Physicians' resistance toward healthcare information technology: a ...

European Journal of Information Systems (2007) 16, 725–737 & 2007 Operational Research Society Ltd. All rights reserved 0960-085X/07 $30.00 www.palgrave-journals.com/ejis

Physicians’ resistance toward healthcare information technology: a theoretical model and empirical test Anol Bhattacherjee1 and Neset Hikmet1 1

College of Business Administration, University of South Florida, Tampa, FL, U.S.A. Correspondence: Anol Bhattacherjee, College of Business Administration, University of South Florida, 4202 East Fowler Avenue, CIS 1040, Tampa, FL 33620, U.S.A. Tel: þ 1 813 974 6760; E-mail: [email protected]

Abstract This paper presents a theoretical model of physician resistance of healthcare information technology (HIT) usage by integrating the technology acceptance and resistance to change literatures, using a dual-factor model of technology usage. This model elaborates the interdependent and asymmetric effects of inhibiting usage perceptions, such as resistance, on HIT usage intentions relative to enabling perceptions, such as perceived usefulness and perceived ease of use. It also proposes perceived threat as a predictor of resistance, perceived compatibility as predicting perceived usefulness, and related knowledge as predicting perceived ease of use. The resulting model is empirically supported using a field survey of a computerized physician order entry system among 129 practicing physicians at a large acute-care hospital. Our study illustrates the importance of incorporating user resistance in technology usage studies in general and HIT usage studies in particular, grounds resistance research within extant theories of technology usage, and provides a preliminary model of resistance that can serve as the starting point for future research in this relatively unexplored yet potentially fertile area of research. European Journal of Information Systems (2007) 16, 725–737. doi:10.1057/palgrave.ejis.3000717 Keywords: healthcare information technology; resistance; dual-factor model; survey research; computerized physician order entry

Introduction

Received: 28 February 2007 Revised: 27 June 2007 2nd Revision: 3 September 2007 Accepted: 27 September 2007

For years, governmental agencies, healthcare insurers, and healthcare advocacy groups have advocated healthcare information technologies (HIT) as the means to reducing medical error rates, improving healthcare delivery quality, and lowering healthcare costs (IoM, 1999; PITAC, 2004).1 However, technologies such as computerized physician order entry (CPOE) systems, electronic health records (EHR), and electronic prescriptions are often strongly resisted by the same community that is expected to benefit from its use. For instance, in 2003, physicians at the prestigious CedarsSinai Medical Center at Los Angeles rebelled against their newly installed CPOE system, complaining that the system was too great a distraction from their medical duties and forcing its withdrawal after it was already online in two-thirds of the 870-bed hospital (Freudenheim, 2004). Similar patterns of resistance were also encountered at the University of Virginia Medical Center (Massaro, 1993) and in a nationwide survey of healthcare 1 An initial version of this paper was presented at the Hawaii International Conference on System Sciences (HICSS), Hawaii, in January 2007.

726 Physicians’ resistance toward healthcare information technology

Anol Bhattacherjee and Neset Hikmet

facilities (Lorence & Richards, 2003). This apparent resistance, given the expected benefits of such systems, not only runs counter much of our current understanding of information technology (IT) adoption and usage research, but may also seriously undermine the potential benefits of HIT systems. Why are seemingly useful technologies sometimes resisted by potential adopters? This is the general problem of interest to this study. More specifically, we examine two related research questions: (1) why do many physicians resist a new HIT, and (2) how does resistance influence their HIT usage decision? Addressing these questions is important for practical as well as theoretical reasons. From a practical standpoint, understanding why physicians resist HIT and how such resistance is manifested in their subsequent behavior can help healthcare administrators devise appropriate intervention strategies for minimizing physician resistance and their effects on the organization. This is important because our own observations of HIT implementation in local hospitals indicates that traditional change management practices, such as software demonstration, user training, and help desk staffing, are only of limited use in mitigating user resistance. Further, most HIT designs tend to focus on system considerations, such as new functionalities, security, and connectivity, rather than user considerations such as the system’s impact on users’ work behaviors and potential user resistance. A better understanding of user resistance of HIT may help design better systems that are both functionally good and also acceptable to their targeted user population. From a theoretical standpoint, though this study examines HIT resistance within a healthcare setting, its findings have broader ramifications for understanding IT resistance in general in other types of organizations, and for building a preliminary theory of IT resistance. As described in the next section, prior research on IT usage has largely ignored the problem of user resistance, and prior research on user resistance has been limited, fragmented, and non-cumulative. Resistance is not quite equivalent to non-usage, because non-usage may imply potential adopters are simply unaware of a new IT or are still evaluating the IT prior to its adoption, while resistance implies that the IT has been considered and rejected by these people. Further, resistance is often marked with open hostility toward the change agents or covert behaviors to stall or undermine change (e.g., Freudenheim, 2004), while non-usage does not generally engender such outcomes. Despite the above differences, IT usage and resistance must be examined jointly within a common theoretical model because user resistance is clearly a barrier to IT usage in organizations. This study proposes a preliminary model of HIT resistance, highlighting its causes and effects, by synthesizing prior findings in the IT usage and resistance to change literatures, using a recently proposed dual-factor structure of IT usage (Cenfetelli, 2004) as the bridge between these two literatures. This

