Effectiveness of Artificial Intelligence–Based Nursing Interventions for Chronic Illness Care: Umbrella Review

Introduction

Background

In recent years, the evolution of artificial intelligence (AI) has led to major transformations in the nursing domain. The World Health Organization (WHO) announced that health care systems should use AI technologies for patient-centered and continuous care [1]. The International Council of Nurses (ICN) has highlighted that the adoption of AI is an innovative approach to health care and a potentially beneficial tool for providing nursing care to patients to enhance global health and wellness [2]. Ultimately, AI technologies adopted in health care systems have been integrated into nursing research, practice, and health care systems.

Globally, chronic illnesses have been a serious public health problem for decades. Referred to as “the Silent Pandemic,” chronic illnesses account for approximately 74% of global deaths and pose challenges to global health systems and the economy [1]. Individuals with chronic illnesses experience multiple comorbidities, require continuous care from nurses, and experience recurrent planned or unplanned hospital visits. Evidence from nursing research has shown that interventions by nursing professionals improve chronic illness care. Nurses are key health care professionals who provide patient-centered and effective coordinated care to improve chronic illness outcomes [3,4].

With the evolution of AI in nursing, extensive efforts have been made to apply AI technologies and algorithms. Specific AI techniques in nursing include machine learning (ML), deep learning (DL), robotics, decision support systems, and natural language processing [5,6]. AI algorithms are computational procedures that enable computers to execute specific tasks by following predefined rules and processing data. Common AI algorithms include decision trees, logistic regression, random forests, neural networks, and Naïve Bayes [7]. AI techniques and algorithms have been used to predict patient conditions and risks, support disease diagnosis, and provide personalized care [8,9].

Several reviews have examined AI-based health care interventions and their impact on patient care [10-12]. Previous evidence has also explored the application of AI in nursing research, practice, and education [5,13,14]. In addition, reviews focusing on AI-based nursing interventions for chronic illness care have been conducted [15-17]. However, the findings regarding the effectiveness of AI-based nursing interventions remain inconsistent because of varied outcome measures and a lack of consensus regarding their clinical impact. This evidence gap may hinder the advancement and implementation of AI-supported nursing care for chronic illnesses. Therefore, a comprehensive synthesis of current evidence is needed. Accordingly, this umbrella review aimed to identify and evaluate the effectiveness of AI-based nursing interventions for chronic illnesses and their impact on health care outcomes in clinical settings.

Objective

This study addressed two research aims: (1) to identify nursing interventions in health care settings that apply AI in caring for chronic illnesses to influence health care outcomes and (2) to synthesize and evaluate the impacts of AI-based nursing interventions on chronic illnesses.

Methods

Study Design

This study followed the umbrella review methodology developed using the Joanna Briggs Institute (JBI) guidelines [18]. An umbrella review is an approach that compiles the best evidence from existing published systematic reviews (SRs) to obtain a single comprehensive review [18]. This umbrella review methodology is useful for aggregating and expanding recent issues and for highlighting synthesized evidence for future directions in health care research [18]. This umbrella review was conducted using the PICO (Population, Intervention, Comparison, and Outcome) framework to define the eligibility criteria and guide paper selection [19]. The review process adhered to the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines (see Checklist 1). The PICO framework was considered appropriate because the review sought to capture a broad range of evidence regarding AI-based nursing interventions across diverse intervention designs, health care settings, and patient outcomes. This review has been registered with the PROSPERO (International Prospective Register of Systematic Reviews) database. No institutional review board approval was required because no humans were included in the study.

Search Strategy

Five electronic bibliographic databases (CINAHL, Cochrane Library, PubMed, Scopus, and Web of Science) were searched to identify relevant studies. The complete search strategy for each database, including Boolean operators and keyword combinations, was developed in consultation with the research team and is provided in Table S1 of Multimedia Appendix 1 to enhance reproducibility.

The following keywords and Medical Subject Headings were used with Boolean operators (AND and OR): (“artificial intelligence” OR “AI” OR “machine learning” OR “technology”), (“systematic review” OR “meta-analysis”), and (“chronic” OR “chronic disease” OR “chronic illness”) or (“nurse” OR “nursing”). Several additional search terms were used: “prediction” or “effectiveness” or “health outcome” or “nursing research.” To capture the most current evidence regarding the rapid evolution of AI in nursing, we limited our selection to SRs published within the past 5 years. Therefore, SRs with and without meta-analyses published between January 2021 and December 2025 were included in this study. This time frame was selected because AI technologies in nursing and health care have rapidly evolved in recent years, particularly following the increased implementation of ML, conversational agents, and predictive analytics in clinical practice after 2020. Restricting this review to recent SRs enabled the inclusion of the most contemporary evidence reflecting current AI capabilities, health care infrastructure, and nursing practices. However, we acknowledge that this restriction may have excluded earlier relevant evidence and potentially introduced selection bias. Articles available ahead of print publications or as online-first publications were also included. The initial search was conducted on December 1, 2025, and a confirmatory search was conducted on February 28, 2026, to identify potentially eligible studies.

Inclusion and Exclusion Criteria

Textbox 1 shows the inclusion and exclusion criteria. Based on the PICO framework, the Population (P) included patients with chronic illnesses (eg, diabetes, hypertension, and chronic kidney disease) who received nursing care services; the Intervention (I) was defined as the integration of AI-based technologies into nursing interventions in health care settings. The Comparison (C) included the usual care, control groups, or non–AI-based interventions. The Outcome (O) included quantitative patient outcomes. The eligible studies were SRs with or without meta-analysis that included primary empirical studies such as cross-sectional studies, quasi-experimental studies, and randomized controlled trials (RCTs).

In this umbrella review, AI-based nursing interventions were operationally defined as nursing-related interventions that incorporated AI technologies to support assessment, clinical decision-making, patient monitoring, communication, education, or care coordination for patients with chronic illnesses. In this review, AI-based nursing interventions refer to those that incorporate AI technologies, such as ML models, predictive analytics, or clinical decision support systems, to support nursing assessment, decision-making, monitoring, or care coordination. These interventions included nurse-led applications implemented directly by nurses or nurse practitioners (NPs), as well as AI-supported systems integrated into nursing workflows within health care settings. Studies were included if AI technologies were integrated into nursing care delivery, nursing assessment, nurse-led chronic disease management, or other nursing-related clinical workflows. However, studies focusing solely on institution-level administrative AI systems without direct relevance to nursing care delivery were excluded. This umbrella review also excluded publications that were not published in English or that indicated the use of generative AI, such as ChatGPT, in manuscript writing.

In this umbrella review, “adverse events” were operationally defined as undesirable clinical outcomes associated with chronic illness progression or health care utilization, including mortality, unplanned hospital readmission, intensive care unit (ICU) admission, and disease-related deterioration. “Complications” referred to disease-related secondary conditions such as neuropathy, nephropathy, retinopathy, or other chronic illness–related sequelae. “Hospital utilization outcomes” included hospital readmissions, emergency department visits, length of hospital stay, and health care–related costs. To improve consistency and interpretability, these terms were used consistently throughout this review.

Study Selection

Two reviewers independently screened the titles, abstracts, and full texts based on the predefined eligibility criteria. Any disagreements regarding study inclusion were resolved through discussion and consensus among all research team members. When consensus could not be reached initially, the final decision was made after a repeated review of the full-text article and methodological criteria. To assess potential double-counting bias, overlap among the included SRs was examined using a citation matrix. The primary studies included in each SR were extracted and compared across reviews to identify overlapping studies and duplicate publications.

Quality Appraisal of Selected SRs

For umbrella reviews, it is recommended to assess the professional and methodological quality of the included SRs using a validated checklist [18,20]. This study used the JBI Critical Appraisal Checklist for SRs [21]. This checklist is recommended for umbrella reviews and provides a structured assessment of the methodological quality of SRs. It was developed to evaluate research methodologies and address the risks of bias in included studies when conducting an umbrella review [20]. Therefore, it was considered appropriate for the methodological approach in this umbrella review. The checklist consists of 11 questions in the following domains: study aim, search strategy, inclusion and exclusion criteria, data selection and extraction, publication bias, and appropriate results. These questions have 4 response options: “yes,” “no,” “unclear,” and “not applicable.” The total sum of responses ranged from 0 to 11; if the answer was “yes,” it was scored as 1; otherwise, it was scored as 0. The JBI checklist does not provide predefined cutoff scores to classify reviews as high, moderate, or low quality. However, high scores indicate high quality and low risk of methodological bias. The first and second researchers independently evaluated the quality of the 8 selected studies using the checklist.

