Exploring the impact of artificial intelligence integration on medication error reduction: A nursing perspective

Abstract

Aim

To systematically evaluate the impact of artificial intelligence (AI) technologies on reducing medication errors in nursing practice, focusing on tools such as clinical decision support systems (CDSS), smart infusion pumps, barcode scanning and automated prescription validation.

Background

Medication errors are a persistent threat to patient safety and a major burden on healthcare systems. Nurses, who are central to the medication administration process, remain vulnerable to human error. AI offers new opportunities to enhance safety through real-time decision support and predictive analytics.

Design

A systematic review following PRISMA 2020 guidelines and using a mixed-methods approach to integrate quantitative outcomes with qualitative insights from nursing practice.

Methods

Studies published in English between January 2013 and March 2024 were retrieved from PubMed, ScienceDirect and CINAHL. Eligibility was guided by the PICO framework. Quality appraisal tools appropriate to study designs were applied.

Results

Twelve studies were included. CDSS reduced operating room errors by up to 95 %, while smart infusion pumps reduced IV medication errors by approximately 80 %. Prescription validation tools led to a 55 % reduction in prescribing errors. AI-driven alert filtering decreased non-actionable alerts by 45 %. Qualitative data revealed both appreciation of AI’s utility and concerns about algorithmic bias, system usability and trust.

Conclusions

AI technologies significantly improve medication safety in nursing. However, successful implementation depends on nurse training, system integration, ethical safeguards and workflow alignment. Further experimental studies are needed to validate efficacy and address barriers such as alert fatigue, algorithm transparency and adoption resistance.

1 Introduction

Medication errors have been shown to be one of the most persistent problems in nursing and their negative impact on patient safety, associated hospitalization costs, morbidity and mortality are major concerns. Any preventable event that may lead to inappropriate use of medication and/or patient harm can be classified as errors, which can occur at any stage of the medication process, including prescribing, transcribing, dispensing, providing and monitoring ( ^{Bonnabry et al., 2023}). Despite scientific advances, fewer than 40 % of infusions have zero errors and studies indicate that IV medication administration errors continue to be a common occurrence ( ^{Leape et al., 2024}).

Errors made in nursing workflow can be mitigated using artificial intelligence (AI), which includes automation, predictive analytics and real-time decision support for modeling nursing workflows. Clinical Decision Support Systems (CDSS) that use Artificial Intelligence (AI) embedded in Electronic Health Records (EHRs) inform nurses in real time about high-risk drug interactions and incorrect dosing ( ^{Classen et al., 2024}). For example, the case of Massachusetts General Hospital led to AI-based CDSS providing immediate alerts on high-risk prescriptions, helping to prevent 4500 adverse medication events per year ( ^{Nanji et al., 2024}). Smart infusion pumps featuring dose error reduction software (DERS) have also been implemented in many hospitals, proving to reduce errors during intravenous medication by nearly 80 % ( ^{Leape et al., 2024}). This potential for using AI to enhance efficiency and reduce patient harm is underscored by these findings.

With increasing patient loads, complex pharmacological treatments and a shortage of nursing staff, the application of AI in medication safety is becoming increasingly relevant. Medication errors are characterized by the Institute of Medicine (IOM) ( ^{Classen et al., 2024}) as a leading cause of preventable harm, contributing to approximately 70,000 patient deaths annually in the U.S. alone. However, traditional manual medication verification processes, which are essential, are susceptible to human error, particularly in high-risk areas such as the ICU and ED, where nurses must make rapid decisions with significant consequences. AI-powered prescription validation algorithms can identify potential prescribing errors and achieve up to 55 percent fewer misadministrations when prescriptions reach patients ( ^{Harrington et al., 2024}).

Moreover, integrating AI is essential to address alert fatigue, which is a significant concern for nurses receiving excessive medication warnings from traditional CDSS. Clinicians dismiss 49–96 % of alerts, including 86 % of clinically relevant ones due to alert desensitization ( ^{Classen et al., 2024}). ^{Harrington et al. (2024)} demonstrate that AI-driven alert filtering systems reduce non-actionable alerts by 45 %, allowing nurses more time to respond to critical alerts.

From a nursing perspective, this study systematically examines the impact of artificial intelligence (AI) on reducing medication errors in clinical settings. The focus is on how AI technologies are integrated into key phases of the medication process, including prescribing, administration and monitoring and how these applications support nursing practice.

The research is guided by a question structured using the PICO framework. Based on this framework, the study addresses two core questions:

1. How effective are AI-based tools in reducing medication errors in nursing practice compared with traditional methods?

2. What factors, including usability, staff training, ethical concerns and system integration, influence the implementation and acceptance of AI in nursing medication management?

The study incorporates both quantitative and qualitative research designs under a mixed-methods approach to examine AI’s effectiveness in minimizing medication errors. The quantitative descriptive method provides statistical analyses of AI error reduction results alongside qualitative case study analysis investigating nurses’ encounters with AI implementation. This research study merges various methodologies to deliver quantitative data along with qualitative details, which guarantees comprehensive knowledge regarding AI's effects on nursing practice. Research reliability can be strengthened by using experimental as well as quasi-experimental methods to validate the current results. Specifically, this study will culminate with a focus on three main applications of AI in medication safety:

• • Prescribing: Ensuring valid prescriptions align with patient-specific factors like individual characteristics, drug allergies and renal function. For instance, in oncology, AI-assisted systems like IBM Watson for Oncology have significantly enhanced the accuracy of chemotherapy dosing, enabling oncologists to administer the correct amounts of chemotherapy while minimizing dosing-related complications for cancer patients ( ^{Nanji et al., 2024}).
• • Administration: AI-driven barcode scanning and robotics improve medication verification accuracy and reduce human error. Automated dispensing cabinets in hospitals have demonstrated a 36 % reduction in opioid-related medication errors, particularly in high-risk areas such as postoperative recovery wards ( ^{Leape et al., 2024}).
• • Monitoring: Using real-time pharmacovigilance tools with AI integration helps monitor patient medication responses to identify early signs of adverse drug reactions (ADRs). A study conducted in ICU settings showed that AI-driven models successfully detected early-stage antibiotic-associated nephrotoxicity, resulting in a 27 % reduction in serious complications ( ^{Nanji et al., 2024}).

Yet, while AI brings numerous benefits, harmonizing it is not an easy task. Another critical issue that raises questions about the interpretation of AI-recommended results is the concern regarding the reliability of the outputs. A survey of 500 nurses in four tertiary hospitals in Canada indicated that 62 % of the respondents were unsure about relying on AI-powered medication verification due to concerns about algorithmic bias and a lack of explainability ( ^{Classen et al., 2024}).

2 Methods

This study employed a systematic literature review design, aligned with the PRISMA 2020 (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) statement. The objective was to evaluate the effectiveness of artificial intelligence (AI)-based interventions in reducing medication errors in nursing practice, specifically across the domains of medication prescribing, administration and monitoring. This review adopted a mixed-methods design, integrating quantitative descriptive analyses and qualitative thematic synthesis to explore both the measurable impact of AI on medication error reduction and the experiential insights from nursing professionals.

The methodology was structured to ensure transparency, reproducibility and methodological rigor by explicitly defining the research question, systematically searching and selecting relevant studies and applying standardized tools for data extraction and quality appraisal. The methodology adhered to established guidelines for transparent reporting, including the PRISMA 2020 and, where relevant, the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) statement for observational studies.

2.1 Research Question and PICO Framework

The central research question guiding this review was: "In hospital-based nursing practice, how effective are AI-powered tools compared with conventional systems in reducing medication errors during prescribing, administration and monitoring phases?"

To structure this question and inform the review protocol, the PICO framework was used:

• • Population (P): Registered nurses and nurse practitioners involved in medication management in hospital or acute care settings.
• • Intervention (I): Artificial intelligence (AI)-based technologies including, but not limited to, clinical decision support systems (CDSS), smart infusion pumps, automated dispensing cabinets, barcode medication administration (BCMA) and AI-enabled pharmacovigilance platforms.
• • Comparison (C): Traditional/manual medication processes without AI support, such as standard electronic prescribing or human-dependent transcription, administration and monitoring protocols.
• • Outcome (O): Reduction in the incidence, frequency, or severity of medication errors (including prescribing errors, administration errors, dose errors and monitoring failures). Secondary outcomes included nurse acceptance, workflow efficiency, alert fatigue and patient safety metrics.

This framework formed the basis for developing the eligibility criteria, search strategy and data synthesis plan.

2.2 Search Strategy

A comprehensive and structured search strategy was developed to identify studies assessing the impact of artificial intelligence (AI) tools on reducing medication errors in nursing practice. The search was conducted across the following major academic databases:

• • PubMed (MEDLINE)
• • ScienceDirect
• • CINAHL (Cumulative Index to Nursing and Allied Health Literature)

The search was limited to studies published between 1 January 2013 and 31 March 2024, to ensure inclusion of the most current evidence following widespread clinical adoption of AI technologies in healthcare.

