Effectiveness of early warning scores in predicting in-hospital mortality and adverse events in adult general-ward patients in low- and middle-income countries: A systematic review

Highlights

• •EWS performance in LMIC wards is heterogeneous and context dependent.
• •NEWS and MEWS show moderate-to-good discrimination for in-hospital mortality.
• •UVA needs HIV testing, limiting comparability with bedside-only scores.
• •Observational designs predominate; implementation evidence is scarce.
• •Multicentre pragmatic trials and local recalibration are essential.

Abstract

Background

Early Warning Score systems (EWSs) based on bedside physiological parameters are widely implemented in high-income countries, yet their performance and utility in low- and middle-income countries (LMICs) remain uncertain, particularly in emergency and acute care settings, due to limited evidence and health system constraints. We systematically reviewed the effectiveness of EWSs for predicting in-hospital mortality and adverse events among adult general-ward patients in LMICs.

Methods

This systematic review followed PRISMA 2020 and PRISMA-S guidance and was prospectively registered in PROSPERO (CRD420251029273). We searched PubMed, Scopus, LILACS and African Journals Online (AJOL) from inception to May 2025, with no language restrictions. We included studies enrolling adult general-ward patients in LMICs, and excluded studies conducted solely in intensive care units, emergency departments, paediatric or obstetric populations. Data were narratively synthesised, and risk of bias was assessed using PROBAST.

Results

Twenty-eight observational studies comprising a total of 36,638 participants — primarily from sub-Saharan Africa and South Asia — met inclusion criteria. The Modified Early Warning Score (MEWS) and National Early Warning Score (NEWS) were most frequently assessed and generally demonstrated moderate-to-good discrimination for in-hospital mortality, with substantial heterogeneity by setting. The Universal Vital Assessment (UVA) showed promising discrimination relative to MEWS and qSOFA in some resource-limited contexts, however its dependence on HIV testing and laboratory support limits comparability with purely bedside scores. Only a minority of studies examined unplanned ICU admission, cardiac arrest or rapid-response activation. Limitations included substantial heterogeneity in methodology, outcomes, cut-off values and follow-up periods, which limited comparability, as well as the absence of studies originating from Central and South America.

Conclusion

Evidence in LMIC wards is largely observational and frequently based on statistical simulations with small samples; it should not be interpreted as proof of clinical effectiveness. Before widespread adoption, rigorous local validation and recalibration, along with multicentre, pragmatic implementation studies, are needed to define thresholds, workflows and escalation pathways appropriate to LMIC health-system capacity.

Graphical abstract

African relevance.

• •High burden of preventable in-hospital deaths in Africa highlights the potential value of EWSs in resource-constrained ward settings.
• •Most included studies were conducted in African countries, ensuring direct applicability of findings to regional healthcare systems.
• •Variable performance of NEWS, MEWS, and UVA underscores the need for local validation and recalibration within African contexts.
• •Context-specific implementation of EWSs could inform national patient safety policies, workforce training, and rapid response systems across Africa.
• •Although Africa-focused, similar challenges arise across LMICs; therefore, implementation lessons and performance characteristics are transferable to other resource-limited healthcare environments globally.

Introduction

Early warning score systems (EWSs) were introduced almost three decades ago to address the high incidence of preventable deaths and adverse events among hospitalised patients [[1], [2], [3], [4], [5]]. These standardised clinical assessment tools use simple physiological parameters — such as respiratory rate, heart rate, blood pressure, temperature, level of consciousness, oxygen saturation, and/or other parameters — routinely collected at the bedside to identify signs of clinical deterioration [[1], [2], [3], [4], [5]]. They provide valuable information regarding the risk of ward cardiac arrest, unplanned admission to intensive care unit (ICU), and in-hospital mortality, and are intended to inform timely monitoring, escalation and intervention on general wards [[1], [2], [3], [4], [5]].

Following their introduction, these tools gained widespread adoption due to their reproducibility, ease of use, and high user acceptability [2,3,[6], [7], [8], [9]]. They are now employed not only on general wards but also in Emergency Departments and, increasingly, in pre-hospital settings for early risk stratification. Despite their widespread use, the extent to which EWSs effectively predict short-term mortality and adverse events remains variable and context-dependent [[10], [11], [12]].

In high-income countries, particularly in the United Kingdom, EWSs are well described and routinely used to identify patients who may benefit from closer monitoring or escalated treatment [[10], [11], [12]]. Their use as screening tools is also recommended by national guidance, such as that issued by the National Institute for Health and Care Excellence (NICE) [12].

Conversely, in resource-limited settings — typically represented by low- and middle-income countries (LMICs) — the healthcare context is markedly different. These countries often face significant constraints in health system management, limited financial resources, and shortage of trained healthcare professionals to meet patient demand. Consequently, the implementation and effectiveness of EWSs in such settings have not been thoroughly investigated. In LMICs (e.g. Angola), available evidence regarding the performance of EWSs remains scarce and fragmented [7,[10], [11], [12], [13], [14]].

Although most of the included studies were undertaken on general inpatient wards, the physiological parameters used in EWSs are routinely measured in Emergency Departments, triage areas and acute medical admission units. There is therefore a clear conceptual and practical link between in-hospital deterioration surveillance and emergency care decision-making. EWSs are relevant because they can support prioritisation of care upon arrival, identify patients at risk of cardiac arrest or in need of high-dependency care and inform escalation to critical care services where available. This is particularly important in African settings, where decisions regarding oxygen therapy, monitoring capacity and early transfer are often made in Emergency Departments at the point of first contact with the health system.

We aimed to evaluate the effectiveness of EWSs in predicting in-hospital mortality and adverse events among adult patients admitted to general wards in LMICs, and to appraise the methodological quality and applicability of the underlying evidence.

Methods

This systematic review was conducted in accordance with PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020 and PRISMA-S guidelines, following a pre-registered protocol in the International Prospective Register of Systematic Reviews (PROSPERO) (https://www.crd.york.ac.uk/PROSPERO/view/CRD420251029273), to ensure methodological transparency and minimise the risk of bias [15,16].

Both randomised and non-randomised studies were included, and using the PICOS framework (Population, Intervention, Comparator, Outcomes, and Study design), we defined inclusion and exclusion criteria. Eligible studies involved adult populations — defined according to local context — admitted to general hospital wards in LMICs (World Bank classification) [17]. The intervention of interest was the use of any EWSs (e.g. National Early Warning Score (NEWS), Modified Early Warning Score (MEWS), VitalPAC Early Warning Score (ViEWS), among others). Comparators included standard care without EWSs or use of an alternative EWSs. Primary outcomes were short-term mortality — defined as death at 24 h, 3 days, 7 days, and 30 days. Secondary outcomes included cardiac arrest and unplanned ICU admission. Studies conducted exclusively in ICUs, Emergency Departments, or focused solely on paediatric or obstetric populations, were excluded. Studies conducted exclusively in intensive care units and emergency departments were excluded because the primary aim of this review was to evaluate prediction of deterioration in general medical and surgical wards, where failure-to-rescue most commonly occurs. Paediatric and obstetric populations were excluded due to the existence of specialised early warning systems — the Paediatric Early Warning Score (PEWS) and the Obstetric Early Warning Score (OEWS) — which would limit comparability. The LMIC filter was applied according to the World Bank classification to ensure consistency in health-system characteristics and comparability between included studies.

We conducted a comprehensive literature search in PubMed, Scopus, LILACS, and African Journals Online (AJOL) from database inception to 8 May 2025, without language restrictions. Grey literature was not systematically searched due to feasibility constraints. Additional citation searching was conducted by screening reference lists of all included studies to identify further eligible records, in accordance with PRISMA-S recommendations.

Search strategies were adapted for each database (Supplementary Appendix 1). We combined terms related to “early warning score systems” or “track and trigger systems” with keywords related to LMICs. Study selection was performed independently by two reviewers, with a third reviewer resolving discrepancies. References were imported into Rayyan (Rayyan Systems Inc., Montreal, Canada) [18]. Duplicates were removed using both automated and manual processes. Rayyan’s automation functionalities assisted with duplicate removal and screening. Titles and abstracts were screened; potentially relevant records proceeded to full-text review. Two reviewers independently extracted data, with discrepancies resolved by a third reviewer. Authors were not contacted for additional information. Data extraction included: study design; sample size; country; participant characteristics; type of EWS assessed; cut-off values; prediction windows; primary and secondary outcomes; funding source; missing data handling; and statistical performance metrics including AUC, sensitivity and specificity.

The planned effect measures listed in the PROSPERO protocol included diagnostic discrimination (AUC), sensitivity, specificity, positive and negative predictive values and likelihood ratios. Given the high level of clinical and methodological heterogeneity across included studies, we conducted a narrative synthesis in accordance with PRISMA-S recommendations. Planned sensitivity analyses were not performed due to inconsistent reporting of statistical estimates across included studies, which precluded meaningful stratification by risk-of-bias characteristics.

We did not perform a formal assessment of publication bias; however, methodological quality was assessed independently by two authors using the Prediction model Risk of Bias Assessment Tool (PROBAST), developed by the Cochrane Prognosis Methods Group, which comprises 23 signalling questions across four domains: participants, predictors, outcomes, and analysis. Each study was rated as having a low, high, or unclear risk of bias in each domain [19].

Application of GRADE was deemed not feasible due to the observational nature of most included studies and the lack of directly comparable effect measures.

The study used previously published data only; no human participants, identifiable data, or unpublished information were involved. No funding received.

Results

The search yielded 6086 records, with 28 studies ulitmately included (Fig. 1).

Fig. 1.

PRISMA flow diagram of study screening and selection.

The inclusion and exclusion criteria outlined in the Methods section were applied to identify eligible studies, and the results are summarised in Table 1.

Table 1.

Baseline characteristics of included studies and study populations.

S/N	Study Design	Citation	Mean or median age	Year of the data	Male/Female	Sample size	Country	EWSs	Inclusion criteria	Outcome evaluated
1	Prospective observational study	Nabayigga, 2016 [46]	47.2 and 69.4	2013–2014	386/397	783	Ugandan/ Denmark	NEWS	Patient aged >12 years; severely ill, and hospitalized in the ward; NEWS>6	Mortality at 7, 30, and 60 days
2	Prospective observational study	Asiimwe, 2015 [20]	34.5	2009; 2011	88/79	317	Ugandan	Restricted cubic splines; MEWS	Patient with sepsis	Mortality
3	Prospective observational study	Kellett, 2019 [21]	50.0	2018	772/ 1038	1 810	Ugandan	NEWS	Patients hospitalized for acute medical conditions	In-hospital mortality
4	Prospective observational study	Broekhoven, 2013 [40]	N/A	2011	N/A	180	Zimbabwe	EWS	Adult patients admitted to the medical and surgical wards with an EWS ≥ 3	Mortality or survival
5	Prospective observational study	Adebusoye, 2019 [38]	71.5	2013–2014	216/234	450	Nigeria	MEWS	Older patients aged 60 years and above who were admitted consecutively to the medical wards	All-cause mortality
6	Retrospective observational study	Wachira, 2015 [45]	59.3	2013	63/45	108	Kenya	MEWS	≥18 years; IHCA	Survival to hospital discharge
7	Cross-sectional study	Kumari, 2024 [44]	46.9	2020	75/41	116	Pakistan	NEWS2; PSI	Patients aged ≥ 12; admitted with CAP	In-hospital mortality.
8	Prospective observational study	Nakitende, 2020 [22]	48.4	2018–2019	583/683	1 266	Ugandan	NEWS2	patients hospitalized for acute medical conditions	In-hospital mortality within 7 days of admission.
9	Prospective observational study	Wheeler, 2013 [32]	39.5	2012	155/147	302	Malawi	HOTEL; MEWS	Patients aged > 18 years	In-hospital mortality within 3 days of admission.
10	Prospective observational study	Beane, 2018 [34]	42.7	2015	6640/ 9710	16 386	Sri Lanka	ViEWS; SEWS; MEWS; CART; AWTTS; SPTTS	Patients aged >17 years	In-hospital mortality, ICU admission, transfer to a tertiary care hospital, Cardiac arrest or CPR
11	Prospective observational study	Kayambankadzanja, 2019 [33]	39	2017–2018	446/648	1 094	Malawi	NEWS	≥18 years; Admitted to the Departments of Medicine, Surgery, Obstetrics and Gynecology, Ophthalmology, and Orthopedics	In-hospital mortality within 30 days of admission.
12	Retrospective observational study	Abbey, 2021 [42]	47	2017–2019	69/43	112	Ghana	MEWS and limited MEWS	Patients aged ≥13 years, admitted to the ward	In-hospital mortality
13	Prospective observational study	Kruisselbrink, 2016 [23]	40.5	2013	241/211	452	Ugandan	MEWS	Patients aged ≥18 years; from medical and surgical wards	In-hospital mortality within 7 days of admission.
14	Case-control study	Tan, 2022 [41]	Cases 63.56 Control 59.75	2018–2019	40/42	82	Philippines	MEWS and CART	Patients aged ≥19 years; admitted to the ward for at least 48 h	Cardiac arrest; Transfer to the ICU
15	Prospective observational study	Nakitende, 2018 [24]	N/A	2016–2018	N/A	2 240	Ugandan	NEWS	All alert acutely ill medical patients	In-hospital mortality
16	Prospective observational study	Bonnewell, 2021 [29]	43	2016–2019	297/300	597	Tanzania	MEWS; NEWS; qSOFA; SIRS; UVA	Patients aged ≥18 years	In-hospital mortality
17	Retrospective observational study	Hazard, 2022 [36]	38	2016–2017	37/63	573	Rwanda	MEWS; qSOFA; UVA	Adults ≥ 18 years with acute infection	In hospital mortality
18	Prospective observational study	Carugati, 2018 [30]	37.3 and 38.9	2007 - 2008	202/ 217	419	Tanzania	MEWS	Patient ≥ 10 years	In hospital mortality
19	Prospective observational study	Klinger, 2021[37]	35	2017	347/300	647	Rwanda	MEWS; qSOFA; UVA	Patient ≥ 15 years; Patients admitted to the ward with suspected infection	In hospital mortality
20	Prospective observation study	Lal, 2024 [47]	32	2018–2021	1113/ 684	2 797	Malawi Mozambique Zimbabwe	MEWS qSOFA UVA	Patient ≥ 15 years with fever	Mortality
21	Prospective observational study	Namujwiga, 2019 [25]	N/A	2016–2018	600/701	1 361	Ugandan	NEWS	Patients aged ≥20 years	In hospital mortality
22	Prospective observational study	Mar Minn, 2021 [43]	48	2019	237/ 272	509	Myanmar	NEWS2 qNEWS qSOFA SOFA, UVA	Patients ≥ 12 years with sepsis	Mortality
23	Prospective observational study	Sikakulya, 2024 [26]	55.9	2023–2024	155/247	402	Ugandan	NEWS	≥ 18 years	In hospital mortality
24	Prospective observational study	Opio, 2018 [27]	50.3	2016- 2017	366/ 533	899	Ugandan	MUAC MEWS	Patients ≥ 16 years critical ill	In-hospital mortality
25	Retrospective observational study	Davwar, 2023 [39]	46.7	2021	204/201	405	Nigeria	NEWS	Adults admitted to the medical and surgical wards	Mortality; Hospital discharge; Transfer to the HDU or ICU
26	Prospective observational study	Perera, 2011 [35]	49.4	2009	117/ 125	242	Sri Lanka	MEWS	Patients ≥ 15 years	Cardiorespiratory resuscitation; Mortality
27	Prospective observational study	Rylance, 2009 [31]	N/A	2005	243/466	1 235	Tanzania	MEWS	Patients ≥ 12 years	Mortality
28	Prospective observational study	Opio, 2013 [28]	45.2	2012	N/A	844	Ugandan	ViEWS	Patients ≥ 12 years old, hospitalized for acute medical conditions	In hospital mortality

