In a surgical ward equipped with continuous wearable sensors, transitioning routine nursing assessments from mandatory scheduled intervals to clinical discretion reduced the overall frequency of intermittent vital sign measurements while increasing the proportion of recorded scores indicating potential patient deterioration. Researchers conducting an observational investigation at Catharina Hospital Eindhoven in the Netherlands evaluated how altering bedside nursing protocols influenced the Modified Early Warning Score, known as MEWS, alongside two remote automated scoring algorithms derived from continuous sensor data. Across the evaluation, moving away from mandatory manual vital sign recording shifted the statistical performance of manual scores, raising measured sensitivity while diminishing specificity, whereas continuous scoring algorithms demonstrated modest overall diagnostic accuracy alongside a notable elevation in alert generation.
Hospital surgical units traditionally rely on intermittent bedside examinations performed by nursing staff two to three times daily to track patient stability and identify signs of postoperative decompensation. These routine assessments compile physiological parameters, including heart rate, respiratory rate, blood pressure, body temperature, and oxygen saturation, into structured scoring frameworks such as MEWS to guide clinical escalation. Although intermittent scoring remains standard in surgical wards, prior investigations have demonstrated that scheduled manual observations exhibit limited sensitivity for capturing subtle or rapid physiological shifts between rounds. The introduction of continuous monitoring devices aims to bridge these surveillance gaps, yet integrating automated physiological streams into existing nursing workflows raises operational questions regarding whether scheduled manual recordings remain necessary when continuous telemetry is already active.
To investigate the clinical and operational consequences of shifting monitoring protocols, investigators analyzed 544 surgical admissions representing 516 individual patients treated between October 2024 and June 2025. The study population was admitted to a general surgical ward that managed non-malignant conditions across fourteen specialties, predominantly general surgery representing 48 percent of admissions, urology accounting for 28 percent, and gynecology comprising 8 percent. Throughout the nine-month evaluation, all admitted patients wore a CE-marked reusable sensor on their upper arm that captured heart rate, respiratory rate, and blood oxygen saturation via photoplethysmography at one-minute intervals. Continuous monitoring data were transmitted wirelessly to hospital systems, aggregated into four-hour medians, and displayed directly within electronic health records for nursing review during routine shifts without automated pop-up alarms.
The study timeline was divided into two distinct operational phases reflecting real-world ward workflow adjustments rather than an experimental intervention. During Phase 1, spanning October 2024 to March 2025, continuous wearable telemetry was active and visible to healthcare staff while routine nurse-measured vital signs and MEWS calculations remained mandatory under institutional protocols. In Phase 2, which ran from March 2025 to June 2025, continuous wearable tracking remained fully accessible, but mandatory schedules for bedside vital signs were eliminated, leaving manual MEWS calculations entirely to staff clinical judgment. Because continuous sensor data were accessible to bedside staff across both phases, the primary structural divergence between the two periods resided in the administrative mandate governing manual nursing checks.
Researchers tracked postoperative complications graded Clavien–Dindo II or higher as the primary clinical deterioration endpoint, identifying 154 deterioration events occurring across 98 admissions, representing 18 percent of total admissions. Clinical records underwent manual verification by two independent researchers, with deterioration timing established by documented diagnostic confirmation, initiation of targeted treatment, or formal physician diagnosis. Patients experiencing postoperative deterioration exhibited a significantly longer median hospital stay of 9 days, compared to a median stay of 2 days among patients without deterioration. The protocol transition between phases coincided with a reduction in median hospital stay from 3 days in Phase 1 to 3 days in Phase 2 with a narrower interquartile span, while patient age showed no significant variation between phases.
The shift to discretionary assessments generated marked changes in bedside documentation patterns. Total recorded intermittent vital sign sets declined from 1,069 measurements during Phase 1 to 956 measurements in Phase 2, demonstrating a statistically significant reduction in manual observation frequency. Concurrently, the proportion of manual vital sign sets reaching an elevated threshold of MEWS 3 or higher surged from 5.9 percent during the mandatory phase to 20.8 percent under discretionary management. This proportional increase caused the calculated sensitivity of manual MEWS for detecting clinical deterioration within 24 hours to rise from 0.12 in Phase 1 to 0.38 in Phase 2, whereas specificity declined from 0.95 to 0.81, reflecting a transition toward measurements taken primarily when nurses suspected clinical decline.
Alongside evaluating nursing assessment protocols, the researchers assessed the diagnostic performance of two wearable-derived scoring models: the Continuous Remote Early Warning Score, or CREWS, and the Remote MEWS-based Continuous score, termed R3EWS. CREWS adapted an established continuous scoring framework to incorporate heart rate, respiratory rate, and oxygen saturation with an alarm threshold set at 3. R3EWS modified the hospital standard MEWS table to score the same three parameters while assigning partial penalty scores to minor physiological deviations and higher penalty scores to severe deviations. When evaluating diagnostic discrimination across all score values, R3EWS yielded an area under the receiver operating characteristic curve of 0.61, while nurse-measured MEWS achieved 0.58 and CREWS demonstrated 0.58.