Peer-reviewed primary research articles published in the English language were included in this review. In addition to peer-reviewed articles, selected secondary data sources such as hospital-based case reports and government health reports (e.g., from the Agency for Healthcare Research and Quality) were reviewed where they contained empirical findings aligned with the eligibility criteria. These were only included if they met standards of methodological rigor and transparency. Additionally, narrative reviews, editorials and opinion pieces were excluded unless they provided empirical findings with clear methodology.

To capture relevant studies, a combination of Medical Subject Headings (MeSH) and free-text keywords was used. Boolean operators (AND, OR) were employed to expand and refine the search ( ^{Table 1}). An example of a search string used in PubMed is as follows:

("Artificial Intelligence"[MeSH Terms] OR "Machine Learning"[MeSH Terms] OR "Clinical Decision Support Systems" OR "Smart Infusion Pumps" OR "Automated Dispensing" OR "Barcode Scanning" OR "Pharmacovigilance") AND ("Medication Errors"[MeSH Terms] OR "Drug Administration Errors" OR "Prescribing Errors" OR "Medication Safety") AND ("Nursing"[MeSH Terms] OR "Nursing Practice" OR "Nurses") AND ("Hospitals"[MeSH Terms] OR "Clinical Settings" OR "Acute Care")

Search strategies were adapted to the syntax and indexing of each database. Additionally, the reference lists of included studies and relevant systematic reviews were manually screened to identify any studies that may have been missed during the database search.

All retrieved records were imported into Zotero reference management software, where duplicates were removed automatically. The remaining citations underwent two-stage screening: (1) title and abstract review, followed by (2) full-text evaluation against the pre-established eligibility criteria.

The full study selection process, including the number of records identified, screened, included and excluded, is illustrated in the PRISMA flow diagram ( ^{Fig. 1}).

2.3 Eligibility Criteria

The eligibility criteria were developed based on the PICO framework and applied consistently throughout the review process to determine the inclusion or exclusion of studies.

Inclusion Criteria:

• • Population: Studies involving registered nurses, nurse practitioners, or nursing workflows directly related to medication administration in hospital-based or acute care settings.
• • Intervention: Implementation of AI-based tools such as clinical decision support systems (CDSS), smart infusion pumps, barcode medication administration (BCMA), automated dispensing systems, or AI-driven pharmacovigilance platforms.
• • Outcome: Studies that quantitatively or qualitatively reported medication error rates or safety outcomes related to prescribing, administration, or monitoring of medications.
• • Study Types: Primary research articles, including randomized controlled trials (RCTs), cohort studies, case-control studies, cross-sectional studies and qualitative studies.
• • Publication Language: English.
• • Publication Type: Peer-reviewed journal articles.
• • Publication Date Range: January 2013 to March 2024.

Exclusion Criteria:

• • Studies unrelated to nursing practice or not involving nurses as participants or end-users.
• • Studies focusing solely on education, simulation, or AI applications outside the medication administration process.
• • Articles that were not peer-reviewed, including conference abstracts, white papers, dissertations, editorials, opinion pieces, or grey literature.
• • Reviews lacking a systematic methodology (e.g., narrative reviews without structured data).

This rigorous set of criteria ensured that only methodologically sound, relevant and current studies were included in the final synthesis.

2.4 Study Selection

The study selection process followed a two-stage screening protocol aligned with PRISMA guidelines.

1. Title and Abstract Screening: Two independent reviewers screened the titles and abstracts of all retrieved records to assess preliminary relevance based on the eligibility criteria.
2. Full-Text Review: Full-text articles were retrieved for all studies deemed potentially eligible. These were then reviewed in detail to determine final inclusion. Any discrepancies between reviewers were resolved through discussion or, when necessary, consultation with a third reviewer.

Studies excluded at this stage were recorded along with specific reasons for exclusion (e.g., not involving nurses, AI not used in medication safety, duplicate data).

The study selection workflow, including the number of studies screened, excluded and included, is presented in the PRISMA flow diagram ( ^{Fig. 1}) in the Results section.

2.5 Data Extraction

A standardized data extraction form was developed and pilot-tested to ensure consistency and completeness of data collection. Two reviewers independently extracted data from each included study. Extracted information included the following:

• • Study identification: First author, year of publication and country
• • Study design and setting: Study type (e.g., RCT, qualitative), healthcare environment (e.g., ICU, general ward)
• • Sample characteristics: Number and roles of nurse participants or nursing units involved
• • AI intervention details: Type of AI tool used (e.g., CDSS, smart pump), integration with hospital systems (e.g., EHR)
• • Medication process phase: Prescribing, administration, or monitoring
• • Outcomes reported: Type and frequency of medication errors, clinical impact, nurse perceptions, workflow changes, implementation challenges
• • Main findings: Summary of results including quantified error reduction or thematic insights
• • Funding and conflicts of interest (if reported)

Disagreements during data extraction were resolved by consensus. If key data were missing or unclear, efforts were made to contact study authors for clarification.

