Introduction

The rise of information communication technology (ICT) globally has given rise to new markets in various industries, including automated health services. Consequently, the utilization of ICT in healthcare systems has been on the rise worldwide [1]. E-health, which refers to the application of ICT in healthcare provision, is crucial for strengthening the healthcare system [2]. One specific ICT solution is the EHR system, designed to capture, store, process, and display patient information [3]. The EHR system facilitates care quality monitoring, electronic alerts, and administrative tasks, thereby enhancing knowledge exchange and decision-making for healthcare professionals [4,5,6,7].

While many countries have invested tremendous effort in integrating e-health initiatives into their healthcare facilities, EHR usage is still lagging behind expectations [8]. The adoption of EHR systems especially in low-income nations remains remarkably low [9], with only 15% successfully implementing such systems [10]. Studies have identified factors contributing to the low adoption of EHR systems in the Sub-Saharan region, including high costs, unreliable power supply, limited internet access, and inadequate computer skills among primary healthcare providers [11, 12].

In Ethiopia, the Federal Ministry of Health (FMOH) has planned various ICT projects, including m-health, human resource information systems, and health management information systems (HMIS) [13]. While the FMOH has made progress in implementing the District Health Information System 2 (DHIS2) and Electronic Community Health Information System (eCHIS), the implementation of a nationwide EHR systems has faced challenges, particularly related to human factors like technical proficiency, negative beliefs, and lack of acceptance [14].

Understanding the factors influencing the adoption of EHR systems is crucial for successful implementation [15], particularly considering the significance of healthcare professionals' adoption behavior [9, 16,17,18]. While studies have been conducted in Ethiopia [19,20,21] and other developing countries [19,20,21,22,23,24,25,26,27,28,29,30] on the intention to use digital health initiatives, the factors influencing the intention to use EHR systems among healthcare providers have not been assessed using the extended UTAUT3 model.

Conducting this study in Ethiopia is crucial for examining the adoption of the EHR within a distinct healthcare context. The insights gained will inform the development of customized approaches for successful EHR implementation, not only in Ethiopia but also in similar resource-constrained settings worldwide. Extending the UTAUT3 model will deepen the understanding of factors influencing EHR adoption, potentially enabling improved predictions and guidance for researchers and digital health implementers in developing customized strategies. Therefore, this study aims to propose an alternative theoretical model, the extended UTAUT3 model, and examine the potential predictors of healthcare professionals' intention to use EHR systems based on this extended model. The study seeks to address the following research questions:

  1. I)

    How effective is our proposed model, the extended UTAUT3, in explaining the intention of healthcare professionals to use the EHR system?

  2. II)

    What are the potential factors that can predict the intention of healthcare professionals to use the EHR system, according to the extended UTAUT3 model?

Theoretical background and the proposed research model

Over the past few decades, there has been a noticeable focus on comprehending how newly introduced technologies will be adopted and embraced by end-users. This emphasis aims to predict levels of acceptance and address any resistance to change that may arise. To achieve these goals, several models have been developed, two of which have gained significant recognition: the Technology Acceptance Model (TAM) [31] and the UTAUT [32] model. The UTAUT model, in particular, has garnered considerable attention from researchers since its development, inspiring many to delve into its intricacies [23, 29, 33].

In the following subsections, we conducted a review of both the original and extended models of UTAUT and proposed a potential extension of the model for our study.

Unified theory of acceptance and use of technology (UTAUT)

The UTAUT model, originally formulated by Venkatesh et al., 2003 [32], is considered more comprehensive and predictive than earlier models. It was developed by integrating and synthesizing eight earlier models in the field. It initially consisted of four core constructs: effort expectancy, performance expectancy, social influence, and facilitating conditions. In addition, Venkatesh et al. considered four moderator variables such as age, gender, experience, and voluntariness of use in the original UTAUT model [32].

