Impact of clinical decision support systems (CDSS) on clinical outcomes and healthcare delivery in low- and middle-income countries: protocol for a systematic review and meta-analysis

Abstract

Introduction

Clinical decision support systems (CDSS) are used to improve clinical and service outcomes, yet evidence from low- and middle-income countries (LMICs) is dispersed. This protocol outlines methods to quantify the impact of CDSS on patient and healthcare delivery outcomes in LMICs.

Methods and analysis

We will include comparative quantitative designs (randomised trials, controlled before–after, interrupted time series, comparative cohorts) evaluating CDSS in World Bank-defined LMICs. Standalone qualitative studies are excluded; mixed-methods studies are eligible only if they report comparative quantitative outcomes, for which we will extract the quantitative component. Searches (from inception to 30 September 2024) will cover MEDLINE, Embase, CINAHL, CENTRAL, Web of Science, Global Health, Scopus, IEEE Xplore, LILACS, African Index Medicus and IndMED, plus grey sources. Screening and extraction will be performed in duplicate. Risk of bias will be assessed with Risk of Bias 2 (randomised trials) and Risk Of Bias In Non-randomised Studies—of Interventions (ROBINS-I) (non-randomised). Random-effects meta-analysis will be performed where outcomes are conceptually/statistically comparable; otherwise, a structured narrative synthesis will be presented. Heterogeneity will be explored using I²/τ² and a priori subgroups/meta-regression (condition area, care level, CDSS type, readiness proxies, study design).

Ethics and dissemination

This review uses published data and does not require ethics approval. Results will be disseminated via peer-reviewed publication, conference presentations and LMIC-oriented policy briefs; extraction templates and analysis code will be shared openly on publication.

PROSPERO registration number

CRD42024599329.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• The protocol is registered (PROSPERO CRD42024599329) and follows Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA)/PRISMA Protocols (PRISMA-P) guidance.

• We use a multi-database search (biomedical+technical sources) with reference list screening to maximise retrieval.

• Dual independent screening and data extraction are planned, with design-appropriate risk-of-bias appraisal (Risk of Bias 2 for randomised controlled trials; Joanna Briggs Institute (JBI) checklists for observational designs).

• A pre-specified quantitative synthesis plan uses random-effects meta-analysis where outcomes are comparable, with subgroup and sensitivity analyses and narrative synthesis otherwise.

• Restricting to English full texts and the last 10 years may introduce language/time window bias and under-represent low- and middle-income country evidence.

Background

Clinical decision support systems (CDSS) are advanced technological tools designed to enhance healthcare decision-making by integrating clinical knowledge with patient data to provide personalised recommendations. CDSS have been widely implemented to aid healthcare providers in clinical decision-making, leveraging digital technologies to improve diagnostic accuracy, guide treatment decisions and optimise healthcare outcomes.¹,³ These are an integral part of health information technologies and are often part of electronic health record (EHR) systems to provide real-time, evidence-based recommendations to clinicians during patient care.⁴

The origin of CDSS dates back to the late 1960s when the first rule-based systems, such as MYCIN, were developed to assist in diagnosing bacterial infections and suggesting appropriate treatments.⁵ Over time, these systems have evolved considerably, incorporating more sophisticated technologies, including artificial intelligence (AI), machine learning and natural language processing. Modern CDSS can analyse vast and complex datasets, enabling healthcare providers to make more informed decisions, whether it be diagnosing a disease, selecting the most appropriate treatment plan or predicting patient outcomes based on clinical history and available medical data.

CDSS encompasses a variety of tools, including computerised alerts and reminders, clinical guidelines, condition-specific order sets, focused patient data reports, diagnostic support and knowledge-based systems that guide treatment decisions. These systems help address common issues in healthcare delivery, such as the need for timely decision-making, the overload of clinical data and variability in clinical practice. CDSS can positively influence patient-level outcomes such as reducing diagnostic errors, improving medication safety and enhancing treatment effectiveness, while also improving healthcare delivery outcomes like patient throughput, resource allocation and overall care quality.¹⁵,⁸

CDSS applications are diverse; for example, in the management of chronic diseases like diabetes, CDSS can suggest appropriate interventions based on blood glucose levels and other clinical parameters. In oncology, CDSS assists in treatment planning by integrating genomic data to recommend targeted therapies. In emergency care, CDSS aids in triage systems, helping clinicians quickly assess patients based on their symptoms and history, ensuring rapid response and intervention. In mental health, CDSS can suggest probable diagnoses and treatment options based on patient-reported outcomes and clinical data, allowing for personalised care strategies.