N/A= Not Applicable; IHCA=In-Hospital Cardiac Arrest; CAP=community-acquired pneumonia; ICU=Intensive Care Unit; CPR=Cardiopulmonary resuscitation; EWS= Early Warning Score; NEWS=National Early Warning Score; MEWS= Modified Early Warning Score; PSI= Pneumonia Severity Index; HOTEL= (Hypotension, Oxygen saturation,lLow Temperature, ECG abnormalities, Loss of independence); ViEWS= VitalPAC Early Warning Score; SEWS= Standardised Early Warning Score; CART= Cardiac Arrest Risk Triage Score; qSOFA= quick Sequential Organ Failure Assessment; SIRS= Systemic Inflammatory Response Syndrome; UVA= Universal Vital Assessment; MUAC= Mid-upper arm circumference; HDU= High Dependency Unit.

AWTTS = Aggregate Weighted Track and Trigger Systems.

SPTTS = Single-Parameter Track and Trigger Systems.

Twenty-eight studies published from 2009 to 2024 met inclusion criteria; all were observational (22 prospective, 4 retrospective, 1 case-control and 1 cross-sectional), comprising a total of 36638 participants.

Most studies were conducted in Uganda (n = 9) [[20], [21], [22], [23], [24], [25], [26], [27], [28]], others were from Tanzania (n = 3) [[29], [30], [31]], Malawi (n = 2) [32,33], Sri Lanka (n = 2) [34,35], Rwanda (n = 2) [36,37] and Nigeria (n = 2) [38,39], with one study each from Zimbabwe [40], the Philippines [41], Ghana [42], Myanmar [43], Pakistan [44], and Kenya [45]. Multicountry studies included Uganda with Denmark [46] and a tri-nation study across Malawi, Mozambique, and Zimbabwe [47]. No eligible studies were identified from Central or South America, which limits geographical representativeness.

