What is the effect of a decision aid in potentially vulnerable parents? Insights from the head CT choice randomized trial

Abstract Objective To test the hypotheses that use of the Head CT Choice decision aid would be similarly effective in all parent/patient dyads but parents with high (vs low) numeracy experience a greater increase in knowledge while those with low (vs high) health literacy experience a greater increase in trust. Methods This was a secondary analysis of a cluster randomized trial conducted at seven sites. One hundred seventy‐two clinicians caring for 971 children at intermediate risk for clinically important traumatic brain injuries were randomized to shared decision making facilitated by the DA (n = 493) or to usual care (n = 478). We assessed for subgroup effects based on patient and parent characteristics, including socioeconomic status (health literacy, numeracy and income). We tested for interactions using regression models with indicators for arm assignment and study site. Results The decision aid did not increase knowledge more in parents with high numeracy (P for interaction [P int] = 0.14) or physician trust more in parents with low health literacy (P int = 0.34). The decision aid decreased decisional conflict more in non‐white parents (decisional conflict scale, −8.14, 95% CI: −12.33 to −3.95; P int = 0.05) and increased physician trust more in socioeconomically disadvantaged parents (trust in physician scale, OR: 8.59, 95% CI: 2.35‐14.83; P int = 0.04). Conclusions Use of the Head CT Choice decision aid resulted in less decisional conflict in non‐white parents and greater physician trust in socioeconomically disadvantaged parents. Decision aids may be particularly effective in potentially vulnerable parents.


| INTRODUC TI ON
Shared decision making (SDM) aims to improve health-care quality by involving patients, parents and clinicians in medical decisions. Decision aids (DAs), patient-centred tools that facilitate SDM, have been shown to improve patients' knowledge, decrease decisional conflict and enhance patient engagement in decision making. [1][2][3][4][5] Although SDM is an emerging trend in paediatrics, few interventions to promote SDM in paediatric emergency care have been rigorously studied. 5 Further, there is limited research on the impact of SDM in various population subgroups such as those with low health literacy or numeracy or among individuals of different ethnic/racial groups.
Recently, a multicentre randomized trial evaluated the impact of a DA, Chest Pain Choice, in patients with chest pain at low risk for acute coronary syndrome. In a planned secondary analysis, the DA aided all subgroups to a similar extent, with greater knowledge transfer in patients with high numeracy and greater physician trust in patients with low health literacy. 6 A qualitative study of African American patients with diabetes further explores the relationship between shared decision making and patient trust. 7 In focus groups and in-depth interviews, participants revealed concerns about potential racial bias and whether their physician might withhold medical information from them. At the same time, patients described specific physician shared decision-making behaviours, such as information sharing and discussing the pros and cons of treatment options, as enhancers of patient trust. Guided by these insights, the authors proposed a conceptual model in which both shared decision-making behaviours and race independently influence physician trust ( Figure 1). In this model, the information sharing, deliberation and decision-making domains of shared decision making have potential to enhance patient trust, even in the context of racial and cultural differences between patients and their physicians.
A recently completed multicentre randomized trial evaluating the impact of a DA in parents of children with minor head trauma, 'Head CT Choice', offers a unique opportunity to evaluate potential subgroup effects of a DA in the context of paediatric emergency care. In this trial, 172 clinicians caring for 971 children at intermediate risk for clinically important traumatic brain injuries (ciTBIs) were cluster-randomized to SDM facilitated by a DA or to usual care. Similar to what has been observed in adult SDM trials, 1-3 parents randomized to shared decision making had greater knowledge, less decisional conflict and were more engaged in decision making compared to usual care. 8 Outcomes Research Institute (PCORI) Award ( . The views presented in this publication are solely the responsibility of the author(s) and do not necessarily represent the views of the Patient-Centered Outcomes Research Institute (PCORI), its Board of Governors or Methodology Committee.

Methods:
This was a secondary analysis of a cluster randomized trial conducted at seven sites. One hundred seventy-two clinicians caring for 971 children at intermediate risk for clinically important traumatic brain injuries were randomized to shared decision making facilitated by the DA (n = 493) or to usual care (n = 478). We assessed for subgroup effects based on patient and parent characteristics, including socioeconomic status (health literacy, numeracy and income). We tested for interactions using regression models with indicators for arm assignment and study site.

