The Acceptability and Impact of Diarrheal Etiology Prediction (DEP) Algorithm

Overview

This is a randomized crossover study, where clinicians will be randomized to periods where they will use a rehydration calculator application with or without the Diarrheal Etiology Prediction (DEP) algorithm. The crossover will include a washout period to reduce carryover effect. The study will be conducted over a 9-week period. The Investigators will use a random number generator to randomize clinicians to DEP (use of the etiology calculator) or control arm (use of a previously-tested rehydration calculator) within site for the first 4 weeks. After the first 4 weeks, there will be 1-week washout period without decision support, after which each clinician will cross-over to the other arm for the next 4 weeks. The Investigators will enroll diarrhea-treating clinicians who treat children presenting with acute diarrhea at sites in Bangladesh and Mali. Utah investigators will only analyze de-identified data provided by our collaborators in Bangladesh and Mali.

Diarrhoeal diseases are a leading cause of morbidity and mortality in children worldwide, with an estimated one billion cases and 500,000 deaths annually. While the majority of deaths due to diarrhoea occur in lower-income countries, infectious diarrhoea remains a significant problem in high-income countries. Aside from the immediate morbidity, potential long-term sequelae of diarrhoea in children in low-resource settings include malnutrition, growth faltering, and deficits in cognitive development. While the cornerstone of diarrhoeal disease management in children is rehydration, a number of other management decisions, including the use of antibiotics and laboratory testing, may impact the course of disease. Overuse of antibiotics may cause side-effects and lead to increased antimicrobial resistance in the community. Underuse of antimicrobials for some bacterial and protozoal pathogens may lead to prolonged duration of illness and facilitate transmission. This can result in increased days of school or parental work missed, and among malnourished children in resource-poor settings, potential for growth faltering or death. Overuse of laboratory testing may have financial impact on both the patient and the healthcare system, and underuse may delay appropriate therapy or prevent recognition of outbreaks. Thus, accurate and cost-effective determination of diarrhoea etiology is important for proper case management in children and for public health. Clinical prediction rules (CPRs) help clinicians interpret clinical information and can improve decision making. A recent systematic review showed that out of 137 studies of conditions for which clinical prediction rules have been developed for children only 2 were for diarrhoea, both of which are for the assessment of dehydration. Similarly, the majority of available guidelines for pediatric diarrhoea are focused on the route, timing, and choice of fluids for rehydration. A few studies in the past 30 years studied the use of clinical predictors to estimate the probability of a bacterial cause of diarrhoea. However, these studies were limited by low rates of pathogen identification, small sample sizes, use of a single study site, and suboptimal prediction performance. Given the lack of guidelines and effective clinical predictors, decisions for use of antibiotics and laboratory testing are mostly empiric in nature, based on a number of "rules of thumb" for which evidence is scant. Unfortunately, physician judgment does very poorly to predict both need for antibiotics and correct type of testing. A recent study of children presenting to Kenyan hospitals with diarrhoea showed that reliance on dysentery as a proxy for Shigella infection led to the failure to diagnose Shigellosis in nearly 90% of cases. Better tools for decision making and evidence-based guidelines regarding use of antibiotics and laboratory testing in children with diarrhoea are clearly needed. The majority of decisions for use of antibiotics in diarrhoeal illnesses are made empirically. In lower- and middle-income countries (LMICs), due to cost constraints, etiological diagnosis is rarely made, and a large number (up to 70%) of patients with acute diarrhoea are prescribed antibiotics. However, in contrast to high resource settings, bacterial pathogens may be very common in low resource settings. In the multicenter Global Enteric Multicenter Study (GEMS) study, the investigators found that detection of Shigella ranged from 16-78% of children with dysentery and 2-43% of children with watery diarrhoea. In both high and low resource settings, inappropriate use of antimicrobials leads to unnecessary toxicity for the individual, increased costs and an increase in antibiotic resistance in the community. Thus, methods for guiding appropriate use of antibiotics for pediatric diarrhoea in both high- and low-resource settings are urgently needed. Clinical prediction rules (CPRs) are decision-making rubrics that help clinicians estimate the likelihood of a patient outcome. A number of prominent prediction scores have been widely adopted for clinical use. Examples include the CHADS2 score for stroke risk in patients with atrial fibrillation, the TIMI score for mortality in patients with NSTEMI, and the CURB-65 Score for mortality in community-acquired pneumonia. Clinical prediction rules integrated into clinical decision-making have the ability to direct clinicians towards more evidence-based behaviors, resulting in improved care and reduction of costs. CPRs can also reduce antimicrobial usage, as shown by the use of scores for strep pharyngitis, and linking CPRs with testing guidance may further reduce antimicrobial usage. Thus, clinical prediction rules have the potential to help healthcare workers worldwide address clinical uncertainty and provide improved care for children with diarrhoea. The investigators have recently use data from GEMS to derive a viral etiology prediction rule with an internal cross-validated area under the curve (AUC) of approximately 0.85. The investigators used a post-test odds formulation method, which takes into account odds from multiple models or tests. First, using data from GEMS, the investigators trained a logistic regression model with viral etiology as dependent variable using the five most predictive clinical variables as independent variables. The investigators then trained models with the same viral etiology response using both local climate and recent clinical trends as independent variables. For each model, odds of viral etiology versus other known aetiologies are generated by estimating the conditional distribution of training predictions using kernel density estimates. The odds generated for each model are multiplied along with a pre-test odds to determine an overall odds of a viral etiology. The investigators have now transferred the calculation of this prediction rule into a smart-phone application, called the Diarrheal Etiology Prediction (DEP) algorithm. The investigators have used TAC data from the VIDA (Vaccine Impact on Diarrhea in Africa) study to externally validate the DEP algorithm's ability to predict viral etiology of diarrhea. In this application, our objective is to determine the acceptability and use of the DEP among clinicians caring for children with diarrhea.