Generative AI in healthcare: redefining clinical practice through digital transformation Authors: Jackie Zhanbiao Li, Xin Li, Qifeng Wu, Qianhui Ting, Yingqian Lao Published: 2026 ABSTRACT AbstractAlthough generative AI (GenAI) has transformative potential for healthcare systems, this timely issue has not yet been sufficiently explored. This study investigates the application effectiveness and underlying mechanisms of GenAI in the healthcare sector, using a dataset comprising 1,440 samples from 160 hospitals across 31 provinces in China, covering the period from 2021 to 2024. The results show that GenAI significantly enhances healthcare effectiveness, including improving diagnostic accuracy, optimizing operational efficiency, and increasing patient satisfaction. The degree of industrial digitalization positively moderates these effects, with highly digitalized regions experiencing greater AI effectiveness. This study not only enriches the theoretical framework of technology diffusion but also provides valuable insights into the practical implementation and policy optimization of GenAI in healthcare.Keywords: Generative AI, Healthcare, Industrial digitalization, Economic development, Technological infrastructure Introduction Generative Artificial Intelligence (GenAI) is rapidly reshaping healthcare by leveraging deep learning and neural network architectures to extract complex patterns from multimodal clinical data and generate high-quality outputs. Its impact is increasingly visible across diagnosis, treatment planning, workflow optimization, and drug discovery. For example, transformer-based large language–vision models have improved early lung cancer detection accuracy by 12% relative to conventional networks [5]. Conditional Generative Adversarial Networks (GANs) have accelerated MRI reconstruction, reducing scan time by 40% while preserving diagnostic fidelity [25, 47]. GPT-4-based transformer models integrated into EMR systems have saved physicians an average of 25 min per patient in documentation time while improving coding consistency [36, 90]. At the system level, retrieval-augmented models have managed more than one million triage requests [56, 107], while generative diffusion models have shown promise in drug discovery by producing fibrosis candidate molecules in only 18 months at lower cost than conventional pipelines [29, 48]. These advances are driven by the close alignment between model families and task requirements. Transformer-based models and retrieval-augmented pipelines are particularly effective for text generation and clinical decision support [59, 64], GANs perform well in realistic image synthesis [6], and diffusion models and VAEs are increasingly important in molecular generation [2, 10]. Yet, despite this rapid technological progress, the real-world value of GenAI is not determined by algorithmic capability alone [20, 30]. Its effectiveness depends critically on whether the surrounding healthcare and industrial environment can support data integration, model deployment, system interoperability, and continuous optimization [1, 68]. In this respect, the degree of industrial digitalization is especially important. It shapes the quality and accessibility of healthcare data, the maturity of digital infrastructure, and the institutional capacity to integrate GenAI into clinical workflows [72, 83]. In other words, the successful implementation of GenAI depends not only on the technology itself, but also on broader digital foundations such as data governance, interoperability, and platform connectivity [7, 72]. This makes GenAI an ideal case for examining how emerging technologies interact with uneven digital ecosystems to produce heterogeneous organizational outcomes. China provides an especially important context for addressing this question. From a global perspective, China combines one of the world’s largest healthcare systems with substantial regional variation in digital infrastructure, industrial upgrading, and healthcare resource allocation [32, 66, 67]. This combination makes China particularly suitable for examining not only whether GenAI improves healthcare outcomes, but also under what structural conditions such improvements are more likely to occur [100, 103]. On the one hand, leading regions such as Beijing, Shanghai, and Guangdong possess advanced digital infrastructures and increasingly interoperable systems, creating favorable conditions for AI deployment. On the other hand, institutions in central and western regions continue to face fragmented data architectures and limited compatibility [33, 35, 55]. This within-country heterogeneity offers a stronger empirical basis for identifying the moderating role of industrial digitalization than studies focused on more digitally homogeneous contexts. Regional disparities in China also extend to policy support, industrial clustering, and innovation ecosystems. Provinces in the Yangtze River Delta and the Greater Bay Area benefit from mature AI industries and stronger innovation synergies, whereas less developed regions face constraints in technological investment and digital transformation capacity [38, 104]. Differences in economic development and healthcare resource allocation further amplify these patterns: wealthier regions are better positioned to adopt AI-assisted diagnostic systems, while resource-constrained areas often face funding and data-access limitations [76, 80, 87]. Accordingly, China is not merely an empirical setting of convenience,it is a theoretically relevant context in which the interaction between GenAI and industrial digitalization can be observed with unusual clarity. Despite growing scholarly interest in GenAI in healthcare, existing studies have focused primarily on localized technical applications and short-term performance gains [22, 73, 102, 108]. Few have systematically examined the mechanisms through which GenAI influences healthcare outcomes, particularly the moderating role of industrial digitalization under conditions of marked regional heterogeneity. This limits theoretical understanding of how