Tricia Morphew

Background: The growing use of computational modeling of real-world data (RWD) in clinical research introduces significant risks for data scientists and AI trainers already managing inconsistency, unexamined bias, and analytical opacity. Building upon more than 600 guideline-based checklists and pre-planning templates for data quality and transparency reporting in observational RWD, we propose a run-time operational execution loop to embed reproducibility analytics within the workflow. Our approach acts as a run-time gatekeeper mapping prespecified analytics goals to a closed-loop LLM semantic code auditor, checking for both logic, bias, and vulnerabilities during code execution. Our framework was designed to bridge the gap between passive disclosure, ensuring proposed requirements, and post-hoc validation. Our structured hierarchical approach coupled with AI auditing is the first phase in the development of full-stack automation of bias-aware RWD analytic validation. Methods: Our analytic framework spans 10 domains across three analytic phases. Each domain includes prespecified criteria governing verification with potential biases, mechanisms, and mitigation strategies systematically mapped throughout. Phase I entails specification (Domains 1–4: objectives and estimands, data inputs and outputs, cohort and filters, and variable coding); Phase II: modeling (Domains 5–7: model testing, diagnostics, and reasonableness checks); and Phase III: robustness and interpretation (Domains 8–10: missing data handling, temporal and stratified checks, and scope boundaries). We utilize specific prompt-directed large language model (LLM) auditing of model elements to ensure data quality. The methodology is illustrated with examples from an empirical characterization of inpatient opioid data from the Cerner Real-World Data (2014–2021). GLMMs and corresponding outputs were generated in R v4.4.2. Curated AI-auditing of first-pass code drafting was performed using ChatGPT 5.2 (OpenAI), with cross-LLM portability assessed using Claude Sonnet 5 (Anthropic). Results: The Cerner RWD database contained 16,734 pediatric cancer patients (76,395 encounters) across 55 U.S. health systems. During the specification phase, total daily dose could not be estimated with sufficient validity from the Cerner RWD, and “as-needed” (PRN) orders, warranted empirically supported exclusion. The modeling phase revealed residual variability in opioid ordering attributable primarily to patient-level differences within health systems (ICC = 19.9%) rather than between health systems (ICC = 2.0%). Temporal analysis revealed divergent opioid-ordering trajectories from 2017 onward, with decreasing morphine and increasing fentanyl ordering. We then utilized curated prompting to enforce an AI-generated coding requirement to produce valid likelihood ratio tests across the prespecified model sequence accounting for temporal trends. After harmonizing the model specification and estimation settings, both LLM workflows produced identical random-effect variance estimates, ICCs, and AIC values evaluated through the third decimal place, with corresponding fixed-effect estimates likewise matching. Conclusions: The 10-domain framework provides a structured, reproducible mechanism for governing RWD analyses by using a closed-loop semantic code auditor to embed run-time analytics directly into the data science workflow, supporting transparent exposure definition, systematic bias assessment, and prespecified analytic decision-making. By bridging the gap between passive transparency checklists and active automation, the framework operationalizes existing data quality guidelines through an executable, bias-aware code-auditing process. Cross-LLM concordance in identifying problematic logic and reproducing complex workflows supports the portability of this run-time gatekeeper. Ultimately, this approach establishes a scalable pathway for embedding expert statistical decision-making into the automation of safe, reliable real-world evidence (RWE) generation.

