Multi-channel laboratory testing and the unexpected abnormal result: a statistical myth corrected.

Multi-channel automatic analysers generate unexpected abnormal test results impacting medical economics, policy and practice. Critics caution that many such abnormal results should be expected due simply to the statistical definition of normal limits. With a test battery of n channels the probability p that a healthy individual will be within limits on all tests, each with probability s, is usually predicted p = sn. In most clinical situations, however, the assumptions upon which this formula rests are not met and it overestimates the occurrence of abnormal results. In three screening programmes involving blood chemistry batteries on 769 patients the formula gave overestimates (p less than 0.001) of the percentage of individuals with at least one abnormal result. For a battery of eight tests 33.7 percent of patients were predicted to have abnormal results but only 20.6 percent were observed, with 14 tests 51.2 percent were predicted and only 37.1 percent observed, for 17 tests 58 percent were predicted and only 33.3 percent observed. Clinical judgment, not misapplied probability theory, should guide the physician's strategy in evaluating abnormal test results.