Data Extraction and Synthesis

In this study, data were extracted by the primary researcher using an Excel data sheet. A standardized Excel-based extraction form was developed prior to data extraction to improve consistency and reproducibility. The form included (1) bibliographic details (authors, types and designs of reviews, authors’ nationality, and publication years), (2) aims, (3) number of primary studies used in an SR, (4) details of databases used, (5) quality assessment tools, and (6) study participants. A summary of the types of AI reported in the selected studies, the algorithms used, and the findings of common health care outcomes was also extracted onto another data sheet. The primary researcher extracted the data, and the third researcher verified the extraction.

Due to substantial heterogeneity across the included SRs in terms of AI techniques, intervention characteristics, patient populations, chronic illness types, outcome measurements, and study designs, a quantitative meta-analysis was considered inappropriate. In particular, the included reviews used diverse outcome measures and varying definitions of health care outcomes, which limited the comparability of pooled effect sizes. Therefore, a narrative synthesis approach was adopted to comprehensively summarize and interpret the findings.

Results

General Characteristics of Included Reviews

Figure 1 presents a summary of the search process and study selection outcomes using a PRISMA flow diagram. A total of 541 records were initially identified from the 5 electronic databases. After removing duplicate records (n=190) using EndNote X9 (Clarivate Analytics), 351 unique records remained for screening. Of these, 284 records were excluded during title screening, and 44 records were excluded after abstract screening based on the predefined eligibility criteria. Full-text eligibility assessment was conducted for the remaining 23 reports, and 15 reports were excluded because they did not meet the inclusion and exclusion criteria. Finally, 8 SRs were included in this umbrella review.

‎

Table 1 presents the general characteristics of the 8 SRs, including the study aim, number of primary studies included, review methods, sample sizes, and year of publication. The 8 SRs were published between 2021 and 2025 [4,15-17,22-25]. These studies investigated the impact of AI-based nursing interventions on patients with chronic illnesses and analyzed their health care–related outcomes resulting from the interventions. The 8 included reviews were conducted in China (n=2), the United Kingdom, the United States, Canada, South Korea, Indonesia, and Slovenia (n=1 each), and they all followed the PRISMA guidelines . Three SRs had registered their review protocols with PROSPERO, while the rest did not report any registration. Three SRs included only RCTs [16,22,25]. Among the included SRs, only 1 SR conducted a meta-analysis [17].

The 8 SRs included a total of 150 primary study records. A citation matrix identified 7 overlapping primary studies across the reviews, corresponding to 8 duplicate publications. After accounting for overlap, 142 unique primary studies were represented in this umbrella review (Table S4 in Multimedia Appendix 1). Given the relatively small number of overlapping studies, the risk of double-counting bias was considered limited. Among these, 41 studies were RCTs, and the rest were quasi-experimental, retrospective cohort, or cross-sectional studies. Each SR included 5 to 65 primary studies published across various continents, such as North and South America, Europe, and Asia, from 2008 to 2024. Although all the SRs were published in English, the primary studies included in the reviews were published in the English, Chinese [17], or Korean [25] languages. These 2 SRs, which included primary studies in languages other than English, used translation and back-translation methods.

Based on the inclusion criteria, the participants were patients aged at least 16 years or older who had at least 1 chronic illness, such as type 2 diabetes mellitus or cancer, had been hospitalized in ICUs, or had visited the emergency department due to their chronic illness. The number of participants in primary studies ranged from 28 to 469,976.

