Continuous diagnostic models for volume deficit in patients with acute diarrhea.

Background: Episodes of acute diarrhea lead to dehydration, and existing care algorithms base treatment around categorical estimates for fluid resuscitation. This study aims to develop models for the percentage dehydration (fluid deficit) in individuals with acute diarrhea, to better target treatment and avoid the potential sequelae of over or under resuscitation.

Methods: This study utilizes data from two prospective cohort studies of patients with acute diarrhea in Dhaka, Bangladesh. Data were collected on patient arrival, including weight, clinical signs and symptoms, and demographic information. Consecutive weights were obtained to determine the true volume deficit of each patient. Data were entered into two distinct forward stepwise regression logistic models (DHAKA for under 5 years and NIRUDAK for 5 years and over).

Results: A total of 782 patients were included in the final analysis of the DHAKA data set, and 2139 were included in the final analysis of the NIRUDAK data set. The best model for the DHAKA data achieved an R of 0.27 and a root mean square error (RMSE) of 3.7 (compared to R of 0.06 and RMSE of 5.5 with the World Health Organization child care algorithm) and selected 6 predictors. The best performance model for the NIRUDAK data achieved an R of 0.28 and a RMSE of 2.6 (compared to R of 0.08 and RMSE of 4.3 with the World Health Organization adolescent/adult care algorithm) and selected 7 predictors with 2 interactions.

Conclusions: These are the first mathematical models for patients with acute diarrhea that allow for the calculation of a patient's percentage dehydration (fluid deficit) and subsequent targeted treatment with fluid resuscitation. These findings are an improvement on existing World Health Organization care algorithms.