Scientific Abstract Presentations: Research & Technology Session 3 Thursday, September 26 Thu, Sep 26 5:00-6:00 p.m. PDT

Purpose/Background: Artificial Intelligence (AI) has emerged as a transformative force in diverse medical applications, spanning health monitoring, drug development, diagnostics, and treatment planning. Notably, AI's proficiency in identifying clinical pathologies from medical imaging has revolutionized diagnostic accuracy. Recent studies have underscored the pivotal role of AI in text-to-image generation. Despite these strides, a gap persists in assessing AI's text-to-image generation in anatomic illustration, for example, in accurately representing facial feminization surgery-an integral component of gender-affirmation addressing gender dysphoria and enhancing quality of life. This study aims to assess the quality of AI generated images by different text-to-image models through comparative analysis with a published medical illustration.

Methods and Materials: In this study, five image-to-text AI models-Muse AI, BlueWillow, Stable Diffusion 2.1 Demo, Gencraft, and Picsart-were employed to generate visual representations of facial feminization surgery. Each AI received a specific prompt: "Please create a visual representation illustrating the outcomes of facial feminization surgery, depicting the facial transformation from its initial masculine state to the post-surgical feminine appearance. Ensure the left image portrays the pre-surgery condition with masculine features, while the right image depicts the post-surgery outcome with feminine features." A gender-affirming surgeon assessed the generated images across five categories: adherence to the prompt, anatomical accuracy, detail orientation, artistic quality, and clear communication. A control image depicting facial feminization surgery was sourced from the Cleveland Clinic website. Data collection was facilitated through the Qualtrics platform.

Results: Gencraft and Picsart exhibited exceptional performance, earning a perfect total score of 25 out of 25 (100%) with the highest ratings (5, strongly agree) in all five assessment categories, comparable to the score of a control image with 100%. In contrast, Muse AI received a score of 40% (10/25), displaying challenges in adherence to the prompt, anatomical accuracy, detail orientation, artistic quality, and clear communication. Bluewillow and Stable Diffusion Demo 2.1 achieved a comparable total score of 64% (16/25). Qualitative analysis highlighted Muse AI's issues with illegible text and "jumbled" features, while BlueWillow captured facial feminization surgery (FFS) essence, retaining some masculine features. Stable Diffusion's image portrayed FFS features but included surgically implausible elements. Gencraft and Picsart stood out, receiving praise as the "best" examples, notably for their effectiveness in patient education and aesthetic features.

Conclusion : Our results indicate that AI has the potential to be a useful tool for generating medical illustrations, but comes with inherent limitations. These limitations can vary depending on the platform used and on the type of content, such as words, that are included in the source images. Therefore, AI needs to be continually improved and tested as a potential tool for patient education.

Purpose Mandibular angle fractures have historically high rates of reoperation and complication leading to a particularly high morbidity rate. Mandible angle fractures are notoriously difficult to operate on due to an inability to properly visualize the site of fixation. Mixed Reality (MR) is novel technology that allows users to interact with and manipulate digital information that is superimposed on the real world. We demonstrate a method of planning trajectories and guiding screw placement by using MR to visualize the entire skull, including the mandible, prior to an incision. Methods Two "targets" made of metal and measuring roughly 4mm x 1mm were fixed to the left mandible angle of a cadaver, which was then CT scanned. The CT was uploaded into the Medivis SurgicalAR system and projected as a hologram using the Microsoft HoloLens 2. Virtual target and entry points were placed on the hologram over both of the metal targets and at corresponding points on the exterior surface of the face, creating two trajectory lines at 90 degree angles to the mandible. Holes were drilled according to concentric circles around the virtual trajectory lines that became green once the distance from the drill to the entry point was less than 1.5mm and the angle of drilling was less than 2 degrees from the intended trajectory. Results We percutaneously drilled Time to plot trajectories, place fiducials, and match the hologram to the cadaver was 4 minutes 53 seconds. Time to locate correct trajectories and drill screw guide holes was 5 minutes and 13 seconds. On assessment of the drilling accuracy, six guide holes were drilled with mean deviation from intended location of 1.67mm, with a minimum of 0mm and maximum of 3mm. Mean deviation from ideal angle trajectory of 90° was 2.8° ± 2.7°.
Conclusions In this first preclinical validation of MR-guided percutaneous screw placement, we found the MR trajectory guidance to be rapid and highly accurate. Inaccuracies were likely due to inaccurate placement of trajectories, which can be improved with increased use. A future direction for MR software for OR usage is the ability to plot trajectories at 90° perpendicular to a given tangent to CT-identified anatomy. We believe, however, this technology is approaching high levels of utility and has the potential to both reduce complication rates as well as operating time in mandible angle fracture fixation.

