topic Re: Median Days To Diagnosis in Statistical Procedures

Median Days To Diagnosis

DanielQuay — Thu, 08 Jun 2023 12:23:31 GMT

Good Morning Everyone, or whatever time of day it is.

I have talked myself into a circle and I'm not sure which path to follow at this point. So, I thought I would ask the experts.

What I am trying to do is determine if there is a statistically significant difference in the time it takes to reach a diagnosis based on the result of the diagnosis.

For example, does it take significantly longer to reach a final diagnosis based on whether a tumor is malignant or benign.

Calculating the median number of days to final diagnosis I believe I have figured out successfully.
But then I have found myself a bit stuck. I considered ANOVA but the time data violates the independence assumption.

I considered proc lifetest, conducting a peto test, and probdf but I'm uncertain which to use in this scenario.

Any guidance would be helpful. Thanks

Re: Median Days To Diagnosis

PaigeMiller — Thu, 08 Jun 2023 12:27:12 GMT

But then I have found myself a bit stuck. I considered ANOVA but the time data violates the independence assumption.

Explain further. If the variable of interest is number of days to diagnosis, there is no violation of the independence assumption as this is not time series data.

Re: Median Days To Diagnosis

DanielQuay — Thu, 08 Jun 2023 12:31:24 GMT

That is where I've talked myself into a circle. Doing a wilcoxon analysis seemed like it would be appropriate, but I haven't used time data in ... 12 years. Can you help me wrap my head around how this isn't time series data? I need a better understanding here so I can defend my methodology if challenged.

Re: Median Days To Diagnosis

PaigeMiller — Thu, 08 Jun 2023 12:32:47 GMT

Person 1 is diagnosed in 12 days

Person 2 is diagnosed in 24 days

Person 3 is diagnosed in 17 days

...

I don't see any time series here.

Re: Median Days To Diagnosis

SteveDenham — Tue, 20 Jun 2023 18:51:07 GMT

PROC LIFETEST is probably the simplest way to calculate the median value for a time to event variable, so long as not more than 50% of the observations are censored. The examples in the documentation are really quite good. Run those, and then see how to change to your variables.

SteveDenham

Re: Median Days To Diagnosis

Reeza — Tue, 20 Jun 2023 19:42:44 GMT

Time series would have multiple measurements at different time for the same individual. You do not have that here.

Because you have to have a final diagnosis for this analysis (benign/malignant) you don't have censored data either, so GLM/ANOVA is suitable. If your errors are not normally distributed, you can then switch to a non-parametric test.