Notice: On March 31, it was **announced** that Statalist is moving from an email list to a **forum**. The old list will shut down on April 23, and its replacement, **statalist.org** is already up and running.

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

From |
"Ariel Linden, DrPH" <ariel.linden@gmail.com> |

To |
<statalist@hsphsun2.harvard.edu> |

Subject |
re: st: ELIXHAUSER Comorbidity: choosing covariates, colinearity, elix_cnt |

Date |
Thu, 18 Apr 2013 10:46:51 -0400 |

Hi Mike, First off, I find the -elixhaus- [ -findit elixhaus-] code to be extremely cumbersome. The user can only input information via a dialog box, not a command line as is more common in Stata. From there, it only gets more cumbersome... So my first suggestion is that you consider using -charlson- (findit charlson) which implements the Charlson comorbidity index. It is much easier to use, and can cycle through as many diagnosis fields as you have available. As to your question about using comorbidities in conjunction with the primary diagnosis. The short answer is "no". A comorbidity, by definition, is not the primary condition under study. Nor was the original intent of the developers of such indices to use them as primary diagnosis indicators. So you should run the code with the intent of NOT categorizing the comorbidity as a primary diagnosis. There are several ways of handling this, and of course, I am advocating -charlson- because of its flexibility. The first and easiest approach is the run -charlson- and then simply eliminate the corresponding comorbidity from the data. For example, if CHF is your primary diagnosis under study, you can drop the CHF comorbidity variable. You can also easily recalculate the index score without this comorbidity (equally as useful, you should reference this comorbidity category to ensure that everyone with the primary diagnosis of CHF actually shows up with a comorbidity of CHF before dropping that variable). Another, more cumbersome, way of dealing with this is to drop from the data all primary diagnoses of the targeted diagnosis (in this example, CHF), and then run -charlson-. You can then merge the files together after the command was run... I hope this helps... Ariel Date: Wed, 17 Apr 2013 00:25:33 -0700 From: Mike Butterfield <mcbutterfield@gmail.com> Subject: st: ELIXHAUSER Comorbidity: choosing covariates, colinearity, elix_cnt Hi Statalisters, I have figured out how to add Elixhauser comorbidity variables to my dataset [ -findit elixhaus-] and used them in my logistic model for predicting death given infection ("infx"). Output is below. I had questions about three things: 1. Elixhauser acknowledges that the coding of the primary diagnosis for a hospital admission might be somewhat arbitrary (Elixhauser, A., Steiner, C., Harris, D. R. & Coffey, R. M. Comorbidity measures for use with administrative data. Med. Care 36, 8-27 (1998)-that is to say, if the primary reason for admission is incorrectly coded, then a complication of the disease may incorrectly be counted as a comorbidity, since the latter should be "conditions present on admission that are not related directly to the main reason for hospitalization." To be conservative, then, if my infection is liver-related, should I exclude comorbidity variable 14 (liver disease) from my elixhauser-adjusted analysis? 