Due to the above constraints, doctors resort to treating infants through the administration of antibiotics, regardless of the severity of the symptoms. We have therefore developed a new model with few non-invasive parameters and critical tests (CRP test, WBC test, etc.) which are available at primary hospital level so that the application will be useful for semi-urban and rural areas. The noninvasive parameters for babies include data points such as birth weight, frequency of stools and gestational age; for mothers, these parameters include mother BMI and blood pressure. The score generated by the platform provides a risk score through which a doctor would be able to make a risk assessment of the baby’s condition (whether the baby is moving towards sepsis or not).
During the time the application was being developed, we collected data in excel files and later data collection was done through the open source application. We then commenced the development of the predictive algorithm by analyzing the data collected. Through inputting different neonate data points, the platform generates a predicate score that doctors can use in their diagnosis of neonatal sepsis. The platform provides three levels of risks – Low, Medium and High — through which a doctor can assess a baby’s probability for an onset of sepsis.
We iterated upon the predictive model and fine-tuned the algorithm to generate the predictive score. We then compared the predictive scores generated via the PresCo platform with the blood culture results obtained from laboratories. With culture tests being the gold standard, we compared our score with the culture tests results that came in after about 2-3 days. We achieved an accuracy ranging from 85% to 90%.
The application was then tested by multiple health care professionals for accessibility, user-friendliness and accuracy of the score. Uuser requirements and methods used for identifying sepsis are quite different between urban and rural hospitals, which was indeed a huge learning during our prototype development.