theoretical model is then empirically validated using survey data from practicing physicians at a large acutecare hospital in the southeastern United States. The rest of the paper proceeds as follows. In the next section, we describe prior research on resistance to set the context for our study. The third section builds a theoretical model of HIT resistance. The fourth section describes our research methods, including instrument construction, site selection, and sampling, for empirical testing of our hypothesized model. The fifth section presents statistical data analysis techniques and results. Our final section provides a discussion of our study’s findings, its limitations, and its implications for future research and practice.

European Journal of Information Systems

Prior research There is no dearth of prior academic studies demonstrating the potential benefits of CPOE, EHR and other HIT in reducing adverse drug events (e.g., Bates et al., 1998), improving healthcare delivery (Mekhijan et al., 2002), and lowering costs (Taylor et al., 2002). Less common, however, are investigations of the behavioral problems associated with HIT usage, such as physician resistance. Instances of physician resistance to HIT systems are documented in the practitioner press (e.g., Freudenheim, 2004), but there has been little formal examination of why or how it occurs. In a recent study, Poon et al. (2004) interviewed senior managers at 26 U.S. hospitals and used the ‘grounded theory approach’ to identify barriers to CPOE implementation and ways of overcoming them. Physician resistance was identified as the primary barrier to CPOE implementation, with high cost of CPOE implementation and vendor/product immaturity being the other barriers. This study recommended four strategies for overcoming resistance: (1) strong hospital leadership to communicate project vision and commitment, (2) customizing the system extensively to fit physicians’ workflow, (3) employing well-respected physicians to champion change and encourage others to see beyond immediate frustrations, and (4) leveraging house staffs (younger physicians) who were exposed to CPOE as medical students and are comfortable with its usage (Poon et al., 2004). However, no theory of HIT resistance was posited linking the causes and outcomes of resistance in a nomological network. Ahmad et al. (2002) examined the case study of successful CPOE implementation at Ohio State Hospital, and suggested that continuous executive support and physician empowerment, an effective implementation team, a user-friendly interface, ongoing user support, breadth of order sets, and elimination of manual ordering process contributed to implementation success. However, most university/academic medical centers tend to have strong IT infrastructure, better support, and a pro-technology culture, since these facilities are geared toward healthcare education and HIT is an important component of such education. In contrast, HIT resistance tends to occur more frequently in non-academic

Physicians’ resistance toward healthcare information technology

hospitals, and it is unclear whether factors cited in the above study can be extended to non-academic hospitals. In an interpretive study of physicians’ reactions to clinical IT in three hospitals, Lapointe & Rivard (2005) observed significant evidence of physician resistance. This study also noted that resistance evolved with time across different stages of the implementation process, and that resistance behaviors are triggered or exacerbated by perceived threats among adopter groups, such as loss of power and reorganization of work. Spil et al. (2004) developed a ‘USE IT’ model describing four dimensions of HIT use: (1) resistance toward using the system, (2) relevance of the system to users’ needs, (3) requirements or extent to which the system meets user needs, and (4) availability of resources to design, operate, and maintain the system. Following a multiple case study of 56 general practitioners, the authors found that electronic prescription systems (EPS) was not used in 72% of the cases, that resistance was the strongest determinant of EPS non-use, and that resistance was the cumulative consequence of the remaining three determinants. A few more studies of resistance can be found in the general (non-healthcare) literature on IT adoption and usage. In a seminal paper, Markus (1983) studied financial accountants’ reaction to one organization’s implementation of a new financial accounting system and observed significant resistance among the firm’s divisional accountants. Using a positivistic case research approach, she inferred that this resistance was caused not by system deficiencies (e.g., inadequate features) or individual limitations (e.g., lack of training), but by the accountants’ fear of loss of control over key accounting data expected from system use and consequent loss of organizational power. Studying household adoption of personal computers, Venkatesh & Brown (2001) observed that: (1) resistors of computers outnumbered acceptors by a significant margin (67% to 33%) in a random sample of U.S. households, (2) factors that predicted resistance (i.e., fear of technological obsolescence, high cost of technology, and lack of requisite knowledge) were unique and distinct from those predicting acceptance (i.e., utilitarian outcomes, hedonic outcomes, and social outcomes), and (3) the association between intention and behavior was asymmetric between the two groups, with resistors behaving more closely with their stated intent than acceptors. In a survey study of individual usage of internet access service, Parthasarathy & Bhattacherjee (1998) noted that end-users who discontinued using internet service at the end of the promotional trial period (a proxy for resistors) were systematically different from continued users (adopters) in terms of service utilization, service-related perceptions (usefulness, ease of use, compatibility, and network externality), and communication influences during initial service trial (mass-media vs interpersonal sources). Likewise, a survey of business managers led Jiang et al. (2000) to report that the reasons for their


