We published the details of the study and the design rationale in a registered study protocol [28] and in the PROSPERO database (registration number: CRD322838). We conducted the review according to the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines and flowchart (Checklist 1) [29] to answer the following questions:

We searched the databases PubMed, Web of Science, and Scopus without restrictions on the document type. We used the MeSH to determine the relevant keywords and their synonyms. Two clinical specialists in the areas of emergency medicine and medical informatics approved the search terms. We composed English terms to search the databases by title, abstract, and keywords to find relevant studies published between January 2010 and December 2025. The search strategy covered two main concept areas relevant to the technical details of EW/TTS: technological and clinical terminology. We provided further details in the registered study protocol (Multimedia Appendix 1) [28].

We predefined the eligibility criteria presented in Textbox 1 in the registered study protocol [28] for the focused analysis of automated features, algorithms, and technologies of EW/TTS.

After searching the databases, we transferred the results into an EndNote library and removed duplicates. We imported the remaining records into Covidence [30] to support systematic screening. Four authors (SRNK, MDT, PH, and VMGS) with expertise in Medical Informatics and Biomedical Engineering independently reviewed the titles and abstracts to identify relevant studies. Subsequently, they assessed the full texts against the predefined inclusion and exclusion criteria following the PRISMA framework. We summarized the selection process in the PRISMA flow diagram (Figure S1 in Multimedia Appendix 2). During data extraction, the reviewers developed an Excel-based data extraction sheet that included variables reflecting the study aims and key features and functions of EW/TTSs. Two additional authors (MH and NG) contributed to data extraction and helped resolve inconsistencies in the extracted data through discussion until consensus was reached.

After completing the selection process, four authors (SRNK, MDT, PH, and VMGS) extracted the relevant parameters from the included studies. These parameters included the first author’s last name, title, publication year, country, study duration, study design, study population, hospital ward, primary and secondary objectives, clinical index, automation level, technology type, predictive algorithms, including statistical and machine learning approaches, and reported outcomes.

The extracted data were analyzed using a descriptive narrative approach, with results organized according to the review objectives. To assess automation levels, we adopted the CMM, as adapted by ISHO [27]. This framework evaluates the stages of maturity of digital health systems and defines six levels: absent, initial, managed, defined, measured, and optimized. Their operational definitions and decision rules are summarized in Table 1. The six levels are defined as follows: level 1, absent (no meaningful automation in data acquisition, analysis, or alerting); level 2, initial (limited, isolated automation without an integrated workflow); level 3, managed (automation implemented in practice without standardization or performance tracking); level 4, defined (structured and integrated automation workflow with limited measurement or feedback); level 5, measured (automation supported by explicit evaluation or performance monitoring); and level 6, optimized (automation continuously refined through feedback or quality improvement). The adapted CMM was used to assess workflow and system automation maturity rather than the sophistication of the underlying analytic method. To improve transparency and reproducibility, we operationalized the adapted CMM maturity levels using explicit decision rules based on the degree of automation reported for each EW/TTS. Classification considered whether the system included automated data acquisition, automated data analysis and/or score calculation, automated alert generation or escalation support, integration with other digital systems, and evidence of performance monitoring or continuous improvement. Reviewers independently assigned each EW/TTS to one maturity level using these criteria and the operational definitions presented in Table 1. Disagreements were resolved through discussion until consensus was reached. When a study did not provide sufficient evidence to support a higher maturity level, the lower level was assigned based on the available evidence. Furthermore, we grouped the applied technologies into six mutually exclusive domains: Data Analytics, Hardware, Data Collection, Communication, Software, and Unimplemented Systems; their operational definitions and representative examples are summarized in Table 2.

We assessed the methodological quality of the included studies using the Joanna Briggs Institute (JBI) critical appraisal checklist with 8 main items [39]. We adapted the JBI case report checklist, which considers the step-by-step cycle of system development (Multimedia Appendix 3).

