Hospital -acquired infections (HAI) affect approximately 3.5 million people in the United States (10% of hospital admittees) and are strongly linked to significant mortality and hospital expenses (Inweregbu, 2005). In fact, patients exposed to such infections can have a hospital stay that is 4x longer than that of a patient without an HAI. One such infection is caused by clostridium difficile bacteria, which can be spread amongst inpatients through contact with surfaces. In addition, when patients come in contact with HAIs, the hospital is assessed an additional charge for readmission of the patient. Therefore, by producing an accurate report of hospital locations that are most prone to be loci for spreading HAIs, hospitals can not only reduce the extraneous costs they incur, but also save patients the agony of prolonged, complicated illness.

What it does

Leveraging electronic health record (EHR) data to identify potential origins of HAIs, we prototyped an interactive dashboard that pinpoints specific locations that may be responsible for person-to-surface-to-person infection transmission. Clear visualizations and records of this data allow healthcare providers to trace the source of an in-house infectious outbreak and take actions to prevent further transmission, or analyze what changes were made that initiated such conditions. Our probability-based approach to the prediction of future infections also enables prospective action to be taken, based on the history.

How we built it

Because of the nature of the existing dataset (see below), we simulated patient locations during the actual stay to create proof-of-concept visualizations. An underlying assumption of this model is that if a patient visits a room that was previously visited by an infected patient within a 24-hour time window, then they are tagged as "exposed," enabling a traceable pathway. Based on data collected over the course of a month for each room, we're able to draw conclusions about risk associated with entering the room and contracting the infection of interest.

We were able to implement the data visualization through a simple histogram which was coded up using d3.js on an online platform called CodePen. In the future, we hope to expand upon this system to make it a fully functional dynamic program which takes in real-time data from the hospital and converts it into a series of histograms.

Challenges we ran into

Limited time and resources meant that it was difficult to find a medical dataset with the exact locations of patients throughout their stay. Instead, we had to mimic what would normally be found in EHR databases using a model. To do this, we generated a set of patients and all the locations they visited, and used a random number generator to determine whether they were exposed to an infection, and whether they went on to develop an infection.

Accomplishments that we're proud of

We're tackling an issue that is often discounted when it comes to the grand scheme of disease propagation within the healthcare industry. While many hospitals have antimicrobial stewardship programs, they still grapple with issues of HAI. Our solution takes a unique approach to a multibillion dollar market by offering an intuitive tool for a complex problem by leveraging data that already exists within hospitals.

What we learned

Interestingly, HAIs are considered on a floor-by-floor scale in hospitals, rather than on a building-wide basis, which makes it difficult to draw conclusions about infection transmission across different departments and diagnosis services (e.g. radiology, X-ray, operating room, intensive care unit, etc). There are also no existing clinical informatics approaches that are currently being leveraged by companies to ameliorate hospital acquired infections. In fact, current strategy involves calling in a pair of epidemiologists that sift through existing EHR data to trace the source and provide recommendations on a case by case basis. Automating this process and moving towards constant surveillance of HAI risk is an important step towards preventing unnecessary healthcare costs.

What's next for HTraxx+

Step 1: Application Precision Enhancement Our current implementation of HTraxx+ does not stratify for confounding variables such as patient age and existence of immunocompromising conditions or ongoing treatments. Doing so would allow for an even more robust model that provides more precise insights for doctors and hospital administration looking to conduct source control for HAIs.

Step 2: Product Diversification Beyond clostridium difficile infections, there's potential to target multiple diseases, including but not limited to: central line associated bloodstream infections, urinal tract infections, and MRSA. We can doing this by querying the EHR API for other diagnoses to scale this product in a very straightforward manner.

Step 3: New Potential Domestic & Global Markets Because of the location-based probability approach of our tool, we can expand sales to urgent care, nursing homes, doctors' offices, and other types of healthcare facilities as well.

Share this project: