Cai Li

Large language models (LLMs) have transformed knowledge discovery, but their use in sensitive domains is limited by privacy concerns and risks of data leakage. Moreover, LLMs may hallucinate when queried about information outside their training data, undermining their reliability for scientific, clinical, or policy-related tasks. We present MyGPT, a fully open-source, privacy-preserving retrieval-augmented-generation (RAG) platform that uses open-source LLMs and operates locally or within secure institutional environments. MyGPT empowers users to query user-defined libraries such as biomedical literature, confidential clinical documents, and multilingual global health policies, while ensuring that sensitive data never leaves the secure environment. We introduce confidence metrics, including “relevance scores” and “hallucination index,” and provide explicit source citation from the library within a single interface to assess contextual support for generated answers. MyGPT offers a cost-effective, modular, open-source alternative to commercial LLM services and democratizes knowledge-seeking across multiple domains, including use at remote facilities without internet access.

Understanding depressive symptoms dynamics and their determinants is crucial for designing effective mental health support initiatives. This study compared two methods for describing youth depressive symptoms trajectories and investigated associations of early-life factors (maternal education, maternal perinatal depression, domestic violence, physical, emotional, or sexual abuse, bullying victimisation, psychiatric disorder) with trajectory features. Prospective data from 8,264 mostly White European participants (54% female), including self-reported Short Moods and Feelings Questionnaires on ten occasions between 10-25 years, were used. Trajectories were summarised using functional principal component analysis (FPCA) and P-splines linear mixed-effect (PLME) models. Estimated derivatives were used to obtain magnitude and age of peak symptoms and peak symptoms velocity. Both methods performed comparably, but PLME models tended to over-smooth trajectories. Peak symptoms and peak velocity were higher and occurred >1 year earlier in females than males. All early-life factors were associated with higher peak symptoms, and most associated with higher and earlier peak velocity. Abuse and bullying additionally associated with earlier age of peak symptoms. FPCA is a useful alternative for characterising depressive symptoms trajectories and informing time-sensitive preventative measures to reduce impact of depression before symptoms reach their peak. Early-life stressors may accelerate timeline and intensity of symptoms escalation during adolescence.

Background and Objectives: Little is known about the final week of life for inpatient pediatric oncology patients. The purpose of this study was to describe inpatient pediatric oncology deaths. Methods: Retrospective chart review of all patients who died in a large pediatric cancer center between 2007 and 2017. Demographic, diagnostic, and proximate cause of death information was extracted. Intensive care unit (ICU) admissions, chemotherapy receipt, medical interventions, and cardiopulmonary resuscitation (CPR) events one week, 48 h, and 24 h prior to death were obtained. Analysis included descriptive and statistical correlations. Results: 344 decedent pediatric oncology patients were included. Half of the patients died in the ICU (51%). The average age was 11.3 years (range 0.13–27.7 years). Most had ongoing evidence of disease (86%) with 20% receiving cure-directed chemotherapy during their final week. Receiving cure-directed chemotherapy was not associated with age, race, ethnicity, or diagnosis but was associated with a significantly increased likelihood of receiving CPR on the last day of life (p = 0.011). The majority (84%) of children did not receive CPR on their final day of life. Receipt of CPR was not associated with race/ethnicity. CPR was associated with younger age (p = 0.006), hematologic malignancies (p = 0.037), and ICU location (p < 0.001). Most patients were not on dialysis (84%), pressors (72%), or ventilated (60%) during the final 24 h of life. Compassionate extubation occurred in <6% of deaths. Conclusions: Most deaths in a pediatric cancer center occur in children with active disease. Continuation of cure-directed chemotherapy, age, diagnosis, and location of death has potential to influence end-of-life inpatient care.

