Modelling intentions to provide smoking cessation support among mental health professionals in the Netherlands

Abstract

Background: Tobacco use prevalence is elevated among people with mental illnesses, leading to elevated rates of premature smoking-related mortality. Opportunities to encourage smoking cessation among them are currently underused by mental health professionals. In this paper, we aim to explore mechanisms to invigorate professionals' intentions to help patients stop smoking.

Methods: Data stem from a recent staff survey on the provision of smoking cessation support to patients with mental illnesses in the Netherlands. Items and underlying constructs were based on the theory of planned behaviour and literature on habitual behaviour. Data were weighted and only data from staff members with regular patient contact (n = 506) were included. Descriptive statistics of the survey items are presented and in a second step using structural equation modelling (SEM), we regressed the latent variables attitudes, subjective norms (SN), perceived behavioural control (PBC), past cessation support behaviour (PB) and current smoking behaviour on intentions to provide support. In optimisation steps, models comprising a subset of this initial model were evaluated.

Results: A sample of 506 mental health workers who had direct contact with patients completed the survey. The majority of them were females (70.0 %), respondents had an average age of 42.5 years (SD = 12.0). Seventy-five percent had at least a BSc educational background. Of the respondents, 76 % indicated that patients should be encouraged more to quit smoking. Respondents were supportive to train their direct colleagues to provide cessation support more often (71 %) and also supported the involvement of mental health care facilities in providing cessation support to patients (69 %). The majority of the respondents feels capable to provide cessation support (66 %). Two thirds of the respondents wants to provide support, however only a minority (35 %) intends to actually do so during the coming year. Next, using SEM an acceptable fit was found of the constructs derived from the theory of planned behaviour and literature on habitual behaviour to the weighted data (χ (2) (322) = 1188, p < .001; RMSEA = 0.067; CFI = 0.983), after removal of insignificant latent variables (SN and current smoking) and inclusion of covariates. Attitudes, PBC and PB of staff are the strongest identified correlates of intention toward providing cessation support to patients. SN and staff smoking behaviour were found to be weaker, non-significant correlates.

Conclusions: To nudge staff towards providing cessation support to people with mental illnesses one should aim at influencing attitudes and perceived behavioural control.

Keywords: Psychiatry; Structural equation modelling; Survey research; Treatment and intervention.

Fig. 1

Measurement items for attitudes, subjective norms, perceived behavioural control and intention

Fig. 2

Initial Model 1, Model 2 with non-significant paths removed, and optimized Model 3. Figure note: SN: subjective norm, PBC: perceived behavioural control; ATT: behavioural attitude; PB: past behaviour (providing cessation support to patients); SMO: current smoking behaviour; INT: intention to provide cessation support to patients in the near future; Model 1: Initial model with all possible relevant constructs included; Model 2: Initial model after insignificant paths and constructs have been removed; Model 3: Final model based on model 2 with covariances specified based on modification indices; The numbers in the paths between two latent constructs are standardized parameter estimates. Covariance matrices for the three models are available as Additional file 2. Analyses are based on unweighted data

Fig. 3

Median model, 95 % CI worst fitting, and 95 % CI best fitting model. Figure note: PBC: perceived behavioural control; ATT: behavioural attitude; PB: past behaviour (providing cessation support to patients); INT: intention to provide cessation support to patients in the near future; Model ‘Median fit’: Final optimized model identified using bootstrapped data (1000 iterations) using the median fitting data (based on χ ²): 50 % of the bootstrapped datasets fitted the model better, 50 % fitted the model worse; Model ‘95 % CI worst fit’: 97.5 % of the bootstrapped datasets fit the model better, 2.5 % fit the model worse (based on χ ²); Model ‘95 % CI best fit’: 2.5 % of the bootstrapped datasets fit the model better, 97.5 % fit the model worse (based on χ ²); The numbers in the paths between two latent constructs represent standardized parameter estimates, Covariance matrices for the three presented fits are available as Additional file 2