SEARCH · PubMed Health

Results for “Explainability”

Explore indexed PubMed citations for clinical trials, systematic reviews and public health research. Read source abstracts and follow each citation to its original PubMed record.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 19 recordsLinked to original sources

PubMed1991

HF-Explain: a natural language generation system for explaining a medical expert system.

Causal models have been used, with considerable success, to reason in the medical domain. While these systems typically have a robust reasoning mechanism and knowledge base about their specific area of expertise, their ability to satisfactorily explain their results in a meaningful, coherent and concise manner has been less impressive then their diagnostic capabilities. This paper describes a program, HF-Explain, that generates natural language explanations of one such system--the Heart Failure Program. HF-Explain, is loosely based on work done by McKeown in the Text system, using augmented transition networks (ATN) as a formalism to guide the explanation process. The result is a coherent, concise, accurate and rich explanation of Heart Failure Programs' diagnostic hypotheses.

Cardiac Output, Low↗

PubMed1991

Surface exclusion and molecular mobility may explain Vroman effects in protein adsorption.

Data on protein adsorption usually show that for increasing surface coverage the adsorption velocity decreases much faster than linearly. This contrasts to the classical Langmuir model with an adsorption velocity proportional to the number of unoccupied binding sites. It has been shown that this non-linearity may explain phenomena like transient adsorption of different proteins from a protein mixture or dilution-dependent changes in binding properties, collectively called Vroman effects. However, the molecular mechanisms explaining this non-linear behavior remain to be established. A Monte Carlo simulation model is presented that incorporates steric hindrance, lateral mobility and mutual interactions of adsorbed molecules. Experimental data on the adsorption kinetics of prothrombin and annexin V, a recently discovered anticoagulant protein, at phospholipid bilayers are analyzed with this model. A major conclusion is that the steep decline in adsorption rates for increasing surface coverage can be explained, without assuming repulsive forces between adsorbed molecules, as a surface exclusion effect combined with lateral mobility of adsorbed molecules. The fact that annexin V shows this effect to a much lesser degree than prothrombin is tentatively explained by clustering of adsorbed annexin V molecules. A qualitative effect of lateral mobility on the adsorption characteristics, predicted by the model, is confirmed in experiments in which the fluidity of the bilayers was manipulated.

Adsorption↗

PubMed2026

Data-centric, robust, and explainable multimodal deep learning for clinical decision support: A systematic review.

PURPOSE: Multimodal deep learning is increasingly proposed for clinical decision support (CDS) under a "data-centric" framing that prioritizes label quality, missing-modality robustness, distribution shift, calibration, and explainability. Prior reviews have examined multimodal medical AI, CDS, and data-centric methods separately, but none address their intersection. We mapped the modalities, fusion strategies, and data-centric and explainability techniques used in this recent literature, quantified how often each is implemented rather than merely mentioned, assessed deployment-relevant evidence (external validation, clinical-outcome measurement, equity), and formally appraised study-level risk of bias. METHODS: Following the PRISMA 2020 statement (PROSPERO CRD420261427815; registered retrospectively), we screened 150 records and included primary, clinical, multimodal studies that applied machine or deep learning to a decision-support task and reported at least one quantitative result. Two reviewers screened and extracted data with consensus adjudication. Each study was coded against pre-specified operational definitions, separating implemented or empirically evaluated techniques from those only mentioned. Study-level risk of bias was assessed with PROBAST + AI. Synthesis was narrative. RESULTS: Thirty-one studies met inclusion; 30 (97%) were published between 2024 and 2026, with a median of three modalities (range 2-6), most commonly structured EHR (71%) and imaging (39%). Data-centric techniques were frequently reported (74-84% across label-noise, distribution-shift, calibration, missing-modality and class-imbalance handling; equity 61%). However, external validation was reported in only 4/31 studies (13%), a clinical or provider outcome in 3/31 (10%), and no study reported routine deployment. Overall risk of bias was high in 27/31 studies (87%), driven by the analysis domain. CONCLUSION: Within this recent, self-selected slice of the field, technical robustness and explainability techniques are widely reported but rarely validated out-of-distribution or against clinical outcomes, and the underlying evidence is at high risk of bias. Progress requires external multi-site validation, clinical-outcome measurement, formal bias appraisal, and adherence to AI reporting standards (e.g., TRIPOD + AI) before deployment can be justified.

Deep Learning↗

PubMed2025

CaXML: Chemistry-informed machine learning explains mutual changes between protein conformations and calcium ions in calcium-binding proteins using structural and topological features.

