S P E C I A L I S T I N S I G H T

CogStack . It ’ s a funny name , but when we first started about seven years ago , we were working on health data in hospitals and pulling it together and making it searchable . We did this for the 100k genome project and since then , we have grown into a large-scale data platform for the hospitals to use only our data for improving patient care , patient safety , population health , research and AI . We took a different approach in building CogStack , including Elasticsearch technology , to allow us to search through all the records . This had not been previously possible using conventional systems .

We found and consented the patients to perform the genetic sequencing . We were one of the fastest recruiting organisations because we could find all the relevant patients . As a result , we were widely trumpeted by the Chief Medical Officer back then about the technical scope of the technology .

However , we realised that once you could search for one thing , you could search for anything . Due to this , we started exploring how we could use searches for treating patients for safety events . We have triggers that sit in the background waiting for the safety event . If a patient falls or certain medication is being used in a particular way or certain side effects are being recorded , the trigger will inform relevant clinical teams and goes in to try to prevent or fix the problem .

The pandemic was an example of a situation where a disease without a consistent name , symptom or known symptoms arrived in the hospital where lots of people didn ’ t know who had it . During the pandemic , an unnamed and diverse disease emerged , causing confusion in hospitals . We swiftly created dashboards to track patients , identifying cases , racial disparities and early symptoms like loss of smell . Our approach , akin to Twitter , transformed health records , aiding in effective patient care .

COVID is a great example , but what are the results of unstructured and siloed electronic healthcare records ( EHRs ), particularly within the NHS ?

Patients typically undergo repetitive storytelling in both GP offices and hospitals , encountering redundant queries due to siloed and incompatible data . The technical challenge of unifying data hindered progress until the ability to search unstructured data emerged .

Consolidating information into a single accessible repository revolutionised patient care by expediting relevant data retrieval , minimising unnecessary duplication and facilitating informed responses . Previously , doctors faced impractical expectations of reviewing extensive , unstructured records during brief encounters . The transformation to structured , consolidated data significantly enhances the efficiency and efficacy of inpatient treatment .

For those who aren ’ t working in the NHS , how would you describe the current systems being used by hospitals that aren ’ t using CogStack ?

Many hospital health record systems , designed in the 90s , hinder efficient data handling . Despite sophisticated interfaces , their outdated back-end limits search capabilities . This is analogous to navigating the pre-Google Internet with directories instead of search engines . Current healthcare professionals , like doctors , operate without a search function , grappling with outdated technology . It ’ s concerning that we subject healthcare professionals to such awkward limitations in today ’ s tech landscape .

Can you elaborate on the role of natural language processing and Artificial Intelligence ( AI ) within CogStack ? Is this a new implementation ?

We started using AI at the end of 2019 . We had been using Coxa for a while , and what we learned was that there was just too much information for a human being to read and we had to use natural language processing to help us with the reading to pick up relevant information . In the early pandemic , we used it to automate the reading process and data cleaning process .

AI and Machine Learning is iterative and improves over time . The more we use it , the more we train it , the better it becomes . And we have witnessed it get better and better .

In 2021 , we developed the ‘ Met GPT ’ AI model and published our findings but deemed it less useful due to frequent inaccuracies – 20 – 30 % of the time . Dismissing it , we underestimated its potential , especially with the less memorable name ‘ GPT .’ However , to our surprise , it gained widespread recognition upon its December launch . Previously handling complex tasks like processing vast information internally , GPT now obviates the need to explain natural language processing . Despite initial scepticism , its efficiency in managing extensive data has driven its on-going success .

Do you have any statistics on how it ’ s improved the time efficiency within CogStack from when you weren ’ t using AI to now ?

Hospitals engage in the labour-intensive task of clinical coding , employing individuals to manually transcribe medical records into spreadsheets . However , we ’ ve harnessed

www . intelligenthealth . tech 49