An Empirical Study of Opinion Leader Effects on Mobile Information Technology Adoption in Healthcare

Abstract

Given the increasing number of applications but slow adoption of IT, including mobile IT, in healthcare, it is important to develop a better understanding of the contextual factors that motivate IT adoption by physicians. Although studies have shown that age or gender may affect physicians’ IT adoption, those factors cannot be controlled when deploying a new IT. Therefore, the current research examines empirical evidence of a contextual factor, opinion leader effects, on IT adoption in healthcare that can be influenced by organizational policies. Using a unique panel dataset of physicians’ usage of a mobile clinical IT from a community hospital, we observe a significant result that physicians under the influence of opinion leaders are three times more likely to adopt the IT than otherwise. This finding suggests that incentivizing a small proportion of opinion leaders to adopt a new IT has the potential to motivate wider adoption across the organization.

Introduction

In an effort to control rising healthcare costs and improve quality of care, recent policy initiatives have promoted the adoption of Information Technology (IT) in the healthcare sector, particularly in clinical care delivery¹. However, despite incentives and penalties, the adoption rate is low at both physician office level and hospital level. The adoption rate of basic IT functionality in physician offices was 11% in 2006 and 21% in 2009, and that of full IT functionality was 3% and 6%, respectively¹. Clinical care takes place in multiple, diverse delivery settings such as inpatient, outpatient, emergency and office practice environments. Hence, mobility is a critical aspect of health care delivery. Information technology solutions such as Electronic Health Records and Electronic Prescribing systems are facilitating the availability and utilization of patient information in some settings more than others, due either to the lack of mobile channels of access to the information or to the lack of usage of such technologies at the point of care. Mobile clinical information systems can significantly improve access to data and information wherever and whenever it is needed. Such systems have been shown to have positive impacts on reducing medical errors, saving costs, improving usability and convenience, and enhancing positive attitudes toward wider use of such applications.^2,3,4 As noted in a recent study⁵, while mobile devices are increasingly being used in healthcare, there are few studies that provide a detailed assessment of the range of mobile clinical applications being deployed, the types of uses and users accessing them and the adoption and usage patterns among large groups of physicians, particularly in community health settings⁶.

Hence, in order to promote both clinical IT and mobile clinical IT adoption through appropriate policy interventions and incentives at the physician level, it is important to develop a better understanding of the critical contextual factors that impact adoption decisions. These factors can subsequently be used to motivate clinical IT adoption in healthcare. There is a very broad literature on technology adoption models in the discipline of social psychology, such as Theory of Reasoned Action⁷, Theory of Planned Behavior⁸, and Technology Acceptance Model⁹. These three models apply similar logic by using survey instruments to collect users’ subjective attitude toward a new technology’s perceived usefulness and perceived ease of use, then model and analyze users’ acceptance based on the measures they have defined. However, one of the major limitations of these models is that while they provide insights into users’ perceptions about a technology, they do not provide practical guidance on policies and interventions that may improve adoption of the technology. Studies have also summarized many surveys indicating that age, gender, and practice specialty may influence physicians’ IT adoption behavior¹⁰. However, these factors are not controllable from a policy perspective. Furthermore, a recent study has demonstrated that social network analysis can yield more objective and accurate assessments of the social contagion effect than perception-based measures, with their findings also suggesting that power of persuasion is more likely to develop through a friendship-based social network¹¹. Despite these studies, the magnitude and mechanism of influence of peers on IT adoption by individual physicians in the context of clinical care are yet to be investigated.

Our ongoing research examines these questions related to the impact of peer effects on mobile clinical IT adoption in a community health setting. These peer effects are not symmetric because some users are early adopters and may exert a stronger influence, hence in our study, we call them “opinion leaders”, following Rogers’ definition¹². . However, Rogers describes opinion leader effects on the diffusion process based on data from qualitative studies¹². The current study summarizes our preliminary research to examine the empirical evidence of opinion leader effects on IT adoption using data that includes individual physician level usage of a mobile clinical IT from the time it was deployed in June 2006 until the end of March 2008, demographic information and practice group membership of physicians, and the identification of opinion leaders by the administrators of the healthcare system. The result may be interesting to technology providers, system implementers and decision makers in a variety of ways including (a) to assist technology providers to determine their marketing strategy to promote their products or services, (b) to assist implementers to improve the utilization of the deployed system, and (c) to assist decision makers to plan their technology diffusion strategies.

Study Setting

The study site is a progressive, community-based healthcare system located in southwestern Pennsylvania. In partnership with more than 500 physicians and nearly 4,000 employees, the health system offers a broad range of medical, surgical and diagnostic services at two hospital locations with over 500 beds and five affiliated community satellite facilities. In June 2006, the health system deployed a Mobile Clinical Access Portal (MCAP), which is a secure, wireless PDA-based client-server solution providing physicians with 3 years of on-line clinical data accessible from PDAs via any broadband or Wi-Fi connection. Physicians are able to use it anywhere, anytime, at their convenience, such as at home, in the office, or during the encounter with patients. MCAP also connects to the Clinical Accessible Portal (CAP), their enterprise Electronic Medical Record (EMR) system, accessible from the clinician’s desktop. MCAP initially provided access to 266 medical applications, such as reviewing patient medical histories, using electronic prescribing, placing lab orders, checking lab results, reviewing patient summary data, performing dictation, and related functionality. Not all the features were made available at the same time, but over 75 percent of the features were tried or used in the first three months of deployment of the system. The requirements analysis and system design were updated and many features were revised and changed over time. After one year, approximately 24 features continued to be frequently used, with lab-related and search-related medical applications being the most frequently utilized. The system currently includes 268 physician users, but not all the users received the PDAs at the same time; however, around half the users received the hand-held devices within the first five months of deployment. Usage is voluntary but it was hypothesized that the convenience of using the device in a variety of care delivery settings would incentivize the physicians to become accustomed to accessing electronic patient information at the point of care, thus facilitating the move to a completely paperless electronic record system in the future.

Data

We compiled three datasets from the healthcare organization for the present study. The first dataset included de-identified demographic and practice group information about 250 physicians, comprising a unique identifier, gender, age, primary specialty, sub-specialty, solo vs. group practice membership, medical title and the date when the hand-held device was received. The second dataset recorded MCAP usage data consisting of approximately 363,000 records, representing all applications used at any time by any physician from June 2006 to March 2008. This included physician identifier, usage date and time, and applications that were accessed, representing MCAP usage over 22 months for approximately 268 physician users of 266 clinical applications (no demographic data was available on 18 of these users). The third dataset contained patient visits, detailing the incidence of four types of visits for each physician over the same time period as the MCAP dataset. It was necessary to exclude 58 physicians with some missing demographic information in the first data set (out of 250), leaving 192 physicians in the merged file for the data analysis reported in this paper. All were full time practitioners. Thus the merged data set analyzed in this study included complete demographic and usage information on 192 physicians, 54 of whom were in solo practice and 138 in group practices of varying sizes. Since almost 23 percent (58 out of 250) of the physician records were dropped due to incomplete data, we performed a series of t-tests to check for non-response bias. None of the t-tests were statistically significant.

We divided the 30 specialty areas into two categories in order to examine and control for how practice specialty may affect physicians’ adoption of MCAP. These were: General Practitioner, which includes internal medicine, family practice, and pediatrics; Specialists, which includes the remaining specialty areas. Since we do not believe that age has a linear impact on technology adoption, we group the physicians into three nominal age cohorts: age under 45 years old, between 46 and 55, and above 56 years old. Besides the demographic information and MCAP usage data, we also have four types of patient visit counts for each physician: emergency, inpatient, outpatient, and physician office visit. Not all the physicians had all four types of visits, for example, some physicians did not have emergency visits while a few others did not have inpatient visits. On average, each physician’s daily or monthly patient visit count of any type was quite stable over time, numbering around several hundred to several thousand each month.