Proteins' flexibility is a feature in communicating changes in cell signaling instigated by binding with secondary messengers, such as calcium ions, associated with the coordination of muscle contraction, neurotransmitter release, and gene expression. When binding with the disordered parts of a protein, calcium ions must balance their charge states with the shape of calcium-binding proteins and their versatile pool of partners depending on the circumstances they transmit. Accurately determining the ionic charges of those ions is essential for understanding their role in such processes. However, it is unclear whether the limited experimental data available can be effectively used to train models to accurately predict the charges of calcium-binding protein variants. Here, we developed a chemistry-informed, machine-learning algorithm that implements a game theoretic approach to explain the output of a machine-learning model without the prerequisite of an excessively large database for high-performance prediction of atomic charges. We used the ab initio electronic structure data representing calcium ions and the structures of the disordered segments of calcium-binding peptides with surrounding water molecules to train several explainable models. Network theory was used to extract the topological features of atomic interactions in the structurally complex data dictated by the coordination chemistry of a calcium ion, a potent indicator of its charge state in protein. Our design created a computational tool of CaXML, which provided a framework of explainable machine learning model to annotate ionic charges of calcium ions in calcium-binding proteins in response to the chemical changes in an environment. Our framework will provide new insights into protein design for engineering functionality based on the limited size of scientific data in a genome space.

Machine Learning↗

PubMed1991

A model for diagnosing and explaining multiple disorders.

The ability to diagnose multiple interacting disorders and explain them in a coherent causal framework has only partially been achieved in medical expert systems. This paper proposes a causal model for diagnosing and explaining multiple disorders whose key elements are: physician-directed hypotheses generation, object-oriented knowledge representation, and novel explanation heuristics. The heuristics modify and link the explanations to make the physician aware of diagnostic complexities. A computer program incorporating the model currently is in use for diagnosing peripheral nerve and muscle disorders. The program successfully diagnoses and explains interactions between diseases in terms of underlying pathophysiologic concepts. The model offers a new architecture for medical domains where reasoning from first principles is difficult but explanation of disease interactions is crucial for the system's operation.

Acid-Base Imbalance↗

PubMed1991

Clinical and pathophysiological observations in migraine and tension-type headache explained by integration of vascular, supraspinal and myofascial inputs.

A vascular-supraspinal-myogenic (VSM) model for pain in migraine based on our previous clinical and pathophysiological observations is proposed. According to the model, perceived pain (headache) intensity is determined by the sum of nociception from cephalic arteries and pericranial myofascial tissues converging upon the same neurons and integrated with supraspinal effects (usually facilitating). Vascular input predominates over myofascial input in migraine, whereas significance of supraspinal facilitation is difficult to estimate. The importance of these 3 effects may vary between patients and in the same individual with time. The model is in accordance with recent experimental studies showing convergence of somatovisceral afferents upon n. caudalis neurons. Also, long term potentiation due to nociceptive activation and sensitization of neurons to input from wider areas and non-nociceptive stimuli are relevant to our model. In tension-type headache, nociception is primarily myofascial, but vascular input cannot be disregarded. Supraspinal facilitation probably plays a large, sometimes dominant role (the MSV model). The model explains much of the complexity of the clinical picture of these disorders as well as their tendency to overlap and to change into one another. Also, a number of pathophysiological observations such as why muscles are tender during migraine, why trigger-point injection may cure migraine attacks and why chronic tension-type headache is often associated with episodes of pulsating pain, can be explained. The model gives a rational explanation of empirically developed, internationally accepted, multimodal treatment strategies for migraine and tension-type headache. It may thus serve a useful purpose in explaining the disorder to patients. Finally, the model points to several avenues of future research in animals and man.

Animals↗

PubMed1992

The neurohormonal hypothesis: a theory to explain the mechanism of disease progression in heart failure.

Because physicians have traditionally considered heart failure to be a hemodynamic disorder, they have described the syndrome of heart failure using hemodynamic concepts and have designed treatment strategies to correct the hemodynamic derangements of the disease. However, although hemodynamic abnormalities may explain the symptoms of heart failure, they are not sufficient to explain the progression of heart failure and, ultimately, the death of the patient. Therapeutic interventions may improve the hemodynamic status of patients but adversely affect their long-term outcome. These findings have raised questions about the validity of the hemodynamic hypothesis and suggest that alternative mechanisms must play a primary role in advancing the disease process. Several lines of evidence suggest that neurohormonal mechanisms play a central role in the progression of heart failure. Activation of the sympathetic nervous system and renin-angiotensin system exerts a direct deleterious effect on the heart that is independent of the hemodynamic actions of these endogenous mechanisms. Therapeutic interventions that block the effects of these neurohormonal systems favorably alter the natural history of heart failure, and such benefits cannot be explained by the effect of these treatments on cardiac contractility and ejection fraction. Conversely, pharmacologic agents that adversely influence neurohormonal systems in heart failure may increase cardiovascular morbidity and mortality, even though they exert favorable hemodynamic effects. These observations support the formulation of a neurohormonal hypothesis of heart failure and provide the basis for the development of novel therapeutic strategies in the next decade.

