Conversational Engagement Using a Short Message Service Chatbot After Total Joint Arthroplasty

Abstract

Background

Utilizing conversational analytics in orthopaedic surgery may provide insights into patients’ experiences and outcomes. This study retrospectively assessed how patients interacted with a perioperative chatbot and whether the topic of patients’ queries could offer insight on their outcomes after total knee or hip arthroplasty.

Methods

We identified 1338 patients (746 knees and 592 hips) who enrolled in a short message service chatbot from 2020-2022 with greater than 3 months of follow-up. The total number and topics of patient-generated text responses to the chatbot were recorded. Independent t-tests, chi-squared tests, and linear regressions were performed to determine if specific patient-generated responses to the chatbot or overall chatbot engagement were associated with demographics or outcomes.

Results

Readmitted patients interacted less with the perioperative chatbot than those who were not readmitted (3.9 messages vs 12.7 messages, P < .0001). Return to emergency department (ED) and reoperation rates were not associated with engagement. Patients who visited the ED within 90 days of their surgery were most commonly seeking advice about walking after surgery (P = .0046) and weaning off their opiate pain medications (P = .0281). Patients who were readmitted to the hospital were similarly seeking advice about walking (P = .0188) and weaning off pain medications (P = .0218). Replying “exercise” was associated with no ED visits or readmissions (P = .0009). Patients with a mental health diagnosis were more likely to reply with high-acuity topics (P = .0052).

Conclusions

The topic of chatbot queries and chatbot engagement were associated with patient outcomes after total knee arthroplasty or total hip arthroplasty and may provide insight to patients’ perioperative courses.

Introduction

Total hip arthroplasty (THA) and total knee arthroplasty (TKA) are well-established and effective interventions for patients suffering from arthritis, offering improved mobility and enhanced quality of life. [1] The volume of primary THAs and TKAs continues to rise, and current projections estimate about 850,000 and 1,921,000 annual THAs and TKAs, respectively. [2,3] Patients’ success following total joint arthroplasty (TJA) extends beyond the operating room and encompasses preoperative preparation, postoperative rehabilitation, and clinical monitoring. Patient engagement is a critical factor in patients’ perioperative progress and several patient engagement platforms now exist ranging in degree of participation, from more passive platforms like patient portals to more active platforms like mobile health applications and chatbots. [4]

Given the recent advancements in artificial intelligence (AI), special attention has been paid to AI-powered chatbots due to their ability to promote patient engagement through the perioperative process. [5,6] Chatbots can simulate conversation and deliver immediate information to patients. [[7], [8], [9]] They can deliver clarification regarding postoperative instructions and answer general questions that commonly arise during the postoperative recovery period. Several commercial AI-powered perioperative chatbots exist for patients undergoing TJA, and their utility in THA and TKA continues to be studied.

The purpose of this study was to retrospectively assess how patients interacted with a perioperative chatbot. Specifically, we analyzed if demographic variables were associated with chatbot use and whether the topics of patient-generated queries were associated with particular outcomes following THA or TKA. We hypothesized that men and women would engage the chatbot equally and that “notify” and “pain” messages to the chatbot would be associated with an increased rate of emergency department (ED) visits. Additionally, we assessed whether the level of patient engagement with the chatbot would correlate to clinical outcomes, such as ED visits or readmissions. We hypothesized that increased engagement with an AI-powered chatbot would correlate with decreased readmissions and ED visits.

Material and methods

Patient selection

A retrospective review was performed on all patients who underwent THA or TKA at a single academic tertiary center from 2020 to 2022. All patients were enrolled in a short message service (SMS) AI chatbot and had follow-up of at least 3 months following THA or TKA.

Data collection

Demographic variables were collected, including age, sex, and ethnicity. Comorbidities including body mass index (BMI), Charlson comorbidity index (CCI) score, and diagnosis of anxiety and depression were collected. The total number of patient-generated text responses to the chatbot was recorded in the perioperative period. Additionally, the total number of patient-generated text responses by topic was recorded.

Our orthopaedic clinic monitors each TJA patient at specific intervals to assess their clinical progress and identify problems early on. These calls are on postoperative days 1-2, 7-10, 30, 60, and 90. All calls are documented within the electronic health record and provide a general assessment of the patient’s postoperative course, mobility, and if they have been presented to an ED or if they have been admitted to a hospital since surgery. Additionally, ED visits, reoperations, and readmissions are documented in postoperative clinic visits. Therefore, we recorded any visit to the ED, readmission, or reoperation within 90 days of surgery at our institution or any ED participating in the Epic Care Everywhere Network, or identified via chart review in postoperative telephone or clinic notes.