Using PRISMA [29], we identified 1181 studies and removed 133 records due to duplication (Figure 1). We further excluded 927 records at the title and abstract level. We screened 121 full-text records, of which 43 studies [1,3,4,17,31-38,40-70] met the eligibility criteria (Figure S1 in Multimedia Appendix 2). We classified the studies according to the Study Design Classification (SDC) framework [71]. The analysis included descriptive studies (cross-sectional: 10/43, 23.3% [1,32,35,38,46,55,60-62,66], qualitative: 5/43, 11.6% [4,33,49,50,59]) and analytical studies. The latter consisted of experimental (randomized clinical trials: 3/43, 7% [36,37,56], nonrandomized clinical trials: 9/43, 20.9% [3,31,34,40,41,51,57,65,70]) and observational designs (cohort: 6/43 [17,42,45,48,53,67], 14%, retrospective: 10/43, 23.3% [43,44,47,52,54,58,63,64,68,69]). There were reports from 16 countries; 26 of 43 (60.5%) [1,33,34,36,37,40,43,45,46,48-50,53,55,57-63,68,69] studies were from the United States, United Kingdom, and the Netherlands. The studies ranged from 0.8 to 78 months, with an average of 22.8 months (Table 3).

Different EWS charts for adults, pediatric patients, and pregnant women reflect each group’s physiological norms and clinical risks, ensuring accurate assessment and timely intervention. This differentiation enhances the sensitivity and specificity of monitoring, improving patient outcomes through tailored clinical responses [2,8,16,17,62]. We identified different participant groups across the included studies: adults older than 18 years (35/43, 81.3% [1,3,4,17,32-38,40-44,46,47,49-56,58-61,65-70]), children, patients younger than 18 years (6/43, 14% [31,35,45,48,63,64]), and pregnant women (2/43, 4.7% [57,62]). Most EW/TTS were applied in health monitoring in hospitals (39 of 46, 84.8% reported settings [1,3,4,17,31,33-37,41-46,48,50-54,56,57,59-65,67,69,70]). Health care providers use them primarily in hospitals: surgery wards 12 of 46 (26.2%) reported settings [1,4,40-43,46,47,49,53,54,65], general wards 10 of 46 (21.7%) reported settings [17,33,34,51,52,56,60,61,67,70], pediatric wards 6 of 46 (13%) reported settings [31,35,45,48,63,64], emergency wards 3 of 46 (6.5%) reported settings [3,50,69], and other wards including ICU, respiratory, maternal, oncology, and short-stay 8 of 46 (17.4%) [36,44,49,56,57,62] reported settings. In addition, 7 of 46 (15.2%) reported settings [32,38,40,46,58,66,68] applied in other settings (Figure S2 in Multimedia Appendix 2); one study applied VS monitoring for CD detection in cars [32] while another applied it in older adult care residences and hospices [38].

The primary objectives of EW/TTS were CD detection in 24 of 44 reported primary objectives (54.5%), ICU transfer in 4 of 44 (9.1%), death in 4 of 44 (9.1%), and other objectives in 12 of 44 (27.3%). The secondary objectives were ICU transfer in 16 of 68 reported secondary objectives (23.5%), death in 10 of 68 (14.7%), clinical intervention in 8 of 68 (11.8%), length of stay in 7 of 68 (10.3%), and other objectives in 27 of 68 (39.7%; Figure 2).

To detect and predict CD, the EW/TTS used various clinical indexes, such as VS, assessment scores, clinical conditions, diagnostic results, demographics, and specialized measurements. VS parameters such as HR, RR, SpO₂, BT, and BP, along with measurements such as stroke volume and cardiac index, dominate the landscape in 29 of 67 reported clinical indexes (43.3%), underscoring their critical role in patient monitoring. Assessment scores (EWS, pediatric EWS, patient status index, AVPU, and modified EWS) accounted for 13 of 67 (19.4%), highlighting their significance in evaluating patients’ conditions. Clinical conditions, such as consciousness, blood infections, and pain level, accounted for 9 of 67 (13.4%), underscoring the role of specific symptoms. Diagnostic results (6 of 67, 9%), demographics and history (6 of 67, 9%), and specialized measurements (4 of 67, 6%) were less frequent but still integral to comprehensive patient assessment. In summary, EW/TTS majorly relied on VS in CD detection or prediction (Figure S3 in Multimedia Appendix 2).

According to ISHO’s adapted CMM [27], 18 of the 43 implemented EW/TTS (41.9%) were classified at the measured automation level. At this level, the EW/TTS data analysis processes are systematically implemented and actively managed. Ongoing efforts are in place to develop and refine the system, with established mechanisms to measure performance, track progress, and ensure coordination among involved teams. The managed automation level, identified in 11 of 43 (25.6%) [8,44,45,47,48,53,56,58,65,69] studies, represented an intermediate stage of automation maturity in which EW/TTS processes were implemented in practice but had not yet been systematically documented or institutionalized.

