Clinical AI systems' lack of interpretability limits their adoption in evidence-based medicine. To address this challenge, we propose a computational framework that harnesses generative AI's medical knowledge to create interpretable structural causal models (SCMs) for clinical decision support, quality improvement evaluation, and population health management. We evaluated our approach through a case study using data from the Midwest Healthcare Conference Causal Diagram Challenge, where we compared transformer-based large language models against human performance on a complex causal reasoning task: estimating COVID- 19 treatment effects through target trial emulation. Both groups designed SCMs to evaluate glucocorticoid treatment effects on 28-day mortality using real-world data from more than 2,000 hospitalized patients, benchmarked against published RECOVERY randomized controlled trial results. The best performing SCMs achieved bootstrap coverage rates exceeding 90% for two of three severity strata. Both human and AI models demonstrated equivalent clinical plausibility (n=3 expert reviewers) and similar statistical performance, though both struggled with critical disease severity. Ablation experiments comparing SCM-based approaches against traditional potential outcomes methods revealed SCMs achieved 76-98% coverage versus 1-37% for traditional methods. These results suggest that structural causal models can effectively bridge the interpretability gap in clinical AI by providing essential scaffolding for reliable causal inference and enabling meaningful human-AI collaboration while preserving methodological rigor essential for evidence-based medicine.
Introduction and Objective: Glucagon has underappreciated effects on lipid, glucose, and amino acid metabolism. This study aimed to assess the comprehensive metabolic effects of glucagon antagonism in type 1 diabetes (T1D). Methods: 30 adults with T1D were randomized 1:1 to receive the glucagon receptor antagonist (GRA) volagidemab via once weekly subq injection for 12 weeks vs. placebo. A 2-stage hyperinsulinemic-euglycemic clamp (8 and 40 mU/m2/min) with indirect calorimetry was conducted at baseline and after 12-weeks to determine changes in insulin sensitivity and substrate oxidation. Results: GRA therapy decreased exogenous insulin use by 16% while maintaining glucose control (no Δ in A1c or CGM metrics). Compared to baseline, GRA decreased circulating FFA concentrations by 30% in the fasting state and 39% in the 1st stage of the insulin clamp, suggesting an increase in adipose tissue insulin sensitivity (p < 0.05). During the high dose step of the clamp, which is representative of skeletal muscle insulin action, glucose disposal increased by 33% (p = 0.052) and respiratory quotient by indirect calorimetry increased by 5% (p = 0.013). Thus, GRA therapy significantly increased glucose utilization and showed a strong trend toward improving skeletal muscle insulin sensitivity. Finally, bioimpedance data showed a significant increase in lean body mass of 1.5 kg after GRA therapy (p < 0.001) with no change in total body weight. As a potential explanation for this finding, GRA therapy significantly increased circulating amino acid concentrations (~2-fold) which may provide additional substrate for muscle tissue synthesis. Conclusion: GRA therapy decreased lipolysis, improved peripheral glucose disposal, increased glucose oxidation, and increased lean body mass. These data highlight the profound effects that glucagon action has on multiple aspects of metabolism that extend far beyond glucose control. S.C. Boeder: Consultant; Cecelia Health. Advisory Panel; Novo Nordisk. Consultant; Lexicon Pharmaceuticals, Inc, Persperion Diagnostics. Research Support; Eli Lilly and Company, Carmot Therapeutics, Inc, REMD Therapeutics, Dexcom, Inc., Lexicon Pharmaceuticals, Inc. R.L. Thomas: Research Support; REMD Therapeutics, Carmot Therapeutics. V. Hamidi: None. E.R. Giovannetti: Advisory Panel; Eli Lilly and Company, Sanofi. A. Armstrong: None. L. Carter: None. T.P. Ciaraldi: None. J.H. Pettus: None. Breakthrough T1D and the Helmsley Charitable Trust (3-SRA-2021-1066-M-B)
