Using explainable machine learning to identify patients at risk of reattendance at discharge from emergency departments.

Short-term reattendances to emergency departments are a key quality of care indicator. Identifying patients at increased risk of early reattendance could help reduce the number of missed critical illnesses and could reduce avoidable utilization of emergency departments by enabling targeted post-discharge intervention. In this manuscript, we present a retrospective, single-centre study where we created and evaluated an extreme gradient boosting decision tree model trained to identify patients at risk of reattendance within 72 h of discharge from an emergency department (University Hospitals Southampton Foundation Trust, UK). Our model was trained using 35,447 attendances by 28,945 patients and evaluated on a hold-out test set featuring 8847 attendances by 7237 patients. The set of attendances from a given patient appeared exclusively in either the training or the test set. Our model was trained using both visit level variables (e.g., vital signs, arrival mode, and chief complaint) and a set of variables available in a patients electronic patient record, such as age and any recorded medical conditions. On the hold-out test set, our highest performing model obtained an AUROC of 0.747 (95% CI 0.722-0.773) and an average precision of 0.233 (95% CI 0.194-0.277). These results demonstrate that machine-learning models can be used to classify patients, with moderate performance, into low and high-risk groups for reattendance. We explained our models predictions using SHAP values, a concept developed from coalitional game theory, capable of explaining predictions at an attendance level. We demonstrated how clustering techniques (the UMAP algorithm) can be used to investigate the different sub-groups of explanations present in our patient cohort.

3D semiconducting nanostructures via inverse lipid cubic phases.

Well-ordered and highly interconnected 3D semiconducting nanostructures of bismuth sulphide were prepared from inverse cubic lipid mesophases. This route offers significant advantages in terms of mild conditions, ease of use and electrode architecture over other routes to nanomaterials synthesis for device applications. The resulting 3D bicontinous nanowire network films exhibited a single diamond topology of symmetry Fd3m (Q227) which was verified by Small angle X-ray scattering (SAXS) and Transmission electron microscopy (TEM) and holds great promise for potential applications in optoelectronics, photovoltaics and thermoelectrics.

Galectin-1 has potential prognostic significance and is implicated in clear cell renal cell carcinoma progression through the HIF/mTOR signaling axis.

BACKGROUND: Metastatic clear cell renal cell carcinoma (ccRCC) patients have <9% 5-year survival rate, do not respond well to targeted therapy and eventually develop resistance. A better understanding of molecular pathways of RCC metastasis is the basis for the discovery of novel prognostic markers and targeted therapies. METHODS: We investigated the biological impact of galectin-1 (Gal-1) in RCC cell lines by migration and invasion assays. Effect of Gal-1 expression on the mitogen-activated protein kinase pathway was assessed by proteome array. RESULTS: Increased expression of Gal-1 increased cell migration while knocking down Gal-1 expression by siRNA resulted in reduced cellular migration (P<0.001) and invasion (P<0.05). Gal-1 overexpression increased phosphorylation of Akt, mTOR and p70 kinase. Upon hypoxia and increased HIF-1α, Gal-1 increased in a dose-dependent manner. We also found miR-22 overexpression resulted in decreased Gal-1 and HIF-1α. Immunohistochemistry analysis showed that high Gal-1 protein expression was associated with larger size tumor (P=0.034), grades III/IV tumors (P<0.001) and shorter disease-free survival (P=0.0013). Using the Cancer Genome Atlas data set, we found that high Gal-1 mRNA expression was associated with shorter overall survival (41 vs 78 months; P<0.01). CONCLUSIONS: Our data suggest Gal-1 mediates migration and invasion through the HIF-1α-mTOR signaling axis and is a potential prognostic marker and therapeutic target.

miRNA profiling in metastatic renal cell carcinoma reveals a tumour-suppressor effect for miR-215.

