Hello,
I have a logistic regression model assessing the impact of COVID-19 on odds of death in patients with Bacteraemias. The model has a categorical vs categorical interaction between cases and ethnicity variables. Below is the outcome of my model:
logistic dead i.cases ib3.age2 i.Sex i.comorbid ib5.IMD ib5.eth ib5.eth#i.cases ib2.onset_encode if comorbid != 5 & IMD != 6 & Sex != 3 & eth != 6
Since I am going to write these results as a table in a paper, I am struggling with certain aspects of the lincom command to decipher the interaction term.
Ideally, I want to get two types of information to present in the paper. The first one is to get the comparison of each ethnic group to its baseline (white) within each case category. This is how I get it via lincom for Asians (1.eth) in Bacteraemia with COVID category (3.cases) of cases variable :
lincom 1.eth + 1.eth#3.cases
The second bit of information I want is the overall odds ratios of 2.cases (Bacteraemia with no COVID) and 3.cases (Bacteraemia with COVID) when compared to baseline of pre-pandemic cases (1.cases). This is where I am struggling. As I understand the interactions, the model output for cases in the table above is for the reference category (Whites) of ethnicity. Now using lincom, I can get the odds ratios for cases within each ethnic group (example for Asians: lincom 3.cases + 3.cases#1.eth).
However, this is not what I ideally want to present. So the question is, is there a way to calculate the overall odds ratios for 2.cases and 3.cases from the model irrespective of the ethnic group category? I have tried this command for 3.cases:
lincom 3.cases + 3.cases#1.eth + 3.cases#2.eth + 3.cases#3.eth + 3.cases#4.eth
I am pretty sure that this is not the right output. Would be very thankful is someone could please let me know if it is possible to get what I am looking for here?
Many thanks,
Taimoor
I have a logistic regression model assessing the impact of COVID-19 on odds of death in patients with Bacteraemias. The model has a categorical vs categorical interaction between cases and ethnicity variables. Below is the outcome of my model:
logistic dead i.cases ib3.age2 i.Sex i.comorbid ib5.IMD ib5.eth ib5.eth#i.cases ib2.onset_encode if comorbid != 5 & IMD != 6 & Sex != 3 & eth != 6
| dead | Odds ratio | Std. err. | z | P>z | [95% conf. | interval] |
| cases (ref: pre-pandemic cases) | ||||||
| Bacteraemia with no COVID (pandemic) | 1.128696 | .0469193 | 2.91 | 0.004 | 1.040383 | 1.224506 |
| Bacteraemia with COVID (pandemic) | 2.657374 | .1814952 | 14.31 | 0.000 | 2.324431 | 3.038006 |
| age_groups (ref: 15 - 44) | ||||||
| 1.650525 | .2187061 | 3.78 | 0.000 | 1.27301 | 2.139992 | |
| 1-14 | .3749888 | .0721286 | -5.10 | 0.000 | .2572118 | .5466959 |
| 45-64 | 1.689975 | .1054362 | 8.41 | 0.000 | 1.495459 | 1.909792 |
| 65-74 | 2.027809 | .1264948 | 11.33 | 0.000 | 1.794441 | 2.291528 |
| 75-84 | 2.505079 | .154768 | 14.86 | 0.000 | 2.219386 | 2.827549 |
| 85+ | 4.059257 | .2549682 | 22.30 | 0.000 | 3.589065 | 4.591048 |
| Sex | ||||||
| Male | .8806917 | .0205493 | -5.44 | 0.000 | .8413229 | .9219027 |
| comorbidity (ref: no comorbidity) | ||||||
| 1-2 | 1.548949 | .0575598 | 11.78 | 0.000 | 1.440144 | 1.665974 |
| 3-4 | 2.171945 | .0855253 | 19.70 | 0.000 | 2.010624 | 2.34621 |
| 5 or more | 3.97011 | .1579815 | 34.65 | 0.000 | 3.672239 | 4.292143 |
| IMD | ||||||
| 1st (most deprived) | 1.106776 | .0403061 | 2.79 | 0.005 | 1.030531 | 1.188662 |
