A decade ago, a major challenge in healthcare was the capture and storage of patient data within the central repository known as the electronic health record (EHR). Today, with many hospital systems regularly collecting data electronically, the challenge has become making sense of these data. The field of analytics has expanded and found a home in healthcare through the use of data for both clinical and population health informatics.
One branch of analytics – predictive analytics—seeks to anticipate future behavior by studying the past and taking into account features and behaviors of patients and health systems through algorithms. These algorithms seek to apply statistics and machine learning techniques to generate actionable insights from massive data sets and facilitate real-time clinical decision making.
Predictive analytics is common in other industries, including retailers, digital technology service companies and professional sports franchises.
Healthcare is a natural fit for the next predictive analytics revolution. The typical hospital is swimming in vast oceans of data—continuous data from bedside monitors, historical values in EHRs and thousands of data points collected during face-to-face clinical encounters.
Fortunately, many hospitals already possess most of the ingredients necessary to anticipate and prevent patient health risks, make better informed clinical decisions, and more readily identify and close care and workflow gaps.
Putting it to use is the real challenge.
The ‘Unseen’ Data Problem
Is the premature baby girl in the NICU vulnerable to sepsis? Is the adult male recovering from surgery experiencing respiratory depression due to undiagnosed central sleep apnea?
While both of these patients would be continuously monitored by physiologic devices at the bedside, the overwhelming majority of the data collected—information that can be used to predict serious illness and prompt life-saving treatment before symptoms ever appear—are never be used.
Episodic vital signs collection, or spot vitals checks, by clinical staff do not result in a continuous stream of data that can be consumed by the EHR or other clinical repository. Episodic data collection can result in data gaps that do not capture key vital signs activity – such as periods of apnea, low hemoglobin oxygenation levels, or periods of bradycardia or tachycardia – that take place in the span of seconds to minutes with some patients. Hence, when data collection are not continuous, events such as these are unknown and unknowable to the clinician unless observed directly.
The inability to anticipate—or even identify—health risks is not just a patient safety problem. Rescuing patients is also costly in terms of resource utilization, morbidity and mortality, and downstream effects related to patients who require emergency transfers to intensive care units can impact other patients in terms of space availability or even cause the cancellation or postponement of procedures due to lack of available space.
The use of continuous electronic monitoring to facilitate predictive analytics—thereby detecting patterns not readily visible through intermittent spot checks—offers clinicians a quantitative estimate of whether a patient’s condition is going to get worse over time.
Key Elements of Continuous Clinical Surveillance and Predictive Analytics
Supporting predictive analytics requires facilities with computer and monitor networks, a centralized data center, the computer capacity to run the algorithms and the ability to distribute the data via displays or mobile devices.
While some net-new technology investments are not out of the question, hospitals with critical care units or ICUs already have a continuous monitoring infrastructure in place. Optimizing that infrastructure’s capabilities and incorporating it into existing clinical workflows is the real “heavy lift.”
Continuous monitoring from multiple data sources—EKGs, vital signs, laboratory tests—will yield better predictive models than data from a single source. However, one of the risks of swimming in that proverbial ocean of data is that it’s easy to drown. One of the objectives of analytics is to seek interrelationships among seemingly unrelated measurements and sources of data to determine whether these interrelationships can yield the detection of the onset of an adverse event that would not normally be visible by observing a single parameter or multiple parameters individually. These interrelationships, which may be revealed through correlation analysis, can lead to multivariate rules that capitalize on the trended behavior of several parameters that herald the onset of a decline in patient state leading towards adverse events or death. This multivariate data collection comprises the overall concept of surveillance, in which the state of the patient is monitored through the objective (and subjective) observations made over time.
For example, most medical device integration (MDI) solutions gather and filter data to support documentation in an EHR. To achieve real-time clinical surveillance, a more clinically significant capability, MDI should be able to collect data from multiple sources and at the frequencies offered by the source data.
Continuous clinical surveillance solutions that analyze real-time patient data can identify clinically relevant time-based (or temporal) trends, sustained conditions, reoccurrences and combinatorial indications that establish the trajectory of the patient state towards an adverse even prior to the violation of the limit threshold of any individual parameter.
Data collection and analysis are further enhanced when including methods for disseminating, analyzing, and distributing these data. These features facilitate better patient care management and clinical workflow by allowing patients to be monitored remotely.
Continuous electronic monitoring can be leveraged to prevent serious deterioration in patients at risk of sepsis or respiratory depression in both high acuity and general care setting. Patients in intensive care units are already continuously monitored. However, many patients who experience adverse events, such as respiratory depression, do not follow any simple criteria for determining whether they will be at risk for obstructive or central sleep apnea. Hence, the safe alternative is to monitor everyone continuously, even on the general care floor.
Combining analysis with real-time data at the point of collection creates a powerful tool for prediction and clinical decision support.