Background The growing use of computational modeling of real-world data (RWD) in clinical research introduces significant risks for data scientists already managing inconsistency, unexamined bias, and analytical opacity. Without the formal theory supporting biostatistical models and despite more than 600 guideline-based checklists for data quality and transparency reporting in observational RWD, reliance on ad hoc model specification and verification is still the norm. No structured guidance exists to help clinicians and researchers systematically assess and manage the risks generated by computational results using RWD. In this paper, we propose a framework that reflects a rigorous, bias-aware methodology for complex analyses of RWD. We also propose the curated use of a large language model (LLM) supporting AI-generated code-auditing to ensure data quality. Our structured hierarchical approach coupled with AI-auditing is the first phase in the development of full-stack automation of RWD analytic validation. Methods Our analytic framework spans 10 domains across three analytic phases. Each domain includes prespecified criteria governing verification with potential biases, mechanisms, and mitigation strategies systematically mapped throughout. Phase 1 entails Specification (Domains 1–4: objectives and estimands, data inputs and outputs, cohort and filters, and variable coding); Phase 2:Modeling (Domains 5–7: model testing, diagnostics, and reasonableness checks); and Phase 3: Robustness and Interpretation (Domains 8–10: missing data handling, temporal and stratified checks, and scope boundaries). We utilize specific prompt-directed large language model (LLM) auditing of model elements to ensure data quality. The methodology is illustrated with examples from an empirical characterization of inpatient opioid data from Cerner Real World (2014–2021). Results The Cerner RWD data base contained 16,734 pediatric cancer patients (76,395 encounters) across 55 U.S. health systems. During the specification phase, total daily dose could not be estimated with sufficient validity from Cerner RWD, and “as-needed” orders, warranted empirically supported exclusion. The modeling phase revealed residual variability in opioid ordering attributable primarily to patient-level differences within health systems (ICC=19.9%) rather than between health systems (ICC=2.0%). Temporal analysis revealed a downward trend in morphine and fentanyl ordering trajectories from 2017-onward. We then utilized curated prompting to enforce an AI-generated coding requirement to produce valid likelihood ratio tests across the prespecified model sequence accounting for temporal trends. Conclusions The 10-domain framework provides a structured, reproducible mechanism for governing RWD analyses, supporting transparent exposure definition, systematic bias assessment, prespecified analytic decision-making, and auditing via curated AI prompting. To our knowledge, this is the first structured framework to 1) identify analytic metrics at risk for unreliability; 2) incorporate AI-audited workflows in RWD research; and 3) demonstrate utility in a large multi-institutional real-world data base. The approach reflects the initial development in automating of RWE data analytics.

Objective Previous research has shown that predictive models can optimize surgical planning. This study investigates whether predictive models can be used to accurately predict compensatory changes in sagittal spinopelvic parameters after all-posterior spinal fusion (PSF) in patients with adolescent idiopathic scoliosis (AIS). Study design Retrospective analysis of medical records was performed to identify pediatric patients with Lenke class 1–4 AIS who underwent PSF with implantation of patient-specific rods (PSR) during the period from July 2018 to December 2020. All included patients (n = 34) were followed for at least two years. Methods Surgical strategies were simulated on preoperative radiographs to achieve desired thoracic kyphosis (TK) for sagittal planning. Previously validated predictive models were used to predict postoperative lumbar lordosis (LL) and pelvic tilt (PT). Pre-contoured patient specific titanium rods were utilized during corrective surgery. Standard radiographic measurements were obtained preoperatively, along with one and two years postoperatively. Results At 2-year follow-up, in the overall cohort, median TK gain was 9.3° vs. preoperative [IQR: −1.9, 20.5] (p = .003) and 4.9° vs. planned [− 1.6, 10.8] (p < .001). Median differences of postoperative vs. predicted in LL and PT were − 3.8° [IQR: − 10.4, 1.4] (p = .006) and − 0.1° [IQR: − 3.7, 3.3] (p = .555), respectively. At 2-year follow-up in the hypokyphotic subgroup (TK < 20°) (n = 12), median TK gain was 22.4° vs. preoperative [IQR: 15.6, 21.2] (p = .002) and 2.1° vs. planned [IQR: − 2.1, 8.1] (p = .239). Median differences of postoperative vs. predicted in LL and PT were − 3.4 [− 9.7, 0.4] (p = .083) and 0.5 [IQR: − 2.0, 2.7] (p = .262), respectively. Conclusion The predictive models accurately predicted compensatory changes in the spinopelvic parameters of unfused segments after AIS surgery in the hypokyphotic subset of patients up to two years post op, and up to one year post-operatively in our entire cohort.

