Cox regression with varying data completion


New Member

Many thanks for providing a fantastic resource in many previous projects, I have often found very helpful answers to simple questions browsing the forums here. Today however I have become unstuck. Please forgive my poor basic understanding of stats.

My usual statistician has been abducted by his fiancée and is taking weeks to answer emails, I have two papers I cannot proceed further with until I have the results so I am trying to go on by myself. One is far too tricky, the other should be possible.

In the analysis I want to perform I have 166 patients with a terminal lung disease, 57 have died over the study period. I want to do a Cox regression to look at hazard ratio given the following variables:
- gradient of two markers of lung function,
- the level of a putative biomarker of disease activity
- the variance of the biomarker level for each patient

The problem comes that only 22 of the patients who died have all the data.

22 are missing one of the lung function parameters, a different but not mutually exclusive 22 are missing the other lung function parameter.

One patient is missing biomarker data.

The biomarker is not the only area of interest here, the cohort were treated in a manner significantly different from the norm, it would therefore be useful to see if the relationship between rate of decline and mortality matches the more common data.

Would regressions for each variable individually be robust if I only included the cases with data for that variable? There is significant variation in the numbers so I don't want to estimate missing values.

Is there some other form of multivariate regression I can do that will work around missing values?

Many thanks for your time and help.