A prediction model for moderate or severe dehydration in children with diarrhoea

A hospital-based unmatched case-control study (387 cases and 387 controls) was carried out at the Government Medical College Hospital, Nagpur, India, to devise and validate a risk-scoring system for predicting the development of moderate or severe dehydration in children, aged less than five years, with acute watery diarrhoea. On unconditional multiple logistic regression, 12 risk factors--infancy, minority religion, undernutrition, not washing hands by mother before preparation of food, frequency of stools > 8/day, frequency of vomiting > 2/day, measles in previous 6 months, withdrawal of breast-feeding/other feedings, withdrawal of fluids during diarrhoea, not giving oral rehydration solutions (ORS), home available fluids and both during diarrhoea--were significant. Based on regression coefficients, these factors were ascribed statistical weights of 5, 5, 4, 4, 22, 9, 11, 13, 5, 5, 5, and 7 respectively. The receiver-operating characteristic curve suggested a total score of 48 to be the best cut-off for predicting the development of moderate or severe dehydration. At this cut-off, the sensitivity, specificity, positive predictive value, Cohen's kappa, and overall predictive accuracy were 0.81, 0.81, 0.81, 0.61, and 0.86 respectively. If substantiated by further validation, this system can be used for predicting the development of dehydration at the earlier stage, thereby reducing the mortality associated with life-threatening dehydration.