Results:
The decision aid did not increase knowledge more in parents with high numeracy (P for interaction [P int ] = 0.14) or physician trust more in parents with low health literacy (P int = 0.34). The decision aid decreased decisional conflict more in nonwhite parents (decisional conflict scale, −8.14, 95% CI: −12.33 to −3.95; P int = 0.05) and increased physician trust more in socioeconomically disadvantaged parents (trust in physician scale, OR: 8.59, 95% CI: 2.35-14.83; P int = 0.04). Based on the summary effect estimates of DAs published in two meta-analyses 2,5 an insightful qualitative study exploring the relationship between physician trust and shared decision making, 7 and a recent subgroup analysis of the effects of a DA in adults with chest pain, 6 we hypothesized that use of the Head CT Choice DA would be similarly effective in all parent/patient dyads but would increase knowledge more in parents with high numeracy and increase physician trust more in parents with low health literacy.

| Study design
This was a planned secondary analysis of a SDM trial in parents of children with minor head trauma. The study protocol has been previously published. 9 The trial was conducted at seven clinical sites across

| Intervention clinicians
The Head CT Choice DA was developed in Rochester, MN, USA, through a participatory action research methodology that involved eliciting input from a multidisciplinary investigative team including clinicians, health services researchers, a graphic designer, a radiation physicist and parent stakeholders. An initial DA prototype was designed based on input from the investigative team and subsequently refined based on feedback received from parents and clinicians after use in clinical encounters. Full details of the DA development process have been described previously. 9 The DA can be accessed at https ://share ddeci sions.mayoc linic.org/decis ion-aid-infor matio n/head-ct-choice-decis ion-aid/.
After enrolment, study coordinators calculated the patient's precise PECARN risk estimate of ciTBI (calculated based on the presence or absence of individual PECARN clinical predictors in isolation, as well as combinations of predictors 10 ) and provided intervention clinicians a DA corresponding to the individual patient's level of risk.
Research staff offered intervention clinicians a brief, just-in-time refresher of DA content and use just prior to the clinical encounter.
Clinicians then brought the DA to the bedside and used it during the clinical encounter to facilitate a SDM discussion with the parents.

| Usual care clinicians
For patients whose clinicians were randomized to usual care, research assistants instructed the clinicians to discuss management options with parents according to each clinician's usual fashion.
Clinicians in the usual care arm were blinded to the precise risk estimates for ciTBI calculated from the PECARN head injury public access database and did not have access to the DA. The usual care arm was not standardized.

| Data collection
Data documenting the process of screening and enrolment were collected in compliance with CONSORT guidelines. 11 Patient characteristics collected included the sex and age of the child as well as the number and type of PECARN risk factors for ciTBI. Parent characteristics collected included their race, highest level of education, health literacy, numeracy and annual household income. We also recorded the number of parents present during the encounter (father only, mother only and/or both parents). Outcome data analysed in this report were obtained from video and audio recordings of the parent-clinician encounter, a pre-encounter parent survey, a post-encounter parent survey, a post-encounter clinician survey, review of the electronic medical record and telephone follow-up initiated 7 days after the ED visit. The parent and clinician surveys have been previously published. 9 On the pre-encounter survey, parent literacy was assessed using the subjective literacy scale 12,13 and numeracy was assessed using the subjective numeracy scale. 14 The subjective literacy scale, which ranges from 3 to 15, consists of three items (each with a five-point Likert response) that were summed to a total score after reverse coding one item, with higher scores indicating higher health literacy. The subjective numeracy scale, which quantifies an individual's ability to understand and use numbers in daily life, consists of an 8-item survey. Responses to all 8 questions were reversed and averaged, creating an overall score ranging from 6 to 48, with higher scores indicating higher numeracy.
The post-encounter parent survey collected data assessing parents' knowledge regarding their child's risk for ciTBI and the available management options.

