Big Data Transforms ICU Care: Real-Time Risk Prediction and Personalized Clinical Decision Support

Big data refers to collections of data that cannot be captured, managed, and processed using conventional software tools within an acceptable timeframe. It requires the application of new processing paradigms to transform into an information asset with enhanced decision-making power, insight discovery capabilities, and process optimization abilities. Currently, industries across the board are striving to improve their data processing capabilities to realize data value addition through such processing. The healthcare industry is no exception; researchers are actively seeking ways to extract useful insights from the wealth of historical case information. In June this year, The Wall Street Journal published an in-depth report on the efforts of U.S. healthcare institutions in this regard, highlighting their commitment to measuring risk levels in intensive care units at any given point in time.

For patients, the Intensive Care Unit (ICU) may soon become a less daunting place.

Hospital intensive care units (ICUs) are dedicated to the treatment of critically ill patients, with varying rates of treatment success. Researchers are now mining vast amounts of ICU data to identify insights that can help improve patient outcomes.

Studies have shown that more than 5 million patients are admitted to intensive care units (ICUs) annually in the United States, with a mortality rate ranging from 10% to 29%. While some deaths are inevitable, others result from preventable complications caused by device- and treatment-related infections. Other complications include blood clots and delirium induced by excessive sedation and prolonged immobility.

In the past, intensive care units (ICUs) successfully reduced patient risks by implementing relatively basic measures such as checklists. Adherence to patient care checklists has helped prevent ventilator-associated pneumonia caused by prolonged intubation, as well as infections associated with catheters and various tubes.

The Data Era
Currently, some hospitals are testing the application of big data methods. By conducting comparative analyses and screening medical records from multiple sources over many years—including data that might never be incorporated into single-method analyses—they aim to identify previously unknown correlations, thereby uncovering more problem areas and potential solutions.

“Checklist verification can address foreseeable risks, but it is the less predictable adverse events that concern us most,” said Kenneth Sands, Chief Quality Officer and Senior Vice President at Beth Israel Deaconess Medical Center (BIDMC) of Harvard Medical School. The medical center is collaborating with integrated systems scientists from the Massachusetts Institute of Technology (MIT) and human factors experts from APTIMA to form the project team “State of Hazard,” which is dedicated to measuring the risk level in specific intensive care units at any given time.

BIDMC has seven intensive care units (ICUs), including those for surgical and coronary patients. By analyzing data from all ICU patients between 2012 and 2014, the hospital’s project team identified various conditions that increase risk, such as high patient admission volumes, an increased number of critically ill patients, a higher proportion of nurses with less than one year of experience working in the ICUs, and excessive patient-to-nurse ratios in the intensive care units.

“We are using data that was previously unimaginable to predict hazards in the intensive care unit,” said Daniel Talmor, Interim Chair of Anesthesia, Critical Care and Pain Medicine at BIDMC. “For example, people generally do not factor nurses’ level of experience into risk assessments.”

The project team identified 30 hazards during and after the risk status phase, including bleeding, medication errors, cardiac arrest, readmission to the intensive care unit (ICU) after discharge, falls, and communication errors.

The application developed by the team can automatically extract data from specialized electronic medical record (EMR) software and allow doctors and nurses to input additional patient- and ward-specific concerns. It then calculates risk scores using a visual “dashboard” and displays them in real time on monitors and handheld devices used by intensive care unit (ICU) staff.

Dr. Sands said, “At present, we can provide early warnings for patients who are about to enter a critical condition.”

Patricia Folcarelli is a nurse and Senior Director of Patient Safety at the hospital. She stated that intensive care units (ICUs) can adjust staffing, defer elective procedures, or transfer patients from overloaded ICUs to those with lighter burdens, rather than relying solely on a generic checklist that may not be suitable for every patient. Therefore, the team is also developing “situation-specific” personalized checklists focused on the needs of individual patients.

BIDMC’s “Danger State” team is a project funded by the Gordon and Betty Moore Foundation. The foundation has also provided research and development funding to several other teams in similar fields, with the aim of helping hospitals improve the outcomes of critical care. Other funded teams include the Johns Hopkins University School of Medicine in Baltimore, which is working to enhance the efficiency of intensive care units (ICUs) by conducting rapid analysis and diagnosis through comparison of current patient data with historical data. Peter Pronovost, Senior Vice President of the Department of Patient Safety and Quality at Johns Hopkins University School of Medicine, stated that the lack of interoperability among medical devices in ICUs forces doctors and nurses to piece together patient data from various devices, which not only wastes human and material resources but may also put patients at risk.

“We need software that can link medical record data with devices to predict risks that may harm patients, recommend corresponding treatments based on those risks, indicate whether the treatments have been implemented, and then monitor the patient’s condition,” said Dr. Pronovost.

Establish Connection
Johns Hopkins University has assigned an expert from the Applied Physics Laboratory to help design a system called “Project Emerge,” which acquires data from electronic records and bedside sensors to determine whether individual patients should receive treatment and to guide therapies for preventing and managing related complications.

Alan Ravitz is the engineering lead for the project. He explained that the project team first conducted thorough consultations with intensive care unit (ICU) staff to identify the information they most critically needed and determine how such information should be displayed. Nurse Rhonda Wyskiel recommended designing the interface in a dial-like format, which became the foundational design for both the “Hazard Monitor” in the ward and the tablet devices used by nursing staff. The display alerts healthcare providers when specific patient care interventions are due, proving particularly valuable when care is not delivered on schedule. Its interface features a seven-color dial, where red indicates imminent danger, yellow signifies emerging issues, and green denotes completed tasks.

One of the sensors installed in the monitor continuously tracks the angle of the patient’s hospital bed. Studies have shown that maintaining the bed at a 30-degree angle in the intensive care unit (ICU) can prevent ventilator-associated pneumonia (VAP). Therefore, if a nurse forgets to return the bed to the 30-degree position after adjusting it, the corresponding section on the monitor will turn red. Additionally, other sensors in the monitor include a device that can be attached to an IV pole to measure the distance the patient walks.

Compiled by Chen Xin
Editor: Mo Renying