Background: Mastectomy skin flap necrosis (MSFN) is a devastating complication of breast reconstruction. MSFN can exacerbate physical disfigurement, prolong recovery, and inflict emotional and financial strain on the patient and healthcare system as a whole. Up to 30% of patients may be affected, and current predictive methodologies fall short (1). In response to this challenge, we aimed to create a machine learning (ML) model to determine factors highly associated with MSFN and to predict future cases with confidence in a patient population with high incidences of comorbidities. The nature of this population presents a unique opportunity to explore the social, in addition to clinical, determinants of MSFN. In the midst of the increasing prevalence of breast cancer, prediction of this frequent complication facilitates the development of safe, personalized treatment plans, particularly for the most vulnerable patients.

Methods: An IRB-approved retrospective review was conducted on patients who underwent total mastectomy paired with immediate breast reconstruction at a single academic medical center in the Bronx, New York, from January 2015 to January 2022. Reconstruction was prosthetic or autologous. This study extracted 24 patient and operative factors to train and test seven machine learning (ML) models. Patient data was randomly divided into training (80%) and testing (20%) sets. Resampling was used to ameliorate class imbalance before exploring a comprehensive parameter sweep for each model. After choosing optimal parameters, models were evaluated on the unseen testing set for final performance, and feature importance analysis was conducted.

Results: A total of 647 breasts (428 patients) were included in this study. The patient population was predominantly non-White Hispanic (30.1%) and Black/African American (29.7%). Mean age (SD) was 49.5 (10.8) years. Notable comorbidities included obesity (mean BMI of 29.6 +/- 5.7 kg/m3), hypertension (37.6%), smoking history (32.3%), and hyperlipidemia (21.3%). The overall MSFN rate was 14.5%. The ML models showed strong predicting power with AUC up to 0.74 and accuracy up to 83%. The models also revealed 6 most significant predictors of MSFN, including race, diabetes, coronary artery disease, hypertension, hyperlipidemia, and regular preoperative use of anticoagulation therapy.

Conclusions: Retrospective review revealed a racially diverse population with high rates of comorbidities, factors which were found to be associated with the development of MSFN by the ML models. These findings underscore the importance of MSFN awareness and proactive preventative care in communities with poorer health outcomes. Overall, high model performance validates the use of artificial intelligence modeling in the prediction of MSFN following mastectomy with immediate prosthetic or autologous breast reconstruction.

References: Nykiel, M., Sayid, Z., Wong, R., & Lee, G. K. (2014). Management of mastectomy skin flap necrosis in autologous breast reconstruction. Annals of Plastic Surgery, 72 Suppl 1, S31-34. https://doi.org/10.1097/SAP.0000000000000174

