Pilot Trial of a Behavioral Economics-Informed Clinical Decision Support Alert to Improve Inpatient Tobacco Use Treatment Rates

Abstract

Introduction

Evidence-based tobacco use treatment (TUT) improves clinical outcomes, yet few clinicians initiate TUT for hospitalized patients who smoke. Clinical decision support (CDS) tools embedded in electronic health records (EHRs) offer opportunities to guide clinicians toward desired behaviors. CDS alerts informed by behavioral economics (BE-CDS) may increase TUT by presenting preselected orders and requiring justification to opt out. We conducted a pilot study to evaluate the impact of a BE-CDS alert on TUT ordering for inpatients who smoke.

Aims and Methods

In a two-arm randomized trial at a large US academic health system, 50 clinicians were randomized to receive either a BE-CDS alert or a standard reminder encouraging inpatient pharmacotherapy and outpatient follow-up. Alerts were triggered upon chart entry for eligible patients. The primary outcome was the rate of any TUT-related order (medication, counseling, or referral). Clinician feedback on alert appropriateness and workflow impact was also collected.

Results

From June 2022 to December 2023, 635 inpatients met the inclusion criteria. There were no significant differences in rates of any TUT order (25.2% vs. 28.5%, p = .27), inpatient medication (17.2% vs. 18.4%, p = .62), or discharge medication (7.1% vs. 6.6%, p = .90). Referral to outpatient follow-up was higher in the BE-CDS group (8.4% vs. 3.3%; OR = 2.54, p = .01). Number of alerts delivered, and White patient race, were associated with increased TUT rates. Clinicians preferred the BE-CDS format but cited alert workflow placement as more influential than alert design.

Conclusions

BE-CDS improved referral rates but did not significantly impact overall TUT. Further research should explore deeper drivers of inpatient TUT decision-making.

Trial Registration

ClinicalTrials.gov. NCT04738643. Registered February 4, 2021. https://clinicaltrials.gov/study/NCT04738643.

Implications

Compared to the standard electronic practice reminders, behavioral economics-informed clinical decision support alerts increased referrals to outpatient tobacco use treatment services for inpatients who smoke, but did not increase tobacco treatment intervention rates during the inpatient stay. Several variables appear to influence rates of tobacco use treatment, including the number of alert interactions and patient race. This pilot study suggests that the barriers to inpatient tobacco interventions may be complex and therefore insensitive to simple behavioral economic nudges.

Introduction

Tobacco use treatment (TUT) is associated with improved cancer survival rates,¹ improved inflammatory markers of cardiovascular disease,² higher overall quality of life,³ and a significantly lower mortality risk.⁴ Inpatient TUT reduces hospital readmissions⁵ and may reduce length of stay,⁶ which could yield financial savings.^7–9 Groups such as the World Health Organization, the United States Preventive Services Task Force, the American Society of Clinical Oncology, the Department of Health and Human Services, and the US Surgeon General recommend that TUT interventions be more fully integrated into patient care.^10–12 Several quality metrics have been implemented to promote integration of TUT into inpatient clinical workflows.^13,14

Unfortunately, TUT interventions remain highly underutilized across healthcare settings. In oncology, only 5.3% of patients newly diagnosed with smoking-related genitourinary cancers had billing claims for smoking cessation interventions.¹⁵ Similarly, 14% of patients hospitalized for myocardial infarction who smoke are prescribed tobacco cessation medications during their inpatient stay.¹⁶ One large study of electronic health records (EHRs) found that only about one in five hospitalized patients who used tobacco were prescribed tobacco cessation medication, with significant racial inequities and wide variability based on medical specialty.¹⁷

Recent implementation studies have leveraged the EHR, behavioral economic (BE) principles, and pragmatic designs to shrink this practice gap.^18,19 BE-informed strategies can harness decision-makers’ biases and leverage them to align behaviors with evidence-based practice.²⁰ These “nudges” facilitate behavior change without restricting autonomy by adjusting the environment in which people make decisions or the presentation of choices within that environment.²¹ For example, EHR-based nudges to outpatient oncologists utilizing a default referral to the health system’s Tobacco Use Treatment Service resulted in a nearly 3-fold increase in TUT engagement.¹⁹ Such defaults make the desired behavior easier²² while preserving choice.²³ Thus, defaults sit atop the “nudge ladder” as an especially impactful behavior change strategy.^20,24 Defaults can further promote desired behaviors when combined with mechanisms that hold ordering clinicians accountable for their decisions, requiring justification for declining the default. Accountable justification requirements disrupt automatic responses and prompt self-reflection, and may lead to a perceived social pressure to participate in the desired behaviors.²⁵

