INTRODUCTION

Nanotechnology is one of the essential technologies of the 21st century [1]. It involves the use of nanomaterials, which are defined as ‘natural, incidental, or manufactured materials containing particles, in an unbound state or as an aggregate or as an agglomerate and where, for 50% or more of the particles in the number size distribution, one or more external dimensions is in the size range 1 nm–100 nm’ [2]. As a result of the unique properties of nanoparticles, nanotechnology has become a promising field that has improved many aspects of human life. However, nanoparticles may also exhibit toxic effects and this raises concerns about possible health and environmental risks [3]. A significant body of research has focused on the unique properties of nanoparticles, their toxicological aspects [4, 5] and the development of reliable tools for assessment of nanotoxicity [6, 7]. By contrast, relatively little research has been carried out regarding the opinions of stakeholders and the general public on nanotechnology and the intention to use innovative materials.

Studies from Europe and the United States have demonstrated that the general public is rather unfamiliar with the topic of nanomaterials [8-11] and that their attitudes toward nanotechnology are either positive or neutral [8, 11, 12]. Moreover, there is evidence indicating that risk perceptions related to nanotechnology are higher among laypersons than among nanotechnology experts, policy makers, and risk managers [9, 10, 13-15]. However, possible environmental pollution and long-term health problems associated with nanotechnology, as well as use of nanomaterials in food, cosmetics, and packaging, have raised higher concerns among scientists than among non-experts [14, 15]. Interestingly, a recent study revealed that nano-scientists and engineers perceive lower risk than the experts involved in risk regulation and management [16]. Considering that nanotechnology is a relatively new and continuously developing field, the opinions of stakeholders and the general public have not been completely established and thus might change in pace with accumulation of new knowledge [14].

Dentistry is among the fields that have been significantly improved by nanotechnology [17]. The current market offers a variety of dental materials modified by nanoparticles, such as restorative composites, glass ionomer cements, adhesives, and bone-regenerative materials, to name but a few [18-20]. Recently, it has been demonstrated that dentists and dental hygienists have moderate knowledge about nanomaterials and perceive both risks and benefits related to their application [21]. Although several studies have reported on public and expert opinion about nanotechnology, few studies have investigated the attitudes of dental health-care professionals toward this technology [8-16]. Thus, our understanding of the reasons why dental health-care workers use or refrain from use of nanomaterials in the context of clinical dental care is incomplete. Investigation of the attitudes of dental health-care workers towards nanomaterials is essential because it plays an important role in their acceptance or rejection of nanotechnology [22, 23]. To assist policy makers in their management practice, we need to identify the psychosocial factors that influence the decision of dental health-care workers on whether or not to use nanomaterials when treating patients in the future.

The theory of planned behavior (TPB) is a well-recognized theoretical framework of the attitude–behavior relationship, which assumes that most conscious behaviors are goal directed [24]. This theory is an extension of the theory of reasoned action (TRA) and has been applied across various populations, contexts, and behavioral domains [25-31]. In addition to the TRA constructs, the TPB includes perceived behavioral control, therefore allowing a better explanation of behaviors which are beyond full volitional control and improved predictive power of the model [24, 32]. Moreover,TPB has proved to be a reliable tool in predicting and explaining occupational behaviors [26, 30, 31, 33–36]. A systematic review revealed consistency of predicted behavior between health-care professionals and non–health-care professionals, indicating that TPB is a valid tool for use in the occupational context of health care [26]. Meta-analyses have shown that the TPB explains (on average) 39%–59% of the variance in behavioral intention, whereas intention explains (on average) 19%–35% of the variance in actual behavior [30, 37, 38].

According to the TPB, behavior is predicted by behavioral intention (summarizing a person's motivation to engage in a particular behavior and indicating how hard the person is willing to try and how much time and effort he or she is willing to devote in order to perform the behavior) and perceived behavioral control (perception of presence or absence of necessary resources and opportunities as well as anticipated obstacles or impediments related to performing the behavior). Intention, in turn, is a joint function of perceived behavioral control, attitudes toward performing the behavior (positive or negative evaluation of the behavior), and subjective norms (perceived social pressure of performing or not performing the behavior). The TPB maintains that the relative importance of the TPB constructs differs according to the particular behavior and populations investigated [32].