Adrenergic beta-Antagonists↗

PubMed1992

[Sudden infant death: from the unexplained to the explained].

In France, 1,300 to 1,500 infants die suddenly each year. Are these deaths explained? Most clinicians agree that in more than two-thirds of the cases death cannot be formally explained by the clinical or paraclinical context or by the findings at post-mortem examination. These infants are the victims of the "sudden infant death syndrome." This syndrome consists of a transient abnormality in the maturation of the vegetative nervous system function, upon which are superimposed non-specific elements that facilitate or precipitate death. The lack of routine examination to detect this background explains why sudden infant deaths cannot be predicted in the majority of cases. Because there is no method sufficiently accurate to diagnose the abnormality of maturation, these deaths cannot for the moment be firmly ascribed to this abnormality.

Autonomic Nervous System↗

PubMed1992

Can age-related decline in speech understanding be explained by peripheral hearing loss?

Speech audiometric scores were compared across the age range from 50 to 90 years in 137 subjects selected in such a way that average audiometric thresholds were matched across four age groups. Thus any age-related changes in speech audiometric scores could not be attributed to age-related differences in peripheral hearing sensitivity. Four speech audiometric measures were studied; phonemically-balanced words (PB), Speech Perception in Noise (SPIN) test for both high- and low-predictability sentences, and the Synthetic Sentence Identification (SSI) test. There was an age trend for all four speech measures but only the change in SSI (30%) was statistically significant. The argument that the change in SSI can be explained by subtle changes in the auditory periphery, not reflected in audiometric thresholds, is weakened by the fact that the change in SSI was greater than the change in either of the two monosyllabic word tests (PB and SPIN-low). The argument that the change in SSI can be explained by concomitant cognitive decline is not supported by correlations among SSI performance and any of several neuropsychological measures of cognitive function in the same subjects. Finally, the lack of a significant interactive effect between hearing sensitivity level and cognitive status does not support a model in which a peripherally degraded speech signal interacts with a deficit in cognitive function to produce the decline in speech audiometric scores. We conclude that the observed age-related decline in SSI performance cannot be satisfactorily explained by peripheral hearing sensitivity loss.

Aged↗

PubMed2026

From prediction to mechanism: Explainable AI uncovers plasma and CSF proteomic signatures of Alzheimer's disease.

Alzheimer's disease (AD) plasma and cerebrospinal fluid (CSF) proteomics can distinguish AD from cognitively normal controls, but the generalizability of machine learning performance and the recurrence of biological signals across datasets require cautious interpretation. We developed an explainable artificial intelligence framework spanning two fluids and four ADNI proteomic datasets, covering 2082 modality specific samples, all analysed internally within ADNI. Phase 1 analysed plasma using a 119 analyte NULISA and targeted UPENN panel (n = 727; 216 CE, 511 controls). Phase 2 extended the analysis to CSF using SOMAscan7k, TMT-MS and targeted SET2, with Elecsys Aβ42, Aβ40, total tau and p-tau181 as anchor biomarkers. Only SOMAscan was subject-independent relative to Phase 1 plasma; TMT-MS and SET2 overlapped with Phase 1 for 96.0% and 97.7% of subjects and therefore are not independent replication cohorts. Under subject-level splits with fold internal preprocessing, we compared Elastic Net, Explainable Boosting Machines and gradient boosted trees with SHAP-based explanations. Among the candidate pipelines, we selected the pipeline with the highest held-out test ROC AUC for each platform; the selected values were 0.927 in plasma and 0.954-0.973 across the three CSF datasets. Because the same held out test performance was used for pipeline selection and headline reporting, these are optimistically selected single-holdout estimates, not unbiased estimates of generalizable or clinical performance. Explanations identified five recurring biological axes within ADNI: cholinergic (ACHE), tau/14-3-3 (YWHAG, YWHAZ, YWHAB, YWHAE), neuro-axonal (NEFL, NEFH), microglial/complement (CHIT1, SMOC1, CHI3L1, C7, CFH) and synaptic (NPTXR, NPTX2, DLG4, SYT5, VSNL1, ELAVL2). CSF analyses showed synaptic vesicle-cycle enrichment (q = 2 × 10-6), and CSF YWHAG correlated strongly with total tau (ρ = 0.87). Cross-fluid directional concordance was modest overall (54-57%) but increased to 73-80% among mapped analyte/protein rows reaching q < 0.05 in CSF. These findings provide hypothesis-generating, internally supported evidence within ADNI. Independent external cohorts with locked pipelines are required to evaluate generalizable performance and biological reproducibility; the overlapping TMT-MS and SET2 analyses should not be interpreted as independent replication.

