What is your sample size and what are you trying to get out of this? Also, is data prospectively collected? Not sure about SPSS, but I would run logistic regression then calculate odds ratios for 1 v 0, 2 v 0, and 3 v 0, and correct the alpha level used in the confidence intervals associated with the odds ratios to address repeated testing and false discovery risk. Some people will report a trend in the ordinal group by also rerunning the model with 0-3 as a continuous variable.
You can use Poison regression to get the relative risks if you have prospectively collected data.
I would like to add that we don't know what variable is the "factor" and what is the "effect". Regression with symptoms as predictors and stay in hospital as dependent variable also can be made, it more simple for interpretation. I think, more appropriate method is Cochrane-Armitage test. It can't be done in SPSS directly, but if you familiar with SPSS syntax, you can calculate it.
See the page: http://www-01.ibm.com/support/docview.wss?uid=swg21480127