Investigating the Antecedents of User Acceptance of Clinical Decision Support Systems: a Case from Saudi Arabia

Background The purpose of this paper is to develop an integrated research model to identify the technological and non-technological factors that inuence user acceptance of the CDSS. This model is empirically tested using a data sample collected from selected healthcare facilities in Saudi Arabia. Methods The research study uses the quantitative approach to evaluate the currently implemented CDSS as a part of Hospital Information System BESTCare 2.0 in Saudi Ministry of National Guard Health Affairs. A survey questionnaire is conducted at all Ministry of National Guard Health Affairs hospitals for data collection. Then, the survey data is analyzed using Structural Equation Modeling (SEM) and AMOS 21.0. This analysis includes: measurement instrument reliability, discriminant validity, convergent validity, and hypothesis testing. Moreover, a CDSS usage data sample is extracted from the data warehouse to be analyzed as an additional data source. The results of the hypotheses testing show that usability, availability, and medical history accessibility are the important factors inuencing user acceptance of CDSS.


Introduction
One of the most signi cant causes of healthcare mistakes is the inability to access patients' medical records due to lack of implementing health electronic systems at healthcare facilities. It is a global issue affecting healthcare quality [1], [2]. Although the adoption of Electronic Medical Records (EMR) and Computerized Physician Order Entry (CPOE) has increased [3], [1], [4], EMR and CPOE are insu cient to prevent a large number of medication errors without a full integration with intelligent module such as (CDSS) [5], [6].
CDSS are information systems used to enhance decision-making process by matching patients' characteristics to a knowledge base and algorithms to generate warnings, alerts, and recommendations [7]. This de nition clari es the power of integrating patients' characteristics from EMR with CDSS. Usually, CDSS capture structured data from EMR through CPOE such as dosage, frequency, duration, and some other information [8]. Therefore, some research studies categorize EMR and CPOE systems as a prerequisite of [45]. The research study [46] applied FITT framework on a fully-integrated health electronic services called "HYGEIAnet". HYGEIAnet is a network of Hospital Information systems, Primary Care Information Systems, and Emergency Information Systems implemented in the hospitals and primary care centers of the Greek island "Crete" [47] [48]. This case study aims to nd the factors which in uence the adoption of IT services throughout ad distributed health environment, and show FITT's applicability to explain implementation's successes and failures. The research team used both quantitative and qualitative methods during the research including extracted data, interviews, documents review, and site observation. The reasons led to increasing the adoption are healthcare practitioners found the system facilitates their job in terms of retrieving patients' data and monitoring them. The integrated hospital information system consists of Clinical Information System, nursing records, Laboratory Information System (LIS), Electronic Health Records (EHR), and Pictures Archieving and Communications Systems (PACS). After case analysis, the overall success of the previously mentioned systems was due to some initiative covering FITT factors from the implementation team. The main initiatives such as on-job training, 24 hours hot-line support, pilot deployments, and managerial support played a signi cant role to achieve a high success rate.

CDSS Impact
CDSS success can be measured by its impact in minimizing prescription errors. Prescription errors can be de ned as "any preventable event that may cause or lead to inappropriate medication or patient harm when the medication is in the control of the health care professional, patient or consumer" [49]. Medication errors have been categorized into four categories as the following: Serious error (Type A), Major error (Type B), Minor error (Type C), and Trivial error (Type D) [50]. An investigation research has been conducted to analyze the handwritten medication errors at 10 Primary Healthcare Centers from public and private sectors in Riyadh [16]. The research team collected the paper medical records for 1182 patients from public primary care centers and 1200 from private primary care centers. This research proved that the rate of errors is high (near to 1/5) which can harm patients. Since this paper is limited to primary care environment, it's expected to have a higher rate of errors at more complicated facilities which provide emergency and critical care services. CDSS can be good solutions to reduce the high rate of medical mistakes even in stressful circumstances [51]. Moreover, most of studies that discussed CDSS net bene ts were focusing on physician's practice. Therefore, there is a need for more research to investigate the CDSS bene ts in terms of minimizing errors, and increasing e ciency and effectiveness [25].