Chatbot engagement

The SMS-based AI chatbot used in this study was StreaMD (Chicago, IL) [10]. This orthopaedic chatbot engages patients throughout the entire perioperative process and sends automated messages to patients preoperatively and postoperatively. One of the authors (K.J.C.) was responsible for the initial conceptualization of this chatbot in addition to its final implementation. He was involved in the interface as well as the functional elements of the chatbot to ensure it was user-friendly and efficient. The automated messages from the chatbot encompass a broad range of topics including preoperative preparation tips, postoperative recovery guidance, therapy videos, and personalized messages directly from their surgeons. Additionally, patients have the ability to initiate conversations with the chatbot and can inquire about various topics, ranging from pain management strategies to showering protocols. The patient-initiated queries were recorded and categorized, and these individual topics are summarized in Table 1. Figure 1 is a typical automated response provided by the chatbot to a patient inquiring about weaning narcotics. StreaMD tracks the volume and topic of patient-generated queries throughout the perioperative process and allows this information to be shared with providers. Although this information remains deidentified to the third party, providers can cross-reference listed telephone numbers and surgical dates to then link this information to patients’ electronic health record.

Table 1.

Chatbot topics generated by patients.

Chatbot topic	Description
Pain	Texts were categorized as “pain” if patients responded to the chatbot with any variation of postoperative discomfort or pain.
Exercise	If patients responded to the chatbot regarding postoperative exercises or rehabilitation, these texts were categorized as “exercise.”
Motivate	If patients wished to receive an encouragement from the chatbot, they were asked to respond with “motivate” to be provided with perioperative motivation from the chatbot.
Emotions	Messages were categorized as “emotions” if patients responded to the chatbot with any emotional or psychological struggles perioperatively.
Notify	Patients were instructed to respond with “notify” to the chatbot if they have had an emergency department or unexpected hospital visit.
Wean	If patients engaged the chatbot regarding discontinuation of any of their medications, these texts were categorized as “wean.”
Walking	Any messages to the chatbot regarding a postoperative ambulation and restrictions were categorized as “walking.”
Pack	If patients had questions regarding preoperative instructions for the day of their surgery, the term “pack” would subsequently generate a comprehensive list of essential items that patients should bring with them to the surgical facility.
High Acuity	“High acuity” was defined as any patient-generated text regarding dislocation, unable to bear weight, chest pain, breathing difficulty, stroke symptoms, blood clot, infection, wound dehiscence, wound redness/warmth, infection, emergency room, fall, or unusual bleeding.

Figure 1.

Example of chatbot responses to weaning narcotics and removing dressings.

Statistical analysis

Independent t-tests, chi-squared tests, and linear regressions were performed to determine if specific patient-generated responses to the chatbot were associated with demographic variables, comorbidities, and clinical outcomes including ED visits, readmissions, and reoperations postoperatively. Statistical analysis was performed using SAS 9.4 (Cary, N.C.). Statistical significance was defined as P < .05 for all tests.

Results

Patient characteristics

A total of 1338 patients were identified (746 knees and 592 hips) who were enrolled in StreaMD. A total of 58.5% of patients identified as female and 44.4% of patients identified as male, while 92.2% of patients were Caucasian and 5.8% of patients were Hispanic. The average age in the StreaMD cohort was 62.7 (standard deviation [SD] 11.9). The average BMI was 32.9 (SD 7.7) and the average CCI score was 1.8 (SD 2.3). A total of 12.1% of patients had a diagnosis of anxiety and 30.4% a diagnosis of depression. Demographic characteristics of this patient cohort are summarized in Table 2.

Table 2.

Patient demographics.