The four core constructs of the original UTAUT model were derived by merging constructs with similar behavior from the earlier eight models of technology acceptance and use. During this process, they identified five constructs perceived usefulness, extrinsic motivation, job fit, relative advantage, and outcome expectations from different models that exhibited similar behavior. These constructs were then integrated into the UTAUT model to explore the concept of effort expectancy [32].

The creators of the original UTAUT model also integrated three constructs perceived ease of use, complexity, and ease of use from the existing models to capture the concept of effort expectancy within the UTAUT model. Similarly, the concept of social influence was addressed by integrating three constructs from the existing models: subjective norm, social factors, and image. Furthermore, facilitating conditions were represented in the UTAUT model by integrating three distinct constructs: perceived behavioral control, facilitating conditions, and compatibility [32].

Extended theory of acceptance and use of technology (UTAUT2)

Venkatesh et al., 2012 [34] extended the UTAUT model to UTAUT2 by incorporating three additional constructs: hedonic motivation, price value, and habit. UTAUT2 demonstrated greater predictive power in explaining the intention to use new technologies compared to the original UTAUT model [32, 34]. The authors also made modifications to the model by considering age, gender, and experience of use as moderator variables, while dropping voluntariness as a moderator, to make the revised model more applicable to voluntary behavior.

The constructs of hedonic motivation, price value, and habit, which were incorporated into the UTAUT2 model, have distinct origins and were derived from different theoretical perspectives. While the concept of hedonic motivation traced back to the earlier concept of psychological hedonism, the significance of hedonic factors, such as enjoyment and pleasure, in shaping consumers' acceptance and use of technology was recognized by Childers et al. in 2001 [35]. The inclusion of the second construct, price value, in UTAUT2 was based on economic and rational decision-making theory [36], which emphasizes individuals' evaluation of the costs and benefits associated with adopting a technology.

The inclusion of the concept of "habit" in the UTAUT2 model was inspired by the understanding that people develop automatic patterns of behavior over time. This recognition was influenced by two perspectives: the Habit/Automaticity Perspective (HAP) [37] and the Instant Activation Perspective (IAP) [38] as highlighted by Venkatesh et al, 2012 [34]. The HAP suggests that habits formed through repeated actions can lead to automatic behavior, while the IAP views automatic use as a faster version of conscious use [37, 38].

Unified theory of acceptance and use of technology 3 (UTAUT3)

UTAUT3 was adapted from UTAUT2 by Farook et al., 2017 [25]. This model includes eight constructs: effort expectancy, performance expectancy, social influence, facilitating conditions, hedonic motivation, price value, habit, and a newly added variable called personal innovativeness [25]. The original UTAUT3 model was tested among business students in Malaysia and explained 58.1% of the variance in behavioral intention to use a lecture capture system [25].

The concept of personal innovation in information technology (IT) has long been acknowledged, with scholars like Agarwal and Prasad [39] recognizing its significance in technology adoption. Expanding on this, Farook et al., 2017 [25] extended the UTAUT model by incorporating personal innovativeness as a variable. Their empirical research aimed to provide insights into how personal innovativeness influences individuals' behavioral intentions and acceptance of technology, specifically focusing on the usage of a lecture capture system.

The proposed research model: identify constructs to extend and modify UTAUT3

Our research aimed to extend the UTAUT3 model by identifying additional constructs and modifying them to our specific research context. Building on previous studies that explored ways to expand the UTAUT model, such as introducing new variables and incorporating moderation and mediation mechanisms [26], we conducted a comprehensive literature review to address the limitations of previous UTAUT models. Our primary focus was on the intention to use, specifically excluding use behavior during the pre-implementation phase of EHR in our study area, as use behavior cannot be studied at this stage. Consequently, we extended the UTAUT3 model by incorporating additional constructs and making necessary adjustments. To achieve this, we drew upon the UTAUT3 model proposed by Farook et al. in 2017 [25] and refined it to better suit our research context.

