Author: Vishal Nangalia
As medical machine learning evolves and new models to predict patient outcome, and recommend clinical interventions are developed a key problem arises, specifically that of how to incorporate these advanced analytics into clinical practice. Culturally, new clinical decision support systems need to provide actionable insight to the clinician rather than just increase their ‘screentime’ on a computer system. Rather than providing a risk of condition occurring, what is needed is a recommendation for the next most appropriate intervention that needs to be undertake. These recommended interventions could range from additional diagnostics, referrals to specialist teams, or even intravenous fluids or drugs. From a technical perspective, hospital information systems vary, some hospitals have electronic health records, while others are still paper based with only basic electronic patient administrative systems. Although, new standards such as FHIR have recently been described, the majority of existing systems still use decade old protocols. Therefore any technical solution needs to work within this diverse environment. The solution we have successfully built addresses both these cultural and technical challenges. Patient Rescue is a cloud based real time event based analytic system. At the time of first implementation in 2015 it was the only IGSOC (UK version of HIPAA) compliant event based healthcare analytic system in existence. Data received via HL7 channels (from version 2.xx onwards to FHIR) is persisted in a database and combined with any bulk imported data (for example, historic patient information). Along with real-time analysis of messages, algorithms can be run on the persisted data in order to perform one-off analyses. The platform has the following features: 1. Data from different sources are mapped to a canonical form before being analysed and persisted. This overcomes the issue of variations in the terminology and meaning of information sent. 2. The components of the platform are loosely coupled. This means that different implementation technologies, algorithms and actions as a result of analysis can be plugged in to the system. 3. The platform is highly scalable: a. New sources of data can be added quickly. b. The database schema is optimised for large quantities of data, allowing exceptionally fast queries to be performed on large datasets. c. Components can be distributed across multiple servers. d. Components that are ‘plugged into’ the system (e.g. algorithms and alerting modules) can subscribe to ‘topics’ of interest. This system was successfully deployed and tested at four hospitals in London, UK. The algorithm deployed related to Acute Kidney Injury, and milliseconds after a patient was admitted, a blood test result being available, or another event of interest occurring the Aki algorithm analysed the patient’s information. Depending on the results of this algorithm a custom report was then generated and communicated directly with the clinician, which highlighted the key clinical features of the patient, and also recommended the next steps. This entire process occurred within 1 second. This platform successfully demonstrates that advanced analytics can be incorporated into the workflow of clinicians while working alongside existing hospital IT.
Co Author/Co-Investigator Names/Professional Title: Simon Brown, Ani Dwarakanath, Prashant Lele
Funding Acknowledgement (If Applicable): NHS England, Medical Research Council