We designate a physician practice group as the peer group; this consists of physicians sharing the same specialty, facility and infrastructure. Since this is a community health system, most practice groups are spread out over a fairly wide region,. Therefore, we assume that physicians from the same practice group will have more social and technical interactions amongst themselves than with outside groups, which provides the theoretical foundation for peer effects, or opinion leader effects. We define Opinion Leaders as the physicians identified by the health system as early adopters of MCAP, i.e., those who started using it within the first month of deployment. These physicians were randomly distributed across multiple practices. We thus examine these opinion leaders’ impact on their peers’ technology adoption by comparing users who have opinion leaders in their practice groups with users who have no opinion leaders in their practice groups. Therefore, we remove opinion leaders themselves from the comparison. Solo practitioners are removed as well. We define a dynamic adoption variable for each user by observing each physician’s usage over time and defining an adopter as any user who used the handheld more than 30 times in any of the first four months, or used it 20 to 40 times per month and continued using it for several months. Overall, 109 out of 192 physicians adopted this mobile technology based on our definition. In particular, all opinion leaders were technology adopters. Furthermore, excluding solo practitioners, there were 18 opinion leaders and 120 of their peers in group practices. Tables 1 and 2 summarize the descriptive statistics for these non-opinion-leaders and opinion leaders, respectively.

Table 1.

Descriptive Statistics about Non-Opinion-Leader Physicians in Group Practices

Variable	Mean (Total Number)	Std. Dev.	Min	Max
Adopted	0.56 (67)	0.50	0	1
Male	0.78 (94)	0.41	0	1
Age	48.94	9.70	30	78
Age less 45	0.35 (42)	0.48	0	1
Age between 46 and 55	0.42 (50)	0.50	0	1
Age above 56	0.23 (28)	0.42	0	1
In OPL Group	0.28 (34)	0.45	0	1
Group Size	4.43	3.04	2	12
General Practitioner	0.49 (59)	0.50	0	1
Total usage	1018	2119	2	13438
Total Months	15	5	1	20
Average monthly usage	61	117	0.11	707
Average monthly visit	1583	1419	11	8618
Average monthly emergency visit	58	137	0	1111
Average monthly inpatient visit	74	74	0	353
Average monthly outpatient visit	702	706	0	3780
Average monthly physician office visit	749	919	0	7063

Note: Total number of observations is 120, after removing opinion leader users.

Table 2.

Descriptive Statistics about Opinion Leaders in Group Practices

Variable	Mean (Total Number)	Std. Dev.	Min	Max
Adopted	1(18)	0	1	1
Male	0.89 (16)	0.30	0	1
Age	49.5	6.69	39	60
Age less than 45	0.33(6)	0.48	0	1
Age between 46 and 55	0.50(9)	0.51	0	1
Age above 56	0.16(3)	0.38	0	1
Group Size	4.11	2.25	2	12
General Practitioner	0.78 (14)	0.48	0	1
Total Usage	8101	14020	150	58768
Total Months	21.6	0.5	21	22
Average monthly usage	378	667	7	2798
Average monthly visit	1802	1199	165	3973
Average monthly emergency visit	57	49	0	175
Average monthly inpatient visit	88	47	18.1	170
Average monthly outpatient visit	1066	741	14.9	2467
Average monthly physician office visit	591	591	0.7	1998

Note: The total number of observations in this dataset is 18.

Analytical Model and Results

MCAP provides several advantages for the study of opinion leader effects on new IT adoption. First, we have the individual usage data from the time MCAP was launched until stable use was achieved across the organization. Second, the PDAs were provided to the physicians free of charge by the health system; therefore, there was no user acquisition cost which might confound technology adoption due to users’ heterogeneous socioeconomic situations or reluctance to invest in an unknown technology with uncertain benefits. Third, the PDAs were designed with a very straightforward menu-click interface, so generally the learning curve was low and the non-adoption should not have been due to “difficult to use” reasons. Fourth, the use of the PDA was not required or dependent on the desktop computer system, but was only a supplement to it. Fifth, the healthcare system provided MCAP access to all the physicians’ offices and work locations. Therefore, usage of this PDA solution avoided many complications typically present in consumer behavior studies, such as purchasing cost, difficulty in using it, or lack of incentives to use. Thus the usage data of MCAP technology should primarily reflect users’ preferences based on the utility of the technology.

