+ Reply to Thread
Results 1 to 4 of 4

Thread: Retrospective review analysis questions

  1. #1
    Points: 10, Level: 1
    Level completed: 19%, Points required for next Level: 40

    Posts
    2
    Thanks
    1
    Thanked 0 Times in 0 Posts

    Retrospective review analysis questions




    Hello! I am new to this forum. Thank you in advance for any help you can give!

    I am doing a prelim analysis for a pilot study I have performed. I will have dedicated stats professionals in the future, but for now I am on my own and need to do some basics.

    This is a retrospective chart review of >2000 consecutive patients undergoing any 1 of 3 types of open heart surgeries (lets say operations 0,1,2) in between two predetermined dates at a single hospital.

    I am looking for a complication known as diaphragm elevation. There are 3 categorical outcomes. No elevation , Left elevation , and Right Elevation.

    I have preoperative demographic data, both categorical (the presence of certain comorbid conditions) and continuous (such as age). I have operative data both categorical (operative approach, type of graft, etc) and continuous (number of grafts, etc). Finally I have outcomes data both categorical (presence/absence of post operative respiratory failure) and continuous (length of stay).

    My primary objective is (1) to calculate the incidence of diaphragm elevation for each of the 3 surgery types. This is straightforward.

    My secondary objectives include (2) determining any significant associations between pre-operative or operative data and diaphragm elevation and (3) any significant associations between diaphragm elevation and post-operative outcomes.

    My question is what statistical tests should I do to achieve 2 and 3? Fisher's exact test for the categorical variables and T-tests for the continuous variables? I am really just looking for guidance on what my general approach should be to this data. I am aware there are tons of limitations in a retrospective cohort review. This is hypothesis generating. Once I have the stats team on board I think they will likely match controls to do a case control analysis but I need to have something to present before this. Ultimately, a prospective study will be performed.

    Thank you so so much.

    Boomer

  2. #2
    Omega Contributor
    Points: 38,374, Level: 100
    Level completed: 0%, Points required for next Level: 0
    hlsmith's Avatar
    Location
    Not Ames, IA
    Posts
    6,998
    Thanks
    398
    Thanked 1,186 Times in 1,147 Posts

    Re: Retrospective review analysis questions

    You state you know the limitations of retrospective data, though these seem to be observational data as well. Were surgical procedures randomized? If not, is there a threat for confounding by indication.

    How are you planning to adress (1)?

    How many patients do you have and what are the proportions of patients per treatment group and outcomes per treatment?

    You can compare the differences between the groups, however - these may not be the groups that get used in the future analyses by the Stats Team. Not controlling for the patient differences in (1) could lead to spurious results. Then just controlling for them is about as comparable to propensity match for about 90% of time, though typically just controlling for variables overestimate actual differences. Big picture, Just comparing and reporting differences in (1) could be detrimental to your field and your reputation, if results are not robust to all of the issues presented.

    Another thing to consider if you compare characteristics of THESE groups, is correcting for multiplicity or false discovery rates given you will be potentially making 3 comparisons if omnibus tests are significant. Fisher's could be used, however for your continuous data you would use ANOVA followed by pairwise t-test or if data normality assumption is nor met, then Kruskal-Wallis followed by Wilcoxon rank sum (medians).

    I applaud your interest, though if you have apprehension about your ability to run the stats, I would just wait or recruit some help. Or play around with these data and not attempt to report them at a conference.
    Stop cowardice, ban guns!

  3. The Following User Says Thank You to hlsmith For This Useful Post:

    boomer (02-25-2015)

  4. #3
    Points: 10, Level: 1
    Level completed: 19%, Points required for next Level: 40

    Posts
    2
    Thanks
    1
    Thanked 0 Times in 0 Posts

    Re: Retrospective review analysis questions

    Quote Originally Posted by hlsmith View Post
    You state you know the limitations of retrospective data, though these seem to be observational data as well. Were surgical procedures randomized? If not, is there a threat for confounding by indication.

    It is observational, taken by sequential convenience sampling. There is no randomization. The procedure groups (coronary artery bypass graft [CABG], aortic procedures, and mitral/tricuspid procedures) all have different indications and are treatments for idfferent disease. They are representative of the majority of invasive CT procedures besides transplant and bridge to transplant. There is no consensus in the literature for the incidence or risk factors for diaphragm elevation or paralysis. Is it thought to be due to injury to the phrenic nerve. These procedures have different techniques which potentially compromise the phrenic nerves in different way
    How are you planning to adress (1)?


    How many patients do you have and what are the proportions of patients per treatment group and outcomes per treatment?
    Dividing the numerator by the denominator for each group
    CABG: N=805. 22 with Left elevation (2.6%) 17 with Right elevation (2%)
    Aortic: N=942. 9 with L elevation (0.9%) 7 with R elevation (0.7%)
    Mitral/Tricuspid: N = 380 3 with L elevation (0.7%) 15 with R elevation (4%).


    You can compare the differences between the groups, however - these may not be the groups that get used in the future analyses by the Stats Team. Not controlling for the patient differences in (1) could lead to spurious results. Then just controlling for them is about as comparable to propensity match for about 90% of time, though typically just controlling for variables overestimate actual differences. Big picture, Just comparing and reporting differences in (1) could be detrimental to your field and your reputation, if results are not robust to all of the issues presented.
    Understood, but I hope my previous replies clarify the structure of the study.

    Another thing to consider if you compare characteristics of THESE groups, is correcting for multiplicity or false discovery rates given you will be potentially making 3 comparisons if omnibus tests are significant. Fisher's could be used, however for your continuous data you would use ANOVA followed by pairwise t-test or if data normality assumption is nor met, then Kruskal-Wallis followed by Wilcoxon rank sum (medians).

    I applaud your interest, though if you have apprehension about your ability to run the stats, I would just wait or recruit some help. Or play around with these data and not attempt to report them at a conference.
    Thank you. Basically I am writing a prelim manuscript on my own for purposes of completing a capstone, which will not be published or rigorously reviewed. This is basically a formality due to the timing of a due date. I just want to be able to say something basically. What I would eventually publish or present would have the input of a stats team.

    Hence Im just looking for simple stats tests that wouldnt be wildly inapproriate, or that would be the starting point for a full professional analysis. For example, can I do a 2x2 table for each group using a Fisher analysis that would have the rows be Diaphragm Elevation/No Diaphrgam Elevation and the columns be REspiratory Failure/No REspiratory failure to see if diaphragm elevation possibly is associated with a greater risk of post operative resp failure. Or, as another example, have the rows be Diabetes/No Diabetes and the columns be Diaphragm Elevation/No Diaphragm elevation to see if pre-operative diabetes is a potential risk factor for diaphragm elevation post operatively (both of these examples have biologically plausible mechanisms).
    Last edited by boomer; 02-25-2015 at 04:22 PM.

  5. #4
    Omega Contributor
    Points: 38,374, Level: 100
    Level completed: 0%, Points required for next Level: 0
    hlsmith's Avatar
    Location
    Not Ames, IA
    Posts
    6,998
    Thanks
    398
    Thanked 1,186 Times in 1,147 Posts

    Re: Retrospective review analysis questions


    Yeah you could play around with those as long as it seems appropriate the collapse left and right together. Good luck and let us know if you have more questions.
    Stop cowardice, ban guns!

+ Reply to Thread

           




Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts






Advertise on Talk Stats