To bridge the gap observed in the original UTAUT, UTAUT2, and UTAUT3 models, we conducted a comprehensive review of the literature that examined various extensions and theories associated with those models. Through this review, we identified individual behavioral factors and integrated them into our proposed model. In line with the TAM model [40], which considered attitude as a fundamental construct, and previous UTAUT extensions that utilized attitude as a mediating variable [21,22,23, 41,42,43,44], we incorporated attitude as a mediator variable in our study. While the original UTAUT model did not include attitude based on empirical findings [32], we argue that its exclusion does not necessarily invalidate its potential relevance as a mediator in specific research contexts.

By considering different theoretical perspectives [45] and our specific research context, we argue that including attitude as a mediator can provide a more comprehensive understanding of the underlying mechanisms involved in technology acceptance and use. We aimed to contribute to the existing literature by uncovering the specific psychological processes that drive individuals' decision-making and behavioral intention formation in technology adoption. This perspective aligns with the Cognitive Evaluation Theory (CET) proposed by Deci and Ryan [45], which emphasize the impact of perceptions on intrinsic motivation.

In our study, we propose including technology anxiety and digital technology self-efficacy as exogenous variables in the context of digital health technologies, building upon the findings of researchers who have extended the UTAUT models [27, 46]. Specifically, Guo et al., 2013 [42] recommended considering inhibiting factors in technology acceptance, with technology anxiety being one such factor. Additionally, previous studies have examined the impact of technology anxiety and technology self-efficacy in various domains of information system development [27, 46,47,48].

Technology anxiety initially recognized and explored by researchers and scholars in the fields of psychology and human–computer interaction, represents the fear or apprehension individuals may experience when using new technologies [48]. On the other hand, technology self-efficacy, introduced by psychologist Albert Bandura, refers to an individual's belief in their ability to successfully perform tasks using technology [49].

While price value was a construct in UTAUT2, we excluded it from our model based on the recommendation of a previous study [19], particularly when participants were not direct purchasers of the system. Even though UTAUT3 is the extension of UTAUT2, it doesn’t consider the moderation effect. We argue that including mediators and moderation can enhance the viability of the models, and therefore, we incorporated the moderator variables of the UTAUT2 model such as gender [7], age [7, 50], and experience [7].

Although UTAUT is not new, the specific UTAUT3 model has not been extensively tested in previous research, particularly in the context of digital health. Additionally, our extended version of UTAUT3 provides a foundation for future researchers examining the behavioral intention and acceptance of to use of new technology. By incorporating additional relevant constructs, our study addressed calls from scholars to enhance the applicability of the UTAUT model [34, 51, 52].

Overall, our study contributes to the theory of health information technology (HIT) use behavior by extending UTAUT3 after a comprehensive review of previous models and literature. Through this process, we introduced two exogenous variables, technology anxiety, and technology self-efficacy, and also the mediator variable of attitude, as well as moderation variables of the UTAUT2 model were introduced, to extend the recent UTAUT3 model by Farook et al., 2017 [25].

Overall, our extended model includes core constructs such as performance expectancy, effort expectancy, social influence, facilitating conditions, hedonic motivation, habit, technology self-efficacy, technology anxiety, and the mediator variable attitude. Furthermore, the model encompasses three moderator variables: age, sex, and experience. The hypotheses were developed based on a thorough review of existing literature and are supported by relevant studies, as summarized in Table 1. Figure 1 provides a visual representation of our proposed research model.

Table 1 Key concepts and hypothesis to be tested
Full size table
Fig. 1
figure 1

The proposed research model. NB: TSE = technology self-efficacy, EE = effort expectancy, PE = performance expectancy, PI = personal innovativeness, FC = facilitating condition, SI = social influence, HM = Hedonic motivation, TAN = technology anxiety, EHR = Electronic health record

Full size image

Methods

Study design and setting

Between September 1 and October 14, 2022, a cross-sectional quantitative study was conducted at governmental hospitals in Illu Aba Bora and Buno Bedele Zones, which are situated in the Oromia regional state of southwest Ethiopia. Previously, the two zones were considered a single administrative zone, but they have since been demarcated separately. Mettu and Bedele are the respective capital cities of Illu Aba Bora and Buno Bedele, located approximately 600 km and 580 km away from Addis Ababa, the capital city of Ethiopia, respectively.

