contingency table: categories are not mutually exclusive, and data is not all paired


I'm confused about what test I should be using and hope someone can help.

I have data from microbiological samples where the bacteria's sensitivity to various antibiotics was tested. A sample might have been tested for sensitivity to loads of different antibiotics (say up to 10), or only a few antibiotics. In each case, the sample might have been sensitive to the antibiotic or resistant (or not tested).

So my data is something like this (only with more antibiotics and samples):
Sample1: amoxicillin-resistant, ciprofloxacin-sensitive, piperacillin-untested, trimethoprim-sensitive
Sample2: amoxicillin-resistant, ciprofloxacin-sensitive, piperacillin-sensitive, trimethoprim-resistant
Sample3: amoxicillin-untested, ciprofloxacin-sensitive, piperacillin-sensitive, trimethopirim-resistant

So multiple tests are carried out on the same sample, and positive results with different antibiotics are not mutually exclusive - a sample could be sensitive to several different antibiotics, a few, or none.

Desired outcome:
I want to test whether one antibiotic (say ciprofloxacin) is significantly more effective than another (say amoxicillin) for the samples in which each was used. (I'd also do something similar comparing other pairs.)

(1) I think I can use a McNemar's test for the samples where both antibiotics were used.
I'd end up with a 2x2 contingency table of amox-sens, amox-res v. cipro-sens, cipro-res.
Would this be a correct choice?

(2) I'd also like to check whether the proportion of the organisms which were tested against ciprofloxacin (i.e. all organisms tested with ciprofloxacin, irrespective of whether they were also tested against amoxicillin) that were ciprofloxacin-sensitive was significantly greater than the proportion of all the organisms tested against amoxicillin that were amoxicillin-sensitive.What is a suitable test for this?
I don't know what I should be using here. Presumably not chi-squared since the categories are not mutually exclusive (a sample could be both amox-sensitive and cipro-sensitive). Is McNemar's ruled out because the data is no longer paired (since some samples were only tested against one drug and not the other)?

I thought of using a 2 x 3 where I'd have sensitive, resistant and untested categories for each drug, but really I want to exclude the untested and just compare the sensitive and resistant proportions. And even if I included an untested category I'm still not sure where that gets me! Still can't use chi-squared? :confused:


[Note: I know that for method 2 I'd need to add a caveat in my results that certain drugs may appear to have performed well but potentially only because they were tested on a small, select group of samples against which they were likely to work whereas if they'd been tested against a wider range they might have been effective against a lower proportion.]

Any advice would be appreciated - thanks!
(I was planning to use SPSS/PASW if that makes any difference.)