727

resistance to transaction processing systems were different from those for decision support systems. Among theory-based studies, Joshi (1991) used an equity theory perspective to compare IT resistance in a hospital, a bank, and a software development firm, and observed that users tend to resist more when they perceive their personal outcomes from the IT to be inequitable or unfair relative to (1) their inputs to the system, (2) outcomes from the system, and (3) outcomes realized by others in their reference group. Martinko et al. (1996) proposed an attributional model suggesting that one’s prior experiences with success or failure at tasks involving similar technologies evoke causal attributions toward new IT, which may lead to its resistance. These above studies point to a general recognition of user resistance as a salient problem for IT usage in general (Venkatesh & Brown, 2001) and HIT usage in particular (Spil et al., 2004). They also suggest that resistance is likely triggered by a different set of causative mechanisms than IT acceptance. However, the above studies were largely conducted in isolation from each other, and hence, their findings do not inform or build upon each other. Part of the problem may have been the lack of a generalized theory of user resistance and its lack of grounding within an established stream of research. In the next section, we attempt to build such a theory, while also grounding it in the IT adoption and resistance to change literatures.

Theory and hypotheses Early thoughts on user resistance can be credited to Kurt Lewin’s (1947) pioneering studies on force-field analysis in the organizational development literature. Lewin suggested that social systems, like biological systems, have a tendency to maintain a status quo by resisting change and reverting back to the original state – a characteristic referred to as ‘homeostasis.’ The status quo represents an equilibrium between the forces favoring and opposing change. Lewin thereby contended that successful change rests on organizations’ ability to first ‘unfreeze’ the equilibrium by altering the dynamics of the favoring and opposing forces, before change can be successfully enacted. Similar ideas were echoed by Zaltman & Duncan (1977), who defined resistance to change as ‘any conduct that serves to maintain the status quo in the face of pressure to alter the status quo’ (p. 63) and Keen (1981) in his definition of resistance as ‘social inertia’ similar to Lewin’s notion of homeostasis. The above definitions underscore some of the unique characteristics of the resistance to change construct and its distinction from IT acceptance. First, while acceptance behavior is targeted at a specific IT and driven by user perceptions related to that IT, resistance is a generalized opposition to change engendered by the expected adverse consequences of change. Resistance is therefore not focused so much on a specific IT, but on the change from the status quo caused by IT usage. Second, while acceptance is a behavior, resistance is not a behavior but a



cognitive force precluding potential behavior (Lewin, 1947). Hence, resistance is not the mirror opposite of IT acceptance, but a possible antecedent to IT acceptance. Therefore, resistance must be first overcome if IT is to be successfully adopted by its target population. Though resistance is an antecedent of IT usage, the exact nomological nature of this relationship is unclear. To understand this complex association, we draw upon Cenfetelli’s (2004) dual-factor model of IT usage. The core argument in this model is that IT usage considerations among potential users is based on a simultaneous examination of enabling and inhibiting factors, similar to Lewin’s (1947) notion of opposing forces. Cenfetelli defined inhibitors as those negative factors that discourage IT usage when present, but do not necessarily favor usage when absent. This ‘one-sided’ or asymmetric effect implies that inhibitors are not quite the opposite of enablers, but are qualitatively distinct constructs that are independent of but may coexist with enablers. Cenfetelli also contended that while IT acceptance is best predicted by enablers, IT rejection tends to be best predicted by inhibitors. Cenfetelli’s (2004) model was motivated by the observation that extant theories of IT usage, such as the technology acceptance model (TAM) (Davis et al., 1989) and the unified theory of acceptance and use of technology (UTAUT) (Venkatesh et al., 2003), have focused almost exclusively on users’ positive (enabling) perceptions related to IT usage (e.g., its perceived usefulness and ease of use), while ignoring negative (inhibiting) perceptions that may hinder IT usage. However, there is empirical evidence that negative perceptions do inhibit usage. For example, Speier et al. (2003) observed that system interruptions, such as popup advertisements on a web site, ‘you’ve got mail’ announcements in an e-mail system, and animation characters offering help with writing letters in Microsoft Word, hinders IT usage and task performance, although the lack of such interruptions does not enhance IT usage. Although Cenfetelli (2004) did not identify any specific inhibitor of IT usage, based on our prior discussion, resistance to change fits the classic definition of an inhibitor as well as portrays similar idealized behaviors. For instance, resistance demonstrates asymmetric behaviors typical of inhibitors, because presence of resistance hurts IT usage but lack of resistance does not necessarily enhance IT usage. Further, the independent nature of enablers and inhibitors in Cenfetelli’s model is consistent with prior empirical findings that IT acceptance and resistance are driven by different motivations (Parthasarathy & Bhattacherjee, 1998; Venkatesh & Brown, 2001). The dual-factor model therefore provides a theoretical bridge to link research on IT usage and resistance to change within an integrated model. Based on Lewin’s (1947) idea of opposing forces and Cenfetelli’s (2004) dual-factor model of IT usage, we propose that physicians’ intention to use a new HIT such as CPOE systems is based on both the traditional enablers