As proposed by Ajzen [32], the original TPB model can be augmented by external variables, such as demographics, moral norms, descriptive norms, and anticipated regret, in accordance with the context and nature of the particular behavior investigated [25, 30, 39]. A number of studies have reported on residual effects of past behavior on intention and future behavior after having controlled for the original TPB constructs, suggesting that these effects reflect the sufficiency of the TPB model [40, 41]. Only a few studies have considered the occupational behavior of dental health-care professionals using a socio-cognitive approach [33, 34, 36, 42, 43].

Whereas knowledge was demonstrated to be an important covariate of the risk perceptions of dental health-care workers related to use of nanomaterials[21], a socio-cognitive model to explain variance in intention to use these materials has yet to be validated among dentists and dental hygienists employed in the public dental health-care service in Norway. As dental health-care workers have been using dental nanomaterials for patient treatment, it seems relevant to investigate whether past behavior predicts the intention to use nanomaterials beyond the effect of the original TPB constructs. In addition, risk perceptions related to nanomaterials might influence behavioral intention, as demonstrated by previous studies [27-29]. Relying on the TPB augmented with past behavior and risk perception, the purpose of this study was to predict the intention of dental health-care workers to use nanomaterials in the future and to explore whether the augmented TPB model operates equivalently across the professional groups of dentists and dental hygienists. In accordance with TPB, it was hypothesized that positive attitudes toward the use of dental nanomaterials, stronger confidence in the ability to use these materials (perceived behavioral control), and higher pressure from significant others (subjective norms) increase the intention to use dental nanomaterials. Furthermore, it was suggested that external variables, in terms of risk perception and previous experience with nanomaterials (past behavior), have both direct and indirect effects on behavioral intention, through attitudes, subjective norms, and perceived behavioral control. The hypothesized model for the present study is depicted in Figure 1.

The hypothesized augmented theory of planned behavior (TPB) model including four latent variables (intention, attitudes, perceived behavioral control [PBC], and subjective norms [SN]) and two observed variables (risk perception and past behavior)

MATERIAL AND METHODS

A census of all dentists and dental hygienists working in the public dental health-care service in Norway (1792 eligible participants) was asked to participate in a cross-sectional self-administered survey in March–May 2017. The questionnaire was developed based on recommendations for TPB questionnaires and relevant literature [44] and was pilot-tested in a public dental clinic in Bergen. The Norwegian Centre for Research Data approved the survey (51053/3/AMS) and was responsible for administration of the questionnaire, data collection, and anonymization of personal information about participants. The questionnaire, together with the informed consent and a short introductory description of nanomaterials (Appendix S1), was distributed by e-post. The main invitation to the survey was supplemented by three consequent reminders in an attempt to increase the response rate.

The questionnaire included the original constructs of the TPB: intention, attitudes, perceived behavioral control, and subjective norms. Each of the TPB constructs was measured by several items, with responses recorded on a seven-point Likert scale that ranged from ‘1 = strongly agree’ to ‘7 = strongly disagree’ (except for item 18 that ranged from ‘1 = very easy’ to ‘5 = very difficult’). The scales of items 7, 10, and 12 were reversed as they represented negative statements (Table 1). Low scores indicated positive cognitions, and high scores indicated negative cognitions. Intention was measured by four items, three of which assessed positive intention, while the fourth asked about the likelihood of using nanomaterials in the future. Attitudes were measured by nine items: six were positively worded and three were negatively worded. Perceived behavioral control and subjective norms were measured by five and four positively phrased items, respectively. In addition, two variables, external to the TPB model, were incorporated: (1) risk perceptions of dental nanomaterials, and (2) past behavior. Risk perception was a summative score of six items, each assessed on a seven-point Likert scale that ranged from ‘1 = very likely’ to ‘7 = very unlikely’, for which low scores represent high perception of risk and high scores represent low perception of risk (Table 1). Past behavior was measured by one item ‘Have you used dental nanomaterials for patient treatment before?’ with response alternatives ‘1 = yes’, ‘2 = no’, and ‘3 = I don't know’, which were further dichotomized into ‘0 = yes’ and ‘1 = no/I don't know’ for the purpose of analysis. In accordance with recommendations, the TPB constructs in the augmented model were measured considering the four elements of action (using), target (nanomaterials), context (for patient treatment), and time (in the future) [24].

