Development of Electronic Health Record-Based Machine Learning Models to Predict Barrett's Esophagus and Esophageal Adenocarcinoma Risk.

INTRODUCTION: Screening for Barrett's esophagus (BE) is suggested in those with risk factors, but remains underutilized. BE/esophageal adenocarcinoma (EAC) risk prediction tools integrating multiple risk factors have been described. However, accuracy remains modest (area under the receiver-operating curve [AUROC] ≤0.7), and clinical implementation has been challenging. We aimed to develop machine learning (ML) BE/EAC risk prediction models from an electronic health record (EHR) database. METHODS: The Clinical Data Analytics Platform, a deidentified EHR database of 6 million Mayo Clinic patients, was used to predict BE and EAC risk. BE and EAC cases and controls were identified using International Classification of Diseases codes and augmented curation (natural language processing) techniques applied to clinical, endoscopy, laboratory, and pathology notes. Cases were propensity score matched to 5 independent randomly selected control groups. An ensemble transformer-based ML model architecture was used to develop predictive models. RESULTS: We identified 8,476 BE cases, 1,539 EAC cases, and 252,276 controls. The BE ML transformer model had an overall sensitivity, specificity, and AUROC of 76%, 76%, and 0.84, respectively. The EAC ML transformer model had an overall sensitivity, specificity, and AUROC of 84%, 70%, and 0.84, respectively. Predictors of BE and EAC included conventional risk factors and additional novel factors, such as coronary artery disease, serum triglycerides, and electrolytes. DISCUSSION: ML models developed on an EHR database can predict incident BE and EAC risk with improved accuracy compared with conventional risk factor-based risk scores. Such a model may enable effective implementation of a minimally invasive screening technology.

An Introduction to Artificial Intelligence in Developmental and Behavioral Pediatrics.

ABSTRACT: Technological breakthroughs, together with the rapid growth of medical information and improved data connectivity, are creating dramatic shifts in the health care landscape, including the field of developmental and behavioral pediatrics. While medical information took an estimated 50 years to double in 1950, by 2020, it was projected to double every 73 days. Artificial intelligence (AI)-powered health technologies, once considered theoretical or research-exclusive concepts, are increasingly being granted regulatory approval and integrated into clinical care. In the United States, the Food and Drug Administration has cleared or approved over 160 health-related AI-based devices to date. These trends are only likely to accelerate as economic investment in AI health care outstrips investment in other sectors. The exponential increase in peer-reviewed AI-focused health care publications year over year highlights the speed of growth in this sector. As health care moves toward an era of intelligent technology powered by rich medical information, pediatricians will increasingly be asked to engage with tools and systems underpinned by AI. However, medical students and practicing clinicians receive insufficient training and lack preparedness for transitioning into a more AI-informed future. This article provides a brief primer on AI in health care. Underlying AI principles and key performance metrics are described, and the clinical potential of AI-driven technology together with potential pitfalls is explored within the developmental and behavioral pediatric health context.

Evaluation of an artificial intelligence-based medical device for diagnosis of autism spectrum disorder.

Autism spectrum disorder (ASD) can be reliably diagnosed at 18 months, yet significant diagnostic delays persist in the United States. This double-blinded, multi-site, prospective, active comparator cohort study tested the accuracy of an artificial intelligence-based Software as a Medical Device designed to aid primary care healthcare providers (HCPs) in diagnosing ASD. The Device combines behavioral features from three distinct inputs (a caregiver questionnaire, analysis of two short home videos, and an HCP questionnaire) in a gradient boosted decision tree machine learning algorithm to produce either an ASD positive, ASD negative, or indeterminate output. This study compared Device outputs to diagnostic agreement by two or more independent specialists in a cohort of 18-72-month-olds with developmental delay concerns (425 study completers, 36% female, 29% ASD prevalence). Device output PPV for all study completers was 80.8% (95% confidence intervals (CI), 70.3%-88.8%) and NPV was 98.3% (90.6%-100%). For the 31.8% of participants who received a determinate output (ASD positive or negative) Device sensitivity was 98.4% (91.6%-100%) and specificity was 78.9% (67.6%-87.7%). The Device's indeterminate output acts as a risk control measure when inputs are insufficiently granular to make a determinate recommendation with confidence. If this risk control measure were removed, the sensitivity for all study completers would fall to 51.6% (63/122) (95% CI 42.4%, 60.8%), and specificity would fall to 18.5% (56/303) (95% CI 14.3%, 23.3%). Among participants for whom the Device abstained from providing a result, specialists identified that 91% had one or more complex neurodevelopmental disorders. No significant differences in Device performance were found across participants' sex, race/ethnicity, income, or education level. For nearly a third of this primary care sample, the Device enabled timely diagnostic evaluation with a high degree of accuracy. The Device shows promise to significantly increase the number of children able to be diagnosed with ASD in a primary care setting, potentially facilitating earlier intervention and more efficient use of specialist resources.