Perceived Compatibility

H7 (+)

Perceived Usefulness H1 (+) H4 (-)

Perceived Threat

H6 (+)

Resistance to Change

H3 (-)

Intention to Use HIT

H5 (-) H2 (+) Related Knowledge

Figure 1

H8 (+)

Perceived Ease of Use

Research model.

of IT usage, such as their perceived usefulness and perceived ease of use of IT usage (Davis et al., 1989; Venkatesh et al., 2003), as well as inhibitors such as their resistance to change (see Figure 1). Perceived usefulness refers to the extent to which users believe that system usage will enhance their job performance, while perceived ease of use is the extent to which users believe that their system usage will be relatively free of effort (Davis et al., 1989). TAM suggests that perceived usefulness and perceived ease of use are the two salient cognitive determinants of IT usage in the workplace, because people want to use systems that benefit their work and that do not cost them a lot of effort. Hence, both perceived usefulness and perceived ease of use tend to be positively related to IT usage intention. Note that while the effect of perceived ease of use on intention was modeled in TAM as being fully mediated by attitude, subsequent extensions of this model such as UTAUT (Venkatesh et al., 2003) have tended to drop attitude entirely and linked perceived ease of use (also called ‘effort expectancy’) directly to intention. Since HIT is one specific instance of workplace IT, the salience and effects of perceived usefulness and perceived ease of use on IT usage should also apply to the HIT context, as enablers of HIT usage. Empirical support for both associations within the HIT context is provided by Tulu et al. (2007), leading us to hypothesize: H1

Physicians’ perceived usefulness of HIT usage is positively related to their intention to use HIT systems.

H2

Physicians’ perceived ease of use of HIT usage is positively related to their intention to use HIT systems.

However, the introduction of a new system often engenders significant changes in users’ existing work processes. If such change is of a sufficiently high magnitude, given natural human proclivity to oppose change, many users will tend to resist the technology, resulting in lower intention of use. Within the HIT


literature, Poon et al. (2004) and Spil et al. (2004) provided support for the negative effect of resistance on HIT usage, based on qualitative interpretation of case study results, although no prior study has attempted to quantify the magnitude of this effect or compared it with enabling effects such as those of perceived usefulness or perceived ease of use. Nevertheless, the above logic and findings lead us to hypothesize: H3 Physicians’ resistance to change is negatively related to their intention to use HIT systems. Cenfetelli (2004) contended that inhibitors influence IT usage not only directly, but also indirectly via enablers as mediators. The indirect effects suggest that salient inhibitors, such as resistance to change, tend to influence (or ‘bias’) enablers, such as perceived usefulness and perceived ease of use, in a negative manner. Note that this ‘biasing effect’ is unidirectional, since enablers do not have any corresponding effect on inhibitors, and is the essence of Cenfetelli’s asymmetric effects hypothesis. There are two plausible reasons for this biasing effect. First, norm theory suggests that negative perceptions and acts garner more cognitive attention, are remembered better, and instigates greater information processing than positive ones (Kahneman & Miller, 1986). Resistance also tends to garner a wider range of emotional reactions ranging on the part of potential users from overt opposition to covert stalling, than do enablers. Second, inhibitors, when present, tend to anchor one’s overall perception toward attitude objects, subsequently biasing all other perceptions, including those of enablers. For instance, a single instance of system failure may lead physicians to view the target HIT as being of overall poor quality, despite more frequent instances of adequate system functioning or its multitude of positive features and capabilities. In light of this asymmetric biasing effect of resistance to change on enabling perceptions of HIT usage, we hypothesize: H4 Physicians’ resistance to change is negatively related to their perceived usefulness of HIT usage. H5 Physicians’ resistance to change is negatively related to their perceived ease of use of HIT usage. Why do people resist change? Piderit (2000) noted, ‘Rarely do individuals form resistant attitudes, or express such attitudes in acts of dissent or protest, without considering the potential negative consequences for themselves’ (emphasis added) (p. 784). In other words, people resist change if they expect it to threaten the status quo, such as a potential loss of power or control over strategic organizational resources. A case in point is Markus’ (1983) study of accountants’ resistance toward a new financial accounting system, which was motivated by their expected loss of control over key accounting data and consequent loss of organizational power resulting


729

from system usage. Markus concluded that despite the best intentions, new IT implementations can fail unless managers can adequately address power imbalances engendered by the new system. The notion of perceived threats is also reinforced in Lapointe & Rivard’s (2005) case study of physician resistance, ‘When a system is introduced, users in a group will first assess it in terms of the interplay between its features and individual and/or organizational-level initial conditions. They then make projections about the consequences of its use. If expected conditions are threatening, resistance behaviors will result’ (p. 461). These expectations lead us to hypothesize: H6

Perceived threat from HIT usage is positively related to physicians’ resistance to change.