2.6 Quality Appraisal

To ensure the credibility and methodological rigor of the included studies, a formal quality appraisal process was conducted. Each study was independently assessed by two reviewers using appraisal tools appropriate to its design:

• • Randomized Controlled Trials (RCTs): Evaluated using the Cochrane Risk of Bias (RoB 2.0) tool, assessing domains such as randomization process, deviations from intended interventions and outcome measurement.
• • Observational Studies (e.g., cohort, cross-sectional, case-control): Appraised using the Critical Appraisal Skills Programme (CASP) checklists, focusing on sample selection, confounding control and outcome validity.
• • Qualitative Studies: Assessed using the CASP Qualitative Checklist, covering credibility, relevance, data collection, researcher reflexivity and ethical considerations.
• • Mixed-Methods Studies: Evaluated using the Joanna Briggs Institute (JBI) Mixed Methods Appraisal Tool (MMAT) to assess both qualitative and quantitative components.

Each study received an overall quality rating (high, moderate, or low). Discrepancies between reviewers were resolved through discussion and consensus. No studies were excluded based solely on quality assessment outcomes; however, appraisal scores were considered when interpreting the strength and reliability of findings during analysis.

2.7 Data Analysis

The analysis involved a descriptive and thematic approach, reflecting the mixed-methods nature of the included studies. Due to the heterogeneity in AI interventions, study designs, outcome measures and clinical settings, meta-analysis was not feasible. Instead, findings were organized and synthesized according to the three main phases of the medication process:

1. Prescribing
2. Administration
3. Monitoring

Quantitative data (e.g., error rates, reduction percentages) were tabulated and analyzed to identify trends in effectiveness of AI technologies. When available, reported effect sizes or relative risk reductions were documented. Frequencies and percentages were used to compare impacts across interventions.

Qualitative data (e.g., nurse experiences, perceived challenges) were analyzed using inductive thematic analysis. Emerging themes were categorized under implementation facilitators, barriers, workflow impacts, trust in AI systems and ethical considerations.

Cross-study comparisons were made to highlight consistencies, contradictions and gaps in the evidence. The integration of both quantitative and qualitative findings aimed to provide a comprehensive understanding of AI’s role in nursing-led medication safety.

2.8 Ethical Considerations

This review was based entirely on published and publicly accessible data and did not involve any direct patient interaction or collection of personal health information. Therefore, formal ethical approval was not required. All included studies were appropriately referenced and screened to ensure integrity, transparency and absence of plagiarism.

3 Results
3.1 Summary of Included Studies

A total of 12 studies were included in this systematic review. These comprised a mix of systematic reviews, mixed-methods evaluations, descriptive studies, case studies and technical or policy reports. The included studies originated from diverse geographic regions, including the United States, China, Germany, Ireland, Finland, Saudi Arabia and the European Union, representing various clinical settings such as hospitals, intensive care units and national health systems.

The studies focused on a variety of artificial intelligence (AI) tools used in medication safety, including clinical decision support systems (CDSS), computerized physician order entry (CPOE) systems, barcode medication administration (BCMA), predictive analytics and AI-driven pharmacovigilance tools. The interventions were assessed for their impact on medication error rates, nurse workflow, system usability and implementation challenges. A comprehensive summary of the included studies is presented in ^{Table 2}.

3.2 Effectiveness of AI in Reducing Medication Errors

Quantitative data across several studies demonstrated the substantial potential of AI technologies in minimizing medication errors in clinical nursing practice:

• • CPOE systems significantly reduced prescribing errors by eliminating issues associated with manual entry, illegible handwriting and transcription. Studies reported notable improvements in dosing accuracy and drug interaction alerts ( ^{Elshayib and Pawola, 2020}).
• • Clinical Decision Support Systems (CDSS) were shown to be effective in delivering real-time alerts related to contraindications, dosage errors and drug interactions. Their implementation was associated with a marked reduction in medication-related incidents ( ^{Varghese et al., 2018}).
• • BCMA systems improved the accuracy of medication administration by verifying the right patient, medication, dosage and time. However, challenges were noted in workflow integration, as some nurses experienced alert fatigue and operational interruptions ( ^{O'Connor et al., 2023}).
• • AI-based text-mining tools were employed in analyzing medication incident reports. This approach facilitated early identification of error patterns and trends, supporting the development of targeted prevention strategies ( ^{Härkänen et al., 2021}).