GenAI creates value and restricts the evidence base available to policymakers and practitioners. To address this gap, this study examines two questions: (1) How does GenAI improve healthcare outcomes, including diagnostic accuracy, operational efficiency, and patient satisfaction? and (2) How does industrial digitalization moderate the relationship between GenAI and healthcare outcomes? By integrating multi-source data and empirical analysis, the study clarifies the mechanisms through which GenAI affects healthcare performance and quantifies its impact on key outcome dimensions. In doing so, it contributes not only to the literature on AI in healthcare, but also to a broader understanding of how the value of emerging technologies depends on the digital maturity of the industrial environment in which they are embedded. More specifically, the study makes two contributions: first, it deepens theoretical understanding of GenAI’s multidimensional effects in healthcare; second, by introducing industrial digitalization as a moderator, it provides new insight into the relationship between technological innovation and industry-level digital transformation. Theory and research hypotheses Depth of generative AI applications and healthcare effectiveness Generative Artificial Intelligence (GenAI) enhances healthcare outcomes through a set of interrelated mechanisms, where the depth of GenAI application directly correlates with measurable improvements in clinical workflows and patient-facing services. The "depth of application" refers to the extent and integration of GenAI technologies across various healthcare domains, such as diagnostics, patient communication, and clinical decision-making. When applied at a deeper level, involving widespread integration across imaging diagnostics, electronic medical record (EMR) generation, intelligent consultations, and treatment optimization, GenAI can significantly enhance its capacity to fuse multimodal data and support comprehensive clinical decision-making. For example, in medical imaging, GenAI can analyze large-scale CT and MRI datasets to identify lesions and generate detailed diagnostic reports, achieving accuracy that in some cases matches or even surpasses that of human experts. At deeper application levels, GenAI aids in personalized medicine, such as tailoring treatment plans by integrating medical history, genetic data, and real-time monitoring, improving institutional resource management, reducing human error, and demonstrating systemic advantages in complex case management [24]. Automation of routine documentation and reporting: GenAI can generate high-quality clinical reports, discharge summaries, and patient consultation notes, significantly reducing physicians' administrative workload. A mixed-method study involving 38 clinicians found that automated systems significantly shortened documentation time while maintaining quality, with physicians preferring moderate automation under human supervision [94, 96, 108]. Automation of routine documentation and reporting: GenAI can generate high-quality clinical reports, discharge summaries, and patient consultation notes, significantly reducing physicians' administrative workload. A mixed-method study involving 38 clinicians found that automated systems significantly shortened documentation time while maintaining quality, with physicians preferring moderate automation under human supervision [94, 96, 108]. Optimization of physician–patient communication: GenAI can quickly generate clear and understandable responses for medication refills, test results, and routine inquiries, reducing response latency and improving communication timeliness and clarity, which alleviates physicians' cognitive burdens and improves patient satisfaction [4, 23, 82]. Optimization of physician–patient communication: GenAI can quickly generate clear and understandable responses for medication refills, test results, and routine inquiries, reducing response latency and improving communication timeliness and clarity, which alleviates physicians' cognitive burdens and improves patient satisfaction [4, 23, 82]. Clinical decision support and error reduction: GenAI, when deployed as a decision-support tool, can validate diagnostic findings, suggest alternative hypotheses, and flag rare or atypical cases. Peer-reviewed studies indicate that properly integrated systems can reduce diagnostic errors and shorten diagnostic delays in complex clinical settings [94]. Clinical decision support and error reduction: GenAI, when deployed as a decision-support tool, can validate diagnostic findings, suggest alternative hypotheses, and flag rare or atypical cases. Peer-reviewed studies indicate that properly integrated systems can reduce diagnostic errors and shorten diagnostic delays in complex clinical settings [94]. Physician-Supervised Conversational Agents for Patient Interaction: In resource-limited settings, physician-supervised conversational agents improve healthcare accessibility and patient comprehension, enhancing satisfaction without compromising safety or empathy [79, 93, 106]. Physician-Supervised Conversational Agents for Patient Interaction: In resource-limited settings, physician-supervised conversational agents improve healthcare accessibility and patient comprehension, enhancing satisfaction without compromising safety or empathy [79, 93, 106]. These mechanisms collectively shorten diagnostic and treatment cycles, reduce preventable errors, and enhance patient experience. Importantly, the effectiveness of these benefits is contingent upon appropriate clinical oversight, rigorous validation procedures, and seamless integration into existing workflows. Deeper applications of GenAI, which are broadly integrated across multiple domains, enhance its capacity to support complex clinical decision-making [77], optimize personalized care, and improve healthcare efficiency, especially in public health emergencies [24]. GenAI enables real-time resource allocation and optimization of strategies through large-scale data analysis [11, 44, 65]. Moreover, deeper