This implementation occurred within Houston Methodist (HM), an 8-hospital health system in Texas encompassing the HM Hospital (a quaternary care academic flagship), 6 community hospitals, and 1 long-term acute care (LTAC) facility serving the greater Houston area, together serving a diverse adult population of more than 140,000 annual admissions and 2 million outpatient visits [1].

During the COVID-19 pandemic, HM established a centralized virtual operations center (VOC) and virtual intensive care unit (VICU) providing 24×7 monitoring for approximately 370 intensive care unit (ICU) beds. This infrastructure created favorable conditions for remote patient monitoring (RPM) and ultimately informed the development of the system-wide continuous vital sign (VS) monitoring (CVSM) program. This CVSM program differs from traditional RPM in its continuous, hospital-based, and fully integrated design, with data integrated into the electronic medical record and supported by 24×7 centralized oversight rather than periodic outpatient review.

In 2022, the health system launched a redesign of inpatient VS practices, with this report detailing the program’s scale-up and deployment, which were designed from the outset for system-wide implementation.

Frequent monitoring of VSs—heart rate (HR), respiratory rate (RR), temperature, blood pressure, and oxygen saturation (SpO₂)—is a cornerstone of inpatient safety [2]. On general hospital floors, VSs are routinely collected manually every 4 hours, reflecting a century-old practice that does not account for patient acuity [3,4]. This approach risks missed deterioration between checks, places substantial time demands on nursing staff, and disrupts patients' sleep, with physiological consequences [5-14]. Deteriorating VSs are well-established predictors of ICU transfer and in-hospital mortality, and CVSM, enabled by wearable RPM devices, now allows for unobtrusive, high-frequency VS measurement [7,11,12]. This work directly addresses health system challenges of limited access to timely clinical data and variable adherence to evidence-based monitoring practices, both of which undermine patient safety and workflow efficiency. However, evidence on broad, system-wide implementation across diverse hospital settings remains limited.

Previous CVSM implementations have typically been limited to ICUs, single hospitals, or narrow patient groups, demonstrating feasibility but limiting generalizability [15-17]. Our initiative differs by extending CVSM across all adult non-ICU inpatients in an 8-hospital system, integrated with Epic and supported by centralized monitoring. This report describes the implementation and scale-up process.

This report follows JMIR’s implementation reporting guidelines for digital health implementation [18] to ensure transparency in describing the implementation context and processes.

The aim of this initiative was to redesign inpatient VS measurement practices to enhance patient safety, patient experience, and staff efficiency through the implementation of CVSM. Specific objectives were to enable early detection of deterioration, reduce nighttime disruptions to promote patient rest and recovery, and improve workflow by decreasing reliance on manual VS checks. Implementation outcomes included system-wide rollout to all eligible beds, 100% nursing staff training completion, optimal device use, and reductions in manual VS collection, with qualitative observations of staff and patient acceptance. Operational metrics (scale-up timeline, device use, alert filtering, and training completion) were directly measured, whereas workflow impacts and staff acceptance were reflected in qualitative, staff-reported estimates.

The implementation strategy was designed for system-wide scalability and organized into 4 phases, with the observed timeline shown in Figure 1.

The Institutional Review Board of Houston Methodist Research Institute reviewed this initiative and deemed it not human research, as it represented a service redesign implemented as part of standard care.

Between April 2023 and February 2024, the program expanded from pilot units to enterprise-wide coverage across all 8 hospitals, encompassing the full adult, non-ICU inpatient population (approximately 2700 beds across the academic flagship, 6 community hospitals, and 1 LTAC facility).

The enterprise rollout followed a phased road map. Pilot implementation began on 4 units at the main hospital (2 internal medicine units, 1 neurosurgery unit, and 1 neurology unit). Four months later, deployment scaled through staggered 8-week waves of 4 to 8 units, with 3-week offsets to permit overlap between staff training and transitions, allowing the incorporation of iterative learning. This cadence enabled continuous optimization while maintaining operational continuity.

Five months after the pilot launch, HM introduced a standardized escalation protocol embedded in centralized monitoring workflows. When BioButton device–triggered alerts appeared on the VOC dashboard, a VOC technician performed a first-line review to add clinical content and either dismissed nonactionable events or escalated alerts to a VICU nurse. Approximately half of the alerts were dismissed at the VOC step, reducing alert burden for frontline staff. Illustrative case examples are provided in Multimedia Appendix 4.

The VICU nurse notified the bedside nurse or activated the clinical emergency response team when intervention was indicated. This tiered model concentrated clinical escalation on higher-value alerts. Data on clinical emergency response team activations during the study period have been reported previously in a related outcomes study [24].

Primary outcomes included system-wide scale-up, reliable Epic integration, and sustained monitoring. Device use exceeded 95%, training completion reached 100%, and approximately 50% of alerts were resolved at VOC review. Secondary outcomes included improved alert specificity, fewer patient refusals, increased staff acceptance, and operational efficiencies (nursing and PCA time savings). Alert thresholds were systematically reviewed with virtual medicine leadership in collaboration with frontline nursing staff, superusers, and the vendor. Baseline thresholds (HR 50-115 beats per minute, RR 11-28 breaths per minute, and skin temperature >36 °C) were operationalized with alerts stratified as single alerts (1 abnormal VS), double alerts (2 abnormalities, typically HR and RR), or triple alerts (3 abnormalities: HR, RR, and skin temperature). Three protocol changes improved the system-wide specificity.

