6. PREMIER ARTICLE

An adaptation of the Theory of Interpersonal Behaviour to the study of telemedicine adoption by physicians

Gagnon M-P., Godin G., Gagné C., Fortin J-P., Lamothe L., Reinharz D., Cloutier A. (2003). International Journal of Medical Informatics, 71 (1-2), pp.103-115.

Abstract

Based upon the Theory of Interpersonal Behavior, this study was aimed at assessing the predictors of physicians’ intention to use telemedicine in their clinical practice. Physicians were mailed a questionnaire to identify the psychosocial determinants of their intention to adopt telemedicine. Structural equation modelling was applied to test the theoretical model. The adapted theoretical model explained 81% (p < .001) of variance in physicians’ intention. The main predictors of intentions were a composite normative factor, comprising personal as well as social norms (β = 1.08; p < .001) and self identity (β = −.33; p < .001). Thus, physicians who perceived professional and social responsibilities regarding adoption of telemedicine in their clinical practice had stronger intention to use this technology. However, the suppression effect of personal identity in the regression equation indicates that physicians’ intention to use telemedicine was better predicted if their self-perception as telemedicine users was considered.

Keywords: Telemedicine, Technology acceptance and adoption, Psychosocial theory, Structural equation models, Medical profession, Telemedicine diffusion.

Résumé

Cette étude a utilisé la Théorie des comportements interpersonnels afin d’identifier les facteurs prédisant l’intention des médecins d’utiliser la télémédecine dans leur pratique. Un questionnaire a été envoyé à tous les médecins afin d’identifier les déterminants psychosociaux de leur intention d’adopter la télémédecine. La modélisation par équations structurales a permis de tester le modèle théorique. Après ajustement, le modèle expliquait 81 % (p < 0,001) de la variance dans l’intention des médecins. Les construits prédisant l’intention étaient un facteur normatif composite (β = 1,08; p < 0,001 ) et l’identité personnelle (β = -0,33; p < 0,001). Ainsi, les médecins ayant une forte perception de leurs responsabilités professionnelles et sociales au regard de l’utilisation de la télémédecine dans leur pratique auraient davantage l’intention d’adopter cette technologie. Cependant, l’effet de suppression de l’identité personnelle suggère que cette relation est influencée par la façon dont les médecins se perçoivent en tant qu’utilisateurs de la télémédecine.

Mots-clés: Télémédecine, Adoption et acceptation de la technologie, Théorie psychosociale, Modélisation par équations structurales, Profession médicale, Diffusion de la télémédecine.

.

Introduction

Over the past years, the adoption of information and communication technologies (ICT) in the health care sector has been the focus of many studies. Telemedicine, defined as the use of information technologies to exchange health information and provide health care services across geographical, time, social, and cultural barriers (Reid, 1996), has the potential to increase quality and access to health care and to lower health expenditures. This technology is considered as a major innovation at the technological, social, and cultural levels. Thus, the introduction of telemedicine as a tool to support the delivery of health care induces numerous changes for professionals, institutions, and for the health care system as a whole that must be accounted for during the implementation process (Hu, Chau & Sheng, 2000). Telemedicine is expected to impact all levels of health care organisations.

Physicians represent one of the principal groups of telemedicine users and their acceptance of this technology constitutes one of the prerequisites to the emergence and sustainability of telemedicine networks (Hu, Chau & Sheng, 2000). However, the decision of physicians to adopt a new technology such as telemedicine can be challenged by their relatively low computer literacy, the possible alteration of their traditional routines, and their high professional autonomy. Many studies have investigated physicians’ acceptance of various telemedicine applications over the last ten years (Hu et al., 1999a). These studies were of exploratory nature and were often limited to the measure of attitudes and perceived barriers. Furthermore, most of these studies were based on small samples and did not used explicit theoretical foundation to test their hypotheses (Hu et al., 1999).

Theoretical background

Among the studies of telemedicine adoption that were based on a theoretical framework, most employed the Theory of Planned Behaviour (TPB) (Ajzen, 1991) or the Technology Acceptance Model (TAM) (Davis, 1989). The validity of the TPB was demonstrated in a study of telemedicine adoption by physicians (Hu & Chau, 1999). This study reported that attitude was the principal determinant of physicians’ intention to use telemedicine, while perceived behavioural control had also a lesser but significant effect on intention. However, social norms were not found to significantly influence intention.

Derived from the TPB, the TAM was specifically designed to study the adoption of technology. In its original version, the model considers intention as the direct determinant of behaviour, while attitude and social norms are the predictors of intention (Davis, 1989). However, the TAM decomposes the attitude construct into two distinct factors: perceived ease of use and perceived usefulness . Many studies have empirically tested the TAM for the prediction of adoption behaviours for various technologies (Hu et al., 1999; Croteau & Vieru, 2002).

An investigation of telemedicine adoption among physicians in Hong Kong (Davis, 1989) found reasonable support for the TAM. The model explained 44 % of the variance in physicians’ intention to use telemedicine. This study has also demonstrated that intention was mostly determined by perceived usefulness. In counterpart, perceived ease of use of the technology did not influence significantly its adoption. These authors argued that other constructs should be added to the TAM for the study of technology adoption by physicians in order to enhance its explanatory power and its applicability in the healthcare sector.

Croteau and Vieru (2002) used an adaptation of the TAM to explore the factors affecting telemedicine adoption by two groups of physicians in Canada. Perceived usefulness was the main predictor of adoption in both groups, while perceived ease of use was significantly associated to adoption in only one group. The concepts of image and perceived voluntariness of use were also added to the original TAM in this study. Image, defined as the perceived impact of technology adoption on one’s social status, was not significant, while perceived voluntariness of use was negatively correlated to adoption (contrary to their hypothesis), but only in one group.