applications improve accessibility in remote or resource-constrained areas by reducing diagnostic and treatment delays [28, 105]. Based on these theoretical and empirical considerations, this study proposes the following hypothesis H1: H1: The depth of GenAI applications significantly enhances healthcare outcomes. H1: The depth of GenAI applications significantly enhances healthcare outcomes. Moderating effect of industrial digitalization The rapid development of generative artificial intelligence (GenAI) in the healthcare sector is largely due to its technological advantages. As a cutting-edge technology grounded in deep learning and generative modeling, the depth of GenAI application refers to the extent and integration of GenAI across healthcare processes, such as diagnostics, treatment planning, and resource allocation. This depth determines how effectively GenAI mechanisms can be translated into improved health outcomes. GenAI enables efficient and precise solutions by simulating and optimizing complex datasets. For instance, in medical imaging, GenAI can accurately identify early-stage lesions and provide detailed diagnostic recommendations, significantly enhancing diagnostic accuracy and efficiency. By analyzing large-scale CT and MRI datasets, GenAI can detect pathological features and generate comprehensive diagnostic reports, achieving accuracy that matches—or even exceeds—that of human experts [34, 78]. In drug discovery, GenAI generates potential molecular structures and accelerates virtual screening, thus shortening development cycles and reducing costs [16, 97]. However, the practical impact of GenAI varies considerably across regions and healthcare systems, with the degree of industrial digitalization being a critical moderating factor. Industrial digitalization refers to the extent to which a healthcare system has adopted digital technologies, integrated data, and optimized workflows. According to technological diffusion theory, industrial digitalization influences the effectiveness of emerging technologies by shaping the environment for their adoption and diffusion. Highly digitalized healthcare systems typically possess advanced infrastructure, robust data integration capabilities, and greater technological adaptability, thus creating favorable conditions for GenAI deployment [31, 42]. First, such systems can efficiently collect, store, and process medical data, providing abundant and high-quality training datasets for GenAI models [84]. Second, healthcare institutions with higher levels of digitalization often operate advanced information platforms that enable the seamless integration of GenAI into clinical workflows, improving both diagnosis and patient management efficiency [43, 49]. In contrast, healthcare systems with lower levels of digitalization often face fragmented infrastructures, poor data interoperability, and insufficient technical support, limiting GenAI’s potential [14, 98]. In addition to infrastructure, industrial digitalization further promotes technology diffusion through real-time data sharing, cloud-based collaboration, and standardized communication protocols. Furthermore, highly digitalized ecosystems are often supported by favorable policy frameworks and driven by strong market demand, creating a positive feedback loop that accelerates technology adoption and iteration [89]. For example, in countries undergoing healthcare digitalization reforms, GenAI has been positioned as a core technology, leading to significant progress in medical imaging diagnostics and personalized medicine. This demonstrates how digitalization amplifies the impact of GenAI technologies [26, 41, 52]. Based on these theoretical and empirical insights, this study proposes the following hypothesis H2: H2: The degree of industrial digitalization significantly moderates the positive relationship between GenAI applications and healthcare outcomes. H2: The degree of industrial digitalization significantly moderates the positive relationship between GenAI applications and healthcare outcomes. The research framework of this study is shown in Fig. 1. Conceptual framework Methods Sample source and selection The dataset used in this study consists of 160 hospitals across 31 provincial-level administrative regions in China, yielding 1,440 hospital-year observations from 2021 to 2024. This dataset was selected because it combines broad geographic coverage, hospital-level variation, and multi-source information on digital transformation, enabling a more comprehensive assessment of how GenAI adoption affects healthcare outcomes across regions. Compared with alternative sources, such as single-province surveys, case-based hospital records, or purely firm-level technology datasets, the present dataset is better suited to the research questions because it jointly captures hospital technology adoption, regional industrial digitalization, and healthcare performance within one analytical framework. Data were compiled from four authoritative sources. First, national and provincial statistical databases provide standardized indicators of healthcare investment and industrial digitalization and are widely recognized as reliable in policy and academic research [85]. Second, audited annual reports of listed healthcare technology firms and related hospital disclosures provide verifiable information on technology deployment and performance, offering greater reliability than unaudited corporate materials [86]. Third, industry research reports were used to capture AI adoption trends and healthcare-sector developments not fully observable in administrative data,only reports with transparent methods were retained [3]. Fourth, publicly available information from technology media platforms such as TechCrunch, VentureBeat, and 36Kr was used only as supplementary contextual evidence and cross-validated against primary sources [41]. The 2021–2024 period was chosen for both theoretical and empirical reasons. It captures the accelerated development, commercialization, and institutional uptake of GenAI following major advances in large