03-15-2013 03:15 PM
Has anyone here used NIS database for analyses? which gives the information about numbers of discharges from each hospital.
I am trying to run a regression model and using a variable called TOTAL_DISC,
I want to use it to indicate the hospital volume in my analyses; but I am facing some issue with it. No matter what independent variables I include in my regression model along with this one, the results always shows as 1.00 as the value for point estimate and the LCL and UCL, as highlighted below. I am not sure why i am getting these results and cannot interpret the Odds ratio value and LCL/UCL for this variable.
Appreciate if anyone who has nay clue, could provide some guidance in this regard and advise if there is any other way to use this variable and get comprehensible result values or if there is anything I am doing incorrectly.
Below is the table, with the value highlighted..
|Effect||Point Estimate||95% Wald|
|FEMALE 1 vs 0||1.784||1.709||1.862|
|race1 2 vs 1||1.095||1.012||1.184|
|race1 3 vs 1||1.151||1.061||1.249|
|race1 4 vs 1||0.875||0.748||1.025|
|race1 5 vs 1||1.13||1.025||1.246|
|ZIPINC_QRTL 2 vs 1||0.985||0.929||1.045|
|ZIPINC_QRTL 3 vs 1||1.055||0.994||1.119|
|ZIPINC_QRTL 4 vs 1||1.198||1.127||1.274|
|HOSP_LOCATION 0 vs 1||0.97||0.882||1.067|
|H_CONTRL 2 vs 1||1.078||0.994||1.169|
|H_CONTRL 3 vs 1||0.892||0.809||0.983|
|HOSP_TEACH 1 vs 0||1.108||1.058||1.161|
|dm_all 1 vs 0||0.791||0.754||0.83|
|CM_HTN_C 1 vs 0||0.837||0.8||0.875|
|morbidobesity 1 vs 0||0.906||0.796||1.031|
|hyperlipidemia 1 vs 0||0.812||0.778||0.848|
|CM_PERIVASC 1 vs 0||2.554||2.421||2.694|
|CHF 1 vs 0||2.142||1.636||2.804|
03-15-2013 04:56 PM
From the HCUP website I see that total_disc takes uniform values
|TOTAL_DISC||Total hospital discharges||5(n)||Total hospital discharges|
What is the distribution of total_disc? Maybe in your sample you have exactly the same #s in each 'group' of total_disc (what I mean is that maybe there is an equal # of total_disc = 1, total_disc = 2...so on...).
03-15-2013 05:15 PM
03-15-2013 05:54 PM
Ok, so giving the distribution of total_disc you have ~ 65% of the data with less than 100 discharges, but the max is 1522(discharges)
Ok, so if you look at the histogram of total_disc you notice that it is not only highly skewed, but it also has a high kurtosis.So, I wonder if you would categorize the total_disc into say 5 or 6 groups, like
0 to 25, 25 to 50, 50 to 75, 75 to 100, then > 100...or 100 to 125...and > 125
Anyway, my point it to sort of create groups such that any total_disc > some cutt-of point (between 100 and 200) gets lumped into one group.
Do you know what I mean?
03-15-2013 06:22 PM
I see what you are saying, yes i would be a good idea to categorize them. But I am not sure where you see that 65% of the data is with <100 discharges and max is 1522. Where do you see those numbers? In fact what i see that 75% of the data shows discharges upto 27408.
03-15-2013 10:58 PM
sorry, I got my numbers confused. I was looking a the distribution of the frequencies of total_disc, which obviously is not the same as total_disc.
But you get my idea, so let me know if the ORs look better once you've grouped the variable.(possibly in quartiles,or quintiles)