Also, another critical issue for estimating opinion leader effects on technology adoption is how to isolate opinion leader effects from other factors by using peer group formation information to disentangle problems of endogeneity. Otherwise, we cannot tell what caused the peer users’ adoption behavior, i.e., the opinion leader’s effect or merely other unobserved similarities in technology adoption among opinion leaders and their peers. This unique dataset from the community healthcare system eliminates these concerns. Opinion leaders (OPL) were selected by the social system’s insiders based on their long-term and deep understanding of the internal social system¹², i.e., the health system administrators identified opinion leaders as the more influential, vocal, and active physicians and provided them with PDAs within the first 35 days of the system deployment. Finally, the physicians’ peer groups are practice groups based on specialty, and formed independent of, and much before, the new technology deployment, hence are unlikely to have any correlation with any physician’s technology preferences.

This dataset thus provides us with an ideal scenario for investigating how a new technology is naturally diffused in a complex organization, as well as the users’ and their opinion leaders’ (OPL) behavior, from which we can examine whether opinion leaders affect their peer users’ adoption and usage behavior of a new IT. To empirically investigate the influence of opinion leaders on their peers’ IT adoption within an organization, we propose the following model:

$${\mathrm{Y}}_{\mathrm{i}}={\beta }_0+{\beta }_1{\mathrm{X}}_{\mathrm{i}}+{\beta }_2{\mathrm{G}}_{\mathrm{j}}+{\beta }_2{\mathrm{E}}_{\mathrm{i}}+{\varepsilon }_{\mathrm{i}}$$ (1)

Y_i is a binary variable, 1 indicating if user i adopted the new technology, 0 otherwise. X_i is user i’s individual level characteristics. G_j is opinion leader effects in group j, 1 indicating that user i is in Group j in which an opinion leader is present, otherwise it is 0. E_i is user i’s contextual factors such as working environment or work load. Error term ɛ_i includes all the unobservable endogenous factors, such as technology preference, personal interest, etc. However, as long as the opinion leader selection criteria are orthogonal to those unobserved factors, we will be able to obtain unbiased estimates of the opinion leader effects (G_j). Further, we assume ɛ_i follows a Gumbel distribution and Model (1) can be estimated by logit regression.

We empirically estimate model (1) as:

$$\begin{array}{l}{\mathrm{Y}}_{\mathrm{i}}={\beta }_0+{\beta }_1\hspace{0.17em}{\text{Male}}_{\mathrm{i}}+{\beta }_2\hspace{0.17em}{\text{Age}\_45}_{\mathrm{i}}+{\beta }_2{\hspace{0.17em}\text{Age}\_55}_{\mathrm{i}}+{\beta }_4\hspace{0.17em}{\text{General}\_\text{Practice}}_{\mathrm{i}}++{\beta }_7\hspace{0.17em}{\text{OPLeader}}_{\mathrm{i}}+{\beta }_8{\hspace{0.17em}\text{Emergency}\_\mathrm{v}}_{\mathrm{i}}+\\ {\beta }_8\hspace{0.17em}{\text{InPt}\_\mathrm{v}}_{\mathrm{i}}+{\beta }_9\hspace{0.17em}{\text{OutPt}\_\mathrm{v}}_{\mathrm{i}}+{\beta }_{10}\text{PhyOff}\_{\mathrm{v}}_{\mathrm{i}}\end{array}$$ (2)

We estimate the adoption decision using logistic regression based on user’s gender, age, specialty area, four types of average monthly patient visit volume, and opinion leader effect (see Table 3).

Table 3.