language models, while also providing a timely window for observing hospitals’ responses to AI-enabled transformation. Earlier years are less suitable because GenAI applications were still limited and reporting practices were less consistent. Data Collection and Screening Procedures: Strict data screening measures were implemented to retain only relevant and reliable information [9, 69]. Hospitals with missing key variables or hospitals unrelated to GenAI healthcare technology were excluded. Outliers were treated, and data processing was conducted using Excel 19.0 and Stata 18.0. After screening, 1,440 valid observations were retained. The final sample covers hospitals distributed across all 31 provincial-level administrative regions, ensuring strong regional heterogeneity and national representativeness. This study therefore provides a comprehensive analysis of how GenAI technology adoption in hospitals affects healthcare outcomes across regions in China. Methods employed To improve the rigor and transparency of variable construction, this study utilized expert scoring for composite indicators of healthcare effectiveness (HE), depth of generative AI application (AID), and the degree of industrial digitalization, ensuring a clear and justifiable methodology. While initially relying on the Delphi method in combination with the Nominal Group Technique (NGT), the study moved to a more objective data-driven approach to mitigate concerns over potential biases and transparency issues. The Delphi method was used for its ability to gather informed judgments from experts across different regions, maintaining anonymity across multiple rounds to reduce the risk of dominant individual influence. However, recognizing the limitations of expert-based methods, objective operational data were prioritized wherever possible to validate these subjective assessments. This allowed for a more empirical foundation for the composite indices, particularly in healthcare settings where direct observation may be limited [19, 27, 99]. The Nominal Group Technique (NGT) facilitated structured discussions in the first phase to generate and refine indicators, contributing to the early development of measurement items. To counter potential biases such as groupthink and dominance by vocal members, diverse data sources were employed in combination with expert feedback to enhance the validity of the indicators and avoid over-reliance on consensus alone [8, 81, 91]. While expert-based consensus methods can be prone to biases, the study introduced statistical controls, such as monitoring convergence in Delphi rounds and ensuring equitable participation in NGT, to safeguard against these issues. This combination of expert judgment and objective data strengthened the overall methodology, improving the robustness and transparency of the study's conclusions regarding the effectiveness of GenAI in hospital settings [13, 40, 71, 74]. The study found that expert-based composite scores were both conceptually valid and based on empirical evidence, with the use of both the Delphi and NGT methods and careful attention to the limitations of any approach based on consensus. Variable description The dependent variable in this study is healthcare effectiveness (HE). Building on previous research, healthcare effectiveness was measured using indicators, such as diagnostic accuracy, operational efficiency, and patient satisfaction [18, 46, 54, 61]. These indicators were scored using the nominal group technique and the Delphi method. After data processing, the values of the dependent variable were obtained (see Table 1). Table 1Measurements of variablesVariable typeVariable nameSymbolMeasurement methodDependent variableHealthcare effectivenessHERepresented by a composite score (0–100) that includes diagnostic accuracy, operational efficiency, and patient satisfactionIndependent variableDepth of GenAI applicationAIDDepth of GenAI application, measures the level of integration of GenAI in healthcare (0–100)Moderating variableDegree of industrial digitalizationDIDDegree of industrial digitalization, is quantified using a digitalization index and informatization coverage rate within the healthcare industry (0–100)Control variablesPolicy Support intensityPSThis variable captures the strategic commitment of local governments to promote the adoption of GenAI in the healthcare sector. It is measured based on the number of official policy documents, financial subsidies, and the implementation of pilot programs. The resulting index is standardized within a range of 5% to 20%, reflecting the intensity of institutional support across regionsMarket demand levelMDMarket demand level is measured by the growth rate of market demand and user activity for GenAI in healthcare (0–100)Data complianceDCData compliance reflects the strictness of regional data privacy and compliance regulations (0–100)Organizational scaleOSOrganizational scale is measured by the average number of hospital beds and technical staff in healthcare institutions (0–100)R&D expenditure ratioRDSRegional R&D expenditure as a percentage of GDP (5–20%)Population agingAPRegional population aging index (30–90)Medical insurance coverageICMedical insurance coverage rate within the region (50–100%)Healthcare expenditure ratioHSHealthcare expenditure as a percentage of GDP (10–50%)YearYearThe model incorporates dummy variables for the years from 2021 to 2024This study employed the nominal group technique and the Delphi method to generate composite scores for the core variables: HE, AID, and DID. A panel of 12 experts—representing the fields of medical services, artificial intelligence research and development, and policy analysis—participated in the evaluation. A five-point Likert scale was used, and the scoring process was conducted over three rounds, with a seven-day interval between each round. Consensus was defined as scores falling within ± 1 standard deviation of the mean. Final scores for each expert were calculated as the mean of their responses and subsequently normalized to a 0–100 scale. The internal consistency