The Joint Commission

Machine learning revolutionizes The Joint Commission

Picture a healthcare surveyor navigating the complex terrain of a healthcare institution, meticulously observing every aspect of patient care. It’s a task that demands precision and attention to detail. Whether related to patient safety or healthcare practices, each observation holds immense significance.

However, the real challenge lies in translating these observations into actionable data, aligning them with myriad standards and Elements of Performance (EPs) that govern the healthcare industry. With thousands of standards in play, this task was nothing short of daunting.

All US health institutions must be certified as “fit to operate.” This certification mandates that healthcare organizations meet compliance requirements while providing optimal service. The Joint Commission is one of the few organizations to issue “fit to operate” certifications, and the company guarantees compliance through inspection. These inspections cover a range of public standards, regulations, requirements, and procedures.

Recognizing the need for transformation, The Joint Commission embarked on a remarkable journey in collaboration with SoftwareOne (formerly Crayon Consulting), a company deeply entrenched in the Azure cloud ecosystem. The goal? To harness the transformative power of machine learning and artificial intelligence to streamline the process of survey findings mapping.

Creating a robust Azure foundation

The decision to leverage Microsoft’s Azure platform was strategic, as it provided the robust foundation required for this ambitious project. Combining Azure DevOps, Azure Data Factory, Azure ML, and Azure Kubernetes Service SoftwareOne constructed the architecture for this groundbreaking system. This ensured Machine Learning Operation (MLOps) processes were in place to support the deployment, monitoring, updating, and retraining of models. All model-building, training, and heavy pre-processing are executed on Azure Machine Learning (AML) using flexible compute clusters with a Graphics Processing Unit (GPU).

The introduction of a ‘search-as-you-type’ mechanism was a pivotal moment in this journey. Surveyors were now equipped with a tool that could instantly suggest the most relevant EPs as they typed their observations. This innovation reduced the time spent on navigating standards and allowed surveyors to focus more on the essence of their surveys, which was ensuring the highest quality healthcare delivery.

Intelligent machine learning: a new approach

Developing an intelligent machine learning solution that could address the unique challenges faced by The Joint Commission was no small feat. The heart of this challenge lay in the ever-evolving nature of EPs, which underwent revisions every six months. A traditional classification-based approach simply wouldn’t suffice.

The visionary team at The Joint Commission approached the problem from a different angle, reframing it as a Question and Answer challenge. They likened each observation made during a survey to a ‘question,’ and the corresponding EP became the ‘answer.’ The task at hand was to establish a connection between these two entities using vector space representations.

The Universal Sentence Encoder (USE) from Google emerged as the tool of choice for this task. By transforming textual findings and EPs into vectors or embeddings, the team was able to create a foundation for matching them accurately. Training a transformation head to align these embeddings further refined the system.

The technology stack was robust, with Python and the PyTorch framework at its core. Rigorous experimentation took place on Data Science Virtual Machines within the Azure environment. Once the model reached a state of perfection, Azure ML pipelines were deployed for its training and deployment.

Improved consistency, efficiency savings and streamlined process

The impact of this machine-learning solution was profound. The Joint Commission witnessed a substantial improvement in the consistency of findings, a pivotal factor in ensuring the highest standards of healthcare. Surveyors could now perform their duties more efficiently, and the process of assigning standards and EPs had never been more streamlined.

Kin Lee, TJC Chief Digital and Information Officer, a key figure behind this transformative project, emphasized, “The ‘Machine Learning for Survey Consistency initiative’ leverages our rich data in improving survey-finding consistency and operational efficiency. This project and associated learnings set the stage for many other opportunities to further leverage data and technologies in advancing The Joint Commission’s mission to improve healthcare quality and reduce patient harms.”

In an era where technology and healthcare intersect, The Joint Commission’s initiative stands as a testament to the incredible potential of machine learning in bridging gaps and ensuring that every individual receives the best possible care.

Unlocking healthcare excellence through technology, innovation, and partnership

The collaboration between The Joint Commission and SoftwareOne is a testament to the power of technology, innovation, and partnership in the pursuit of excellence in healthcare. It showcases how a forward-thinking approach to AI and machine learning can reshape an industry that is fundamental to the well-being of every individual. This project serves as an inspiring example for healthcare organizations worldwide, highlighting the transformative impact of technology on the journey to provide safer, higher-quality healthcare.