Impact of artificial intelligence integrations on empathy in healthcare interactions between patients and practitioners: protocol for a systematic review and thematic synthesis of qualitative studies Authors: Louisa Conradi, Jeremy Howick, Amber Bennett-Weston Published: 2026 ABSTRACT AbstractIntroductionEmpathic healthcare improves patient satisfaction with care, anxiety and pain, while reducing practitioner burnout. Artificial intelligence (AI) is continuously advancing and changing the context of empathy in healthcare. While AI may improve diagnostic accuracy or help streamline processes to reduce workload, there is a concern about how AI will impact human patient–practitioner relationships. However, patient and practitioner experiences of AI in healthcare are underexplored. We therefore aimed to synthesise the findings of qualitative studies which explore patient and practitioner experiences of empathy in AI-supported encounters in healthcare.Methods and analysisWe will include any qualitative study in which patient or practitioner experiences of empathy with AI-assisted healthcare are explored. Secondary studies, quantitative studies and those exploring other stakeholders’ experiences will be excluded. The search will include records from database inception in any language. The search strategy is based on the Population, Phenomenon of Interest, Context framework, built around the keywords: artificial intelligence, empathy, healthcare professionals and patients. The following databases will be searched: MEDLINE, Scopus, APA PsycINFO and CINAHL. Additionally, grey literature searches in BASE and OpenAIRE. Forward and backward citation chasing will also be conducted. Records will be screened by two independent reviewers, data extraction will be conducted by one reviewer and checked by another. The risk of bias assessment will be conducted in duplicate using the Joanna Briggs Institute appraisal tool for qualitative studies. The results will be synthesised using thematic synthesis. The number of records identified from the search and the exclusions to reach the total number of included records will be presented in a Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram. The included studies will be listed, along with summaries of relevant study characteristics and risk of bias assessments. Confidence in the evidence will be assessed using the Grading of Recommendations Assessment, Development and Evaluation - Confidence in the Evidence from Reviews of Qualitative research framework.Ethics and disseminationThe systematic review will include only previously anonymised data from primary studies. For this reason, ethical approval is not required for this study. Dissemination of the findings of the final systematic review will occur through publishing in a peer-reviewed journal.PROSPERO registration numberCRD420261301427. STRENGTHS AND LIMITATIONS OF THIS STUDY The systematic review will provide the most up-to-date insight into the experience of empathy in artificial intelligence-assisted healthcare. The systematic review will provide the most up-to-date insight into the experience of empathy in artificial intelligence-assisted healthcare. This review will include both patient and practitioner views, providing a more holistic view on the experience of empathy in this context. This review will include both patient and practitioner views, providing a more holistic view on the experience of empathy in this context. The search will be conducted in several large databases to attempt to collect all relevant studies, although due to limited access to databases from Asia and Africa, relevant records from these regions may be missed. The search will be conducted in several large databases to attempt to collect all relevant studies, although due to limited access to databases from Asia and Africa, relevant records from these regions may be missed. A limitation of this review is that the amount of research on this topic is likely to increase rapidly and literature emerging after the search but prior to the dissemination of the findings of this study may be substantial. A limitation of this review is that the amount of research on this topic is likely to increase rapidly and literature emerging after the search but prior to the dissemination of the findings of this study may be substantial. Introduction Therapeutic empathy is an important element in patient-centred approaches,1 to support patient–practitioner relationships and build trust.2 3 More empathic healthcare delivery is associated with better clinical and patient-centred health outcomes.4 5 Empathy has been increasingly recognised as important over recent years, as patient-centred care has become more central to healthcare delivery.68 The nature and components of empathy have been a topic of debate since the introduction of the construct in philosophy.9 While empathy is now widely accepted as a construct that can be studied and measured, there have been debates about the nature of empathy as an automatic or deliberate process.1012 A recent systematic review and thematic analysis found that the differences in how empathy has been conceptualised in the healthcare setting may have been exaggerated.13 The analysis found that therapeutic empathy involves exploring, seeking to understand, feeling, and therapeutic action, all while maintaining professional boundaries.14 Meanwhile, artificial intelligence (AI) technology is expanding into healthcare, in ways that affect communication in general and empathy in particular. AI is being used in various ways, including consultations and decision-making to assist human healthcare professionals.15 16 With the rapid application of this technology in patient care, concerns have arisen over how AI may impact patients’ and practitioners’ experiences of empathy in consultations.1722 Although a recent meta-analysis found that AI written responses are often rated as more empathic than human practitioner written responses,23 little is known about how patient and practitioner experiences of AI in therapeutic