital
    • HOSP_TEACH is the teaching status
    • income is income
    • OPERATOR_VOLUME is the operator procedure volume which is devided in 1 (operators who have less done 20 procedures per year) , 2 (20-40 procedures/year)
    (This is the .dta sample in HERE )

    I greatly appreciate your suggestions on how to perform and interpret this analysis and its results in Stata.


    Thank you so much!
    Reza
    Last edited by Reza Hosseini; 13 Feb 2016, 08:32. Reason: regression

  • #2
    On hierarchical (multilievel) mixed-effects logistic regression: help melogit
    You have issued a very general plea for help, and readers are unlikely to oblige. You are much more likely to get a response if you provide examples of particular problems that you are having, also reporting what you have read (Stata help, Stata manuals, relevant literature) and why it doesn't appear to help you, and also what Stata code you have used. Before posting on that, please also read the Forum FAQ (hit the black bar at top of the page) for recommendations on how to ask questions most likely to elicit helpful responses, and for instructions on how to post your Stata input and output.

    Comment


    • #3
      Hello Reza,

      Please don't take it amiss, but I fear the odds are not high that you promptly get some sort of "Applied Mixed Models in a Nutshell" in a couple of lines. Logically, things can be presented in a didatic way, and sometimes they may become simpler than we expected. However, I really don't know whether it is possible to subsume all these heavyweight demands under a short message. Had it been a specific doubt, I bet there would be several replies so far.

      Generally speaking, in similar circunstances, and now sharing what I believe are the best options (at least, they keep doing wonders to me), I usually:

      1) Read carefully the Stata Manual. Yes, that's unavoidable. Full of applied examples. What is more, it's free.

      2) Call for immediate help, i.e., I really meant it, when I do type:
      Code:
      . help mixed
      3) Watch some video tutorials: StataTube under Chuck Huber is really something!

      4) Get a couple of referential books such as: Multilevel and Longitudinal Modeling Using Stata, Third Edition, 2 volumes, StataPress, 2012 (Sophia Rabe-Hesketh and Anders Skrondal); also, for an introduction, Multilevel Modeling in Plain Language, Sage, 2016 (Karen Robson and David Pevalin)

      5) Attend a "crash" course: by the way, among the Stata Training Courses, there is a net course on panel data.

      6) No wonder, doubts may persist. Being this the case, I believe this forum is one of the best places to solve doubts when it comes to Stata!


      This was the best I could possibly do, considering the scope of the subject. Crossing fingers, I hope you'll get more clarifying advice from the fellows.

      Best,

      Marcos
      Last edited by Marcos Almeida; 13 Feb 2016, 11:14.
      Best regards,

      Marcos

      Comment

      Working...
      X