Early-onset colorectal cancer (EOCRC) incidence is rising in a predominantly symptomatic population of young adults. Effective triage tools are needed to identify high-risk individuals in this relatively low incidence population. We examined fecal immunochemical test (FIT) use among adults ages <50 with red flag signs and symptoms for EOCRC and evaluated whether FIT use is predictive of EOCRC risk. Retrospective cohort study of US Veterans (ages 18-49) receiving Veterans Health Administration (VHA) care during 1999-2022 with a documented EOCRC red flag sign or symptom (abdominal distension, abdominal pain, anemia [non-specific and iron-deficiency], change in bowel habits, constipation, diarrhea, hematochezia, nausea/vomiting) based on International Classification of Diseases, 9th (ICD-9) or 10th (ICD-10) Revision codes, or lab results. The primary exposure was FIT uptake and result, documented via lab results, shown as a three-level variable (no FIT use, negative FIT or positive FIT). The primary outcome was EOCRC diagnosis, derived from linkages to the VA Oncology Domain and National Death Index. Covariates included age at symptom presentation, sex, race and ethnicity, and number of symptoms within 60 days of first symptom presentation. Participants entered the study at first symptom onset and were followed until the first of: incident or fatal EOCRC diagnosis, non-EOCRC-related death, 2 years follow-up, age 50 or December 31, 2022. We derived cumulative CRC incidence estimates using Kaplan-Meier estimation. Multivariable, mixed-effects Cox regression models were used to estimate adjusted hazard ratios (aHR) and 95% confidence intervals (CI) for CRC risk among those who received a FIT test. Among 751,116 Veterans, 38,019 (5.1%) received a FIT. The most common symptoms yielding a FIT were abdominal pain (29.6%), anemia (20.4%), hematochezia (18.5%), and diarrhea (16.4%). Approximately 76% of patients who received a FIT had one symptom at presentation. Among 38,019 patients who received a FIT, 6,191 (16.3%) had a positive finding. Approximately 1,295 (21%) of 6,191 FIT-positive patients received a diagnostic colonoscopy compared to 46,693 (6.6%) of 713,097 non-FIT patients. After two years of follow-up, patients with a positive FIT had a 1.44% cumulative EOCRC incidence (95% CI: 1.12%-1.77%), compared to a 0.12% among those with a negative FIT (95% CI; 0.08%-0.16%) and 0.12% among those who did not receive a FIT (95% CI: 0.12%-0.13%). The findings correspond to an aHR for EOCRC of 12.81 (95% CI: 8.47-19.36) for FIT-positive patients compared to those with a negative FIT. Among adults ages 18-49 presenting with a red flag sign or symptom to VHA care, FIT use was low. However, a positive FIT result was linked to a substantially elevated EOCRC risk relative to a negative result. To address potential concerns about generalizability, future research should confirm whether systematic use of FIT as a clinical triage tool could help identify symptomatic adults at high EOCRC risk who need a diagnostic colonoscopy. Daniel Sabater Minarim, Kylie Morgan, Lin Liu, Matthew P. Banegas, Maria Elena Martinez, Samir Gupta, Josh Demb. FIT for red flag signs and symptoms of early onset colorectal cancer: low value or viable diagnostic tool? [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: The Rise in Early-Onset Cancers—Knowledge Gaps and Research Opportunities; 2025 Dec 10-13; Montreal, QC, Canada. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(23_Suppl):Abstract nr PR014.
