Citations

Abstract Objective The fast accumulation of vast pharmacogenomics data of cancer cell lines provide unprecedented opportunities for drug sensitivity prediction (DSP), a crucial prerequisite for the advancement of precision oncology. Recently, Generative Large Language Models (LLM) have demonstrated performance and generalization prowess across diverse tasks in the field of natural language processing (NLP). However, the structured format of the pharmacogenomics data poses challenge for the utility of LLM in DSP. Therefore, the objective of this study is multi-fold: to adapt prompt engineering for structured pharmacogenomics data toward optimizing LLM’s DSP performance, to evaluate LLM’s generalization in real-world DSP scenarios, and to compare LLM’s DSP performance against that of state-of-the-science baselines. Methods We systematically investigated the capability of the Generative Pre-trained Transformer (GPT) as a DSP model on four publicly available benchmark pharmacogenomics datasets, which are stratified by five cancer tissue types of cell lines and encompass both oncology and non-oncology drugs. Essentially, the predictive landscape of GPT is assessed for effectiveness on the DSP task via four learning paradigms: zero-shot learning, few-shot learning, fine-tuning and clustering pretrained embeddings. To facilitate GPT in seamlessly processing the structured pharmacogenomics data, domain-specific novel prompt engineering is employed by implementing three prompt templates (i.e., Instruction, Instruction-Prefix, Cloze) and integrating pharmacogenomics-related features into the prompt. We validated GPT’s performance in diverse real-world DSP scenarios: cross-tissue generalization, blind tests, and analyses of drug-pathway associations and top sensitive/resistant cell lines. Furthermore, we conducted a comparative evaluation of GPT against multiple Transformer-based pretrained models and existing DSP baselines. Results Extensive experiments on the pharmacogenomics datasets across the five tissue cohorts demonstrate that fine-tuning GPT yields the best DSP performance (28% F1 increase, p-value= 0.0003) followed by clustering pretrained GPT embeddings (26% F1 increase, p-value= 0.0005), outperforming GPT in-context learning (i.e., few-shot). However, GPT in the zero-shot setting had a big F1 gap, resulting in the worst performance. Within the scope of prompt engineering, performance enhancement was achieved by directly instructing GPT about the DSP task and resorting to a concise context format (i.e., instruction-prefix), leading to F1 performance gain of 22% (p-value=0.02); while incorporation of drug-cell line prompt context derived from genomics and/or molecular features further boosted F1 score by 2%. Compared to state-of-the-science DSP baselines, GPT significantly asserted superior mean F1 performance (16% gain, p-value<0.05) on the GDSC dataset. In the crosstissue analysis, GPT showcased comparable generalizability to the within-tissue performances on the GDSC and PRISM datasets, while statistically significant F1 performance improvements on the CCLE (8%, p-value=0.001) and DrugComb (19%, p-value=0.009) datasets. Evaluation on the challenging blind tests suggests GPT’s competitiveness on the CCLE and DrugComb datasets compared to random splitting. Furthermore, analyses of the drug-pathway associations and log probabilities provided valuable insights that align with previous DSP findings. Conclusion The diverse experiment setups and in-depth analysis underscore the importance of generative LLM, such as GPT, as a viable in silico approach to guide precision oncology. Availability https://github.com/bioIKEA/SensitiveCancerGPT

Migraine with aura(MwA) is associated with an increased risk of stroke and adverse vascular outcomes compared to those with migraine without aura (MwoA). AI-ECG prediction models developed at our institution can evaluate the probability of atrial fibrillation (AF) and estimate a patient’s age based on a normal sinus rhythm (NSR) ECG. Delta age, AI-ECG estimated age minus chronological age, indicates endothelial dysfunction and is associated with cardiovascular mortality. We aim to assess how AI-ECG prediction model outputs, specifically the AF probability and delta age, are associated with adverse vascular outcomes in patients with migraine. Adult patients diagnosed with MwA and MwoA from 2000-2020 with at least one digital, standard 12-lead ECG were identified. The first ECG that showed NSR is used as the index ECG. Patients with any adverse vascular outcomes before the index ECG were excluded. Adverse vascular outcomes used as endpoints include acute ischemic stroke, acute myocardial infarction, deep vein thrombosis/pulmonary embolism, AF, and carotid artery dissection. A total of 31301 patients (13783 MwA, 17518 MwoA) were included in our analysis. The mean age at the time of the index ECG was 44.3 (14.5), and the average follow-up time was 71.6 (69.8) months. Based on the index ECG, the mean AF probability was significantly higher in MwA compared to MwoA (2.39 vs 2.06, p< 0.001), while the delta age was lower in MwA compared to MwoA (3.7 vs 4.2, p <.001). The prevalences of all vascular outcomes are summarized in Table 1. Multivariate Cox regression models showed that a higher AF prediction model output (every 10% increase), as well as a higher delta age (every 10-year increase), are associated with an increased risk of composite vascular outcomes (HR1.20, 95%CI 1.17-1.23, p<.001, and HR1.19, 95%CI 1.14-1.23, p < .001) [Table 2]. The Kaplan-Meier curve of adverse vascular outcome-free survival by AF prediction model output, using 1% as the cutoff (HR 1.49, 95% CI 1.41, 1.57, p<.001), and by delta age, using 2-year as the cutoff (HR 1.14, 95% CI 1.08, 1.21, p <. 001) are presented in Figure 1. Having aura, older age, male, coronary artery disease, congestive heart failure, diabetes, hypertension, and tobacco use were also associated with a higher risk of adverse vascular outcomes. Our results demonstrated the potential of using AI-ECG AF prediction model output and delta age to identify migraine patients at risk for adverse vascular outcomes.