Project Rosetta: a childhood social, emotional, and behavioral developmental feature mapping.

BACKGROUND: A wide array of existing instruments are commonly used to assess childhood behavior and development for the evaluation of social, emotional and behavioral disorders such as Autism Spectrum Disorder (ASD), attention-deficit/hyperactivity disorder (ADHD), and anxiety. Many of these instruments either focus on one diagnostic category or encompass a broad set of childhood behaviors. We analyze a wide range of standardized behavioral instruments and identify a comprehensive, structured semantic hierarchical grouping of child behavioral observational features. We use the hierarchy to create Rosetta: a new set of behavioral assessment questions, designed to be minimal yet comprehensive in its coverage of clinically relevant behaviors. We maintain a full mapping from every functional feature in every covered instrument to a corresponding question in Rosetta. RESULTS: In all, 209 Rosetta questions are shown to cover all the behavioral concepts targeted in the eight existing standardized instruments. CONCLUSION: The resulting hierarchy can be used to create more concise instruments across various ages and conditions, as well as create more robust overlapping datasets for both clinical and research use.

Multi-modular AI Approach to Streamline Autism Diagnosis in Young Children.

Autism has become a pressing healthcare challenge. The instruments used to aid diagnosis are time and labor expensive and require trained clinicians to administer, leading to long wait times for at-risk children. We present a multi-modular, machine learning-based assessment of autism comprising three complementary modules for a unified outcome of diagnostic-grade reliability: A 4-minute, parent-report questionnaire delivered via a mobile app, a list of key behaviors identified from 2-minute, semi-structured home videos of children, and a 2-minute questionnaire presented to the clinician at the time of clinical assessment. We demonstrate the assessment reliability in a blinded, multi-site clinical study on children 18-72 months of age (n = 375) in the United States. It outperforms baseline screeners administered to children by 0.35 (90% CI: 0.26 to 0.43) in AUC and 0.69 (90% CI: 0.58 to 0.81) in specificity when operating at 90% sensitivity. Compared to the baseline screeners evaluated on children less than 48 months of age, our assessment outperforms the most accurate by 0.18 (90% CI: 0.08 to 0.29 at 90%) in AUC and 0.30 (90% CI: 0.11 to 0.50) in specificity when operating at 90% sensitivity.

In-Home Speech and Language Screening for Young Children: A Proof-of-Concept Study Using Interactive Mobile Storytime.

Early identification and intervention of speech and language delays in children contribute to better communication and literacy skills for school readiness and are protective against behavioral and mental health problems. Through collaboration between the data science and clinical teams at Cognoa, we designed Storytime, an interactive storytelling experience on a mobile device using a virtual avatar to mediate speech and language screening for children ages 4 to 6 years old. Our proof-of-concept study collects Storytime session footage from 71 pairs of parents and children including 57 typically developing children and 14 children with a current or prior history of communication impairments. Initial findings suggest that participating children verbally engaged with the video avatar without significant differences in performance across age, gender, and experimental location, leading to promising implications for using Storytime as a future tracking tool with automated feature analyses to detect speech and language delays.

Machine learning approach for early detection of autism by combining questionnaire and home video screening.

Background: Existing screening tools for early detection of autism are expensive, cumbersome, time- intensive, and sometimes fall short in predictive value. In this work, we sought to apply Machine Learning (ML) to gold standard clinical data obtained across thousands of children at-risk for autism spectrum disorder to create a low-cost, quick, and easy to apply autism screening tool. Methods: Two algorithms are trained to identify autism, one based on short, structured parent-reported questionnaires and the other on tagging key behaviors from short, semi-structured home videos of children. A combination algorithm is then used to combine the results into a single assessment of higher accuracy. To overcome the scarcity, sparsity, and imbalance of training data, we apply novel feature selection, feature engineering, and feature encoding techniques. We allow for inconclusive determination where appropriate in order to boost screening accuracy when conclusive. The performance is then validated in a controlled clinical study. Results: A multi-center clinical study of n = 162 children is performed to ascertain the performance of these algorithms and their combination. We demonstrate a significant accuracy improvement over standard screening tools in measurements of AUC, sensitivity, and specificity. Conclusion: These findings suggest that a mobile, machine learning process is a reliable method for detection of autism outside of clinical settings. A variety of confounding factors in the clinical analysis are discussed along with the solutions engineered into the algorithms. Final results are statistically limited and will benefit from future clinical studies to extend the sample size.

Efficiency and tolerance of mitotane in Cushing's disease in 76 patients from a single center.

