Parent Preferences for Pediatric Clinician Messaging to Promote Smoking Cessation Treatment

Study Design, Setting, and Participants

We conducted a cross-sectional study of 180 parent smokers to assess their perceptions regarding the relative importance of distinct messages to promote smoking cessation. Each parent was asked to provide sociodemographic data using a questionnaire as well as complete a discrete choice experiment (DCE) to choose the most meaningful messages for them to start smoking cessation treatment.

Parent smokers were recruited through 4 outpatient primary care practices from the Children’s Hospital of Philadelphia (CHOP) Pediatric Research Consortium, a primary care practice-based research network.²⁴ To ensure generalizability to other clinic settings, we recruited from 4 diverse, high-volume practices, ranging in setting (urban and suburban), with patient populations that varied by insurance status and race and/or ethnicity. Previous work identified 10% to 15% of parents who present with their child for care as smokers.²⁵

Potential study participants were identified through standard screening for SHS exposure embedded within clinical practice, which included parent assessment of smoking status at each visit.²⁶ On days study staff were recruiting at practice sites, all eligible smokers were approached at their child’s office visit to ascertain interest in participating, obtain informed consent, collect all measures, and conduct the DCE. Inclusion criteria included parents or caregivers (hereafter “parents”) being ≥18 years in age, present at their child’s health care visit during the study period, able to communicate in English, and a self-reported smoker. We focused on parents who were cigarette smokers rather than electronic cigarette (e-cigarette) users considering the established clear health harms of cigarette use and SHS exposure, evidence-based treatment options available for cigarette use, and prevalence of use: Philadelphia and the surrounding suburban areas have high rates of adult cigarette smoking (22%)²⁷ compared with fewer than 5% of adults identifying as e-cigarette users.²⁸ Only English speakers were recruited because the practice population is overwhelmingly English speaking, and previous health message framing work was conducted among English speakers.¹³ In total, we recruited 180 parent smokers through the 4 practices from September 2017 to February 2019. Parents provided written informed consent and were financially compensated ($15) for their participation. The study protocol was approved by the Institutional Review Board at CHOP.

Development of Framed Messages

We developed the messages used in the DCE using tobacco control literature; expert feedback from pediatricians, statisticians, and psychologists; and input from parents from the target population.^16,27,29 Draft messages were pretested through one-on-one interviews with parent smokers from the target population and revised according to their feedback, a development approach used in other DCEs^30,31 and message framing studies.³² Messages were modified to ensure they were easy to understand, were meaningful to the reader, and were intelligible to those with a basic health literacy.^33,34 A total of 30 parent smokers from one of the practices reviewed iterations of the messages. The result of the iterative development process and testing was a final list of 26 items. Messages were tailored on the basis of 3 dimensions of interest: whom the message featured (child, parent, or family), whether the message was gain or loss framed, and inclusion of a specific outcome (general health, cancer, respiratory illnesses, child becoming a smoker, or financial impact) (Table 1). Of the 26 items, 10 items targeted the child (5 gain framed, 5 loss framed; 2 items for each of the 5 outcomes); 8 items targeted the parental respondent (4 gain framed and 4 loss framed, with 2 items for each of the 4 outcomes); and 8 remaining items targeted the entire family (with the same breakdown as the items targeting the parent).

DCE

A DCE is a well-established methodology for eliciting preferences that would otherwise be difficult or impossible to get. A DCE consists of a series of tasks in which respondents are forced to choose the responses that they prefer; as a result, the DCE enables a quantification of the relative importance of each attribute. This methodology circumvents the known problems with techniques that require the individual to rank alternatives, such as Likert-style rating scales, presenting a straightforward task that more closely resembles a real-world decision.^35,36 This methodology has been used in other pediatric health care studies to support development of future information-based interventions.^31,32,37 The DCE was administered by using a tablet computer. Parent respondents were shown 4 messages at a time, selecting the message in each subset that was the most and least important (see Fig 1). The DCE was designed in such a way that each message is shown an equal number of times.