The influence of social factors has not been significant in most of the studies of telemedicine adoption by physicians (Hu, Chau & Sheng, 1999; Hu & Chau, 1999; Croteau & Vieru, 2002). It has been recognised that the medical profession is characterised by the relative autonomy of physicians and their independence in decision-making (Tanriverdi & Venkatraman, 1999). However, a technology that could interfere with physicians’ traditional practice could affect their perception of their professional role. Furthermore, as other professionals, physicians are committed to their profession and look to their peers for acceptable standards of performance (Tanriverdi & Venkatrama, 1999). As suggested by Succi and Walter (1999), the addition of specific determinants to psychosocial models, such as the perceived impact of using the technology on professional status, should be tested in further studies of telemedicine adoption by physicians.

In response to some of these concerns, the purpose of this study is to propose and to test a model that explores telemedicine acceptance determinants among physicians using a conceptual framework specifically adapted to the particular characteristics of the medical profession and the health care sector.

Conceptual model and research hypotheses

Among the psychosocial theories developed to understand the adoption of behaviours, Triandis’ Theory of Interpersonal Behaviour (Triandis, 1980) encompasses many of the behavioural determinants found in other models such as the TPB and the TAM. However, the TIB has a wider scope in that sense that this model also considers cultural, social, and moral factors that are not accounted for in other theories (Facione, 1993).

According to Triandis (1980), behaviour is determined by three dimensions: intention, facilitating conditions, and habit. Intention refers to the individual’s motivation regarding the performance of a given behaviour. Facilitating conditions represent objective factors that can make the realisation of a given behaviour easy to do. Conversely, barriers consist of factors that can impede or constrain the realisation of the behaviour. Habit constitutes the level of routinization of a behaviour, i.e. the frequency of its occurrence. As suggested by Triandis, habit can also exerts an influence on the emotive component of attitude (affect).

In the TIB, intention is formed by attitudinal, normative, and identity beliefs. Affect represents an emotional state that the performance of a given behaviour evokes for an individual. It is considered as the affective perceived consequences of the behaviour, whereas perceived consequences refer to the cognitive evaluation of the probable consequences of the behaviour. Perceived consequences encompass the perceived usefulness construct found in the TAM. The TIB incorporates two different normative dimensions: social and personal norms. Perceived social norms are formed by normative and role beliefs. Normative beliefs consist of the internalisation by an individual of referent people or groups’ opinion about the realisation of the behaviour, whereas role beliefs reflect the extent to which an individual thinks someone of his or her age, gender and social position should or should not behave. The other normative component of the TIB is the personal normative belief that represents the feeling of personal obligation regarding the performance or not of a given behaviour. Finally, self identity refers to the degree of congruence between the individual’s perception of himself or herself and the characteristics he or she associates with the realisation of the behaviour.

To the best of our knowledge, this model has not previously been applied to the study of telemedicine adoption by physicians. However, the TIB was used in some studies of information technology adoption by different groups of workers (Bergeron et al., 1995; Thompson, Higgins & Howell, 1991). For instance, Thompson et al. (1991) have tested the TIB in relation to personal computer use. Their model explained 40% of the variance in the behaviour. Paré and Elam (1995) employed a subset of Triandis’ model to explore determinants of computer use among knowledge workers. They found limited support for the TIB with less than 30% of explained variance in behaviour. The main predictors of computer use were beliefs, affect, social norms, facilitating conditions and habit. Finally, a study by Bergeron et al. (1995), found that knowledge workers’ internalisation of an information system was predicted by their affect towards the system and the perceived consequences of using it (R2 = .52). However, the TIB variables could not explain significantly information systems utilisation in this study. Although these results moderately support the use of the TIB, this model was adopted in the present study since the target population (physicians) differs on many respects from other studies. This conclusion is further supported by the observation that the constructs of role beliefs, self identity, and personal normative belief found in the original TIB were excluded in all of the reported studies.

Telemedicine adoption refers to physician’s psychological state with regard to his or her intention to use telemedicine in his or her practice (Croteau & Vieru, 2002). Telemedicine acceptance can be defined in different manners and adoption (or utilisation) represents a common indicator of the degree of telemedicine acceptance. Thus, the dependant variable measured in the present study is intention to use telemedicine . An individual’s intention to use telemedicine is considered as an appropriate measure of his or her actual use of the technology (Hu, Chau & Sheng, 1999). Moreover, meta-analysis on the use of psychosocial models in the study of health behaviours found high correlation between the intention to perform a given behaviour and the actual behaviour (Godin & Kok, 1996).

Figure 6.1. Conceptual model (adapted from Triandis, 1980)

For the purpose of this study, behavioural determinants of the intention were adapted from the original Triandis’ model with minor modifications. Firstly, two constructs, habit and facilitating conditions, were hypothesised to be linked directly to intention in our model while they are conceptualised as direct antecedents of behaviour in the original model. This was done because previous studies that employed Triandis’ theory have found that facilitating conditions and habit were important predictors of intention (Boots & Treloar, 2000). Also, a mediating effect of affect on the association between habit and intention was tested in our model, as suggested by Triandis.

The following hypotheses were tested:

Affect is a predictor of physicians’ intention to use telemedicine

Perceived consequences are predictors of physicians’ intention to use telemedicine

Perceived social norms are predictors of physicians’ intention to use telemedicine

Personal normative belief is a predictor of physicians’ intention to use telemedicine

Self identity is a predictor of physicians’ intention to use telemedicine

Facilitating conditions are predictors of physicians’ intention to use telemedicine

Habit is a predictor of physicians’ intention to use telemedicine

Affect has a mediating effect on the relation between habit and intention

Methods

Instrument development and validation

As recommended by Davidson et al. (1976), an etic-emic approach, inspired from the field of anthropology, was used to develop research instruments according to the TIB’s constructs. Firstly, a survey was realised among a convenience sample of physicians attending a conference on telehealth. An open-ended questionnaire, comprising ten questions, was distributed to a total of 60 physicians. The questions dealt with: a) physicians’ perceived telemedicine pros and cons; b) barriers and facilitating conditions affecting telemedicine use; c) emotions related to telemedicine utilisation; d) individuals or groups favourable or unfavourable to one’s utilisation of telemedicine; e) characteristics of telemedicine users; f) personal values related to telemedicine; and g) information and communication technologies used in practice. Forty-two questionnaires were returned completed (70%). A content analysis was performed to extract the salient modal beliefs among physicians, i.e. the beliefs that are common in this subgroup. This step constituted the emic component. The responses given by more than 25% of physicians were kept to form the items for each of the theoretical constructs that represent the etic component. A content analysis was performed independently by two researchers who had to reach an agreement on the classification and labelling of themes extracted. Thus, the number of items composing each constructs varied according to the number of popular responses given by physicians. The questions were formulated following social psychology theorists’ consensus for the development of questionnaires (Ajzen, 1991).