Skin temperature was removed as an isolated escalation criterion and retained only for triple alerts. The RR lower limit was reduced from 11 to 8 breaths per minute to reduce clinically insignificant overnight alerts. The upper HR limit was increased from 115 to 120 beats per minute to reduce nonactionable alerts reported by community hospitals and the LTAC facility. Other parameters (eg, body position and activity) contributed to low motion alerts but were not primary triggers.

Collectively, these adjustments improved the balance between sensitivity and specificity, as reflected by observed reductions in nonactionable alerts and staff-reported decreases in false positives during enterprise rollout. The final escalation workflow and logic are depicted in Figures 3 and 4, with detailed initial and revised threshold values in Multimedia Appendix 5.

Technical and workflow challenges identified during the pilot informed the enterprise rollout. Barcode scanning difficulties (eg, poor label contrast) occasionally impeded device-to-patient matching; staff used device packaging barcodes and Workstations on Wheels carts as interim work-arounds, and the vendor subsequently improved barcode resolution. Rare double HR artifacts (concurrent S1 and S2 detections) produced nonactionable alerts and were resolved via algorithm updates.

Variation in device application sometimes caused skin irritation, which was addressed via staff reeducation and the use of adhesive remover spray. Early staff skepticism about using median CVSM values for documentation declined as experience accumulated, and staff perceived CVSM values to be accurate and consistent with routine bedside measurements; patient refusals decreased with standardized education. These mitigations and training reinforcements were propagated system wide during enterprise rollout.

As noted previously, unintended consequences, including minor skin irritation, initial device hesitancy, and early false positives, were observed and addressed through staff education, supplies, and vendor refinements; no significant harms were reported.

To our knowledge, this initiative represents the first large-scale deployment of wearable CVSM across a diverse, multihospital health system as part of a system-wide redesign initiative [17]. Our goal was to modernize inpatient VS practices by introducing continuous monitoring to support earlier detection, improve outcomes, and enhance both patient and clinician experience. Several units extended overnight VS intervals from every 4 hours to 6 to 8 hours, saving approximately 4 hours per nursing shift and allowing PCAs to focus on patient mobility and personal needs. Together with early clinical benefits, these efficiencies strengthened adoption and staff buy-in.

Weller et al [17] evaluated the same FDA-cleared device in 11,977 patients across urban and rural hospitals, with initial ICU use before expansion to non-ICU settings. They reported low RR alarms most frequently, while in our implementation, elevated RR emerged as the most common alert, underscoring RR’s role as a sensitive marker of cardiopulmonary function. This difference likely reflects patient mix, as sedation in ICU patients tends to lower RRs, while elevated RRs in ward patients more often indicate deterioration.

A key distinction of this program was its breadth and scope. Unlike previous efforts limited to ICUs, daytime hours, or narrow diagnostic groups, this program encompassed all adult non-ICU inpatients, deployed continuously (24×7) across the flagship academic hospital, 6 community hospitals, and 1 LTAC facility. This design addressed gaps in every 4-hour VS collection, enabled earlier detection across diverse patient populations, and supported workflow redesign amid nursing staff shortages.

Device selection prioritized accurate RR measurement, given its strong predictive value. Although the BioButton devices do not measure blood pressure or SpO₂, they remained available through existing bedside monitoring. Continuous SpO₂ monitoring was avoided to minimize alarm burden. The device offered unobtrusive, trended monitoring without disturbing patients’ rest while capturing key physiological parameters for early detection. The final escalation protocol’s emphasis on HR and RR reflected clinical prioritization and not device limitation. Although formal satisfaction and quality metrics are forthcoming, early stakeholder feedback suggests that reduced overnight disruptions improved sleep quality, potentially enhancing patient satisfaction and reducing overnight nursing workload. Prospective studies are needed to evaluate long-term effects on clinical outcomes, alert trends, workflow efficiency, and satisfaction, including patient-reported outcomes.

This report has several limitations. No formal satisfaction survey was conducted, and estimates of time savings were based on staff feedback. BioButton device–derived VSs were not independently validated, as accuracy testing of the FDA-cleared device was beyond the scope of this work. The escalation protocol was intentionally delayed in pilot units; therefore, early data reflect monitoring without active escalation, with finalized protocols introduced later and applied retrospectively. Although guided by an internal business case projecting nursing and PCA time savings, no formal health technology assessment was conducted.

Generalizability may be limited by HM’s centralized monitoring model; other health systems may face different challenges. Observed alert trends may also differ across settings, as illustrated by differences between ICU-focused studies and our broad non-ICU deployment. Outcome measures, such as clinical impact, workflow efficiency, and adoption, were not prospectively defined, constraining evaluation but informing priorities for future analyses.

Several transferable lessons emerged: (1) structured change management with engagement, champions, and feedback loops supported adoption; (2) phased rollout with iterative refinements reduced risk and enabled optimization; (3) centralized monitoring enabled 24×7 oversight, while hospitals without such infrastructure could adapt using telemetry staff, rapid response teams, or external partners; and (4) alignment of CVSM with institutional priorities secured leadership support and momentum. Table 1 summarizes key lessons and success factors to guide future implementers in scaling similar initiatives.

This initiative demonstrates the feasibility and operational value of scaling CVSM across a large, diverse health system. Embedding continuous monitoring into standard workflows can improve patient safety, enhance staff efficiency, and build infrastructure for digitally enabled care models.

Future work should rigorously assess clinical outcomes, economic impact, and patient-reported measures while exploring application in postacute and ambulatory settings. Although growing evidence supports continuous monitoring, prospective randomized trials are still needed to confirm clinical and economic benefits. As health systems address workforce shortages and rising patient acuity, wearable monitoring technologies, such as CVSM, could serve as core infrastructure for safer, more efficient, and patient-centered care.