The inpatient setting may be ideal to test implementation strategies to enhance TUT.⁷ Patients who smoke are especially sensitive to risk after smoking-related health problems,²⁶ and an inpatient admission may serve as a time when patient motivation to engage in TUT is heightened.^27,28 Hospitals are smoke-free environments, requiring patients to abstain from smoking, removing smoking-related environmental cues affecting acute craving^29,30 and promoting long-term cessation.^31,32 One randomized trial tested an inpatient EHR alert that presented clinicians with available orders for tobacco medications and a quit-line referral found that about a third of patients in the EHR alert condition received tobacco medications, a significant increase over the control.³³

Varenicline is among the most effective TUT medications,³⁴ safe and effective for smokers with cancer,³⁵ HIV,³⁶ or major depression.³⁷ It also facilitates smoking cessation among patients uninterested in quitting when treatment begins,³⁸ making it easy to initiate during inpatient care without workflow disruption.^39,40 Current clinical practice guidelines identify varenicline as the optimal “first choice” controller when initiating pharmacotherapy,⁴¹ and despite its delayed onset of peak activity, hospitalized patients started on varenicline at admission experience a significant increase in likelihood of abstinence at 1 year.⁴²

In this pilot trial, we sought to evaluate the utility of an EHR clinical decision support (CDS) alert, enhanced by two BE principles: (1) order for varenicline and referral to an in-house varenicline management service were presented as default, and (2) a required accountable justification mechanism presented if the default order was declined. The comparator was a standard alert reminding clinicians to treat tobacco use, offering optional acknowledged reasons for deferral, and providing convenient links to relevant EHR orders.

Materials and Methods

Design

We conducted a two-arm pilot pragmatic, randomized clinical trial within a single inpatient service line of a large academic hospital. Fifty clinicians providing inpatient care were identified and randomized 1:1 by the study statistician utilizing random number generation to receive the standard alert or BE-enhanced CDS alert (BE-CDS) (Figure 1). Clinicians received their assigned alert on each instance of chart engagement until an order for TUT was placed or the patient was discharged. The primary outcome was the proportion of included inpatients who received any TUT order, including tobacco use medication, nurse counseling, or tobacco treatment service consultation. Secondary outcomes included the proportion of patients for whom tobacco medications were prescribed at discharge (ie, for outpatient use) and the proportion of engaged inpatients who followed through with outpatient tobacco treatment services. Following the intervention period, a clinician survey assessed their priorities for inpatient care, barriers to and facilitators of inpatient TUT, and preferences for alert structure. The Institutional Review Board reviewed and approved all study procedures, providing a waiver of informed consent and HIPAA authorization requirements for both patients and clinicians during the intervention period, and waiving the written consent requirement for the follow-up survey.

Figure 1.

Standard (top) and enhanced (bottom) practice alerts. Two clinical decision support alerts are displayed. The standard alert contains only an option to open or not open a tobacco treatment order set and declination reasons of “patient declined” and “deferred,” while the enhanced alert includes more options for orders (tobacco treatment order set, varenicline order, inpatient consult to the smoking cessation program) and more acknowledgements for declining the order.

Participants

To minimize the potential effects of variation by specialty or patient illness severity, eligible clinicians were drawn from the hospitalist section of the hospital’s Department of Medicine. Included clinicians practiced independently and were prescheduled to provide inpatient care during the trial period, licensed to prescribe medications in Pennsylvania, experienced caring for >1 patient with tobacco use within 30 days of the trial, and English speaking. The alerts were programmed to deploy only during the care of inpatients identified as current smokers. Data on all inpatients for whom an alert was deployed were included in the analysis.

Procedures

Prior to randomization, clinicians received an email explaining the study purpose and procedures, with a detailed explanation of adjustments made to create the BE-CDS. All were provided links to a website outlining the relevance of inpatient tobacco treatment and the appropriateness of varenicline in this context, as well as study team contact information for questions. All clinicians had the ability to decline participation without explanation; none did.

