How can medical device integration enhance prediction?
Data are the heart of decision making. The old adage “what cannot be measured cannot be controlled” is apt here. Historically, clinically quantifiable benefits of connected medical devices within the healthcare enterprise have been measured in terms of time-in-motion studies and workflow relating to time saving associated with accomplish a specific task or end goal. These are valid measures. Yet, the question remains as to whether there is something more tangible clinically that can be used as a measure of effectiveness related to medical device integration. The analysis of data made available from these sources is temporal in nature (i.e., time-varying and collected over time), is multidimensional (i.e., is a vector and represents the patient cardiovascular and respiratory system evolution over time), and is objective if collected automatically from the bedside.
Are Data Access is Key to Improved Prediction and Predictive Analytics?
The attached white paper captures some of the references and measures for improvement relative to medical device integration. Prediction of patient clinical outcome has been the subject of much research and many papers over the years. The difference between large scale data mining and predictive analytics in this context is that data from medical devices are multidimensional time series. Hence, temporal trends in behavior as measured by patient state changes over time provide the ability to track how a patient is “evolving” with time.
The figure below is from a presentation (“FDA Regulatory Submission Prototype Use Case”) I gave at the 2nd Annual Medical Connectivity Conference (San Diego). The figure depicts multiple sources of medical device data, from physiologic monitors to mechanical ventilators to infusion pumps and laboratory systems. The data from each of these medical devices are brought forward to an integrator, whereupon the data can be combined, processed, analyzed and then the output of which can be individual indices, measures, or time predictions. These outputs are, in ensemble, the objective measures of the patient, time-based, and comparative. The scope of the analysis is limited only by the needs and imagination of the researcher and the clinical end user.
What types of algorithms can be fed using data from these sources?
- Weaning algorithms (i.e., weaning from post-operative mechanical ventilation)
- Sepsis algorithms (i.e., modified early warning scores combining vital signs, laboratory and visual observations)
- Respiratory sufficiency assessments and ventilator acquired events (e.g.: ARDS, COPD, PCA management, Extubation Criteria, VAP, etc.)
Many other methods and analysis can be performed, as well, to provide predictive assessments of the patient while in-situ in the critical care, medical surgical, or operating room.