Quantitative outcomes reported by these studies are summarized in the following ^{Table 3}.

3.3 Implementation of AI Tools in Nursing Workflows

The integration of AI tools such as CDSS, BCMA and predictive analytics has significantly influenced nursing workflows. These technologies aid in decision-making, minimize manual tasks and improve medication verification processes. Nurses using CDSS and predictive analytics tools reported increased confidence in clinical decisions, particularly in high-risk environments like intensive care units.

A study by ^{Alowais et al. (2023)} highlighted the role of AI-driven predictive analytics in decreasing hospital readmissions and cognitive workload by allowing nurses to identify potential complications in real time. Similarly, BCMA systems automated routine verification tasks; however, inconsistent system integration occasionally disrupted workflows.

Despite these advantages, challenges in adoption persist. ^{Buchanan et al. (2020)} noted that limited training, low digital literacy and inadequate involvement of nursing staff during system design contributed to resistance and underusation. Nurses stressed the importance of hands-on training and ongoing technical support to ensure successful AI implementation.

3.4 Qualitative Findings: Nurse Perspectives and User Feedback

Several studies included in this review highlighted qualitative data reflecting nursing staff perspectives on the integration of AI technologies into medication safety workflows. Common themes emerged around trust, usability, training needs and perceived workload.

Nurses expressed concerns about algorithmic opacity, particularly regarding the reliability of AI-generated alerts and recommendations. Many participants reported experiencing alert fatigue, especially when faced with high volumes of non-critical warnings generated by CDSS and BCMA systems ( ^{O'Connor et al., 2023}). This contributed to desensitization and occasional bypassing of alerts, undermining system safety benefits.

In contrast, nurses who received structured training reported higher confidence in using AI tools and greater appreciation for their role in reducing manual burden. ^{Buchanan et al. (2020)} found that comprehensive training programs not only improved the functional use of technology but also promoted greater acceptance by demystifying AI mechanisms and demonstrating tangible benefits.

Trust and usability were noted as critical factors influencing adoption, underscoring the importance of involving nurses in the design, testing and feedback stages of AI system development.

3.5 Barriers to Integration

Despite the potential of AI technologies in improving medication safety, several barriers to integration were consistently identified across the reviewed literature:

• • Technical Limitations: Complex AI systems often encountered compatibility issues with existing electronic health record (EHR) platforms. ^{Haddad et al. (2022)} noted that poor interoperability and system malfunctions disrupted clinical workflows and eroded user confidence.
• • Training and Digital Literacy: Inadequate training was a recurrent theme. ^{Buchanan et al. (2020)} emphasized the need for structured, role-specific education to support effective use. Nurses without prior exposure to advanced digital tools were particularly vulnerable to underusation and misuse.
• • Data Security and Ethical Concerns: The use of AI in clinical environments raises ethical questions about data privacy and patient consent. AI systems often require large-scale data access, which, if not adequately protected, could compromise confidentiality. ^{Danezis et al. (2015)} highlighted the necessity of integrating privacy-by-design principles to ensure compliance with data protection regulations and maintain public trust.

These barriers illustrate the importance of strategic planning, technical infrastructure and interprofessional collaboration to ensure the safe, ethical and sustainable adoption of AI in nursing settings.

3.6 Real-World Applications and Case Study Evidence

The practical application of AI in clinical environments has yielded encouraging results. One illustrative case comes from ^{Härkänen et al. (2021)}, who evaluated an AI-driven system implemented at a Finnish university hospital to analyze free-text medication incident reports. The AI application successfully identified high-risk trends and process gaps, enabling hospital administrators to design preventive interventions. This case underscores the utility of AI not only in real-time decision support but also in retrospective incident analysis for quality improvement.

Another real-world example includes the work by ^{Alowais et al. (2023)}, who examined AI-powered predictive analytics used by nursing teams to anticipate adverse outcomes and guide early interventions. The study reported improvements in patient safety, reduced readmissions and enhanced nurse workflow efficiency, supporting the broader potential of predictive models in clinical decision-making.

These cases reflect how AI can transition from theoretical potential to practical benefit when thoughtfully integrated, adequately resourced and aligned with frontline clinical needs.

