>LONDON (Dec 15, 1995 - 21:00 EST) - Parkinson's Disease sufferers >treated with a combination of two drugs, selegiline and levodopa, >showed a higher mortality rate in a study than those treated >with levodopa alone, British researchers said on Friday. >The researchers working for the Parkinson's Disease Society studied >520 patients and found that 76 out of 271 taking both drugs died >over a 10-year period compared with 44 patients out of 249 taking >levodopa alone. Quick and dirty analysis for statistical significance: The total death % is (76+44)/520 = 23.1%. Test the null hypothesis that 76/271 is just random variation. For large samples, the normal distribution approximates the binomial (the binomial applies to "% nonconforming" or "% fatalities"- i.e. the proportion of yes/no or good/bad events in an experiment.) The standard deviation of the binomial is SQRT(np(1-p)), where n is the sample and p is the event fraction (0.231) Compute (observed-expected)/SQRT(np(1-p)) =(76-0.231*271)/SQRT(0.231*271(1-0.231))= 1.93 so the mortality rate for the eldepryl + sinemet group is 1.93 standard deviations above the expected rate. The chance of this happening due to random variation is 2.7%. Based on pure statistics, the results are significant. That is, there is only a 2.7% chance the selegline/levodopa group was just "unlucky"- there is a 97.3% chance that their mortality rate is really higher. The danger in accepting these results as significant is failure to look at other factors, like age. Is it possible that the selegline/levodopa group was, on average, older than the levodopa-only group? One needs to look at all the risk factors. A statistical analysis assumes the only difference between the groups is selegline vs. no selegline; if this assumption is not true, we cannot rely on the results. -Bill Levinson (industrial statistician)