Alzheimer Disease↗

PubMed1978

A tunnelling model to explain the reduction of ferricytochrome c by H and OH radicals.

The kinetics of the reaction of OH radicals with ferricytochrome c was studied in the time range 1 microsecond to 1 s by means of pulse radiolysis. The OH radicals reduce ferricytochrome c by 40% +/- 10%. The time course of the reduction is explained by a mechanism whereby a radical formed after hydrogen has been abstracted from the outer surface of the protein reduces the iron by electron tunnelling. We have calculated that the reducing electron in the radical is bound with an energy of at least 1.75 eV and that the frequency factor of the tunnelling process is v=10(11.5)s-1. This model accounts for the observed absorbance change in time range 5 . 10(-6)--10(-1)s. The time course of the reduction of ferricytochrome c by H radicals (Lichtin, N.N., Shafferman A. and Stein, G. (1974) Biochim. Biophys. Acta 357, 386--398) is explained by the same model.

Calorimetry↗

PubMed1992

General slowing alone cannot explain age-related search effects: reply to Cerella (1991)

Cerella (1991) has argued that the performance of older adults in the Fisk and Rogers (1991) study is a linear function of the performance of younger adults that is independent of task-specific cognitive requirements. We demonstrate that this is not the case. First, we show that the scatter plot analyses used by Cerella can hide the very task-specific age-related slowing they were designed to reveal. Second, we demonstrate that the percentage of variance explained by such analyses can be misleading. Third, we show that there are reliable differences across tasks in the parameters relating younger and older adults' performance. Finally, we argue that the general, task-independent proportionate slowing that Cerella suggested explains so much of the variance in age-related performance is actually an average slowing that is a function of a relatively small task-independent and a relatively large task-dependent factor.

Adult↗

PubMed2025

SetBERT: the deep learning platform for contextualized embeddings and explainable predictions from high-throughput sequencing.

MOTIVATION: High-throughput sequencing (HTS) is a modern sequencing technology used to profile microbiomes by sequencing thousands of short genomic fragments from the microorganisms within a given sample. This technology presents a unique opportunity for artificial intelligence to comprehend the underlying functional relationships of microbial communities. However, due to the unstructured nature of HTS data, nearly all computational models are limited to processing DNA sequences individually. This limitation causes them to miss out on key interactions between microorganisms, significantly hindering our understanding of how these interactions influence the microbial communities as a whole. Furthermore, most computational methods rely on post-processing of samples which could inadvertently introduce unintentional protocol-specific bias. RESULTS: Addressing these concerns, we present SetBERT, a robust pre-training methodology for creating generalized deep learning models for processing HTS data to produce contextualized embeddings and be fine-tuned for downstream tasks with explainable predictions. By leveraging sequence interactions, we show that SetBERT significantly outperforms other models in taxonomic classification with genus-level classification accuracy of 95%. Furthermore, we demonstrate that SetBERT is able to accurately explain its predictions autonomously by confirming the biological-relevance of taxa identified by the model. AVAILABILITY AND IMPLEMENTATION: All source code is available at https://github.com/DLii-Research/setbert. SetBERT may be used through the q2-deepdna QIIME 2 plugin whose source code is available at https://github.com/DLii-Research/q2-deepdna.

Deep Learning↗

PubMed1991

Multilevel modelling of longitudinal cephalometric data explained for orthodontists.

Multilevel modelling of longitudinal data is an important new statistical technique. In this article some of the basic concepts and ideas of multilevel modelling are explained. The model is introduced by showing how individual and average growth can be modelled. The intercept, linear and quadratic coefficient, between and within variance, fixed and random part, and other concepts of multilevel modelling are explained. Attention is also given to the reading of statistical tables of the results of multilevel analysis. In the conclusion some of the advantages of multilevel modelling of cephalometric data are mentioned.

Aging↗

PubMed2004

Genomic regionality in rates of evolution is not explained by clustering of genes of comparable expression profile.

