what is the best statistical analysis ? outcome of surgery

Dear All,

I would like some advice on the most appropriate statistical analysis for some data I have.

I have a data set which comes from a group of 51 patients who combined have had 251 operations, and Im interested in presenting the data looking at timing of surgery and complication free outcomes.

Each patient has had an accident and each patient has had an operation , now some patients had more than one operation.

Those operations have been groups as early , mid or late depending on the time between injury and operation

Each patient then has an outcome from each operation of early complication, late complication or no complication

I am interested in a statistical analysis of the 3 groups of operation timings

So independant variables I think is time of surgery (3 in total = early , mid, or late ) nominal data

depdent variable (3 in total) I think are outcome of surgery ( early complication, late complication or no complication ) nominal data

As there are some of the 51 patients who perhaps have had both early and mid or early and late surgery then the data is "within group"?

Statistical question

Is operating early compared to mid or late associated with a significantly lower or high likelyhood of complication ?

on the same tack - does operating late carry a significantly lower or higher likelyhood of complication ( early or late) as compared to the 2 other operation timings

Is operating early associated with higher or lower chance of complication (early or late) if then operated mid or late ?

I'm totally at a loss as to what test to use so any advice would be most welcome

Im thinking ANOVA or multiple t tests ?

The data set looks like

(see attached file )

in total 51 patients had a total of 251 operations

Re: what is the best statistical analysis ? outcome of surgery

Hi,
this looks like a logistic regression to me, probably even a binary one.

The data you showed seems a bit complex, at least for me. I wonder if I can read it correctly,
e.g in the image you sent patient1 had no early operation, was operated mid term and late, but on the late operation an early complication was identified? Is there a way to organize the data like operation time (early, mid, late) complication (yes, no) ?

Re: what is the best statistical analysis ? outcome of surgery

that is correct - essentially

Patient 1 - had 2 operations , one mid term which had no complications and one late term which had an early complication. Patient 1 did not undergo an early operation.

what do you think is the best way of organising the data as some patients had 3 operations each of the 3 different time periods. No patients had more than 1 operation in 1 time period.

the outcome of the complications is not just yes or no but - yes early problem , yes late problem or no problem. So the outcome is not binary ( complication yes or no )

Re: what is the best statistical analysis ? outcome of surgery

Hi,
thinking about it a bit more, it cannot be a logistic regression because we do not have any continuous X-s. maybe a simple chi-squared analysis would do here - I hope someone will correct me if I am wrong.

So, I would organize the data in a table - the columns would be operation timing (early, mid, late) the rows complication timing (none, early, mid, late) and in the cell just the number of cases that are described by that set. E.g. if you have 5 cases where patients were operated early and had no complications the cell (early, none) would contain the number 5. The same patient will of course be counted several times, e.g, if the person was operated mid term without complications then he will be counted once in that cell, and if he was operated later with an early complication then counted again in that cell.

The method supposes that we can consider these the events to be independent, if there is some kind of dependency then it will not work (e,g, if only those will be operated late who were operated early and had some complication then this will not be the right way to analyze).

Re: what is the best statistical analysis ? outcome of surgery

the majority of the time the events are independent but in a few incidents a complication in an operation meant that a second operation which falls into a new time catagoriy was undertaken

Re: what is the best statistical analysis ? outcome of surgery

hi,
youre welcome!
If there are only a few cases of non-independence I would run the chi-squared anyway and see if the results make sense. An alternative would be to run the analysis twice, once with the full data set and once without the dependent cades and check if the conclusions differ.
regards
rogojel

Re: what is the best statistical analysis ? outcome of surgery

How standard are the "accidents"? I have done a number of trials involving significant trauma and there are certain types of injuries that simply don't heal very well. A skilled surgeon can look at an x-ray or examine the patient and decree the likely prognosis based on the pattern of injury. If you mean this to be a serious analysis, you should think about controlling for severity. If you don't have the data available, think about having an independent trauma physician (i.e. blinded) review the patient chart as of admission and give an impression of the case on a numerical scale.