Acceptability of Using Large Language Models to Support Smoking Cessation Attempts: a Qualitative Study Among African Americans Who Smoke

Abstract

Introduction

African Americans face significant disparities in tobacco-related diseases and access to smoking cessation resources. Digital health interventions using large language models (LLMs) show promise in increasing engagement and successful quit attempts. However, impressions regarding acceptability of LLM-based interventions among African Americans are unknown and evidence suggests such interventions require tailoring for underserved communities. Our goal was to evaluate perceptions of LLM-based interventions among African Americans to inform development of an LLM-based chatbot to support quit attempts.

Methods

We conducted remote focus groups with adult African Americans residing in Minnesota who smoke cigarettes to determine acceptability of LLM-based health interventions and inform the functionality of an LLM-based chatbot supporting quit attempts. Discussions followed a structured guide and were recorded to facilitate analysis. Transcripts were coded independently by two analysts with a shared codebook.

Results

21 adults participated in focus group discussions. 57.1% of participants were female and the mean age was 45.6 (SD, 10.5). Three themes united participants’ feedback: (1) participants limited familiarity with LLM and related AI technology, (2) willingness to use an LLM-based app to support quit attempts, and (3) perceptions of features that impact the acceptability of an LLM-based chatbot to support quit attempts.

Conclusion

African Americans have complex perceptions and expectations regarding the use of LLMs to deliver health interventions. While many recognize the potential benefits, significant concerns and knowledge gaps remain. To build trust and acceptability, intervention developers should prioritize community engagement and ethical guidelines to ensure that LLM-driven solutions are inclusive and effective for multiple communities.

Implications

Large language models and related AI technologies have shown promise for improving access and engagement when deployed through digital health interventions. Focus groups with African Americans who smoke combustible cigarettes revealed factors impacting the acceptability in using LLM-based interventions to support quit attempts. Efforts to build digital health interventions should prioritize community engagement to reduce apprehensions and improve alignment with community needs and values.

Introduction

Background

Large language models (LLMs) have shown significant advancements in recent years that have broad implications for their application to healthcare. Recently trained LLMs are able to offer personalized responses to users based on preferences, patterns, location, and other information provided by the user.¹ A highly publicized example of an LLM is the OpenAI conversational agent (CA) ChatGPT, which can have conversations with users and provide tailored information. Feedback from the user then allows the LLM to continue being trained to produce responses that are increasingly acceptable to the user. These recently developed features have shown success across LLM-based chatbots and artificial intelligence (AI)-based interventions related to depression, anxiety, and other behavioral health needs.^2,3 This may allow for increased access to counseling-such as services where LLMs supplement the efforts of human counselors in some circumstances. As this possibility becomes a reality, more information is needed on the acceptability of such interventions to potential end users. Further, LLMs have been found to have the potential to perpetuate racial bias and disparate outcomes due to dataset bias influencing model performance (e.g., training data).^4,5 The current study focused on evaluating the acceptability of an LLM-based chatbot smoking cessation intervention delivered via a smartphone app among African Americans who smoke cigarettes.

L‌LM-Based Digital Health Interventions

LLM-based chatbots could potentially be integrated into existing digital smoking cessation interventions such as text messaging and Just In Time Adaptive Interventions (JITAI). Text messaging interventions using a static bank of pre-written messages are an already validated approach to smoking cessation intervention.^6–8 Similarly, JITAI are a class of digital health interventions that deliver support at predictive-model-defined times of need (e.g., when cravings are the highest) or opportunity (e.g., when coping skills are able to be used).⁹ Because recent LLMs are more lifelike and flexible than previous models, they show great promise to support JITAIs in identifying situations where participants are at increased risk of smoking and providing in-the-moment support through tailored text messages. Improving the acceptability and effectiveness of digital interventions is particularly important for African Americans. African Americans who smoke are at increased risk of smoking related disease and have less access to evidence-based smoking cessation treatments.^10–12 Acceptable digital interventions would have broad accessibility and reach.

The aim of the current study is to gather feedback from African Americans who smoke regarding the acceptability of an LLM based smartphone app intervention for managing smoking triggers during a cessation attempt. In particular, we were interested in (1) familiarity and impressions of LLMs and related AI-based technologies, (2) the acceptability of a large-language model-based chatbot, and (3) acceptability of the planned intervention features. Data from this study will be used to inform design of a future app using an LLM to help African Americans manage triggers during a smoking cessation attempt.

