Validation of large language models for detecting pathologic complete response in breast cancer using population-based pathology reports.

AIMS: The primary goal of this study is to evaluate the capabilities of Large Language Models (LLMs) in understanding and processing complex medical documentation. We chose to focus on the identification of pathologic complete response (pCR) in narrative pathology reports. This approach aims to contribute to the advancement of comprehensive reporting, health research, and public health surveillance, thereby enhancing patient care and breast cancer management strategies. METHODS: The study utilized two analytical pipelines, developed with open-source LLMs within the healthcare system's computing environment. First, we extracted embeddings from pathology reports using 15 different transformer-based models and then employed logistic regression on these embeddings to classify the presence or absence of pCR. Secondly, we fine-tuned the Generative Pre-trained Transformer-2 (GPT-2) model by attaching a simple feed-forward neural network (FFNN) layer to improve the detection performance of pCR from pathology reports. RESULTS: In a cohort of 351 female breast cancer patients who underwent neoadjuvant chemotherapy (NAC) and subsequent surgery between 2010 and 2017 in Calgary, the optimized method displayed a sensitivity of 95.3% (95%CI: 84.0-100.0%), a positive predictive value of 90.9% (95%CI: 76.5-100.0%), and an F1 score of 93.0% (95%CI: 83.7-100.0%). The results, achieved through diverse LLM integration, surpassed traditional machine learning models, underscoring the potential of LLMs in clinical pathology information extraction. CONCLUSIONS: The study successfully demonstrates the efficacy of LLMs in interpreting and processing digital pathology data, particularly for determining pCR in breast cancer patients post-NAC. The superior performance of LLM-based pipelines over traditional models highlights their significant potential in extracting and analyzing key clinical data from narrative reports. While promising, these findings highlight the need for future external validation to confirm the reliability and broader applicability of these methods.

HIGH THROUGHPUT QUANTITATION OF HUMAN NEUTROPHIL RECRUITMENT AND FUNCTIONAL RESPONSES IN AN AIR-BLOOD BARRIER ARRAY.

Dysregulated neutrophil recruitment drives many pulmonary diseases, but most preclinical screening methods are unsuited to evaluate pulmonary neutrophilia, limiting progress towards therapeutics. Namely, high throughput therapeutic screening systems typically exclude critical neutrophilic pathophysiology, including blood-to-lung recruitment, dysfunctional activation, and resulting impacts on the air-blood barrier. To meet the conflicting demands of physiological complexity and high throughput, we developed an assay of 96-well Leukocyte recruitment in an Air-Blood Barrier Array (L-ABBA-96) that enables in vivo -like neutrophil recruitment compatible with downstream phenotyping by automated flow cytometry. We modeled acute respiratory distress syndrome (ARDS) with neutrophil recruitment to 20 ng/mL epithelial-side interleukin 8 (IL-8) and found a dose dependent reduction in recruitment with physiologic doses of baricitinib, a JAK1/2 inhibitor recently FDA-approved for severe COVID-19 ARDS. Additionally, neutrophil recruitment to patient-derived cystic fibrosis sputum supernatant induced disease-mimetic recruitment and activation of healthy donor neutrophils and upregulated endothelial e-selectin. Compared to 24-well assays, the L-ABBA-96 reduces required patient sample volumes by 25 times per well and quadruples throughput per plate. Compared to microfluidic assays, the L-ABBA-96 recruits two orders of magnitude more neutrophils per well, enabling downstream flow cytometry and other standard biochemical assays. This novel pairing of high-throughput in vitro modeling of organ-level lung function with parallel high-throughput leukocyte phenotyping substantially advances opportunities for pathophysiological studies, personalized medicine, and drug testing applications.

Antiretroviral treatment outcome following genotyping in Thai children who failed dual nucleoside reverse transcriptase inhibitors.

OBJECTIVE: To evaluate outcomes in dual nucleoside reverse transcriptase inhibitor (NRTI) pretreated children after genotyping (GT). METHODS: We assessed CD4 and viral load (VL) in children three years after baseline GT at the time of dual NRTI failure. Baseline high grade resistance (HR) was defined as >or=4 nucleoside analogue mutations (NAMs)+/-Q151M or 69 insertion complex, and low grade resistance (LR) was defined as <4 NAMs. Genotypic susceptibility scores (GSS) were determined. The current selection of antiretrovirals (ARV) was based on physician judgment and ARV availability. RESULTS: Seventy-two children were enrolled, with a mean age of 9.3 years; 61% were female. Baseline median CD4 was 18%, VL was 1.7 log(10) with HR 37.5%, LR 56.9% and no mutation (NR, no resistance) 5.6%. Sixty-five (90.3%) switched ARV: 46.2% non-nucleoside reverse transcriptase inhibitor (NNRTI), 30.8% protease inhibitor (PI), and 23.1% PI+NNRTI based highly active antiretroviral therapy (HAART). The choice of regimen did not differ based on baseline HR, LR, and NR. The median duration from dual NRTI therapy to HAART was 5.4 years (interquartile range (IQR) 4.0-6.9 years) and the mean (SD) duration of current HAART regimen was 1.51 (1.78) years; both were similar between ARV groups. Five children continued dual NRTI, two interrupted therapy. The GSS score was significantly higher in the PI group (3.1) vs. PI+NNRTI (2.5) vs. NNRTI (2.6) groups. Sixty-three percent of the HR group used PI or PI+NNRTI-based HAART compared to 41% of the LR group, p=not significant. At follow-up, median CD4 changes from baseline were +5% and VL -2.2 log(10) (p<0.001). VL <1.7 log(10) was seen in 59.3% of HR, 58.5% of LR, and 50.0% of NR groups (no significant difference). More children on PI (75%) and PI+NNRTI (80%) based HAART had VL <50 compared to NNRTI-based HAART (50%), p=0.003. CONCLUSION: PI-based regimens showed a higher rate of undetectable VL compared with NNRTI-based regimens. Having GT may not affect second-line treatment choices in developing countries, most likely due to late VL failure and limited availability of PIs.