When assessed at the clinical escalation threshold of 3 or higher, the algorithms revealed distinct trade-offs between event detection and operational workload. R3EWS captured 18 percent of clinical deterioration events occurring within the subsequent 24 hours, compared to 13 percent identified by manual MEWS and 9 percent detected by CREWS. R3EWS achieved an overall sensitivity of 0.16 and a specificity of 0.93, whereas CREWS produced fewer total alarms, yielding the highest specificity at 0.98 but the lowest sensitivity at 0.05. However, both wearable-derived scoring algorithms produced a substantially higher absolute volume of alerts than intermittent nursing assessments, driven by continuous data calculation across four-hour epochs that yielded 121 percent more assessment points than manual charting during Phase 1.
An exploratory temporal analysis examining one-minute data streams during the 24 hours preceding clinical deterioration revealed physiological warning patterns. The proportion of alarms generated by wearable scores increased progressively in the hours approaching documented complications, displaying distinct activity peaks approximately 6 to 8 hours prior to formal diagnosis. Furthermore, threshold-matching analyses indicated that to equal the 11 percent true positive rate achieved by conventional MEWS, the operational threshold would need adjustment to approximately 2.8 for R3EWS and 1.0 for CREWS. These numerical disparities underscore how algorithms tracking three vital signs require lower numerical thresholds than comprehensive five-parameter bedside scores to achieve comparable sensitivity.
The authors highlighted several methodological limitations, noting that the single-center before-and-after design precluded isolating the independent effect of wearable technology from protocol modifications. Because continuous data were available throughout both study phases, reductions in manual measurements and changes in MEWS sensitivity reflected combined organizational adjustments rather than sensor deployment alone. Furthermore, defining deterioration timing by clinical confirmation rather than physiological onset, the retrospective application of algorithms, and the modest overall discriminative power observed across all models emphasize the necessity for multicenter prospective evaluations before automated scoring can be integrated into routine ward escalation pathways.
Because postoperative clinical deterioration occurred in only 18 percent of admissions, the investigators incorporated precision-recall curve analyses to evaluate score discrimination under pronounced class-imbalanced conditions. When assessing the area under the precision-recall curve, conventional nurse-measured MEWS reached 0.12, with a 95 percent confidence interval spanning 0.09 to 0.17. In comparison, continuous algorithmic monitoring demonstrated numerically lower values, with CREWS achieving an area under the precision-recall curve of 0.09 and R3EWS recording 0.08. Statistical comparisons between the manual and wearable scoring approaches indicated that these differences did not achieve statistical significance, reflecting the persistent challenge of accurately predicting infrequent adverse events regardless of measurement continuity.
Stratifying diagnostic performance across the two monitoring periods demonstrated distinct operational differences between the algorithms. During Phase 1, wearable data generated 2,372 four-hour scoring epochs, representing a 121 percent increase in available evaluations over the 1,069 nurse-measured assessments. In this mandatory phase, R3EWS yielded an area under the receiver operating characteristic curve of 0.58, significantly exceeding the 0.54 achieved by CREWS. Both R3EWS and manual MEWS identified 12.5 percent of clinical deterioration events, whereas CREWS produced a sensitivity of just 0.031. During Phase 2, under discretionary nursing measurements, R3EWS anticipated 20.2 percent of deterioration events, nearly twice the 10.6 percent captured by CREWS, and achieved a sensitivity of 0.172, outperforming both discretionary nurse charting at 0.120 and CREWS at 0.061. Across Phase 2, R3EWS maintained a significantly higher receiver operating characteristic area of 0.62 compared to 0.59 for CREWS.
Technical and observational factors also influenced the evaluation. Prior assessments of the upper-arm sensor noted acceptable agreement for heart rate measurements in surgical cohorts, but recorded greater variability for respiratory rate and oxygen saturation readings. Analytically, the investigators retained scoring calculations recorded after complications emerged. While preserving post-deterioration scores mirrored real-world ward conditions where staff continue reviewing patient data following adverse events, this approach may have inflated cumulative alarm counts and slightly depressed specificity. Additionally, the study did not perform interrupted time series modeling or differentiate seasonal variations in hospital capacity and nursing staffing. Because the exploratory 24-hour temporal analysis lacked an unaffected control group, potential circadian patterns or time-dependent biases could not be excluded.
Subject of Research: Evaluation of discretionary nurse MEWS assessments and wearable-derived early warning scores in surgical ward monitoring
Article Title: Transition from mandatory to discretionary nurse-measured MEWS assessment in a continuously monitored surgical ward: Diagnostic performance of wearable-derived early warning scores
Article References: Jerry, E., Afonso, M., De Vries, J., Akkermans, A., Turco, S., Bouwman, A., & Nienhuijs, S. (2026). Transition from mandatory to discretionary nurse-measured MEWS assessment in a continuously monitored surgical ward: Diagnostic performance of wearable-derived early warning scores. Journal of Clinical Monitoring and Computing. https://doi.org/10.1007/s10877-026-01488-y
Image Credits: AI Generated
DOI: 10.1007/s10877-026-01488-y
Keywords: continuous monitoring, wearable technology, surgical ward, Modified Early Warning Score, clinical deterioration, patient monitoring, vital signs, photoplethysmography, CREWS, R3EWS, postoperative complications, hospital workflow, nursing assessment, diagnostic performance, AUROC, health informatics, clinical decision support

