There is an increasing interest in clinical prediction tools that can achieve high prediction accuracy and provide explanations of the factors leading to increased risk of adverse outcomes. However, ...approaches to explaining complex machine learning (ML) models are rarely informed by end-user needs and user evaluations of model interpretability are lacking in the healthcare domain. We used extended revisions of previously-published theoretical frameworks to propose a framework for the design of user-centered displays of explanations. This new framework served as the basis for qualitative inquiries and design review sessions with critical care nurses and physicians that informed the design of a user-centered explanation display for an ML-based prediction tool.
We used our framework to propose explanation displays for predictions from a pediatric intensive care unit (PICU) in-hospital mortality risk model. Proposed displays were based on a model-agnostic, instance-level explanation approach based on feature influence, as determined by Shapley values. Focus group sessions solicited critical care provider feedback on the proposed displays, which were then revised accordingly.
The proposed displays were perceived as useful tools in assessing model predictions. However, specific explanation goals and information needs varied by clinical role and level of predictive modeling knowledge. Providers preferred explanation displays that required less information processing effort and could support the information needs of a variety of users. Providing supporting information to assist in interpretation was seen as critical for fostering provider understanding and acceptance of the predictions and explanations. The user-centered explanation display for the PICU in-hospital mortality risk model incorporated elements from the initial displays along with enhancements suggested by providers.
We proposed a framework for the design of user-centered displays of explanations for ML models. We used the proposed framework to motivate the design of a user-centered display of an explanation for predictions from a PICU in-hospital mortality risk model. Positive feedback from focus group participants provides preliminary support for the use of model-agnostic, instance-level explanations of feature influence as an approach to understand ML model predictions in healthcare and advances the discussion on how to effectively communicate ML model information to healthcare providers.
ABSTRACT The precision of photometric and spectroscopic observations has been systematically improved in the last decade, mostly thanks to space-borne photometric missions and ground-based ...spectrographs dedicated to finding exoplanets. The field of eclipsing binary stars strongly benefited from this development. Eclipsing binaries serve as critical tools for determining fundamental stellar properties (masses, radii, temperatures, and luminosities), yet the models are not capable of reproducing observed data well, either because of the missing physics or because of insufficient precision. This led to a predicament where radiative and dynamical effects, insofar buried in noise, started showing up routinely in the data, but were not accounted for in the models. PHOEBE (PHysics Of Eclipsing BinariEs; http://phoebe-project.org) is an open source modeling code for computing theoretical light and radial velocity curves that addresses both problems by incorporating missing physics and by increasing the computational fidelity. In particular, we discuss triangulation as a superior surface discretization algorithm, meshing of rotating single stars, light travel time effects, advanced phase computation, volume conservation in eccentric orbits, and improved computation of local intensity across the stellar surfaces that includes the photon-weighted mode, the enhanced limb darkening treatment, the better reflection treatment, and Doppler boosting. Here we present the concepts on which PHOEBE is built and proofs of concept that demonstrate the increased model fidelity.
Severe sepsis in immunocompromised children is associated with increased mortality. This paper describes the epidemiology landscape, clinical acuity, and outcomes for severe sepsis in pediatric ...intestinal (ITx) and multi‐visceral (MVTx) transplant recipients requiring admission to the pediatric intensive care unit (PICU). Severe sepsis episodes were retrospectively reviewed in 51 ITx and MVTx patients receiving organs between 2009 and 2015. Twenty‐nine (56.8%) patients had at least one sepsis episode (total of 63 episodes) through December 2016. Bacterial etiologies accounted for 66.7% of all episodes (n = 42), occurring a median of 122.5 days following transplant (IQR 59–211.8 days). Multidrug‐resistant organisms (MDROs) accounted for 73.8% of bacterial infections; extended spectrum beta‐lactamase producers, vancomycin‐resistant enterococcus, and highly‐resistant Pseudomonas aeruginosa were the most commonly identified. Increased mechanical ventilation and vasoactive requirements were noted in MDRO episodes (OR 3.03, 95% CI 1.09–8.46 and OR 3.07, 95% CI 1.09–8.61, respectively; p < .05) compared to non‐MDRO episodes. PICU length of stay was significantly increased for MDRO episodes (7 vs. 3 days, p = .02). Graft loss was 24.1% (n = 7) and mortality was 24.1% (n = 7) in patients who experienced severe sepsis. Further attention is needed for MDRO risk mitigation and modification of sepsis treatment guidelines to ensure MDRO coverage for this population.