CONTEXT: Alternatives to transsphenoidal pituitary surgery may be required in Cushing's disease (CD) as a first- or second-line treatment. Mitotane is a potent anti-cortisolic drug but has been rarely investigated in the treatment of CD. OBJECTIVE: Evaluation of the efficacy and tolerance of mitotane in CD patients. DESIGN AND SETTING: Retrospective analysis of 76 patients treated with mitotane from 219 patients diagnosed with CD between 1993 and 2009 in a single center. MAIN OUTCOME MEASURE: Remission was defined as normalization of 24-h urinary free cortisol (24-h-UFC). RESULTS: Remission was achieved in 48 (72%) of the 67 long-term treated patients, after a median time of 6.7 (5.2-8.2) months. Mean plasma mitotane concentration at the time of remission was 10.5 ± 8.9 mg/l, with a mean daily dose of 2.6 ± 1.1 g. A negative linear relationship was observed between plasma mitotane concentration and 24-h-UFC (P<0.0001). Seventeen of 24 (71%) patients with durable remission subsequently experienced recurrence, after a median time of 13.2 (5.0-67.9) months. At the time of treatment discontinuation, ACTH concentration was statistically associated with a lower recurrence probability (hazard ratios 0.57 (0.32-1.00), P=0.05). Intolerance leading to treatment discontinuation occurred in 19 patients (29%). A pituitary adenoma became identifiable during mitotane treatment in 12 (25%) of the 48 patients with initial negative pituitary imaging allowing subsequent transsphenoidal surgery. CONCLUSION: Mitotane is useful at different stages of CD. Mitotane dose adjustment based on plasma concentration monitoring and side effects could control hypercortisolism in the majority of CD patients.

Saturable absorption of sorafenib in patients with solid tumors: a population model.

Sorafenib is an oral tyrosine kinase inhibitor approved for the treatment of advanced renal cell carcinoma and hepatocellular carcinoma. By using a population approach, this study aimed to characterise its pharmacokinetics. Plasma concentration-time data (n = 372) from 71 patients under sorafenib were analysed using nonlinear mixed-effect modelling to estimate population pharmacokinetic parameters, as well as relationships between these parameters and different covariates (demographic, biological). Simulations were done to compare different daily dosing regimens in a context of dose-escalation. A 1-compartment model with saturated absorption, first-order intestinal loss and elimination best described the pharmacokinetics of sorafenib. Absolute bioavailability significantly dropped with increasing daily doses of sorafenib. AUC increased less than proportionally with increasing doses [47.3 (41.3-63.3), 60.3 (56.3-64.4), 71.4 (51.3-99.1), 75.9 (45.5-100.9) mg/L.h for 400, 800, 1,200 and 1,600 mg/day, respectively]. According to the simulations, dividing the daily dose in three or four doses for daily dose >800 mg would significantly increase AUC compared with a twice daily dosing regimen (101.7 vs 81.6 mg/L.h for 400 mg q8h and 600 mg q12h respectively; 131.6 vs 91.5 mg/L.h for 400 mg q6h and 800 mg q12h, respectively). Thrice daily regimen may be most suitable in a context of dose-escalation (>800 mg/day) in non-responders to standard-dosing regimen.

Functional and clinical evidence of the influence of sorafenib binding to albumin on sorafenib disposition in adult cancer patients.

PURPOSE: Sorafenib, an oral multitargeted tyrosine kinase inhibitor, is highly bound to plasma proteins (>99.5%). Little is known about the influence of variations in sorafenib protein binding on its disposition. The aims of this study were to characterize in vitro sorafenib binding properties to albumin using the quenching fluorescence method and investigate the influence of albuminemia and bilirubinemia on sorafenib disposition in 54 adult cancer patients. RESULTS: In vitro estimate of sorafenib dissociation constant (Kd) for albumin was 0.22 µM [CI95 0.20-0.23]. In physiological conditions, sorafenib unbound fraction would increase 1.7-fold as albuminemia decreased from 45 g/L (680 µM) to 30 g/L (453 µM). In presence of bilirubin, apparent Kd of sorafenib was ~1.5-fold greater for bilirubin/albumin molar ratio of 1:4. In clinical settings, median sorafenib clearance (CL) was 1.42 L/h (0.75-2.13 L/h). In univariate analysis, sex, body mass index, and albuminemia were associated with CL (p = 0.04, 0.048, and 0.008, respectively). In multivariate analysis, albuminemia (p = 0.0036) was the single parameter independently associated with CL. CONCLUSION: These findings highlight the major influence of albuminemia on sorafenib clearance and its disposition in cancer patients.

Variability in free mycophenolic acid exposure in adult liver transplant recipients during the early posttransplantation period.