Seven of 43 (16.3%) studies [3,37,43,52,62-64] were classified at a defined automation level. Systematic, ongoing efforts to develop EW/TTS are underway; however, there is no comprehensive method to measure progress or ensure coordination. On the other hand, only 4 of 43 (9.3%) studies [32,40,59,68], 2 of 43 (4.7%) studies [35,36], and 1 of 43 (2.3%) studies [31] were classified as initial, optimized, and absent automation levels, respectively. The absent level represents a complete lack of automation capability, whereas the initial level is characterized by ad hoc and informal automation efforts. In contrast, the optimized level corresponds to the most advanced stage of automation, characterized by systematic, ongoing development of EW/TTS, supported by continuous quality improvement activities that align EW/TTS outcomes with the strategic objectives and performance metrics (Figure 3).

More than half of the included (24/43, 55.8%) studies [1,3,4,17,32,36,40,44-46,48-52,55,56,58,60,62-64,67,69] focused on data analytics technologies. Across the measured, defined, and managed automation levels, data analytics was the dominant technological focus in EW/TTS implementations. Data analytics included machine learning and AI, statistical analysis, visualization tools, and cloud services. Cloud services and visualization tools supported real-time dashboards for visual analytics. Predictive algorithms, including statistical and machine learning approaches, were identified in 11 of 43 (25.6%) studies [32,34,40-44,47,51,60,67].

Hardware technologies were identified in 7 of 43 (16.3%) studies [33,54,57,59,66,68,70], primarily comprising wearable devices and sensors. Wearable devices include HealthPatch [33], Hexiwear [38], ViSi Mobile [68], and waterproof patches [59], all of which are essential for continuously monitoring VS and other health metrics. These devices use sensors to record photoplethysmography (PPG) and other physiological data. Additionally, communication technologies accounting for 5 of 43 (11.6%) studies [34,41-43,47] use Bluetooth [32,36,52], Wi-Fi [33,52], and Bluetooth low energy (BLE) [32] to ensure seamless connectivity. Software technologies encompass mobile apps, eHandover, Node.js API, and eObs for remote health monitoring (Figure 3).

The trend in EW/TTS automation indicated a clear long-term progression toward increased automation. During the early years (2010‐2015), there was minimal automation. Since 2016, there has been a consistent shift toward managed and measured automation levels. This transition drives a sharp rise in the weighted total, which peaks around 2020‐2021 and temporarily declines in 2022‐2023. We observed a strong rebound in 2024‐2025, with high levels of managed, defined, and measured automation signaling mature, data-driven, and continuously improving automation practices (Figure 4).

Data analysis using predictive algorithms, including statistical and machine learning approaches, occurred in only 11 of 43 (25.6%) [32,34,40-44,47,51,60,67] studies, all published in 2019 or later (Table 4). The applied algorithms included eCART [43], a multivariate probabilistic model [44], logistic regression-based models [34,41,42,47], logistic regression, decision trees, neural networks, and ensemble models [51], hybrid k-nearest neighbors least-squares support vector machines (SVMs) [40], natural language processing for text analysis, decision trees, and logistic regression [60], and a time-aware MARS model [67]. Furthermore, Chowdhury et al applied an SVM [32] for electrocardiogram (ECG) analysis and CD prediction within EW/TTS. Table 4 summarizes the predictive algorithms identified in the included studies, together with their reported data sources, prediction targets, time horizons, and performance metrics, where available. Thirty of 43 systems (69.8%) [3,17,33,34,36-38,41-44,47,49-51,53,54,56-58,60-65,67-70] had a data exchange feature and connectivity with other systems. Connectivity with EHRs was reported in 21 of 43 systems (48.8%) [3,17,34,37,41-44,47,50,51,53,56-58,61-64,67,69], wearable-device connectivity in 3 of 43 systems (7%) [33,38,54], and connectivity through a proprietary cloud platform in 1 of 43 systems (2.3%) [70]. Among connected systems, 5 of 43 (11.6%) [36,49,60,65,68] reported interoperability standards or communication platforms, including Fast Healthcare Interoperability Resource (FHIR), Health Level Seven (HL7)-based connectivity, and Ascom. Eight of 43 systems (18.6%) [1,4,31,32,35,40,48,59] had no connectivity, and connectivity was not specified in 5 of 43 systems (11.6%) [45,46,52,55,66].