A changing climate indirectly threatens health by acting as a stressor from perceived threats to well-being. Few studies assess youth-centered climate anxiety in clinical settings. The emergency department (ED) is a prominent source of adolescent mental health triage. The primary objective was to understand demographic and clinical predictors of climate anxiety among ED pediatric patients. Secondarily, to externally validate an abbreviated version of the climate change anxiety scale (CCAS). In this cross-sectional study, data was collected from 2023 to 2024 in a single-center pediatric ED, with 860 patients ages 12–17 years approached via convenience sampling; 116 did not complete study surveys. Exposures included demographic factors (age, gender, race, ethnicity, insurance payor), chief complaint type (mental health or medical), temporal season, and General Anxiety Disorder–7 (GAD-7) score. The main outcome was climate anxiety, measured on the CCAS. 744 participants, with 79.8% presenting with medical complaints and 28.1% reporting climate anxiety. Females had 1.49 times higher odds of experiencing greater climate anxiety compared to males (95% CI 1.05, 2.12), p =.027. Having public health insurance increased the odds of experiencing climate anxiety (OR = 1.82, 95% CI 1.21, 2.73), p =.004, as well as warm seasonality (OR = 1.51, 95% CI (1.07, 2.13), p =.018. Participants with moderate or severe GAD-7 had five times higher odds of elevated climate anxiety than those with minimal generalized anxiety (p <.001). The CCAS-short form (CCAS-S) demonstrated high sensitivity (91.4%–96.8%) and specificity (85%–95.5%) in detecting mild to severe levels of distress compared to the full CCAS. A four-tiered threshold of CCAS-S severity showed excellent reliability (quadratic weighted kappa = 0.838). We identified clinical predictors of climate anxiety and externally validated the CCAS-S in a large pediatric cohort. Given the increasing impact of the changing climate on human health, we advocate utilizing the tool alongside routine pediatric mental health assessments.

The Mon Valley near Pittsburgh, Pennsylvania, consistently reports some of the poorest air quality in the United States. Recent studies have linked air pollution in this region to poor asthma outcomes but did not examine the impact on other respiratory conditions or vulnerable populations. This retrospective study examined the relationship between the air quality index (AQI) and respiratory exacerbations of asthma, bronchitis, and chronic obstructive pulmonary disease (COPD) in the Mon Valley between January 2018 and February 2020. We linked daily Air Quality Index (AQI) values for ozone, PM_2.5, SO₂ and NO₂, plus temperature and wind speed to healthcare utilization for these conditions. Using a Poisson generalized linear model, we quantified the association between pollutant levels and same-day exacerbation rates, stratified analyses by age, sex, and insurance type to identify vulnerable subgroups. Results indicated that higher AQI scores, driven primarily by PM_2.5 and SO₂, were significantly associated with increased asthma exacerbations on the day of exposure. Children and individuals with public insurance experienced the greatest impact. Bronchitis exacerbations showed a delayed response to SO₂. Our findings affirm PM_2.5 and SO₂ as key drivers of acute asthma events in the Mon Valley and extend this observation to include impacts on bronchitis and vulnerable populations. They also demonstrate the AQI’s value for public health surveillance and underscore the importance of tailored interventions such as issuing timely air quality alerts, strengthening emissions regulations, and improving access to preventive care to protect at-risk populations from adverse air pollution effects.