Study participants, sample size, and sampling procedure

The study population for this research consisted of all health professionals permanently working in the Illu Aba Bora and Buno Bedele zones in southwest Ethiopia. The sample size was calculated using a single population proportion formula, as follows:

$$\text{Sample size }(\text{n})= {\text{Z}}^{2} \cdot \text{P }\cdot (1-\text{P})/{\text{d}}^{2}$$

In this study, a prevalence (P) of 50% (P = 0.5) was assumed for health professionals' behavioral intention to use EHR, as previous studies in a similar setting did not reveal the exact prevalence. The researchers also used a 95% confidence level (Z = 1.96) and a margin of error (d) of 5% (d = 0.05).

Based on these assumptions and the aforementioned formula, a sample size of 384 was calculated. To account for the expected non-response rate of 10%, the total sample size was increased by 10% of the initial sample size, resulting in a final total sample size of 423.

In Illu Aba Bora and Buno Bedele Zones, there were a total of five fully functional public hospitals, employing 1,398 healthcare providers. To ensure representation, the sample size of 423 was proportionally allocated among the five public hospitals. Subsequently, health professionals were randomly selected from each hospital to participate in the study.

Measurement instrument

Data collection was conducted using a self-administered questionnaire that had undergone pretesting. The questionnaire was specifically developed for this study, taking into consideration an extensive review of previous UTAUT models, theories, and studies conducted elsewhere [27, 43, 48, 55, 58, 59, 66, 69,70,71]. It consisted of three parts, with a total of 47 items. Part one assessed socio-demographic factors using five items. Part two focused on UTAUT-based behavioral predictors with 39 items. Part three addressed the behavioral intention to use EHR with three items (See Supportive file 1 for additional information).

Statistical analysis

For statistical analysis, the data were entered into EpiData version 4.6 and then exported to SPSS AMOS 23. Descriptive statistics, including frequencies and percentages, were used to summarize the variables. The assumptions of linearity and normality were checked and found to be fulfilled. Scatterplots were examined and tests such as the Shapiro–Wilk test were conducted to assess these assumptions. The results confirmed that the data exhibited a relatively linear pattern and approximated a normal distribution.

To ensure measurement validity, the outer loadings, cross-loadings, and item loadings of each item were assessed. The reliability of the measurement model was evaluated using Construct Composite Reliability (CR) and Cronbach's alpha (Cα) for each construct. Convergent validity was assessed using the Average Variance Extracted (AVE), while discriminant validity was examined using the Heterotrait-Monotrait Ratio (HTMT). Common method bias was evaluated using Harman's single-factor test. Multicollinearity between predictor constructs was checked using the Variance Inflation Factor (VIF).

To evaluate the structural model, R2, standardized estimates, and model fit indices were utilized. Model fit was assessed using indices such as the Root Mean Square of Standardized Residual (RMSEA), Adjusted Goodness of Fit Index (AGFI), Goodness of Fit Index (GFI), Comparative Fit Index (CFI), and Normal Fit Index (NFI). The associations between exogenous and endogenous variables, as well as significant moderation and mediation effects, were examined using standardized regression coefficients (β), 95% confidence intervals, and p-values with a significance level of p-value < 0.05.

The explanatory power of UTAUT2, UTAUT3, and the proposed model of Extended UTAUT3 were compared using R2. Statistical mediation analysis was employed to estimate the strength of mediator-outcome effects, while moderation analysis was used to observe whether moderator variables influenced the association between independent and dependent variables. In the analysis of the moderating effect of experience, experience was treated as a continuous variable to capture the potential variation in the relationship between predictors and the outcome variable as experience increases. Direct, indirect, and total effects were reported, and mean-centered values were used for moderation effect analysis.