4 Discussion

The investigation employs both quantitative descriptive data to demonstrate AI effectiveness in reducing medication mistakes and qualitative case study information to assess nurses' perspectives and barriers to AI implementation. There is considerable efficacy in the application of AI technologies that reduce medication errors in nursing practices. For instance, AI implementation of Clinical Decision Support Systems (CDSS) has shown a significant reduction in medication error incidence due to their ability to detect automatic alerts for dosage and interaction checks, which has improved patient safety ( ^{Jetske Graafsma et al., 2024}). Similarly, an AI-powered dispensing system for verifying medication now confirms high accuracy and efficacy in the medication dispensing process while also reducing potential errors ( ^{Zheng et al., 2023}).

4.1 Interpretation of Quantitative Findings

AI demonstrated statistical proof of reducing medication errors through quantitative descriptive assessments which supports its role in enhancing medication safety. Previous studies confirm that automation reduces human errors during medication administration and CDSS along with smart infusion pumps demonstrate this pattern.

4.2 Interpretation of Qualitative Findings

A qualitative case study approach gave insights about the problems faced by nurses who want to use AI in medication administration. The research showed that AI capabilities enhance operational speed but health professionals need extra training to use systems successfully.

4.3 Strengths and Limitations of Study Design

The research design employed quantitative descriptive methods with qualitative case study approaches to yield complete AI-related results for medication errors. Future studies will benefit from using either experimental methods or quasi-experimental methods since these designs permit for direct comparisons between conventional medication procedures and AI-based intervention approaches in controlled healthcare facilities.

Areas, where AI excels over traditional methods, are the area of predictive analytics due to its ability to check through huge datasets and determine whether certain medication errors could potentially take place before the occurrence of the error. For instance, machine learning (ML) algorithms can predict adverse drug reactions (ADR) based on the history and genetic data of the patients and thereby suggest preemptive change in treatment plans ( ^{Yang and Kar, 2023}). While such reactive measures are common, this proactive approach stands out because it is a great step forward from the norm and puts AI at center stage for the prevention of errors. AI also takes the lead in the automatization of medication administration using working smart infusion pumps and it ensures the accuracy of dosage considering and avoiding manual adjustments with errors. Dose-error reduction software (DERS) is used in these systems to check actual doses to the dose prescribed in real-time, thereby eliminating underdosing or overdosing in high-risk dosages i.e., anticoagulants and opioids ( ^{Leape et al., 2024}). At Massachusetts General Hospital, a study showed a 75 % reduction in medication administration errors by hospitals using AI infusion systems as compared with traditional manual calculations ( ^{Nanji et al., 2024}).

Finally, AI enhances barcode-assisted medication administration (BCMA) to increase real-time rune verification and therefore reduce patient-drug mismatches. AI-enabled barcode scanning helps nurses to match the patient to drug, dose, route and time (5 rights of medication administration) and aids with the practice of minimizing human oversight in busy clinical settings ( ^{Owens et al., 2020}). An AI-enhanced BCMA was tested in a large-scale trial in a multi-hospital system in the United. However, rather than just reducing errors, AI has made it possible to embed real-time medication safety alerts in Electronic Health Records (EHR). By sending these alerts, the nurses and prescribers are being alerted toward potential contraindications, duplicate prescriptions, or alternative recommendations, which results in more than precise medication management. Beyond traditional methods based on cross reference and human recall, this level provides an increased chance of oversight ( ^{Classen et al., 2024}).

4.4 Challenges and Barriers
4.4.1 Nurse Resistance

While it is not ideal to resist change, the introduction of AI technologies has a high probability of nursing staff feeling apprehensive that technology will replace human judgment ( ^{George et al., 2024}). A study examining registered nurses’ attitudes towards the use of AI in the Kingdom of Saudi Arabia identified that AI might be perceived to take away skills currently possessed by nurses and that care would not require human expertise ( ^{Alotaibi and Federico, 2017}). What needs to be minimized is this and, as such, it is important to highlight AI as a supportive tool, rather than one that replaces clinical judgment ( ^{Table 4}).

4.4.2 Ethical Concerns

There are ethical issues to consider when deploying AI for medication management, one of which is accountability in the AI driven decision-making. Who is to blame when AI systems err is also questioned: developers, the healthcare institution, or the practitioner? Moreover, handling vast amounts of patient data required for the use of AI raises the question of privacy and, thus, also informed consent. The ^{American Nurses Association (ANA) (2022)} emphasizes that we believe that American nurses must be knowledgeable about artificial intelligence to provide proper education to the patients and their families by supporting the ethical use of AI for optimal health outcomes ( ^{Riley, 2024}) ( ^{Table 4}).