TABLE 1. Descriptive statistics for the theory of planned behavior (TPB) measurement model Latent factor Itemno. Na Question Scale Mean SD Intention; α = 0.93 1 712 I intend to use dental nanomaterials for patient treatment in the future b 3.2 1.3 2 715 I plan to use -//- b 3.2 1.4 3 712 I have decided to use -//- b 3.5 1.3 4 718 How likely is that you will use -//- c 2.8 1.3 Attitudes; α = 0.93 5 754 To use nanomaterials for dental treatment in the future is a good idea b 3.4 1.2 6 751 -//- is important b 3.4 1.2 7e 749 -//- is dangerous b 3.9 1.0 8 734 -//- is responsible b 3.5 1.1 9 729 -//- is reasonable considering the quality of treatment b 3.2 1.1 10e 735 -//- is irresponsible considering the patient's health b 3.8 1.1 11 705 -//- is valuable b 3.3 1.1 12e 709 -//- is useless b 3.2 1.1 13 713 -//- is interesting b 2.9 1.3 Perceived behavioral control; α = 0.80 14 668 If I want, I have the possibility to use dental nanomaterials for patient treatment in the future b 3.0 1.3 15 673 It is totally up to me if I use -//- b 3.9 1.5 16 673 I have all the resources I need to use -//- b 3.7 1.4 17 669 I am sure that I am able to use -//- b 3.2 1.3 18 672 How easy or difficult you think it is to use -//- d 2.7 0.7 Subjective norms; α = 0.87 19 661 Colleagues who influence my clinical practice think that I should use dental nanomaterials for patient treatment in the future b 3.9 1.2 20 661 Colleagues who are important to me think that I should use -//- b 3.8 1.2 21 655 The chief dentist of my clinic thinks that I should use -//- b 3.9 1.1 22 659 The chief dentist of the county thinks that I should use -//- b 3.8 1.0 Risk perceptionf; α = 0.89 23 660 How likely is that you subject yourself to health damage by using dental nanomaterials in the future c 3.9 1.1 24 657 How likely is that you increase your own risk to get cancer by using -//- c 4.0 1.1 25 658 How likely is that you inhale nanoparticles that accumulate in your body if you use -//- c 3.7 1.2 26 647 How likely is that you contribute to the uncontrolled spreading of nanoparticles if you use -//- c 3.6 1.2 27 649 How likely is that you contribute to patient's health damage if you use -//- c 4.1 1.1 28 646 How likely is that you contribute to environmental pollution if you use -//- c 3.4 1.3 aNumber of participants does not add up to 851 in the questions because of missing values (11%–24% in separate items). b7-point Likert scale ranging from 1 (strongly agree) to 7 (strongly disagree). c7-point Likert scale ranging from 1 (very likely) to 7 (very unlikely). d5-point Likert scale ranging from 1 (very easy) to 5 (very difficult). eScale of items 7, 10, and 12 was reversed as they represent negative statements. fRisk perception is a summative score (range 6–42), incorporated as an observed variable in the structural equation model. Statistical analysis

Descriptive statistical analysis, in terms of frequencies and mean distributions, was conducted using SPSS, version 25.0 (IBM). Structural equation modelling was performed using the Lavaan package [45] in R (R Core Team). Structural equation modelling is an advanced statistical technique that enables us to investigate whether the hypothesized augmented TPB model has acceptable fit to the data, testing simultaneously the interrelationships between the constructs specified in the hypothesized model [46].

In the present study, a two-stage modelling approach was used to test the hypothesized augmented TPB model [47]. First, confirmatory factor analysis (CFA) was performed to test the factorial validity of the latent constructs and the adequacy of the measurement model. In the first stage, four latent constructs comprising the original TPB model were used (intention, attitudes, perceived behavioral control, and subjective norms), excluding risk perception and past behavior as they were used as observed variables in the model. Potential sources of misfit were examined with the help of modification indices, which provided a basis for the re-specification of the measurement model.

Second, following the specification of the measurement model, structural equation modelling was performed to examine whether the hypothesized augmented TPB model has acceptable fit to the data and to estimate direct, indirect, and total effects of relationships in the model. The following statistical parameters were used to measure how well the hypothesized model fit the data – chi-square (χ2) test, comparative fit index (CFI), root mean square error of approximation (RMSEA), and standardized root mean square residual (SRMR) [48]. A statistically non-significant chi-square test result (i.e., P > 0.05) indicates good fit of the model. However, because this test is highly sample-size sensitive (large samples can lead to a significant P-value of the chi-square test, even with trivial misspecifications), the emphasis was set on the remaining fit indices. In line with conventional recommendations of Hu and Bentler [49], values of CFI > 0.90 and > 0.95, of RMSEA < 0.08 and < 0.06, and of SRMR < 0.08 and < 0.05 indicate acceptable fit and good fit, respectively. The maximum likelihood estimator with robust standard errors was applied to account for non-normally distributed data. Missing data were handled by the full information maximum likelihood, which is most often superior to handling missing data by use of standard ad hoc routines, such as mean replacement and listwise or pairwise deletion [50].

