Background: Colorectal cancer is now the leading cause of cancer-related deaths among young Americans. Accurate early prediction and a thorough understanding of the risk factors for early-onset colorectal cancer (EOCRC) are vital for effective prevention and treatment, particularly for patients below the recommended screening age. Objective: Our study aims to predict EOCRC using machine learning (ML) and structured electronic health record data for individuals under the screening age of 45 years, with the aim of exploring potential risk and protective factors that could support early diagnosis. Methods: We identified a cohort of patients under the age of 45 years from the OneFlorida+ Clinical Research Consortium. Given the distinct pathology of colon cancer (CC) and rectal cancer (RC), we created separate prediction models for each cancer type with various ML algorithms. We assessed multiple prediction time windows (ie, 0, 1, 3, and 5 y) and ensured robustness through propensity score matching to account for confounding variables including sex, race, ethnicity, and birth year. We conducted a comprehensive performance evaluation using metrics including area under the curve (AUC), sensitivity, specificity, positive predictive value, negative predictive value, and F₁-score. Both linear (ie, logistic regression, support vector machine) and nonlinear (ie, Extreme Gradient Boosting and random forest) models were assessed to enable rigorous comparison across different classification strategies. In addition, we used the Shapley Additive Explanations to interpret the models and identify key risk and protective factors associated with EOCRC. Results: The final cohort included 1358 CC cases with 6790 matched controls, and 560 RC cases with 2800 matched controls. The RC group had a more balanced sex distribution (2:3 male-to-female) compared to the CC group (2:5 male-to-female), and both groups showed diverse racial and ethnic representation. Our predictive models demonstrated reasonable results, with AUC scores for CC prediction of 0.811, 0.748, 0.689, and 0.686 at 0, 1, 3, and 5 years before diagnosis, respectively. For RC prediction, AUC scores were 0.829, 0.771, 0.727, and 0.721 across the same time windows. Key predictive features across both cancer types included immune and digestive system disorders, secondary malignancies, and underweight status. In addition, blood diseases emerged as prominent indicators specifically for CC. Conclusions: Our findings demonstrate the potential of ML models leveraging electronic health record data to facilitate the early prediction of EOCRC in individuals under 45 years. By uncovering important risk factors and achieving promising predictive performance, this study provides preliminary insights that could inform future efforts toward earlier detection and prevention in younger populations.

Assessment of eye movements is the cornerstone of diagnosing vestibular disorders and differentiating central from peripheral causes of dizziness. Nonetheless, accurate assessment of eye movements is challenging, especially in the emergency department and primary care settings. To overcome this challenge, clinicians objectively measure eye movements using devices like video-oculography (VOG) goggles, which provide a video recording of the eye and quantified eye position traces. However, despite the value of VOG goggles in studying eye movements, barriers such as high prices and the need for dedicated operators have limited their use to subspecialty clinics. Recent advancements in the hardware and software of smartphones have positioned them as potential alternatives to VOG goggles that can reliably record and quantify eye movements. Although currently not as accurate as VOG goggles, smartphones can provide a cheap, widely available tool that can be used in various medical settings and even at home by patients. We review the current state and future directions of the devices that can be used for recording and quantifying eye movements.

Importance: Artificial intelligence (AI) is under active development to support diagnosis and prognostication of glaucoma from visual fields (VF). These systems do not audit for vertical- meridian-respecting field loss patterns—known sequelae of stroke, hemorrhage, and neoplasm. Objective: To develop a self-supervised encoder of automated perimetry that learns anatomically interpretable visual field structure without labels and to evaluate its capacity to identify suspected neurologic VF patterns in an independent public glaucoma dataset. Design, Setting, and Participants: Diagnostic study (TRIPOD+AI). Pretraining: 23,223 unlabeled Humphrey VFs (patient-grouped training split of 28,943 fields from 3,871 University of Washington patients; UWHVF, all-comers perimetry). External evaluation: Harvard-Glaucoma Fairness dataset (Harvard-GF; 3,300 patients with paired VF and OCT from a single academic center). The encoder was never exposed to Harvard-GF during training. Exposures: A 128-dimensional masked autoencoder of monocular Humphrey VF pattern- deviation data, with a supervised linear classifier on vertical-midline latent dimensions trained on per-eye expert neurological/nonneurological labels. Main Outcomes and Measures: Primary: classifier accuracy under hard-negative evaluation (cross-validated balanced accuracy and AUC). Secondary: held-out specificity on structurally separated UWHVF controls; expert confirmation and OCT structural correlates of classifier- identified Harvard-GF suspects. Results: Masked reconstruction recovered structure concordant with retinal neuroanatomy; 50 of 128 latent dimensions emerged spatially specialized (vs 23 for the TD encoder). Under hard- negative evaluation the classifier achieved cross-validated balanced accuracy 0.78 (95% CI, 0.75-0.82) and AUC 0.85 (95% CI, 0.82-0.89), with no false positives among 100 held-out controls. Applied to Harvard-GF without fine-tuning, it identified a top-20 of 1,748 glaucoma- labeled patients (1.1%) with morphology inconsistent with glaucoma; all 20 were NHT-positive (mean 62.4), and OCT showed preserved superior (Cohen d = +0.68; P < .001) and inferior (d = +0.63; P = .003) RNFL versus severity-matched controls. Conclusions and Relevance: A self-supervised VF encoder learned anatomically interpretable visual field structure from unlabeled data and identified suspected neurological cases in a curated glaucoma dataset, with expert, rule-based, and OCT corroboration. Visual field datasets used to train glaucoma AI may benefit from neurological screening; the encoder reported here supports such audits and provides a foundation for neuro-ophthalmic AI beyond fundus photography and OCT. Key Points: Question: Can self-supervised learning learn anatomically interpretable visual field structure from unlabeled visual fields and identify neurological field loss? Findings: Trained on 28,943 unlabeled visual fields, the encoder learned anatomically interpretable visual field structure and revealed ≥20 of 1,748 patients (≥1.1%) in an external glaucoma dataset with vertical-midline loss and retinal nerve fiber layer preservation. Meaning: A self-supervised encoder learned anatomically interpretable visual field structure, and can audit glaucoma datasets for neurological field loss, reducing the risk that future AI misclassifies intracranial disease.

Background/Purpose: Downbeat nystagmus (DBN) is an uncommon finding in comatose patients, especially as a manifestation of epileptiform activity. We report a 59-year-old man who developed DBN in the context of myoclonic status epilepticus following anoxic brain injury secondary to cardiac arrest. The DBN was phase-locked with generalized periodic epileptiform discharges (GPDs) on electroencephalography (EEG) and resolved with pharmacologic burst suppression. Conclusion: This case suggests a potential link between DBN and cortical epileptiform activity, which we hypothesize may be due to bilateral cortical hyperexcitability and cerebellar disinhibition. The presence of DBN in this setting may indicate a poor prognosis.