Neurodevelopmental disorders are thought to arise from intrinsic brain abnormalities. Alternatively, they may arise from disrupted crosstalk among tissues. Here we show the local reduction of two vestibulo-cerebellar lobules, the paraflocculus and flocculus, in mouse models and humans with 22q11.2 deletion syndrome (22q11DS). In mice, this paraflocculus/flocculus dysplasia is associated with haploinsufficiency of the Tbx1 gene. Tbx1 haploinsufficiency also leads to impaired cerebellar synaptic plasticity and motor learning. However, neural cell compositions and neurogenesis are not altered in the dysplastic paraflocculus/flocculus. Interestingly, 22q11DS and Tbx1^+/– mice have malformations of the subarcuate fossa, a part of the petrous temporal bone, which encapsulates the paraflocculus/flocculus. Single-nuclei RNA sequencing reveals that Tbx1 haploinsufficiency leads to precocious differentiation of chondrocytes to osteoblasts in the petrous temporal bone autonomous to paraflocculus/flocculus cell populations. These findings suggest a previously unrecognized pathogenic structure/function relation in 22q11DS in which local skeletal deformity and cerebellar dysplasia result in behavioral deficiencies.

Serial prognostic evaluation after allogeneic hematopoietic cell transplantation (allo-HCT) might help identify patients at high risk of lethal organ dysfunction. Current prediction algorithms based on models that do not incorporate changes to patients’ clinical condition after allo-HCT have limited predictive ability. We developed and validated a robust risk-prediction algorithm to predict short- and long-term survival after allo-HCT in pediatric patients that includes baseline biological variables and changes in the patients’ clinical status after allo-HCT. The model was developed using clinical data from children and young adults treated at a single academic quaternary-care referral center. The model was created using a randomly split training data set (70% of the cohort), internally validated (remaining 30% of the cohort) and then externally validated on patient data from another tertiary-care referral center. Repeated clinical measurements performed from 30 days before allo-HCT to 30 days afterwards were extracted from the electronic medical record and incorporated into the model to predict survival at 100 days, 1 year, and 2 years after allo-HCT. Naïve-Bayes machine learning models incorporating longitudinal data were significantly better than models constructed from baseline variables alone at predicting whether patients would be alive or deceased at the given time points. This proof-of-concept study demonstrates that unlike traditional prognostic tools that use fixed variables for risk assessment, incorporating dynamic variability using clinical and laboratory data improves the prediction of mortality in patients undergoing allo-HCT.

Aggressive cancer treatments that affect the central nervous system are associated with an increased risk of cognitive deficits. As treatment for pediatric brain tumors has become more effective, there has been a heightened focus on improving cognitive outcomes, which can significantly affect the quality of life for pediatric cancer survivors. This paper is motivated by and applied to a clinical trial for medulloblastoma, the most common malignant brain tumor in children. The trial collects comprehensive data including treatment-related clinical information, neuroimaging, and longitudinal neurocognitive outcomes to enhance our understanding of the responses to treatment and the enduring impacts of radiation therapy on the survivors of medulloblastoma. To this end, we have developed a new mediation model tailored for longitudinal outcomes with high-dimensional imaging mediators. Specifically, we adopt a joint binary Ising-Gaussian Markov random field prior distribution to account for spatial dependency and smoothness of ultra-high-dimensional neuroimaging mediators for enhancing detection power of informative voxels. By exploiting the proposed approach, we identify causal pathways and the corresponding white matter microstructures mediating the negative impact of irradiation on neurodevelopment. The results provide guidance on sparing the brain regions and improving long-term neurodevelopment for pediatric cancer survivors. Simulation studies also confirm the validity of the proposed method.

