The following post was written several years ago. Although more recent developments have changed the field of clinical laboratory science since the original posting, the information contained was deemed to be of historical interest.
Most clinicians rely on TSH to screen for thyroid disease, especially hypothyroidism (1). As the methods for TSH have improved, the number of cases with abnormal TSH values, but minimal symptoms has increased (2). Most physicians have relied on a TSH range of 0.3 to 5.0 mU/L; but the American Association of Clinical Endocrinologists (AACE) has recently advocated tightening up this range (0.4 to 2.5 mU/L), reasoning that many patients have subclinical hypothyroidism (3). The proposal has stirred some controversy in the endocrine community.
We all know what will happen if we move the upper end of the reference interval for TSH from 5.0 to 2.5 mU/L. We will increase the sensitivity of the test, but decrease its specificity, because there is overlap between normal patients and those who are hypothyroid. Biological Variation complicates this picture by spreading a patient’s values over a range and making the values bounce around the patient’s true value. While the CV for FT4 is only 3.5%, TSH has a one-week CV of 19.3%. If a patient had a true TSH of 3.0 mU/L, their 95% confidence interval would range from 1.4 to 4.6 mU/L, thus confusing the issue. A central two-thirds confidence interval would range from 2.2 to 3.8 mU/L, which could still be misinterpreted. Luckily TSH has a positively skewed distribution with its tail pointing towards higher values among euthyroid patients, so many patients would have values not straddling the 3.0 value. But still relying on one value could result in misinterpretation. It is not a common practice to obtain three TSH values and average them, but that might be a much better approach. Should we try to influence our clinicians that an average is better than one result? We would have to persuade insurers that obtaining multiple samples is the correct course.
In addition, if we had adequate baselines on our patients, we could detect individual changes over time, which would be a convenient way to personalize medicine. Is this a good time to ask for more testing, given the current high cost of healthcare?