BACKGROUND: Renal cell carcinoma (RCC) is the most common neoplasm of the adult kidney. Metastatic RCC is difficult to treat. The 5-year survival rate for metastatic RCC is ≤10%. Recently, microRNAs (miRNAs) have been shown to have a role in cancer metastasis and potential as prognostic biomarkers in cancer. METHOD: We performed a miRNA microarray to identify a miRNA signature characteristic of metastatic compared with primary RCCs. We validated our results by quantitative real-time PCR. We performed experimental and bioinformatic analyses to explore the involvement of miR-215 in RCC progression and metastasis. RESULTS: We identified 65 miRNAs that were significantly altered in metastatic compared with primary RCCs. We validated our results by examining the expression of miR-10b, miR-126, miR-196a, miR-204 and miR-215, in two independent cohorts of patients. We showed that overexpression of miR-215 decreased cellular migration and invasion in an RCC cell line model. In addition, through gene expression profiling, we identified direct and indirect targets of miR-215 that can contribute to tumour metastasis. CONCLUSION: Our analysis showed that miRNAs are altered in metastatic RCCs and can contribute to kidney cancer metastasis through different biological processes. Dysregulated miRNAs represent potential prognostic biomarkers and may have therapeutic applications in kidney cancer.

Three dysregulated miRNAs control kallikrein 10 expression and cell proliferation in ovarian cancer.

BACKGROUND: Kallikrein-related peptidases (KLKs) are a family of serine proteases that have been shown to be dysregulated in several malignancies including ovarian cancer. The control of kallikrein genes and their physiological function in cancer is not well understood. We hypothesized that microRNAs (miRNAs) represent a novel mechanism for post-transcriptional control of KLK expression in cancer. METHODS: We first analysed miRNA expression in ovarian cancer in silico. A total of 98 miRNAs were reported to have altered expression in ovarian cancer. Three of these miRNAs were predicted to target KLK10. We experimentally verified the predicted miR-KLK10 interaction using two independent techniques, a luciferase assay with a construct containing the KLK10 3' untranslated region (UTR), pMIR-KLK10, and measuring KLK10 protein levels after transfection with miRNA. RESULTS: When we co-transfected cells with pMIR-KLK10 and either let-7f, miR-224, or mR-516a, we saw decreased luciferase signal, suggesting that these miRNAs can target KLK10. We then examined the effect of these three miRNAs on KLK10 protein expression and cell growth. Transfection of all miRNAs, let-7f, miR-224, and miR-516a led to a decrease in protein expression and cellular growth. This effect was shown to be dose dependent. The KLK10 protein levels were partially restored by co-transfecting let-7f and its inhibitor. In addition, there was a slight decrease in KLK10 mRNA expression after transfection with let-7f. CONCLUSION: Our results confirm that KLKs can be targeted by more than one miRNA. Increased expression of certain miRNAs in ovarian cancer can lead to decreased KLK protein expression and subsequently have a negative effect on cell proliferation. This dose-dependent effect suggests that a 'tweaking' or 'fine-tuning' mechanism exists in which the expression of one KLK can be controlled by multiple miRNAs. These data together suggest that miRNA may be used as potential therapeutic options and further studies are required.

KLK6 and KLK13 predict tumor recurrence in epithelial ovarian carcinoma.

BACKGROUND: The human kallikrein-related peptidase family consists of 15 genes. Twelve of these genes are overexpressed in ovarian cancer and may represent potential markers for diagnosis, prognosis, and/or response to treatment. The aim of this study was to determine the prognostic significance of kallikrein-related peptidase 6 (KLK6) and kallikrein-related peptidase 13 (KLK13) in epithelial ovarian cancer by quantifying gene expression levels with tumour pathology and patient survival data. METHODS: Total RNA was isolated from 106 patients diagnosed with primary ovarian cancer, as well as 8 normal ovary controls. Samples were analysed by quantitative real-time PCR for KLK6 and KLK13 expression. Correlation between kallikrein gene expression and clinical characteristics was evaluated with the chi(2)-test. Survival analysis was performed using Kaplan-Meier and Cox proportional hazards regression models. RESULTS: Expression levels of both KLK6 and KLK13 mRNA were significantly increased in invasive cancers relative to normal ovaries (P=0.002 and 0.039 respectively). High KLK6 and KLK13 expression was an indicator of poor prognosis, with patients having a shorter recurrence-free survival (P=0.002 and 0.027 respectively). High KLK6 expression was also significantly associated with lower overall survival (P=0.011). When subjected to multivariate analysis, patients with either high KLK6 or KLK13 were 3- and 2.2-fold, respectively, more likely to have a recurrence than patients with low kallikrein expression. CONCLUSION: These data show increased mRNA expression of KLK6 and KLK13 in ovarian cancer compared to normal ovarian tissues. High KLK6 or KLK13 expression in primary ovarian tumours can significantly predict prognosis in terms of recurrence-free survival and overall survival. In all, this study shows KLK6 and KLK13 as potential biomarkers and may be therapeutic targets for treatment of ovarian cancer.