CDSS has a potential to transform healthcare. By integrating knowledge from clinical guidelines, research and patient data, these systems provide decision support that is not only evidence based but tailored to individual patient needs also. For example, in acute stroke management, it can predict patient outcomes and guide interventions, which can lead to improved survival rates and reduced healthcare costs.⁷ Moreover, the dynamic decision-making capabilities of CDSS where the system can update itself with new data and guidelines allow for continuous improvements in diagnostic accuracy and treatment efficacy, making it an invaluable tool in evolving clinical environments.

There are few challenges in the application of CDSS. For example, with frequent system alerts, clinicians can become desensitised and this can reduce the effectiveness of the system. CDSS must be carefully integrated into clinical workflows to avoid disruptions in medical care. Sometimes, complex interfaces can lead to user frustration and hinder clinical adoption. Interoperability with existing EHRs and healthcare systems is essential to ensure seamless data exchange and system functionality. For successful implementation of CDSS, alignment with local clinical guidelines and healthcare infrastructure is important.

In high-income countries, CDSS are popular owing to their impact on improving healthcare delivery. AI-driven CDSS systems have shown improved diagnostic accuracy and decision-making in fields such as cardiology, oncology and emergency care. These systems also reduce the cognitive load on clinicians by providing automated, evidence-based recommendations, allowing for more consistent and timely care, reducing medication errors, decreasing unnecessary diagnostic testing and enhancing overall patient safety and care. CDSS in low- and middle-income countries (LMICs) have potential to improve clinical outcomes. Electronic decision algorithms, such as e-ALMANACH, have been used to manage febrile illnesses in children, resulting in improved clinical outcomes and more rational use of antibiotics. These systems can guide healthcare workers through decision-making processes, reducing clinical failure rates and unnecessary treatments. However, despite these benefits, the use of CDSS in LMICs remains underutilised, largely due to a lack of awareness, training and infrastructure.

Implementation of CDSS in LMICs is challenging due to poor infrastructure, scarce healthcare personnel and resource constraints.⁹ There are barriers in adopting CDSS like limited access to digital health infrastructure, lack of technical expertise and financial constraints.¹⁰ It should be noted that most of CDSS are developed considering their use in high-resource settings, making them less applicable to the healthcare realities in LMICs, where diagnostic tools and clinical guidelines may differ significantly.

Most of the existing research focuses on high-income countries, leaving a critical need for evaluating the effectiveness of CDSS in resource-poor nations. Understanding the barriers in their implementation can help to strengthen their applications in these countries. Also, evidence on the scalability and sustainability of CDSS in these regions is limited, highlighting the need for more research to explore how these systems can be adapted to fit the needs of LMICs. Current Systematic Review is planned to understand the effect of using CDSS in LMICs and to bridge the gap in the evidence-based clinical management especially in LMICs and possible measures that can be considered for their use in the best possible ways.

Methods/design

We will follow the PRISMA guidelines. The protocol has been registered with PROSPERO (CRD42024599329). Studies will be identified through searches of the following databases: PubMed, Embase, Cochrane Library, Medline, IEEE Explore and Google Scholar.

Study dates

Searches will cover database inception to 30 June 2025; screening/extraction will occur July–September 2025 with analysis and submission of results targeted for Q1 2026.

Eligibility criteria and study selection are shown in table 1.

Table 1. Summary of eligibility criteria.

Category	Inclusion criteria	Exclusion criteria
Population	Patients, healthcare personnel, or healthcare systems in low- and middle-income countries	Studies focusing only on high-income countries
Intervention	Implementation or use of clinical decision support systems (CDSS) using digital technology, including artificial intelligence-based tools for healthcare delivery and decision support	Studies focused on the development of CDSS models without real-world implementation, or those limited to framework proposals
Comparator	Healthcare systems using CDSS compared with systems without CDSS, or traditional decision-making processes without digital support	Absence of a control group or inappropriate comparator
Outcome	Primary: patient outcomes (diagnostic accuracy, treatment adherence, improved clinical outcomes) Secondary: healthcare delivery outcomes (efficiency, workflow optimisation, quality of care)	Studies using non-validated methods to measure the outcomes, or lacking patient-level or healthcare workflow outcomes
Study design	Randomised controlled trials (RCTs), cohort studies, prospective interventional studies	Non-RCTs, qualitative studies, reviews, commentaries and non-peer-reviewed articles

Inclusion criteria

• Studies involving digital technology (including, but not limited to AI) as the intervention.

• Knowledge-based and non-knowledge-based CDSS.

• Cohort, cross-sectional, prospective observational, interventional studies and randomised control trials.

• Language: studies must be published in English. Records in any language will be searched and screened at title/abstract level; however, only English full texts will be assessed for inclusion. We will record the number of non-English full texts excluded at eligibility and summarise their key characteristics (eg, country, condition area) when available from the abstract. Searches will be executed without language limits to map potentially relevant non-English literature; eligibility will apply the English-only full-text criterion.

