Inspiration
Aura stemmed from a desire to shift hospital operations from a reactive to a proactive model. We saw that hospitals have access to vast amounts of patient data, but it's often used to look backward at what has already happened. We were driven by the question: What if we could use this data to look forward? The idea of empowering healthcare providers with the foresight to anticipate patient needs, optimize resource allocation, and ultimately improve patient flow and care was the core motivation behind this project.
What it does
Aura - Predictive Patient Pathways is an interactive Tableau dashboard that forecasts a patient's likely length of stay from the moment they are admitted. By analyzing key patient data points such as age, severity of illness, and pre-existing health conditions, Aura calculates a "Long Stay Risk Score." This score provides hospital administrators and staff with an at-a-glance understanding of which patients may require more resources, allowing for better-informed decisions on bed management, staffing, and care coordination.
How we built it
Our process was grounded in a data-first approach.
- Data Preparation: We began by cleaning and transforming the raw 'Nicolybgs/healthcare_data' file obtained from https://huggingface.co/datasets/Nicolybgs/healthcare_data.
- Exploratory Analysis: Using the cleaned data, we moved into Tableau to identify the key drivers of patient stay duration. We built several worksheets to visualize the relationships between variables like age, department, and illness severity against the length of stay.
- Predictive Modeling: The insights from our analysis fueled the creation of our primary feature: the "Long Stay Risk Score." We engineered this using a weighted formula in a Tableau Calculated Field, which became the predictive engine of our dashboard.
- Dashboard Design: Finally, we brought all the elements together into a single, interactive dashboard. We designed it to be an intuitive and actionable command center. Users can filter by department to see risk scores change in real-time, while specialized graphs automatically highlight departments and associated doctors with a "Long Stay Risk Score" of 5 or higher, flagging them for proactive intervention. Additionally, the dashboard generates a priority list of patients who are in low-resource wards and have a high-risk score of 10 or more, enabling swift and data-driven decisions for ward resource reallocation.
Challenges we ran into
We encountered a few significant challenges on our journey. Firstly, navigating the tools themselves presented a learning curve. Our initial vision included a "data action" feature to trigger workflows directly from the dashboard, but we discovered this wasn't readily achievable and had to pivot our strategy to focus on creating an interactive dashboard instead.
Secondly, we faced a data relevance challenge. While we had a synthetic dataset, it was difficult to integrate into our primary predictive goal because it lacked the key admit/discharge dates column. This made it hard to generate synthetic data that was truly useful for augmenting our model. It taught us a valuable lesson: having more data is only beneficial if it's the right data, and that generating truly representative synthetic data is a complex task in itself.
Accomplishments that we're proud of
We are proud of successfully translating a limited dataset into a simple, actionable, and predictive tool. Creating a reliable "Long Stay Risk Score" using only Tableau Next's native features was a major accomplishment that demonstrated the power of creative data engineering.
What we learned
This project reinforced the classic lesson that data preparation is 80% of the work. The quality of our predictive model was entirely dependent on the quality of our initial data preparation. We also learned the importance of adaptability when our initial technical plans hit a roadblock. Most significantly, we learned that you don't need a full-scale machine learning pipeline to generate powerful predictive insights; the right visualizations and a well-crafted analytical approach can unlock immense value from data.
What's next for Aura - Predictive Patient Pathways
The future for Aura is focused on expanding its intelligence and integration.
- Algorithm Refinement: We plan to incorporate more variables into our risk score to test and refine the algorithm's accuracy.
- Prescriptive Analytics: We want to move beyond prediction to prescription. The goal is for Aura to not only identify a high-risk patient but also to recommend specific actions, such as "Consider specialist consultation" or "Allocate to high-resource ward."
- Broader Integration: While we faced challenges, we aim to explore deeper integration with other tools to make Aura an even more seamless part of the clinical workflow.
Log in or sign up for Devpost to join the conversation.