In pediatric intestinal and multi‐visceral transplant recipients, MDROs account for a large majority of severe sepsis episodes related to bacterial infections and are associated with an increased need for mechanical ventilation, use of vasoactive medications, and length of intensive care unit stay.
Background
We performed a propensity score matched study comparing patients’ short- and long-term results after laparoscopic omentum-preserving gastrectomy and open surgery with omentectomy with UICC ...stages 0–IV.
Methods
Between 2015 and 2022, 311 patients with gastric cancer underwent surgery at the University Clinical Centre Maribor. Of these, 249 met the inclusion criteria and 198 were included in the study group after PSM.
Results
Patients in both groups were well-balanced in demographic and pathological characteristics after PSM. There was no significant difference in the 5-year survival between groups (LAP: 62.2% vs. OPN: 54.4%;
p
= 0.950). The Cox regression model identified UICC stage and age as significant predictors for survival. In both groups, peritoneal dissemination was the most common site of recurrence. The multivariate analysis identified the UICC stage as a significant predictor for peritoneal recurrence, while omental preservation was not associated with a higher risk of peritoneal dissemination. Omentum preservation was not associated with more intestinal obstruction. Patients in the LAP group had significantly shorter hospital stays (LAP: 9(6) vs. OPN: 10(5);
p
= 0.009), less postoperative morbidity (LAP: 17% vs. OPN: 23.4%;
p
= 0.009), and significantly more extracted LNs per operation compared to open surgery (LAP: 31 ± 11 LNs vs. OPN: 25 ± 12 LNs;
p
= 0.002).
Conclusion
Based on our results, we recommend the use of laparoscopic omentum-preserving gastrectomy in patients with early and advanced gastric cancer.
Graphical abstract
There are no therapies shown to improve outcome after severe traumatic brain injury (TBI) in humans, a leading cause of morbidity and mortality. We sought to verify brain exposure of the systemically ...administered antioxidant N-acetylcysteine (NAC) and the synergistic adjuvant probenecid, and identify adverse effects of this drug combination after severe TBI in children.
IRB-approved, randomized, double-blind, placebo controlled Phase I study in children 2 to 18 years-of-age admitted to a Pediatric Intensive Care Unit after severe TBI (Glasgow Coma Scale GCS score ≤8) requiring an externalized ventricular drain for measurement of intracranial pressure (ICP). Patients were recruited from November 2011-August 2013. Fourteen patients (n = 7/group) were randomly assigned after obtaining informed consent to receive probenecid (25 mg/kg load, then 10 mg/kg/dose q6h×11 doses) and NAC (140 mg/kg load, then 70 mg/kg/dose q4h×17 doses), or placebos via naso/orogastric tube. Serum and CSF samples were drawn pre-bolus and 1-96 h after randomization and drug concentrations were measured via UPLC-MS/MS. Glasgow Outcome Scale (GOS) score was assessed at 3 months.
There were no adverse events attributable to drug treatment. One patient in the placebo group was withdrawn due to adverse effects. In the treatment group, NAC concentrations ranged from 16,977.3±2,212.3 to 16,786.1±3,285.3 in serum and from 269.3±113.0 to 467.9±262.7 ng/mL in CSF, at 24 to 72 h post-bolus, respectively; and probenecid concentrations ranged from 75.4.3±10.0 to 52.9±25.8 in serum and 5.4±1.0 to 4.6±2.1 μg/mL in CSF, at 24 to 72 h post-bolus, respectively (mean±SEM). Temperature, mean arterial pressure, ICP, use of ICP-directed therapies, surveillance serum brain injury biomarkers, and GOS at 3 months were not different between groups.
Treatment resulted in detectable concentrations of NAC and probenecid in CSF and was not associated with undesirable effects after TBI in children.
ClinicalTrials.gov NCT01322009.
