Faculty Bibliography

PURPOSE:To evaluate risk factors associated with severe retinopathy of prematurity ROP (sROP) in two separate cohorts of infants. METHODS:We performed a retrospective study of the Merative MarketScan Commercial Database between 2003 and 2022. Infants with ROP were stratified into two cohorts: group A (defined as infants with both birth weight [BW] <1000 g and gestational age [GA] <29 weeks) and group B (defined as infants with either BW ≥1000 g or GA ≥29 weeks). Infants with sROP were defined as those who received treatment for ROP. Outpatient and inpatient claims using International Classification of Diseases and Current Procedural Terminology codes were used to identify neonatal comorbidities. Multivariable logistic regression was used to calculate odds ratios (ORs) and 95% confidence intervals (CIs) for sROP requiring treatment with respect to various comorbidities. RESULTS:A total of 425 of 8,789 infants with ROP developed sROP (4.8%). We identified 2,726 infants in group A, of whom 387 (14.2%) required a procedure, and 6,063 in group B, of whom 38 (0.6%) required a procedure. In group A, intraventricular hemorrhage (OR = 1.38; 95% CI, 1.09-1.74) and patent ductus arteriosus ligation (OR = 1.65; 95% CI, 1.25-2.16) were the comorbidities that significantly increased the odds of sROP on multivariable analysis. In group B, infection (OR = 1.96; 95% CI, 1.01-3.80) was the only comorbidity that significantly increased the adjusted odds of sROP. CONCLUSIONS:Risk factors for sROP may differ between smaller and larger infants. Infection may serve as an important risk factor for ROP progression amongst larger infants due to its connection with poor postnatal growth.

PURPOSE/UNASSIGNED:This is a case report of a 31-year-old male patient presenting progressive and markedly asymmetric keratoconus treated with sequential intracorneal ring segment (ICRS) implantation followed by accelerated corneal cross-linking (CXL). OBSERVATIONS/UNASSIGNED:post-CXL) was 19.1 D. CONCLUSIONS AND IMPORTANCE/UNASSIGNED:This original case reports how CXL followed by ICRS implant may result in an early and extreme corneal remodeling. Moreover, such an unusual combination of extreme corneal flattening, thinning, and opacification may imitate a clinical manifestation of central toxic keratopathy and suggests that eyes with ICRS implantation must be followed closely if CXL is performed sequentially after.

BACKGROUND:There are limited data to guide the diagnosis and management of vasa previa. Currently, what is known is largely based on case reports or series and cohort studies. OBJECTIVE:(s): To systematically collect and classify expert opinions and achieve consensus on the diagnosis and clinical management of vasa previa using focus group discussions (FGD) and a Delphi technique. STUDY DESIGN/METHODS:A four-round FGD and a three-round Delphi survey of an international panel of experts on vasa previa were conducted. Experts were selected based on their publication record on vasa previa. First, we convened an FGD panel of 20 experts and agreed on which issues were unresolved in the diagnosis and management of vasa previa. A three-round anonymous electronic survey was then sent to the full expert panel. Survey questions were presented on the diagnosis and management of vasa previa that the experts were asked to rate on a 5-point Likert scale (from strongly disagree = 1 to strongly agree = 5). Consensus was defined as a median score of 5. Following responses to each round, any statements that had median scores of 3 or less were deemed to have had no consensus and excluded. Statements with a median score of 4 were revised and re-presented to the experts in the next round. Consensus and non-consensus statements were then aggregated. RESULTS:Sixty-eight international experts were invited to participate in the study, of which 57 participated. Experts were from 13 countries on five continents and have contributed to over 80% of published cohort studies on vasa previa, as well as national and international society guidelines. Completion rates were 84%, 93%, 91% for the first, second, and third rounds, respectively, and 71% completed all three rounds. The panel reached a consensus on 26 statements regarding the diagnosis and key points of management of vasa previa, including: 1) While there is no agreement on a distance between the fetal vessels and the cervical internal os to define vasa previa, the definition should not be limited to a 2 cm distance; 2) All pregnancies should be screened for vasa previa with routine examination for placental cord insertion and a color Doppler sweep of the region over the cervix at the second-trimester anatomy scan; 3) When a low-lying placenta or placenta previa is found in the second trimester, a transvaginal ultrasound with Doppler should be performed at around 32 weeks to rule out vasa previa; 4) Outpatient management of asymptomatic patients without risk factors for preterm birth is reasonable; 5)Asymptomatic patients with vasa previa should be delivered by scheduled cesarean between 35- and 37-weeks of gestation; and 6) There was no agreement on routine hospitalization, avoidance of intercourse, or use of 3-dimensional ultrasound for diagnosis of vasa previa. CONCLUSIONS:Through FGD and a Delphi process, an international expert panel reached consensus on the definition, screening, clinical management, and timing of delivery in vasa previa, which could inform the development of new clinical guidelines.

Ensuring reliability of Large Language Models (LLMs) in clinical tasks is crucial. Our study assesses two state-of-the-art LLMs (ChatGPT and LlaMA-2) for extracting clinical information, focusing on cognitive tests like MMSE and CDR. Our data consisted of 135,307 clinical notes (Jan 12th, 2010 to May 24th, 2023) mentioning MMSE, CDR, or MoCA. After applying inclusion criteria 34,465 notes remained, of which 765 underwent ChatGPT (GPT-4) and LlaMA-2, and 22 experts reviewed the responses. ChatGPT successfully extracted MMSE and CDR instances with dates from 742 notes. We used 20 notes for fine-tuning and training the reviewers. The remaining 722 were assigned to reviewers, with 309 each assigned to two reviewers simultaneously. Inter-rater-agreement (Fleiss' Kappa), precision, recall, true/false negative rates, and accuracy were calculated. Our study follows TRIPOD reporting guidelines for model validation. For MMSE information extraction, ChatGPT (vs. LlaMA-2) achieved accuracy of 83% (vs. 66.4%), sensitivity of 89.7% (vs. 69.9%), true-negative rates of 96% (vs 60.0%), and precision of 82.7% (vs 62.2%). For CDR the results were lower overall, with accuracy of 87.1% (vs. 74.5%), sensitivity of 84.3% (vs. 39.7%), true-negative rates of 99.8% (98.4%), and precision of 48.3% (vs. 16.1%). We qualitatively evaluated the MMSE errors of ChatGPT and LlaMA-2 on double-reviewed notes. LlaMA-2 errors included 27 cases of total hallucination, 19 cases of reporting other scores instead of MMSE, 25 missed scores, and 23 cases of reporting only the wrong date. In comparison, ChatGPT's errors included only 3 cases of total hallucination, 17 cases of wrong test reported instead of MMSE, and 19 cases of reporting a wrong date. In this diagnostic/prognostic study of ChatGPT and LlaMA-2 for extracting cognitive exam dates and scores from clinical notes, ChatGPT exhibited high accuracy, with better performance compared to LlaMA-2. The use of LLMs could benefit dementia research and clinical care, by identifying eligible patients for treatments initialization or clinical trial enrollments. Rigorous evaluation of LLMs is crucial to understanding their capabilities and limitations.