Secondly, a test-retest was performed to assess the reliability of the questionnaire with a sample representative of the studied population. A total of 20 physicians completed the same version of the questionnaire with a two-week interval. Results indicated good construct reliability, with Cronbach alpha varying from to .71 to .90 for the theoretical variables, which is considered satisfactory for an exploratory study. The temporal stability was assessed by calculating the intra-class correlation coefficients for each theoretical construct. Results varied from .46 to .98, which represent moderated to very good coefficients of agreement. Minor modifications were made to the final version of the questionnaire, following commentaries of the respondents.

Variables measured

In this research, the targeted behaviour was the intention of physicians to use telemedicine in their practice. The following definitions were printed on the questionnaires:

Telemedicine refers to any medical service provided at distance via an electronic communication.

In your practice refers to any act of consultation, diagnosis, treatment, or follow-up provided to a patient on site or at distance.

The intention (α = .84) to use telemedicine was measured by means of three items: “I estimate that my chances of using telemedicine in my practice are...” (7-point scale: very high, 7; very low, 1); “If I have the opportunity, I will use telemedicine in my practice” (7-point scale: strongly agree, 7; strongly disagree, 1); and “ I intent to use telemedicine in my practice” (7-point scale: strongly agree, 7; strongly disagree, 1).

The determinants of the affective dimension (affect) of attitude were obtained using a semantic differential 7-point scale made up of two pairs of adjectives, appearing after the sentence: “ For me, using telemedicine in my practice would be...”. The bipolar adjectives proposed were: stressful-relaxing and satisfying-dissatisfying. The Spearman correlation coefficient for this construct was .49 (p < .001).

For the cognitive component of attitude, or the perceived consequences (PC ), only one arm of the belief-based measure was obtained, that is b. As suggested by Gagné and Godin (2000), this method yields high coefficients of correlation with the direct determinant. This is also consistent with other studies based on the TIB that used a direct measure of the perceived consequences associated with the realisation of the behaviour (Boots & Treloar, 2000). Thus, seven items were used to assess the perceived consequences of using telemedicine (α = .72). Five items were worded as follows: “Using telemedicine in my practice would...” 1) facilitate access to expertise; 2) necessitate more time; 3) allow to update my knowledge; 4) reduce patients transfers; and 5) help my decision making. The other two items were: 6) “The definition of my professional roles and responsibilities would not be clear if I were using telemedicine in my practice”; and 7) “My relationships with patients would be less human if I were using telemedicine in my practice”. Each item was measured on a 7-point scale, ranging from 1 (strongly disagree) to 7 (strongly agree). Reverse score was computed for the items negatively formulated (items 2, 6, and 7).

The normative beliefs (NB) (α = .76) were assessed by asking the respondents to indicate their level of agreement, on 7-point scales, with the following four statements: 1) “If I were using telemedicine in my practice, my patients would...” (strongly approve, 7; strongly disapprove, 1); 2) My colleagues would recommend that I use telemedicine in my practice” (strongly agree, 7; strongly disagree, 1); 3) “The consulting specialists would recommend that I use telemedicine in my practice” (strongly agree, 7; strongly disagree, 1); and 4) “The hospital managers would encourage me to use telemedicine in my practice” (strongly agree, 7; strongly disagree, 1).

The measure of role beliefs (RB) (α = .85) was obtained by using three items. These items were worded as follows: “I consider that using telemedicine is correct for a physician of...” 1) my speciality, 2) my region; and 3) my age. All three items were measured on a 7-point scale, ranging from 1 (strongly disagree) to 7 (strongly agree). Consistent with Triandis’ theory, items measuring role beliefs and normative beliefs were aggregated into a single construct, perceived social norms (SN) (α = .85).

Personal normative belief (PNB) (α = .75) was measured by means of three items. Respondents were asked to evaluate, on 7-point scales, the following statements: 1) “I would feel guilty if I was not using telemedicine in my practice” (strongly agree, 7; strongly disagree, 1); 2) “Using telemedicine would be in my principles” (strongly agree, 7; strongly disagree, 1); and 3) “It would be unacceptable to not use telemedicine in my practice” (strongly agree, 7; strongly disagree, 1).

Measure of self identity (SI) (α = .66) was obtained by calculating the difference between physicians’ beliefs regarding characteristics of telemedicine users and their evaluation of the importance of these characteristics for themselves. Firstly, the three characteristics assessed were: 1) “A physician who uses telemedicine shows an innovative mind”; 2) “Using telemedicine is a proof of a physician’s competence”; and 3) “A physician who uses telemedicine is concerned by the quality of patients care”. Respondents’ level of agreement with each item was assessed on 7-point scales (strongly agree, 7; strongly disagree, 1). Secondly, the importance of each of these characteristics was assessed by asking the respondents to evaluate themselves, on 7-point scales (strongly agree, 7; strongly disagree, 1), on these statements: 1) “I consider myself as someone with an innovative mind”; 2) “ I consider myself as competent”; and 3) “I am concerned by the quality of patients care”. Finally, the absolute value of the difference between the two scores was calculated to form the personal identity construct. Thus, the possible scores vary from 0 indicating a high agreement between characteristics of telemedicine users and physicians’ self-evaluation, to 6, indicating a poor agreement.