• Date of publication: search will cover articles published in the last ten years in the databases up to 30^th September 2024.

Exclusion criteria

• Studies limited to developing/proposing models or frameworks.

• Studies not reporting patient-level outcomes or healthcare workflow outcomes.

• Non-peer-reviewed or qualitative reviews, viewpoints, editorials, abstracts, conference papers, treatment guidelines and prevalence studies. Standalone qualitative studies are excluded; mixed-methods studies are eligible only if they report comparative quantitative outcomes, in which case we will use the quantitative component.

Population (P)

Patients, healthcare personnel or healthcare systems in LMICs¹¹ (online supplemental table 1).

Intervention (I)

Implementation or use of CDSS that consist of at least one digital technology (including, but not limited to artificial intelligence).

Comparison (C)

Healthcare systems with CDSS versus those without CDSS or traditional decision-making processes.

Outcomes (O)

Primary outcome

• Patient outcomes (eg, diagnostic accuracy, time to diagnosis, treatment decisions, adherence to treatment guidelines, self-management, mortality reduction, etc.).

• Healthcare workflow outcomes (eg, efficiency, quality, burden, delivery outcomes).

Search Strategy

The search will be performed using a combination of index terms, including “clinical decision support system”, “CDSS”, and “healthcare delivery” across the databases mentioned above (online supplemental table 1). Additional studies will be identified by examining reference lists of reviews and eligible publications.

Risk of bias assessment

To evaluate the risk of bias in included studies, pairs of reviewers will independently assess each study using design-appropriate, current tools.

For randomised controlled trials (RCTs), we will use the Cochrane Risk of Bias 2 (RoB 2) tool, which assesses bias across five domains: randomisation process, deviations from intended interventions, missing outcome data, outcome measurement and selection of the reported result. Both reviewers will apply RoB 2 at the outcome level following the official guidance; disagreements will be resolved by discussion or a third reviewer.¹² For non-randomised comparative studies (eg, cohort, case-control, controlled before–after, interrupted time series), we will use the Risk Of Bias In Non-randomised Studies—of Interventions (ROBINS-I) tool, which evaluates bias due to confounding, participant selection, intervention classification, deviations from intended interventions, missing data, outcome measurement and selective reporting.

Where available, we will provide visual summaries (traffic light plots) of risk-of-bias judgments. Sensitivity analyses will exclude studies judged to be at high or critical risk of bias to assess the robustness of pooled estimates.¹³ In cases of disagreement between the reviewers, discussions will be held to reach a consensus, with input from a third reviewer if required. The results of the risk-of-bias assessments, as well as the extracted data, will be entered into a standardised sheet for further synthesis and statistical analysis.

Data extraction

Data extraction will be independently conducted by two reviewers. Discrepancies will be resolved through discussion. For our study, we will use a standardised data abstraction form accompanied by a comprehensive instruction manual. To ensure consistency and accuracy in data extraction, calibration exercises will be conducted prior to the extraction process. Two reviewers, working in pairs, will independently extract the following information from the eligible studies: (1) study characteristics, including the names of authors, publication year, country of origin and funding sources; (2) population-related details such as the number of participants, average age, gender distribution, the type of clinical condition or disease studied and duration of follow-up; (3) intervention and comparator details, such as the type of intervention, mode of delivery, duration of treatment and timing of interventions; and (4) the outcomes of interest related to CDSS, including clinical outcomes, healthcare delivery outcomes and any adverse events reported. A meta-analysis will be conducted if sufficient homogeneous data are available; otherwise, a narrative synthesis will be performed.

Data synthesis

Data synthesis will follow a structured approach to combine and analyse findings from the included studies. We will first perform a narrative synthesis for all studies, summarising the characteristics, interventions, outcomes and key findings of each study. This will provide a broad overview of the evidence.

For quantitative data, we will perform a meta-analysis if the studies are sufficiently homogeneous in terms of population, interventions and outcomes. This will involve pooling effect sizes using a random-effects model, which accounts for variations between studies. For dichotomous outcomes, such as clinical outcomes or adverse events, we will calculate relative risks with 95% CI. For continuous outcomes, such as treatment effects or health improvements, we will use mean differences (MD) or standardised MD with 95% CI depending on the nature of the data.

Heterogeneity among studies will be assessed using the Cochran’s Q test and I² statistics, with significant heterogeneity explored further through subgroup analyses and sensitivity analyses. Subgroup analyses may include factors such as population characteristics, setting (eg, rural vs urban) and type of intervention (eg, provider-led vs self-managed interventions). Sensitivity analyses will be conducted to test the robustness of the findings, particularly by excluding studies with a high risk of bias or outlier results.