| Outcomes
The primary outcome, which was selected by parent stakeholders, was parent knowledge regarding their child's risk for ciTBI and the available diagnostic options. Parent knowledge was assessed by immediate post-visit survey. Secondary outcomes, also obtained by post-visit survey, included the degree of uncertainty parents experienced related to choosing between diagnostic options with which they were unfamiliar using the validated Decisional Conflict Scale (DCS) 15 and parents' trust in their clinician measured using the validated Trust in Physician Scale (TPS). 16 Trained research assistants viewed encounter video recordings to assess the degree to which clinicians engaged parents in the decision-making process using the 12-item 'observing patient involvement' (OPTION) scale. 17 To measure health-care utilization, data were collected on the proportion of children who underwent cranial CT scanning during the ED visit, the most immediate utilization decision. Finally, the safety of DA use was assessed by comparing the rate of ciTBI in each arm of the trial.

| Subgroups
We dichotomized parent and patient characteristics to assess the differential effect of the DA. The following patient characteristics were dichotomized: the sex and age in years (<2 years and 2-18 years, as there are two different PECARN prediction rules based on this age cut-off) and the number of PECARN risk factors (1 vs 2). The following parent characteristics were also dichotomized: race, highest level of education, health literacy, numeracy, annual household income and the number of parents present during the encounter.
To explore the differential effectiveness of the DA in potentially vulnerable parents while simultaneously limiting the risk of bias associated with multiple testing, we created a combination variable to identify a socioeconomically disadvantaged parent subgroup. If the parent was of non-white race, low health literacy or numeracy, and low income (<$40 000), they were classified as socioeconomically disadvantaged. We dichotomized the data for two reasons: (a) to avoid subgroups that were too small to analyse and (b) to simplify the analysis and interpretation of subgroup effects by way of interactions. 18 Classifications for each variable were selected based on the distribution of the data, which we report in full, and conceptual considerations regarding the mostly likely contrasts to show heterogeneity of effect. Patients/parents missing a subgroup variable were excluded from the analysis for that subgroup. For race, the 'Other' category was included with 'Non-White'. For education, the 'Other' category was excluded from the dichotomous groups, as we did not assume that 'other' indicated either of the two categories. For the parent(s) present during the encounter, the 'other' category was excluded from the analysis for that subgroup (n = 8), where 'other' could indicate a family member, caregiver or friend.

| Statistical analysis
Sample size and power estimates for the primary and secondary outcomes of the trial have been published elsewhere. 9 Briefly, we estimated that enrolling 950 patients would provide 99% power to detect a 16% difference in parent knowledge between the DA and usual care arms. This percentage difference was selected a priori, as it had been observed in a prior pilot trial conducted in the ED setting. 3 As with the primary analysis of the trial, 8 outcome assessments for this analysis were measured using regression models (linear for continuous outcomes, multinomial for categorical outcomes) that included indicators for arm assignment and study site. To assess the effect of the DA in each subgroup, we included an interaction term for group assignment. To improve interpretation, we also replicated the primary trial analysis (ie, without an interaction term) within each subgroup and report whether the group effect differed significantly from zero. This group effect was reported as a coefficient for continuous outcomes and as odds ratios for dichotomous or multinomial outcomes.
The analytic approach was informed by publication guidelines for reporting subgroup analyses. 18 Interaction testing between parent/patient characteristics and the outcomes of parent knowledge, decisional conflict, parent engagement in decision making and physician trust were pre-specified, and a significance level of 5% was used to identify significant interactions for these subgroup effects. Significant interactions identified in subgroup analyses that were not pre-specified were considered hypothesis generating. All analyses were performed using Stata 14.1 (2016.

Stata Corporation).
The funder of the study had no role in study design, data collection, data analysis, data interpretation or writing of the report. All researchers' maintained independence from the funder of the study.  similarly across all patient and parent subgroups ( Table 2). There was a significant interaction between patient race and the outcome of decisional conflict. The DA decreased decisional conflict to a relatively greater degree in non-white parents compared to white parents (Table 2; Figure 2). There was also a significant interaction between whether the parent was socioeconomically disadvantaged and physician trust. The degree to which trust improved with use of the DA was greater in socioeconomically disadvantaged parents (Table 2; Figure 3). Interestingly, there was an inverse correlation between decisional conflict and physician trust (correlation = −0.40, P < .001). There were no significant interactions between parent/patient characteristics and the outcomes of knowledge or parent engagement in decision making (OPTION score).

