Two-step interpretable modeling of ICU-AIs.

We present a novel methodology for integrating high resolution longitudinal data with the dynamic prediction capabilities of survival models. The aim is two-fold: to improve the predictive power while maintaining the interpretability of the models. To go beyond the black box paradigm of artificial neural networks, we propose a parsimonious and robust semi-parametric approach (i.e., a landmarking competing risks model) that combines routinely collected low-resolution data with predictive features extracted from a convolutional neural network, that was trained on high resolution time-dependent information. We then use saliency maps to analyze and explain the extra predictive power of this model. To illustrate our methodology, we focus on healthcare-associated infections in patients admitted to an intensive care unit.

Physics captured by data-based methods in El Niño prediction.

On average once every four years, the Tropical Pacific warms considerably during events called El Niño, leading to weather disruptions over many regions on Earth. Recent machine-learning approaches to El Niño prediction, in particular, Convolutional Neural Networks (CNNs), have shown a surprisingly high skill at relatively long lead times. In an attempt to understand this high skill, we here use data from distorted physics simulations with the intermediate-complexity Zebiak-Cane model to determine what aspects of El Niño physics are represented in a specific CNN-based classification method. We find that the CNN can adequately correct for distortions in the ocean adjustment processes, but that the machine-learning method has far more trouble in dealing with distortions in upwelling feedback strength.

Toward fully automated genotyping: genotyping microsatellite markers by deconvolution.

Dense genetic linkage maps have been constructed for the human and mouse genomes, with average densities of 2.9 cM and 0.35 cM, respectively. These genetic maps are crucial for mapping both Mendelian and complex traits and are useful in clinical genetic diagnosis. Current maps are largely comprised of abundant, easily assayed, and highly polymorphic PCR-based microsatellite markers, primarily dinucleotide (CA)n repeats. One key limitation of these length polymorphisms is the PCR stutter (or slippage) artifact that introduces additional stutter bands. With two (or more) closely spaced alleles, the stutter bands overlap, and it is difficult to accurately determine the correct alleles; this stutter phenomenon has all but precluded full automation, since a human must visually inspect the allele data. We describe here novel deconvolution methods for accurate genotyping that mathematically remove PCR stutter artifact from microsatellite markers. These methods overcome the manual interpretation bottleneck and thereby enable full automation of genetic map construction and use. New functionalities, including the pooling of DNAs and the pooling of markers, are described that may greatly reduce the associated experimentation requirements.

Nimodipine in acute ischemic stroke: a double-blind controlled study.

Nimodipine (BAY e 9736), a new dihydropyridine derivative, has been shown to reduce neurological deficits and mortality induced by acute cerebral ischemia in experimental studies. We investigated the effects of this calcium antagonist in patients with acute ischemic stroke through a randomized, double-blind, parallel-designed trial in which nimodipine was compared with placebo. Forty-one of 54 screened cases were found to fulfil the inclusion criteria (sudden occurrence of a focal neurological deficit secondary to an acute ischemic event in the carotid area diagnosed after a complete neurological work-up) and entered the study. Nineteen of them were treated with nimodipine (40 mg t.i.d. administered for 28 days) and the remaining 22 with placebo, given in identical tablets. In all patients the treatment started within 12 h after the onset of the symptoms. Course and intensity of the neurological deficit were evaluated by the Mathew Scale (slightly modified). Forty patients concluded the trial. Nimodipine was withdrawn in one case following the occurrence of a skin rash whose causative relation with the test drug could not be clarified. Altogether, however, nimodipine was well tolerated and no severe cardiovascular adverse reactions were observed. In terms of efficacy, the scores obtained by the Mathew Scale showed a higher rate of improvement on nimodipine than on placebo, thus indicating that patients receiving the latter drug did not fare as well as those receiving the test medication. Our data suggest that nimodipine may be beneficial in the treatment of acute stroke.