Results

Socio-demographic characteristics

Out of the 423 respondents, 394 participants (93.1%) responded during the data collection period. The largest proportion of participants fell within the age range of 31 to 40 (49.5%), identified as male (67.8%), worked as nurses (30.9%), and held a degree (66.2%). Additionally, the majority of respondents (43.7%) reported having 6 to 10 years of experience (refer to Table 2).

Table 2 Socio-demographic characteristics of health professionals at public hospitals in southwest Ethiopia, 2022
Full size table

Evaluating the measurement model

In the measurement model, we evaluated reliability, convergent validity, and discriminant validity. Reliability was assessed by examining the Construct Composite Reliability (CR) and the internal consistency reliability (Cα) for each construct. Our findings indicated that Cα values ranged from 0.727 to 0.94, while CR values ranged from 0.906 to 0.989. These values exceeded the suggested threshold of 0.7 [72] (Refer to Table 3).

Table 3 Measurement model statistics
Full size table

We assessed convergent validity using the Average Variance Extracted (AVE). The AVE is calculated as the sum of the squared loadings divided by the number of indicators. Prior researchers in this field have suggested that a factor loading above 0.7 is generally considered sufficient for establishing convergent validity [73]. While a factor loading of 0.7 is commonly considered indicative of a strong association between indicators and constructs, it is widely accepted among scholars to use a cut-off point of 0.5 as acceptable [31, 74]. Therefore, we included indicators with factor loadings above 0.5 to ensure a comprehensive assessment of convergent validity. Items below this threshold were excluded to maintain the reliability and validity of the measurement model.

Based on Table 3, it can be inferred that the AVE values for all variables, except for effort expectancy, habit, and technology anxiety, were above 0.7. The AVE values for effort expectancy, habit, and technology anxiety were 0.685, 0.653, and 0.602, respectively, falling below the threshold of 0.7. However, these AVE values still demonstrate reasonable convergent validity according to the argument put forth by certain scholars, who stated that an AVE value above 0.5 might be acceptable [75, 76].

The concept of discriminant validity aims to assess the distinctiveness and individual existence of each unobservable latent variable [77]. This evaluation can be performed using criteria such as the Fornell-Larcker criterion and the HTMT method. Notably, recent studies have highlighted that the HTMT approach is a superior and more contemporary method [78]. When the HTMT value falls below 0.90, it indicates the presence of discriminant validity between two related constructs. As shown in Table 4, the HTMT values ranged from 0.208 to 0.864, which is below the recommended threshold for establishing discriminant validity [79].

Table 4 Heterotrait-monotrait ratio (HTMT)
Full size table

Common method bias (CMB) arises when respondents' answers to behavioral, opinion, and perception measurements are influenced by their situation and context. In our study, we employed Harman's single-factor test [80] to examine the presence of CMB. The results of Harman's one-component test indicated a dispersion of 41.7%, which falls below the acceptable threshold of 50% for this test [80]. Based on these findings, we concluded that there was no evidence of CMB in our study.

To evaluate multicollinearity among predictor constructs, we conducted an assessment. The results indicated that all constructs had VIF values below 4.0 and tolerance levels above 0.2. These findings suggest that there is no evidence of multicollinearity among the predictor constructs, indicating that they are independent of each other.

Evaluating the structural model

We assessed the structural model using R2 values, standardized estimates, and model fit indices. To evaluate the model fit, we considered popular indices such as RMSEA, AGFI, GFI, CFI, and NFI. Table 5 shows that all model fit indices indicated an acceptable level of fit.