Introduction and Objective: Circulating cell-free mitochondrial DNA (cf-mtDNA) activates innate immunity and has been implicated in animal models of diabetes. This study aimed to assess how mitochondrial DNA levels correlate with diabetes disease state and hospitalization in humans. Methods: Using droplet digital PCR (ddPCR), we quantified plasma cf-mtDNA concentrations in 52 adults classified into 4 groups: non-diabetes (Non-DM), obesity without diabetes (ObND), outpatient Type 2 Diabetes (T2D) and hospitalized T2D (T2D-Hosp). ObND group participants had BMI >30 and hyperinsulinemia without hyperglycemia (~2-fold fasting insulin vs. Non-DM, p = 0.028). Results: Compared to Non-DM controls, logarithmic analysis revealed a progressive increase in plasma cf-mtDNA concentrations in ObND (p = 0.042), T2D (p = 0.006), and T2D-Hosp groups (p <0.001) (Figure). Conclusion: Here we show that cf-mtDNA is elevated in people with ObND or T2D compared to Non-DM controls. This effect is exacerbated by physiologic stress, with an additional ~3-fold increase in cf-mtDNA during hospitalization. Early cf-mtDNA elevation is detectable in obese patients with hyperinsulinemia before significant hyperglycemia develops. Plasma cf-mtDNA may thus provide a biomarker for early diagnosis and clinical severity in T2D. R.L. Thomas: Research Support; REMD Therapeutics, Carmot Therapeutics. J.D. Garcia: None. A.J. Kumar: None. L. Carter: None. J.A. Masso-Silva: None. V. Hamidi: None. S.C. Boeder: Consultant; Cecelia Health. Advisory Panel; Novo Nordisk. Consultant; Lexicon Pharmaceuticals, Inc, Persperion Diagnostics. Research Support; Eli Lilly and Company, Carmot Therapeutics, Inc, REMD Therapeutics, Dexcom, Inc., Lexicon Pharmaceuticals, Inc. T.P. Ciaraldi: None. J.H. Pettus: None. M.L. Hepokoski: None. R.L. Thomas: Altman Clinical and Translational Research Institute Pilot Project National Institutes of Health, CTSA Grant (UL1TR0014422); National Institute of Diabetes and Digestive and Kidney Diseases (P30 DK063491); Foundation for the National Institutes of Health T32 (2T32DK007044-41); National NIH K12 DiabDocs Program (K12DK133995)
Introduction and Objective: The success of longitudinal studies depends on the retention of participants. We examined sociodemographic, clinical and psychosocial characteristics as predictors of retention among participants with prediabetes and type 2 diabetes (T2D) in the DPP/DPPOS. Methods: 3218 adults who joined the DPP (1996-1999, mean age 51±10y) at high risk of T2D were randomized to a lifestyle, metformin or placebo intervention, and followed in the DPPOS through 2020 with lifestyle offered to all and metformin continued open label. Logistic regression models estimated the association between baseline sociodemographic, clinical and psychosocial characteristics (life events, family functioning, social support) and short-term retention (~3y during DPP). Cox proportional hazards models, censoring at death, estimated the association between baseline and time-varying characteristics and time to drop-out over 20 years. Results: After 3 years, 93% of surviving participants were retained; 76% of the surviving cohort remained engaged over ~17y. Older age was associated with short-term study retention (p<0.001). Older age, female sex, minority race/ethnicity, full or part-time employment, and lack of depressive symptoms at baseline were associated with long-term retention. Over time, better health state (SF-6D, SF-36 survey) (0.31; CI: 0.15, 0.63) were associated with retention; greater BMI (1.12; CI 1.01, 1.25), higher number of recent life events (social, personal, financial) (1.08; CI: 1.02, 1.14) and depression symptoms (Beck Depression Index) (1.02, CI: 1.01, 1.03) were associated with slightly reduced retention. Among adults 45-59y at baseline, development of diabetes was associated with decreased retention (0.75, CI: 0.58, 0.97). Conclusion: Twenty-year retention of a racially and geographically diverse T2D-related cohort is possible. Retention may be influenced by age, psychosocial factors, diabetes development, and weight change. A.H. Tjaden: None. S. Golden: Advisory Panel; Abbott. N.M. Butera: None. M.G. Araneta: None. O. Carmichael: Research Support; Eli Lilly and Company. E. Groessl: Consultant; Fisher Paykel Health. H.P. Hazuda: None. M.A. Hoskin: None. U.N. Ibebuogu: None. M.F. Magee: Other Relationship; Lilly USA LLC. M. Mau: None. T. Stich: None. D. Soliman: None. A. Wallia: Research Support; UnitedHealth Group. B.H. Braffett: None. M. Temprosa: None. D. Research Group: None. National Institute of Diabetes and Digestive and Kidney Diseases of the National Institutes of Health (U01 DK048489, U01 DK048339, U01 DK048377, U01 DK048349, U01 DK048381, U01 DK048468, U01 DK048434, U01 DK048485, U01 DK048375, U01 DK048514, U01 DK048437, U01 DK048413, U01 DK048411, U01 DK048406, U01 DK048380, U01 DK048397, U01 DK048412, U01 DK048404, U01 DK048387, U01 DK048407, U01 DK048443, and U01 DK048400)