| RE SULTS
Regarding diagnostic decisions, we identified significant interactions between whether a cranial CT was obtained and parent literacy, age of the child, and father only present during the encounter ( Table 3). Children of parents with low health literacy had a lower odds of having a cranial CT performed compared to parents with high health literacy. On further analysis, 58 (24%) of nonwhites and 84 (12%) of whites had low health literacy (P < .001).
If the child was less than 2 years of age, there was a higher odds of cranial CT compared to children 2-18 years of age. Conversely, there was also a lower odds of cranial CT among encounters in which only the father was present compared to encounters with both parents and only the mother present ( Figure 4).   20 A telephone survey of adults who had a recent primary care visit also supports these findings. In this study, African

| D ISCUSS I ON
American patients rated their visits as less participatory compared to whites. However, patients with race-concordant relationships with their physicians rated their physicians as significantly more participatory. 21 Engaging parents with higher levels of decisional conflict at baseline may result in a relatively greater improvement when engaged in SDM.
Why did use of the DA generate greater physician trust in socioeconomically disadvantaged parents? In our prior shared decision-making trial in ED patients with chest pain, we observed greater physician trust in patients with low health literacy. 6 Although the current trial was conducted in parents of children with minor head trauma, the The odds of CT imaging was also higher in children younger than 2 years of age. However, there were no differences in knowledge, decisional conflict, physician trust or OPTION scores among parents of F I G U R E 3 Trust in Physician Scale subgroup effects. Forest plot demonstrating the effect of the Head CT Choice decision aid on parent trust in the physician in subgroups according to patient and parent characteristics children younger than two or older than 2 years of age. This suggests that the higher rates of CT imaging in the younger age group may be related to greater parental anxiety and/or clinician uncertainty when caring for preverbal children who are unable to clearly express their symptoms.

| Limitations and strengths of the study
The primary limitations of this study relate to issues of multiple testing and imprecision (lack of power) around estimates of subgroup effects. Given that a total of 55 comparisons were performed, one would expect 2-3 (5% of 55) tests to be significant at the 5% level due to chance alone. To mitigate the risk of spurious findings related to multiple testing, we pre-specified hypotheses based on prior observations in SDM trials. 3,5,6 We also followed guideline recommendations for reporting subgroup analyses in clinical trials 18

| Implications for practice and future research
What are the implications of the findings of this subgroup analysis?
It is possible that use of the Head CT Choice DA increased clinicians' efforts to share information and engage parents in deliberations regarding whether imaging should be obtained for their child, and that this change in clinician behaviour mitigated parent distrust related to socioeconomic and racial disparities. For practicing clinicians, it is important to note that efforts to engage parents in SDM have potential to increase trust, particularly in socioeconomically disadvantaged parents, and that this trust can positively affect the therapeutic relationship. For researchers, the observations of lower decisional conflict in non-white parents and greater physician trust in socioeconomically disadvantaged parents are exploratory.
Additional research is needed to replicate these findings and to gain greater insight into how the SDM process alters the experiences and perceptions of care in socioeconomically disadvantaged populations.

| CON CLUS IONS
Use of the Head CT Choice DA was similarly effective in all parent/ patient dyads but did not increase knowledge more in parents with high numeracy or physician trust more in parents with low health literacy. In exploratory analyses, we found that decisional conflict was lower in non-white parents, physician trust was greater in socioeconomically disadvantaged parents, and CT rates were lower in children whose parents had low health literacy. DAs may be particularly effective in potentially vulnerable parents.

ACK N OWLED G EM ENTS
We thank the patients and clinicians who participated in our trial and all other contributors.

CO N FLI C T S O F I NTE R E S T S
None of the authors have conflicts of interest that are directly relevant to the content of this article.

AUTH O R CO NTR I B UTI O N S
RS was responsible for drafting the manuscript, and all authors were responsible for revising the manuscript critically for important intellectual content. EH was responsible for this study's conception. JH was responsible for data analysis. All authors were responsible for the acquisition and interpretation of data. All authors read and approved the final manuscript.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.

O RCI D
Victor M. Montori https://orcid.org/0000-0003-0595-2898 Erik P. Hess https://orcid.org/0000-0002-6499-9592 F I G U R E 4 Cranial CT rate subgroup effects. Forest plot demonstrating the effect of the Head CT Choice decision aid on the rate of cranial computed tomography imaging obtained in subgroups according to patient and parent characteristics