PURPOSE The RECIST guidelines provide a standardized approach for evaluating the response of cancer to treatment, allowing for consistent comparison of treatment efficacy across different therapies and patients. However, collecting such information from electronic health records manually can be extremely labor-intensive and time-consuming because of the complexity and volume of clinical notes. The aim of this study is to apply natural language processing (NLP) techniques to automate this process, minimizing manual data collection efforts, and improving the consistency and reliability of the results. METHODS We proposed a complex, hybrid NLP system that automates the process of extracting, linking, and summarizing anticancer therapy and associated RECIST-like responses from narrative clinical text. The system consists of multiple machine learning–/deep learning–based and rule-based modules for diverse NLP tasks such as named entity recognition, assertion classification, relation extraction, and text normalization, to address different challenges associated with anticancer therapy and response information extraction. We then evaluated the system performances on two independent test sets from different institutions to demonstrate its effectiveness and generalizability. RESULTS The system used domain-specific language models, BioBERT and BioClinicalBERT, for high-performance therapy mentions identification and RECIST responses extraction and categorization. The best-performing model achieved a 0.66 score in linking therapy and RECIST response mentions, with end-to-end performance peaking at 0.74 after relation normalization, indicating substantial efficacy with room for improvement. CONCLUSION We developed, implemented, and tested an information extraction system from clinical notes for cancer treatment and efficacy assessment information. We expect this system will support future cancer research, particularly oncologic studies that focus on efficiently assessing the effectiveness and reliability of cancer therapeutics.

11118 Background: Lung cancer survivors experience more physical and mental challenges, heavier symptom burden, and poorer quality of life than survivors of other cancers. Many symptoms persist or worsen for survivors beyond 5 years of cancer diagnosis, referred as long-term lung cancer survivors (LTLCS). Social determinants of health (SDoH) factors negatively impact patient reported outcomes (PROs) among cancer survivors, but their role in outcomes is under reported in lung cancer survivors and unknown in LTLCS. We investigated the impact of SDoH on PROs in a 20-year prospective lung cancer cohort. Methods: Our cohort includes 19,251 lung cancer patients (cases) diagnosed from 1997 through 2016 and followed up to 27 years, of whom 4799 (25%) became LTLCS. SDoH information was obtained by a lung cancer study PRO questionnaire or an institution SDoH questionnaire. Analyzed in this report are 2982 cases and 480 controls who had a low-dose chest CT scan and were ruled out for cancer, all answered at least one PRO questionnaire (first response); and 1366 LTLCS answered 5 years after diagnosis (long-term response). Along with fatigue and dyspnea, 5 SDoH variables, i.e., financial concerns, legal concerns, family and friends support, spiritual well-being, and social activity were scored on a 0 (worst) - 10 (best) scale and dichotomized as 0-5 (deficit/worse/lower) or 6-10 (no deficit/better/higher). We also assessed employment and health-related work experience. A difference of 1.0 unit or more in score was considered clinically meaningful along with a 2-sided test P-value less than 0.05; Chi-square or Fisher’s Exact Test and Kruskal-Wallis test were used accordingly. Results: Compared to controls, cases were younger (mean age: 62 vs. 65 years, P<0.05); more women than men reported worse financial concerns (56% vs. 38%, P<0.01) and lower spiritual well-being (43% vs. 26%, P=0.04) at first response; this sex disparity reduced but remained significant at long-term. Worse dyspnea (4.4-4.8 vs. 5.6) and fatigue (4.7-4.8 vs. 5.8-5.9) were seen in cases with more financial and legal concerns compared to those with fewer concerns (Ps<0.01). Cases with higher family and friends support, spiritual well-being and social activity had less dyspnea (5.7-6.0 vs. 4.2-4.5) and fatigue (5.5-5.9 vs. 3.9-4.4) than those with lower scores (Ps<0.04). Employment and health-related work experience data were provided by 2213 responders: Comparing 1735 cases to 478 controls, a lower proportion of cases was employed (28% vs. 39%), more cases “often felt tired after work” (49% vs. 28%), and more cases had “quite a lot” of difficulty in daily work or could not work at all” (26% vs. 9%, Ps<0.01). Conclusions: We found that in lung cancer survivors, more women than men reported worse financial concerns, and lower SDoH status was correlated with worse dyspnea and fatigue, indicating SDoH-driven interventions are important to improve lung cancer survivors’ PROs.

