Paul Roscoe is the Chief Executive Officer of CLEW Medical.
Prior to joining Clew, Mr Roscoe was CEO of Trinda Health, and was responsible for establishing the company as the industry leader in quality oriented clinical documentation solutions.
CLEW Medical offers hospitals, healthcare systems and intensive care units advanced clinical intelligence and patient diagnostics using AI-powered, FDA-cleared predictive analytics and proprietary critical care models.
Could you start by telling us a bit more about CLEW Medical’s AI-enabled platform and its unique capabilities in the MedTech industry?
CLEW’s founding was based on the premise that data analytics and AI can significantly improve patient outcomes and clinician experience in high-acuity care settings. The clinical surveillance platform we’ve built is the first to have FDA-cleared AI-driven prediction models for critical care. Our system obtains data by integrating with all clinical data sources within a hospital and builds a near real-time physiological profile of each patient to continuously monitor their status. It then uses this data to provide predictive insights to identify patients who will likely have an adverse event – such as respiratory failure – and alert clinicians to intervene up to eight hours before the anticipated event. The platform’s high degree of accuracy also reduces the excessive number of false alarms, enabling clinicians to practice at the top of their license and focus on patients most in need of immediate intervention.
What were the key factors that contributed to the FDA clearance of CLEW’s AI-driven predictive models?
CLEW has embraced AI since its inception. Our founders and developmental leaders recognized the significance of fostering trust with caregivers, the individuals responsible for utilizing our technology to care for their most vulnerable patients. It was imperative that our technology undergo the same level of scrutiny and diligence in design, development, testing, and validation as the devices already in use by our users. To encourage the adoption of an AI solution for critical care settings, our team understood the necessity of building models with meticulous product development and quality systems. As a result, our AI model development leverages robust MLOPS (machine learning operations) infrastructure to meet regulatory expectations, such as the PCCP (pre-authorized change control plan) guidance from the FDA. Our AI models are methodically designed, while undergoing all necessary experiments for medical device regulatory clearance.
The robustness of the models and our internal processes resulted in the FDA classifying our solution as a class II medical device in early 2021, which exemplified a landmark, first-of-its-kind achievement. FDA medical device clearance serves as a testament to the quality of our end-to-end development process, which includes clinical validation studies conducted in real patient populations.
The recent study published in CHEST® Journal highlighted the predictive accuracy of your AI models. Can you discuss the methodology and the specific findings of this study?
A CLEW-trained ML algorithm was deployed in 14 intensive care units (ICUs) across two major health systems to predict intubation and vasopressor initiation events – in other words, events that require life-saving intervention – among critically ill adult patients. Its performance was measured against existing bedside monitoring alarms and the predictive effectiveness of telemedicine system alerts.
The study, designed to evaluate the tool’s accuracy and utility of alerts in ICUs, found that CLEW’s models for predicting patient deterioration were five times more accurate than and produced 50 times fewer alarms than the leading telemedicine system. The findings also show that the ML model has superior accuracy compared to traditional monitoring systems and drastically reduces unnecessary interruptions to clinician workflows.
How do the AI predictions made by CLEW’s platform potentially transform care delivery in the ICU? Could you elaborate on how these predictions improve outcomes and reduce complications?
CLEW’s platform produces opportunities for early interventions in high-risk patients and supports capacity management by identifying low-risk individuals who may be ready for step-down or discharge. This, in turn, decreases mortality and readmission rates, reduces complications caused by patient deterioration, and minimizes patients’ length of stay.
For example, within the first 24 hours of deployment at a major health system, our technology predicted hemodynamic instability in an ICU patient, which triggered a provider evaluation. Upon evaluating the patient, the provider ordered a CT scan and detected an abdominal bleed. The patient was rushed to the operating room for emergency surgery, infused with fluids and blood, and their life was ultimately saved. 24 hours later the patient was in stable condition.
Your system was found to be five times more accurate than a leading telemedicine monitoring system. What makes CLEW’s technology more effective in predicting critical patient deteriorations?
In general, ML-generated notifications are less frequent, have higher levels of accuracy and lower rates of errors such as false positives, and create longer pre-event lead times than other telemedicine system alerts and bedside monitoring system alarms. CLEW’s alerts are more accurate and functional and provide time for the care team to adopt countermeasures to prevent predicted outcomes. The sophisticated intelligence that CLEW provides is made possible by its ability to mine patient data from a health system’s electronic medical record (EMR), combined with ML models that have been rigorously tested and validated through peer-reviewed research and FDA clearance.