Background/Objective: Millions of children were infected with SARS-CoV-2, and a small proportion progressed to severe disease, especially those with underlying risk factors. Adult COVID-19 studies showed mortality benefits with Remdesivir. Data on Remdesivir use in pediatrics are limited. We report on the safety and tolerability of Remdesivir in pediatric patients seen at our institution. Methods: This was a retrospective cohort study of patients p < 0.001), and 55 (51, 59) (p = 0.479); for ALT, they were 42 (38, 47), 59 (95% CI 52, 66) (p < 0.001), and 46 (95% CI 41, 52) (p = 0.054); and for bilirubin, they were 0.56 (95% CI 0.50, 0.62), 0.67 (95% CI 0.61, 0.74) (p < 0.001), and 0.44 (95% CI 0.40, 0.48) (p < 0.001), respectively. During Remdesivir treatment, we did not observe marrow suppression or renal toxicity. Conclusions: No clinically significant hematological or renal toxicity was noted. Mean liver enzymes increased modestly and returned to baseline without interrupting treatment. Remdesivir was well tolerated in patients <19 years old.

Background: A possible explanation for the continued risk of asthma exacerbations while utilizing current asthma guideline‐based management is the failure to consider small airway dysfunction (SAD) and eosinophilic airway inflammation in assuming asthma‐controlled status. Objective: To construct a practical algorithm that potentially identifies additional risks of not well‐controlled (NWC) asthma and exacerbations beyond guideline criteria with oscillometry and fractional exhaled nitric oxide (FENO) determination. Methods: This was a retrospective, posthoc analysis of 183 children, ages 4–18 years, with moderate‐to‐severe asthma, characterized by demographic factors, National Health Lung and Blood Institute (NHLBI) determined asthma‐controlled status, therapy step, and lung function status. Impulse oscillometry (IOS) and FENO were performed before spirometry. Abnormal lung function values were determined for spirometry, SAD, and FENO based on established thresholds. Statistical analysis included a decision tree and Venn diagram. Results: For 53 patients with guideline‐based NWC, 15 (28.3%) had an FEV1 ≤ 80% predicted. An additional 30 patients (56.6%) had abnormal SAD and/or FENO with a preserved FEV1. Of 130 patients, well‐controlled (WC) by guidelines 76 (58.5%) exhibited SAD and/or abnormal FENO. Conclusions: These data demonstrate that over 50% of WC patients with moderate‐to‐severe asthma, with few symptoms and preserved FEV1, had SAD and/or abnormal FENO levels, suggesting the potential of early phenotype‐specific therapy to prevent risk of future exacerbations.

Background: Despite advancements in minimally invasive repair of pectus excavatum (MIRPE), Nuss procedure, postoperative pain control remains challenging. This report covers a multimodal regimen using bilateral single-shot paravertebral block (PVB) and bilateral thoracoscopic intercostal nerve (T3–T7) cryoablation, leading to significant reduction in length of stay (LOS) and high rate of same-day discharge. Methods: This is a comparative study of pain management protocols for patients undergoing the Nuss procedure at a single center from 2016 through 2020. All patients underwent the the same surgical technique for the treatment of pectus excavatum at a single center. Patients received bilateral PVB with continuous infusion (Group 1, n = 12), bilateral PVB with infusion and right-side cryoablation (Group 2, n = 9), or bilateral single-shot PVB and bilateral cryoablation (Group 3, n = 17). The primary outcome was LOS with focus on same-day discharge, and the secondary outcome was decreased opioid usage. Results: Eleven of 17 patients in Group 3 (65%) (bilateral single-shot PVB and bilateral cryoablation) were discharged the same day as surgery. The remaining Group 3 patients were discharged the following day with no complications or interventions. Compared to Group 1 (no cryoablation), Group 3 had shorter LOS (median 4.4 days vs. 0.7 days, respectively, p < 0.001) and significantly decreased median opioid use on the day of surgery (0.92 mg/kg vs. 0.47 mg/kg, p = 0.006). Conclusion: Findings demonstrate the feasibility of multimodal pain management for same-day discharge after the Nuss procedure. Future multisite studies are needed to investigate the superiority of this approach to established methods. Level of Evidence: III.