Research Model And Hypotheses
A research model has been developed as a result of integrating the FITT framework into the Hot-t model.
This integrated model is used to analyze the factors of CDSS user acceptance. The developed model utilizes the three domains of FITT framework: Technology, Task, and Individual; and it utilizes the in uence between different domains factors of CDSS adoption from Hot-t model. The model consists of eight independent variables, one mediating variable and one dependent variable as depicted in Fig. 2. The description of these variables is as follows.
System's usability, is de ned as the degree to which the system is friendly and accessible [52]. The users feel that the system is easy-to-use and will help them to perform their tasks without extra effort. This attribute in uences the t between individual and technology. Research studies show that more than half of medical information systems fail due to usability issues [53].
System's availability, is de ned as the correct technical functioning of the system [54]. The system should be available and accessible anytime and anywhere within the organization. Otherwise, the tasks will not be performed in the required time. This attribute in uences the t between individual and technology, and between task and technology.
Medical history accessibility, is de ned as the coverage of the extent to which complete, accurate, organized, understandable, up-to-date, and timely information is provided in the system for the health practitioners to obtain information about any of their intended objectives [55]. Since it is very important to utilize the medical history of the patient to support decision making process and to avoid any order duplication, the medical history should be accessible and clearly stated in the system. This attribute in uences the t between individual and technology, and between task and technology.
Therefore, three hypotheses related to technology are identi ed as follows: H1. System's usability has a positive in uence on intention to use.
H2. System's availability has a positive in uence on intention to use.
H3. Medical history accessibility has a positive in uence on intention to use.
Task impact, is de ned as the users' perceptions about the extent to which the system allows them to complete their tasks effectively and to improve their work [56]. The users feel that the system allows them to accomplish more work than would otherwise be possible. This attribute in uences the t between individual and task.
Task-Technology Fit, is de ned as the degree to which the system assists users in performing their work or coursework [57]. The user nds that the system's functions are t for the requirements of tasks or coursework. This attribute in uences the t between task and technology.
Thus, two hypotheses related to task are identi ed as follows: H4. Task impact has a positive in uence on intention to use.
H5. Task-Technology Fit has a positive in uence on intention to use.
Training sessions construct, is de ned as the extent to which an individual has been trained about the system through courses, training, manuals, and so on [52]. This attribute in uences the t between individual and technology. The lack of training is reported as an obstacle to using CDSS for supporting healthcare decisions [58] [59].
User support, is de ned as the perception of how the system's provider delivers the service to the user [52]. The user will be more satis ed when the provider solved the system's issues rapidly. This attribute in uences Override Justi cation, is de ned as the reason for rejecting system's alerts [60]. This attribute in uences the t between individual and technology. Findings from IS research suggest that physicians will accept systems that allow them to have professional autonomy and practice individual judgment [61].
Intention to use, is de ned as the user intends to use the system [62]. The user is willing to let the system assist her/him in deciding which medication to prescribe.
Net bene t, is de ned as the bene ts of the system as perceived by the user [63]. The system has reduced the time and effort it takes to support decision making.
Therefore, four hypotheses related to individual are identi ed as follows: H6. Override justi cation has a positive in uence on intention to use.
H7. Training sessions have a positive in uence on intention to use.
H8. User support has a positive in uence on intention to use.
H9. Intention to use has a positive in uence on Net bene t.

Research Methodology
The research methodology followed in this research is a quantitative approach to achieve the research goals. A survey questionnaire is developed as the main measurement instrument to collect the health practitioners' responses measuring their behaviors towards the implemented CDSS. The survey questionnaire is used because it has many advantages in the IS research. These advantages include the ease of reuse, comparing different perspectives, capability of predicting behaviors and capability of testing types of theoretical propositions objectively [36]. The survey questionnaire provides a clear picture of health practitioners' experience with such systems. In addition, this study presents the CDSS alerts popped up to health practitioners during the medication prescribing process in EMR, and their actions and behaviors towards these alerts. The medication prescribing process requires the physician's order and pharmacist's veri cation or change. The research team used Oracle Data warehouse for extraction and Tableau for Visualization. The study implementation went through the following phases: understanding the business work ow, identi cation of the scope of the required data, data extraction, data modeling, and identi cation of dimensions and measures, and dashboards design.