During the intervention period, clinicians received their alert at first inpatient encounter, with repeat exposure on subsequent encounters until a TUT order was placed or the patient was discharged. Because the admitting attending hospitalist served as the unit of randomization and not the patient, all clinicians subsequently engaging that patient’s chart received the assigned alert throughout the intervention period. At the end of the trial, clinicians received an email with questions about their experience with the alerts and the trial. Respondents were compensated $25.

The Alerts

The standard (control) alert (Figure 1, top) consisted of a notice that the patient screened positive for tobacco use and had not received TUT in the past year. This alert was the standard of practice across the health system. Prior to launching the trial, we used a “rapid-cycle” iterative approach to planning and refining the BE-CDS alert design.⁴³ The proposed BE-CDS design was shared with nonsample inpatient clinicians and health system leaders to gather feedback on the content, design, and implementation of the alert within the EHR. Interviewees recommended ensuring that orders are available throughout the admission and do not require phone calls to facilitate consultation. The recommended elements of educational support about the project led to the development of an information sheet that was electronically disseminated to clinicians. The final BE-CDS alert (Figure 1, bottom) featured a similar notice of both smoking status and absent TUT along with default orders to initiate varenicline and referral for varenicline management services. The referral order prompted a consultation with a certified tobacco treatment specialist experienced in both in- and outpatient care within 24 h to assist with cessation counseling, along with navigation to medications and follow-up care. The BE-CDS alert included the same two justification options (ie, “Patient declined” and “Deferred”) as in the standard alert to maintain baseline equipoise and compliance with hospital regulations. These two baseline justification options were felt to predominantly reflect the patient’s perspective on intervention. In addition to designing the BE-CDS justification mechanism as required, the justification options were expanded to include two that reflected the clinician’s point of view to account for the tendency of respondents to rely on their own perspective when working within system-wide rules (ie, egocentric bias).⁴⁴

Measures

The primary outcome measure was the placement of any TUT-related order during the admission. Secondary outcome measures included placement of orders for any inpatient TUT medication, discharge TUT medications, inpatient smoking cessation consult requests, and referral for outpatient follow-up. Summary outcomes are expressed as proportions, with the number of patients with the outcome of interest divided by the total number of patients for whom each alert fired. Additionally, we assessed rates of engagement with tobacco treatment following referral for outpatient management.

After the trial, a brief survey (Supplementary Material 1) was deployed in REDCap to assess standard implementation measures (eg, ease, appropriateness of the BE-CDS alert) and concerns related to varenicline within the inpatient workflow (eg, risk, time burden). Survey measures were operationalized using a 5-point Likert scale (1 = “strongly disagree” to 5 = “strongly agree”). Following each question, participants could explain their ratings via open-ended text boxes. The survey concluded by asking participants whether they preferred the BE-CDS or the standard alert.

Statistical Analysis

We characterized the sample of patients in terms of available demographic information and compared rates of TUT (overall, medication, counseling) across study arms using chi-square tests. Logistic regression compared outcomes across arms while controlling for covariates (eg, demographics and the number of times the alert fired per patient admission). Among patients referred for TUT, we compared rates of treatment engagement across the study arms. Analyses were conducted in Stata version 18. Prior to the trial, our power assessment indicated that with 23 clinicians per arm, we were powered to detect a 20% increase in TUT rates. Post-trial survey responses were characterized using descriptive statistics.

Results

Sample Characteristics

Twenty-eight (56%) clinicians were randomized to the standard alert condition and 22 (44%) to the BE-CDS. A total of 635 patients identified as using tobacco were engaged by clinicians who participated in this trial. On average, patients were 56.5 years old, 42.8% were female, 74.8% were African American, and 2.2% were Hispanic. Of these patients, 397 (62.5%) were seen by clinicians randomized to receive the standard alert and 238 (37.5%) by clinicians receiving the BE-CDS alert. Table 1 displays the characteristics of the patient sample by study arm. At baseline, patients whose clinicians were randomized to receive the BE-CDS alert were older (58.0 vs. 55.7 years, t(633) = 2.05, p = .04) but did not differ based on race, ethnicity, or gender. Figure 2 shows the CONSORT diagrams of the trial clinicians and patients.

