Chronic Respiratory Diseases (CRDs) affects an estimate of over a quarter of the Australian population. CRDs can be present in variations across all age groups such as asthma, allergies, hayfever, and Chronic Obstructive Pulmonary Disease (COPD). These variations are attributable to factors including air quality, weather changes, patient health condition, patient lifestyle, and patient self-prevention and self-protection measures. A hospital in a NSW rural health district wanted a solution to analyse patient data to identify patients at-risk of developing CRDs and help doctors distinguish the factors related to respiratory issues. Development of CRDs in returning patients is a pre-existing problem that doctors struggle to predict, and results in a less efficient use of money and resources. The solution developed by our data superstars would assist doctors to identify patients at-risk of developing a CRDs and the factors that catalyse these.
The hospital wanted an end-to-end AI solution to take as many of the potential factors as possible into consideration, assist doctors to distinguish factors most related to respiratory issues, and help predict re-presentations of CRDs for potential patients. The hospital provided over ten years of patient electronic medical records, which were a sufficient data set to perform most of the predictive analytics involved in the two-part project. The first part was to find out the factors most likely to cause respiratory issues, and second to develop a predictive data model to classify potential patients into two groups – at low risk of readmission and at high risk of readmission, and provide logical reasons for these classifications. Given these circumstances, AI Canberra decided to perform exploratory data analysis for the first part of the project, and use machine learning algorithms on the second part. Our team collaborated with the hospital at each step of the development process through Scrum meetings to receive feedback and revise the product to provide the most precise and customised solution.
AI Canberra has over 20 years’ experience working with data and over 10 years’ experience providing data-related solutions in the health care industry. We are uniquely positioned to ensure you get the most powerful insights out of everyday information using health-focused intelligent agents. It’s just what we do. Our experience demonstrating the ROI of AI in rural health districts makes us the perfect partner to transform your health initiative.
Using the hospital data, a correlation analysis was performed and a regression model was created to explore factors that were most related to the re-presentation of respiratory issues. A machine learning predictive model was developed to analyse patient data including: demographic information; heath, treatment, and admission history; air quality index (AQI); and climate changes to predict re-presentation risk of patients with CRDs, especially with asthma issues under external environmental conditions.
The hospital staff received immediate benefit from the model, which supported real-time decisions on SQL Server on premise with Microsoft SQL Server R Services. The final results were presented with Microsoft Power BI wrapped in exquisite LiveTiles SharePoint pages.
The Project was run as a Proof of Concept in a crawl before you walk approach spanning a total of one-month propeller time. It took one week to understand the client’s needs and identify the most related factors to respiratory issues; two weeks to collect proper data, build predictive models and perform model selection and refinement; and one week to deliver the results.
When test data at the base hospital was first applied to our trained model, the model didn’t provide results confident enough in predicting re-presentation risks within the next week for patients with CRDs, due to a lack of information. The Product evolved in a number of ways over a quick timeframe of a few weeks. The model performance was improved using factors such as: using more specific air quality measures; incorporating data from pharmacies that indicated which medicine had taken progress; and including data from homecare which provided insights on how patients take self-protection actions to prevent disease presentation.