Methods

Design and Setting

We organized five focus groups with adult (21+) African Americans residing in Minnesota who use combustible cigarettes to determine the acceptability of LLM-based health interventions, and to inform the design and functionality of an LLM-based conversational agent to support craving management during a smoking cessation attempt. Focus groups were conducted remotely using Zoom. Discussions were led by an African American male social scientist trained in qualitative methods, urban ethnography and African American studies (W.M.) and a White female clinical psychologist with expertise in behavioral medicine and nicotine dependence (S.J.). A White male coordinator with a background in psychology was present to take notes and respond to logistical questions. Each participant received a $40 honorarium for their participation in the focus groups and $10 for completing a demographic survey. The study was determined to be exempt by the IRB. This manuscript was organized according to the COREQ guidelines for reporting qualitative research.¹³

Participants Criteria

Inclusion criteria included the following characteristics: self-identified African American/Black race, adults 21 and older, possessing a smartphone, and use of 5 or more combustible cigarettes per day. Exclusion criteria were inability to provide consent, residing outside of Minnesota, and not being a native English speaker. Exclusion of non-native English speakers was motivated by the current limitations of the English language automatic speech recognition technology with variable performance on non-native English speech.¹⁴ Adults under the age of 21 were not included as they are unable to legally purchase tobacco products in the state of Minnesota. Individuals not based in Minnesota were excluded due to priorities established by the study sponsor. Potential participants were approached using a clinical trial recruitment service (Build Clinical), Craigslist, and from lists of individuals interested in participating in studies using IRB approved materials for communicating information about the study.

Data Collection and Analysis

Focus group sessions followed a discussion guide organized around three priority areas: (1) acceptability of LLMs and related artificial intelligence (AI), (2) acceptability of an LLM-based smoking cessation app (Personalized Assistant for Smoking Cessation using AI and Large language models [PASCAL]), and (3) acceptability of planned clinical trial procedures to evaluate the acceptability and feasibility of using an LLM-based smoking cessation app to manage smoking triggers among African Americans making a quit attempt. This manuscript will focus on the first two priority areas. Participants were also asked to review low fidelity mockups of PASCAL that highlighted the app’s functionality and navigation. Mockups showed how a user would set alerts to potential high risk smoking times, report smoking events and triggers, and an example of a text exchange demonstrating how the CA would respond to a user’s request for in-the-moment support (Figures 1 and 2). Qualitative data consisted of direct quotations from the study participants. The data was transcribed verbatim and coded to identify common themes.

Figure 1.

Example of PASCAL mockups showing smoking event reporting.

Figure 2.

Example of PASCAL CA responding to user.

Transcripts of study sessions were analyzed using Dedoose Version 7.0.23 (Los Angeles, CA: Sociocultural Research Consultants, LLC). A senior analyst (W.M.) developed a preliminary code book using a combination of deductive and inductive approaches to organize direct responses to prompts on the discussion guide and to allow for emergent themes. A junior analyst (D.F.) was trained to use the codebook using a sample of study transcripts to refine code definitions and confirm alignment with the original codebook. Afterwards, W.M. and D.F. divided the remaining transcripts for independent coding. Questionnaires administered prior to the start of each focus group session collected participant demographics (e.g., age, gender, education level), self-reported smoking history including the number of cigarettes smoked per day and prior quit attempts, information on behavioral factors such as motivation to quit assessed using questions adapted from the theory of planned behavior,¹⁵ and healthcare literacy as measured by the Single Item Healthcare Literacy Screener.¹⁶

Results

Participant Characteristics

A total of 21 adults participated in 5 focus group discussions, with the number of participants per group ranging from 3 to 6 individuals. Recruitment and scheduling of additional study sessions was suspended after achieving saturation with regards to feedback on mockups and impressions of LLM-based health interventions. Participant characteristics are reported in Table 1. A majority of participants identified as female (57.1%), mean age was 45.6 (SD, 10.5), and education attainment varied with 9.5% having less than a high school degree, 28.6% of participants having completed high school, 23.8% completed some college, 28.6% attained an Associate’s degree and 9.5% attained a Bachelor’s degree. A majority of participants report Black/African American race alone (85.7%). 9.5% and 4.8% of participants also identified as Multiracial or Native American/American Indian/Alaska Native/Indigenous, respectively. On average participants smoked 11.2 combustible cigarettes per day (SD, 8.0) and 57% indicated that they had previously attempted to quit. Past supports for quitting included using a smartphone based quit app (9.5%), nicotine replacement therapy (NRT) (54.5%), Bupropion (14.3%) and counseling (19.0%). Mean scores for a single item evaluating motivation to quit from 1-10 was 6.0 (SD, 2.9) indicating moderate motivation. Mean scores for self-reported healthcare literacy were 1.6 (SD, 1.0) and 1.5 (SD, 1.0) when evaluated using items that assess the need for assistance reading and completing medical forms, respectively. Scores greater than 2 indicate a need for assistance in these areas.

