A transformer-based diffusion probabilistic model for heart rate and blood pressure forecasting in Intensive Care Unit

BACKGROUND AND OBJECTIVE

Vital sign monitoring in the Intensive Care Unit (ICU) is crucial for enabling prompt interventions for patients. This underscores the need for an accurate predictive system. Therefore, this study proposes a novel deep learning approach for forecasting Heart Rate (HR), Systolic Blood Pressure (SBP), and Diastolic Blood Pressure (DBP) in the ICU.

METHODS

We extracted 24,886 ICU stays from the MIMIC-III database which contains data from over 46 thousand patients, to train and test the model. The model proposed in this study, Transformer-based Diffusion Probabilistic Model for Sparse Time Series Forecasting (TDSTF), merges Transformer and diffusion models to forecast vital signs. The TDSTF model showed state-of-the-art performance in predicting vital signs in the ICU, outperforming other models' ability to predict distributions of vital signs and being more computationally efficient. The code is available at https://github.com/PingChang818/TDSTF.

RESULTS

The results of the study showed that TDSTF achieved a Standardized Average Continuous Ranked Probability Score (SACRPS) of 0.4438 and a Mean Squared Error (MSE) of 0.4168, an improvement of 18.9% and 34.3% over the best baseline model, respectively. The inference speed of TDSTF is more than 17 times faster than the best baseline model.

CONCLUSION

TDSTF is an effective and efficient solution for forecasting vital signs in the ICU, and it shows a significant improvement compared to other models in the field.

Prediction of fluid flow in porous media by sparse observations and physics-informed PointNet

We predict steady-state Stokes flow of fluids within porous media at pore scales using sparse point observations and a novel class of physics-informed neural networks, called "physics-informed PointNet" (PIPN). Taking the advantages of PIPN into account, three new features become available compared to physics-informed convolutional neural networks for porous medium applications. First, the input of PIPN is exclusively the pore spaces of porous media (rather than both the pore and grain spaces). This feature diminishes required computer memory. Second, PIPN represents the boundary of pore spaces smoothly and realistically (rather than pixel-wise representations). Third, spatial resolution can vary over the physical domain (rather than equally spaced resolutions). This feature enables users to reach an optimal resolution with a minimum computational cost. The performance of our framework is evaluated by the study of the influence of noisy sensor data, pressure observations, and spatial correlation length.

Meta-computational techniques' for managing spare data: An application in off-pump heart surgery

BACKGROUND AND OBJECTIVES

This research looked at the key considerations to remember when selecting a model for working with sparse data. In the presence of sparse evidence, it proposes ideal conditions for conducting meta-analysis.

METHODS

Monte Carlo simulations were used to produce study results, and three forms of continuity correction were used in the research. Besides, meta-analytical approaches were used to measure the cumulative effect of treatment and estimate each method's efficiency. A clinical trial in off-pump surgery met the main objectives of this research. Meta-analysis methods were used to determine the outcome of postoperative risk results. After that, with a total population of 3030, Monte Carlo simulations were used to produce research data to run fixed and random-effect models with three continuity correction forms. The type of consistency adjustment used, group imbalances, statistical analysis used, and variance values between studies all affect meta-analytical methods' results.

RESULTS

MSE values for balanced groups are normally zero. While the Arc-sine variation approach does a decent job of coping with inconsistent results on the effect of treatment, it has concerns with boundary estimates of variance between tests. Furthermore, using continuity correction methods introduces bias and imprecise medication outcome calculations. The spectrum of statistical analysis, such as fixed effects and random effects, can be inferred as completely based on data in samples. The sensitivity analysis of correction decisions could increase the reliability of meta-analysis approaches by enabling researchers to analyze various effect estimation findings.

CONCLUSION

This research study can be expanded upon by identifying alternative approaches to continuity correction methods and resolving boundary estimate problems. The range of statistical analysis, such as fixed effects and random effects, can be entirely dependent on the samples' type of data. The sensitivity analysis of correction decisions could improve the efficiency of meta-analysis methods by allowing researchers to investigate a wide range of effect estimation results.