The reported outcomes of EW/TTS were earlier warning in 14 of 70 reported outcomes (20%), higher accuracy in 12 of 70 (17.1%), lower specificity in 9 of 70 (12.9%), more timely reaction in 7 of 70 (10), and reduced mortality in 6 of 70 (8.6%; Figure S4 in Multimedia Appendix 2).

We critically appraised the quality of the studies using the adapted JBI checklist comprising eight criteria [39]. Most studies received a high proportion of “Yes” responses, suggesting good compliance, particularly in “System evaluation,” “Lessons learned,” “Characteristics,” “Post-implementation,” “Pre-implementation,” and “Implementation.” However, criteria related to “Development history” and “System failure” were less well reported. Only 19 of 43 (44.2%) studies [1,3,4,17,31,32,35,36,42,45,48,49,51,55-57,59,62,70] clearly addressed each of these items, whereas 24 of 43 (55.8%) studies [33,34,37,38,40,41,43,44,46,47,50,52-54,58,60,61,63-69] were rated as No, Unclear, or Not Applicable (Table 5).

This systematic review explored the automated features, algorithms, and technologies of electronic EW/TTS. We identified 43 systems that revealed the transition from manual charts to electronic data. We found only 16 countries implementing electronic EW/TTS; the majority were reported in the United States (15/43, 34.9%), the United Kingdom (6/43, 14%), and the Netherlands (6/43, 14%). This distribution may partly reflect differences in digital health readiness, interoperability capacity, and policy support for the implementation of standardized early warning systems. In the United States, near-universal hospital EHR adoption and increasing interoperable health information exchange may have facilitated the integration of EW/TTS into routine care workflows [72]. In the United Kingdom, the national standardization of NEWS/NEWS2 across the National Health Services probably made them easier to use by ensuring that escalation criteria, documentation practices, and clinical training were consistent [73]. In the Netherlands, sustained national efforts to promote eHealth and electronic health data exchange, including the proposed Wegiz framework, may likewise have strengthened readiness for digitally enabled monitoring and alerting systems [74]. In combination, these factors may have made it easier for EW/TTS to be developed, implemented, and published than in places with lower digital maturity or less standardized escalation frameworks.

Nevertheless, we assume that many countries still consider electronic EWS calculations or have not yet reported their EW/TTS. Since 2016, electronic EW/TTS have advanced gradually through integration with machine learning and AI models, particularly in critical care and emergency medicine [19].

The automation levels of electronic EW/TTS ranged from electronic VS monitoring to real-time risk score prediction. Current EW/TTS largely focus on EWS calculation [2], with only a limited proportion of the included studies applying predictive algorithms to support earlier detection of deterioration. However, shorter prediction intervals and more accurate models are still needed. Although several tools to predict CD already use deep learning [75], they are not yet integrated into current EW/TTS.

The adapted CMM categories reflected differences in how automation was implemented and governed in practice. Managed systems generally showed operational automation in routine use but limited standardization or performance tracking, whereas measured systems were supported by explicit evaluation, outcome monitoring, or both. From a clinical perspective, this distinction is important because more mature automation may support better workflow integration, greater transparency of system performance, and more reliable decision support, even when the underlying predictive intelligence remains limited.

Automation maturity and algorithmic intelligence should not be interpreted as equivalent. In this review, the adapted CMM reflected the maturity of workflow automation, implementation, integration, and performance monitoring, whereas intelligence referred to the sophistication of the analytic approach, ranging from automated rule-based score calculation to statistical, machine learning, and more advanced AI-based models. Accordingly, a system could achieve a relatively advanced automation level through structured data acquisition, score calculation, alert delivery, and outcome monitoring without necessarily using machine learning. Conversely, the presence of predictive or AI-based models did not by itself indicate mature workflow integration (Figure 5).