Facilitating conditions (FC) (α = .77) were assessed by asking the respondents to indicate to what extend the following elements could impede telemedicine utilisation in their practice: time, technology quality, clinicians resistance, consultants availability, lack of qualified personnel, technology availability, remuneration, costs, and clinical complexity. All these items were rated on a 7-point scale, from extremely likely (1) to extremely unlikely (7). For the purpose of structural equation modelling, the facilitating conditions construct was decomposed into two variables. This was done firstly because a principal component analysis has indicated that this construct was formed by two distinct factors. Moreover, Dwyer et al. (1998) have established a distinction between control factors that depends on the individual’s resources and skills (internal barriers), and those that are external to the individual. Thus, the two constructs were created by combining the nine items of the facilitating conditions construct into two categories: external factors (α = .79) and internal factors (α = .67).

Finally, habit (H) was measured by asking the respondents if they have used telemedicine in the past, as well as their frequency of use. The respondents’ scores were grouped into the following categories: 0 (never); 1 (once); 2 (two to four times); and 3 (5 times or more). Since habit was assessed by a single item, its error variance was estimated by multiplying the construct’s measurement error found in the test-retest (.22) by its variance (1.28), as suggested by Kline (1998). Thus, the error variance parameter for habit was fixed to .28.

Studied population and sample

The survey questionnaire was distributed to all general practitioners (GPs) and specialists of the 32 hospitals involved as telemedicine services requestors in the RQTE (the extended provincial telemedicine network of Quebec). This network was created in 1998 to provide specialised consultations in paediatric cardiology to hospitals across the Province of Quebec. With the expected diffusion of telemedicine technology to other medical specialities within this provincial network, the study sample included all active physicians (physicians in administrative position or in public health were excluded). Data from the “Régie de l’assurance maladie du Québec” (RAMQ), the official government agency responsible for the administration of health care services in Quebec, were obtained to estimate the number of full-time equivalent physicians practising in the 32 hospitals of the RQTE. Furthermore, a validation of the total number of physicians (part- and full-time) was made with hospitals. Data compiled from each of the 32 hospitals indicated that the total number of active physicians was 3,832. This number, however, overestimate the actual number of physicians targeted in the present study because several physicians have more than one hospital affiliation.

Out of the 3,832 mailed questionnaires, 538 were received. Among those received, seven were returned uncompleted, six were received by physicians in community health or in administrative position, three physicians refused to complete the questionnaire, two were returned by a physician who had received three copies of the questionnaire, and one was received by a dentist. Thus, 519 questionnaires were satisfactorily completed. Also, the variation in response rate was considerable between hospitals, with proportions of respondents varying from 7% to 50%. Hospitals from remote areas were those with the highest response rate while urban hospitals had the lowest participation.

As the study questionnaires were entirely anonymous, it was not possible to identify physicians who did not return their questionnaire. However, the possibility of non-response bias was assessed by comparing the respondents with the Quebec physicians’ population. As presented below, physicians in the sample had similar characteristics than the general population of physicians in Quebec on most of the control variables measured (age, gender, and speciality). The only exception pertains to physicians’ region of practice: outlying and remote regions were over-represented in the sample.

Data collection procedure

Contacts with the CMDP (Council of Physicians, Dentists and Pharmacists) or the DSP (Professional Services Direction) had been made previously to identify a local contact person in every hospital. This contact person was responsible for the promotion of the study in the hospital, the distribution of questionnaires, and the follow-up.

Each contact person of the 32 hospitals was mailed a number of packets corresponding to the total number of physicians practising in the hospital. The packets contained a letter explaining the purpose of the study, a consent form, a questionnaire and a pre-stamped envelope. This pre-stamped envelope was to be mailed directly to the researchers with the completed questionnaire and the signed consent form. The contact persons were responsible for the distribution of packets to the physicians by internal mail. Two weeks later, the contact person of each hospital distributed recalls to all physicians. Another recall was sent by the same procedure three weeks after the first recall. This last letter indicated that another copy of the questionnaire could be obtained if needed from the local contact person identified.

A unique identification number that indicated the code of the hospital and the code of each respondent was printed on the questionnaire. However, questionnaires were completely anonymous since physicians’ names were not linked to their identification number. Thus, none of the material sent to the physicians was personalised. This study was approved by the ethic committee of the local university.

Statistical analyses

A structural equation modelling approach (SEM) was performed using the EQS version 5.7. Analyses were conducted in two major stages. The first step consisted in a confirmatory factor analysis (CFA) to assess the measurement model. Thus, the correspondence between observed variables and the latent constructs hypothesised was tested. In the second step, the adequacy of the TIB in explaining intention to use telemedicine by physicians was tested. Also, an analysis was performed to test whether affect had a mediating effect on the relationship between habit and intention. This significant mediation effect was incorporated into the global model.

The maximum likelihood method was used to estimate the parameters of the model. As recommended by Byrne (1994), the following statistics were considered to assess the fit of the model: the chi-square value (χ2), the Satorra-Bentler scaled statistic (S-Bχ2), the corrected Comparative Fit Index (*CFI), the corrected Nonnormed Fit Index (NNFI*), and the corrected Root Mean Squared Approximation of Error (RMSEA*). The chi-square statistic is sensitive to sample size and is not recommended for data with non normal distribution. The Satorra-Bentler scaled statistic (S-Bχ2) corrects the χ2 when the assumptions of normality in data distribution are violated (Byrne, 1994). The corrected values of the indexes *CFI, NNFI*, and RMSEA* were computed from the S-Bχ2. Based on the recommendation of Hu & Bentler (1995), a cut-off value close to .95 for NNFI* and *CFI and a cut-off value close to .06 for RMSEA* were used to assess the goodness of the model to fit the data. A raw data file was submitted to EQS in order to obtain the Satorra-Bentler scaled statistic. Then, the program created the covariance matrix used for analyses.