Survey Design and Instruments Development
The survey design process started by reviewing the related literature to nd suitable survey questionnaire items for each model construct. After identifying the questionnaire items, the survey was designed and sent to six domain experts (University professors of Computer Science and Information Systems) in order to test its face and content validity. The experts thankfully provided the researchers with some notes to enhance the survey questionnaire. In addition, the process of survey evaluation continued with survey pilot study administered to some users chosen randomly to evaluate the questions in terms of clarity, precision, and time taken to complete the survey. Moreover, the data used in this study is collected from ve hospitals that belong to Ministry of National Guard. The hospitals are located in Riyadh, Jeddah, Ahsa, Dammam, and Madinah. The researchers assume that the questions are suitable for all hospitals as the system is standardized and follows the regulations and legislation of the ministry. The users received the same training materials and support process. After that, the survey was written in English language, published online and a noti cation sent to around 350 users through the department's managers. The scope of this survey includes physicians and pharmacists from all experience levels. Appendix 1 shows the latent construct items. Five points Likert scale [64] with anchors of strongly disagree to strongly agree was used to measure each item. The other part of this research which is Datawarehouse (DW) data extraction was conducted at King Abdulaziz Hospital -Al Ahsa, Saudi Arabia between January 1, 2018, and December 31, 2018. Physicians and pharmacists from all medical departments and within different experience levels were included in this study.      Factor loadings for each variable have to be at least 0.5 or the variable becomes a candidate for deletion [69]. The factor loadings have been calculated using AMOS for each construct.

The structural model
Structural model is a technique used to analyze the relationships between latent constructs and measured variables. Figure 3 shows the structural model results. All beta path coe cients are positive and statistically signi cant (at p < 0.001).
In the structural model, the value R 2 is the square of the correlation between the predicted values and the observed values and it indicates the percentage of variation explained by the regression line out of the total variation. Compared to prior models of CDSS acceptance, the model reported better explanatory power of variance in behavior intention to use CDSS. While the model in [70] explained 28% of variance and the model in [79] explained 47% of variance, the explanatory power of the model developed in this study is 75% of variance in behavior intention. Moreover, the model explained 84% of the variance in perceived net bene ts.
The value β is the correlation coe cient between two variables. Table 6 presents hypothesis testing results.  Finally, hypothesis 9 stated that intention to use has a positive in uence on net bene t. A positive path coe cient (β = 0.57, P < 0.001) supported hypothesis 9.

CDSS Usage Data analysis
In addition to the survey questionnaire data, a sample of 46212 medication alerts is extracted from the actual system usage history. The most occurred alerts are related to (Single Dose Maximum) and (Drug & Drug Severity Major) as shown in Table 7. In order to explore the physician's behavior during different cases and situations, a new important attribute which is visit type was added to this descriptive study. Table 8 shows the total and percentage of alerts and physician's override by visit type. As shown in the Table 8, the physicians show high acceptance rate for the received alerts. In the most critical area at any hospital which is ER, physicians accepted (78.14%) of the alerts and overrode (21.86%). Similar results occurred at inpatient and outpatient since (74.18%) and (71.17%) of the alerts were accepted by physicians respectively. These override rates are very low when compared to high alert overrides rates ranging between 49% and 96% as reported in recent research studies [70], [71].