Objective: To investigate the clinicopathological, therapeutic, and survival data on pediatric major salivary gland cancers. Materials and Methods: National Cancer Database (NCDB) query from 2004 to 2018. Results: In total, 967 cases of individuals under the age of 21 were identified. Most cancers affected the parotid gland (86%). Mucoepidermoid carcinoma (41.3%) and acinic cell adenocarcinoma (33.6%) were the most common. Tumors occurred more often from age 11 to 21, and females were more affected. Histology varied by age, gender, and race. In the 0–5 age group, mucoepidermoid carcinoma and myoepithelial carcinoma/sarcoma/rhabdomyosarcoma were the most common pathologies. In patients over 5 years old, mucoepidermoid carcinoma was the most frequent tumor in boys, while acinic cell adenocarcinoma was more common in girls. African American patients had a higher incidence of mucoepidermoid carcinoma, while White patients in the 0–5 age group had a higher incidence of myoepithelial carcinoma/sarcoma/rhabdomyosarcoma tumors. Low-grade tumors were commonly diagnosed at stage I, but the 0–5 age group had a high frequency of stage IV tumors. The overall 5-year survival rate was 94.9%, with 90% for the 0–5 years age group and 96% for the 11–15 years age group. Negative margins were associated with higher 5-year survival rates in high-stage tumors (93%) compared to positive margins (80%). Submandibular malignancies had worse 5-year survival rates across all age groups. Conclusions: Major salivary gland malignancies in pediatric patients exhibit variations in histopathologic characteristics by age, gender, and race. Negative margins impact 5-year survival rates, especially in high-stage tumors.

The analysis of tumor-related and perivascular CT radiomic features improved the detection of superior mesenteric artery involvement in patients with pancreatic ductal adenocarcinoma. Background Current imaging methods for prediction of complete margin resection (R0) in patients with pancreatic ductal adenocarcinoma (PDAC) are not reliable. Purpose To investigate whether tumor-r...

Human intelligence is usually measured by well-established psychometric tests through a series of problem solving. The recorded cognitive scores are continuous but usually heavy-tailed with potential outliers and violating the normality assumption. Meanwhile, magnetic resonance imaging (MRI) provides an unparalleled opportunity to study brain structures and cognitive ability. Motivated by association studies between MRI images and human intelligence, we propose a tensor quantile regression model, which is a general and robust alternative to the commonly used scalar-on-image linear regression. Moreover, we take into account rich spatial information of brain structures, incorporating low-rankness and piecewise smoothness of imaging coefficients into a regularized regression framework. We formulate the optimization problem as a sequence of penalized quantile regressions with a generalized Lasso penalty, based on tensor decomposition, and develop a computationally efficient alternating direction method of multipliers algorithm (ADMM) to estimate the model components. Extensive numerical studies are conducted to examine the empirical performance of the proposed method and its competitors. Finally, we apply the proposed method to a large-scale important dataset—the Human Connectome Project. We find that the tensor quantile regression can serve as a prognostic tool to assess future risk of cognitive impairment progression. More importantly, with the proposed method we are able to identify the most activated brain subregions associated with quantiles of human intelligence. The prefrontal and anterior cingulate cortex are found to be mostly associated with lower and upper quantile of fluid intelligence. The insular cortex associated with median of fluid intelligence is a rarely reported region.

Regional human brain volumes including total area, average thickness, and total volume are heritable and associated with neurological disorders. However, the genetic architecture of brain structure and function is still largely unknown and worthy of exploring. The Pediatric Imaging, Neurocognition, and Genetics (PING) data set provides an excellent resource with genome‐wide genetic data and related neuroimaging data. In this study, we perform genome‐wide association studies (GWAS) of 315 brain volumetric phenotypes from the PING data set including 1036 samples with 539,865 single‐nucleotide polymorphisms (SNPs). We introduce a nonparametric test based on K‐sample Ball Divergence (KBD) to identify genetic risk variants that influence regional brain volumes. We carry out simulations to demonstrate that KBD is a powerful test for identifying significant SNPs associated with multivariate phenotypes although controlling the type I error rate. We successfully identify nine SNPs below a significance level of 5 × 10⁻⁵ for the PING data. Among the nine identified genetic variants, two SNPs rs486179 and rs562110 are located in the ADRA1A gene that is a well‐known risk factor of mental illness, such as schizophrenia and attention deficit hyperactivity disorder. Our study suggests that the nonparametric test KBD is an effective method for identifying genetic variants associated with complex diseases in large‐scale GWAS of multiple phenotypes.