| 2nd | 1.032491 | .0374681 | 0.88 | 0.378 | .9616054 | 1.108602 |
| 3rd | .9744415 | .0358835 | -0.70 | 0.482 | .9065892 | 1.047372 |
| 4th | 1.040487 | .0385931 | 1.07 | 0.285 | .96753 | 1.118945 |
| ethnicity (ref: White) | ||||||
| Asian | .7602339 | .0443648 | -4.70 | 0.000 | .678069 | .8523552 |
| Black | .7619127 | .0632906 | -3.27 | 0.001 | .6474371 | .8966292 |
| Mixed | .6901543 | .1227129 | -2.09 | 0.037 | .4870777 | .9778993 |
| Other | 1.401999 | .1940003 | 2.44 | 0.015 | 1.068965 | 1.838789 |
| eth#cases | ||||||
| Asian#Bacteraemia with no COVID (pandemic) | 1.359244 | .1417933 | 2.94 | 0.003 | 1.107904 | 1.667604 |
| Asian#Bacteraemia with COVID (pandemic) | 2.098096 | .3253626 | 4.78 | 0.000 | 1.548193 | 2.843319 |
| Black#Bacteraemia with no COVID (pandemic) | 1.286516 | .1876806 | 1.73 | 0.084 | .966584 | 1.712344 |
| Black#Bacteraemia with COVID (pandemic) | 1.350835 | .3001279 | 1.35 | 0.176 | .8739418 | 2.087958 |
| Mixed#Bacteraemia with no COVID (pandemic) | 1.511664 | .4347204 | 1.44 | 0.151 | .8603374 | 2.656083 |
| Mixed#Bacteraemia with COVID (pandemic) | 1.246824 | .647318 | 0.42 | 0.671 | .450696 | 3.449264 |
| Other#Bacteraemia with no COVID (pandemic) | .6595776 | .1689423 | -1.62 | 0.104 | .3992469 | 1.089658 |
| Other#Bacteraemia with COVID (pandemic) | .8575516 | .2796592 | -0.47 | 0.637 | .4525556 | 1.624982 |
| onset_encode (ref: Community onset) | ||||||
| COHA | 1.383543 | .0401135 | 11.20 | 0.000 | 1.307114 | 1.464441 |
| Hospital_Onset | 1.871458 | .0510639 | 22.97 | 0.000 | 1.774004 | 1.974266 |
| _cons | .5396891 | .2860624 | -1.16 | 0.245 | .1909695 | 1.525188 |
Since I am going to write these results as a table in a paper, I am struggling with certain aspects of the lincom command to decipher the interaction term.
Ideally, I want to get two types of information to present in the paper. The first one is to get the comparison of each ethnic group to its baseline (white) within each case category. This is how I get it via lincom for Asians (1.eth) in Bacteraemia with COVID category (3.cases) of cases variable :
lincom 1.eth + 1.eth#3.cases
| dead | Odds ratio | Std. err. | z | P>z | [95% conf. | interval] |
| (1) | 1.595044 | .229646 | 3.24 | 0.001 | 1.202877 | 2.115067 |
The second bit of information I want is the overall odds ratios of 2.cases (Bacteraemia with no COVID) and 3.cases (Bacteraemia with COVID) when compared to baseline of pre-pandemic cases (1.cases). This is where I am struggling. As I understand the interactions, the model output for cases in the table above is for the reference category (Whites) of ethnicity. Now using lincom, I can get the odds ratios for cases within each ethnic group (example for Asians: lincom 3.cases + 3.cases#1.eth).
However, this is not what I ideally want to present. So the question is, is there a way to calculate the overall odds ratios for 2.cases and 3.cases from the model irrespective of the ethnic group category? I have tried this command for 3.cases:
lincom 3.cases + 3.cases#1.eth + 3.cases#2.eth + 3.cases#3.eth + 3.cases#4.eth
| dead | Odds ratio | Std. err. | z | P>z | [95% conf. | interval] |
| (1) | 8.052774 | 5.521473 | 3.04 | 0.002 | 2.100438 | 30.87317 |
I am pretty sure that this is not the right output. Would be very thankful is someone could please let me know if it is possible to get what I am looking for here?
Many thanks,
Taimoor