Table 5 Model fitness assessment for intention to use
Full size table

In terms of explanatory power, we analyzed the R2 values of the endogenous constructs. The results revealed that the modified UTAUT3 constructs explained 84.5% (0.845) of the variance in attitude towards EHR and 86.2% (0.862) of the variance in intention to use EHR. When compared to earlier UTAUT models, UTAUT3 explained 66.9% (0.669) of the variance in intention to use EHR, while UTAUT2 explained 62.9% (0.629). These findings suggest that the proposed model (Extended UTAUT3) has a stronger explanatory power than the earlier UTAUT3 and UTAUT2 models (See Supportive File 1).

Furthermore, the evaluation of the structural model revealed significant direct effects. Specifically, personal innovativeness (β = 0.468, p-value < 0.001), hedonic motivation (β = 0.135, p-value < 0.016), technology anxiety (β = -0.083, p-value < 0.027), and performance expectancy (β = 0.782, p-value < 0.001) had significant direct effects on attitude towards EHR. Similarly, personal innovativeness (β = 0.215, p-value < 0.018), performance expectancy (β = 0.245, p-value < 0.001), and attitude (β = 0.611, p-value < 0.001) had significant direct effects on the intention to use EHR (See Table 6).Table 5

Table 6 Structural model analysis
Full size table

Mediation analysis

We used mediation analysis to examine whether attitude plays a mediating role. Mediation occurs when the independent variable's impact on the dependent variable is partially or completely transmitted through the mediator. Complete mediation happens when the predictor's influence is entirely transmitted through the mediator, while partial mediation occurs when the predictor's influence is partially transmitted through the mediator.

Our findings reveal that attitude mediates the relationship between hedonic motivation (β = 0.512, p-value < 0.01), technology anxiety (β = -0.360, p-value < 0.05), and performance expectancy (β = 0.701, p-value < 0.01) with the intention to use EHR, as shown in Table 7.

Table 7 Mediation effect of attitude
Full size table

Moderation analysis

The moderating effect of age

According to the results presented in Table 8, the moderation analysis indicated that age does not have a moderating effect on the relationship between the independent variables and behavioral intention.

Table 8 Moderation analysis summary of age
Full size table

The moderating effect of gender

The results from the moderation analysis, as presented in Table 9, indicate that gender moderates the association between social influence (χ2 = 14.630, p-value < 0.001) and personal innovativeness (χ2 = 5.011, p-value < 0.001) with behavioral intention. Specifically, being male (β = -0.784, p-value < 0.001) strengthens the association between social influence and behavioral intention compared to being female (β = -0.086, p-value > 0.05). Conversely, being female (β = 0.407, p-value < 0.001) strengthens the association between personal innovativeness and behavioral intention compared to being male (β = 0.199, p-value < 0.05).

Table 9 Moderation analysis summary of gender
Full size table

The moderating effect of experience

According to the information presented in Table 10, no moderation effect was observed. The moderating variable did not have any effect on either strengthening or weakening the association between the independent and dependent variables.

Table 10 Moderation analysis summary of experience
Full size table

Discussion

In this study, we proposed an extended UTAUT3 model to analyze the predictors of behavioral intention to use EHR among health professionals in public hospitals in southwest Ethiopia. The model exhibited strong explanatory power and explained 84.5% (R2 = 0.845) of the variance in behavioral intention to use EHR. Following Leguina et al., 2015 [86] classification of models as weak, moderate, and strong based on their explanatory power, we can confidently classify the proposed model as strong.

In our analysis, we compared the explanatory power of recent UTAUT models, utilizing the current data, with the extended UTAUT3 model. Our findings revealed that UTAUT3 explained 66.9% of the variance in intention to use EHR, while UTAUT2 explained 62.9%. These results indicate that the newly proposed model exhibits a higher level of explanatory power compared to the earlier models, suggesting its effectiveness in understanding behavioral intention.