Multigroup analyses were performed with CFA and structural equation modelling to test whether the model was invariant across the two groups of employees. Before investigating the invariance of predictive paths (using structural equation modelling), the configural and metric invariance was assessed in the final measurement model (using CFA). The configural invariance (equal forms) was tested by fitting the final measurement model across dentist and dental hygienists. Configural invariance was supported if the model had a satisfactory fit (based on the above-mentioned fit indices). Metric invariance (equal factor loadings) was tested by constraining factor loadings in both groups and by comparing the constrained model with the baseline model (configural invariance model) in which factor loadings were free to vary. Metric invariance was supported if the chi-square change was non-significant and the CFI change was less than 0.002 [51]. Invariance of predictive paths was tested by comparing a structural equation model in which both factor loadings and regression paths were constrained across the groups with a baseline structural equation model in which factor loadings were constrained and regression paths were free to vary. The criteria for invariance of predictive paths were insignificant chi-square change and CFI change less than 0.002.

RESULTS

A total of 851 participants responded to our survey (response rate 47.5%). Descriptive statistics of all variables measuring the TPB constructs and risk perceptions are presented in Table 1. As reflected by mean values of item score measuring different constructs, participants exhibited the following: moderate-to-strong intention to use nanomaterials; somewhat positive attitudes; slightly positive perceived behavioral control and subjective norms; and moderate risk perceptions. Cronbach's alpha values ranged from 0.80 for perceived behavioral control to 0.93 for intention and attitudes, indicating high internal consistency.

Table 2 depicts sociodemographic characteristics stratified according to professional status. In line with the gender and professional distribution in the census of Norwegian dental health-care workers in the public dental healthcare service, 18.6% were male and 71.0% were dentists. The mean ± SD age of the participants was 41.5 ± 11.9 years. Of all respondents, 54.0% (63.7% dentists and 28.7% dental hygienists) confirmed that they had previously used dental nanomaterials.

TABLE 2. Sociodemographic factors stratified according to professional status in the total sample Factor Dentist n = 570 % (n) Dental hygienist n = 228 % (n) Total n = 798a, % (n) Gender** Male 25.6 (139) 1.4 (3) 18.6 (142) Female 74.4 (404) 98.6 (218) 81.4 (622) Work experience* ≤ 5 years 28.2 (161) 19.3 (44) 25.7 (205) 6–20 years 44.7 (255) 43.4 (99) 44.4 (354) > 20 years 27.0 (154) 37.3 (85) 29.9 (239) Place of education** Norwegian institution 68.7 (389) 96.5 (220) 76.7 (609) Foreign institution 31.3 (177) 3.5 (8) 23.3 (185) County regionns South-East 40.9 (233) 42.7 (97) 41.4 (330) West 30.2 (172) 24.7 (56) 28.6 (228) Middle-North 28.9 (165) 32.6 (74) 30.0 (239) Past behavior** Yes 63.7 (311) 28.7 (54) 54 (365) No/I don't know 36.3 (177) 71.3 (134) 46 (311) aNumber of participants is not 851 in each question because of missing values. Testing the association between factor and professional status: ns, not significant; *P < 0.05; **P < 0.001. Measurement model

Standardized factor loadings of all items were significant (P < 0.001) and ranged from 0.385 to 0.948 (results not shown). Standardized correlation coefficients ranged from 0.444 to 0.782 and were below the cut-off point of 0.85 (results not shown), indicating satisfactory discriminant validity of the latent constructs in the model [52].

The hypothesized correlated four-factor model approached acceptable fit, as indicated by fit indices (Table 3, Model 1). According to modification indices, the model fit could be improved by allowing correlation between residuals of items in the attitude construct (item 5 with item 6, item 7 with item 10) and in the subjective norms construct (item 21 with item 22) (Table 1). These residual correlations made theoretical sense and were therefore added to the model, one by one (Model 2 – Model 4). The final measurement model thus achieved a good fit (Table 3, Model 4).

TABLE 3. Overall goodness-of-fit indices for the theory of planned behavior (TPB) measurement models (Models 1–4) and full structural model (Model 5) Fit indices Model 1 Model 2 Model 3 Model 4 Model 5 χ2 782.3 680.6 612.8 555.9 665.5 df 203, P < 0.001 202, P < 0.001 201, P < 0.001 200, P < 0.001 236, P < 0.001