of the expert evaluations was assessed using Cronbach’s α, yielding values between 0.82 and 0.91, which indicates high reliability and strong inter-rater agreement Measurements of variables This study employed the nominal group technique and the Delphi method to generate composite scores for the core variables: HE, AID, and DID. A panel of 12 experts—representing the fields of medical services, artificial intelligence research and development, and policy analysis—participated in the evaluation. A five-point Likert scale was used, and the scoring process was conducted over three rounds, with a seven-day interval between each round. Consensus was defined as scores falling within ± 1 standard deviation of the mean. Final scores for each expert were calculated as the mean of their responses and subsequently normalized to a 0–100 scale. The internal consistency of the expert evaluations was assessed using Cronbach’s α, yielding values between 0.82 and 0.91, which indicates high reliability and strong inter-rater agreement The independent variable in this study is the depth of GenAI applications (AID). Drawing on previous research, the depth of GenAI application was measured by its level of integration in the healthcare domain [50, 51, 88]. These indicators were scored using the nominal group technique and the Delphi method. After data processing, the values of the independent variable were derived (see Table 1). The moderating variable in this study is the degree of industrial digitalization (DID). Building on prior research, industrial digitalization was measured using two indicators: the digitalization index of the healthcare industry and the informatization coverage rate [12, 37, 66, 67, 70, 95]. These indicators were scored using the nominal group technique and the Delphi method. After data processing, the values for the moderating variable were obtained (see Table 1). This study draws upon existing research literature to select control variables that influence healthcare outcomes, including policy support intensity (PS), market demand level (MD), data compliance (DC), organizational scale of medical institutions (OS), proportion of R&D expenditure (RDS), degree of population aging (AP), medical insurance coverage rate (IC), proportion of healthcare expenditure (HS), and yes (Year) [39, 45, 53, 63]. These variables are chosen due to their demonstrated impact on healthcare outcomes (see Table 1). Model design To test research hypothesis H1, this study constructs the following regression model (1): In model (1), HE represents the dependent variable, indicating healthcare effectiveness; AID serves as the independent variable, reflecting the depth of GenAI application; Control denotes a series of control variables; and Year accounts for fixed effects, representing time-fixed factors. Lastly, α represents the random disturbance term, with subscripts i and t referring to individual institutions and time periods, respectively. To test research hypothesis H2, the following regression model (2) was constructed: In model (2), HE represents the dependent variable, indicating healthcare effectiveness; AID serves as the independent variable, reflecting the depth of GenAI applications; and DID acts as the moderating variable, representing the degree of industrial digitalization. Control refers to a series of control variables, while Year denotes fixed effects, accounting for time-fixed factors. Finally, α represents the random disturbance term, with subscripts i and t referring to individual institutions and time periods, respectively. Results Descriptive statistics Table 2 presents the descriptive statistics of the main variables related to the impact of GenAI application depth on healthcare outcomes. The analysis of Table 2 indicates that the control variables—policy support intensity (PS), market demand level (MD), and organizational scale (OS) of medical institutions—exhibit relatively large standard deviations. The similarity between their mean and median values suggests notable variability among these variables across the sample institutions, potentially influenced by extreme values. In contrast, other variables show mean and median values that are closely aligned, with moderate standard deviations, uniform distribution, and stable data characteristics, indicating no significant variability within the sample institutions. Table 2Descriptive statisticsVariablesObsMeanSDMinMedianMaxHE144075.28814.40150.00076.000100.000AID144065.09120.49530.00066.000100.000DID144059.54223.36920.00060.00099.000EDL144069.67117.50840.00069.500100.000TI144065.98420.06130.00067.000100.000PS144054.59926.37311.00053.00099.000MD144056.47624.69415.00055.000100.000DC144060.54723.43620.00061.000100.000OS144055.33726.42411.00055.000100.000RDS144012.6734.3575.20512.81219.849AP144059.96117.98830.00059.00090.000IC144075.42214.54850.00076.000100.000HS144029.52911.25510.49229.20749.374 Descriptive statistics Correlation analysis This study primarily investigates the impact of GenAI application depth on healthcare outcomes in 160 hospitals in China. To examine the relationship between the depth of GenAI application (AID) and healthcare effectiveness (HE), as well as to determine potential multicollinearity among variables, Pearson correlation analysis was conducted [57, 58]. The results are presented in Table 3. As shown in Table 3, the Pearson correlation analysis indicates that AID is significantly positively correlated with HE at the 0.01 level, aligning with the study's expectations and providing preliminary support for Hypothesis H1. To further assess the presence of multicollinearity, a variance inflation factor (VIF) test was performed for all variables [15, 59, 60]. According to established criteria, a Pearson correlation coefficient greater than 0.8 suggests potential multicollinearity, while a VIF value exceeding 10 indicates the presence of severe multicollinearity [62]. The VIF test results for all variables in this study range from 1.015 to 1.048, which are well below the critical threshold of 10. This confirms that the constructed model