Effects of development on runoff and pollutant export.

Runoff from two similar drainage areas in the Piedmont physiographic region of North Carolina was monitored simultaneously for 5.6 years. One of the drainage areas was developed as part of a large residential subdivision, while the other remained in woods and agricultural fields. Runoff volume was 68% greater for the developed compared with the undeveloped area, and baseflow as a percentage of overall discharge was approximately 0% compared with 25% for the undeveloped area. Overall annual export of sediment was 95% greater for the developed area, while export of nitrogen and phosphorus forms was 66 to 88% greater for the developed area. These results document the significant increases in runoff, sediment, and nutrient export associated with residential development.

Investigation of two-dimensional acoustic resonant modes in a particle separator.

Within an acoustic standing wave particles experience acoustic radiation forces, a phenomenon which is exploited in particle or cell manipulation devices. When developing such devices, one-dimensional acoustic characteristics corresponding to the transducer(s) are typically of most importance and determine the primary radiation forces acting on the particles. However, radiation forces have also been observed to act in the lateral direction, perpendicular to the primary radiation force, forming striated patterns. These lateral forces are due to lateral variations in the acoustic field influenced by the geometry and materials used in the resonator. The ability to control them would present an advantage where their effect is either detrimental or beneficial to the particle manipulation process. The two-dimensional characteristics of an ultrasonic separator device have been modelled within a finite element analysis (FEA) package. The fluid chamber of the device, within which the standing wave is produced, has a width to height ratio of approximately 30:1 and it is across the height that a half-wavelength standing wave is produced to control particle movement. Two-dimensional modal analyses have calculated resonant frequencies which agree well with both the one-dimensional modelling of the device and experimentally measured frequencies. However, these two-dimensional analyses also reveal that these modes exhibit distinctive periodic variations in the acoustic pressure field across the width of the fluid chamber. Such variations lead to lateral radiation forces forming particle bands (striations) and are indicative of enclosure modes. The striation spacings predicted by the FEA simulations for several modes compare well with those measured experimentally for the ultrasonic particle separator device. It is also shown that device geometry and materials control enclosure modes and therefore the strength and characteristics of lateral radiation forces, suggesting the potential use of FEA in designing for the control of enclosure modes in similar particle manipulator devices.

Bacterial contamination associated with estuarine shoreline development.

AIMS: To examine the relationships among increasing estuarine shellfish closings due to bacterial contamination, adjacent shoreline land uses and environmental variables. METHODS AND RESULTS: A 1 year study of faecal coliform bacterial contamination of a small estuary in central NC, USA was done relative to adjacent land uses. The area has experienced rapid growth in residential shoreline development including the installation of adjacent, separate docking facilities for larger boats, each <11 slips (pseudomarina) that appear to be a single marina (individual facilities of >10 slips). Six near-shore sites were selected [old developed shore (OD), undeveloped shore (UD), two pseudomarinas (P1, P2), newly developed shore (ND) and a real marina (RM)]. Five locations were spaced along the shore near each site. Paired Thursday/Monday samples were collected biweekly (summer) and monthly (other seasons). Results indicate that OD had the highest bacteria counts followed by ND, RM and P1 & P2. Three sites (OD, ND and RM) failed to meet NC shellfishing waters standards at all locations. At the pseudomarina sites 4 of 10 locations failed to meet shellfish standards while two locations at UD failed to meet these standards. There were no significant differences between paired Thursday/Monday samples. At three sites (OD, UD and P2) bacteria counts were positively correlated with increased water level due to wind tides. CONCLUSIONS: Any type of estuarine shoreline development may result in closing of adjacent shellfishing waters. ND had bacterial counts second only to OD in spite of the retention of vegetated shoreline buffers and very new septic systems. As expected, the RM also failed to meet shellfish standards. Unexpectedly, only four of the 10 pseudomarina locations failed to meet the standards. Weekend boat use had no effect on bacterial counts. Surface runoff from rain and shoreline flooding from increased water levels increased bacterial counts, probably as a result of suspension of surface deposited faeces from wildlife and domestic animals. SIGNIFICANCE AND IMPACT OF THE STUDY: Multiple docking facilities do not necessarily result in violations of shellfish water quality standards. However, the elevated bacterial counts observed along the newly developed shore suggest caution in approving the practice of allowing individual 'oyster gardening' off private piers if the oysters are intended for human consumption. The practice of automatic closure of shellfish waters around RMs was supported. Correlations of bacterial counts with time following significant rainfall suggests a sampling strategy to separate local sources of bacteria from more remote sources thus focusing limited remedial resources more effectively.