Findings And Discussion
This study examines how certain variables affect healthcare practitioners' intention to use CDSS and how these factors affect their performance and clinical decisions. The researchers developed a model based on the integration between FITT and Hot-t models by incorporating some additional variables.
The technology variables: usability, availability, and medical history accessibility were found to be important factors to accept using CDSS.
First, we asserted that usability has a positive in uence on intention to use CDSS. Extensive use of questionnaires to examine CDSS usability and user satisfaction is crucial for integrating user feedback into the CDSS development process [72]. The results support this hypothesis. This nding is consistent with observations made by previous research investigating acceptance of CDSS in other countries [73]. However, the result is con icting with the ndings of [33] where no signi cant effect of effort expectancy on intention when considering user experience as a moderating variable when investigating Saudi users acceptance of IT.
This might be because the users interaction with CDSS is different than their interaction with other information systems.
Second, we hypothesized that system's availability has a positive in uence on intention to use CDSS. The results indicate that system's availability has signi cant impact on user acceptance of CDSS. Prior research has shown that system availability has a positive signi cant impact on the perceived quality of health care systems [74]. In turns, perceived quality is a signi cant predictor of CDSS acceptance.
Third, as expected, the results show that Medical history accessibility and information quality has a positive in uence on intention to use CDSS. This nding is consistent with some prior research [75], [76], [77].
However, some previous research ndings report no signi cant effect of information quality on HIS acceptance as mediated by perceived usefulness and perceived ease of use [78].
The second domain of the developed model includes individual (human) variables. This domain included the override justi cation, user training and technical support variables. This study results show that override justi cation has a positive in uence on intention to use CDSS. Although a number of previous studies has examined factors in uencing acceptance of CDSS alerts [60], our study is the rst study that examines the association between override justi cation and alert acceptance. Physicians thinks that they should be allowed to override the CDSS recommendations and provide their supporting evidence [79]. Without having the option to override CDSS alerts, physicians may consider CDSS as real threat to their professional autonomy [79]. Results from previous studies show that if the physicians consider CDSS as a real threat to their professional autonomy and individual judgment then their acceptance of CDSS will be impacted negatively [31]. Actual usage data extracted from the BESTCare 2.0 data warehouse shows that physicians accepted (78.14%) of the alerts and overrode (21.86%) in the ER. Similar results occurred at inpatient and outpatient since (74.18%) and (71.17%) of the alerts were accepted by physicians respectively.
User training is considered as the rst contact between the healthcare practitioner and the system. A quali ed instructor who has a knowledge of system's functionalities, a clearly designed material and documents, and learning management system were the main objectives of MNGHA to be accomplished through the training process. We asserted that training sessions has a positive in uence on intention to use CDSS. The results support this hypothesis. This nding is consistent with other studies [80]. In [81], [82], the research team found that training has a positive impact on intention to use the system. Therefore, it is essential to consider the amount of training needed before the CDSS implementation [80]. Moreover, [83] conducted a qualitative study that suggested certain organization characteristics, such as training, are in uencing use of CDSS. As a result, (56%) of survey questionnaire respondents accepted the training approach followed by MNGHA, and (21%) didn't like it, and the rest were neutral.
Another important factor is the technical support provided by Information Systems and Informatics Department (ISID) to the end-users after system's implementation. The results show that user support has a positive in uence on intention to use CDSS. Based on survey results, more than (45%) of the respondents consider ISID employees able to solve the technical problems and are helping them with courtesy. However, (18%) of the respondents disagree.
The third domain of the developed model covers task variables which measure the impact of system adoption on healthcare practitioners' daily tasks and productivity. The results support the assertion that task impact has a positive signi cant effect on physicians' intention to use CDSS. This nding is in line with results in prior research in general [31], [84], [85]. In particular, this result validates the nding of the qualitative analysis in [86]. This result can be interpreted by the fact that (70.5%) of the respondents declared that the system helped them to meet patient's needs, and (74%) of them stated that the system allows them to accomplish more work than before. Also actual usage data extracted from the DW shows that consultants faced 5796 alerts out of 160725 orders (3.60%) which are classi ed as the best performance among all job titles. Staff physicians prescribed 466158 medication orders and received 25138 alerts (5.40%). Finally, residents who have 5 years or less of experience prescribed 205455 orders and there were 15233 alerts (7.41%) generated to them. Results from prior research indicate that system usability as measured by effort expectancy has less signi cant effect than task impact as measured by performance expectancy on use intention [31], [73]. However, our results show that CDSS usability (β = 0.35, P < 0.001) has more signi cant than task impact (β = 0.32, P < 0.001). This is might be because most of the respondents are willing to use the system to help them through decision-making process and assist them to choose the most suitable medication for each case. Moreover, a signi cant impact of the system on healthcare practitioner's performance and patient outcomes has been proven since the results show that the system reduced the time and effort taken by healthcare practitioners to accomplish their work and make clinical decisions.
Concerning perceived net bene ts of CDSS, most of studies that discussed CDSS net bene ts were focusing on physician's practice. There is a very limited research that discussed net bene ts in terms of increasing CDSS e ciency and effectiveness [25]. Results from the current study show that there is a signi cant positive correlation between CDSS acceptance and the net bene ts (β = 0.57, P < 0.001). Thus, when CDSS users realize the net bene ts of CDSS and believe that the system has changed their job signi cantly, then they be more likely to accept the system.
A successful design and implementation of the CDSS requires careful consideration of these three mentioned domains in order to shorten treatment process and minimize the time health practitioner spent to perform the daily tasks. This is one of the critical factors required by the clinical environment which leads to increase the acceptance rate.