Missing values

To assure construct validity of the measure, data were retained only for subjects having answered at least 70% of the items for each construct. Since the EQS program requires that there are no missing data, mean imputation was applied for respondents having 30% or less missing items scores for a given construct. In such cases, the missing values were replaced by the total sample mean for this item. The final sample size for structural equation modelling was 506.

Results

Descriptive statistics of the sample are presented in Table 6.1. The mean age of physicians in the sample was 43.9 years. More males than females participated in the study. Also, specialists accounted for 57% of the sample. These proportions are similar to the “College des médecins” (College of Physicians of Quebec) data on Quebec physicians’ profile. However, the practice location of physicians in the sample differed from the provincial average. A majority of respondents was practising in hospitals located in suburbs or small towns. These proportions are consistent with the characteristics of the telemedicine network under study, which includes mainly local or regional hospitals from the different health regions of the Province of Quebec.

Table 6.1. Demographic characteristics of respondents

Sample characteristics (n = 506)

Frequency

Gender *

Male

Female

311 (62.2%)

189 (37.8%)

Type

GP

Specialist

220 (43.5%)

286 (56.5%)

Region Urban

Suburban

Remote

63 (12.5%)

281 (55.5%)

162 (32.2%)

Mean age (sd)

43,9 (±9.9)

Mean years of practice (sd)

16,2 (±10.6)

GP: General practitioners

* n = 500

Measurement model

The first step of data analysis was to assess the adequacy of the measurement model by a confirmatory factor analysis (CFA). The model tested included original theoretical constructs (intention, affect, perceived consequences, self identity, habit) and a composite normative construct (personal normative belief + perceived social norms). This normative construct was created to take into account the multicollinearity between the perceived social norms and the personal norm constructs. According to Kline (1998), multicollinearity is present when the correlation between two independent variables is greater than .85. After adjustment, the coefficients of correlation between each independent variable of the model were all satisfactory. A Cronbach alpha of .86 indicated adequate internal consistency for the composite normative construct. However, the facilitating conditions construct was excluded because of its poor fit in the measurement model. Thus, hypothesis 7 was rejected. The CFA performed on this modified model indicated a relatively good fit of the data. Values of the corrected indexes of fit were satisfactory: *CFI = .93; NNFI* = .91; and RMSEA* = .06.

Structural model

During the second step of data analysis, various structural models were compared. Firstly, the mediation effect of affect on the relation between habit and intention was assessed following Baron and Kenny’s procedure (1986). As hypothesised, affect had a mediation effect on the relation between habit and intention. However, this effect was only partial, since there was still a significant association between habit and intention, after taking affect into account. Consequently, both direct and indirect effects of habit on intention were tested in the structural model. Secondly, a complete model, including the partial mediation effect of affect, was tested in order to assess the validity of the TIB for the prediction of physicians’ intention to use telemedicine. This model is shown in Figure 6.2.

Figure 6.2. Complete TIB structural model

The scaled χ2 value was significant (S-Bχ2 = 325.23; df = 120, N = 506), but lower than the null model χ2 (3858.54; df = 153, N = 506). The fit indexes indicated a relatively good fit for this model, with values of .93 for the *CFI and the NNFI*, and a value of the .06 for the RMSEA*. Overall, the TIB proved to be an acceptable model to explain intention to use telemedicine. However, some of the structural coefficients were not significant in this model (Figure 2). Even if the direct effect of habit on affect was significant, neither the affect nor the habit constructs were significant predictors of intention in the model. Since three of the theoretical constructs – affect, habit, and perceived consequences – were not significant predictors of intention, another model was tested with only the significant predictors of intention. In structural equation modelling, it is suggested to reestimate a model, removing the nonsignificant parameters of the original model by fixing them to zero (Hays, 1989). In the final model, shown in Figure 6.3., only the significant predictors of intention were kept. This approach did not change the initial findings of the study, but the value of the fit indexes increased for the parsimonious model. Hence, the model scaled χ2 was nonsignificant (S-Bχ2 = 44.25; df = 17, N = 506) and the value of the *CFI was .98. The NNFI* and the RMSEA* were also satisfactory, with respective values of .97 and .05.

Figure 6.3. Final structural model

The standardised structural coefficients for the final model as well as the variance in intention to use telemedicine explained by this model are presented in Table 6.2. The strongest predictors of intention are normative factors (β = 1.08), which encompass social as well as personal norms. Self identity is also a significant predictor of intention (β = −.33), but in the opposite way of what hypothesised. Surprisingly, self identity has a negative weight in the prediction of intention. Given the positive correlation between the two predictors (r = .64), a net suppression effect (Cohen & Cohen, 1983) of the self identity construct was suspected. This indicates that a part of the variance in the normative factors that is not relevant for the prediction of intention has been removed by including the self identity construct into the equation. Together, these two constructs explain 81% of the variance in physicians’ intention to use telemedicine.

Table 6.2. Standardised structural model coefficients (final model)

Standardised coefficient

Value (corrected standard error)

Factor correlations  

NF – int

.88*

SI – int

.37*

NF – SI

.64* (.03)

   
Path coefficients  

NF – Int

1.08* (.18)

SI – Int

-.33* (.24)

   
Explained variance  

Int

81%

NF; normative factors; SI: self identity; Int: intention

* p-value < .001

Discussion

In this study, the TIB was adopted as a basis for examining the predictors of physicians’ intention to use telemedicine in their practice. Specifically, normative factors – comprising social as well as personal norms – are the best predictors of intention. In addition, self identity has been found to have a suppression effect in the prediction of physicians’ intention to use telemedicine. These findings have several implications, at the theoretical as well as the practical level. Each of these aspects is presented below, followed by a discussion on the study’s limitations.