Summary of AI-Based Nursing Interventions in the SRs

Table 2 summarizes the characteristics of AI-based nursing interventions used in the studies. Additionally, Tables S2 and S3 in Multimedia Appendix 1 summarize the characteristics of AI-based nursing interventions used in the studies. Table S2 in Multimedia Appendix 1 provides the number of studies that use different types of AI interventions. All the primary studies clearly reported the utilization of AI techniques and AI algorithms in their nursing interventions. In this study, the AI-based nursing interventions in the reviews were classified as AI techniques [26]. The main AI interventions retrieved from the studies included ML, which was used in all 8 studies, followed by DL, clinical decision support, and natural language processing. The types of AI algorithms used were the decision tree model, logistic regression, neural networks, and Naïve Bayes. Several AI algorithms, such as regressions, Naïve Bayes, and neural networks, were used as prediction models. The most frequently used AI algorithm was logistic regression (used in all 8 SRs). Two SRs reported that chatbots and conversational agents [22,23] were used in their nursing interventions, and 1 SR used real-time voice AI support [25].

Most AI-based interventions were led by nurses or NPs [22,24,25] and delivered by nurses or NPs within multidisciplinary health care settings [4,15-17]. These interventions were directly integrated into nursing workflows, including patient monitoring and education, symptom assessment, discharge planning, follow-up, and chronic illness management [4,15-17,22,24]. Some SRs incorporated AI decision support systems and predictive algorithms [16,17,25]; however, these AI technologies were implemented within supporting nursing interventions such as assessment, clinical decision-making, and care coordination in nursing practices. The primary studies were conducted in clinical care settings, such as long-term care settings, intensive care, or emergency departments in hospitals or primary care settings.

Seven SRs reported the usual or manual care group as the control groups; however, in the remaining SR [16], such information was not clearly reported. Li et al [23] specifically described the dosage of nursing interventions, which were assisted by AI-driven conversational agents, as reported in the primary studies.

Methodological Evaluation Results

Table 3 presents the quality assessment results of the 8 SRs. The JBI Critical Appraisal Checklist tool does not have cutoff scores; however, study scores ranging from 8 to 11 are referred to as high-quality studies, indicating a low risk of bias [27]. The review by O’Connor et al [24] scored 9 because they did not clearly and adequately report the resources that were used for the study searches and the methods used to minimize data extraction errors. The other 7 SRs scored between 10 and 11, clearly indicating the methodological directions in their reviews. Thus, all the SRs indicated a low risk of methodological bias.

Main Patient Outcomes

Overview

Among the 8 included SRs, the primary patient outcomes of AI-based nursing interventions were categorized into 3 common domains: (1) predictive outcomes (n=5), (2) psychosocial outcomes (n=5), and (3) hospital utilization outcomes (n=2; Tables S2 and S3 in Multimedia Appendix 1). Table 2 summarizes the common outcomes of these reviews. Due to the heterogeneity of the results, the findings are synthesized narratively.

Predictive Outcomes

In the 5 SRs that reported predictive outcomes, the predictions of adverse events, such as disease-related complication risks, mortality risks, and possible hospital utilization, were commonly reported using predictive algorithms [15-17,24,25].

Three SRs reported predicted possible complications related to diseases using AI-based nursing interventions [15,16,24]. In a high-quality review, Gosak et al [16] reported that nursing interventions applying AI techniques could predict the possible complications of type 2 diabetes mellitus. In this review, the risks of neuropathy, nephropathy, and retinopathy associated with diabetes were precisely predicted using AI prediction algorithms, such as penalized regression, random forest, and neural networks. Bressler et al [15] predicted the mortality risks of patients at 6 months, 1 year, and 2 years. These predictions enabled the early and timely detection of nurses’ care interventions. O’Connor et al [24] reported the predicted risk of complications in patients with cancer in 17 primary studies. Yi et al [25] reviewed emergency triage predictions and found that intervention groups for which AI models were applied showed 0.9% more accurate predictions in emergency triage than those shown by control groups. In addition, the accuracy of emergency triage prediction rates ranged from 80.5% to 99.1% in the 4 primary studies. Thus, the results indicated that AI-based interventions had an impact on predicting adverse outcomes in patients.