Background Analysis of the real-world data of low-dose spiral CT screening of healthy physical examination population in Southwest China from 2006 to 2017, discussed the standard of low-dose spiral CT screening for high-risk population of lung cancer in China. Methods Incorporate all team subjects who have undergone LDCT screenings at the Health Management Center of West China Hospital of Sichuan University between 2006 and 2017, and analyze the cumulative incidence and age-standardized (Segi population) incidence of lung cancer in the whole population; finally, analysis the age-specific, smoking and non-smokers’ lung cancer incidence. Results A total of 15,996 people who met the enrollment conditions and participated in the LDCT lung cancer screening were included. The average screening interval was 1.12±0.26 years, and the average follow-up time was 5.43±0.82 years; The cumulative number of lung cancer cases with negative baseline screening was 56 cases, and the cumulative incidence rate of lung cancer was 122.1/100,000 (the world standard rate is 60.6/100,000), of which males were 100.0/100,000 (the world standard rate is 45.3/100,000) and women were 156.6 /100,000 (the world standard rate is 94.7/100,000); the cumulative incidence of lung cancer in the smoking group is higher than that in the non-smoking group for both males and females, but the cumulative incidence of lung cancer in the non-smokers of the female population (146.3/100,000) is higher than that of the male population (86.5/10 Million). Conclusion In the real world study of lung cancer screening in Chinese population, we should consider setting different standards for high-risk groups of lung cancer by gender or smoking and non-smokers.

Abstract Background: Clinical, molecular, and genetic epidemiology studies displayed remarkable differences between ever- and never-smoking lung cancer. Methods: We conducted a stratified multi-population (European, East Asian, and African descent) association study on 44,823 ever-smokers and 20,074 never-smokers to identify novel variants that were missed in the non-stratified analysis. Functional analysis including expression quantitative trait loci (eQTL) colocalization and DNA damage assays, and annotation studies were conducted to evaluate the functional roles of the variants. We further evaluated the impact of smoking quantity on lung cancer risk for the variants associated with ever-smoking lung cancer. Results: Five novel independent loci, GABRA4, intergenic region 12q24.33, LRRC4C, LINC01088, and LCNL1 were identified with the association at two or three populations (P &lt; 5 × 10−8). Further functional analysis provided multiple lines of evidence suggesting the variants affect lung cancer risk through excessive DNA damage (GABRA4) or cis-regulation of gene expression (LCNL1). The risk of variants from 12 independent regions, including the well-known CHRNA5, associated with ever-smoking lung cancer was evaluated for never-smokers, light-smokers (packyear ≤ 20), and moderate-to-heavy-smokers (packyear &gt; 20). Different risk patterns were observed for the variants among the different groups by smoking behavior. Conclusions: We identified novel variants associated with lung cancer in only ever- or never-smoking groups that were missed by prior main-effect association studies. Impact: Our study highlights the genetic heterogeneity between ever- and never-smoking lung cancer and provides etiologic insights into the complicated genetic architecture of this deadly cancer.