Logistic Regression Results for Opinion Leader Effects on IT Adoption

	Odds Ratio	95% Wald Confidence Limit
Opinion Leader Group	3.125*	1.178	8.288
Male	3.211*	1.083	9.522
Age under 45 years old	2.519	0.816	7.777
Age between 46 and 55 years old	2.593	0.901	7.463
General Practitioner	0.687	0.209	2.26
Average monthly physician office visit	1	0.999	1.001
Average monthly Outpatient visit	1.001*	1	1.003
Average monthly Inpatient Visit	0.985	0.968	1.002
Average monthly Emergency Patient Visit	1.003	0.994	1.011

Note 1: The data size for this regression is N = 120 and pseudo R-square = 0.1517

Note 2:

^*indicates 5% significant level.

From the regression results shown in Table 3, we observe that OPL Group has a statistically significant and positive impact on their peers’ adoption decision. Since all the opinion leaders are adopters, the positive opinion leader effects suggest that opinion leaders may actively influence their peers’ adoption behavior. The odds ratio results show that a user in an opinion leader group would be more than three times more likely to adopt the PDA than a user who has no opinion leader in his or her group, holding other variables constant. Also, the variable Male shows a positive and statistically significant effect on adoption; male physicians are more than three times more likely than female physicians to adopt this new technology, holding other variables constant. Overall, the logistic regression results suggest that opinion leaders do have a strong influence on their peer users’ IT adoption decision.

Another significant factor is the average monthly outpatient visits, which are both continuous variables. The average monthly outpatient number has a statistically significant positive effect on PDA adoption. This may suggest that physicians may prefer to use PDAs when they encounter a patient in the outpatient setting. This might be because using portable PDAs provides convenience for moving around different exam rooms in the outpatient. Both Emergency Visits and Inpatient Visits represent a very small proportion among the all four types of visits as the descriptive statistics show. Neither of them has a statistically significant impact on PDA adoption. The emergency department may accept patients needing urgent care or immediate attention, which depend on emergency care skills but may not require a number of the functions available in MCAP. The number of physician’s office visits is another type of patient visit, the largest proportion of visit among the four types of visits. Its impact is negative, though not statistically significant.

Although several contextual factors show statistically significant effects on physicians’ new technology adoption, we have shown that opinion leader effects, in particular, can have significant impact and can be influenced through policy initiatives. The opinion leaders’ impact on users’ learning have important policy implications for a number of reasons. First, if opinion leader effects on peers’ technology adoption behavior exist, then decision makers can concentrate on working with a finite set of opinion leaders to promote adoption of a new technology within an organization with limited resources through social multiplier effects. That is, an organization may target the opinion leaders to promote new technology adoption to generate wider and multiplied adoption effects. Second, other factors such as gender, age, and work characteristics also have impact on technology adoption. Technology providers, implementers, and decision makers should be aware of these factors, even though it may not always be possible to accommodate them to encourage IT adoption in the clinical care delivery environment.

Conclusion and Limitations

This paper uses objective, individual usage data of a newly implemented mobile technology from a local healthcare system to investigate opinion leader effects on technology adoption in the clinical care setting. The model results show that physicians who were under opinion leader’s influence were more than three times more likely to adopt a new technology than physicians who were not under an opinion leader’s influence, controlling for gender, age, specialty, and workload variables. Besides opinion leader effects, gender and outpatient visit characteristics also have statistically significant impact on users’ technology adoption decision. But, opinion leader effects are the only policy-adjustable variable in the model, which provides practical policy implications on how to motivate new technology diffusion in a healthcare organization.

There are some limitations of this research. First, we do not have qualitative data, such as a survey instrument, on the exact social networking information, i.e., for each user, whose opinions do they seek, particularly regarding this newly deployed PDA system, but we assume the physicians in the opinion leader groups are under the opinion leaders’ influence. We do not know whether any physicians have special connections outside of their practice group, or within the health system. Also, we do not have patient information or detailed workload controls, which might affect physicians’ PDA usage behavior. Another limitation is that we have not explored the mechanism by which opinion leaders influence their group members’ adoption behavior such that their peers learn about the technology and modify their adoption decisions accordingly. These limitations may be solved in future study, such as with on-site surveys and interviews to collect additional social networking information and developing extensions of the econometric models that incorporate a learning component.