Background: Diabetes mellitus is a group of diseases characterized by persistent hyperglycemia. Under these conditions, wound healing is impaired. Hyperglycemic Wounds are characterized by chronic inflammation, epithelial and endothelial cell injury, and a pathologic angiogenic process. (1) Alpha1-antitrypsin (AAT) is a circulating acute phase glycoprotein that redirects inflammatory signals toward inflammatory resolution (2,3), epithelial gap repair, mature blood vessel formation (4), and expedited wound healing. (5) In hyperglycemic individuals, circulating AAT is glycated and inactive. Clinical-grade human AAT has a remarkable safety record and is furthermore a candidate for treating wounds in individuals with poorly controlled hyperglycemia. Methods: Mice were treated with STZ to induce hyperglycemia, followed by an 8 mm dorsal excisional wound. The wound was treated with topical AAT (50 micro gram/wound) or albumin at the onset and every three days. Wound closure was measured by image analysis, and the wound was further investigated by histology and gene expression. In vitro, epithelial scratch assays were exposed to diabetic patient serum with and without clinical-grade AAT, and gap closure rates were documented. AAT was also glycated under laboratory conditions; cells were treated with Glycated-AAT, and inflammatory profiles were measured. Results: Hyperglycemic mice showed a delay in wound closure compared to normoglycemic mice. hyperglycemic mice receiving hAAT treatment demonstrated a significantly more robust healing response than albumin-treated wounds (Day 6: 17.55±7.98 vs. 30.97±9.86 percent from day 0, p<0.001). In vitro, cell re-epithelialization was hindered by hyperglycemic patient serum. However, adding AAT to the serum led to a twofold increase in cell density within 12 hours. Moreover, macrophages that were exposed to glycated-AAT showed elevated secretion of pro-inflammatory cytokines that were alleviated by naïve-AAT administration. Discussion: In hyperglycemic conditions, glycated-AAT turns pro-inflammatory, impairing wound healing. Yet, introducing naïve-AAT notably enhances wound closure and reduces inflammation, suggesting its role in treating diabetic wound complications. Reference 1. Burgess JL, Wyant WA, Abujamra BA, Kirsner RS, Jozic I. Diabetic wound-healing science. Medicina (Lithuania). 2021;57(10). doi:10.3390/medicina57101072 2. Schuster R, Motola-Kalay N, Baranovski BM, et al. Distinct anti-inflammatory properties of alpha1-antitrypsin and corticosteroids reveal unique underlying mechanisms of action. Cell Immunol. 2020;356. doi:10.1016/j.cellimm.2020.104177 3. Lewis E. a1-Antitrypsin Therapy for Non-Deficient Individuals: Integrating and Mitigating Cross-Pathology Inflammatory and Immune Responses To the Injured Cell. Internal Medicine Review. 2017;3(5):1-28. doi:10.18103/imr.v3i5.451 4. Bellacen K, Kalay N, Ozeri E, Shahaf G, Lewis EC. Revascularization of pancreatic islet allografts is enhanced by α-1-Antitrypsin under anti-inflammatory conditions. Cell Transplant. 2013;22(11):2119-2133. doi:10.3727/096368912X657701 5. Gimmon A, Sherker L, Kojukarov L, et al. Accelerated Wound Border Closure Using a Microemulsion Containing Non-Inhibitory Recombinant α1-Antitrypsin. Int J Mol Sci. 2022;23(13):7364. doi:10.3390/ijms23137364

Background: The integrated plastic surgery residency program is one of the most competitive residencies to match into. Therefore, prospective plastic surgery applicants may harbor a lot of uncertainty about their odds of matching. The Texas STAR (Seeking Transparency in Application to Residency) database is a new residency match database that provides data and subjective advice from successful and unsuccessful integrated plastic surgery applicants so prospective applicants can gauge where they stand. This study aims to use the Texas STAR database from 2020 – 2023 to provide an updated analysis of the trends in plastic surgery applicants, to elucidate characteristics of ideal applicants, pitfalls of unsuccessful applicants, and areas of an application where programs are putting increasing emphasis.

Methods: Self-reported survey data from applicants to integrated plastic surgery residency between 2020 and 2023 were obtained from the Texas STAR database. Objective statistics from all applicants were extracted to determine their significance with regards to matching using linear and logistic regression models. The primary outcome of analysis is applicant match success. Subjective advice from successful and unsuccessful applicants was extracted and summarized into general themes.

