We agree with Dr. Hortin that some hyperthyroid patients have detectable TSH levels, but we disagree with his conclusion that a decision limit of 0.1 mU/L should be used to screen for thyroid dysfunction in primary care practice. Dr. Hortin's statement that when third-generation TSH assays are used, 4% to 17% of patients with overt thyrotoxicosis have a detectable TSH level applies to samples of patients with known thyroid disease seen at tertiary teaching hospitals; it does not apply to the general population as he suggests. Moreover, the functional sensitivity of most "ultrasensitive" assays in current use is 0.05 to 0.2 mU/L (1), not 0.01 or 0.02 mU/L.

The relation between the functional sensitivity of an assay and its cost or effectiveness as the initial test for screening is by no means direct (2). In clinical studies of screening with assays that have a functional sensitivity of 0.05 to 0.2 mU/L, the use of a decision limit higher than the detection limit adds little or nothing to the yield of overt hyperthyroidism; however, it adds considerably to the number of false-positive screening test results. The optimal decision limits for the newest assays will not be known until they are evaluated as a tool for identifying unsuspected disease in a generally healthy population.

We agree that there is a trend toward an association between subclinical hypothyroidism and atrial fibrillation in the Framingham data, but it is not as strong or as certain as association as Dr. Zelmanovitz implies. In patients with subclinical hypothyroidism who did not take thyroid hormone, the relative risk for atrial fibrillation (1.6) was not statistically significant (P = 0.06) and, rather than "similar," was much lower than the relative risk for atrial fibrillation in patients with a low TSH level (3.1; P < 0.001) (3).

Dr. Friedman's hypothesis, that baseline data on healthy young persons might prove useful later in the event of depression, has not been tested in clinical studies.

We thank Dr. Dieter for his comments. It is plausible that patients who have a markedly elevated TSH level may be more likely to have a total cholesterol level more than 6.2 mmol/L. Advocates of broader screening also used the estimate we used (4), but the literature is incomplete on this point. We hope that future studies resolve this question.