Where meta-analysis is not possible due to significant heterogeneity or lack of sufficient comparable data, we will present a descriptive synthesis. In such cases, we will synthesise results narratively, emphasising trends and patterns in the data, and using tables and figures for visual representation. Additionally, we will use tools such as forest plots to illustrate effect sizes and funnel plots to assess potential publication bias.

By using these techniques, our data synthesis will provide a comprehensive evaluation of the impact of CDSS in the specific settings of LMICs.

Assessing the quality of evidence

We will use the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) (table 2) framework to systematically assess the quality of evidence in our study.¹⁴ GRADE will help evaluate the certainty of the evidence across different domains including study design, risk of bias, consistency, directness and precision. Evidence from RCTs will initially be considered as high quality, but may be downgraded based on identified limitations such as high risk of bias, inconsistency in results or imprecision. For non-randomised studies, the starting certainty will be lower, and we will assess the risk of bias using tools like the ROBINS-I. The GRADE framework will also enable us to consider the overall strength of recommendations, particularly in resource-limited settings where evidence might be sparse. By using GRADE, we ensure that the conclusions drawn from our review are transparent and based on the best available evidence, providing clear guidance for both clinical practice and policy making.

Table 2. GRADE evidence profile of the impact of clinical decision support systems on healthcare outcomes in low- and middle-income countries.

Outcome	No. of studies (design)	Risk of bias	Inconsistency	Indirectness	Imprecision	Publication bias	Effect size (95% CI)	Overall certainty of evidence
Patient outcomes (eg, diagnostic accuracy, treatment adherence)	X (RCTs, cohort studies)	X	X	X	X	X	X	X
Healthcare workflow efficiency	X (RCTs, cohort studies)	X	X	X	X	X	X	X
Mortality reduction	X (RCTs, cohort studies)	X	X	X	X	X	X	X
Quality of care improvement	X (RCTs, cohort studies)	X	X	X	X	X	X	X
Adverse events (eg, decision-making errors)	X (RCTs, cohort studies)	X	X	X	X	X	X	X

GRADE, Grading of Recommendations, Assessment, Development and Evaluation; RCTs, randomised controlled trial.

Patient and public involvement

Patients or the public were not involved in the design, conduct, reporting or dissemination plans of this research.

Ethics and dissemination

This study involves secondary analysis of published data and does not require ethics approval. Findings will be disseminated via journal publication, conference presentations and LMIC policy briefs; upon publication, de-identified extraction sheets and analytic code will be made available in an open repository.

Discussion

CDSS have become an integral part of modern healthcare, particularly in high-income countries where digital infrastructure is more robust. The potential of CDSS to improve clinical outcomes, enhance decision-making and streamline healthcare workflows is well documented. However, the evidence supporting its implementation in LMICs is limited. This systematic review aims to address this gap by assessing the impact of CDSS in resource-poor settings and providing insights into its effectiveness, scalability and adaptability.

The review aims to evaluate the impact of CDSS on healthcare systems in resource-poor settings, potentially guiding future implementations and helping healthcare providers and policy makers develop strategies tailored to the unique challenges of LMICs. Our study offers several strengths compared with existing reviews. First, we will comprehensively evaluate the comparative effectiveness of CDSS in a variety of healthcare settings, particularly those in resource-constrained environments. This focus on LMICs will provide valuable insights into how these systems can be adapted to fit local healthcare needs. Second, by using the GRADE approach, we will systematically assess the quality of evidence across all included studies, ensuring that the conclusions drawn are based on the highest possible quality of evidence. This will enhance the reliability of our findings and their applicability in guiding healthcare policies and decision-making processes in LMICs.

The potential impact and strengths of this review extend beyond providing evidence on CDSS implementation. By synthesising the existing data, we aim to highlight the areas where CDSS can significantly contribute to improving healthcare outcomes, particularly in managing chronic diseases, enhancing diagnostic accuracy and supporting clinical workflows. Anticipated challenges and limitations of this study include the availability and diversity of the studies needed to perform robust data synthesis. The limited number of high-quality studies focusing on CDSS in LMICs might pose a constraint. Additionally, the heterogeneity of healthcare settings, infrastructure and population characteristics across LMICs may limit the generalisability of the findings. These factors will need to be carefully addressed through subgroup analyses and sensitivity analyses during data synthesis.

Despite these challenges, this review has the potential to significantly contribute to the understanding of CDSS in resource-limited settings. By highlighting both the benefits and the barriers to implementation, this study will provide crucial insights for future research and policy development. Moreover, the findings will be disseminated widely through peer-reviewed publications and presentations at national and international conferences to ensure broad accessibility and impact, thereby helping bridge the knowledge gap on CDSS in LMICs and guiding future efforts to optimise healthcare delivery in these regions.

We used the PRISMA-P checklist when writing our report.¹⁵