Two SRs reported predictions of unplanned hospital utilization [17,24]. In the study by O’Connor et al [24], the length of hospital stay was predicted using 2 primary studies, with a conclusion of a significant improvement in the reduction of length of hospital stay predictions. Hu et al [17] predicted ICU readmissions using pooled results from a meta-analysis of 56 primary studies. They found that ICU readmissions within 30 days showed higher sensitivity (0.789, 95% CI 0.751-0.824) and specificity (0.835, 95% CI 0.805-0.860); sensitivity and specificity within 7 days were 0.462 (95% CI 0.118-0.846) and 0.788 (95% CI 0.448-0.945), respectively. Hu et al [17] reported that using short-term expectations to predict ICU readmissions is more complex.

Psychosocial Outcomes

Five SRs reported patients’ psychosocial outcomes as results of AI-based nursing interventions [4,15,22-24]. In the SRs, patients’ self-reported outcomes, such as quality of care, patient safety, patient satisfaction, and depressive symptoms, were measured.

Regarding the quality of care, 2 SRs demonstrated that the quality of care increased [15,24]. However, Raymond et al [4] reported mixed results in a high-quality SR. There was an improvement in the quality of care provided by NPs in 8 primary studies; however, no improvement was reported in 1 study. The results of patient safety outcomes were mixed. In the study by Kurniawan et al [22], AI-based nursing interventions could not support patient safety. Raymond et al [4] reported improvements in patient safety in 3 primary studies. Regarding patient satisfaction outcomes, Kurniawan et al [22] reported significant differences between groups in 7 primary studies after applying AI-based nursing interventions. Two SRs that reported depression as the primary outcome of AI-based nursing interventions for chronic illnesses [4,23] reported a significant reduction in depression in their 5 included primary RCTs. However, Kurniawan et al [22] reported no differences between the intervention and control groups in 3 studies. Therefore, the impact of AI-based nursing interventions on psychosocial outcomes remains inconclusive.

Hospital Utilization Outcomes

Two SRs reported total hospital utilization outcomes resulting from AI-based nursing interventions for chronic illnesses [4,15]. Both SRs investigated hospital readmissions and health care–related costs, and the results were conclusive. In the review by Bressler et al [15], 60- and 90-day hospital readmissions were significantly decreased in 1 primary study. Similarly, Raymond et al [4] reported that 2 primary studies showed significant reduction in rehospitalizations. Both SRs reported cost savings in intervention groups compared with control groups in hospital-based settings [15] and hospital care, especially in emergency care settings [4]. Both SRs, which were high-quality reviews, concluded the cost-saving capabilities of AI-based nursing interventions.

Discussion

Principal Findings

The integration of AI techniques into patient care has been used by nurses in various health care settings and has been reported to improve patient health care outcomes. This umbrella review expands on the current inconclusive results of AI-based nursing interventions to provide robust evidence on AI-based nursing interventions for chronic illnesses. The first aim of this umbrella review was to summarize nursing interventions that applied AI in caring for patients with chronic illnesses in health care settings, which may have affected patients’ health care outcomes by analyzing individual SRs conducted in 7 countries. In this study, we synthesized the findings from 8 SRs focusing on AI-based nursing interventions for chronic illnesses that were conducted using 143 primary studies. All 8 SRs were published recently within the past 5 years, from 2021 to 2025. This study identified the types of AI techniques and algorithms used in recent evidence-based research. Regarding the AI techniques applied in nursing interventions, ML, decision-support systems, and DL, which is a subset of ML [28], were identified in the included SRs. These findings should be interpreted in the context of the specific AI modality rather than as evidence for a single homogeneous category of AI-based nursing interventions. Among the AI algorithms, logistic regression models were most frequently used in the included reviews, as 5 of the SRs focused on predicting adverse outcomes.

Based on the secondary aim of this review, we synthesized 3 primary and common patient outcomes—predictive, psychosocial, and hospital utilization outcomes. This review identified the advantages of AI-based nursing interventions in predicting adverse events and hospital utilization outcomes. According to the SRs, applying AI algorithm models to intervention groups enabled better prediction of potential risks in patients with chronic illnesses, such as complications and unplanned hospital utilizations, compared with control groups without AI assistance. Chronic illnesses are associated with multiple comorbidities and complications [29]; thus, early-detection systems using AI are important [8,30]. Consistent with previous reviews, AI-based predictive models impact disease-related risk predictions [31,32]. In this umbrella review, we found that nurses could identify the possible risks of adverse events during AI-assisted nursing interventions. Subsequently, the nurses perceived patients’ care needs with AI algorithms and proactively provided appropriate care for patients with chronic illnesses to prevent possible adverse events [16,25]. From the included SRs, we pooled data to predict unplanned ICU readmissions using the AI technique for critical care nurses [17]. The results showed that applying AI techniques affected nurses’ practice, ensuring the prioritization of high-risk patient care. However, there could be accuracy issues and an algorithm risk of bias in AI-assisted predictions [17,25]. Thus, more rigorous AI models that can support nurses’ interventions and larger datasets to critically predict chronic illness risks are required.

