Health Risk and Underweight

Abstract

Anorexia nervosa is associated with a significant risk of morbidity and mortality. In clinical practice, health risk is assessed and estimated using routinely collected laboratory data. This study will develop a risk score using clinically relevant laboratory parameters. The related question is how to estimate the health risk associated with underweight using body weight, height and age. Methods: We used routinely collected laboratory parameters from a total of 4087 patients. The risk score was calculated on the basis of electrolytes, blood count, transaminases and LDH. The nine parameters used were summed as zlog-transformed values. Where appropriate, the scales were inverted so that high values represented higher risk. For statistical prediction of the risk score, weight/height and age reference values from the WHO, the CDC (Center of Disease Control) and representative studies of German children and adults (KIGGS and NNS) were used. Results: The score calculated from nine laboratory parameters already shows a convincing relationship with BMI. Among the weight measures used for height and age, the z-score from the CDC reference population emerged as the best estimate, explaining 34% of the variance in health risk measured by the laboratory score. The percentile rank for each age-specific median weight from the KIGGS/NNS still explained more than 31% of the variance. In contrast, percentiles explained less variance than BMI without age correction. Conclusions: The score we used from routine laboratory parameters appears to be an appropriate measure for assessing the health risk associated with underweight, as measured by the quality of the association with BMI. For estimating health risk based on weight, height and age alone, z-scores and percentages of age-specific median weight, as opposed to percentiles, are appropriate parameters. However, the study also shows that existing age-specific BMI reference values do not represent risk optimally. Improved statistical estimation methods would be desirable.