Theoretical implications

Overall, results suggest that the TIB is an appropriate model to predict physicians’ intention to use telemedicine in their practice, considering the high proportion of variance explained by the structural model. Relative to other studies that have explored telemedicine acceptance among physicians with structural equation modelling (Ajzen, 1991; Hu & Chau, 1999; Croteau & Vieru, 2002), the 81% of variance explained by the structural model is noteworthy. A confirmatory factor analysis was performed to test the measurement model and has indicated strong relationships between the latent constructs and their corresponding measurement items. However, multicollinearity was present between some of the theoretical constructs proposed in the original TIB model. Thus, the social and the personal normative factors were aggregated to form a single normative indicator. This is consistent with what some authors have suggested concerning the presence of a general normative construct, comprising social and personal dimensions, that influence intention of individuals to perform a given behaviour (Fishbein, 1967).

Furthermore, many of the constructs proposed in Triandis’ original model did not significantly predict intention in our study. In previous studies based upon the TIB (Bergeron et al., 1995; Thompson, Higgins & Howell, 1991), perceived consequences and affect were strong predictors of technology acceptance. In contrast, the present study has shown that these attitudinal components did not significantly influence telemedicine acceptance by physicians. The fact that telemedicine is a different technology than the one analysed in other studies, i.e. personal computer or Internet, could explain this finding. Similarly, the target populations in those studies were knowledge workers or students and thus, differ from the population studied here. Indeed, the decision to use telemedicine implies not only a personal evaluation of its benefits by physicians, but highly depends upon the context in which the clinical act is performed where physicians must comply to hospital managers’, colleagues’, and patients’ expectancies. Moreover, the feeling of professional responsibility is central to physicians’ decision-making (Tanriverdi & Venkatraman, 1999) and therefore, influences their acceptance of telemedicine technology.

In their study of telemedicine acceptance, Hu and collaborators (1999) found that physicians’ perceived control over telemedicine utilisation, as measured by proper training, technology access, and in-house technology expertise, was positively associated with intention (Hu & Chau, 1999). Unfortunately, the impact of facilitating conditions (FC) on intention could not be tested in the present study since this construct was removed from the structural model. The CFA performed has indicated a poor fit of the FC items with the measurement model. Thus, the effect of facilitating conditions and barriers may not be adequately captured by the items found in the questionnaire. A plausible explanation to this could be that the FC items were selected from a survey among physicians attending a congress on telemedicine and who were more familiar with this technology. Hence, those items may not had the same meaning for all physicians in the sample. The limited penetration of telemedicine technology in most of the surveyed hospitals could make it difficult for physicians to anticipate potential barriers or facilitating conditions to its utilisation. Consequently, special attention should be given when selecting facilitating conditions items for the study of telemedicine adoption in order to take the degree of exposure to the technology into account. Furthermore, in Triandis’ original model, facilitating conditions are hypothesised as direct behavioural determinants and not as predictors of intention. Thus, other studies should investigate the impact of facilitating conditions on telemedicine utilisation by physicians in the context of a larger diffusion of this technology.

Habit, measured by the frequency of telemedicine utilisation in the past, did not appear as a strong predictor of future utilisation. This is consistent with Bergeron et al. (1995) who found that neither frequency of use nor internalisation of information systems was predicted by past experience. As Thompson et al. (1991) have stated, the measure of habit as the frequency of a behaviour’s occurrence may not be appropriate. They advocate that the frequency of technology utilisation was identical to the measure of utilisation itself (behaviour). In the present study, habit was assessed by a single item, which may have been insufficient to capture the effect of physicians’ past experiences with information and communication technologies. Paré and Elam (1995) have employed a multidimensional measure of habit and found a significant relationship between this variable and personal computer utilisation.

Contrary to our findings, studies that have investigated the determinants of telemedicine acceptance by physicians have found limited support for the impact of social factors on intention to use this technology (Hu et al., 1999; Hu & Chau, 1999; Croteau & Vieru, 2002). As suggested by Succi and Walter (1999), the measure of social norms in psychosocial models may not be appropriate to assess the normative dimension of telemedicine acceptance by physicians. Integrating physicians’ perceived impacts of telemedicine utilisation on their professional status and their beliefs regarding moral responsibility to use this technology could thus improve the measure of the normative construct.

On other respects, the unexpected relationship between the self identity construct and intention to use telemedicine deserve attention. As some researchers have suggested (Courville & Thompson, 2001), it is important to interpret the correlation coefficient between a given predictor and the criterion variable in conjunction with standardised beta weights. In the measurement model, there was a positive correlation of .37 between self identity and intention, while the beta weight was -.33. As the self identity and the normative factors constructs were positively correlated (r = .64), a net (Cohen & Cohen, 1983) suppression effect was detected. As Maasen & Bakker (2001) indicate, it is important to acknowledge the occurrence of a suppression situation in structural models and to consider it when interpreting the results. A suppressor variable increases the predictive validity of another variable by its inclusion in the regression equation (Maassen & Bakker, 2001). In fact, including the self identity score with a negative weight suppressed irrelevant variance in the latent normative construct, thus enhancing prediction of intention by this item. The beta weight of 1.08 for the normative factors is explained by the presence of this suppressor variable (Deegan, 1978).

In this study, self identity refers to the degree of correspondence between physicians’ perception of telemedicine users’ characteristics and their auto-evaluation of these characteristics for themselves. When included in the regression equation, this construct clears out the variance reflecting self-identity concept from the variables measuring professional as well as social normative beliefs. In social psychology, a distinction is made between private, social, and collective self (Triandis, 1989; Ybarra & Trafimow, 1998). The private self represents the assessment of the self by the self (e.g. “I am competent”); the public self is an assessment of the self by the generalised other (e.g. “People think I am competent”); and the collective self corresponds to the assessment of the self by a specific reference group (e.g. “My co-workers think I am competent”) (Triandis, 1989; p.509). These three facets of the normative construct may influence individual behaviours in different manners depending on the context [31]. As noted by Triandis (1989), the cultural context influences which normative cognitions are “sampled” by individuals in the formation of salient beliefs. Thus, intention to use telemedicine is principally influenced by public and collective normative factors and removing the effect of physicians’ self-perception as telemedicine users (or private self) could increase the predictive validity of the normative construct.