Results: A total of 281 integrated plastic surgery applicants responded to the Texas STAR survey from 2020 to 2023. Of these applicants, 208 matched (74.0%) and 73 did not match (26.0%). Matched applicants had a significantly higher number of interviews offered (16.3 vs. 8.1, p = 0.000), USMLE Step 1 score (248.7 vs. 245.9, p = 0.043), USMLE Step 2 score (256.7 vs. 253.2, p = 0.009), class quartile rank (p = 0.036), number of clerkships honored (4.4 vs. 3.5, p = 0.005), number of abstracts/posters/presentations (9.0 vs. 7.9, p = 0.007), number of peer reviewed publications (6.6 vs. 5.4, p = 0.007), and number of interviews attended (14.0 vs. 9.5, p = 0.000). Matched applicants also had a significantly higher percentage of applicants who honored their plastic surgery sub-internship (p = 0.032) and/or took a research year (p = 0.043). Differences in AOA status, GHHS status, couples match status, secondary degrees, number of research experiences, number of volunteering experiences and number of leadership positions were insignificant. Matched and unmatched applicants most frequently mentioned away rotations, letters of recommendation, connections, and mentors as important. Research was next most frequently mentioned. The most common advice from matched applicants was to strategically apply to away rotations they would potentially attend or where is a good fit based on their application and apply broadly to residency programs. For virtual interviews, matched applicants most frequently noted good lighting as important.

Conclusion: Objective factors like number of publications or presentations, board scores, and clinical grades play a significant role in applicant success. Subjective factors include strategy in applying to away rotations, letters of recommendation, mentorship, connections, and research. Important tips include applying to a broad range of residency programs and having good lighting in interviews.

Background: Prior literature has demonstrated that 8% of unplanned 30-day reoperations following posterior spinal surgery are attributed to fluid collections such as hematoma and seroma.(1) Given this high rate of reoperation it is common for surgeons to place drains to decrease the likelihood of associated complications.(2) However, prolonged indwelling time carries a unique infection risk that must be balanced.(3) While the time of drain removal is surgeon dependent, all rely upon accurate documentation of drain outputs, which is commonly performed using paper recording sheets.(4) This form of recording may have inherent limitations related to ease of use, legibility, and availability at postoperative follow-up.(5) Given these barriers, this study sought to determine whether recording drain output via a cellphone application would lead to improved adherence to drain output monitoring as well as earlier removal of drains. Methods: All pediatric patients who underwent plastic surgery closure by the senior author after posterior spinal surgery from April 2023 to February 2024 were randomized to record their drain output via the traditional paper sheet or a specialized smartphone application. The parents of these patients were instructed to record the output daily from the day of discharge until drain removal. Patient demographics were retrospectively reviewed and outcomes of interest included whether the record was brought to the first postoperative visit, and whether drain outputs were recorded daily. Outcomes of interest with regard to drain output included the number days indwelling and number of days the drain output was less than 30 ml. Additionally, a 1-question Likert survey (0-10) assessing satisfaction with the designated recording method was sent to parents following drain removal. All statistical tests were conducted using John Macintosh Project (JMP) software version 17 with significance defined as p < 0.05. Results: Following exclusion of patients who had their drains removed prior to discharge from the hospital, 64 pediatric patients were prospectively enrolled. At the initial postoperative visit it was determined that 7% of patients in the smartphone application group and 14% of patients in the paper method group did not record their drain output (p = 0.30). There was no significant statistical difference between groups with regard to the number of patients who had output logged for all post-discharge days (p = 0.21). The survey did demonstrate greater satisfaction in patients who used the smartphone application for ease of recording output (p < 0.01). Regarding drain removal timing, there were no statistically significant differences in total drain indwelling time or when available, days with output less than 30 ml between the app and paper groups. Conclusion: The utilization of smartphone applications for recording drain outputs may provide improved clinical information to guide surgeon decision making by offering greater patient convenience. However, our findings of similar drain removal timing suggest further developments are needed for a clinically meaningful impact. Future applications should consider medical record integration that would allow providers to track output records in real time and thereby allow for more timely drain removal while balancing prevention of potential fluid collections and their associated complications.

