The Medical Analytics Group has developed numerous strategies, applications, models, and solutions that are used everyday at Massachusetts General Hospital. Our projects are driven by the needs of doctors and patients; while we use cutting-edge techniques, our sole goal is practical applicability. Below is a selection of our current and previous projects in use throughout the hospital:
A Predictive Engine for Estimating Patient Wait Times and Delays
To enhance the patient experience, we have developed an automated, scalable system to (1) train and optimize predictive models using various machine learning techniques, (2) retrain as necessary in response to new data and shifts in long-term trends, and (3) display estimated wait time and delays for exams in select radiology waiting rooms. A dashboard interface allows hospital administration to examine historical wait time and delay trends as well as the accuracy of the predictive model.
An Information Portal that Identifies Real-time Bottlenecks
Tempus Fugit (Latin for time flies) allows hospital administration to track workflow parameters (exams that fall behind in schedule, patients that have waited an abnormally long time, etc…), to identify and swiftly fix any problems that may arise. Tempus Fugit also automatically contacts appropriate staff members about the most pressing delays in the radiology workflow. This tool is highly utilized by hospital administration, and by one measurement, has decreased the rate of problem cases by 70%.
Strategies for Optimal Scheduling of Patient Appointments
Patient care is always the highest priority at any hospital, but as a result, the time a patient spends in a hospital can be notoriously unpredictable. In addition, facility utilization fluctuates greatly throughout the day as patients arrive at varying times (early or late for an appointment; in walk-in facilities, at random times). This creates bottlenecks, idling, and other operational pitfalls that negatively affect both doctors and patients. To address this challenging problem, we are using advanced mathematical algorithms and machine learning models to design optimal patient processing scheduling strategies, which will reduce stress, decrease waiting times, and allow for more predictable, consistent treatment.
Staffing Radiologists Based on Machine Learning Prediction of Exam Volume
To properly staff radiologists, the division head must have a good idea about how many exams the team will have to read on any given day. Traditionally this was largely based on guesswork, but understaffing can cause massive delays for patients and lead to unnecessary stress among radiologists. To solve this problem, we developed a machine learning algorithm which learns from historical data to estimate the volume of exams on any given date. The application is accurate in 99.9% of days tested, and staffing can now be finely tuned to the precise prediction available.
Analyzing Facilities to Find Operational Bottlenecks
We routinely perform comprehensive facility studies to determine the causes of operational bottlenecks—such as late exams or delayed patients. In-person observations of the workflow as well as thorough mathematical analysis and modeling of historical data are combined to produce a list of inefficiencies and possible solutions to improve exam on-time performance.
A Smartphone App to Efficiently Assign Exams to Available Resources
In facilities where multiple rooms perform the same exams, patients are matched to available rooms based on their arrival and appointment time. We developed eQueue, a smartphone app that uses advanced queuing strategies to automate the matching process and fairly assign patients to technologists. The automatic assignment removes human biases that can lead to unintentional bottlenecks and unnecessarily long wait times.
Medical Analytics Group 