settings. One recent review examined perceptions of the impact of AI on healthcare22; however, it included mostly opinion pieces and editorials and thus lacked insights from primary sources. Another recent review21 investigated the beliefs and experiences of various stakeholders in regard to AI in healthcare; however, this qualitative synthesis did not include grey literature or papers not written in English. Moreover, none of the previous studies in this field examine the relationship between AI integration and empathic healthcare delivery specifically. Additionally, and relatedly, no previous reviews have synthesised patients’ and practitioners’ experiences of AI and empathy in healthcare in the same study. Given that empathy is relational, it is important for a single study to combine the views of these two important stakeholder groups. Finally, the searches for these reviews were conducted in 2022; since then, several new studies exploring patients’ and practitioners’ experiences of AI and empathy in healthcare have been published.2427 This review aims to address these gaps by synthesising the evidence from empirical studies on the populations who will be most affected by the rapidly growing use of AI in healthcare: patients and healthcare practitioners. The findings of this review will demonstrate how AI integrations are perceived to enhance or obstruct empathy in healthcare encounters by patients and practitioners. Aims and objectives The overarching aim of the proposed systematic review is to synthesise the findings of qualitative studies exploring patients’ and healthcare professionals’ experiences of empathy in AI-supported encounters in healthcare. To achieve the aim, the specific objectives for this systematic review are: Identify common themes in the patients’ experience and perception of empathy in AI-supported healthcare encounters. Identify common themes in the patients’ experience and perception of empathy in AI-supported healthcare encounters. Identify common themes in the practitioners’ experience and perception of empathy in AI-supported healthcare encounters. Identify common themes in the practitioners’ experience and perception of empathy in AI-supported healthcare encounters. Methods and analysis Reporting guidelines This systematic review protocol is reported in line with the Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols (PRISMA-P)28; the completed checklist for this protocol can be found in online supplemental file 1. The systematic review will be reported in line with the Enhancing transparency in reporting the synthesis of qualitative research statement29 for reporting the synthesis of qualitative research. Eligibility criteria Study designs Studies will be eligible for inclusion if they are primary qualitative or mixed-methods studies. Mixed-methods studies must report qualitative findings separately from quantitative findings to be eligible. Population The population or sample must be patients or qualified healthcare practitioners. For healthcare practitioners, this includes any healthcare professional, such as doctors, nurses, midwives, dentists, mental health practitioners and allied healthcare professionals. Studies with the population of support workers, scientists without a clinical role, management roles and other stakeholders will not be included. Phenomenon of interest and outcome To be eligible, the phenomenon of interest of the study must be experiences or perceptions of empathy in AI-assisted healthcare. Therefore, studies that examine experiences and perceptions of empathy in healthcare settings without the use of AI will not be included. For this study, the term ‘artificial intelligence’ (AI) will be used to mean any form of AI that is used to support healthcare. This includes machine learning AI’s (unsupervised and reinforcement learning algorithms), natural language processing AI’s, neural networks, and deep learning algorithms. The applications which are eligible include specific systems (assisted intelligence and automated intelligence) and adaptive systems (augmented intelligence, autonomous intelligence).3032 Eligible AI assistance may be in note taking, diagnostic assistance or medical imaging interpretation. To examine how AI integrations influence empathy between patients and healthcare practitioners, only studies in which AI is assistive will be included; this excludes studies in which AI serves as the sole provider and there is no interaction between patients and healthcare practitioners. This review will use the definition of therapeutic empathy, as provided in the introduction.14 The term “empathy” (or related terms such as “empathic” or “empathetic”) must be explicitly mentioned in the methods or results section for the studies to be included. The definition of empathy in the included records may differ from the definition provided above; this is not grounds for exclusion. The definitions of empathy provided in the included studies will be checked against the definition provided above. Context The setting of the study must be in healthcare for the study to be included, including consultations, discussions of care plans and treatment sessions. Language Studies in all languages will be included to allow identification of as many studies as possible. Translation of the studies will be conducted by bilingual team members and support from outside the team and technology, as required. Report characteristics Unpublished studies will be considered. All records from database inception will be eligible,33 since no such review has been conducted yet and no clear ‘start date’ of AI assistance in healthcare can be determined. The eligibility criteria are presented following the Population, Phenomenon of Interest, Context (PICo) format for qualitative studies34 in table 1. Table 1Eligibility criteriaElementInclusionExclusionPopulationPatients, healthcare practitionersOther stakeholdersPhenomenon of interestExperiences of empathy–ContextAI-assisted healthcare–Study designQualitative studies, mixed-methods studies where qualitative