Huang, Yuting

Abstract Objectives To comprehensively evaluate the validity of International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) codes for both prevalent diagnoses and less common diseases, and to assess the performance of a large language model (LLM)-based system in validating these codes. Materials and Methods This retrospective study analyzed hospital admissions from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. We developed a validated LLM-based system using GPT-4o, refined through iterative prompt engineering, to assess ICD-10-CM code validity. We measured the positive predictive value (PPV) of ICD-10-CM codes, PPV of principal and secondary diagnoses, and the performance of an LLM-based system in code validation. Results Among 865 079 assigned codes, the PPV was 84.6% (95% CI, 84.5%-84.6%). Principal diagnoses had a PPV of 93.9% (95% CI, 93.7%-94.1%), while secondary diagnoses had a PPV of 83.8% (95% CI, 83.7%-83.9%). The LLM system demonstrated high performance in validating ICD codes, achieving 93.6% accuracy, 95.4% sensitivity, and 85.2% specificity. Among correctly assigned secondary diagnoses, the majority (67.9%) represented historical or baseline conditions, while 32.1% reflected active conditions that deviated from baseline status; 22.3% of these emerged after hospital admission. PPV decreases with later diagnosis positions, with the largest decline occurring between principal and secondary diagnoses. Discussion and Conclusion In this large-scale evaluation, ICD-10-CM codes exhibited generally high accuracy, though variability existed by position and condition type. A validated LLM system performed comparably to physician review and offers a scalable means to improve coding accuracy. These findings support the potential for integrating LLM-based auditing into routine workflows to strengthen the quality of administrative and research data.

851 Background: Diffuse idiopathic skeletal hyperostosis (DISH) has been associated with increased risk of vertebral fractures in elderly men with prostate cancer. However, its potential role in the development of malignancies is poorly defined. In particular, the relationship between DISH and gastrointestinal (GI) cancers—including esophageal, gastric, colorectal, liver, biliary system, and pancreatic cancers—has not been examined. This study aimed to explore the risk of GI cancers among patients with DISH compared to those without DISH. We also assessed survival and incidence of vertebral fractures in these groups. Methods: Using the Mayo Clinic Platform (Full AMC1 dataset), we identified patients with and without a documented history of DISH. Outcomes included (1) prevalence and risk of GI cancers, (2) survival among GI cancer patients, and (3) vertebral fracture incidence. Relative risks (RRs) and hazard ratios (HRs) were calculated using SPSS20.0 and platform-based statistical tools, respectively. Kaplan–Meier survival analyses were performed to evaluate cancer survival outcomes. Results: A total of 6,189,860 patients were included, of whom 2,802 (0.05%) had DISH and 6,187,058 (99.95%) did not. Patients with DISH had a significantly higher prevalence of GI cancers compared with those without (RR = 1.63, 95% CI: 1.377–1.930). Survival analysis suggested worse outcomes among GI cancer patients without DISH (HR = 1.72, 95% CI: 1.25–2.35). Vertebral fracture incidence appeared lower among DISH patients, although low event numbers limit the reliability of this observation. Conclusions: DISH is associated with increased risk of GI cancers, while GI cancer patients without DISH may experience poorer survival. The observed trend toward lower vertebral fracture incidence among DISH patients should be interpreted cautiously due to limited event numbers. These findings highlight the importance of integrating musculoskeletal and oncologic perspectives in patient management and warrant further mechanistic and interventional studies. Survival analysis. Statistic Without DISH With DISH Patient Count 214,180 109 1 Month Survival 95.3% 98.2% 5 Month Survival 82.2% 87.9% 10 Month Survival 72.4% 85.1% 30 Month Survival 54.3% 74.1% 60 Month Survival 44.9% 60.6% Log-Rank 11.90 P-Value 0.00056 Hazard Ratio 1.72 (95% CI: 1.25 - 2.35)

331 Background: The relationship between plant-based dietary patterns and upper gastrointestinal (GI) cancers remains unclear. The Healthful Plant-Based Diet Index (hPDI) is a tool that measures adherence to plant-based diet and generates “scores” for clinical comparison. We conducted a systematic review and meta-analysis to examine the association between hPDI and the risk of esophageal and gastric cancers. Methods: A PubMed search identified relevant studies reporting on the association between hPDI and upper GI cancers. Reference lists were manually searched for additional studies. Study selection followed PRISMA guidelines, and quality was assessed using the Newcastle–Ottawa Scale (NOS). Pooled risk estimates were calculated using inverse-variance fixed-effects models, with subgroup analyses by cancer type. Sensitivity analyses were conducted using leave-one-out approach and by switching between fixed- and random-effects models. Forest plots and funnel plots were generated with RevMan 5.4, while Begg’s and Egger’s tests for publication bias were performed using Stata 15. Results: A total of four studies were included, all with high NOS scores. Overall, higher hPDI scores were associated with a significantly lower risk of upper GI cancers (RR = 0.64, 95% CI: 0.56–0.75). In subgroup analyses, the inverse association was consistent for both esophageal cancer (RR = 0.66, 95% CI: 0.54–0.79) and gastric cancer (RR = 0.63, 95% CI: 0.50–0.79). Between-study heterogeneity was low (I² = 35%, P = 0.16). Publication bias assessment indicated no evidence of small-study effects, as Begg’s test was non-significant (P = 0.88) and Egger’s test showed no significant bias (P = 0.76). Sensitivity analyses confirmed the robustness of the findings, with consistent results after sequential exclusion of individual studies and when using random-effects models. Conclusions: Higher hPDI scores were associated with a significantly reduced risk of esophageal and gastric cancers, indicating that adherence to a healthy plant-based diet may reduce upper GI cancer risk. These findings highlight the potential role of dietary modification in upper GI cancer prevention.

ABSTRACT Objective To comprehensively evaluate the validity of ICD-10-CM codes for both prevalent diagnoses and less common diseases, and to assess the performance of a large language model (LLM)-based system in validating these codes. Materials and Methods This retrospective study analyzed hospital admissions from the Medical Information Mart for Intensive Care (MIMIC-IV) database. We developed a validated LLM-based system using GPT-4o, refined through iterative prompt engineering, to assess ICD-10-CM code validity. We measured the PPV of ICD-10-CM codes, PPV of principal and secondary diagnoses, and the performance of an LLM-based system in code validation. Results Among 865,079 assigned codes, the PPV was 84.6% (95% CI, 84.5%-84.6%). Principal diagnoses had a PPV of 93.9% (95% CI, 93.7%-94.1%), while secondary diagnoses had a PPV of 83.8% (95% CI, 83.7%-83.9%). The LLM system demonstrated high performance in validating ICD codes, achieving 93.6% accuracy, 95.4% sensitivity and 85.2% specificity. Among correctly assigned secondary diagnoses, the majority (67.9%) represented historical or baseline conditions, while 32.1% reflected active conditions that deviated from baseline status; 22.3% of these emerged after hospital admission. PPV decreases with later diagnosis positions, with the largest decline occurring between principal and secondary diagnoses. Discussion and Conclusion In this large-scale evaluation, ICD-10-CM codes exhibited generally high accuracy, though variability existed by position and condition type. A validated LLM system performed comparably to physician review and offers a scalable means to improve coding accuracy. These findings support the potential for integrating LLM-based auditing into routine workflows to strengthen the quality of administrative and research data.