Sepsis is a leading cause of death in United States hospitals. Sepsis-related mortality is approximately 150% higher when compared to other mortality causes. Early and ongoing prediction of sepsis is crucial in preventing mortality. Sepsis mortality increases significantly with each hour of delay in administration. Further, septic patients have an average length of stay estimated at 60% longer than other conditions. Clinical evidence indicates that patients with sepsis manifest clinical signs or symptoms several hours before the condition worsens. Delay in detection and communication among clinicians, nurses, and pharmacist exacerbates delay in sepsis management.
Predictive AI uses machine learning in an attempt to determine the future and prevent bad outcomes. Sepsis Predictive AI model learns from 80+ key indicators and classifications to determine early signs. Both generative AI and and predictive AI use machine learning. However, generative AI turns machine learning inputs into content whereas predictive determines probability of future
HiPaaS data Converter with real time clinical streaming enables Hospitals to build production grade, highly available infrastructure for deployment of the sepsis or any predictive AI models. In addition, HiPaaS enables notification back to EHR system for the nursing staff and physicians. To meet the production grade, HiPaaS platform was developed with iterative approach and included learning from unsucessful endeavors. Following is the implementation patterns:
Train the Predictive AI Model: Large amount of historical data is required for model to train itself. The scientists and medical staff define the key indicators for the model and create a set of decision trees and then combining many trees together to predict outcome . The model once defined, has to then be fed with terabytes of historical data with right indicators, conditions, and outcomes. HiPaaS enables the conversion of vast amounts of patient data, including medical records, vital signs, and lab results which is in various formats , from varied timelines and separate storage mechanisms.
Real Time Streaming of data: Once the model is trained and tested, to enable it in production, it needs feeds from EHR system to run the model and predict. HiPaaS clinical streaming gets HL7 or medical data feeds from the EHR and further filters data in real time with required indicators like vital signs ( like systolic blood pressure, diastolic blood pressure, heart rate, respiratory rate, temperature, weight, etc.), medication(like anti-infective, cardiovascular agent, etc.), demographics (race, marital status, gender, age), laboratory values(like albumin, calcium, creatinine, glucose, hematocrit, hemoglobin, etc.), comorbidities and miscellaneous ( hours since admission, etc.) . The filtered data is then provided to model via python wrapper or web services. IN some cases the R model is converted to python code.
Real-time Results and Early Warning Systems: By detecting conditions at an early stage, healthcare providers can intervene promptly and initiate appropriate treatments, potentially preventing the progression of diseases and reducing mortality rates. HiPaaS executes the model every 5 to 15 minutes (based on model requirements) and integrates the results to EHR to display it in real time to the nursing staff or medical care givers on floor. The result is simplified in to probability metrics of 0 to 1 where above certain probability needs immediate manual intervention. The values are color coded to help the staff to focus on actions rather than interpretations.
Once HiPaaS enables production grade, highly available infrastructure for deployment of the sepsis predictive model, it is also extensible to other models and indicators with same datasets. HiPaaS can be deployed on premise or in your existing high available private cloud.
You can find more information at HiPaaS FHIR Converter for Amazon Healthcare and Clinical Streaming. It is also available on AWS marketplace
More about HiPaaS at www.hipaas.com. HiPaaS is Trademark of HiPaaS Inc.