Table 1.

Participant Characteristics

Variable	Subcategory	Mean/N	SD/%
Race [N (%)]	Black/African American	18	85.7%
	Hispanic/Latino/Latina	0
	Asian/Asian American	0
	Native Hawaiian/Pacific Islander	0
	Native American/American Indian/Alaska Native/ Indigenous	1	4.8%
	White/Caucasian	0
	Multiracial	2	9.5%
Gender [N (%)]	Man	8	38.1%
	Woman	12	57.1%
	Non-binary	1	4.8%
	Transgender	0
Education [N (%)]	Less than high school	2	9.5%
	High School	6	28.6%
	Some College	5	23.8%
	Associate’s	6	28.6%
	Bachelor’s	2	9.5%
	Master’s	0
Age [M (SD)]		45.6	10.5
Cigarettes per day [M (SD)]		11.2	8.0
Time to first cigarette in the morning [N (%)]	Within 5 minutes	9	42.9%
	6-30 minutes	7	33.3%
	30-60 minutes	4	19.0%
	After 60 minutes	1	4.8%
Menthol cigarettes use (N(%))		20	95.2%
Current use of other tobacco product (N (%)]	Cigars	8	36.4%
	Pipe Tobacco	6	27.3%
	Snus (pouch tobacco)	4	18.2%
	Chewing tobacco	4	18.2%
	Hookah	5	22.7%
	Vape or e-cigarette	12	54.5%
Previous quit attempt [N (%)]	Yes	12	57.1%
	No	9	42.9%
Previous use of smartphone quit app (N(%))		2	9.5%
Previous smoking cessation medication use [N (%)]	NRT	12	54.5%
	Bupropion	3	14.3%
	Varenicline	0
Previous smoking cessation counseling use [N (%)]		4	19.0%
Motivation to quit M (SD)*		6.0	2.9
Help reading medical material M (SD)**		1.6	1.0
Confidence filling out medical forms M (SD)**		1.5	1.0

^*The Motivation to quit measure assesses and is scored on a 1-10 scale

^**The Single Item Health Literacy Scale assesses subjective health literacy. The measure has 2 versions. Both are scored on a 1-5 scale with scores greater than or equal to 2 indicating some need for assistance when reading medical materials and completing forms.

Thematic Analysis

Three major themes emerged from the analysis of focus group data. First, African Americans who use combustible cigarettes had limited familiarity with LLMs and related AI technologies. Second, African Americans who use combustible cigarettes generally expressed a willingness to use an LLM-based app to support future quit attempts. Third, participants shared their impression of how different features impact the acceptability of an LLM-based chatbot to support quit attempts. See Table 2 for additional quotations representative of themes.

Table 2.

