Letter to the editor

Alternative Medicine Review,  Sept, 2004  by Alan R. Gaby,  Patricia M. Herman

• E-mail
• Print
• Link

In their single-subject experiment evaluating nine different food sensitivity tests (Altern Med Rev 2004:9:198-207), Drs. Herman and Drost conclude that "a number of tests may be useful in identifying foods to which a patient is reactive: however, no one test is likely to identify all reactive foods." Such a conclusion may give more credibility to these tests than is warranted.

The sensitivity of at least some of the lab tests may actually be far lower than the authors reported because their calculations were based on a very small number of food challenges that were not chosen at random. Rather, some of the foods were selected for challenge specifically because they had been reported by one or more labs to be reactive. Such a selection method is likely to result in erroneously high sensitivity values.

The study subject performed individual challenges on eight foods, of which five evoked symptoms. In contrast, a total of 294 foods and other substances were tested by the various laboratories, 156 (53.1%) of which were considered reactive by one or more tests. The proportion of positive test results reported by the different labs ranged from as low as 0.5 percent to as high as 61 percent.

Because the study subject reacted to 62.5 percent of the foods she challenged, one might reasonably conclude that she is a multiple food reactor. In the absence of additional data, the best estimate one can make is that she would have reacted to 62.5 percent of the 294 substances tested by the various labs, if she had performed individual challenges on all of them.

Consider now the results from Lab B. The sensitivity of the test was said to be 40 percent because it reported a positive result for two of the five foods to which the subject reacted. Looking at the entire report from Lab B, however, reveals that it tested 90 foods, but identified only three reactive foods. Because two of those three reactive foods were selected arbitrarily for food challenges, the results were biased toward finding greater test sensitivity. Had the subject challenged all 90 of the foods screened by Lab B, she presumably would have reacted to approximately 56 foods (62.5% of 90). Lab B detected only three of an estimated 56 reactions, so the true sensitivity of its test was only 5.4 percent (3 divided by 56), rather than the reported 40 percent.

Lab A tested 115 foods, of which 12 (10.4%) were positive. Using the same assumptions as described previously concerning the subject's frequency of true food reactions, one would have expected 72 positive results (62.5% of 115). Thus, the test sensitivity for Lab A should have been 16.7 percent, rather than 60 percent as reported by Herman and Drost. Similar reasoning would reduce the reported sensitivity of several of the other lab tests evaluated.

The specificity of some of the lab tests may also have been overestimated. For example, the Carroll test was given a specificity rating of 100 percent because the subject reacted to the only food that came out positive on that test. However, for a person who reacts to 62.5 percent of foods, any randomly generated report that identifies only a single reactive food would have a 62.5-percent chance of being assigned a specificity of 100 percent. The test from Lab B was also said to have a specificity of 100 percent, based on two positive tests, both of which were correct. For a person who reacts to 62.5 percent of foods, the probability that any randomly generated pair of positive results would both have been correct is approximately 36 percent. Thus, these 100-percent specificity ratings may have occurred purely by chance, and the statistical 95-percent confidence interval includes zero-percent specificity for both of these lab tests.

The fact that very little consistency was found among different labs that purportedly measured the same parameters, combined with previous observations that some labs could not even duplicate their own results on the same blood samples, (1) raises further doubt as to clinical usefulness of some of these tests.

Sincerely,

Alan R. Gaby, M.D.

Carlisle, PA

Reference: 1. Miller SB. IgG food allergy testing by ELISA/EIA: what do they really tell us? Townsend Letter for Doctors and Patients 1998(January):62-65.

To the Editor:

Dr. Gaby makes a number of valid points. It is true that if we had been able to challenge all the foods found reactive, the sensitivity and specificity results may have been significantly different than reported, and his letter illustrates this well. Our challenge sample was purposely not all-inclusive, nor random. It was drawn in a way to maximize the differences across tests with regard to sensitivity and specificity. It was also (as noted in the article) only one sample drawn on one subject and tested one time. Therefore, perhaps what we should take from this letter and the paper is that sensitivity and specificity seem to vary widely across food sensitivity tests. However, due to the fact that different tests measure different underlying mechanisms of food sensitivity (as explained in the paper), a more important point is that it is unlikely that any one test can be assumed to identify all reactive foods.