Modelling of particle paths passing through an ultrasonic standing wave.

Within an ultrasonic standing wave particles experience acoustic radiation forces causing agglomeration at the nodal planes of the wave. The technique can be used to agglomerate, suspend, or manipulate particles within a flow. To control agglomeration rate it is important to balance forces on the particles and, in the case where a fluid/particle mix flows across the applied acoustic field, it is also necessary to optimise fluid flow rate. To investigate the acoustic and fluid forces in such a system a particle model has been developed, extending an earlier model used to characterise the 1-dimensional field in a layered resonator. In order to simulate fluid drag forces, CFD software has been used to determine the velocity profile of the fluid/particle mix passing through the acoustic device. The profile is then incorporated into a MATLAB model. Based on particle force components, a numerical approach has been used to determine particle paths. Using particle coordinates, both particle concentration across the fluid channel and concentration through multiple outlets are calculated. Such an approach has been used to analyse the operation of a microfluidic flow-through separator, which uses a half wavelength standing wave across the main channel of the device. This causes particles to converge near the axial plane of the channel, delivering high and low particle concentrated flow through two outlets, respectively. By extending the model to analyse particle separation over a frequency range, it is possible to identify the resonant frequencies of the device and associated separation performance. This approach will also be used to improve the geometric design of the microengineered fluid channels, where the particle model can determine the limiting fluid flow rate for separation to occur, the value of which is then applied to a CFD model of the device geometry.

Dorsal hippocampal function in unreinforced spatial learning.

This study examined learning about the spatial environment by rats during a single 10 min period of exploration on an eight-arm radial maze. Because no specific behaviors were learned during this procedure, the existence of learned spatial information was inferred from its retarding effect on subsequent conditioned cue preference (CCP) learning on the same maze. Previous experiments have shown that this form of spatial learning, measured in this way, requires an intact fimbriafornix and functional N-methyl-D-aspartate receptors. However, in the present experiments, large neurotoxic lesions of the dorsal hippocampus that impaired win-shift learning failed to eliminate the retarding effect of exploration on CCP learning. This result was obtained in three independent replications. These findings fail to confirm the hypothesis that the hippocampus is involved in spatial learning when that learning occurs in the absence of reinforcers and does not produce any specific learned behaviors. Previous work showed that this form of "pure" spatial learning requires an intact fimbria-fornix for acquisition but not for expression; the present findings suggest that the hippocampus is not required for either of these processes. The fimbria-fornix may interact with other temporal lobe structures in mediating this form of learning. The function of the hippocampus may be limited in some way to situations that involve reinforcers and/or situations in which specific behaviors are learned.

cato encodes a basic helix-loop-helix transcription factor implicated in the correct differentiation of Drosophila sense organs.

In Drosophila neurogenesis, proneural genes encode bHLH proteins that are required for neural precursor selection. But many vertebrate homologues are expressed later and are postulated to have multiple roles during neurogenesis. We have isolated a new Drosophila gene, cato, which encodes a protein with a bHLH domain that is closely related to that of the proneural protein Atonal. cato expression is restricted to the developing PNS, where it is expressed in between the stages of precursor selection and terminal differentiation (and therefore later than the proneural genes). We present evidence from loss-of-function and misexpression experiments that cato is involved in sensory neurone morphology. Moreover, in prospero mutants, in which axon and dendrite outgrowth is defective, cato is strongly derepressed in the developing CNS.