2. Comorbidity 6 refers to hypertension, complicated or not. Comorbidity 6A refers to uncomplicated hypertension, 6B to complicated hypertension. Is there any reason that in my analyses (below) comorbidities 6B should have problems with collinearity? Is this a dataset-specific problem? There isn't this problem with colinearity with diabetes and complicated diabetes (Comorbidities 10+11) for example. Here's an excerpt of the ado file which can be found here: http://fmwww.bc.edu/repec/bocode/e/elixhaus.ado /* Set uncomplicated hypertension to FALSE if complicated hypertension is TRUE. */ replace elix6A = 0 if elix6B == 1 3. There is a variable called elix_cnt, which counts the total number of elixhauser comorbidities. The code is (http://fmwww.bc.edu/repec/bocode/e/elixhaus.ado): replace elix_cnt = elix1 + elix3 + elix4 + elix5 + elix6 + elix7 + elix8 + elix9 + elix10 +/* */ elix11 + elix12 + elix13 + elix14 + elix15 + elix16 + elix17 + elix18 + elix19 + elix20 + elix21 + elix22 +/* */ elix23 + elix24 + elix25 + elix26 + elix27 + elix28 + elix29 + elix30 I don't know how this variable could be used. Would you make a categorical variable, for example, (0-5 comorbidities, 6-10, etc), to analyze how your predictor(s)-outcome relationship is affected by different numbers of comorbidities? OUTPUT 1. Unadjusted logistic regression: . logistic died infx Logistic regression Number of obs = 253010 LR chi2(1) = 441.55 Prob > chi2 = 0.0000 Log likelihood = -62605.076 Pseudo R2 = 0.0035 - ---------------------------------------------------------------------------- -- died | Odds Ratio Std. Err. z P>|z| [95% Conf. Interval] - -------------+-------------------------------------------------------------- -- infx | 3.269661 .1630761 23.75 0.000 2.965163 3.605428 _cons | .0714099 .0005723 -329.35 0.000 .0702971 .0725404 - ---------------------------------------------------------------------------- -- 2. Adjusted logistic regression (all variables) logistic died infx elix1 elix10 elix11 elix12 elix13 elix14 elix15 elix16 elix17 elix18 elix19 elix20 elix21 elix22 elix > 23 elix24 elix25 elix26 elix27 elix28 elix29 elix3 elix30 elix4 elix5 elix6 elix6A elix6B elix7 elix8 elix9 note: elix6B omitted because of collinearity Logistic regression Number of obs = 253010 LR chi2(31) = 15927.65 Prob > chi2 = 0.0000 Log likelihood = -54862.026 Pseudo R2 = 0.1268 - ---------------------------------------------------------------------------- -- died | Odds Ratio Std. Err. z P>|z| [95% Conf. Interval] - -------------+-------------------------------------------------------------- -- infx | 2.327502 .1247853 15.76 0.000 2.095339 2.585389 elix1 | 1.752128 .0466361 21.07 0.000 1.663066 1.845959 elix10 | .7261666 .0174254 -13.33 0.000 .6928041 .7611357 elix11 | .5437848 .0314756 -10.52 0.000 .4854645 .6091113 elix12 | .6945754 .0258834 -9.78 0.000 .6456531 .7472045 elix13 | 1.653751 .0539559 15.42 0.000 1.551309 1.762957 elix14 | .5573842 .0095212 -34.22 0.000 .5390319 .5763612 elix15 | .7121403 .2504525 -0.97 0.334 .3574429 1.418811 elix16 | .6592312 .0613643 -4.48 0.000 .5492929 .791173 elix17 | 1.734889 .1220954 7.83 0.000 1.511357 1.991481 elix18 | 2.676777 .0939822 28.04 0.000 2.49877 2.867464 elix19 | 1.399867 .0555123 8.48 0.000 1.295185 1.513009 elix20 | .8096719 .0571044 -2.99 0.003 .7051404 .9296993 elix21 | 2.027206 .0376051 38.09 0.000 1.954826 2.102267 elix22 | .6329491 .0275517 -10.51 0.000 .5811882 .6893199 elix23 | 1.597156 .0396977 18.84 0.000 1.521214 1.676888 elix24 | 2.712489 .0459804 58.87 0.000 2.62385 2.804123 elix25 | .5913501 .0371296 -8.37 0.000 .522877 .6687902 elix26 | .5529848 .012882 -25.43 0.000 .5283043 .5788183 elix27 | .6438405 .0144374 -19.64 0.000 .6161565 .6727683 elix28 | .5353797 .0239031 -13.99 0.000 .4905217 .5843398 elix29 | .4455346 .0249444 -14.44 