The study also noted a significant reduction in false alarms. How does reducing alarm fatigue benefit ICU staff, and what has been the feedback from healthcare professionals using your system?
98% of bedside monitoring notifications are false positives, leading to alarm fatigue and exacerbating historically high levels of clinician burnout. CLEW addresses alarm fatigue by reducing the number of auditory interruptions, increasing the percentage of actionable notifications for necessary provider intervention, and creating an overall calmer ICU environment. In essence, the platform’s accuracy and ability to reduce unnecessary workload via advanced ML models significantly improves ICU burnout. As part of the implementation process, CLEW’s customer success teams focus on the importance of clinical change management to ensure the technology is appropriately incorporated into the overall clinical decision-making process. The feedback from clinicians has been extremely positive.
How does the early notification feature of CLEW’s platform work, and what kind of interventions has it facilitated in real-world ICU settings?
Based on the incoming stream of information from bedside monitoring and life-support devices, as well as from the Electronic Health Record (EHR), the CLEW AI models can make predictions about the risk of patient deterioration and death over the next eight hours. With these predictive assessments, experienced clinicians can evaluate patients more closely and determine if there are applicable countermeasures to prevent the predicted deteriorations, instead of responding to them on an emergency basis.
For example, the CLEW platform can notify clinicians that a patient is highly likely to enter respiratory failure, which typically leads to intubation and mechanical ventilation. Upon receiving the alert, caregivers can then identify the patient has an excess of fluid that could start backing up into the lungs, and initiate diuretic therapy to reduce the fluids, thus preventing an intubation later. Our model can also anticipate whether a post-surgical patient is likely to become hemodynamically unstable and require vasoactive medication support. Armed with this knowledge in the absence of obvious symptoms, a CT-scan determined the patient had internal bleeding and was taken back to surgery to repair it. Ultimately, this intervention resulted in the patient being stabilized.
CLEW’s AI-enabled predictions also support hospitals with capacity management needs. Some patients will no longer require critical care and can be transferred to lower-acuity care units, freeing up beds to manage more critically ill patients. This allows the health system to improve capacity management and create access for more patients. This also increases contribution margin for the health system.
What are the next steps for CLEW Medical in terms of further developing and expanding the use of your AI-driven models in different healthcare settings?
We have already expanded the CLEW platform outside of critical care settings to include step-down units and emergency departments, and we are currently in the process of expanding across the remaining acute care beds of hospitals, including post-anesthesia care units (PACU) and general medical/surgical & specialty beds. The eventual ubiquity of inexpensive wearable monitors providing frequent vital signs information, along with our PCCP clearance, enables CLEW to expand its AI surveillance capabilities more broadly throughout acute care hospitals.
Additionally, as CLEW predictions are complementary to many other HIT systems including the EHR, we are working on delivering our insights via integration into a health system’s existing toolkit. We have joined the Epic developers’ network and have demonstrated successful integration of advanced CLEW capabilities such as AI-driven predictions into the clinical user experience.
CLEW is also embarking on a novel, AI-driven approach to sepsis management, a devastating and sometimes deadly complication.
Where do you see the future of AI in improving ICU care over the next decade, and how does CLEW plan to be a part of this future?
Hospital patient populations are sicker than they used to be. With increasing age and lifestyle-related chronic illnesses alongside widespread caregiver shortages, the need for intelligent clinical surveillance continues to grow. Since many patients end up in ICUs because of missed opportunities to intervene earlier in the care process, CLEW is not only focused on using its AI to improve ICU care, but also on partnering with health system and industry innovators to improve all acute care. Our programmatic pipeline for AI development (MLOPS) will harness partner capabilities to grow FDA-cleared AI models beyond what CLEW develops on its own.
However, technology is only a part of solution. The use of AI in healthcare is not about replacing caregivers. In fact, AI can offer superior information to support their decision making to provide optimal clinical care, such as reducing noisy alerts that waste their time. CLEW is working with health systems and partners to learn from and educate caregivers on how AI tools can be effectively adopted and accepted into clinical practice. Research that validates the accuracy and efficacy of AI is required, so CLEW works with its customers to generate this proof with their own patient populations. This focused research effort supports implementation and adoption by bedside caregivers who would otherwise be skeptical.
To expedite new clinical implementations, we have the ability to update our platform to include newly discovered best practices within a month, something that typically takes years. Over the next decade, CLEW will be at the forefront of working with health systems to make effective clinical AI the informed and prescient partner of the human caregivers who may someday care for us or our loved ones.
Thank you for the great interview, readers who wish to learn more should visit CLEW Medical.