Table 1.

Baseline Patient Demographics by Arm

	Standard alert	BE-CDS alert	p
Age (Mean, SD; years)	55.66 (13.73)	57.99 (14.02)	.04
Gender (% female)	175 (44.1%)	97 (40.8%)	.41
Race (%)			.71
Black or African American	294 (74.1%)	181 (76.1%)
White	82 (20.7%)	43 (18.1%)
All other racial groups	21 (5.3%)	14 (5.9%)
Ethnicity (% Hispanic)	9 (2.3%)	5 (2.1%)	.88

BE-CDS = behavioral economics-clinical decision support.

Figure 2.

CONSORT flow diagram of patients and clinicians. Fifty clinicians were initially assigned to study arms. Twenty-eight were assigned to the standard alert, and these clinicians saw 397 engaged patients and completed 113 tobacco treatment orders. Twenty-two clinicians were assigned to the enhanced alert, and these clinicians saw 238 engaged patients and completed 60 tobacco treatment orders.

Effects of Alerts on TUT

Inpatient TUT (ie, any medication, counseling, or referral) was ordered for patients whose clinicians received the BE-CDS alert at similar rates as those whose clinicians received the standard alert (25.2% vs. 28.5%; χ²[1] = 0.79, p = .37) see Table 2. In the logistic regression, TUT rates remained similar across arms (OR = 0.81, 95% CI = 0.56% to 1.18%, p = .27). White patients were significantly more likely to be ordered TUT than non-White patients (39.2% vs. 24.3%; OR = 2.09, 95% CI = 1.37% to 3.20%, p = .001) but did not differ by gender (OR = 0.75, 95% CI = 0.53% to 1.07%, p = .12). TUT rates increased with the number of alerts fired per hospital admission (OR = 1.03, 95% CI = 1.00% to 1.07%, p = .02).

Table 2.

Study Outcomes by Arm

	Standard alert		BE-CDS alert		p
	Totals	Percent	Totals	Percent
Overall TUT	113/397	28.5%	60/238	25.2%	0.37
Inpatient TUT medication	73/397	18.4%	41/238	17.2%	0.71
Outpatient/discharge TUT medication	26/397	6.6%	17/238	7.1%	0.77
Referrals to TUTS counseling service	13/397	3.3%	20/238	8.4%	0.005

BE-CDS = behavioral economics-clinical decision support.

Focusing on medication orders, the alert’s impact on both inpatient and discharge orders was similar between groups. During the inpatient stay, 17.2% vs. 18.4% (χ²[1] = 0.14, p = .71) received medication orders, while 7.1% vs. 6.6% (χ²[1] = 0.08, p = .77) received orders at discharge for BE-CDS and standard alerts, respectively. Logistic regression also suggested no effect of study arm (OR = 0.90, 95% CI = 0.58% to 1.39%, p = .62). The number of alert exposures per admission was marginally related to increased rates of tobacco medication ordering for both inpatient (OR = 1.03, 95% CI = 1.00% to 1.06%, p = .08) and discharge (OR = 1.05, 95% CI = 1.02% to 1.10%, p = .006). During the hospital stay, White race was related to an increased likelihood of receiving TUT medication (30.4% vs. 14.9%; OR = 2.51, 95% CI = 1.58% to 3.99%, p < .001), while male gender was marginally related to a decreased likelihood (15.7% vs. 21.0%; OR = 0.67, 95% CI = 0.44% to 1.02%, p=.06).

Only 33 (5.2%) patients were referred to the tobacco treatment service during their inpatient stay. The proportion of patients referred was significantly higher among clinicians who received the BE-CDS alert vs. control (8.4% vs. 3.3%, χ²[1] = 7.94, p = .005). After controlling for demographic variables and number of alert firings, arm assignment remained significantly related to referrals (OR = 2.54, 95% CI = 1.23% to 5.25%, p = .01). No measured covariates were significant.

Justifications for overriding an alert varied. Of the 732 overridden alerts (25.8%), the most common reasons were “Deferred” (59.7%) and “Patient declined” (22.7%). Of the 732, few were accompanied by additional written justification in the BE-CDS (6.8%) or standard (5.3%) arm.

