Generalising uncertainty improves accuracy and safety of deep learning analytics applied to oncology.

Uncertainty estimation is crucial for understanding the reliability of deep learning (DL) predictions, and critical for deploying DL in the clinic. Differences between training and production datasets can lead to incorrect predictions with underestimated uncertainty. To investigate this pitfall, we benchmarked one pointwise and three approximate Bayesian DL models for predicting cancer of unknown primary, using three RNA-seq datasets with 10,968 samples across 57 cancer types.

BladMetrix: a novel urine DNA methylation test with high accuracy for detection of bladder cancer in hematuria patients.

Background: Cystoscopy is the gold standard for bladder cancer detection, but is costly, invasive and has imperfect diagnostic accuracy. We aimed to identify novel and accurate DNA methylation biomarkers for non-invasive detection of bladder cancer in urine, with the potential to reduce the number of cystoscopies among hematuria patients.

Results: Biomarker candidates (n = 32) were identified from methylome sequencing of urological cancer cell lines (n = 16) and subjected to targeted methylation analysis in tissue samples (n = 60).

Automated AI labeling of optic nerve head enables insights into cross-ancestry glaucoma risk and genetic discovery in >280,000 images from UKB and CLSA.

Cupping of the optic nerve head, a highly heritable trait, is a hallmark of glaucomatous optic neuropathy. Two key parameters are vertical cup-to-disc ratio (VCDR) and vertical disc diameter (VDD). However, manual assessment often suffers from poor accuracy and is time intensive.

Characterizing the underlying diagnoses for exploratory laparotomies to improve risk-adjustment models of postoperative mortality.

Background: Surgeons perform emergent exploratory laparotomies (ex-laps) for a myriad of surgical diagnoses. We characterized common diagnoses for which emergent ex-laps were performed and leveraged these groups to improve risk-adjustment models for postoperative mortality.

Methods: Using American Association for the Surgery of Trauma criteria, we identified hospitalizations where the primary procedure was an emergent ex-lap in the 2012 to 2014 (derivation cohort) and 2015 (validation cohort) Nationwide Inpatient Sample.

Assessing regional variations in the effect of the removal of user fees on facility-based deliveries in rural Zambia.

Background: Maternal health remains a concern in sub-Saharan Africa, where maternal mortality averages 680 per 100,000 live births and almost 50% of the approximately 350,000 annual maternal deaths occur. Improving access to skilled birth assistance is paramount to reducing this average, and user fee reductions could help.

Objective: The aim of this research was to analyse the effect of user fee removal in rural areas of Zambia on the use of health facilities for childbirth.