The distinction between managed and measured automation also had practical implications beyond technical classification. Managed systems generally indicated that core automation functions were operational in routine care, but they often showed limited evidence of standardized evaluation, feedback, or outcome monitoring. By contrast, measured systems were supported by explicit assessment of performance, outcome tracking, or both. From a patient-safety perspective, this distinction is important because systems with measured automation may provide greater transparency regarding alert behavior, reliability, and unintended consequences. From a workflow perspective, measured systems may also be better positioned to support integration into escalation pathways and more consistent clinical use. In this sense, higher automation maturity may strengthen decision-making reliability not only through automation itself but through monitoring, validation, and governance of the automated process. Although the heterogeneity of study designs, outcomes, and performance metrics prevented a formal comparative analysis of maturity level against effectiveness, the reviewed evidence suggested that more mature systems were more often associated with structured evaluation, earlier warning, and workflow-oriented implementation features. At the same time, outcome trade-offs such as false alarms and lower specificity remained important, indicating that higher maturity should not be interpreted as uniformly better performance across all settings.

Predictive algorithms in the reviewed studies were primarily used to stratify patients according to their risk of CD, ICU transfer, AEs, or death [32,40,60]. For example, Chowdhury et al [32] reported an accuracy of 97.4% using SVM-based ECG analysis in a real-time wearable system. In this case, the high accuracy reflects the model’s ability to identify ECG patterns associated with myocardial infarction within a 10-second analysis window. This highlights that integrating AI-based analysis into EW/TTS enables real-time early warning of CD, extending beyond traditional EWS calculation. Preventive care and CD control aim not only at CD detection or prediction [76] but also at improving the quality of care, such as EWS error reduction [64], more timely reaction, and better decision-making for patient transfer to ICU, activating RRT, or optimizing the length of hospitalization [16]. Research in critical care has shifted the focus of EW/TTS’s outcomes from mortality reduction toward quality improvement.

Most EW/TTS operate in hospitals, especially in surgery wards (Figure S2 in Multimedia Appendix 2). Only 2 studies [32,38] considered CD detection in smart cars or in older adult care facilities. However, turning our medical systems from a curative to a preventive approach is essential [20]. Home care, particularly for older adults after hospital discharge or those living alone, increasingly relies on electronic EW/TTS to ensure safety monitoring within the private environment [77]. Deserno has already documented the general idea of transforming such places into a diagnostic space [20], and Wang et al [78] provided a comprehensive review on unobtrusive health monitoring in smart homes. Bringing this preventive approach to prehospital care may enhance the overall health care level and reduce AE, such as heart or respiratory failure, which are often precursors to death [79].

Gerry et al [16] assessed the efficacy of manual systems, whereas our systematic review evaluated electronic systems. The main reported outcomes of EW/TTS evaluation included early warning, high accuracy, and lower specificity (Figure S4 in Multimedia Appendix 2). These outcomes were generally positive, primarily for accurate, timely decision-making. Despite these positive results, electronic EW/TTS still need to improve through automation using AI technology.

Several studies reported the use of deep learning or machine learning to calculate predictive EWS [80,81]. The predictive models generated higher-level knowledge to support clinical decision-making. Hence, clinicians should replace the EWS calculation based on charts [2] with valid predictive models using AI methods [75]. Scores from charts indicate CD at the point of care, whereas the predicted value provides a forecast of CDs, which gives the nurse enough time to support the patient and activate RRT quickly. However, directly comparing algorithm performance across studies remains challenging because of substantial heterogeneity in clinical settings, input variables, prediction targets, and reported outcome measures. Cross-study comparability remained limited because the included EW/TTS differed substantially in clinical setting, prediction target, time horizon, and reported performance metrics. Prediction targets included CD, ICU transfer, AEs, and mortality; time horizons ranged from near-real-time alerts to forecasts several hours in advance; and performance was reported inconsistently using measures such as AUC, sensitivity, specificity, and PPV. As a result, the reported performance values should be interpreted within the context of each individual study rather than as directly comparable measures of superiority across systems or settings.

For instance, the system described by Alam et al [8] predicted EWS earlier based on different clinical, demographic, and physiological VS. The output alerted, reminded, and notified users based on their roles to ensure more sensitive, accurate, and timely actions.

Besides the remarkable advancements and achievements in AI, data analytics, cloud computing, and mobile technologies, significant improvements have occurred. Furthermore, high-speed internet, the Internet of Things, and standardized protocols facilitated seamless data exchange. Therefore, the full potential of AI and data analytics in electronic EW/TTS has not yet been realized [79].

Recent reviews have reported limited predictive performance of traditional aggregate-weighted EW/TTS [82]. Muralitharan et al [83] demonstrated that machine learning-based EW/TTS outperformed the conventional risk scoring approach in predicting CD. These findings align with the results identified in our review, which indicate a growing shift toward an automated EW/TTS approach.

