50944_en-US

liliconcom

measurements

For group of cancer patients measured the number of metastases before and after treatment.

The character of changes (less metastases or more metastases after treatment) is an assessment of the effectiveness of the treatment.

What kind of stat. analysis conduct?

 

Re: measurements

Thank you for so careful consideration of data!

Re: paired measurements

As Cameron has already pointed out, a simple matched pairs (before-after) would be a natural approach.  The only thing (aside from lack of any timing) that bothers me is that the number of metastases decreased for all the patients except the 2 that died.  So, there may be another relevant analysis - a classification model as to whether the treatment causes death.  But your sample is too small and I don't think there is enough information to say anything about whether that is the case (is there any kind of control group?).  It appears from the data you have that the treatment reduces the number of metastases except when the patient dies.

Re: measurements

I understand well the notes about additional elements of the model (in particular, time). Thank you!

Re: Repeated measurements

Isn't there a time variable missing from this data?  I'm not sure how meaningful it is to measure the number of metastases without recording when he measurements are taken.  Were they all taken at the same time after treatment?  Given that two patients died, it seems to me that there is at least some time variation in the data (as well as some censoring).  So, while the matched pairs approach may give you a crude estimate of whether or not the treatment reduces metastases, I'm not sure the precise measurements are meaningful without knowing more about the timing of the measurements.

Re: measurements

Cameron, my deep gratitude for your approach.
"Are these patients listed in any kind of time order?"     No!

Re: Repeated measurements

Matched Pairs is the first thing I would think of, but I you need to check the appropriate assumptions.  If you plot the difference (before-after) against the before count, you get this:

The good news is, whatever this treatment is, sure seems to reduce the number of metstases.  However, the fact that the effect of the treatment is also a function of how bad the patient's case is beforehand, you need to account for that in the test/model.

In the wide-format data, add a column that computes the difference.  This is your response.  Then, fit a regression line to the differences by the before count.  You'll find that the slope term is 0.357, which means for every additional pre-existing metastasis, the treatment will reduce 0.357 more mestastases on average.  See the attached script.

My only concern with this approach is the independence plot.  Are these patients listed in any kind of time order?  If so, there might be an issue there, especially since the last 2 patients are the ones with increased metastases.

Re: measurements

Thanks Jim! I will try

Re: Repeated measurements

Using your wide format, check out

     Analyze==>Specialized Modeling==>Matched Pairs