Inspiration
Length of stay is a key indicator of hospital quality CDC, and being able to plan around disruptions such as sick days is disproportionately important to people in low income brackets who may already be struggling to make ends meet. We decided that we wanted to focus on developing an algorithm that could predict length of stay for given patient conditions, giving healthcare providers a means to compare their services versus the national average as well as providing an easily accessible resource for people who need to carefully budget time and money for hospital stays due to issues of e.g. childcare and job security.
What it does
We created an algorithm using the publicly available summary of the National Hospital Discharge Survey to give a 95% confidence interval for the length of a stay for a patient based on their condition, age, and sex. We deployed the resulting estimator on a freely accessible website for hospitals, patients, doctors, etc.
This accurate prediction for length of stay can then be used from the patient point of view as a metric for measuring hospital quality and for forecasting their sick day needs. For physicians and nurses to better allocate resources and inpatient rooms. For hospitals and insurance companies to reduce costs, increase efficiency, which will be inevitably useful when facing an increasing patient population.
How I built it
Data were taken from the National Hospital Discharge Survey Summary, which gave mean stay length with standard errors for various combinations of basic demographics. By combining patient-provided information weighted inversely to the uncertainty of the NHDS's estimate for each of those factors, we can provide the top and bottom of a 95% confidence interval for a given person's stay. Additionally, patients can freely choose to withhold information, and our site will simply use population averages.
Challenges I ran into
The biggest challenge was finding HIPPA compliant data that was also free and accessible for hackers (like us!). Developing an formula for estimating stay range was also tricky since our data came as mean & SEM rather than standard deviation. Finally, we spent a lot of time debugging the site itself, including issues with making sure hosting and UI ran smoothly.
Accomplishments that I'm proud of
Makin' it through the weekend.
What's next for Sick Day Forecast
Since this is a hackathon, we have only developed the MVP for this project and have a lot of room to grow. Here are some specific areas that we would like to focus on: Adding in additional factors for more accurate length of stay prediction Adding financial status into the model Adding more obscure illnesses into the model
Acknowledgements
The team: Oliver Stanley, Hamza Sheikh, Charlene Yinei, Claudia Zhu. We would like to thank Chris Truong and Dan Selver for helping us debug. Spiders forever :))
Log in or sign up for Devpost to join the conversation.