Quotations Representative of Themes

Theme	Excerpt
Theme 1: Limited familiarity with LLMs and related technologies	“I’m not a big fan of it. Nothing replaces the interaction of the human, the empathy that you get from a human and everything, their ability to switch up and bend a rule maybe or not go by the book. Because you know, every situation is different.” “Don’t get me wrong; I like to be able to take advantage of technology but not at human expense” “I mean, if I have to, yes. But, if I had a choice, I’d rather try to do it the old-fashioned way – medicine and everything else or therapy and stuff. But I know, in the future, this is what’s coming to us. So I’m preparing myself to learn it and understand it. But I don’t agree with it.” “I don’t know how much information can be stored onto the AI, like what personal information it’s going to take from me or if it’s – I don’t know. I guess it’s just reading the terms of service, just like any other thing. So I just haven’t found myself too familiar with it. So I don’t really have concerns. But that would be a main concern of just data sharage [storage] and just personal information.”
Theme 2: Willingness to use LLMs to support quit attempts	“I’m not sure. It’s all new, especially technology. So I’m open to learning, seeing how it goes.” “I definitely would mess with this [PASCAL]. The way your presentation the app just did… Like I told you, I don’t agree with AI. But I feel that it’s kind of necessary right there. Because a reminder – somebody to distract you to remind you to do something different. Like, when I eat a nice meal, that’s the first thing I think of is a cigarette. So it would be good with AI to text me and say, “Go outside, go on a walk, or chew some bubblegum or something.” So I think I can mess with this.” “Yes. For me, I think it wouldn’t be a good idea to have a conversation if I’m having a craving or something and I’m feeling that I’m getting ready to light up. My apprehension is talking to a machine, you know what I’m saying? But if it’s helpful, you know, I’m open, you know, to suggestions.”
Theme 3: Acceptability of smartphone app and LLM-chatbot to support quit attempts	“So maybe keeping up with how much you smoke would be a good idea, like using it for the app, because then you could see how much you actually smoke and how often you smoke, kind of like psychologically, just to see how often you doing it.” “That would help. That would help. So long as the chatbot is decent, like I’ve had experience with, like, I mean, the main one that pops up when I think of one is like a crisis center, stuff like that. Their chatbots are kind of mid. Like, I honestly can’t think of a professional way of how it’s worth that, but it’s just kind of, eh. You can tell you’re talking to a robot, you know what I mean? It’s just like, if there is a way to make it where it doesn’t feel as much bad, like how it is with other call centers and stuff of that nature, then I can see it working. I can see it being helpful.” “The one I use right now is kind of like a stopwatch, if you will. You can kind of check it on your own free will. It doesn’t really notify you a whole lot. You know what I mean? But with stuff like this, with something more detail oriented, it lets you recognize your habits that lead to you smoking, and then you can kind of break that cycle that way. I think that’s honestly kind of cool.” “That would be a good idea, the keeping track with it [cigarettes not smoked] to see how much you save, how much money you save and stuff like that.” “For me, I would like for it to consistently remind me of the health issues that I’m going through and the consequences behind them. Which I already know – I know the consequences, I do – but I do need that support system from that and that constant reminder.” “Yeah. I would say, yeah. I would think, yeah. Like if the app could maybe set something up maybe three times or a Monday, Wednesday, Friday or whatever times that you can just make like, okay, the group is be open from, we’ll say, 6:00 to 8:00 or whatever. And it would just be a Zoom conference that everybody in the program can come to and we can just voice and talk about the app, talk about what we’re going through. Just have the app send us that link just to have an open forum. So, if we are having like a day and just say I don’t have anybody to talk to, I know that it’s Thursday, I know that the group is open from 6:00 to 8:00 or 6:00 to whatever time. I can just go in and just vent about the apps, talk about a glitch. Maybe it’s something that I need to work on or communicate like that. If the app had something like that, that would be actually pretty cool, I think.”

Theme 1: Limited Familiarity with LLMs and Related Technologies

Most participants were unfamiliar with LLMs and only a few reported using one in professional or personal settings, for instance to draft letters or messages for social media. As a result, participants who were unfamiliar with LLMs and related-AI technologies often conflated them with automated voice messaging and automated tellers, for instance those used in customer service call lines and in grocery store checkout lanes, respectively. Participants’ frustrations with these non-LLM based technologies were transferred to the potential of LLM-based technologies being used in digital health interventions. For instance, participants were concerned that LLMs would “replace” people and take jobs, and that technology was not sufficiently responsive or personalized to effectively replace the need for human agents. Multiple participants shared their impressions that these services represented reduced quality of service and frustration with having to seek out assistance from a live agent to meet their needs. Among the participants who had heard of LLMs but not used one, many reported being exposed via social and broadcast media reporting. Such reporting covered a range of topics including risks associated with privacy/data security, workforce considerations, and election interference.

“Okay, I’ve never heard anything about AI myself, but I’m willing to be open-minded about it, you know what I’m saying?”

“Like oh, it’s [media reporting] really pushing something that they just may replace humans. Because that’s – I mean, replace with the AI things. They got them all out in restaurants. They got by McDonalds that they’re down in Texas somewhere where they take orders and the robots or AIs or whatever are taking your orders and stuff like that.”

“I ain’t never used it. They got it on Google where you can use it and ask it questions portions and all that type of stuff.”