Sparse data embedding and prediction by tropical matrix factorization

BACKGROUND

Matrix factorization methods are linear models, with limited capability to model complex relations. In our work, we use tropical semiring to introduce non-linearity into matrix factorization models. We propose a method called Sparse Tropical Matrix Factorization (STMF) for the estimation of missing (unknown) values in sparse data.

RESULTS

We evaluate the efficiency of the STMF method on both synthetic data and biological data in the form of gene expression measurements downloaded from The Cancer Genome Atlas (TCGA) database. Tests on unique synthetic data showed that STMF approximation achieves a higher correlation than non-negative matrix factorization (NMF), which is unable to recover patterns effectively. On real data, STMF outperforms NMF on six out of nine gene expression datasets. While NMF assumes normal distribution and tends toward the mean value, STMF can better fit to extreme values and distributions.

CONCLUSION

STMF is the first work that uses tropical semiring on sparse data. We show that in certain cases semirings are useful because they consider the structure, which is different and simpler to understand than it is with standard linear algebra.

Quantifying the impacts of dams on riverine hydrology under non-stationary conditions using incomplete data and Gaussian copula models

Across the world, the assessment of environmental impacts attributable to infrastructure and development projects often require a comparison between observed post-impact outcomes with what "would have happened" in the absence of the impact (i.e., the counterfactual). Environmental impact assessment (EIA) methods traditionally determine the counterfactual based on strong assumptions of stationarity (e.g., using before and after comparisons) and can be particularly challenging to use in the context of substantial data gaps, a vexing problem when combining several time-series data from different sources. Here we propose and test a widely applicable statistical approach for quantifying environmental impacts that avoids the stationarity assumption and circumvents issues associated with data gaps. Specifically, we used a Gaussian Copula (GC) model to assess the hydrological impacts of the Tucuruí dam on the Tocantins River in the Brazilian Amazon. Using multi-source water level and climate data, GC predictions of pre-dam hydrology for the validation period were excellent (Nash-Sutcliffe coefficients of 0.83 to 0.98 and 93-96% of observations within the 95% predictive intervals). In the post-dam period, the river had higher dry-season water levels both upstream and downstream relative to the predicted counterfactual, and the timing and duration of wet-season drawdown was delayed and extended, substantially altering the flood pulse. These impacts were evident as far as 176 km away from the dam, highlighting widespread hydrological impacts. The GC model outperformed standard multiple regression models in representing predictive uncertainty while also avoiding the stationarity assumption and circumventing the issue of sparse and incomplete data. We thus believe the GC approach has wide utility for integrating disparate time-series data to quantify the impacts of dams and other anthropogenic phenomena on riverine hydrology globally.

Studies in the use of data mining, prediction algorithms, and a universal exchange and inference language in the analysis of socioeconomic health data

While clinical and biomedical information in digital form has been escalating, it is socioeconomic factors that are important determinants of health on the national and global scale. We show how collective use of data mining and prediction algorithms to analyze socioeconomic population health data can stand beside classical correlation analysis in routine data analysis. The underlying theoretical basis is the Dirac notation and algebra that is a scientific standard but unusual outside of the physical sciences, combined with a theory of expected information first developed for analyzing sparse data but still largely confined to bioinformatics. The latter was important here because the records analyzed (which are for US counties and equivalents, not patients) are very few by contemporary data mining standards. The approach is very unlikely to be familiar to socioeconomic researchers, so the theory and the advantages of our inference nets over the Bayes Net are reviewed here, mostly using socioeconomic examples. While our expertise and focus is in regard to novel analytical methods rather than socioeconomics per se, a significant negative (countertrending) relationship between population health and equity was initially surprising, at least to the present authors. This encouraged deeper exploration including that of the relationship between our data mining methods and traditional Pearson's correlation. The latter is susceptible to giving wrong conclusions if a phenomenon called Simpson's paradox applies, so this is also investigated. Also discussed is that, even for very few records, associative data mining can still demand significant computational resources due to a combinatorial explosion.