data is reported separatelySystematic reviews, narrative reviews, quantitative studiesLanguageAny–AI, artificial intelligence. Eligibility criteria AI, artificial intelligence. Information sources The search will be conducted primarily in databases, including MEDLINE, Scopus, APA PsycINFO and CINAHL. To increase the likelihood of finding all relevant research, references of relevant studies will be searched and authors of relevant papers contacted to find more potentially relevant studies. Hand searches and grey literature searches in BASE, and OpenAIRE, will be conducted. Grey literature will be included, and where appropriate, authors of included studies may be contacted to ask about similar studies, in line with the recommendations for qualitative systematic reviews and methodology of thematic synthesis.3538 Search strategy The search strategy was developed based on the PICo framework for qualitative studies34 using MeSH terms for databases and keywords based on preliminary searches, built around the keywords: artificial intelligence, empathy, healthcare professionals and patients. The strategy will be adapted for each database after it is finalised. The search strategy was based on two filters from the ISSG Search Filter Resource.39 40 A draft of the search strategy for MEDLINE can be found in online supplemental file 2. Study records Selection process This process will be summarised in a PRISMA flow diagram. The records identified by the search will be collected in EndNote V.21, and duplicates will be removed. Two reviewers will screen the records independently in two phases: title and abstract screening, followed by a full-text screen. Differences will be resolved through discussion. Data management The search results will be transferred from EndNote V.21 to Rayyan for screening. Data collection process The data extraction will be conducted using the extraction form in online supplemental file 3, in duplicate by two reviewers. Data items Outcomes and prioritisation The primary outcome will be themes emerging from patients’ or practitioners’ experiences and perceptions of empathy with AI-assisted healthcare. General The data extraction will include study type, sample, setting/context, philosophical assumptions, data collection methods, data synthesis methods, limitations or conflicts of interest and findings. If a definition of empathy is provided in the study, this will also be extracted and presented in a table in the full report. The extraction of findings will include data from any text labelled ‘results’ or ‘findings’. If mixed-methods studies are included, only the sections relevant to the qualitative substudy will be extracted. Risk of bias assessment The Joanna Briggs Institute tool for appraisal of qualitative studies41 will be used to assess the included studies at the study level. No studies will be excluded based on the quality score to allow for sensitivity analysis to explore the effects of omitting studies with high risk of bias. In the discussion, the results of the risk of bias assessment will be taken into consideration. The risk of bias assessment will be conducted in duplicate by two reviewers. Any discrepancies will be resolved through discussion.42 Study quality will be discussed, and quality scores will be presented in a table. Dissemination bias is likely to impact the findings. Grey literature searches will help to reduce this bias.35 43 Synthesis methods The results will be synthesised using thematic synthesis as described by Thomas and Harden,36 as the preliminary searches suggest that this will be most suitable for the type of data in studies which will be included; additionally, the approach is commonly used in the field of healthcare research.44 First, the findings of the included studies will be free coded in line-by-line coding. All text included in ‘results’ or ‘findings’ will be considered for coding. This process will be started by two reviewers to build an initial set of codes, which will be agreed on. One reviewer will then code the rest of the data. This coding will be checked by another reviewer. Following coding, the codes will be constructed into descriptive themes. After discussion between the reviewers, these descriptive themes will be developed into analytical themes. If sufficient suitable studies are identified and it is feasible, several sensitivity analyses will be conducted,45 such as synthesis of studies above a certain quality threshold and separate syntheses for studies with patient populations and studies with practitioner populations to provide more context-specific insights.46 A section on reflexivity will be included with a discussion of the background and roles of the review team members to transparently explore how the researchers’ subjectivity and context influence the review. This practice will allow readers to see how the biases the researchers bring to the review affect the construction of knowledge and interpretation of the data. Certainty assessment The Grading of Recommendations Assessment, Development and Evaluation–Confidence in the Evidence from Reviews of Qualitative research framework will be used to assess the quality and certainty of the evidence.47 48 Protocol deviations Any deviations from the protocol will be appropriately documented and dated, and the reasoning for the deviations will be provided. Patient and public involvement This protocol was developed without patient and public involvement and engagement. A patient and public involvement group will be consulted at the point of data analysis of the final review. Their input will provide further context for the authors prior to the finalising of themes. The input of the group will also be taken into consideration for the interpretation of the findings. Ethics and dissemination Ethics approval is not necessary for this project, as all data used are publicly available and have been previously anonymised. Dissemination of the findings of the final systematic review will occur through publication in a peer-reviewed journal. Supplementary material