4.4.3 Financial Constraints

AI Technologies have heavy budget requirements for purchasing, training and maintenance of those technologies. Integration of AI is not something that can be fully planned and implemented in healthcare facilities, especially those with constrained resources, for which it may be difficult to transport their budget into AI integration. For instance, AI-driven systems in nursing homes or long-term care facilities have not been adopted for fear of money, even though they could benefit in improving patient care ( ^{Pailaha, 2023}). But to justify required AI investments, such calculations must be made; error reduction and improved patient well-being give long-term benefits.

4.5 Policy and Future Research Recommendations

To solve these problems and take advantage of AI in nursing, several strategies are suggested:

Educational programs need to be developed to provide nurses with knowledge about artificial intelligence technologies. To make nurses proficient in the usage of AI tools, AI literacy can be integrated into nursing curricula and continuous professional development provided to them. Using this approach not only decreases the resistance involved but also facilitates nurses to adopt AI and further aid in improving patient care ( ^{Buchanan et al., 2020}). To accommodate the nonstatic environment of clinical settings, the AI systems need to be designed to be flexible. Nursing professionals and AI developers need to collaborate to achieve user-friendly interfaces that are in line with the current workflows. Indeed, nurses can get involved in the designing and development of AI tools to ensure it is facilitating the real challenges faced in clinical practice ( ^{Zheng et al., 2023}).

Setting up clear frameworks for ethical issues involved in the integration of AI in healthcare makes it go a long way. Issues related to accountability, data privacy, informed consent and others should be dealt with in guidelines. Thus, the ANA's position statement on the ethical use of AI in nursing practice provides a foundation to guide the use of AI for nursing practice applications that support and further the core values and ethical obligations of the profession ( ^{American Nurses Association (ANA), 2022}). These guidelines are something that can further developed to help in the implementation of AI in healthcare organizations responsibly. Finally, AI promises to bring great innovations in minimizing aspects of medication error in nursing practice, but the implementation of such AI needs to conquer human factors, ethics and financial means. ^{Rony et al. (2024)} proposed that the nursing profession can effectively incorporate AI to boost patient safety and quality of care through a focus on education, adaptable technology design and robust ethical standards.

In addition to structural integration and policy oversight, efforts must also focus on preparing the clinical workforce to adopt these tools effectively. The integration of artificial intelligence (AI) into medication safety underscores a critical need to enhance nursing and midwifery education. The successful adoption of technologies such as clinical decision support systems (CDSS), smart infusion pumps and barcode medication administration (BCMA) depends heavily on frontline professionals' familiarity, trust and competence with these tools. Several studies, including ^{Buchanan et al. (2020)}, emphasize that insufficient training is a key barrier to effective AI implementation, often resulting in misuse, underusation, or resistance.

To address this, AI literacy should be embedded into undergraduate and postgraduate nursing curricula, with a focus on core concepts such as algorithmic logic, bias recognition, data ethics and digital safety. Educational institutions should also introduce simulation-based learning environments where students can interact with AI-enabled technologies such as CDSS and smart pumps in realistic clinical scenarios. These hands-on experiences have been shown to improve skill retention and confidence in using high-risk systems ( ^{Alowais et al., 2023}).

Moreover, interprofessional training modules that incorporate AI alongside pharmacology, patient safety and decision-making can help bridge the knowledge gap between technology and clinical reasoning. By aligning AI training with competency-based education models, nurses and midwives can be better prepared to critically assess, manage and safely use AI in practice.

5 Conclusion

This systematic review examined the impact of artificial intelligence (AI) integration on medication error reduction in nursing practice, with a focus on prescribing, administration and monitoring processes. The findings support the clinical value of AI technologies—such as clinical decision support systems (CDSS), smart infusion pumps, barcode medication administration (BCMA) and predictive analytics—in enhancing medication safety and reducing error rates. Evidence from quantitative studies demonstrated significant reductions in prescribing and administration errors, while qualitative data revealed both opportunities and challenges from the nursing perspective, including alert fatigue, usability concerns and varying levels of digital confidence.

However, the successful implementation of AI in nursing practice extends beyond technological deployment. A recurring theme across the included studies is the importance of adequate training, user engagement and professional development. Without targeted educational initiatives, the benefits of AI may be underused or even compromised. Nurses often face unfamiliar technologies with limited guidance, which can lead to resistance, misapplication, or an erosion of trust in automated systems. These barriers highlight the urgent need for a parallel investment in educational infrastructure.

To that end, this review underscores the critical role of nursing and midwifery education in enabling safe and effective AI adoption. Academic and clinical training programs should incorporate AI literacy, covering core topics such as algorithmic logic, ethical considerations, data privacy and human–machine interaction. Simulation-based training with technologies like CDSS and smart pumps should be embedded in curricula to ensure experiential learning. Additionally, continuing professional development (CPD) must provide opportunities for practicing nurses to update their digital competencies and engage with evolving systems in a supported environment.