“I have a little apprehension about it, too. Heaven forbid technology feels that they don’t need humans anymore. That’s my greatest fear there. And then, the replacement of jobs and opportunities, too, you know ”

“Yeah, just messaging, like texting in Grammarly and software like that, just writing letters and stuff like that, I've been using it and stuff like that.”

Theme 2: Willingness to Use LLMs to Support Quit Attempts

While many participants were unfamiliar with or suspicious of LLMs and related technologies, they did support the development of an LLM-based app to provide just-in-time support during a quit attempt. Participants’ consideration for what constitutes effective in-the-moment support ranged from the chatbot providing suggestions for distractions (e.g., games, math problems, stories) to serving as a virtual quit smoking coach. There was no clear preference for engaging with the chatbot via text or speech, but participants did support facilitating both approaches to engaging with the chatbot. While the majority of participants voiced support for using LLMs to support quit attempts, there were those who were apprehensive and voiced concerns about how effective and accurate a non-human agent would be for providing support.

“But like I said, this kind of technology – I think that there’s good in that. You know, there’s nothing insidious or something that’s telling me to do something that’s harmful to myself or others. So this is probably – I’m for this. I believe good can come up out of this.”

“Yes, I would like to do it [use the app to report smoking events and cravings to train the model] like pretty much every time I had the urge to smoke just because that would be the only way it would help me is if it could catch my urge to smoke. That in itself would be something that would replace smoking, being on the app”

“Yes. For me, I think it wouldn’t be a good idea to have a conversation if I'm having a craving or something and I'm feeling that I'm getting ready to light up. My apprehension is talking to a machine, you know what I’m saying? But if it’s helpful, you know, I'm open, you know, to suggestions.”

“I think it’d be appealing. Like if I'm craving, I have somebody like the text box where it can text me or to call me to help my craving.”

Theme 3: Acceptability of Smartphone App and LLM-Chatbot to Support Quit Attempts

Several of the proposed features influenced the acceptability of using smartphone app and LLM chatbot to support quit attempts. Mockups of PASCAL included walkthroughs of interactive features for reporting smoking incidents, documenting triggers, scheduling alerts at potential high risk smoking times and receiving support from the chatbot via text. Overall, feedback on these features was positive and suggested that they would close gaps in support that hindered previous quit attempts and be responsive to their in-the-moment needs. Participants’ feedback also included suggestions for developing additional features to provide users with a patient-centered report of their smoking behavior (e.g., averted smoking lapses, money saved, life years gained), words of encouragement / stories of success, and opportunities to support other users. One participant reported using a smartphone app to support a past quit attempt and that its primary function included tracking their smoking habits.

“They said, oh, AI is making fake people talk, saying … like the way your presentation is of this and how you broke it down, it’s only right that I do give it a try. I can't lie and say I don't like it now it’s broken down and, especially, I’m trying to quit and what it’s doing here. So I've got to give it a chance and see.”

“Like, when I eat a nice meal, that’s the first thing I think of is a cigarette. So it would be good with AI to text me and say, “Go outside, go on a walk, or chew some bubblegum or something.” So I think I can mess with this.”

“I'm not sure if anybody asked this, but is there going to be like a forum or something like that? I mean, an app where you can speak to like who's ever using the app, you can speak to one another for support?”

“Somehow that we can all communicate and talk about our day or how we get away from smoking or why you choose to smoke. Because, sometimes, you might need to talk to somebody that’s going through the same thing you're going through.”

“For me, I would like for it to consistently remind me of the health issues that I'm going through and the consequences behind them. Which I already know – I know the consequences, I do – but I do need that support system from that and that constant reminder.”

Discussion

To our knowledge, no published studies have examined patient beliefs about the use of LLM-based chatbot interventions for smoking cessation. In the current study, we established that African Americans who smoke had limited familiarity with LLMs and related technologies and that exposure was most often driven by media reporting and in professional and social media capacities. Despite limited familiarity, participants considered it acceptable to use LLMs to support quit attempts if accompanied by features that are responsive to their needs and motivations to quit. Participants voiced concerns over the efficacy of a non-human agent, yet participants found the ability to receive adaptive support in the moment to be an improvement over unassisted quit attempts and digital interventions that primarily track smoking behavior. While many of these concerns were grounded in the user experience, a minority of participants did voice concerns regarding data security and to a lesser extent receiving false information from the LLM.