Bayesian hierarchical model of ceftriaxone resistance proportions among Salmonella serotype Heidelberg infections

The National Antimicrobial Resistance Monitoring System (NARMS) at the Centers for Disease Control and Prevention tracks resistance among Salmonella infections. The annual number of Salmonella isolates of a particular serotype from states may be small, making direct estimation of resistance proportions unreliable. We developed a Bayesian hierarchical model to improve estimation by borrowing strength from relevant sampling units. We illustrate the models with different specifications of spatio-temporal interaction using 2004-2013 NARMS data for ceftriaxone-resistant Salmonella serotype Heidelberg. Our results show that Bayesian estimates of resistance proportions were smoother than observed values, and the difference between predicted and observed proportions was inversely related to the number of submitted isolates. The model with interaction allowed for tracking of annual changes in resistance proportions at the state level. We demonstrated that Bayesian hierarchical models provide a useful tool to examine spatio-temporal patterns of small sample size such as those found in NARMS.

The Yusuf-Peto method was not a robust method for meta-analyses of rare events data from antidepressant trials

OBJECTIVES

The aim of the study was to identify the validity of effect estimates for serious rare adverse events in clinical study reports of antidepressants trials, across different meta-analysis methods.

STUDY DESIGN AND SETTING

Four serious rare adverse events (all-cause mortality, suicidality, aggressive behavior, and akathisia) were meta-analyzed using different methods. The Yusuf-Peto odds ratio ignores studies with no events and was compared with the alternative approaches of generalized linear mixed models (GLMMs), conditional logistic regression, a Bayesian approach using Markov Chain Monte Carlo (MCMC), and a beta-binomial regression model.

RESULTS

The estimates for the four outcomes did not change substantially across the different methods; the Yusuf-Peto method underestimated the treatment harm and overestimated its precision, especially when the estimated odds ratio deviated greatly from 1. For example, the odds ratio for suicidality for children and adolescents was 2.39 (95% confidence interval = 1.32-4.33), using the Yusuf-Peto method but increased to 2.64 (1.33-5.26) using conditional logistic regression, to 2.69 (1.19-6.09) using beta-binomial, to 2.73 (1.37-5.42) using the GLMM, and finally to 2.87 (1.42-5.98) using the MCMC approach.

CONCLUSION

The method used for meta-analysis of rare events data influences the estimates obtained, and the exclusion of double-zero event studies can give misleading results. To ensure reduction of bias and erroneous inferences, sensitivity analyses should be performed using different methods instead of the Yusuf-Peto approach, in particular the beta-binomial method, which was shown to be superior through a simulation study.

Examining accommodation effects for equity by overcoming a methodological challenge of sparse data

BACKGROUND

As accommodation itself is an equity issue in varied contexts in and beyond education (e.g., the provision of assistive technology, extended time, or read-aloud), it is critical to examine the equitability of accommodation policies and practices by examining their effects on student performance.

AIMS

This study sought to assess the effectiveness of thirty-one bundled accommodations for students with learning disabilities, emotional or behavioral disorders, or multiple exceptionalities writing a provincial literacy test in Ontario, Canada.

METHODS AND PROCEDURES

We employed quantitative methods of log-linear analysis and odds ratio to examine the data. To analyze sparse data, we compared three different adjustment methods to meet this methodological challenge.

OUTCOMES AND RESULTS

Our findings suggest that the problems with sparse data can be overcome by an adjustment method. We also found that the likelihood of achieving the provincial standards may differ among students with special needs depending on whether they did or did not receive certain combinations of accommodations for the literacy test.

CONCLUSIONS AND IMPLICATIONS

We recommend that education stakeholders review the accommodations that produced significant differential effects to address the concerns regarding whether the test results were interpreted validly and fairly for students with special needs.