References: 1. Butler, A. J., Mohile, N., & Phillips, F. M. (2023). Postoperative spinal hematoma and seroma. JAAOS-Journal of the American Academy of Orthopaedic Surgeons, 31(17), 908-913. 2. Pennington, Zach, et al. "Prolonged Post-Surgical Drain Retention Increases Risk for Deep Wound Infection after Spine Surgery." World Neurosurgery, vol. 130, 2019, https://doi.org/10.1016/j.wneu.2019.07.013. 3. Yu, J., Dumaine, A. M., Poe-Kochert, C., Thompson, G. H., & Mistovich, R. J. (2020). Seromas following pediatric spinal deformity surgery: is operative management necessary?. Journal of Pediatric Orthopaedics, 40(4), e277-e282. 4. Janis, J. E., Khansa, L., & Khansa, I. (2016). Strategies for postoperative seroma prevention. Plastic and Reconstructive Surgery, 138(1), 240–252. https://doi.org/10.1097/prs.0000000000002245 5. Lyons, Nicholas, et al. "Improving the Recording of Surgical Drain Output." BMJ Quality Improvement Reports, vol. 4, no. 1, 2015, https://doi.org/10.1136/bmjquality.u209264.w3964.

Purpose: Social media platforms, such as TikTok, have great potential to influence a user's attitudes towards certain behaviors. The purpose of this study is to analyze TikTok users' reactions to content featuring tanning bed use. Videos that either promote or discourage the use of tanning beds were analyzed to see if users have an overall positive or negative response. Further, this study explores what specific content of these videos tends to garner a positive or negative reaction by commenters, such as photos of skin cancer lesions or glamorizing tan skin.

Methods: The hashtags #tanningbed and #sunbed were searched on TikTok. The top first 100 videos for each of these hashtags were analyzed. Data for each video was collected, including engagement data (likes, comments, shares) and the content of the video (promoting tanning, against tanning, humor, fear, tanning beds, exposed skin). The first 500 comments per video were extracted to R using the TKCommentExport extension. A VADER sentiment analysis was performed on the comments, categorizing the overall user response to each video as positive, negative, or neutral. After eliminating stop words, the top 10 words among all comments posted on all videos were collected. Videos were analyzed to assess video characteristics, sentiments, and interactions (comments, views, likes and shares) to compare VADER scores and interactions in each category.

Results: Videos that showed active cancer lesions or a story about surviving skin cancer had greater numbers of likes and comments (p< 0.05). Videos that instilled fear and humor both had higher viewer interaction than those that did not (both p< 0.05). Videos where creators were against tanning had a higher number of comments, views, likes and shares as compared to creators disclosing positive opinions on tanning (all p<0.05 respectively). Whether the video was promoting or discouraging tanning bed use was not significantly associated with a positive or negative reaction in the comments by TikTok users (all p>0.05), however the most-viewed and most-liked videos were associated with negative comments (p=0.001, p=0.02, respectively).

Conclusion: This information provides a better understanding of viewer interactions with tanning bed content on TikTok. The data can be applied to develop influential social media campaigns and public health initiatives better tailored to educate users on the risks of tanning beds use. The methods of this study can be expanded to analyze the efficacy and impact of other public health campaigns via TikTok. Also, as new machine learning technology develops, future studies can use natural language processing beyond sentiment analysis to characterize the reaction of social media users to social media content.

PURPOSE – Pressure ulcers are a frequent complication of surgeries and hospitalizations, becoming exceedingly costly and debilitating for both patients and hospitals. Previous research has shown that the prevention of these ulcers is a much more effective solution than care after ulcer development. Current options available for ulcer prevention include many adjuvant pressure-offloading devices such as beds and mattress pads, but these expensive tools can be incredibly cost ineffective for prophylaxis: research has shown current risk stratification methods fail to detect 60% of patients who go on to develop an ulcer. A lower-cost option is the manual repositioning of patients at set time intervals but to do so efficiently would require a more accurate tool to objectively assess risk in each patient. The purpose of this study was to establish feasibility of the Mercury Patch: a cost-effective, wearable, wireless device for monitoring sacral pressure in a clinical setting.