In general, end user beliefs about LLMs did not include concerns previously raised by professionals in the medical field. Although LLMs and related AI-technologies may yield numerous benefits in the medical field, previous op-eds have enumerated several potential pitfalls identified by healthcare providers and system leaders as well as potential end-users. First, providers express concerns over liability in the event of AI making mistakes in critical situations.^17,18 While none of the participants in the current study commented on AI-introduced biases in decision making, a previously conducted randomized experiment revealed that social identity (i.e., race) was associated with end users’ perceptions of and responses to algorithmic bias.¹⁹ Second, AI can potentially “hallucinate” and deliver false or misleading information with confidence. Hallucinations can be difficult for the user to differentiate from factual information and if acted upon could lead to harm or increased risk of harm.²⁰ These hallucinations may be especially worrisome among African American patients, as recent LLMs have been shown provide responses that are biases towards African American users.²¹ Further, they may provide responses that lack evidence and instead perpetuate disparities, causing more harm to the user. Thus, being aware of these types of hallucinations is critical when using LLMs to prevent acting on of false and biased responses. Third, like participants in the current study, health system leaders and administrators have raised concerns about data security and concerns about LLMs and their developers having access to confidential patient data. While protecting identifiable health data is possible, there are significant costs associated with developing the IT infrastructure and technical expertise required to keep data local to the health system and ensure data safety.²² Furthermore, the general public has expressed concern over aspects of AI, including allowing AI access to confidential health information which may act as a deterrent to patients consenting to studies and procedures and utilizing digital health interventions.¹⁸

The following limitations should be considered when interpreting the results: first, participants’ limited familiarity with LLMs and AI technologies resulted in the potential conflation of these technologies with each other and unrelated technologies. Participants often used “AI” as a stand in for LLMs and other automated technologies during study sessions leading to a degree of imprecision. We feel this risk was reduced by presenting participants with mockups demonstrating how the smoking cessation app would function and how LLMs would be implemented in the interactive support features. Second, the overrepresentation of older participants may impact the generalizability of the study findings as younger individuals may have a different level of familiarity with LLMs in professional and social settings. Third, the study was conducted in Minnesota and focused on African Americans who smoke. People from other regions or cultures may have different concerns about LLMs. Future research should assess the acceptability of LLMs and AI across other demographic groups and use cases.

Conclusions

This study highlights the complex perceptions African Americans hold regarding the use of large language models to deliver health interventions. While many recognize the potential benefits of improved access and personalization, significant concerns remain. To foster greater trust and acceptability, healthcare providers and intervention developers must prioritize equity-inclusive user engagement and clear ethical guidelines for LLM-driven healthcare solutions. Current study findings support future research to evaluate the efficacy of LLM based smoking cessation interventions to increase quit rates among African Americans who smoke. Feedback collected during this study will be used to inform app development.

Abbreviations

CA – Conversational agent

CVD – Cardiovascular disease.

LLM – Large language model.

NRT – Nicotine replacement therapy.

PASCAL – Personal assistant for smoking cessation using artificial intelligence and large language models.

Author Contributions

Warren McKinney (Formal analysis [lead], Investigation [equal], Writing - original draft [lead], Writing—review & editing [lead]), Devyn Fernholz (Formal analysis [supporting], Project administration [supporting], Writing—original draft [supporting]), Serguei Pakhomov (Conceptualization [supporting], Writing—review & editing [supporting]), Michael Kotlyar (Conceptualization [equal], Funding acquisition [equal], Project administration [equal], Writing—review & editing [equal]), Sheena Dufresne (Project administration [equal]), Paul Thuras (Conceptualization [supporting], Writing—review & editing [supporting]), Emily Welle (Data curation [supporting], Project administration [supporting]), and Sandra Japuntich (Conceptualization [lead], Formal analysis [supporting], Funding acquisition [lead], Investigation [supporting], Methodology [lead], Project administration [lead], Supervision [lead], Writing—review & editing [equal])

Funding

This work was supported by the National Institute on Minority Health and Health Disparities P50MD017342 (PI: Allen ML; Hardeman RR; Patten CA).

Declarations of Interests

The authors have no conflicts of interest to disclose.

Data Availability

The deidentified data and data collection protocols that support the findings of this study are available on request from the corresponding author, SJ. The data are not publicly available due to their containing information that could compromise the privacy of research participants.