Implications for telemedicine diffusion

The normative factors involved in physicians’ intention to use telemedicine are both personal and social, but are primarily of professional nature. The “significant others” who could exert an influence on the decision of physicians to use telemedicine are colleagues, consulting specialists, hospital managers and patients. Similarly, the way physicians perceive their social role as professionals influences their acceptance of telemedicine. For instance, those who believe that using telemedicine is normal for physicians of their region would be more likely to use this technology. Thus, in their decision to use telemedicine, physicians seem to be mostly influenced by cognitions from the “collective self”, i.e. their perception of what the social groups to which they belong expects from them. Hence, to promote the diffusion of telemedicine, campaigns should include messages from peers, patients and telemedicine specialists, and insist on the relevance of telemedicine for physicians of different regions and specialities.

The feeling of professional responsibility also exerts a strong influence on physicians’ intention to use telemedicine in their practice. The promotion of telemedicine diffusion in the health care system should target the benefits for patients with respect to equity in access to specialised medical services, quality and continuity of care. Consequently, physicians would be more likely to perceive the use of telemedicine as a professional obligation towards the well being of their patients.

However, since self identity (or private self) plays a suppression role in the relationship between normative factors and intention to use telemedicine, it is important to take its influence into account. Practically, physicians who do not perceive themselves as telemedicine users would more likely be concerned by messages addressing normative beliefs towards telemedicine use. For physicians who consider they have the attributes of telemedicine users, i.e. those who sample cognitions from the private self in their decision-making (Triandis, 1989), messages focusing on collective norms would be less efficient. Thus, messages promoting the use of telemedicine in medical practice should be selected with caution and tailored to the characteristics of physicians.

Limits of the study

This study has some limitations. First, in spite of a strategy involving local contact persons in each hospital, the response rate was low and varied a lot between hospitals. Low participation has been frequently reported in previous studies of telemedicine acceptance by physicians (Hu et al., 1999; Hu & Chau, 1999). Modest financial incentives have proved effective to increase physicians’ participation in mail surveys (Donaldson et al., 1999). Other strategies could also be explored to ensure better response rates in subsequent studies, such as involving departments chiefs of service (Hu & Chau, 1999) or promoting the study during CMDP meetings.

A second limitation of this study pertains to the generalisability of the results. The population under study was formed by all the physicians practising in hospitals of the RQTE. Thus, the sample is composed of volunteers from this population and is representative of a certain type of physicians. Therefore, responses to this study are subject to self-selection biases. Globally, characteristics of physicians in the sample correspond to those of the whole Quebec physicians population, with the exception of the over-representation of physicians from remote and outlying regions. Since telemedicine is primarily aimed at improving access to specialised healthcare services in remote regions, physicians from these regions are generally more aware of the different applications of this technology and could have a better opinion towards it. Although comparisons between physicians’ responses across regions have not indicated significant differences in intention to use telemedicine, more research is needed to explore contextual factors that impact technology acceptance by healthcare providers.

Thirdly, although the TIB has been satisfactory in predicting a large proportion of variance in physicians’ intention to use telemedicine, some of the theoretical hypotheses were rejected. Also, a global normative item was created because of multicollinearity between social and personal normative components. Furthermore, facilitating conditions were deleted from the structural model since the measurement model of this construct was unsatisfactory. These theoretical limitations call for the use of structural equation modelling in prospective empirical studies based on the TIB in order to validate the model.

Despite these limitations, the present study contributes considerably to the understanding of telemedicine acceptance by physicians. This study was the first, to the best of our knowledge, to employ Triandis’ Theory of Interpersonal Behaviour to investigate the determinants of physicians’ intention to use telemedicine. This model has the advantage of considering cultural variations in the formation of behaviours (Facione, 1993). Physicians represent a particular group of professionals and items measuring theoretical constructs must be adapted to their reality. This was done by applying a qualitative emic-etic approach (Davidson et al., 1976) to the development of the research instrument. Moreover, the structural modelling approach has permitted to assess the validity of the TIB for predicting telemedicine acceptance by physicians. The measurement model was also satisfactory tested by a confirmatory factor analysis. Thus, this research responds to calls for additional theory-testing efforts to extend the results from prior studies by proposing a conceptual framework that considers the particular characteristics of the medical profession. Finally, the present study provides avenues for promoting telemedicine acceptance among physicians and thus, for supporting the diffusion of this technology in the health care system.

Conclusion

The rapid advancements in information and communication technologies over the last years have spurred the development of various telemedicine experiments in Canada. However, the diffusion of this technology to the whole healthcare system remains a major challenge. As a professional group, physicians have an important influence on the integration of telemedicine applications in different clinical settings. In the past, models such as the TPB and the TAM have been applied with limited support to the study of telemedicine acceptance by physicians. The TIB appears as a more comprehensive model since it integrates many psychosocial dimensions involved in the formation of individuals’ behavioural intention.

From a practical standpoint, this study has indicated some avenues for the diffusion of telemedicine in the healthcare system. Thus, communicating positive opinions towards telemedicine from groups such as colleagues and patients, demonstrating the relevance of using this technology in a variety of clinical contexts, and addressing the benefits of using telemedicine for improving patient care could be used as strategies to promote physicians’ acceptance of telemedicine.

From a theoretical standpoint, the findings of this study call for the development of alternatives to measure normative factors that influence physicians’ decision to use telemedicine. Theory refinement is still needed and the integration of constructs from different models represents a promising approach. Also, qualitative research could be conducted to explore more extensively the formation of physicians’ cognitions with respect to telemedicine acceptance. Finally, further studies should compare the determinants of physicians’ acceptance of telemedicine for different clinical or educational purposes and investigate the potential variations among various cultural settings in order to gain a broader understanding of the conditions under which this technology could be implemented on a large scale.

