purpose: This study was undertaken to assess whether circadian variation of thyrotropin (TSH) is affected by the severity of a nonthyroidal illness. patients and methods: On the second day of admission to a medical intensive care unit, 20 consecutive patients with the major diagnosis of acute respiratory failure underwent TSH measurements at 8 A.M., 9 P.M., 11 P.M., and 1 A.M., with two sensitive assays. results: Six patients died, five of whom had hypothyroxinemia (thyroxine [T4] less than 5.5 μg/dL) (83%) on the day of the study, whereas only three of the 14 survivors had low T4 (21%; p <0.05). Baseline 8 A.M. TSH measured with the two assays was similar in both groups and there was a progressive increase in TSH in survivors and a decrease in nonsurvivors at 9 P.M. and 11 P.M. However, the difference at these time points was not statistically significant. At 1 A.M., TSH levels were significantly lower among nonsurvivors (0.75 ± 0.34 μU/mL with assay 1, and 0.7 ± 0.4 μU/mL with assay 2) than in survivors (2.3 ± 0.46 μU/mL with assay 1, and 2 ± 0.5 μU/mL with assay 2; p <0.005; Wilcoxon test). Five of the nonsurvivors and none of the survivors had a suppressed 1 A.M. TSH level (p <0.001), suggesting a good correlation between suppressed 1 A.M. TSH and mortality. After exclusion of patients receiving drugs known to affect TSH levels (two nonsurvivors and four survivors), the same dissociation in TSH changes was observed, and significantly lower 1 A.M. TSH levels were observed in nonsurvivors than in survivors (0.13 ± 0.08 μU/mL versus 2.7 ± 0.6 μU/mL with assay 1; p <0.01). Cortisol levels were significantly higher only at 8 A.M. in nonsurvivors whether patients receiving drugs were included in the analysis (41.6 ± 3.2 versus 28.4 ± 2.7 μU/mL; p <0.01) or not (45.3 ± 4.6 versus 30.5 ± 3.6 μU/mL; p <0.01). At other times, cortisol levels were similar in both groups. The 24-hour TSH areas under the curve were also lower in nonsurvivors than in survivors whether patients receiving drugs known to affect TSH levels were included or not. However, cortisol areas under the curve were similar in both groups. conclusion: It is concluded that fatal illness is associated with a suppression of the late night TSH surge. © 1990.