Our study demonstrated a superior level of explanatory power compared to the original UTAUT model of Venkatesh et al., 2003 [32], the UTAUT2 model of Venkatesh et al., 2012 [34], and the UTAUT3 model of Farooq et al., 2017 [25]. These models originally reported explanatory powers of 69%, 74%, and 58.1% of the variance in intention to use, respectively. The strong explanatory capacity of our extended UTAUT3 model could be attributed to various factors. Firstly, the inclusion of individual factors in the model allows for a more comprehensive understanding of the specific determinants of behavioral intention. Additionally, the specific technology type being studied, the characteristics of the study participants, and the sample size may have contributed to the model's strong performance in explaining behavioral intention.

Based on our comprehensive literature review, we found that many scholars who utilized modified and extended UTAUT models across various disciplines reported lower levels of explanatory power compared to our study [13, 19,20,21,22,23, 27, 28, 30, 56, 58, 66, 68, 87, 88]. This suggests that the inclusion of new constructs in the modified UTAUT3 model may have contributed to achieving a higher level of explanatory power. Furthermore, upon examining the measurement model, we found that all of the constructs demonstrated commendable levels of validity and reliability. This indicates the robustness and reliability of the constructs used in our study.

The study's structural model analysis revealed a significant relationship between performance expectancy and the behavioral intention to use EHR. This finding is consistent with previous research conducted in various regions [19,20,21, 23, 24, 27, 42, 58]. Additionally, our mediation analysis indicated that performance expectancy indirectly influenced intention through the mediator variable of attitude, which is consistent with a previous finding [30].

These results implied that an individual's expectation of the desired outcome plays a crucial role in shaping their positive or negative feelings. These feelings, in turn, impact their motivation to engage in a specific task. The expectancy theory of motivation supports this finding, as it proposes that an individual's perception of the desirability of the consequences directly influences their motivation to choose a particular course of action [89].

The study found a clear and positive link between personal innovativeness and the intention to use EHR. This finding is consistent with previous research conducted in various regions [25, 28, 56, 57]. The possible reason behind this connection is that individuals who are open to trying out new digital health solutions are more likely to have a strong desire to use and benefit from these technologies. In simpler terms, being willing to embrace innovation in the field of digital health enhances people's motivation to adopt and utilize digital health tools.

The study findings revealed that individuals' attitude significantly influences their intention to use EHR, which is consistent with previous research conducted in various regions [19, 21, 68]. Building upon prior studies, this research identifies several pivotal factors, including personal innovativeness [61], hedonic motivation, technology anxiety [47], and performance expectancy [41, 42, 44], that play a crucial role in shaping individuals' attitudes towards EHR. These findings highlight the critical importance of comprehending and addressing these factors, as they directly impact individuals' attitudes. To effectively influence individuals' attitudes, healthcare organizations can promote personal innovativeness, emphasize the positive aspects of EHR, alleviate technology anxiety, and enhance performance expectations. By doing so, healthcare organizations can significantly enhance individuals' intention to adopt and utilize EHR in their healthcare practices.

The mediation analysis conducted in this study revealed that hedonic motivation and technology anxiety had a complete mediation effect. While the structural model analysis indicated that the direct effect of hedonic motivation on behavioral intention was not significant, the mediation analysis demonstrated that the effect of hedonic motivation on behavioral intention occurs through attitude, which is referred to as an indirect effect. This finding is reasonable as the level of pleasure or satisfaction influences individuals' positive or negative feelings and strengthens their inclination towards a particular behavior. It is worth noting that previous studies on behavioral aspects of digital health technology did not extensively explore mediation analysis, and thus we did not find significant mediation effects consistent with our findings. However, unlike our study, a few studies reported that attitude mediates the relationship between effort expectancy and behavioral intention [21, 41, 43].

The mediation analysis conducted in this study revealed that technology anxiety had a significant indirect negative effect on behavioral intention. This implies that individuals' fear or anxiety towards technology creates negative emotions, which in turn hinder their willingness to engage in the desired behavior. Interestingly, while the existing literature has not extensively explored the indirect effect of technology anxiety on intention through the mediator variable of attitude, a few studies highlighted its direct effect [46].