FIGURE 1

Sample DCE screen. Parent respondents were presented with 4 messages at a time, selecting the message in each grouping that they deemed “most” or “least” important.

Relative Importance Scores

Both the DCE (data collection) and the quantitative analysis to estimate individual preferences for each message were done by using Lighthouse Studio (version 9.7.0; Sawtooth Software, Provo, UT). The software uses a multinomial logistic regression to calculate the intensity of importance for each attribute and then transforms the scores into probability scaled scores, which allows for comparisons across attributes.³⁸ For each person, there are 26 probability scaled scores (one for each message) that sum to 100. These ratio-scaled scores have an easy interpretation; for example, a message with a score of 6 is twice as important as a message with a score of 3.^31,35

Demographic and Clinical Measures

The child’s (patient) age, sex, race, ethnicity, presence of a chronic health condition (prematurity, asthma, chronic lung disease, allergies, or attention-deficit/hyperactivity disorder), and insurance payer were obtained via parent respondent self-report. The parent’s age, sex, and education level were also obtained via parent self-report. Parents also completed measures of health literacy (via the Newest Vital Sign, a reliable and validated measure³⁹) and nicotine dependence (measured by the Fagerström Test for Nicotine Dependence⁴⁰). Quit motivation was measured before the DCE by using the Contemplation Ladder, an efficient and face-valid measure, generalizable to diverse populations and associated with objective measures of readiness to quit smoking and actual quit attempts.^41–44

Statistical Analysis

Characteristics of parents and patients are presented with proportions for categorical variables and means and SDs for continuous variables. The relative importance scores for each message across all parent smokers are presented in a ranked order list, with 95% confidence intervals (CIs) for each message.

To understand the relative importance of messages within the dimensions of interest, we coded each of the 26 messages by their item properties: who was featured (child, family, or parent), framing (gain or loss), and outcome included (general health, respiratory health, cancer, risk of child becoming an adult smoker, and financial impact). We created a person-item level data set for analysis and analyzed the relationship between message dimensions and scores by using clustered data methodology, adjusting our model for the variability in outcome across respondents. Using the importance scores as the outcome, we ran multivariable mixed-effects linear regression models to measure the association of message dimensions on message importance score. The models included indicator variables for the subcategories within each dimension of interest: family and child (with parent as the reference category) in the dimension “whom the message featured,” gain framing (with loss framing as the reference category) in the dimension “how the message was framed,” and child becoming an adult smoker, general health, cancer, respiratory health (with financial impact as the reference category) in the dimension “outcome included in the message.” The regression parameter for a particular indicator variable within a dimension represents the difference in expected relative importance score for patients in that subcategory versus the reference category when the comparison is made between respondents who are in the same subcategories for the other dimensions. For example, the coefficient for gain framing represents the difference in expected scores between the 2 messages within a row in Table 1 while accounting for the correlation within the multiple scores on each respondent. We ran additional sensitivity analyses to test 2 properties of our mixed-effects modeling: (1) the normality assumption and (2) the dependence of scores when using all 26 scores per person. To test if the normality assumption could have affected our results, we fit a generalized estimating equation model, which does not assume normality, by using an identity correlation structure for the repeated measures per person and robust SEs. We also compared mixed-effects linear regression and generalized estimating equation models using all 26 scores per person (full data) as well as 25 scores per person (partial data) to ensure that the dependence across the 26 scores did not affect our results. All 4 models provided similar quantitative coefficient estimates and P values.

Statistical analyses were conducted in R version 3.6.1 (The R Foundation, Vienna, Austria)⁴⁵ and Stata version 15 (Stata Corp, College Station, TX). For a sample size calculation, previous literature indicates ∼30 to 60 respondents are needed per dimensions for investigational work in a new area.^31,32,46 Because we were investigating messages across 3 dimensions of interest (whom the message featured, how the message is framed, and the outcome included), we therefore chose an a priori sample size of 180 parent smokers as adequate for a reliable estimate of preferences.