AbstractBackgroundRecent therapeutic advances and screening technologies have improved survival among patients with lung cancer, who are now at high risk of developing second primary lung cancer (SPLC). Recently, an SPLC risk‐prediction model (called SPLC‐RAT) was developed and validated using data from population‐based epidemiological cohorts and clinical trials, but real‐world validation has been lacking. The predictive performance of SPLC‐RAT was evaluated in a hospital‐based cohort of lung cancer survivors.MethodsThe authors analyzed data from 8448 ever‐smoking patients diagnosed with initial primary lung cancer (IPLC) in 1997–2006 at Mayo Clinic, with each patient followed for SPLC through 2018. The predictive performance of SPLC‐RAT and further explored the potential of improving SPLC detection through risk model‐based surveillance using SPLC‐RAT versus existing clinical surveillance guidelines.ResultsOf 8448 IPLC patients, 483 (5.7%) developed SPLC over 26,470 person‐years. The application of SPLC‐RAT showed high discrimination area under the receiver operating characteristics curve: 0.81). When the cohort was stratified by a 10‐year risk threshold of ≥5.6% (i.e., 80th percentile from the SPLC‐RAT development cohort), the observed SPLC incidence was significantly elevated in the high‐risk versus low‐risk subgroup (13.1% vs. 1.1%, p < 1 × 10–6). The risk‐based surveillance through SPLC‐RAT (≥5.6% threshold) outperformed the National Comprehensive Cancer Network guidelines with higher sensitivity (86.4% vs. 79.4%) and specificity (38.9% vs. 30.4%) and required 20% fewer computed tomography follow‐ups needed to detect one SPLC (162 vs. 202).ConclusionIn a large, hospital‐based cohort, the authors validated the predictive performance of SPLC‐RAT in identifying high‐risk survivors of SPLC and showed its potential to improve SPLC detection through risk‐based surveillance.Plain Language Summary Lung cancer survivors have a high risk of developing second primary lung cancer (SPLC). However, no evidence‐based guidelines for SPLC surveillance are available for lung cancer survivors. Recently, an SPLC risk‐prediction model was developed and validated using data from population‐based epidemiological cohorts and clinical trials, but real‐world validation has been lacking. Using a large, real‐world cohort of lung cancer survivors, we showed the high predictive accuracy and risk‐stratification ability of the SPLC risk‐prediction model. Furthermore, we demonstrated the potential to enhance efficiency in detecting SPLC using risk model‐based surveillance strategies compared to the existing consensus‐based clinical guidelines, including the National Comprehensive Cancer Network.

e16326 Background: Cholelithiasis or Gallstone disease (GSD), is one of the most common and costly digestive disorders in adults worldwide, presenting 10-20% of the general population but in greater than 30% of Latina and Hispanic American women aged 40 years or older. Moreover, GSD individuals contribute significantly to the incidence and mortality of gallbladder and other digestive system cancers; specifically, the excessive risk of gallbladder cancer is 3-10-fold. Parallel to gallbladder cancer, an appalling female preponderance of GSD has been documented for decades; thus, precisely defining high-risk populations and properly managing patients with GSD would lead to reduced disease burden from GSD and multiple cancers. Several genome-wide association studies (GWAS) in the past have identified genes associated with GSD but few were focused on Latinx population. Methods: To identify genetic risk factors for GSD in Latinx population, we used the Sangre Por Salud cohort (SPS) where participants are Latinx adult individuals aged 18-85 years who were current patients in the Mountain Park Health Center in Arizona. This sample represents historically underrepresented population with low socioeconomic status that are primarily Spanish-speaking, majority being born outside of the US, with low rates of health insurance coverage. A total of 172 GSD cases were identified, each matched with 6 controls (n = 1032) who were people from the same community but without history of GSD diagnosis; matching variables are age at enrollment to SPS, sex, height and weight, dyslipidemia, and HbA1C level indicative of type 2 diabetes. We characterized genome-wide interrogation for biomarkers predicting GSD and contributing to the observed gender disparity by performing a GWAS using a matched case-control design. Results: We identified several novel loci associated with GSD and validated few loci that were reported in previous GWAS. The top 3 loci (MATN2, GPRIN3, GPC6) were strongly associated with GSD in our overall analysis; follow-up pathway enrichment analysis suggests enrichment of GO terms that are associated with immunological pathways, enzymatic processes in gallbladder, liver and gastrointestinal tract and GSD pathology. In a sex-specific analysis, CDHR3 and RPL31P35-SEC61G were identified for females, which have been associated respectively with rhinovirus C infection and HbA1c level in previous studies. Conclusions: Our findings suggest directions towards better and deeper understanding of sex differences in GSD pathology, biological mechanisms, and disease progression among Latinx people. Knowing approximately 80% of gallbladder cancer patients harbored gallstones, our optimal goal is to provide evidence for an innovative strategy to early detect and effectively intervene the modifiable risk factors to reduce gallbladder cancer burden in high risk populations.