The moderation analysis showed that being male strengthened the association of social influence and behavioral intention than being female. This finding is consistent with previous research conducted in Canada [7]. The difference in the strength of this association might be attributed to variations in social factors and pressures experienced by males and females. Additionally, the likelihood of individuals imitating actions and behavior due to social influence may differ between the two genders. This finding is further supported by scholars who emphasized that responses to social influence learning strategies differ between males and females [90, 91].

Conversely, being female strengthened the association between personal innovativeness and behavioral intention than being male. This finding is supported by a study conducted in Taiwan [92]. The variation in the desire to explore and adopt emerging digital technologies between males and females may explain this difference.

Limitations and future research directions

The findings of this study should be applied cautiously by digital health implementers, considering the study's limitations. Future researchers are encouraged to address these limitations in their work. The first limitation of this study pertains to its generalizability. As our research was conducted among health professionals in public hospitals within resource-limited countries, the findings may not directly apply to technologically advanced nations. Thus, caution should be exercised when extrapolating the results to different contexts.

Another limitation is that this study focused on assessing behavioral intention during the pre-implementation phase, which may not fully represent actual acceptance and utilization after real implementation. Future research should aim to investigate acceptance and usage patterns during the post-implementation phase to gain a comprehensive understanding of the technology's impact. Given that the soundness of variables related to behavioral intention is influenced by the specific technology and discipline being studied, it is recommended that future researchers extend this model by incorporating different concepts and theories. Furthermore, it is important to verify the proposed model in diverse settings and disciplines to ensure its applicability across various contexts.

Despite these limitations, this study provides a baseline for the development of a novel model to examine behavioral intention toward digital health technologies. By addressing these limitations and building upon the proposed model, digital health implementers and researchers can further advance our understanding of behavioral intention and facilitate the successful implementation and adoption of digital health technologies.

Conclusion

Our study has provided valuable insights regarding the factors influencing behavioral intention in the adoption of EHR systems, with a specific focus on low-income countries. Notably, we introduced a new proposed model known as the extended UTAUT3 model, and its superior explanatory power was successfully demonstrated in comparison to previous models. Consequently, we have firmly established the extended UTAUT3 model as a robust framework for a comprehensive understanding of technology acceptance in healthcare, particularly about EHR systems.

Our findings revealed that personal innovativeness, performance expectancy, and attitude directly influenced the behavioral intention to use EHR systems. Interestingly, performance expectancy played a dual role, exerting both direct and indirect effects. Conversely, the impact of hedonic motivation and technology anxiety on behavioral intention operated solely through their indirect influence on attitude. This highlights the importance of personal innovativeness, hedonic motivation, technology anxiety, and performance expectancy in shaping individuals' attitudes towards EHR systems.

While social influence did not exhibit direct or indirect effects on behavioral intention, our moderation analysis unveiled significant gender differences, indicating the need for gender-specific interventions. Specifically, being male strengthened the association between social influence and behavioral intention more than being female. Consequently, social influence emerges as a crucial variable that should be targeted in interventions aimed at enhancing behavioral intention, particularly among males. Additionally, we found that the importance of personal innovativeness varied between males and females. Notably, being female amplified the association between personal innovativeness and behavioral intention, highlighting the necessity of addressing female individuals' personal innovativeness in interventions targeting behavioral intention.

Based on these findings, we recommend that, digital health implementers, policymakers, and concerned bodies consider the comprehensive range of direct, indirect, and moderating effects identified in our study. By addressing personal innovativeness, performance expectancy, attitude, hedonic motivation, technology anxiety, and the gender-specific impact of social influence, interventions can effectively enhance behavioral intention towards EHR systems.

This research makes a significant contribution to the existing literature by shedding light on the determinants of behavioral intention in the adoption of EHR systems, particularly in low-income countries. The insights gained can inform digital health implementers, policymakers, and concerned bodies in designing targeted interventions that promote the successful adoption and utilization of EHR systems in healthcare settings.