Demographic	StreaMD (N = 1338)
	N (%)
Sex
Men	554 (44.4)
Women	781 (58.5)
Ethnicity
Hispanic	77 (5.8)
Not Hispanic	1258 (94.2)
White
White	1200 (92.2)
Non-white	102 (7.8)
Surgery
TKA	746 (55.8)
THA	592 (44.3)
Anxiety
Yes	162 (12.1)
No	1176 (87.9)
Depression
Yes	407 (30.4)
No	931 (69.6)

Demographic	Mean (standard deviation)
Age	62.7 (11.9)
BMI	32.9 (7.7)
CCI	1.8 (2.3)

Number of texts sent

Patients with insurance other than Medicaid (12.5 total messages, non-Medicaid vs 9.7 total messages, Medicaid, P < .0309), females (14.0 females vs 10.0 males, P < .0001), and those undergoing TKA (15.1 for TKA vs 8.9 for THA, P > .0001) were significantly more likely to engage the perioperative chatbot when analyzing total text messages sent as well as messages sent after surgery (P < .05) (Table 3). Increasing age also significantly correlated with increased total perioperative and postoperative chatbot engagement with a β coefficient of 0.77 (P = .005) (Table 3). Patient ethnicity (P = .5185), race (P = .9067), native language (P = .9589), and BMI (P = .9721) did not correlate to the total number of texts sent. Additionally, diagnosis of comorbidities including diabetes (P = .205), rheumatoid arthritis (P = .5955), and CCI score (P = .278) did not correlate to the total number of texts sent (Table 3).

Table 3.

Perioperative chatbot engagement.

Clinical variable	Total texts sent		Texts sent before surgery		Texts sent after surgery
	Mean (standard deviation)	P-value	Mean (standard deviation)	P-value	Mean (standard deviation)	P-value
Medicaid
Yes	9.7 (13.9)	.0309	1.6 (3.3)	.3323	8.1 (12.1)	.0101
No	12.5 (12.2)		1.2 (2.0)		11.3 (11.5)
Ethnicity
Hispanic	13.3 (14.6)	.5185	1.3 (2.1)	.3501	12.2 (13.9)	.342
Not Hispanic	12.3 (12.1)		1.1 (1.7)		11.0 (11.4)
White
White	12.4 (12.2)	.9067	1.3 (2.0)	.7598	11.1 (11.5)	.947
Non-white	12.5 (14.0)		1.3 (2.8)		11.2 (1.3)
English speaking
Yes	12.3 (16.6)	.9589	1.3 (2.1)	.1096	11.1 (11.4)	.761
No	12.2 (12.1)		0.8 (1.2)		11.5 (16.0)
Gender
Female	14.0 (13.3)	<.0001	1.4 (2.0)	.0074	12.6 (12.6)	<.0001
Male	10.0 (10.3)		1.1 (2.3)		8.9 (9.5)
Surgery
TKA	15.1 (13.5)	<.0001	1.3 (2.3)	.768	13.9 (12.7)	<.0001
THA	8.9 (9.5)		1.3 (1.9)		7.6 (8.7)
Anxiety
Yes	12.9 (12.4)	.6859	1.5 (2.3)	.1422	11.4 (12.8)	.8434
No	12.4 (12.2)		1.2 (2.1)		11.2 (11.5)
Depression
Yes	12.5 (11.8)	.9532	1.3 (1.9)	.9447	11.2 (11.2)	.9596
No	12.5 (12.7)		1.3 (2.2)		11.2 (11.9)
Diabetes
Yes	13.5 (14.4)	.205	1.5 (2.3)	.1622	12.1 (13.6)	.269
No	12.2 (11.9)		1.2 (2.1)		10.9 (11.2)
RA
Yes	11.6 (10.6)	.5955	1.7 (1.8)	.1514	10.0 (9.6)	.4074
No	12.5 (12.5)		1.3 (2.1)		11.3 (11.7)

Clinical variable	β coefficient	P-value	β coefficient	P-value	β coefficient	P-value
Age	0.077	.005	0.002	.697	0.075	.0034
BMI	0	.9721	0	.4555	0	.862
CCI	−0.159	.278	−0.006	.826	−0.154	.264

Return to ED/readmissions

Patients who were readmitted within 90 days of surgery were significantly less likely to engage with the chatbot (3.9 total messages vs 12.7 total messages, P < .0001) (Table 4). There was no significant difference in chatbot engagement of patients who required either reoperation (P = .3483) or had an ED visit (P = .6203) within 90 days of surgery (Table 4).

Table 4.

Chatbot engagement and clinical outcomes.

