Innovating Antibiotic Stewardship: A Collaborative Journey
At Enolink, our recent collaboration with Seoul National University Bundang Hospital (SNUBH) and Seoul National University (SNU) represents a significant step in our quest to innovate Antimicrobial Stewardship Programs (ASP). This joint effort, blending SNUBH’s medical expertise with our data science expertise, aimed to create a state-of-the-art ASP intervention recommendation system.
Through Enobase, our secure data platform, SNUBH provided access to their unique datasets from various hospital systems, offering invaluable opportunities for our research. The Enobase platform enabled us to safely and efficiently harness this wealth of information, facilitating collaborative analysis. The initial phase of data exploration was more than gathering numbers; it was about understanding the story behind each dataset. We conducted thorough exploratory analyses to validate the quality and relevance of this data.
The Enolink Data Science team, in a series of iterative steps, transformed raw data into meaningful insights. We trained models using various machine learning techniques, ensuring that the output resonated with clinical reasoning.
A key component of our collaborative success was the regular meetings with the experts from SNUBH. These meetings were crucial for sharing progress, refining our models, and incorporating a layer of clinical intuition. The importance of these regular interactions cannot be overstated; they facilitated a deep, mutual understanding and joint decision-making, which was essential in guiding our project smoothly from one stage to the next. This continuous exchange of ideas and expertise was fundamental in weaving data into a coherent and impactful story.
Challenges of Label Outcome Generation
In our early research stages, we initially tried to define outcome variables using intervention data from ASP pharmacists. However, this approach hit a roadblock: the records were incomplete, and labeling data not reviewed by pharmacists proved to be a significant challenge. This highlighted the limitations of relying solely on existing intervention records.
Confronted with this, we shifted our strategy. Instead of relying on unreviewed intervention data, we defined outcome variables like de-escalation, utilizing existing prescription data. This task was intricate, involving the precise definition of intervention items as outcome variables from real clinical data. A key aspect was processing and defining extensive test and vital sign data to accurately reflect patients' clinical conditions.
This phase of the project was challenging but also enlightening. Enolink's exploratory data analysis and the invaluable consultations from SNUBH’s Infectious Disease Department played a crucial role in overcoming these hurdles. Their insights helped us to navigate through the complexities of clinical data, ensuring that our models were not only technically sound but also clinically relevant.
Enobase: The Catalyst of Our Workflow
Enobase was not just a tool; it was our companion in transforming data into solutions. By standardizing data into a ‘clinical dataset,’ we could focus more on innovation and less on the intricacies of data handling. This standardization, coupled with Enobase’s modular design, allowed us to navigate through the complexities of feature generation and model building with ease.
The Future of ASP: Beyond the Boundaries of a Single Hospital
Dr. Ju-Yeun Lee, the lead researcher on ASP at SNUBH, underscores the need for real-world application: “We are excited about the potential of applying our findings in clinical settings. A computer program integrated with the hospital's EMR system could transform the way we approach antibiotic stewardship.”
Our vision extends beyond SNUBH. We are exploring the prospects of employing federated learning to build robust, privacy-preserving models using data from multiple hospitals. This approach is not just about technology; it’s about adapting to the diverse realities of healthcare settings worldwide.
Our journey with SNUBH is just the beginning. As we continue to explore new frontiers in healthcare technology, we remain committed to creating solutions that are not only technically proficient but also deeply rooted in clinical reality.