The most frequently applied EWSs were the MEWS (n = 15) and NEWS (n = 11). Additionally, five studies used the Universal Vital Assessment (UVA) score, and five evaluated the quick Sequential Organ Failure Assessment (qSOFA). Other scoring systems were reported in three or fewer studies.

Regarding the outcomes, most studies assessed in-hospital mortality (n = 26). Fewer reported unplanned ICU admission (n = 3), cardiopulmonary arrest or resuscitation (n = 3), or transfer to a tertiary care hospital (n = 1).

Using PROBAST, overall applicability to LMIC ward settings was high. Most studies [20,25,29,37,38,42,44,47] demonstrated consistently low risk of bias across all domains and high applicability, making them methodologically robust and highly relevant to the review question. However, other studies [[21], [22], [23], [24],26,28,30,[33], [34], [35], [36],[39], [40], [41],45,46] had high risk in one or more domains, especially participant selection and analysis, limiting reliability and generalisability. Full details are provided in Supplementary Appendix 2.

Despite heterogeneity, most studies reported a good predictive ability of MEWS for in-hospital mortality [23,35,42,45]. Routine calculation of MEWS on admission in severely ill elderly patients was associated with significant improvements in survival [38]. In one analysis involving 16,178 patients, MEWS, qSOFA, and UVA scores showed no significant difference in predicting hospital mortality [37]. However, a multicentre study found UVA score may outperform both MEWS and qSOFA in predicting mortality, suggesting value for early identification, triage, and clinical decision-making in resource-limited settings [26].

Several studies indicated that higher NEWS scores — typically ≥3 — are clinically significant predictors of in-hospital mortality [21,24,33,39,46]. Compared with the Pneumonia Severity Index (PSI) in community-acquired pneumonia, NEWS demonstrated higher sensitivity (but lower specificity) for predicting hospital mortality [44]. A study across hospitals in Uganda and Denmark found no statistically significant differences in 30-day mortality prediction by NEWS [46]. By contrast, a Malawian study evaluating MEWS and the HOTEL score suggested that EWSs in resource-limited environments need to be locally validated and assessed [32].

A case–control study comparing Cardiac Arrest Risk Triage (CART) score and MEWS for predicting ICU admission and cardiopulmonary arrest found CART to be more sensitive and specific, although differences were not statistically significant [41]. For composite adverse outcomes (high dependency/ ICU admission; need for cardiopulmonary resuscitation; length of hospital stay; and mortality), approximately 89.5 % of patients had MEWS score ≥5; however, MEWS did not predict hospital length of stay [26].

Discussion

This systematic review evaluated the efficacy of EWSs in predicting in-hospital mortality and adverse events among adult patients admitted to general wards in LMICs. Although EWSs are well established in high-income settings, their application in LMICs is comparatively recent and remains emergent. Differences in healthcare infrastructure, staffing, resource allocation, and clinical practice may significantly influence their effectiveness and generalisability [7]. For emergency medicine practitioners in Africa, these findings emphasise that EWSs can support triage and early escalation decisions at the point of first contact, particularly in resource-limited emergency units where critical care resources are scarce.

Across 28 observational studies MEWS and NEWS were the most frequently implemented tools in LMIC ward settings, generally showing moderate-to-good discrimination for in-hospital mortality [23,35,42,45]. Considerable heterogeneity in performance was evident, likely driven by differences in case mix, disease burden, and health-system capacity [26,33,38,39]. Higher scores (e.g. NEWS ≥3 or MEWS ≥5) were consistently associated with increased mortality risk, consistent with results reported in high-income settings [2,3]. The UVA score emerged as a potentially useful alternative in sub-Saharan Africa: its derivation and validation demonstrated robust discriminatory ability (AUC ≈ 0.77), compared with MEWS (AUC ≈ 0.70) and qSOFA (AUC ≈ 0.69). External validation in Rwanda showed similar predictive accuracy; dynamic reassessment at 6 h post-resuscitation also predicted in-hospital mortality in Uganda. However, its reliance on HIV testing and other laboratory data challenges direct comparability with scores based solely on vital signs, such as NEWS, and caution is warranted when interpreting claims of superiority without adequately powered, head-to-head multicentre studies, particularly against NEWS as the current industry standard [36,[48], [49], [50]].