Ultimately, AI has the potential to transform medication safety in nursing. Yet its true impact will depend on how well nurses are prepared—educationally, ethically and practically—to integrate these tools into their clinical reasoning and workflows. Future research and policy should prioritize interdisciplinary efforts that advance not only the technology itself but also the educational frameworks that empower nurses and midwives to use it confidently, critically and competently.

CRediT authorship contribution statement

Khan Waleed: Data curation. Muhyeeddin Alqaraleh: Conceptualization, Formal analysis, Investigation, Project administration, Software, Validation, Writing – original draft, Data curation, Funding acquisition, Methodology, Resources, Supervision, Visualization, Writing – review & editing. Almagharbeh Wesam: Data curation, Funding acquisition, Methodology, Resources, Supervision, Visualization, Writing – review & editing, Conceptualization, Formal analysis, Investigation, Project administration, Software, Validation, Writing – original draft.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Table 1

Keywords and Synonyms Used in the Literature Search.

Concept	MeSH Terms	Free-Text Keywords / Synonyms
Artificial Intelligence	"Artificial Intelligence" [MeSH] "Machine Learning" [MeSH]	artificial intelligence, AI, machine learning, clinical decision support, smart pumps, automated dispensing, algorithm, predictive analytics
Medication Errors	"Medication Errors" [MeSH] "Drug Administration Errors" [MeSH]	medication error, drug error, prescribing error, administration error, dosing error, adverse drug event, ADE
Nursing	"Nursing" [MeSH] "Nursing Staff" [MeSH]	nurse, nurses, nursing practice, registered nurse, bedside nurse
Patient Safety	"Patient Safety" [MeSH]	safety, error prevention, clinical safety, risk reduction, harm reduction

Table 2

A summary of the included studies.

Author (Year)	Country	Study Type	AI Technology Used	Outcomes Measured	Main Findings
Al Khatib and Ndiaye (2025)	Unspecified	Quantitative descriptive	CDSS, alert filtering	Medication error reduction (operating rooms)	CDSS reduced OR errors by 95 %
Elshayib and Pawola (2020)	USA	Systematic review	CPOE systems	Reduction in prescribing errors	CPOE systems significantly reduced errors
Zheng et al. (2023)	China	Prototype evaluation study	AI dispensing verification	Dispensing accuracy and trust	Improved dispensing accuracy, enhanced pharmacist trust
Varghese et al. (2018)	Germany	Systematic review	CDSS	Effectiveness of CDSS	CDSS effectively reduced medication errors
O'Connor et al. (2023)	Ireland	Mixed-methods evaluation	BCMA systems	Workflow impact, administration error reduction	BCMA reduced errors but introduced workflow disruptions
Khinvasara et al. (2024)	Global	Systematic review	Various AI applications	Reduction in adverse drug events (ADEs)	AI interventions reduced ADEs across settings
Härkänen et al. (2021)	Finland	Case study	Text-mining incident reports	Identification of medication safety risks	AI identified common error types from incident reports
Alowais et al. (2023)	Saudi Arabia	Descriptive study	Predictive analytics	Workflow efficiency, readmission reduction	AI improved nurse productivity and clinical outcomes
Haddad et al. (2022)	Unspecified	Technical report	System integration analysis	Barriers to AI-EHR integration	Highlighted compatibility and implementation challenges
Buchanan et al. (2020)	UK	Review & survey study	Nurse technology training	Training needs and adoption barriers	Training improves AI adoption and confidence in nursing
Danezis et al. (2015)	EU	Policy & technical report	Privacy-by-design models	Data privacy and regulatory compliance	Emphasized strong privacy frameworks for AI in healthcare
Naeem and Coronato (2022)	Italy	Model development	Reinforcement learning	Clinical decision support safety modeling	RL models improved AI accuracy and risk reduction strategies

Table 3

Summary of Quantitative Outcomes.

AI Intervention	Error Reduction (%)
Computerized Physician Order Entry (CPOE)	Significant reduction in prescribing errors
Clinical Decision Support Systems (CDSS)	High reduction in prescribing and interaction-related errors
Barcode Medication Administration (BCMA)	Reduction in administration errors; noted workflow disruptions
AI Text-Mining for Incident Reports	Enhanced identification of medication error patterns

Table 4

The identified challenges and their impact.

Challenge	Impact (%)
Nurse Resistance	62 %
Alert Fatigue	49–96 % deflection rate
Data Privacy Concerns	High
Algorithmic Bias	Medium-High
Lack of External Validation	Moderate