Other key motivations for HIT usage cited in the HIT literature (Poon et al., 2004; Spil et al., 2004), which were also confirmed during our own interaction with physicians in local hospitals, are the system’s perceived compatibility with physicians’ work processes and physicians’ prior exposure to similar systems. Clearly, medical diagnosis and treatment is a complex work process involving multiple tests, synthesis of information from multiple sources, discussion with colleagues, trying out different courses of treatment, and conducting multiple treatments simultaneously. HIT systems, designed by technical specialists, may sometimes be unable to adequately capture this complex process and may even force their own process on physicians, creating a disincentive for HIT usage. The salience of compatibility, as perceived by physicians, on their HIT usage intention has been confirmed by Chau & Hu (2002), Tulu et al. (2007) and others. However, these studies differ in the way the effect of compatibility on intention is conceptualized. While Tulu et al. (2007) viewed compatibility as directly influencing intention, independent of perceived usefulness, Chau & Hu (2002) found that the effect of compatibility on intention is not direct but fully mediated by perceived usefulness. IT usage researchers have argued that one cannot find a system useful if that system is incompatible with his or her needs, and hence the effect of compatibility should be mediated by perceived usefulness. In view of these arguments, we propose: H7

The perceived compatibility between HIT usage and physicians’ work processes is positively related to their perceived usefulness of HIT usage.

Poon et al. (2004) observed differential response to CPOE adoption among different physicians. While younger physicians in this study were comfortable about using the system by virtue of their likely exposure to CPOE during medical education, resistance was more prominent among senior physicians lacking such an exposure. Given the relative recency of CPOE systems and the consequent lack of direct first-hand experience




with such systems among many practicing physicians, their familiarity with and knowledge of computer-based tools in general may serve as a proxy of their aptitude toward a specific HIT system such as CPOE. A general familiarity with IT, termed related knowledge in this paper, can help physicians transition into using sophisticated HIT systems, by improving their ease of use or comfort with learning and using HIT systems. This expectation leads us to our last hypothesis:

The new CPOE system was intended to work as follows. Physicians would electronically log into the system via a password-protected interface (the system tracked login date and time) from any hospital floor or remotely from their homes or private clinics. They could then access complete medical history of patients assigned to them, place new work orders such as laboratory or radiological tests, check real-time status information on prior work orders, and retrieve results if available. When new results came in, floor nurses were flagged who would inform the ordering physician or forward a printout of the results. The system was voice-activated so that physicians could dictate notes into the systems, for transcription by thirdparty vendors. Value-added features of the CPOE system included adverse drug alerts, which cross-checked physician prescriptions against patients’ allergy records for possible unfavorable interactions, and workflow management, which tracked patients’ medication schedule and alerted the attending physician or nurse whenever a new dose was needed. Repetitive ordering of multiple labs, procedures, and medications for common medical conditions (by ICD code) was automated via ‘order sets,’ which could be further customized to physicians’ personal preferences. At the time of this study, about 25% of physicians at this hospital were using the CPOE system for entering about half of their orders. In other words, 75% of the target population did not use the system, and about 87% of physician orders were not entered or tracked in the system. This high level of non-usage was surprising, given the high level of priority accorded to the CPOE implementation by the hospital administration and the considerable amount of resources devoted to technical and user support. For instance, the technical support staff worked around physicians’ busy schedules, and provided direct one-on-one hands-on support to physicians who were too busy to attend scheduled CPOE training classes. A large technology staff ensured that there were no infrastructure limitations afflicting system use. Further, the steering committee in charge of the CPOE implementation process included practicing physicians to ensure that physicians’ concerns were adequately represented and addressed. During our initial site visits to this facility and direct interactions with the physicians, the common reasons cited for user resistance were ‘it is new and difficult’, ‘it takes too long to learn’, ‘every patient is different, so a common system won’t help,’ and ‘there was nothing wrong with what we had before (i.e., paper-based ordering).’ One younger physician liked the CPOE system and used it regularly, logging in from home in the morning to check on patient charts and retrieving up-todate results and status reports on his patients before his hospital rounds. He indicated that the system helped him optimize his rounds better, saved him considerable time with paperwork processing at the hospital, and let him spend more time with his patients. However, many other physicians hated the system and devised innovative