Outpatient TUT Engagement

Of the 33 patients referred to the outpatient cessation program, 10 (30.3%) completed their appointment within 3 months of hospital discharge, with rates similar across study arms (χ²[1] = 2.64, p = .1). Eight outpatients (80%) were recommended a TUT medication and 9 (90%) were recommended a follow-up appointment. There were no differences in rates of recommendations by study arm (χ²[1] = 1.94, p = .16; χ²[1] = 1.17, p = .28).

Clinician Survey Responses

Seventeen clinicians (34%) completed the postintervention survey (mean age = 43.9 years; 59% female, 71% White); 65% of responding clinicians indicated that they preferred the BE-CDS over the standard alert. Aggregate clinician ratings of the BE-CDS alert were 3.24/5 for “ease of use” and 2.88/5 for “reduces burden.” Regarding the general idea of integrating TUT within inpatient care, clinicians were mixed as to whether this should be a clinical priority (3.18/5); they somewhat disagreed that there was enough time to provide TUT (2.59/5) while rating TUT as a safe option for the inpatient setting (3.82/5). Respondents were mixed on whether varenicline was appropriate (3.24/5); open-ended responses revealed concerns about new daily medication, mental health side effects, and the time needed to engage patients, pharmacies, and insurance authorizations. More than content, respondents emphasized that alert design should focus more on where alerts are deployed within the workflow. Multiple clinicians expressed a need for more education, with respondents previously trained in TUT reporting higher mean response values regarding reduction in burden (3.14/5 vs. 2.71/5), sufficient time to start treatment (3.00/5 vs. 2.43/5), and priority of TUT (3.57/5 vs. 2.57/5).

Discussion

A substantial practice gap persists in the treatment of tobacco use in the inpatient setting. In this pragmatic randomized pilot trial, prompting inpatient clinicians with a TUT alert that included default orders and accountable deselection justification yielded a limited effect on this gap. Though we failed to detect a difference in overall inpatient TUT rates or orders for medication, a significantly higher proportion of patients whose clinicians received the BE-CDS alert were referred for outpatient TUT, suggesting the BE-CDS alert was more effective at making the need for TUT more cognitively available during periods of more focused clinical decision making, such as the period preceding hospital discharge. Interestingly, our BE-CDS alert was designed to require more “clicks” to decline than to accept, suggesting an affirmative decision was being made to deprioritize TUT during routine day-to-day care interactions. An effect on cognitive availability was also supported by our observed association between the number of alert interactions and TUT intervention. This effect, if real, warrants further study, particularly when balanced against the possibility of inducing alert fatigue during complex inpatient care episodes.⁴⁵ It is plausible that improving the alignment between the timing of alerts and the processes of inpatient care may increase their impact, though the most impactful time point remained poorly defined by our respondents.

A second important finding from this pilot relates to persistent racial disparities in inpatient TUT rates. Non-White patients were significantly less likely to receive TUT across both alert conditions. Past studies of inpatient TUT have found similar disparities,^16,17 as have studies in various clinical and nonclinical settings.^46,47 Our study did not collect information to help understand this disparity. This association could have been driven by previously reported clinician⁴⁸ or patient⁴⁹ biases or by unanticipated differences in reason for admission.⁵⁰ Because our pilot employed a convenience sample of clinicians working in a single hospital, socioeconomic disparities and other social determinants of health inherent to the surrounding West Philadelphia community may have played a role.⁵¹ Future research designed to increase TUT in the inpatient setting must include specific efforts to ensure equitable engagement in TUT.

Lastly, the results from this study underscore the magnitude of persistent gaps in the provision of TUT among inpatients who smoke. Overall, only about one in four patients received any TUT intervention, with evidence-based medication provided for less than one in five. Clinicians are not capitalizing on the structural facilitators of smoking cessation in the hospital setting, including the potential utility of supervised medication initiation and the availability of in-house tobacco treatment services. In open-ended comments, respondents did suggest that more education about TUT could increase their willingness to provide TUT. This observation tracks well with previously reported associations between clinicians’ perception of comfort and competence, with a greater likelihood of providing TUT,⁵² and lack of knowledge as a barrier to inpatient TUT.⁵³

Providers commonly perceive behavior change barriers and facilitators in terms of “capability” and “opportunity.”⁵⁴ Similarly, in our sample, requests for TUT education accompanied disagreement with “enough time” to provide TUT. For this reason, simple educational interventions are unlikely to be effective. More intensive implementation strategies, such as facilitation from trained support staff,⁵⁵ system-level education,⁵⁶ efforts to address culpability bias,⁵⁷ or performance feedback⁵⁸ may be required to significantly increase TUT in this context. Though reliability was limited by low response rates, previous TUT training was associated with more positive attitudes about TUT in our survey, suggesting that multicomponent implementation strategies merging CDS with education and role modeling may be required.