does not suffer from serious multicollinearity issues. Table 3Correlation analysisVariablesHEAIDDIDEDLTIPSMDDCOSRDSAPICHSYearHE1AID0.008***1DID−0.027−0.0221EDL0.009−0.027−0.0101TI−0.023−0.005−0.006−0.0011PS0.000−0.001−0.059**−0.032−0.0041MD0.0120.020−0.019−0.0040.0010.0091DC0.0050.007−0.004−0.0080.000−0.0547**0.0261OS−0.006−0.0250.0478*−0.002−0.0320.004−0.0310.0331RDS−0.024−0.0130.0220.031−0.051*−0.048*0.0370.017−0.0231AP0.0000.0150.003−0.0420.0613**0.032−0.031−0.012−0.0190.0081IC0.0330.040−0.0030.0140.040−0.0020.0050.0340.0492*0.0050.0511*1HS−0.0070.0250.042−0.0120.0020.0120.029−0.0200.0060.013−0.009−0.0241Year−0.017−0.0200.0140.0280.0380.006−0.013−0.033−0.021−0.0350.0090.014−0.0261*** p < 0.01, ** p < 0.05, * p < 0.1 Correlation analysis *** p < 0.01, ** p < 0.05, * p < 0.1 Main regression and mechanism tests Table 4 presents the results of the mian regression analysis and and mechanism tests. Column (1) shows the regression results between AID and HE after accounting for fixed time and province effects (FTPEs) but without including control variables. Column (2) reports the regression results between AID and HE with FTPEs and the inclusion of control variables. Column (3) illustrates the regression results between AID and HE without FTPEs or control variables, but with the addition of the moderating variable DID. Column (4) displays the regression results between AID and HE without FTPEs, with the inclusion of control variables and the moderating variable DID. Column (5) presents the regression results between AID and HE with FTPEs, without control variables, and the inclusion of the moderating variable DID. Finally, column (6) provides the regression results between AID and HE with FTPEs, control variables, and the moderating variable DID. Table 4Main regression and mechanism testsVariables(1)(2)(3)(4)(5)(6)HEHEHEHEHEHEAID0.005**0.003**0.005**0.019**0.022***0.020***(2.05)(2.01)(2.00)(2.02)(2.80)(2.85)DID × AID0.0003**0.0003**0.0003***0.0003***(2.80)(2.80)(2.90)(3.00)ControlYESYESYESYESYESYESProvinceYESYESYESYESYESYESYearYESYESNONOYESYESN144014401440144014401440R20.220.230.240.240.220.22Adj. R20.040.050.060.060.040.04Although the effect sizes of R2 and adjusted R2 are not particularly large, the conclusions remain robust after controlling for multiple variables and thus hold practical significance* p < 0.1, ** p < 0.05, *** p < 0.001 Main regression and mechanism tests Although the effect sizes of R2 and adjusted R2 are not particularly large, the conclusions remain robust after controlling for multiple variables and thus hold practical significance * p < 0.1, ** p < 0.05, *** p < 0.001 The results presented in columns (1) and (2) of Table 4 indicate that in model (1), the depth of GenAI applications (AID) is significantly positively correlated with healthcare effectiveness (HE) at the 0.05 level. This finding suggests that the depth of GenAI applications has a significant positive impact on healthcare effectiveness. Specifically, GenAI enhances healthcare outcomes by improving diagnostic accuracy, optimizing operational efficiency, and increasing patient satisfaction. Therefore, the study supports Hypothesis H1. This finding supports Hypothesis H1, confirming that AID has a significant positive impact on HE. The primary reason is that deeper application of GenAI enhances organizational efficiency, facilitates knowledge sharing, strengthens learning capabilities, optimizes resource allocation, and improves patient experiences. Under the influence of technological diffusion theory, these factors collectively contribute to improved HE. The results from columns (3) and (4) and columns (5) and (6) of Table 4 demonstrate the moderating degree of industrial digitalization (DID) on the relationship between AID and HE. In model (2), after introducing the interaction term (DID × AID), the coefficients of the interaction term in columns (3) and (4) are significantly positive at the 0.05 level, and the coefficients of AID's impact on HE in the same columns are also significantly positive at the 0.05 level. Similarly, in columns (5) and (6), the coefficients of the interaction term are significantly positive at the 0.01 level, and the coefficients of AID’s impact on HE are also significantly positive at the 0.01 level. These findings confirm the positive moderating effect of DID on the relationship between AID and HE. Specifically, AI applications in highly digitalized regions exhibit significantly greater effectiveness than those in less digitalized regions, supporting Hypothesis H2. The primary explanation is that higher levels of digitalization provide robust support for technology diffusion by optimizing technological infrastructure, facilitating data flow, reducing technological costs, enhancing absorptive capacity, and reinforcing technological synergy. Consequently, under the moderating effect of high DID, the positive impact of generative AID on HE is significantly strengthened. Robust tests To further assess the robustness of the results, three methods were employed: [21, 75, 101, 109] the lagged variable approach (LVA), the instrumental variable method (IVM), and the sample range adjustment method (SRAM) (see Table 5). These techniques were used to address potential issues of reverse causality and endogeneity. Specifically, LAV introduced a one-year lag of the core independent variable (AID), denoted as AID1, to evaluate its delayed effect on healthcare effectiveness and the moderating effect of DID. The results, presented in columns (1) and (2) of Table 5, are consistent with the main findings of this study and reinforce their validity. IVM introduces a new variable, Z, which is strongly correlated with the actual independent variable (AID) but does not directly influence the dependent variable (HE). In this study, the instrumental variable (IV1) is constructed as the mean value of AID in other regions (excluding the focal region) within the same year, to capture the external pressure of technology diffusion or industry-level influence, without directly affecting the healthcare outcomes in the region under