Despite this advancement, the literature provides a limited synthesis of the extent of automation across EW/TTS components, particularly regarding data acquisition, algorithmic processing, and alert generation. Our review identified key enabling technologies and mapped their implementation across EW/TTS components. Based on the synthesized evidence, we propose a conceptual framework describing intelligent EW/TTS integration across three functional domains: automated data collection (afferent limb), inference and predictive analytics (inference engine), and automated output (efferent limb). In this conceptual model, the afferent limb is responsible for automated data acquisition, whereas the inference engine performs analytic functions such as VS interpretation, critical score calculation, and predictive modeling. The efferent limb is responsible for output functions, including alert generation, escalation, and communication to clinicians or RRTs (Figure 6). Intelligent EW/TTS integrates automated data acquisition within the afferent limb. This enables an embedded inference engine to continuously monitor VS, calculate and predict EWS, and apply advanced analytics to support actionable clinical decision-making. The inference engine links the afferent limb to the efferent limb by generating trigger outputs when predefined thresholds are exceeded. The efferent limb generates timely alerts to clinicians and RRTs, coordinates escalation pathways, and facilitates prompt intervention. By accelerating recognition and enhancing informed decision-making, these systems may improve patient outcomes. We therefore suggest that future intelligent EW/TTS should align both limbs with this model [11] and incorporate automation together with smart features. Sensors and wearable technologies for VS data collection may improve data quality. Research has shown that patch sensors provide higher precision for HR and RR [1].

The quality appraisal findings also affected the strength of evidence in this review. While most studies adequately described system characteristics, implementation, and evaluation, many provided limited information on development history and system failure. This gap makes it harder to assess how mature, reproducible, and thoroughly tested a system was before deployment. It also makes it more difficult to identify technical flaws, safety risks, and unintended effects that could arise during real-world use. As a result, the evidence may overemphasize reported benefits while underrepresenting implementation challenges and system-level risks. Readers should consider this limitation when interpreting the results.

However, this review also has limitations. First, most of the studies included in this review were conducted in hospital settings, which may limit the generalizability of the findings to nonclinical environments. Second, the collection of VS using wearable devices poses challenges, as some wearable monitoring systems still require further clinical evaluation and validation before broader adoption [52,59,68]. This may affect measurement accuracy and reliability. Third, regulatory and ethical factors, as well as the complexity of the health care system, may constrain widespread adoption of the EW/TTS. We also did not conduct a formal comparative analysis of country-level health system characteristics, including digital infrastructure readiness, EHR adoption, interoperability maturity, and national policy frameworks for early warning implementation, which may partially elucidate the concentration of reported EW/TTS in a limited number of countries.

Finally, the possible outcomes of implemented EW/TTS may include no change, statistically insignificant effects, or inconclusive results. However, we did not extract these outcomes because they were outside the scope of this review, which may limit the generalizability of our findings.

Future research should focus on collecting high-quality, real-time VS data across diverse settings beyond hospitals, including smart home and in-vehicle environments, to improve the validity and accuracy of predictive models with acceptable sensitivity and specificity. The use of clinically validated wearable devices and the development of a standardized data collection framework may further improve model accuracy and reliability. Furthermore, adopting modern technologies such as smart devices, cloud computing, AI, robotics, and the Internet of Medical Things can revolutionize EW/TTS. AI technology powers intelligent EW/TTS to enhance decision-making and automate interventions. These advances offer opportunities to leverage modern technologies for transformative applications in health care, ultimately enhancing the quality of care and saving lives. As EW/TTS increasingly integrate AI-driven analytics and decision support, their deployment must also be evaluated within the context of evolving regulatory and digital health safety frameworks. Recent FDA guidance on AI-enabled device software functions and clinical decision support software emphasizes the need for clinical AI systems to be transparent, to undergo lifecycle oversight, to support performance monitoring, and to enable safe model updates.

Current EW/TTS primarily focused on patient monitoring in hospital surgery wards and most commonly demonstrated measured automation. The primary objective of EW/TTS was to detect CD, while ICU transfers and mortality prediction were the most common secondary objectives. EW/TTS predominantly relied on VS and assessment scores to detect and predict CD. The included studies primarily used data analytics technologies. More than half of EW/TTS featured data exchange capabilities and connectivity with other systems. The primary reported outcomes of EW/TTS implementation included early warning, high accuracy, and lower specificity.