Comparison of Thyroid Hormone and Cortisol Measurements with APACHE II and TISS Scoring Systems as Predictors of Mortality in the Medical Intensive Care Unit

Abstract

Thyroid and adrenal functions were evaluated in 49 consecutive critically ill, mechanically ventilated patients admitted to the medical intensive care unit (MICU) at Ben Taub General Hospital in Houston, TX. On admission, severity of illness was assessed by Acute Physiological Assessment and Chronic Health Evaluation II (APACHE II) and Therapeutic Intervention Scoring System (TISS) scores. Thyrotropin Stimulating hormone (TSH) measured with a sensitive assay, thyroxine (T4), tri-iodothyronine (T3), T3 resin uptake, and Cortisol levels were measured on the first MICU day at 8 am, and results were assessed based on ability to predict mortality. It was shown that severe hypothyroxinemia (T4 < 3 μg/dL) was associated with a 75% mortality as opposed to a mortality rate of 19% in patients with a T4 level of 3 μg/dL or more ( p < 0.01), and a high 8 am Cortisol level (>30 μg/mL) was associated with a 56% mortality versus 0% in patients with Cortisol values less than 30 μg/dL ( p < 0.005), whereas a high APACHE II (>25) score and a high TISS (>25) score were only associated with a mortality of 47% ( p = 0.06) and 32% (NS), respectively, versus 21 and 20%, respectively, in patients with scores less than 25. Logistical regression analysis revealed that of all the variables, 8 am Cortisol level had the best predictive value of outcome, followed by T4 levels and T3 resin uptake levels. Although high Cortisol and low T4 levels used alone had a sensitivity of 56 and 75% and a specificity of 100 and 80.5%, respectively, combined low T4 and high Cortisol levels had a sensitivity of 100% and a specificity of 81.5% in predicting mortality. Very low T4 levels, due to severe illness, combined with high Cortisol levels (implying high physiological stress), are more sensitive predictors of mortality than APACHE II, TISS, Cortisol levels, or T4 levels alone, or any other combination of predictors.