study.Regression analyses were then conducted to examine the relationships between IV1, the interaction term DID × IV1, and HE. The results, reported in columns (3) and (4) of Table 5, are consistent with the main findings and further support the conclusions of this study. SRAM applies a subsample consisting of observations with values below the mean of the dependent variable (HE) to test the effect of AID on HE and to examine the moderating role of DID in the relationship between AID and HE. This approach assesses whether the model's findings remain robust within a lower-bound subsample. The results, presented in columns (5) and (6) of Table 5, are consistent with the primary conclusions of the study. As observed in Table 5, the findings from all three methods align with the original analysis, providing robust evidence supporting Hypotheses H1 and H2. Specifically, the results confirm that the depth of GenAI applications significantly improves healthcare outcomes, and that industrial digitalization positively moderates this relationship. Table 5Robust testsVariables(1)(2)(3)(4)(5)(6)HEHEHEHEHEHEDID0.167**0.249***0.012**(2.10)(2.70)(2.25)AID10.189**2.815***(2.05)(2.90)DID × AID10.045***(2.65)IV10.572***2.691**(2.85)(2.10)DID × IV10.055***(2.60)AID0.074**0.085***(2.15)(2.85)DID × AID < 0.001***(3.20)ControlYESYESYESYESYESYESProvinceYESYESYESYESYESYESYearYESYESYESYESYESYESN1440144014401440714714R20.250.260.270.280.180.19Adj. R20.070.070.080.080.030.03Although the effect sizes of R2 and adjusted R2 are not particularly large, the conclusions remain robust after controlling for multiple variables and thus hold practical significance* p < 0.1, ** p < 0.05, *** p < 0.001 Robust tests Although the effect sizes of R2 and adjusted R2 are not particularly large, the conclusions remain robust after controlling for multiple variables and thus hold practical significance * p < 0.1, ** p < 0.05, *** p < 0.001 Discussion Main findings This study systematically analyzed the application effects of GenAI in the healthcare sector and explored the moderating roles of industrial digitalization. The results indicate that the depth of GenAI applications has a significant positive impact on healthcare effectiveness. Furthermore, the moderating effects of industrial digitalization on the effectiveness of GenAI applications were found to be significant. Below is a detailed expansion and discussion of the study’s findings. The findings demonstrate that the depth of GenAI applications plays a critical role in optimizing medical resource allocation, improving diagnostic and treatment efficiency, and enhancing patient experiences. Compared to traditional medical technologies, GenAI leverages deep learning and complex algorithms to process massive datasets with precision, providing more accurate diagnostic recommendations and enabling breakthroughs in disease prevention and treatment optimization. These results align with prior research by Blezek et al. [17] and Soenksen et al. [92], which highlighted the role of GenAI in optimizing clinical workflows and processing multimodal medical data. This study further emphasizes that the effects of deep GenAI applications extend beyond specific medical scenarios, achieving systemic improvements across diverse healthcare tasks through comprehensive integration. This broad impact provides a robust theoretical and empirical foundation for the widespread adoption of GenAI in the healthcare industry. The degree of industrial digitalization has been identified as a significant moderating role in the relationship between the depth of GenAI applications and healthcare outcomes. This finding indicates that high levels of industrial digitalization play a crucial role in supporting the efficient operation of GenAI by optimizing the technological environment. On one hand, advanced information infrastructure and efficient data processing capabilities enable seamless integration of GenAI in highly digitalized settings. On the other hand, highly digitalized healthcare systems exhibit greater technological adaptability and user acceptance, thereby facilitating the diffusion of GenAI technologies. This perspective aligns with the findings of previous scholar [70], who observed a strong correlation between digital maturity and the practical effectiveness of technological innovation. Extending this theoretical framework, the current study highlights the amplifying effect of digitalization on the outcomes of GenAI applications. From a policy perspective, enhancing industrial digitalization is not only a necessary condition for supporting GenAI applications but also a critical pathway for improving the overall efficiency of healthcare systems. Research implications This study holds significant theoretical and practical implications. By constructing a systematic analytical framework, it expands the research scope of GenAI in the healthcare sector and provides a new perspective on technology diffusion theory. More importantly, it deepens the understanding of how contextual factors shape the effectiveness of emerging technologies by explicitly incorporating industrial digitalization as a boundary condition. For the first time, this study introduces industrial digitalization as a moderating variable to systematically explore its impact on AI application effectiveness, thereby advancing the literature from a purely adoption-oriented perspective toward a contingency-based and context-sensitive analytical framework. This approach addresses a critical gap in research on technology–industry integration. From a practical perspective, this study provides targeted guidance for policymakers, healthcare institutions, and technology enterprises. In regions with high degree of industrial digitalization, efforts should focus on accelerating the deep integration of AI with digital technologies to drive intelligent healthcare upgrades and improve medical resource allocation efficiency. In addition, policymakers in these regions should emphasize institutional coordination, data interoperability, and governance mechanisms to maximize the