H8 Physician’s related knowledge of technology usage is positively related to their perceived ease of use of HIT usage. In summary, we proposed a model of HIT usage intention by synthesizing key enablers from the IT usage literature with key inhibitors from the resistance to change literature. This synthesis was accomplished using Cenfetelli’s (2004) dual-factor model of IT usage, which specified the general nature of relationships between IT usage enablers and inhibitors, but did not identify any of these constructs. The resulting model was further extended by identifying the antecedents of the IT usage enablers and inhibitors in perceived threat, perceived compatibility, and related knowledge. This preliminary model, shown in Figure 1, represents our initial attempt at elaborating the causes and effects of physician resistance of HIT usage, which may require further refinement and extensions in future research.

Research design Empirical setting Our hypothesized research model was empirically tested using data collected from a field survey of practicing physicians at a large acute-care hospital in the southeastern United States. The specific technology examined was a CPOE system. These systems are intended to be used directly by physicians for ordering laboratory tests (e.g., blood culture, urine analysis), radiological tests (e.g., X-rays, magnetic resonance imaging, etc.), pharmacy prescriptions, and special procedures (e.g., bronchoscopy, biopsy) for inpatient medical care. The original system was first introduced in 1997 on a trial basis in one unit of this hospital. However, ongoing technical and implementation problems, including network connectivity problems, lack of workflow integration, and training setbacks, led to its discontinuation in 1998. An improved version of the same system was reintroduced in late 2003, following 18 months of process reengineering. The new system included workflow support for physicians, wireless connectivity, and full integration with an EHR system that stored complete medical history of all patients, including admission history, vital signs (e.g., blood pressure, pulse rate), discharge data, lab test results, radiology, and other results ordered using the CPOE. System rollout was completed hospital-wide by late 2005.



strategies to avoid its use, such as ‘smuggling’ in paperbased order sheets, calling in an order to a nurse (for system entry on the physician’s behalf), requesting work assignments on floors where the system was not yet installed, devising workarounds such as sticking ‘Post-It’ notes to patient charts (expecting the nurse to notice it and file it properly), and/or occasional tantrums. Given widespread resistance toward the system despite organizational conditions favoring its use, this hospital was an ideal site for testing our model of HIT resistance.

Measurement of constructs The seven constructs of interest to this study were: perceived usefulness, perceived ease of use, perceived compatibility, intention to use IT, perceived threat, related knowledge, and resistance to change. Each construct was measured using multiple-item, five-point Likert scales anchored between ‘strongly disagree’ and ‘strongly agree.’ Wherever possible, prevalidated construct measures were adapted from prior research, and reworded to fit the current context. Measurement items are listed in the Appendix. Perceived usefulness was measured using Davis et al.’s (1989) four-item scale. Three of these items assessed subjects’ perceptions of productivity, performance, and effectiveness gains from CPOE usage, and the fourth item examined their overall perceptions of its usefulness. Perceived ease of use was also measured using Davis et al.’s (1989) scale that captured the ease with which subjects believed that they can learn to use the CPOE system, be skillful at its use, get the system to do what they wanted to do, and their overall perceptions of its ease of use. Perceived compatibility was measured using three items from Brancheau & Wetherbe (1990), modified slightly to fit the current context. These items examined the extent to which subjects viewed the CPOE system to be compatible to the way they worked, the way they order tests for patients, and other systems they used. Intention to use IT was measured using an adapted version of Bhattacherjee & Sanford’s (2006) intention scale. The three items on this scale examined subjects’ intent to use the CPOE system, use more features of the system, and use the system for more of their job responsibilities. Related knowledge was patterned after Bhattacherjee & Sanford’s (2006) user expertise scale, which asked users to self-rate their knowledge of using computers, software programs, electronic mail systems, and online ordering systems prior to their exposure to the CPOE system. This construct was relabeled as ‘related knowledge’ because we were not measuring physicians’ prior expertise of the CPOE system (as most of them had never used a CPOE system before), but their knowledge of related technologies such as computers, software programs, and e-mails that could potentially help them easily transition to CPOE systems. Given the absence of pre-validated scales for perceived threat and resistance to change constructs, new multipleitem scales were created for these constructs. Nunnally’s

731


(1978) ‘domain sampling technique’ was employed for this purpose, in which the different domains of physicians’ work were identified and individual items were created to represent each work domain. Based on our initial interviews with physicians, perceived threat of HIT usage was interpreted as physician’s loss of control over their work, similar to that observed by Lapointe & Rivard (2005), in the way they made clinical decisions, ordered patient tests, accessed lab results, and worked in general. Hence, we created a four-item Likert scale that examined the extent to which physicians feared that they might lose control over these four domains of their work if they used the CPOE system. Likewise, resistance to change was measured using a similar four-item scale that asked respondents the extent to which they did not want the CPOE system to change the way they ordered clinical tests, made clinical decisions, interacted with other people on their job, and the overall nature of their job.