Mycophenolate mofetil (MMF) pharmacokinetics variability in liver transplant recipients during the early posttransplantation period may be related to changes in mycophenolic acid (MPA) protein binding. This study aimed at characterizing the variation of free MPA exposure with respect to time since transplantation. Three groups (A, B, C) were compared. The median posttransplantation time was 12 days (A, n = 26 pharmacokinetic sessions), 36 days (B, n = 25), and 867 days (C, n = 21). The median MPA AUC(0-12) in group A (26.8 mg x h/L) was significantly lower than in groups B (45.2 mg x h/L, P = .031) and C (43.5 mg x h/L, P = .004). Free MPA AUC(0-12) was comparable whatever the time (0.41, 0.34, and 0.33 mg x h/L, respectively). MPA apparent clearance (CL/F) was significantly correlated with MPA free fraction (r = 0.60, P < .0001) and approximately 1.7-fold higher in group A compared to groups B and C (P < .05). Enhanced CL/F in relation with an increase in MPA free fraction results in a low AUC of total MPA during the first postoperative month, but on average, at the population level, the exposure to free MPA is not altered, suggesting that total MPA AUC should not be used to adapt MMF dosing during this period.

Management of Cushing's syndrome due to ectopic adrenocorticotropin secretion with 1,ortho-1, para'-dichloro-diphenyl-dichloro-ethane: findings in 23 patients from a single center.

CONTEXT: Effective treatment for the ectopic ACTH secretion syndrome (EAS) remains a therapeutic challenge. Immediate curative surgery of the responsible nonpituitary tumor is often not possible. OBJECTIVE: The objective of the study was to evaluate 1,ortho-1, para'-dichloro-diphenyl-dichloro-ethane (O,p'DDD) therapy in EAS. DESIGN AND PATIENTS: Patients included 36 consecutive patients with EAS from a single center treated between 1990 and 2006. Twenty-three of these patients, including 18 women aged 53.7 +/- 12.9 yr (mean +/- sd), were treated with O,p'DDD. Patient follow-up was 8.04 +/- 9.6 yr. RESULTS: A mean daily O,p'DDD dose of 3.3 +/- 1.2 g Lysodren equivalent was given for a mean duration of 1.8 +/- 2.1 yr. Urinary cortisol decreased from 2603 +/- 3443 microg/d before treatment to 79 +/- 169 microg/d at the time of maximal O,p'DDD efficacy. Urinary cortisol was normalized in 21 of the 23 patients. Adrenal insufficiency was observed in 20 patients. This was associated with clinical improvement of Cushing's syndrome manifestations, including diabetes, hypertension, and hypokalemia. O,p'DDD plasma levels were 10.4 +/- 6.5 microg/ml in the 12 patients tested at the time of adrenal insufficiency. Side effects were observed during the first 6 months in seven of 15 patients (46%). National Cancer Institute-Classification Common Toxicity Criteria grade 1 or 2 digestive or neurologic toxicity resolved after withdrawal or reduction of O,p'DDD. Careful monitoring was essential to long-term control, clinical improvement, and good tolerability. Medical control of the disease allowed the subsequent characterization of tumors in eight of 13 patients with initially occult tumors. CONCLUSION: With close monitoring, O,p'DDD could be a potent medical treatment for long-term control and management of EAS.

Comparison of a new enzymatic assay with a high-performance liquid chromatography/ ultraviolet detection method for therapeutic drug monitoring of mycophenolic acid in adult liver transplant recipients.

Mycophenolic acid (MPA) is used to prevent graft rejection. The methods used for determining the plasma MPA concentration in liver transplant recipients are the enzyme-multiplied immunoassay technique (EMIT), high-performance liquid chromatography with ultraviolet detection (HPLC-UV), and most recently mass spectrometry. EMIT has been reported to overestimate the MPA concentration by 30% to 35% in comparison with HPLC-UV. Recently, a new automated enzymatic assay based on inosine monophosphate dehydrogenase inhibition has been designed. The aim of the present investigation was to compare this technique with validated HPLC-UV in adult liver transplant recipients treated with tacrolimus or cyclosporine. One hundred seventy-six samples from 50 adult liver transplant recipients were analyzed with both techniques. Patients received mycophenolate mofetil (2 or 3 times daily) coadministered with cyclosporine microemulsion (n = 18) or tacrolimus (n = 32). Samples were drawn over an interdose interval during the early or late posttransplantation period. The Passing-Bablok regression and Bland-Altman plot were used to compare the 2 techniques. The Passing-Bablock regression, calculated from 166 samples, showed very good agreement between the enzymatic assay and the HPLC-UV method: enzymatic assay = 1.0204 (95% confidence interval, 0.9942, 1.0478) x HPLC-UV + 0.0201 (-0.0442, 0.0882). No significant bias was found between the techniques (Bland-Altman plot), and the median relative difference was 2.7% (95% confidence interval, -0.4, 6.6). In conclusion, the enzymatic assay showed an excellent correlation with HPLC-UV. Therefore, this method was proved valid and reliable for the monitoring of the plasma MPA concentration in adult liver transplant recipients treated with cyclosporine microemulsion or tacrolimus.