The results of our pilot contrast with studies focused on the outpatient setting, yielding significant increases in TUT using similar BE-CDS interventions.¹⁹ Notable limitations should be recognized. First, results from our convenience sample may have been influenced by social desirability bias, with survey responses favoring TUT while behaviors favoring consistency with local norms. Our pilot was powered to detect a substantial shift in behaviors—an expected effect size that may have been too ambitious, given the cultural elements of inpatient care. Though clinician assignment to the two arms was relatively equivalent, the number of patients meeting criteria for alert firing opportunities differed significantly between the groups. This appears to have been a function of the pragmatic pilot design, with natural variations in patient admission characteristics and smoking prevalence affecting the number of opportunities presented. The greater number of opportunities presented to standard alert clinicians could conceivably have biased our results toward the null.

Second, we opted to have the alerts activated immediately upon chart opening. Though we included repeated alert presentations to offset this possibility, this may have led to presentation at nondecision-making moments and the canceling of the default orders. The timing of alerts and their potential to disrupt clinical workflow is important to consider in future work in this area. Third, the impact of accountable justification may have been undermined by the potential for identified workarounds. Future studies should identify the optimal balance between the value of open-ended, individualized justifications and the increased workflow burden this may cause. Finally, the structure of the study, in which all randomized clinicians practiced within one hospital, could have permitted contamination between study arms if clinicians discussed the intervention alert with their colleagues or if patients were seen by multiple clinicians.

Going forward, studies that evaluate implementation strategies via the EHR to increase inpatient TUT should consider broadening their implementation design to include care process-specific strategies, and supplement electronic implementation with clinician-directed strategies such as enhanced education or direct trainer facilitation to address both capability and opportunity domains. In addition, future research in this area should prioritize evaluating multilevel implementation strategies that include community representation in alert development^59,60 and directed quality improvement initiatives aimed at promoting the equitable provision of evidence-based TUT for all patients.⁶¹

Author Contributions

Robert Schnoll (Conceptualization [lead], Formal analysis [equal], Funding acquisition, Investigation, Methodology, Project administration, Supervision, Writing—original draft, Writing—review & editing [lead]), Daniel Blumenthal (Investigation [supporting], Project administration [equal], Writing—original draft, Writing—review & editing [lead]), E. Paul Wileyto (Data curation [equal], Formal analysis [lead], Methodology, Writing—original draft, Writing—review & editing [supporting]), Anna-Marika Bauer (Investigation [equal], Project administration [lead], Supervision [equal], Writing—review & editing [supporting]), Sarah Evers-Casey (Project administration, Supervision [equal], Writing—review & editing [supporting]), Nathaniel Stevens (Project administration, Writing—review & editing [equal]), Tierney Fisher (Project administration [equal], Supervision, Writing—review & editing [supporting]), Susan Ware (Data curation [lead], Software [equal], Writing—review & editing [supporting]), Brian P. Jenssen (Investigation, Methodology, Supervision, Writing—review & editing [supporting]), Colin Wollack (Software [lead], Writing—review & editing [supporting]), Spencer Schwartz (Project administration, Writing—review & editing [supporting]), and Frank T. Leone (Conceptualization [equal], Investigation [lead], Methodology [equal], Resources, Supervision [lead], Writing—original draft, Writing—review & editing [equal])

Funding

This trial was funded by the Commonwealth of Pennsylvania Department of Health (grant SAP number 4100083101 to author RS).

Declaration of Interests

None declared.

Data Availability

Participant data reported in the article, after de-identification, will be available at the time of publication, along with a data dictionary. A methodologically rigorous proposal to use the data should be provided to the corresponding author. Data will be made available with limited investigator support after investigator approval of the proposal and approval by the University of Pennsylvania.