Data analysis and results The survey questionnaire was distributed in paper form to a total of 700 practicing physicians at this hospital, along with a cover letter from the hospital’s chief executive officer emphasizing the importance of their participation in this study and a postage-prepaid envelope to return completed responses. Respondent anonymity was guaranteed and no identifying information was recorded. A second round of survey was repeated 1.5 months later, to encourage participation from nonrespondents. The survey questionnaire was reviewed and approved by the institutional review boards at the researchers’ university and at this hospital. Following two rounds of survey, 131 responses were obtained from a total of about 700 practicing physicians, for a response rate of about 19%. Two of these responses consisted of mostly missing values, and were therefore discarded. Respondents included physicians of all specialties, including internal medicine, pediatrics, gynecology, pathology, general surgery, anesthesiology, radiology, neurology, oncology, and cardiology. Respondents had a mean age of 49.5 years (SD ¼ 10.1 years), fulltime work experience in the medical profession of 20.1 years (SD ¼ 11.4 years), and a computer-related work experience of 9.7 years (SD ¼ 6.2 years). Sample demographics, in contrast with the population demographics for all 700 physicians at this hospital (obtained directly from the hospital administration), are shown in Table 1. Two tailed t-tests found that our sampled respondents did not differ significantly from the overall physician population in age (t ¼ 0.39) or specialty distribution (t ¼ 0.01), alleviating concerns of potential non-response bias in our data sample. Another common source of bias in single-respondent cross-sectional surveys is common method variance (CMV). To test for this bias, we used Lindell & Whitney’s (2001) marker-variable technique, which provides a quantitative estimate of the magnitude of CMV and was recommended by Malhotra et al. (2006) as a more robust



Table 1


Sample and population demographics Sample data

Size (N) Mean age Distribution by specialty Anesthesiology Cardiology Emergency care Internal medicine Obstetrics/gynecology Pathology Pediatrics Psychiatry Radiology Surgery

Population data

Count

Percentage

Count

Percentage

129 49.4

18.43 —

700 49

100 —

1 10 5 62 5 9 9 3 12 13

0.78 7.75 3.88 48.06 3.88 6.98 6.98 2.33 9.30 10.08

10 49 26 276 40 10 72 42 25 150

1.43 7.00 3.71 39.43 5.71 1.43 10.29 6.00 3.57 21.43

test of CMV over alternative techniques such as Harmon’s one-factor test. In this technique, a variable theoretically unrelated to the remaining variables in the study is selected as the marker variable, whose mean correlation with other variables in the study (rM) is an estimate of the extent of CMV in the sample. CMV-adjusted correlations between the variables of interest (rA) can be computed by partialling out rM from the uncorrected bivariate correlations (rU), and the difference between rU and rA can be statistically estimated using a t-statistic. In our study, we used respondents’ total medical work experience (at this and other facilities) as the marker variable. This variable had a mean absolute correlation of 0.088 (rM) with the remaining items of interest in this study, which was fairly typical of IT usage studies (e.g., Malhotra et al., 2006). We computed the CMV-adjusted correlations (rA) by partialling out the effect of rM from the unadjusted bivariate correlations (rU) among the remaining items. The mean change in correlations (rUrA) due to this adjustment was 0.081. The CMV-adjusted correlations were not significantly different from the unadjusted correlations, suggesting the lack of CMV bias in our observed data sample.

Scale validation The validity for our measurement scales was assessed using confirmatory factor analysis (CFA). This was performed using the partial least-squares (PLS) technique using Visual PLS 1.04. While we did not have an adequate sample size to conduct covariance-based structural equation modeling approaches such as LISREL, which requires 10 times the total number of indicators in a model, our sample size was adequate for the component-based PLS approach, which requires 10 times the number of indicators in the largest block (Chin, 1998). Raw data were used as input to the PLS program, and path significances were estimated using the bootstrapping resampling technique with 200 sub-samples.


Table 2 Scale item CMP1 CMP2 CMP3 PU1 PU2 PU3 PU4 PT1 PT2 PT3 PT4 RES1 RES2 RES3 RES4 RK1 RK2 RK3 RK4a EOU1 EOU2 EOU3 EOU4 INT1 INT2 INT3

CFA results

Item mean

Item SD

Factor loading

T-statistic

3.77 3.91 3.53 4.01 3.53 3.80 4.21 3.46 3.40 3.62 3.16 5.18 5.71 5.67 5.22 4.50 4.45 4.81 4.98 4.62 4.85 4.06 3.90 5.22 4.92 4.80