Impaired preference conditioning after anterior temporal lobe resection in humans.

Research with animals suggests that structures within the amygdaloid nuclear complex (ANC) are critical for acquiring associations between rewarding events and neutral stimuli, a form of conditioning often manifested in a subsequent preference for those (conditioned) stimuli. In this study, we investigated the relationship between the ANC and preference learning in humans. Three abstract monochrome patterns were presented to each subject over 180 trials in the context of a counting task requiring working memory. One pattern was paired with food reward on 90% of the trials in which it was presented and with no food reward on the other 10% of trials. The other patterns were similarly reinforced, but at ratios of 50:50% and 10:90% with reward and nonreward, respectively. Subsequently, a group of 21 normal participants preferred the pattern paired most often with reward to that paired least often with reward, and they did not explicitly relate their preferences to the conditioning procedure, but instead attributed them to the characteristics of the patterns themselves. Unlike the normal controls, a group of patients with unilateral surgical lesions that included the ANC (15 left, 18 right) did not show conditioned preferences, but performed normally on a measure of working memory. In contrast, 13 patients with unilateral damage confined to frontal cortex exhibited normal conditioned preferences but were impaired on the working memory task. This double dissociation provides clear evidence that, in humans as in other animals, reward-related learning (conditioned reward) critically depends on a circuit involving inferotemporal cortex and the ANC.

Drosophila atonal controls photoreceptor R8-specific properties and modulates both receptor tyrosine kinase and Hedgehog signalling.

During Drosophila eye development, the proneural gene atonal specifies founding R8 photoreceptors of individual ommatidia, evenly spaced relative to one another in a pattern that prefigures ommatidial organisation in the mature compound eye. Beyond providing neural competence, however, it has remained unclear to what extent atonal controls specific R8 properties. We show here that reduced Atonal function gives rise to R8 photoreceptors that are functionally compromised: both recruitment and axon pathfinding defects are evident. Conversely, prolonged Atonal expression in R8 photoreceptors induces defects in inductive recruitment as a consequence of hyperactive EGFR signalling. Surprisingly, such prolonged expression also results in R8 pattern formation defects in a process associated with both Hedgehog and Receptor Tyrosine Kinase signalling. Our results strongly suggest that Atonal regulates signalling and other properties of R8 precursors.

Progesterone and prostaglandin H synthase-2 involvement in alcohol-induced preterm birth in mice.

BACKGROUND: Recently, an association between alcohol consumption during pregnancy and shortened gestational length has been reported, but the underlying mechanisms remain unknown. Progesterone (P4) and prostaglandins have been shown to play important roles in parturition in both human and animal models. Recently, it has been suggested that prostaglandin H synthase-2 (PGHS-2) is responsible for prostaglandin changes associated with term and preterm labor. It is possible that alcohol induces preterm birth by altering P4 or PGHS-2 levels. These studies were designed to determine the role of P4 and PGHS-2 in alcohol-induced preterm labor in mice. METHODS: Experiment 1: Pregnant dams treated with either vehicle or alcohol (6 g/kg, intragastrically) on gestational day (GD) 16 were killed at various times in gestation up to the time of delivery. Plasma P4 levels were measured by radioimmunoassay and uterine PGHS-2 mRNA expression was measured by Ribonuclease Protection Assay. Results indicated that alcohol treatment was associated with an earlier decline in plasma P4 levels and an earlier rise in uterine PGHS-2 mRNA levels during gestation. Experiment 2: Pregnant C57BL/6J females were treated with either P4 (2.0 mg, subcutaneously) or vehicle (sesame oil) 2 hr before receiving either 6 g/kg alcohol (intragastrically) or vehicle (isocaloric sucrose) on gestational day (GD) 16. Results indicate that P4 pretreatment effectively antagonized alcohol-induced preterm delivery. Experiment 3: On GD16, pregnant dams received either 100 mg/kg nimesulide (a specific PGHS-2 inhibitor) or vehicle (saline) subcutaneously, 2 hr before treatment with either 6 g/kg alcohol (given intragastrically) or isocaloric sucrose. Nimesulide was effective in antagonizing alcohol-induced preterm labor. CONCLUSIONS: Together, these data suggest that both P4 and PGHS-2 may play roles in alcohol-induced preterm birth.