0.000 .3992315 .4972081 elix3 | .7877983 .0413448 -4.54 0.000 .7107924 .8731468 elix30 | .4118454 .0172667 -21.16 0.000 .3793564 .4471167 elix4 | 1.345587 .0667588 5.98 0.000 1.220903 1.483005 elix5 | 1.634475 .0644454 12.46 0.000 1.512922 1.765794 elix6 | .7575117 .0286522 -7.34 0.000 .7033854 .815803 elix6A | .7157584 .030474 -7.85 0.000 .6584546 .7780493 elix6B | 1 (omitted) elix7 | 1.382628 .0815414 5.49 0.000 1.231701 1.552049 elix8 | .9966491 .035557 -0.09 0.925 .9293393 1.068834 elix9 | .9835117 .0232028 -0.70 0.481 .9390704 1.030056 _cons | .0739074 .0013526 -142.34 0.000 .0713033 .0766065 3. Adjusted logistic regression (excluding liver-related comorbidity code) logistic died infx elix1 elix10 elix11 elix12 elix13 elix15 elix16 elix17 elix18 elix19 elix20 elix21 elix22 elix23 elix24 elix25 elix26 elix27 elix28 elix29 elix3 elix30 elix4 elix5 elix6 elix6A elix6B elix7 elix8 elix9 note: elix6B omitted because of collinearity Logistic regression Number of obs = 253010 LR chi2(30) = 14726.61 Prob > chi2 = 0.0000 Log likelihood = -55462.545 Pseudo R2 = 0.1172 - ---------------------------------------------------------------------------- -- died | Odds Ratio Std. Err. z P>|z| [95% Conf. Interval] - -------------+-------------------------------------------------------------- -- infx | 2.089176 .1114439 13.81 0.000 1.881781 2.319429 elix1 | 1.758834 .0465648 21.33 0.000 1.669896 1.852508 elix10 | .7034331 .0168009 -14.73 0.000 .6712628 .7371452 elix11 | .5094977 .0293895 -11.69 0.000 .4550321 .5704826 elix12 | .6857739 .025488 -10.15 0.000 .6375944 .7375939 elix13 | 1.611201 .0523636 14.68 0.000 1.511771 1.717171 elix15 | .7448917 .2608519 -0.84 0.400 .374984 1.479699 elix16 | .6179654 .0573438 -5.19 0.000 .5152017 .7412266 elix17 | 1.719985 .1202377 7.76 0.000 1.499755 1.972554 elix18 | 2.818681 .0985196 29.65 0.000 2.632052 3.018543 elix19 | 1.463343 .0578473 9.63 0.000 1.354245 1.581229 elix20 | .776072 .0545794 -3.60 0.000 .6761436 .890769 elix21 | 2.012992 .0371307 37.93 0.000 1.941517 2.087098 elix22 | .5976372 .0259134 -11.87 0.000 .5489461 .6506471 elix23 | 1.633748 .0404475 19.83 0.000 1.556365 1.714978 elix24 | 2.826368 .0476871 61.58 0.000 2.734432 2.921396 elix25 | .5671278 .0355622 -9.04 0.000 .5015402 .6412925 elix26 | .5518668 .0128213 -25.59 0.000 .527301 .577577 elix27 | .6337118 .0141403 -20.44 0.000 .6065946 .6620412 elix28 | .5047475 .0224505 -15.37 0.000 .4626088 .5507247 elix29 | .4325598 .0241672 -15.00 0.000 .3876941 .4826174 elix3 | .780964 .04086 -4.73 0.000 .7048491 .8652983 elix30 | .4006842 .0167638 -21.86 0.000 .3691388 .4349253 elix4 | 1.337082 .0660036 5.88 0.000 1.213779 1.472912 elix5 | 1.642004 .0643415 12.66 0.000 1.520617 1.77308 elix6 | .7599071 .0286418 -7.28 0.000 .7057937 .8181694 elix6A | .7023825 .0298091 -8.32 0.000 .6463217 .7633059 elix6B | 1 (omitted) elix7 | 1.373044 .0804276 5.41 0.000 1.224121 1.540084 elix8 | .9909208 .0352274 -0.26 0.798 .9242268 1.062428 elix9 | .9434958 .0221438 -2.48 0.013 .9010779 .9879106 _cons | .0572066 .0009789 -167.20 0.000 .0553198 .0591578 - ---------------------------------------------------------------------------- -- Best, - -Mike b. * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/faqs/resources/statalist-faq/ * http://www.ats.ucla.edu/stat/stata/

- Prev by Date:
**Re: st: How to display independent variables after regression** - Next by Date:
**Re: st: Stata 13?** - Previous by thread:
**st: ELIXHAUSER Comorbidity: choosing covariates, colinearity, elix_cnt** - Next by thread:
**st: Labelling axes as percentages** - Index(es):