This umbrella review also revealed reductions in hospital readmissions and health care–related costs. Two SRs reported that AI-based nursing interventions significantly decreased hospitalization and health care costs for chronic illnesses [4,15]. The results regarding hospital utilization outcomes are supported by multidisciplinary studies on patients at high risk or receiving postoperative care [9,33]. In this review, we concluded that hospital utilization prediction was effective; however, 2 of the included SRs (4 primary studies) reported an impact on chronic illnesses. Furthermore, none of the SRs specifically compared the full economic analysis related to hospital use. More robust evidence that includes a large dataset and an elaborate analysis of the impact of AI-based nursing interventions on hospital utilization and cost analysis for chronic illness in health care settings is recommended.

The results regarding psychosocial outcomes were inconclusive. This umbrella review found mixed results on the safety, quality of care, satisfaction, or depression status among patients with chronic illness. Owing to the complex and wide range of psychological outcome measurements, SRs reported varied outcomes. One study demonstrated a false alarm regarding patients’ adverse outcomes; thus, it reported no effectiveness of AI-based interventions on patient safety [22]. Recently, some nursing reviews have explored patient safety outcomes using AI [30,34]. However, the results have been inconclusive. Therefore, more studies on clear psychosocial measures of psychological outcomes after AI-based nursing interventions are required to provide evidence.

In the included SRs, evidence regarding the effectiveness of AI-based theoretical frameworks, specifically which AI type or algorithm effectively affects patient outcomes, is limited. In evidence-based practice, it is imperative to have theoretical frameworks that can be accepted as part of AI technology in nursing. One SR by Hu et al [17] concluded that ML outperformed other algorithms in predicting chronic illness outcomes. This finding is congruent with previous reviews, indicating that ML has been the most useful AI technology for nursing interventions [5]. Recent studies have demonstrated that AI models with AI algorithms are highly accurate in predicting patient outcomes for specific diseases in nursing research [5,10]. However, in this umbrella review, none of the SRs indicated that AI algorithms outperformed others in patient outcome predictions, nor did any show that AI algorithms were mostly used to predict patient outcomes. Further empirical studies are required to identify successful implementations of AI algorithm-optimized nursing interventions that are most useful for chronic illness care provision. Furthermore, theoretical frameworks for AI, targeting chronic illness management, should be established to implement AI-based nursing interventions.

When analyzing such AI-based nursing interventions regarding patient outcomes, it is necessary to clearly and robustly include complex datasets. Moreover, it is imperative to combine and refine all nurses’ clinical notes, including texts from other health care professionals, to identify the most effective AI algorithms or types [24]. It is important to include nursing notes or unstructured texts when analyzing patient outcomes in AI-based nursing interventions for robust data analysis [4,15,17,24]. Several SRs have reported that nurses’ free-text documentation and clinicians’ notes are useful for AI-based model development and for predicting patient outcomes [4,15]. Hu et al [17] suggested that ML outperformed other methods in predicting patient outcomes; however, they used only structured datasets and ignored clinical notes. Thus, it may be insufficient to conclude that ML is a promising AI technique for nursing interventions in care provision for chronic illness. It is recommended to use both structured and unstructured data, such as nursing notes, to predict outcome measurements, along with AI-based techniques to eliminate the risk of data bias.

Strengths and Limitations

One strength of this umbrella review is that the 8 SR results included 41 RCTs and experimental design studies, which is a little less than half of the 143 primary studies. A recent review reported that there is a dearth of experimental studies on AI-based intervention in nursing, and it may be possible to prove the effectiveness of AI on patient outcomes in nursing research [5]. In addition, this review provides evidence that elucidates the impact of AI-based nursing interventions on related patient outcomes due to the high methodological quality of SRs.