Computational Modeling for Enhancing Soft Tissue Image Guided Surgery: An Application in Neurosurgery

With the recent advances in computing, the opportunities to translate computational models to more integrated roles in patient treatment are expanding at an exciting rate. One area of considerable development has been directed towards correcting soft tissue deformation within image guided neurosurgery applications. This review captures the efforts that have been undertaken towards enhancing neuronavigation by the integration of soft tissue biomechanical models, imaging and sensing technologies, and algorithmic developments. In addition, the review speaks to the evolving role of modeling frameworks within surgery and concludes with some future directions beyond neurosurgical applications.

Using Retrospective Sampling to Estimate Models of Relationship Status in Large Longitudinal Social Networks

Estimation of longitudinal models of relationship status between all pairs of individuals (dyads) in social networks is challenging due to the complex inter-dependencies among observations and lengthy computation times. To reduce the computational burden of model estimation, a method is developed that subsamples the "always-null" dyads in which no relationships develop throughout the period of observation. The informative sampling process is accounted for by weighting the likelihood contributions of the observations by the inverses of the sampling probabilities. This weighted-likelihood estimation method is implemented using Bayesian computation and evaluated in terms of its bias, efficiency, and speed of computation under various settings. Comparisons are also made to a full information likelihood-based procedure that is only feasible to compute when limited follow-up observations are available. Calculations are performed on two real social networks of very different sizes. The easily computed weighted-likelihood procedure closely approximates the corresponding estimates for the full network, even when using low sub-sampling fractions. The fast computation times make the weighted-likelihood approach practical and able to be applied to networks of any size.

Marginally specified priors for non-parametric Bayesian estimation

Prior specification for non-parametric Bayesian inference involves the difficult task of quantifying prior knowledge about a parameter of high, often infinite, dimension. A statistician is unlikely to have informed opinions about all aspects of such a parameter but will have real information about functionals of the parameter, such as the population mean or variance. The paper proposes a new framework for non-parametric Bayes inference in which the prior distribution for a possibly infinite dimensional parameter is decomposed into two parts: an informative prior on a finite set of functionals, and a non-parametric conditional prior for the parameter given the functionals. Such priors can be easily constructed from standard non-parametric prior distributions in common use and inherit the large support of the standard priors on which they are based. Additionally, posterior approximations under these informative priors can generally be made via minor adjustments to existing Markov chain approximation algorithms for standard non-parametric prior distributions. We illustrate the use of such priors in the context of multivariate density estimation using Dirichlet process mixture models, and in the modelling of high dimensional sparse contingency tables.

Regularization Design and Control of Change Admission in Prior-Image-based Reconstruction

Nearly all reconstruction methods are controlled through various parameter selections. Traditionally, such parameters are used to specify a particular noise and resolution trade-off in the reconstructed image volumes. The introduction of reconstruction methods that incorporate prior image information has demonstrated dramatic improvements in dose utilization and image quality, but has complicated the selection of reconstruction parameters including those associated with balancing information used from prior images with that of the measurement data. While a noise-resolution tradeoff still exists, other potentially detrimental effects are possible with poor prior image parameter values including the possible introduction of false features and the failure to incorporate sufficient prior information to gain any improvements. Traditional parameter selection methods such as heuristics based on similar imaging scenarios are subject to error and suboptimal solutions while exhaustive searches can involve a large number of time-consuming iterative reconstructions. We propose a novel approach that prospectively determines optimal prior image regularization strength to accurately admit specific anatomical changes without performing full iterative reconstructions. This approach leverages analytical approximations to the implicitly defined prior image-based reconstruction solution and predictive metrics used to estimate imaging performance. The proposed method is investigated in phantom experiments and the shift-variance and data-dependence of optimal prior strength is explored. Optimal regularization based on the predictive approach is shown to agree well with traditional exhaustive reconstruction searches, while yielding substantial reductions in computation time. This suggests great potential of the proposed methodology in allowing for prospective patient-, data-, and change-specific customization of prior-image penalty strength to ensure accurate reconstruction of specific anatomical changes.