Clinical outcome	Total texts sent		Texts sent before surgery		Texts sent after surgery
	Mean (standard deviation)	P-value	Mean (standard deviation)	P-value	Mean (standard deviation)	P-value
ED visit
Yes	11.3 (17.4)	.6203	0.9 (1.4)	.0662	10.4 (16.5)	.7158
No	12.6 (12.2)		1.3 (2.2)		11.2 (11.4)
Readmission
Yes	3.9 (7.3)	<.0001	1.3 (3.2)	.9678	5.6 (6.0)	<.0001
No	12.7 (12.6)		1.3 (2.1)		11.5 (11.8)
Reoperation
Yes	11.6 (10.8)	.3483	1.4 (3.2)	.5999	10.1 (10.1)	.3074
No	12.7 (12.6)		1.3 (2.0)		11.4 (11.8)

Patient-reported outcome measures (PROMs)

Table 5 summarizes the correlations between perioperative chatbot engagement and PROMs. Notably, patients with higher preoperative physical function (β = 0.156, P = .0009) and lower preoperative pain scores (β = −0.037, P = .0378) had significantly higher chatbot engagement. Additionally, postoperative physical function was associated with increased chatbot engagement (β = 0.038, P = .0419). However, when assessing the change in preoperative to postoperative PROMs, there were no significant associations between chatbot engagement and PROMs.

Table 5.

Postoperative PROs and engagement with chatbot.

Patient reported outcome	Total texts sent		Texts sent before surgery		Texts sent after surgery
	β coefficient	P-value	β coefficient	P-value	β coefficient	P-value
Preoperative PROs
Physical function	0.156	.0009	−0.021	.0093	0.176	<.0001
Physical health	0.033	.1294	−0.005	.1448	0.038	.0615
Mental health	0.074	.1285	−0.006	.444	0.08	.0794
Pain	−0.037	.0378	0.002	.4775	−0.04	.0193
KOOS JR	0.048	.2991	0.006	.3645	0.042	.3364
HOOS JR	0.068	.0366	−0.004	.4021	0.072	.017
Postoperative PROs
Physical function	0.038	.0419	−0.198	.0963	0.047	.0154
Physical health	0.035	.4623	−0.141	.6338	0.043	.3911
Mental health	0.049	.0422	−0.005	.9734	0.054	.0316
Pain	−0.053	.3822	0.865	.0026	−0.085	.19
KOOS JR	0.083	.0791	0.087	.7825	0.091	.0704
HOOS JR	0.039	.6435	−1.07	.018	0.088	.3311
Change in PROs
Physical function	0.035	.5400	0.017	.0888	0.018	.7393
Physical health	0.101	.1674	0.019	.1185	0.082	.2313
Mental health	0.021	.7825	0.003	.7968	0.018	.8032
Pain	0.009	.6738	−0.004	.2360	−0.005	.8120
KOOS JR	0.010	.8741	0.002	.8512	0.008	.8908
HOOS JR	0.013	.7616	0.000	.9902	0.013	.7433

KOOS JR, Knee Dysfunction and Osteoarthritis Outcome Score for Joint Replacement; HOOS JR, Hip Dysfunction and Osteoarthritis Outcome Score for Joint Replacement.

Chatbot queries

Patients’ queries were categorized into specific content messages and correlated with specific patient demographics and clinical outcomes. Patients who were seen in the ED postoperatively most commonly engaged the chatbot about ambulation (P = .0046) and weaning narcotics (P = .0281). Patients who were readmitted after initial discharge queried about ambulation (P = .0188) and weaning narcotics (P = .0218) as well. Of the patients who did not present to the ED after surgery, perioperative exercises, tips to stay motivated, and advice about emotional management were the most frequent text messages sent by patients. Querying "exercise" was associated with patients not being readmitted (P = .0009) nor having any ED visits (P = .0001) postoperatively. Patients with mental health diagnoses, such as major depressive disorder or general anxiety disorder, were more likely to reply with a high acuity topic (eg, chest pain, dyspnea, calf pain, etc.) (P = .0052).

Discussion

The primary finding in this study is that Medicaid patients, older patients, females, and patients undergoing TKA were significantly more likely to interact perioperatively with the SMS-based chatbot. Additionally, patients who were readmitted within 90 days of their TJA had significantly less interaction with the chatbot. There was no significant association between chatbot engagement and changes in PROMs from preoperatively to postoperatively.

