RESUMEN

Uncertainty quantification in machine learning can provide powerful insight into a model's capabilities and enhance human trust in opaque models. Well-calibrated uncertainty quantification reveals a connection between high uncertainty and an increased likelihood of an incorrect classification. We hypothesize that if we are able to explain the model's uncertainty by generating rules that define subgroups of data with high and low levels of classification uncertainty, then those same rules will identify subgroups of data on which the model performs well and subgroups on which the model does not perform well. If true, then the utility of uncertainty quantification is not limited to understanding the certainty of individual predictions; it can also be used to provide a more global understanding of the model's understanding of patient subpopulations. We evaluate our proposed technique and hypotheses on deep neural networks and tree-based gradient boosting ensemble across benchmark and real-world medical datasets.

Asunto(s)

RESUMEN

Precision medicine informatics is a field of research that incorporates learning systems that generate new knowledge to improve individualized treatments using integrated data sets and models. Given the ever-increasing volumes of data that are relevant to patient care, artificial intelligence (AI) pipelines need to be a central component of such research to speed discovery. Applying AI methodology to complex multidisciplinary information retrieval can support efforts to discover bridging concepts within collaborating communities. This dovetails with precision medicine research, given the information rich multi-omic data that are used in precision medicine analysis pipelines. In this perspective article we define a prototype AI pipeline to facilitate discovering research connections between bioinformatics and clinical researchers. We propose building knowledge representations that are iteratively improved through AI and human-informed learning feedback loops supported through crowdsourcing. To illustrate this, we will explore the specific use case of nonalcoholic fatty liver disease, a growing health care problem. We will examine AI pipeline construction and utilization in relation to bench-to-bedside bridging concepts with interconnecting knowledge representations applicable to bioinformatics researchers and clinicians.

RESUMEN

Fungi are the primary agents of terrestrial decomposition, yet our understanding of fungal biogeography lags far behind that of plants, animals and bacteria. Here, we use a trait-based approach to quantify the niches of 23 species of basidiomycete wood decay fungi from across North America, and explore the linkages among functional trait expression, climate and phylogeny. Our analysis reveals a fundamental trade-off between abiotic stress tolerance and competitive ability, whereby fungi with wide thermal and moisture niches exhibit lower displacement ability. The magnitude of this dominance-tolerance trade-off is partially related to the environmental conditions under which the fungi were collected, with thermal niche traits exhibiting the strongest climate relationships. Nevertheless, moisture and thermal dominance-tolerance patterns exhibited contrasting phylogenetic signals, suggesting that these trends are influenced by a combination of niche sorting along taxonomic lines in tandem with acclimation and adaptation at the level of the individual. Collectively, our work reveals key insight into the life history strategies of saprotrophic fungi, demonstrating consistent trait trade-offs across broad spatial scales.

Asunto(s)

RESUMEN

The distributed Graph rule automata for iterative linkage (dGrail) toolkit is a software package for deterministic record linkage. This toolkit allows for iterative development of linked record sets. While intended for the generation of gold standard record sets, the toolkit is applicable to any offline deterministic record linkage task. The dGrail toolkit embodies a flexible rule engine allowing the user to implement a wide variety of record matching rules, including those found in the literature.

Asunto(s)

RESUMEN

Triage is a key component of trauma care. Unfortunately, mistriage rates remain high. Machine learning techniques have the potential to improve triage. Our experiment showed that while decision tree induction was as accurate as the most widely accepted trauma triage guidelines, they performed differently with respect to over- and undertriage.

Asunto(s)

RESUMEN

OBJECTIVE: Authors evaluated whether displaying context sensitive links to infrequently accessed educational materials and patient information via the user interface of an inpatient computerized care provider order entry (CPOE) system would affect access rates to the materials. DESIGN: The CPOE of Vanderbilt University Hospital (VUH) included "baseline" clinical decision support advice for safety and quality. Authors augmented this with seven new primarily educational decision support features. A prospective, randomized, controlled trial compared clinicians' utilization rates for the new materials via two interfaces. Control subjects could access study-related decision support from a menu in the standard CPOE interface. Intervention subjects received active notification when study-related decision support was available through context sensitive, visibly highlighted, selectable hyperlinks. MEASUREMENTS: Rates of opportunities to access and utilization of study-related decision support materials from April 1999 through March 2000 on seven VUH Internal Medicine wards. RESULTS: During 4,466 intervention subject-days, there were 240,504 (53.9/subject-day) opportunities for study-related decision support, while during 3,397 control subject-days, there were 178,235 (52.5/subject-day) opportunities for such decision support, respectively (p = 0.11). Individual intervention subjects accessed the decision support features at least once on 3.8% of subject-days logged on (278 responses); controls accessed it at least once on 0.6% of subject-days (18 responses), with a response rate ratio adjusted for decision support frequency of 9.17 (95% confidence interval 4.6-18, p < 0.0005). On average, intervention subjects accessed study-related decision support materials once every 16 days individually and once every 1.26 days in aggregate. CONCLUSION: Highlighting availability of context-sensitive educational materials and patient information through visible hyperlinks significantly increased utilization rates for study-related decision support when compared to "standard" VUH CPOE methods, although absolute response rates were low.

Asunto(s)

RESUMEN

We present a model for a health resource locator to help rural primary healthcare providers care for patients. We identify some unique needs of rural providers, argue that a grassroots effort, driven by the community, is the optimal way to address some of those needs, and propose a centralized Internet-based system to drive the whole process.

Asunto(s)

RESUMEN

Predictive modeling based on gene expression data is complicated by the high dimensionality (number of genes) of microarray data given the number of available samples. We investigate a method for reducing the dimensionality of the data using singular value decomposition.

Asunto(s)

RESUMEN

We present an implementation model for pharmaceutical computerized decision support (CDS) that enables a hospital to incrementally target specific "high value" projects as needs are identified and support is secured. Our model, which we are currently implementing in a rural medical center, allows the hospital and its staff to quickly reap some benefits from CDS in spite of resource limitations.

Asunto(s)