In mammalian genomes, linked genes show similar rates of evolution, both at fourfold degenerate synonymous sites (K4) and at nonsynonymous sites (KA). Although it has been suggested that the local similarity in the synonymous substitution rate is an artifact caused by the inclusion of disparately evolving gene pairs, we demonstrate here that this is not the case: after removal of disparately evolving genes, both (1) linked genes and (2) introns from the same gene have more similar silent substitution rates than expected by chance. What causes the local similarity in both synonymous and nonsynonymous substitution rates? One class of hypotheses argues that both may be related to the observed clustering of genes of comparable expression profile. We investigate these hypotheses using substitution rates from both human-mouse and mouse-rat comparisons, and employing three different methods to assay expression parameters. Although we confirm a negative correlation of expression breadth with both K4 and KA, we find no evidence that clustering of similarly expressed genes explains the clustering of genes of comparable substitution rates. If gene expression is not responsible, what about other causes? At least in the human-mouse comparison, the local similarity in KA can be explained by the covariation of KA and K4. As regards K4, our results appear consistent with the notion that local similarity is due to processes associated with meiotic recombination.

Animals↗

PubMed1991

Explaining outputs of primary health care: population and practice factors.

OBJECTIVE: To examine whether variations in the activities of general practice among family health service authorities can be explained by the populations characteristics and the organisation and resourcing of general practice. DESIGN: The family health services authorities were treated as discrete primary health care systems. Nineteen performance indicators reflecting the size, distribution, and characteristics of the population served; the organisation of general practice (inputs); and the activities generated by general practitioners and their staff (output) were analysed by stepwise regression. SETTING: 90 family health services authorities in England. MAIN OUTCOME MEASURES: Rates of cervical smear testing, immunisation, prescribing, and night visiting. RESULTS: 53% of the variation in uptake of cervical cytology was accounted for by Jarman score (t = -3.3), list inflation (-0.41), the proportion of practitioners over 65 (-0.64), the number of ancillary staff per practitioner (2.5), and 70% of the variation in immunisation rates by standardised mortality ratios (-6.6), the proportion of practitioners aged over 65 (-4.8), and the number of practice nurses per practitioner (3.5). Standardised mortality ratios (8.4), the number of practitioners (2.3), and the proportion over 65 (2.2), and the number of ancillary staff per practitioner (-3.1) accounted for 69% of variation in prescribing rates. 54% of the variation in night visiting was explained by standardised mortality ratios (7.1), the proportion of practitioners with lists sizes below 1000 (-2.2), the proportion aged over 65 (-0.4), and the number of practice nurses per practitioner (-2.5). CONCLUSIONS: Family health services authorities are appropriate systems for studying output of general practice. Their performance indicators need to be refined and to be linked to other relevant factors, notably the performance of hospital, community, and social services.

England↗

PubMed1992

Differential initiation of translation of a single estrogen receptor mRNA could explain some estradiol resistance cases.

Cell response to steroid stimulation is generally acknowledged to be mediated by an intracellular protein known as a receptor. Response intensity is related to the affinity of the receptor and to the number of sites occupied by its specific ligand. Although verified in the majority of experimental and clinical studies, certain phenomena of steroid hormone resistance would seem to challenge this assertion. Application of gene molecular biology to determine the action mechanisms of steroid hormones has partially explained cell resistance in terms of genetic modifications. The work presented here shows that in certain cases, estrogen resistance could be explained by regulation of translation of the single messenger RNA coding for the receptor.

Animals↗

PubMed1992

Are race and sex differences in lung function explained by frame size? The CARDIA Study.

Using the CARDIA cohort of 20- to 32-yr-old black and white men and women, FVC and FEV1 were standardized for standing height, sitting height, leg height, elbow breadth, and biacromial diameter in such a way that the standardized lung function showed minimal statistical dependence on these measures of frame size. Race and sex differences in lung function have been reported even after adjustment for height; however, these differences might depend on aspects of frame size other than height. We found that within this age group height2 provided robust standardization for FVC and FEV1 for all race and sex strata of the population. Height explained approximately 40% of the variance of FVC and FEV1 in whites, 30% in black women, and 20% in black men. In black men only, standardization for the combination of sitting height, leg height, elbow breadth, and biacromial diameter improved explained variance to nearly 40% for FVC and nearly 30% for FEV1. After standardization for height, FVC and FEV1 were found to be 14 to 19% higher in whites than in blacks, and in men than in women. Standardization of FVC and FEV1 for sitting height, leg height, elbow breadth, and biacromial diameter combined reduced these differences to 13-16%. Thus, race and sex differences in lung function exist even after detailed adjustment for frame size.

Adult↗