Notwithstanding these promising findings, important challenges remain. Study designs varied considerably, with inconsistencies in outcome definitions, predictor selection, and data collection. Although several studies demonstrated low bias in outcome assessment and predictor definitions [20,25,29,37,38,42,44,47,51], others had limitations in participant selection or statistical methods [[21], [22], [23], [24],26,28,30,[33], [34], [35], [36],[39], [40], [41],45,46]. Many were single-centre, retrospective, or small, limiting external validity. Furthermore, most evidence was generated through observational studies and statistical simulations rather than pragmatic, real-world implementation trials, making causal inference and claims of superiority between tools inappropriate.

Most studies focused primarily on mortality, which is consistent with existing EWS research; however, fewer studies evaluated other outcomes relevant to emergency care, such as unplanned ICU admission, cardiopulmonary arrest, or rapid response activation [3,41,43]. Implementation factors — staff training, fidelity to protocols, workflow integration, and the availability of escalation pathways — were rarely reported, yet are essential for achieving clinical benefit [10,11]. Indeed, a rapid review suggests that EWSs and Rapid Response Systems may lead to little or no improvement in hospital mortality or adverse events when implementation is suboptimal or inconsistent [51,52]. Published literature from high-income settings describes integration of EWSs with electronic health records, automated monitoring alerts and digital dashboards. However, there is almost no research describing digital implementation of EWSs in African LMICs, representing a significant and actionable research gap.

Furthermore, EWSs were initially developed and validated in high-income contexts with continuous monitoring capabilities, access to laboratory support, and well-resourced escalation teams [1,52]. By contrast, many LMIC settings lack sufficient personnel, monitoring equipment, and intensive care capacity, so early identification of deterioration may not translate into improved outcomes without timely and appropriate intervention [7,53]. This raises concerns about the potential for implementation failure, even when predictive performance is adequate [13].

Taken together, our findings underscore the importance of validating and, where necessary, adapting EWSs for LMICs. This should include both recalibration of score thresholds causing local workflows, staffing models, and escalation capacity. The promising performance of tools like the UVA score [35,36] warrants further investigation in large, multicentre prospective studies and may serve as a model for developing context-specific early warning systems [35,36,48].

All included studies were observational, limiting causal inference between EWSs implementation and outcomes. Heterogeneity in score types, thresholds, settings, and outcome hindered comparability. Although the literature search was comprehensive, relevant grey literature from LMICs may have been missed. Variability in data quality and risk of bias in areas like participant selection and analysis may have affected interpretation. We did not undertake sensitivity analyses despite variability and high risk of bias in some included studies, owing to lack of consistent statistical reporting across studies. The limited number of studies assessing outcomes beyond mortality restricts evaluation of broader clinical and economic impact of EWSs. Finally, the geographical imbalance restricts generalisability to Latin American LMICs, which may have differing patterns of disease burden, emergency care infrastructure and escalation pathways.”

Conclusion

This systematic review demonstrates that EWSs—particularly NEWS and MEWS—provide useful predictive information for in-hospital mortality in LMIC general-ward settings, but evidence remains predominantly observational and context-dependent. No tool should be assumed superior without adequately powered, head-to-head, multicentre implementation studies. Rigorous local validation, recalibration and system-level readiness are prerequisites for safe, effective and equitable scale-up. Future research should prioritise large, prospective, multicentre implementation studies in LMICs, evaluating the real-world effectiveness of EWSs. Efforts should focus on validating or adapting existing scoring to reflect local healthcare conditions, including workforce limitations, infrastructure challenges, and diverse patient populations. Integration of EWSs into digital health solutions and routine workflows — particularly for non-physician providers — warrants investigation. Studies should extend their focus beyond mortality to include patient-centred outcomes, time-to-intervention, and cost-effectiveness, to better inform national policies and support sustainable health system improvements.

Author contributions (CRediT)

ET: Conceptualisation, Methodology, Data Curation, Formal Analysis, Visualization, Writing – Original Draft, Project Administration; CA, AJ: Conceptualisation, Writing – Review & Editing, Supervision; CP: Conceptualisation, Formal Analysis, Visualisation, Writing – Original Draft; EU: Conceptualisation, Data Curation, Writing – Review & Editing; MLA: Writing – Review & Editing. All authors approved the final version for publication and agree to be accountable for all aspects of the work.

Dissemination of results

Findings from this research were shared with clinical and managerial staff at participating hospitals through an onsite seminar and ward-based feedback sessions. Key messages were also presented at a regional patient-safety meeting to inform policy and practice across comparable settings.

Declaration of competing interest

The authors declared no conflicts of interest.