Conclusion
This study found predictive factors in uencing healthcare practitioners' intention to use BESTCare 2.0 to provide healthcare services and assist them in decision making based on the developed model. As expected, the study found that the ten used variables are critical and predictive factors in CDSS acceptance. The results con rmed that the variables played an important role in the outcomes of CDSS acceptance. Results from hypotheses testing show that system's usability, system's availability, medical history accessibility, task impact and task-technology-t positively correlate with user intention to use.
This study provides prudence pertaining to healthcare facilities and their higher management to adopt This research has some limitations. First, the study used a cross-sectional survey questionnaire to collect the data sample making it di cult to investigate causal relationships [31] among the research model constructs.
Future research might use a longitudinal survey questionnaire to collect the data over a longer period of time in order get more reliable interpretations.
Second, our study only used the quantitative method. Mixing quantitative and qualitative methods might be a good future research direction to validate the results of this study.
Third, subjects of this study are from ve hospitals but not all hospitals are equally represented in sample.
That might lead to some bias in the data set. Although, the ve hospitals are underneath one administration and are sharing the same policies, regulations, standards, salary scales bene ts, facilities, and working hours, future research studies should consider uniform representation of all the ve hospital in the data set to avoid any chance for bias.

Declarations Authors Contributions
MSB is responsible to review literature from previous researches in the elds of medical informatics and decision support systems. Moreover, he is responsible for data collection, data analysis, and discussions section. MES and MSB are responsible to design the acceptance model and de ne survey items. In addition, both researchers approved the nal version. All authors have read and approved the manuscript.

Constructs and items
System's usability SU1. This system is easily used [52]. SU2. This system has a quick response [52].
System's availability SA1. This system is always available for business [54]. SA2. This system launches and runs right away [54].
Medical history accessibility MH1. Information provided in the system is up-to-date [55].
MH2. Information provided in patient's pro le is easy to understand [55].

MH3
. The system provides all patient relevant information necessary to ful ll my needs [55].
Training sessions TS1. I consider the training used for this system is adequate [52].
TS2. I completely accept the training approach of this system [52].
User support US1. The provider (ISID) is very sophisticated with this system [52].
US2. The provider (ISID) of this system is able to rapidly solve the operating problems [52].
US3. Generally, the provider (ISID) of this system treats its customers with courtesy [52].
Override justi cation OJ1. If I override a drug alert, it is because the risk of the drug (or drug combination) is acceptable after considering the therapeutic bene t [60].
OJ2. If I override a drug alert, it is because this drug alert is not clinically important for the given patient [60].
Task impact TI1. The system helps me to meet patient's needs [56]. TI2. The system allows me to accomplish more work than would otherwise be possible [56].
Task-Technology Fit TT1. In my opinion, the system's functions are suitable for helping me complete my task [57].
TT2. In my opinion, the system's functions are enough to help me complete my task [57].
TT3. In my opinion, the system's functions are t for the requirements of my work or coursework [57].
Intention to use IU1. I am willing to use the system as an aid to help with my decisions about which medication to prescribe [62].
IU2. I am willing to let the system assist me in deciding which medication to prescribe [62].
IU3. I am willing to use the system as a tool that suggests to me a number of medications from which I can choose [62].
Net bene t NB1. The system has changed my job signi cantly [63]. NB2. The system has reduced the time it takes to support decision making [63]. NB3. The system has reduced the effort it takes to support decision making [63]s.  The research model with the hypotheses