Alcohol plus cocaine prenatally is more deleterious than either drug alone.

A C57BL/6J mouse model was used to examine the coteratology of alcohol and cocaine. Plugged female mice were assigned to one of four treatment groups: control, cocaine only, alcohol only, or alcohol-cocaine. Experimental animals were treated from gestation day (GD) 6-18 and were killed the morning of GD 19. Alcohol was administered in a liquid diet containing 25% ethanol-derived calories (25% EDC), and cocaine was administered daily in subcutaneous injections of 60 mg/kg. All groups were pair-fed to the alcohol-cocaine group. The results showed that the cocaine-only and the alcohol-cocaine group had fewer successful pregnancies. The alcohol-only group had the lowest maternal weight gain from GD 1-19. There were no treatment group effects on litter size, sex ratio, or prenatal mortality. Importantly, fetuses in the alcohol-cocaine group weighed less than all other groups and had the greatest occurrence of fetal anomalies. These data confirm the teratogenic effects of alcohol and cocaine and suggest that the combination of the two drugs, if administered chronically, is more deleterious to pregnancy and fetal outcome than either drug alone.

Effects of medial and lateral caudate-putamen lesions on place- and cue-guided behaviors in the water maze: relation to thigmotaxis.

Rats with dorsomedial or dorsolateral caudate-putamen lesions and sham-operated controls were trained on the standard hidden platform (place) task in the water maze. Compared to controls, rats with dorsomedial, but not dorsolateral lesions were slower to escape to the hidden platform and spent significantly more time swimming near the wall of the pool (thigmotaxis) on the early trials, but eventually achieved control levels of performance. When the platform was removed from the pool, all groups exhibited a significant bias for swimming in the training quadrant and crossing the former location of the platform. In the second phase of the experiment rats were given visible platform (cue) training in a different room/pool with the platform moved to a new location each day. Rats with dorsomedial, but not dorsolateral lesions required more trials to reach criterion; again, thigmotaxis was observed on the early trials. The third phase, carried out in the original room/pool, included a place-retention trial followed by a place-cue competition test, (i.e. a choice between the learned spatial location of the hidden platform and the visible platform in a new location). The rats with dorsomedial, but not dorsolateral lesions swam to the visible platform more frequently than the controls. In the final phase, the rats in both lesion groups exhibited slightly lower thigmotactic tendencies than controls in a standard dry-land open field, a finding inconsistent with the hypothesis that thigmotaxis in the water maze is due to increased fear or anxiety. Taken together with other behavioral and anatomical findings, the results suggest that the dorsomedial caudate-putamen, by virtue of its connections with limbic and prefrontal cortical regions, may mediate a response selection process that integrates cognitive information with stimulus-response tendencies.

Parallel information processing in the dorsal striatum: relation to hippocampal function.

We investigated the effects of localized medial and lateral CPu lesions and fornix/fimbria lesions on responses to a local cue and to behavior based on cognitive-spatial information in the water maze. Rats were trained concurrently on the cue (visible platform) and spatial (submerged platform) components of the task, followed by a test in which responses to the two types of information were dissociated by a measure of competing response tendencies. Bilateral lesions of lateral CPu did not affect acquisition of either cue or spatial responding but produced a preference for the spatial response on the competition test. Bilateral lesions of the medial CPu retarded but did not prevent learning both components and produced a preference for the cue response on the competition test. The latter effect was accompanied by increased thigmotaxis (swimming in the periphery of the pool), primarily during the early acquisition trials, which was attributed to an impaired ability to respond to learned spatial information. Fornix/fimbria lesions prevented spatial but not cue learning and produced a preference for the cue response on the competition test. Asymmetric lesions (unilateral hippocampus and contralateral medial CPu) produced mild retardation of acquisition of both the cue and spatial tasks and a preference for the cue response on the competition test. These findings dissociate the functions of the lateral and medial CPu and suggest that the hippocampus and medial CPu may be parts of a system that promotes responding based on learned cognitive-spatial information, particularly in competitive cue-place response situations.