2.12 2.16 2.01 2.12 2.22 2.16 2.13 1.82 1.82 1.85 1.83 1.57 1.44 1.45 1.59 1.92 1.86 2.00 1.74 1.88 1.66 1.82 1.93 1.95 1.94 1.86

0.96 0.97 0.92 0.95 0.96 0.98 0.94 0.88 0.92 0.91 0.75 0.87 0.89 0.87 0.86 0.96 0.97 0.95 0.12 0.88 0.89 0.90 0.89 0.93 0.97 0.95

115.17 165.48 46.28 120.44 119.95 326.03 74.92 35.57 48.77 55.64 12.12 33.39 28.61 21.56 29.53 111.63 218.39 87.61 1.39 27.65 33.50 36.64 43.02 43.45 125.85 114.48

CMP: perceived compatibility; PU: perceived usefulness; PT: perceived threat; RES: resistance to change; RK: related knowledge; EOU: perceived ease of use; INT: intention. a Dropped from further analysis due to low factor loading.

Scale validation proceeded in two phases. In the first phase, convergent validity of scale items was assessed using three criteria suggested by Fornell & Larcker (1981): (1) all item factor loadings should be significant and exceed 0.70, (2) composite reliabilities (rc) for each construct should exceed 0.80, and (3) average variance extracted (AVE) for each construct should exceed 0.50. As seen from Table 2, standardized loadings for all scale items, except one, in the CFA model met the minimum loading criterion of 0.70 and were significant at Po0.001. The errant item was the fourth item for related knowledge (RK4), which referred to physicians’ prior knowledge of online ordering (see the Appendix). Given that our target physicians had no prior exposure to CPOE systems, we had expected their prior online ordering experience in general (from websites such as Amazon.com) to correlate with their CPOE ordering. However, it turned out that most physicians had little time outside of their busy work schedules to explore online ordering from websites, resulting in this item loading poorly with the remaining three items of this scale: physicians’ prior knowledge of using computers, software programs, and e-mail. In view of its low factor loading and conceptual ambiguity, the errant item was dropped from the remainder of our analysis.


Table 3 Construct

CMP PU PT RES RK EOU INT

rc

0.86 0.97 0.92 0.92 0.87 0.93 0.86

Scale properties

AVE

0.68 0.91 0.75 0.76 0.69 0.79 0.67

733


Inter-construct correlations CMP

PU

PT

RES

RK

EOU

INT

0.82 0.70 0.49 0.43 0.57 0.62 0.65

0.95 0.51 0.46 0.63 0.56 0.71

0.86 0.54 0.30 0.24 0.50

0.87 0.27 0.22 0.49

0.83 0.53 0.50

0.89 0.39

0.82

CMP: perceived compatibility; PU: perceived usefulness; PT: perceived threat; RES: resistance to change; RK: related knowledge; EOU: perceived ease of use; INT: intention. Diagonal terms (in italics) represent square roots of AVE for that construct.

Composite reliabilities of all constructs (after dropping the fourth related knowledge item) exceeded the required minimum of 0.80, with the lowest value being 0.86 for perceived compatibility and intention (see Table 3). Further, AVE values for all seven constructs exceeded 0.50, with the lowest values being 0.67 for usage intention (see Table 3). Hence, all conditions for convergent validity were met. Discriminant validity between constructs was assessed using Fornell & Larcker’s (1981) recommendation that the square root of AVE for each construct should exceed the correlations between that and all other constructs. In our CFA model, the highest bivariate correlation in our CFA model was 0.71 between perceived usefulness and intention (see Table 3). This correlation was lower than the lowest square root of AVE among all constructs, which was 0.82 for perceived compatibility and intention constructs. Hence, the discriminant validity criterion was also met, assuring that our data sample was of adequate quality for hypotheses testing.

Hypotheses testing The next step in our data analysis was to examine the explanatory power of our hypothesized research model (Figure 1) and the strengths and significances of individual paths in this model. This analysis was also conducted using PLS, after respecifying the CFA model to match the causality hypothesized in Figure 1. The research model explained 55% of the variance in the dependent variable–intention to use HIT – while variance explained in perceived usefulness, perceived ease of use, and resistance to change were 53, 29, and 29% respectively (see Figure 2). Examining individual path coefficients, we find that six of the eight hypothesized paths in our research model were significant at Po0.05, with one path (from resistance to change to perceived ease of use) being marginally significant (Po0.10) and one path (from perceived ease of use to intention) being non-significant. The directionality of each significant path (positive or negative) was also as hypothesized, providing overall support to our research model.

Perceived Compatibility

0.62***

Perceived Usefulness (R2=0.53)

0.62***

-0.20**

Perceived Threat

0.54***

Resistance to Change

-0.21***

(R2=0.29)

Intention to Use HIT (R2=0.55)

-0.09 0.02ns Related Knowledge

0.51***

Perceived Ease of Use (R2=0.29)

Path Significance: *** p