systemic benefits of AI deployment. Conversely, in less developed regions, priority should be given to improving infrastructure, enhancing data management capabilities and network coverage, and promoting cross-regional collaboration and resource sharing to facilitate the efficient deployment and application of GenAI, thereby increasing its adaptability and effectiveness. This also implies the need for staged development strategies, where foundational digital capacity building precedes large-scale AI implementation. Furthermore, healthcare institutions should expand AI applications in medical imaging, intelligent diagnosis, and patient management to enhance service efficiency. They should also invest in workforce training and organizational adaptation to ensure effective human–AI collaboration.Meanwhile, technology enterprises should optimize AI solutions based on regional needs to improve market adaptability and support the widespread adoption of GenAI in the healthcare sector. In particular, developing modular and scalable AI systems can help accommodate heterogeneous regional digital conditions and reduce implementation barriers. Limitations and future research Despite the significant contributions that emerged from this study, some limitations are acknowledged. First, the research sample is primarily focused on China, which may limit the generalizability of the findings. The effectiveness of GenAI applications may vary across different cultural and policy environments. Future research should expand the sample to include a broader range of countries and regions to test the external validity of the results. Second, although the panel data used in this study provides a static view of GenAI's effectiveness, it does not capture the dynamic processes of technological application. Future research could utilize longitudinal data to further explore the effects of GenAI at different stages and its developmental trends. Additionally, the rapid iteration of GenAI may introduce new variables or influencing factors, such as ethical concerns, data privacy, and security challenges, which require further investigation in future studies. Conclusion This study systematically analyzes the application effects of GenAI in the healthcare sector, revealing its significant positive impact on healthcare outcomes through deeper application integration. The findings confirm the critical role of industrial digitalization as a moderating variable on the effectiveness of GenAI. The results indicate that GenAI provides robust technical support for the healthcare industry by optimizing diagnostic and treatment processes, improving patient experiences, and enhancing operational efficiency. Under the moderation of industrial digitalization, the effectiveness of GenAI applications is significantly amplified. High degrees of industrial digitalization create an ideal environment for the deep integration of GenAI through advanced information infrastructure, enhanced technological adaptability, and accelerated technology diffusion. These findings further validate the theory of technological diffusion, which posits that an optimized external environment can substantially amplify the value of new technologies. The theoretical contribution of this study lies in expanding the framework for understanding GenAI by emphasizing the critical roles of digital transformation in adoption technology. The practical implications provide policymakers with differentiated recommendations based on regional characteristics: regions with high degree of digital transformation should prioritize accelerating the integration of GenAI with digital technologies, whereas regions with lower levels should focus on improving infrastructure to enhance the adaptability and effectiveness of GenAI applications. Although this study has yielded valuable insights, it has certain limitations, including the regional sample constraints, insufficient attention to dynamic analysis of panel data, and a lack of focus on dynamic technological variables. Future research could address these gaps by employing cross-national comparisons, longitudinal analyses, and dynamic modeling approaches to provide more robust empirical evidence for the optimized application of GenAI and the equitable development of healthcare services. In conclusion, GenAI holds immense potential for the healthcare industry, and its continued advancement is expected to drive the digital transformation of global healthcare systems and significantly enhance service quality. Acknowledgements We would like to express our sincere gratitude to all members of our research team for their hard work and dedication. Authors’ contributions Jackie Zhanbiao Li and Xin Li analyzed data and drafted the manuscript. Yingqian Lao designed the study and performed the data collection. Xin Li polished the manuscript and performed the data collection. Xin Li, Qifeng Wu and Qianhui Ting made substantive intellectual contributions to the conception of the work and the interpretation of the data and revised the manuscript. These authors contributed equally to this work. Funding This work was not funded by any funding. Data availability Data will be made available upon request by contacting the primary corresponding author. Declarations Ethics approval and consent to participate The data used in this study are secondary data, all of which are sourced from publicly available and legally authorized databases, with no direct or indirect human involvement. The data collection process adhered to relevant legal regulations to ensure the protection of individual privacy and other legal rights. Therefore, ethical approval from an ethics committee is not required for this study. Competing interests The authors declare no competing interests. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Jackie Zhanbiao Li, Email: lizhanbiao123@163.com. Xin Li, Email: 15179123142@163.com. Yingqian Lao, Email: lyq130816@126.com. References Associated Data Data Availability Statement Data will be made available upon request by contacting the primary corresponding author.