Background: Many patients who have involuntary weight loss have cancer. The Hernandez prediction rule includes 5 variables (elevated levels of alkaline phosphatase and lactate dehydrogenase, low albumin, high white blood cell count, and age >80 years). The purpose of this study was to evaluate the validity of the prediction rule.
Methods: We prospectively evaluated 290 consecutive inpatients and outpatients who had involuntary weight loss. Clinical, hematologic, and biochemical parameters were determined. There were 259 patients who had follow-up at 6 months to determine the cause of involuntary weight loss, and 31 other patients were lost to follow-up. The 5 variables were introduced into a regression logistic model with cancer as a dependent variable.
Results: Cancer was diagnosed in 72 of the 290 patients (25%) who had involuntary weight loss. Bivariate analysis showed that serum albumin, C-reactive protein, erythrocyte sedimentation rate, alkaline phosphatase, iron, lactate dehydrogenase, white blood cell count, hemoglobin, and ferritin levels were associated with cancer (range of area under the receiver operating characteristic curve, 0.589 to 0.688). Multivariate analysis showed that albumin, erythrocyte sedimentation rate, iron, white blood cell count, and lactate dehydrogenase levels were associated with cancer. When dichotomized, only low albumin (odds ratio, 2.6, CI [1.3-5.2]) and high alkaline phosphatase (odds ratio, 2.3, CI [1.7-4.7]) were associated with cancer. The area under the receiver operating characteristic curve of the 5-variable prediction rule was only 0.70 (95% confidence interval, 0.61-0.78). The negative predictive value of this model with 3 variables (age >60 y, alkaline phosphatase, and albumin level) increased from 85% to 95% when all tests were negative.
Conclusions: In patients who had involuntary weight loss, those who have cancer are likely to have ≥1 abnormal laboratory test. The 5-variable prediction rule had a significantly lower accuracy than originally reported. Further evaluation of the 3-variable modification of the prediction rule may be useful.