This study has a few limitations. First, we identified and synthesized 3 common patient outcomes from 8 SRs. This review identified 3 dominant patient outcomes as the impact of AI-based nursing interventions. There could be more positive outcomes and benefits of AI-based technologies. Moreover, substantial heterogeneity in AI intervention types, outcome measures, patient populations, chronic illness types, and study designs limited the feasibility of conducting a quantitative meta-analysis. Therefore, this review could only aggregate and summarize the findings narratively. More frequently, pooled results reporting patient outcomes with meta-analytical methods are desirable. Given the heterogeneity of AI technologies, intervention characteristics, and outcome measures, these findings should not be interpreted as evidence supporting the effectiveness of AI-based nursing interventions as a homogeneous entity. Second, the 8 included SRs focused only on the effectiveness of AI technologies in participants with chronic illnesses. There was limited information about the dosage of the interventions, indicating a dearth of information about the frequency or duration of AI-based interventions. Only the review by Li et al [23] reported the dosage of interventions in their primary RCTs. Thus, this umbrella review had limited information from the included SRs to perform a dosage analysis of AI techniques and nursing interventions and its impact on patient outcomes. Third, this umbrella review included SRs published in peer-reviewed journals from 5 electronic databases in the past 5 years and excluded gray literature to meet the aim of this review. Thus, some data may have been missing. Furthermore, this study may have a potential overlap of primary studies across the included SRs, which may introduce double-counting bias and affect the validity of the findings. Fourth, participants with chronic illnesses included in the SRs were recruited within health care settings, such as emergency departments or long-term care. There was limited information about their age, number of chronic diseases, or any complications. Fifth, potential publication bias could have been present in the 8 SRs, as they had the tendency to report positive results. Finally, this review included only SRs published in English.

Several challenges should also be considered when integrating AI-based technologies into nursing practice. Algorithmic bias, data quality limitations, and the limited explainability of some AI models may affect the reliability and transparency of AI-assisted clinical decision-making. In addition, interoperability with existing electronic health record systems remains a significant barrier in many health care settings. Clinician trust, ethical considerations, data privacy, and regulatory compliance must also be addressed to support the safe and effective implementation of AI-enabled nursing interventions. Future research should evaluate not only effectiveness but also implementation feasibility, acceptability, and sustainability in clinical practice.

Implications for Nursing Practice and Policy

For frontline nurses caring for patients with chronic illnesses, the findings of this umbrella review will enable them to identify patient risks and effectively integrate AI-based techniques into clinical nursing practice. Nurses and NPs can use the results of this umbrella review as evidence for practice in chronic illness care provision. The ICN emphasizes that nurses should participate in and be educated on the development of AI-based interventions in health care systems and implement these AI technologies in their clinical practice to achieve better patient outcomes and patient safety [2]. Evidence exists on risk predictions, hospital utilization, and health care costs in AI-based nursing interventions for chronic illnesses. Thus, nurses need to explore more patient outcomes, including the investigation of inconclusive psychological outcomes of chronic illnesses. Furthermore, it is recommended that nurses should implement and lead the application of AI in clinical practice to improve patient-centered care [2].

Nursing leaders in global nursing organizations prioritize AI in nursing research, education, and practice [1,2]. The ICN has announced the importance of digital technology education in nursing and nurses’ competency in using AI systems [2]. They must continuously leverage the empowerment of the nursing workforce within an AI health care system. They require stable support in nursing research, practice, and policy development domains to continuously realize the value of AI in clinical practice.

Conclusion

This umbrella review is the first to synthesize recent SRs examining the impact of AI-based nursing interventions on chronic illnesses in health care settings. The findings suggest that AI-based nursing interventions may have potential benefits for predictive outcomes and for reducing unplanned hospital utilization. However, evidence regarding psychosocial outcomes remains conflicting and inconclusive. These results should be interpreted cautiously due to the limited number of included reviews and the heterogeneity across studies. Therefore, further high-quality studies are needed to strengthen the identified evidence and optimize nurse-led AI-based interventions across different outcome domains in clinical practice.