METHODS – Patients scheduled to undergo surgery for at least 4 hours were approached for informed consent. The Mercury Patch is placed on top of an Allevyn foam dressing and held in place using Tegaderm. In the operating room, the Mercury Patch and dressing was applied to the sacral skin when patients were transferred to the operating table. An accelerometer was then placed on the side of the operating table in line with the sacrum to record changes in bed position. An iPhone was used to wirelessly record pressure and temperature data from the Mercury Patch as well as positional data from the accelerometer during the case. Patients' lower backs were photographed to document skin integrity prior to device placement, at the conclusion of the operative case when the device was removed, and at the participants' first postoperative clinic visit. The pressure and positional data during each case was subsequently plotted and analyzed.

RESULTS – The Mercury Patch was successfully utilized in 11 patients' surgeries, recording sacral pressure throughout the full duration of each case without disrupting the flow of clinical care. The average pressure across all sensors was 16.7 mm Hg for deep inferior epigastric perforator flap cases, 28.1 mm Hg for breast reductions, and 36.9 mm Hg for mastectomies. None of the 11 patients developed a sacral wound or pressure injury. The patch was also able to successfully record the minimization of pressure changes during microsurgery.

CONCLUSIONS – This study demonstrated the feasibility of the Mercury Patch for real-time monitoring of sacral interface pressure in a clinical setting. The device itself did not result in pressure injury. Further studies are needed in higher-risk patients in order to determine pressure values and durations that are concerning for pressure injury formation.

Background Becoming a physician can be a challenging and stress evoking journey, and completing residency may be the most difficult portion. Emotional exhaustion is a well-documented challenge for residents. Compared to the public they frequently report higher levels of stress, anxiety, and depression. Poor resident well-being has also been linked with worse patient outcomes. For these reasons resident mental health is a vital concern. Emotional intelligence (EI) is a rapidly growing topic in the medical community because of the correlation with improved mental health. There are a limited number of studies assessing EI levels between residency specialties. Therefore, the aim of this study was to determine differences in EI between residency specialties and determine the correlation with stress, anxiety, and depression. A secondary aim of this study was to determine significant associations between resident demographic factors and EI, stress, anxiety, and depression.

Methods A prospective survey study was performed. All residents from a single tertiary academic medical center were invited to participate in the study. In total the survey was 60 questions and consisted of resident demographic information, the Schutte self-report emotional intelligence test, the perceived stress scale, and the patient health questionnaire 4 (which screens for anxiety and depression). Resident work hours and time dedicated to work-related tasks at home were also included. Parametric comparison tests were used to assess differences in EI, stress, and anxiety/depression scores. Linear regression analysis was used to determine associations between demographic and work hour variables.

Results There were 146 completed surveys with a response rate of 34%. Residents from 19 specialties responded with 35% being from a surgical specialty. The mean SSEIT score for the resident cohort was 118.6±10.8. Overall, no between group differences were found, however when specialties were separated into upper, middle, and lower terciles a significant difference was found (123.8 vs 119.1 vs 115.0, P < .001). No significant difference in EI was noted between surgical and non-surgical specialties, however surgical specialties reported higher perceived stress (22.7 vs 21.4, P = .006). Residents further along in training were associated with higher EI scores (P = .024). On linear regression analysis, EI was negatively correlated with anxiety/depression (r = -.32, P < .001), and weakly correlated with perceived stress (r = .17, P = .042). Higher resident working hours were associated with higher anxiety/depression scores (P < .001).

Conclusion At a single tertiary academic medical center, residency specialty was found to be a significant factor in EI, stress, and anxiety/depression scores. Emotional intelligence was found to be protective against anxiety/depression. Residency specialties with lower levels EI should participate in EI training to decrease the risk of poor mental health and well-being. Higher reported residency work hours were a significant contributor to anxiety/depression, emphasizing the importance for programs to monitor and manage these hours to mitigate the associated risks that arise from these mental health issues.