With the continual reduction in reimbursements for TJA and simultaneous increasing difficulty in hiring and keeping employees devoted to patient engagement such as medical assistants and nurses, offloading some of the communication and education burden to an automated technology has obvious benefits. Furthermore, TJA outcomes are now more than ever linked to return to ED and readmission and patient perceptions, rather than just the historical data points of complications, range of motion, and PROMs. The challenge to the arthroplasty surgeon is not just completing the surgery but guiding a team committed to both improving patient outcomes and patient perception of that process. This holds especially true with TKA, as nearly 20% of patients are dissatisfied following TKA. [11,12] There has been notable attention to technology in the operating room with increasing prevalence of robotic-assisted surgery in recent years. [13] Technology outside of the operating room has additionally received increased attention in an attempt to improve patient engagement perioperatively. These forms of technology are commonly referred to as patient engagement platforms, and the most common examples are patient portals, mobile health applications, and chatbots. [4] Each platform has its unique strengths and limitations and should be tailored to the needs of a particular practice.

Patient portals are online platforms that allow patients to log into portions of their electronic health record. They can also function as a secure email to query their providers. Patient portals facilitate information access and communication through the perioperative process. However, these portals are typically accessed through personal computers or smartphones, which a number of Americans do not possess. [14] Additionally, patient portals are unable to engage patients proactively and require motivation from the patient to engage this platform. Mobile health applications are smartphone applications that encourage patient engagement and monitor overall clinical care. They offer a wide range of features that can proactively engage patients and can also be synced to wearable technology to track clinical progress. For example, patients can sync a particular mobile health application to their Apple Watch, which then allows patients' daily step count, gait speed, and gait asymmetry to be recorded, among other metrics. [15] Mobile health applications are also capable of collecting PROMs, which may be helpful for research purposes. For patients to utilize these applications, they must own a smartphone. This technology barrier can be prohibitory to a significant minority of the US population. [14]

In contrast to patient portals and mobile health applications, chatbots have several unique advantages. First, chatbots utilize AI to simulate conversation and immediately answer patients’ questions. [16] TJA continues to increase in volume in the United States, and current projections predict over 850,000 and 1,920,000 THAs and TKAs by 2030, respectively. [3] This increase in clinical volume presents challenges to surgeons and clinics in reducing available time and resources dedicated to perioperative patient engagement. Chatbots may help offload the patient communication workload of surgeons and clinics. This is highlighted in prior investigation, which found that use of an SMS-based chatbot improved patient experience and enhanced patient mood while reducing office telephone calls. [17] The chatbot was also shown to improve time spent on home exercises and decrease narcotic use, which further emphasizes the effectiveness of this platform. [17]

The second advantage of chatbots is their widespread availability and the minimal technology ownership required to communicate with them. The chatbot utilized in this study is an SMS-based chatbot, such that patients with cellphones capable of basic texting are able to interact with this engagement platform. [10] According to the Pew Research Center, around 15% of Americans do not possess a smartphone, but around 97% of people possess a phone capable of basic text messaging. [14] Prior studies using patient portals have shown that young, Caucasian, employed, and wealthy patients have significantly higher engagement rates with the communication platform. [18,19] They hypothesize that this population is more comfortable with the use of a computer and has more access to the internet. [18,19] On the contrary, our study found that Medicare patients have significantly higher engagement with the SMS-based chatbot, which could be due to a cell phone being more easily accessible than routine internet use in this patient population. Additionally, the SMS-based chatbot utilized in this retrospective review is available in 20 languages and allows patients who are limited-English proficient to actively engage with this technology. In a prior study published by the current authors, patients with limited-English proficient who enrolled in the chatbot had fewer readmissions (0% vs 8.3%, P = .13) and trended toward fewer ED visits (0.9% vs 8.0%, P = .085) compared to those not enrolled. [20] The limited technology requirement of this particular chatbot and its availability in multiple languages therefore allow for increased inclusivity among patients undergoing TJA.

Lastly, SMS-based communication is extremely effective in capturing individuals’ attention. For example, it has been documented that nearly 99% of marketing text messages are opened within the first 20 minutes while email marketing campaigns struggle with a mean open rate of only 20%. [[21], [22], [23]] An SMS-based chatbot would theoretically follow suite in capturing patients' attention to a similar degree.

Although chatbots have notable strengths, they are not without their unique limitations. Chatbots require a broad clinical conversational database in order to simulate conversation with patients and several require sophisticated natural language processing. [24] Many of these commercial applications are not integrated with Health Insurance Portability and Accountability Act-compliant messaging platforms and cannot be integrated within a patient’s electronic health record. This can make research cumbersome and fragmented if a clinic has not thoughtfully planned how to integrate this platform into their daily practice.