Acknowledgements

The study on which this paper is based was substantially supported by a grant form the Canadian Institutes of Health Research (Project No. 49452). The realisation of this research was also made possible with the support of a doctoral scholarship from the FCAR/FRSQ to Marie-Pierre Gagnon.

References

Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human Decision Processes , 50, 179-211.

Baron, R.M. & Kenny, D.A. (1986). The moderator-mediator variable distinction in social psychological research : Conceptual, strategic and statistical considerations. Journal of Personality and Social Psychology , 51, 1173-1182.

Bergeron, F., Raymond, L., Rivard, S. & Gara, S. (1995). Determinants of EIS use: Testing a behavioral model. Decision Support Systems , 14 (2), 131-146.

Boots, R.J. & Treloar, C. (2000). Prediction of intern attendance at a seminar-based training program : A behavioural intention model. Medical Education , 34, 512-518.

Byrne, B.M. (1994). Structural equation modelling with EQS and EQS/Windows : Basic concepts, applications and programming. Thousand Oaks : Sage Publications.

Cohen, J. & Cohen, P. (1983). Applied multiple regression/correlation analysis for the behavioral sciences, 2nd Edition . Hillsdale : Erlbaum.

Courville, T. & Thompson, B. (2001). Use of structure coefficients in published multiple regression articles : (Beta) is not enough. Educational Psychology Measure , 61, 229-249.

Croteau, A-M. & Vieru, D. (2002). Telemedicine adoption by different groups of physicians. 35th Hawaii International Conference on System Sciences , Maui : IEEE Computer Society.

Davidson, A. R., Jaccard, J. J., Triandis, H. C., Morales, M. L., & Diaz-Guerrero, R. (1976). Cross-cultural model testing toward a solution of the etic-emic dilemna. International Journal of Psychology, 11 , 1-13.

Davis, F.D. (1989). Perceived usefulness, perceived ease of use, and user acceptance of information technology. MIS Quarterly , 13 (2), 319-339.

Deegan, J. (1978). On the occurence of standardized regression coefficients greater than one. Educational Psychology Measure , 38, 873-888.

Donaldson, G.W., Moinpour, C.M., Bush, N.E., Chapko, M., Jocom, J., Siadak, M., Nielsen-Stoeck, M., Bradshaw, J.M., Bichindaritz, I., Sullivan, K.M. (1999). Physician participation in research surveys : A randomized study of inducements to return mailed research questionnaires. Evaluation & the Health Professions , 22, 427-441.

Dwyer, J.J., Allison, K.R. & Makin, S. (1998). Internal structure of a measure of self-efficacy in physical activity among high school students. Social Science & Medicine , 46, 1175-1182.

Facione, N.C. (1993). The Triandis model for the study of health and illness behavior : A social behavior theory with sensitivity to diversity. Advances in Nursing Science , 15 (3), 49-58.

Fishbein, M. (1967). Attitude and the prediction of behavior. In M. Fishbein (Ed.) : Readings in attitude theory and measurement . New York : John Wiley. 477-492.

Gagné, C. & Godin, G. (2000). The theory of planned behavior : Some measurement issues concerning belief-based variables. Journal of Applied Social Psychology , 30, 2173-2193.

Godin, G. & Kok, G. (1996). The theory of planned behavior : A review of its application to health-related behaviors. American Journal of Health Promotion , 11, 87-98.

Hays, R.D. (1989). Robustness of a model of exercise. Journal of Behavioral Medicine, 12, 599-603.

Hu, L. & Bentler, P.M. (1995). Evaluating model fit in structural equation modeling : Concepts, issues, and applications . Thousand Oaks, CA : Sage Publications.

Hu, P.J. & Chau, P.Y.K. (1999). Physician acceptance of telemedicine technology : An empirical investigation. Topics in Health Information Management , 19 (4), 20-35.

Hu, P.J., Chau, P.Y.K., Sheng, O.L. & Tam, K.Y. (1999). Examining the technology acceptance model using physician acceptance of telemedicine technology. Journal of Management Information Systems , 16 (2), 91-112.

Hu, P.J., Chau, P.Y.K. & Sheng, O.L. (2000). Investigation of factors affecting healthcare organization‘s adoption of telemedicine technology. 33rd Hawaii International Conference on System Sciences , Maui : IEEE Computer Society.

Kline, R.B. (1998). Principles and practice of structural equation modeling. New York : Guilford Press.

Maassen, G.H., & Bakker, A.B. (2001). Suppressor variables in path models : Definitions and interpretations. Sociological Methods and Research , 30, 241-270.

Paré, G. & Elam, J. (1995). Discretionary use of personal computers by knowledge workers : Testing of a social psychology theoretical model. Behavior & Information Technology , 14, 215-218.

Reid, J. (1996). A telemedicine primer: Understanding the issues . Billings, Montana : Artcraft Printers.

Succi, M.J. & Walter, Z.D. (1999). Theory of user acceptance of information technologies : An examination of health care professionals. 32nd Hawaii International Conference on System Sciences , Maui: IEEE Computer Society.

Tanriverdi, H. & Venkatraman, N. (1999). Creation of Professional Networks : An emergent model using telemedicine as a case. 32nd Hawaii International Conference on System Sciences , Maui : IEEE Computer Society.

Thompson, R.L., Higgins, C.A & Howell, J.M. (1991). Personal computing : Towards a conceptual model of utilization. MIS Quarterly , 15 (1), 125-142.

Triandis, H.C. (1980). Values, attitudes and interpersonal behavior. In M.M. Page (Ed.) : Nebraska Symposium on Motivation, 1979 : Beliefs, attitudes and values . Lincoln : University of Nebraska Press.

Triandis, H.C. (1989). The self and social behavior in differing cultural contexts. Psychological Review , 96, 506-520.

Ybarra, O. & Trafimow, D. (1998). How priming the private self or the collective self affects the relative weights of attitudes and subjective norms. Personality and Social Psychology Bulletin , 24, 362-370.