Although pituitary thyrotropin (TSH) and thyroid hormones are physiologically interrelated, interpretation of measurements is conventionally done separately. Classification of subclinical thyroid dysfunction depends by definition solely on an abnormal TSH. This study examines a composite multivariate approach to disease classification.MethodsBivariate and trivariate reference limits were derived from a thyroid-healthy control group (n=271) and applied to a clinically diverse sample (n=820) from a prospective study, comparing their diagnostic efficiency with the conventional method.ResultsThe following 95% reference limits were derived from the control group: (i) separate reference intervals for TSH, free thyroxine (FT4) and free triiodothyronine (FT3); (ii) bivariate composite reference limits for the logarithmically transformed TSH and FT4, and (iii) trivariate composite reference limits including all three parameters. A multivariate approach converts the “rectangular” or “cuboid” graphical representations of the independent parameters into an ellipse or ellipsoid. When applying these reference limits to the clinical sample, thyroid dysfunctions were classified differently, compared with the separate method, in 6.3 or 12% of all cases by the bivariate or trivariate method respectively. Of the established dysfunctions according to the separate intervals, 26% were reclassified to “euthyroid” by using the bivariate limit. Discrepancies from the laboratory-evaluated reference range were less pronounced.ConclusionsFrequent divergencies between composite multivariate reference limits and a combination of separate univariate reference intervals suggest that statistical analytic techniques may heavily influence thyroid disease classification. This challenges the validity of the conjoined roles of TSH currently employed as both a sensitive screening test and a reliable classification tool for thyroid disease.