Ultimately, the primary purpose of any patient engagement platform is to help improve patient outcomes. Patient outcomes can be measured in various ways including rates of return to the ED, readmission rates, reoperation rates, PROMs, and functional outcomes. Our study showed a significant correlation between low patient engagement with the chatbot and higher readmission rates, with no significant correlation to rates of return to ED or reoperation. These patients who were readmitted or returned to the ED inquired more about walking after surgery and weaning off their opiate medications, whereas patients who did not get readmitted or return to the ED inquired more about specific exercises to the chatbot. Contrary to our study, one randomized controlled trial of 452 patients found a significant reduction in ED visits using a smartphone-based platform after primary partial and TKA (8.2% vs 2.5%, P = .014), while no difference was seen in readmission rates. [25] In a retrospective review comparing outcomes before and after the implementation of a perioperative patient engagement and pathway management solution (PES), Higgins et al. found that the PES cohort had a significantly shorter length of stay, fewer reoperations within 60 days of surgery, and higher PROMs than the pre-PES cohort. [26] The shorter length of stay in their study resulted in a secondary benefit of reduced overall cost. [26] Rosner et al. similarly found that their digital patient engagement platform resulted in significantly lower cost as well as a 54.4% significant relative reduction in 90-day complications. [27] Higher patient engagement using a web-based portal was also associated with significant improvement in postoperative physical function after TKA and led to a more rapid recovery after THA compared to those with less portal engagement. [18] Our study adds to the current body of literature available that demonstrates superior patient outcomes using a patient engagement portal.

This study is not without limitations. Given its retrospective design, only inferences can be drawn as to how patients engage with a perioperative chatbot and their correlation with clinical outcomes. Additionally, all patients were treated at a single, academic institution, and the demographic composition of our patient population may not be applicable to other geographic regions of the United States. Furthermore, the patient follow-up period was limited to 90 days following surgery, and we are unable to make definitive conclusions about the potential effects of perioperative chatbot interactions and long-term outcomes for patients undergoing THA or TKA. To evaluate associations between perioperative chatbot engagement and clinical outcomes beyond 3 months postoperatively, long-term data will need to be collected and analyzed.

Conclusions

Medicaid patients, females, and patients undergoing TKA were significantly more likely to interact perioperatively with this SMS-based chatbot. Patients who were readmitted within 90 days following THA or TKA interacted significantly less with the chatbot compared to those who were not readmitted. Although preoperative physical function and pain scores correlated with chatbot engagement, there was no significant association between chatbot engagement and changes in PROMs from preoperatively to postoperatively.

Conflicts of interest

L. A. Anderson is a speaker bureau and paid consultant for Medacta; has stock options in OrthoGrid; and receives research support from Stryker and Zimmer. M. J. Archibeck receives research support from Zimmer and is a paid consultant for Zimmer Biomet. B. E. Blackburn is a board/committee member of AAHKS. K. J. Campbell is an unpaid consultant for BoneFoam and has stock options in STREAMD. J. M. Gililland receives royalties from OrthoGrid; is a paid consultant for Stryker, OrthoGrid, and Enovis; has stock options in OrthoGrid, CoNextions, and MiCare Path; receives research support from Zimmer Biomet, Stryker, and Medacta; is an editorial board member of the Journal of Arthroplasty; and is a board/committee member of AAHKS. C. E. Pelt receives royalties from Total Joint Orthopaedics and Smith and Nephew; is a speaker bureau of Total Joint Orthopaedics; is a paid consultant for 3M and Total Joint Orthopaedics; has stock options in Joint Development, LLC; receives research support from Zimmer Biomet, Peptilogics, and Smith and Nephew; and is a board/committee member of AAHKS and AAOS. The other authors declare no potential conflicts of interest.

For full disclosure statements refer to https://doi.org/10.1016/j.artd.2024.101484.

CRediT authorship contribution statement

Joshua P. Rainey: Writing – review & editing, Writing – original draft, Visualization, Resources, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Emily A. Treu: Writing – review & editing. Kevin J. Campbell: Writing – review & editing, Data curation. Brenna E. Blackburn: Writing – review & editing, Methodology, Formal analysis, Data curation. Christopher E. Pelt: Writing – review & editing. Michael J. Archibeck: Writing – review & editing. Jeremy M. Gililland: Writing – original draft, Resources, Investigation, Conceptualization. Lucas A. Anderson: Writing – review & editing.