3
$\begingroup$

I'm a complete novice in statistics, but while doing my first publication in medical research my advisor gave me the task to analyse a dataset. So the problem:

I'm comparing the time spent in a hospital after surgery between two surgical methods (robotic assisted and traditional open technique). The dataset is quite large (7960 cases). Basically I'm comparing two groups (which surgery = nominal data) by the days spent in hospital (scale), while taking into account about 7 other nominal and scale variables (eq. lymphnode dissection, was the surgery done in an university hospital, the year of the surgery and so on).The point of the study is to see weather the chosen method has an effect on the post op days and how big is that effect.

Which statistical method should I use? Preferably one that can be done with SPSS with some practice.

Improve this question
$\endgroup$
5
  • $\begingroup$ Is days in hospital skewed? Does it ever have the value zero? $\endgroup$
    – mdewey
    Commented Feb 7, 2017 at 18:22
  • $\begingroup$ Logistic regression would not be appropriate. The outcome variable is days. Time is not categorical here. $\endgroup$
    – Marcus Morrisey
    Commented Feb 7, 2017 at 18:24
  • 1
    $\begingroup$ No, minimum is 1 $\endgroup$
    – Monnimies
    Commented Feb 7, 2017 at 18:24
  • $\begingroup$ This question could be addressed with some form of regression analysis. It would be helpful if you could add more detail about all the variables involved. $\endgroup$
    – Marcus Morrisey
    Commented Feb 7, 2017 at 18:28
  • 1
    $\begingroup$ Monnimies: To add detail or otherwise clarify your question, please edit it. $\endgroup$
    – whuber ♦
    Commented Feb 7, 2017 at 18:33

3 Answers 3

Reset to default
1
$\begingroup$

This sounds like a job for survival analysis. In this case, "survival" is the number of days to discharge, and shorter "survival" is better. If the appropriate assumptions are met, a Cox proportional hazards regression could evaluate the open/robotic-assisted variable of main interest while accounting for the influences of your other covariates.

Improve this answer
$\endgroup$
0
$\begingroup$

This is standard ANCOVA or general linear model if you'd rather call it that. Since times tend to be skewed, you should examine the distribution of the residuals and, perhaps transform the data using log or something similar. In SPSS I believe it is under General Linear Model but I'm not sure because I don't use SPSS.

Improve this answer
$\endgroup$
-1
$\begingroup$

I agree with @MissMonicaE. A logistic regression should work for you, just use the group as the dependent variable and put all the others in as covariates. See the link below for step-by-step instructions in SPSS, and help interpreting and reporting the results. https://statistics.laerd.com/spss-tutorials/binomial-logistic-regression-using-spss-statistics.php

Improve this answer
$\endgroup$
7
  • 2
    $\begingroup$ Logistic regression would not be appropriate. The outcome variable is days. Time is not a categorical variable here. $\endgroup$
    – Marcus Morrisey
    Commented Feb 7, 2017 at 18:27
  • 1
    $\begingroup$ @Marcus Although time is not categorical, length of stay as recorded in the data is a nonzero count of days. Although (because the counts cannot be zero) logistic regression clearly is not applicable (and therefore this answer isn't quite on the mark), some form of a GLM likely will be a good procedure. $\endgroup$
    – whuber ♦
    Commented Feb 7, 2017 at 18:30
  • $\begingroup$ @MarcusMorrisey, for testing differences between two groups when data are non-normal (length of stay tends to follow a gamma distribution) and wanting covariates, logistic regression would work. The interpretation of the odds ratios would be that for each one unit change in length of stay, the patient is x% more likely to be in one group than the other. $\endgroup$
    – JRF1111
    Commented Feb 7, 2017 at 18:32
  • $\begingroup$ The variables in question are: the type of surgery (1=robotic,2=traditional), the year it was performed, lymphnode dissection (1=y,0=n, indicates a more demanding surgery), was the hospital a university hospital (1=y, 0=n), was some other invasive treatment performed (eq. Hernia surgery or hemicolectomy, 1=y,0=n), does the patient's home town have more than 100k residents (affects the typical job and socioeconomy , 1=y,0=n), the patients age at the moment of leaving the hospital and were there more than 50 of said operations performed in the same hospital that year (1=y, 0=n) $\endgroup$
    – Monnimies
    Commented Feb 7, 2017 at 18:40
  • 1
    $\begingroup$ @Marcus It would have to be a truncated Poisson response to avoid having $0$ as a possibility. $\endgroup$
    – whuber ♦
    Commented Feb 7, 2017 at 19:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.