PM2.5 is an air pollution metric widely used to assess air quality, with the European Union having set targets for reduction in PM2.5 levels and population exposure. A major challenge for the ...scientific community is to identify, quantify and characterize the sources of atmospheric particles in the aspect of proposing effective control strategies. In the frame of ICARUS EU2020 project, a comprehensive database including PM2.5 concentration and chemical composition (ions, metals, organic/elemental carbon, Polycyclic Aromatic Hydrocarbons) from three sites (traffic, urban background, rural) of five European cities (Athens, Brno, Ljubljana, Madrid, Thessaloniki) was created. The common and synchronous sampling (two seasons involved) and analysis procedure offered the prospect of a harmonized Positive Matrix Factorization model approach, with the scope of identifying the similarities and differences of PM2.5 key-source chemical fingerprints across the sampling sites. The results indicated that the average contribution of traffic exhausts to PM2.5 concentration was 23.3% (traffic sites), 13.3% (urban background sites) and 8.8% (rural sites). The average contribution of traffic non-exhausts was 12.6% (traffic), 13.5% (urban background) and 6.1% (rural sites). The contribution of fuel oil combustion was 3.8% at traffic, 11.6% at urban background and 18.7% at rural sites. Biomass burning contribution was 22% at traffic sites, 30% at urban background sites and 28% at rural sites. Regarding soil dust, the average contribution was 5% and 8% at traffic and urban background sites respectively and 16% at rural sites. Sea salt contribution was low (1–4%) while secondary aerosols corresponded to the 16–34% of PM2.5. The homogeneity of the chemical profiles as well as their relationship with prevailing meteorological parameters were investigated. The results showed that fuel oil combustion, traffic non-exhausts and soil dust profiles are considered as dissimilar while biomass burning, sea salt and traffic exhaust can be characterized as relatively homogenous among the sites.
Display omitted
•Source apportionment was applied for 15 sites from 5 cities in Europe.•PMF harmonized application was based on common sampling and analysis protocol.•Traffic and biomass burning were the major control-demanding sources.•The homogeneity of PM2.5 source chemical profiles was examined.
Background. Despite inadequate empirical validation, sit-to-stand (STS) performance is often used as a proxy measure of lower limb strength among older adults. Furthermore, the relationships between ...bilateral isokinetic hip, knee, and ankle joint strength and their contributions to STS performances among older adults have not been established. The authors evaluated these relationships on 2 STS tests (5-chair STS test and 30-second chair STS test) in sexagenarian women. Methods. 47 women (mean age, 64.50 years) performed both STS tests on the same day and bilateral isokinetic (60°/second) hip extensor, hip flexor, knee extensor, knee flexor, ankle plantar flexor, and ankle dorsiflexor strength testing within 7 days after STS testing. Regression analyses were performed using the average weight-adjusted isokinetic hip, knee, and ankle joint strength scores as the independent variables and both STS test scores as the dependent variables. Results. Regression analyses including all 6 leg strength variables explained 48% (p =.0001) and 35% (p =.007) of the variance in 5-chair STS test scores and 30-second chair STS scores, respectively. Ankle plantar flexor, hip flexor, and knee extensor strength were the strongest predictors for both STS tests. Conclusions. Although ankle plantar flexor, hip flexor, and knee extensor strength play essential roles in performing the STS movement, most STS variance was unexplained, suggesting that important additional variables are also involved in completing the movement.
Multiple studies among adults have suggested a non-linear relationship between arterial partial pressure of oxygen (PaO
) and clinical outcomes. Meta-analyses in this population suggest that high ...levels of supplemental oxygen resulting in hyperoxia are associated with mortality. This mini-review focuses on the non-neonatal pediatric literature examining the relationship between PaO
and mortality. While only one pilot pediatric randomized-controlled trials exists, over the past decade, there have been at least eleven observational studies examining the relationship between PaO
values and mortality in critically ill children. These analyses of mixed-case pediatric ICU populations have generally reported a parabolic ("u-shaped") relationship between PaO
and mortality, similar to that seen in the adult literature. However, the estimates of the point at which hyperoxemia becomes deleterious have varied widely (300-550 mmHg). Where attempted, this effect has been robust to analyses restricted to the first PaO
value obtained, those obtained within 24 h of admission, anytime during admission, and the number of hyperoxemic blood gases over time. These findings have also been noted when using various methods of risk-adjustment (accounting for severity of illness scores or complex chronic conditions). Similar relationships were found in the majority of studies restricted to patients undergoing care after cardiac arrest. Taken together, the majority of the literature suggests that there is a robust parabolic relationship between PaO
and risk-adjusted pediatric ICU mortality, but that the exact threshold at which hyperoxemia becomes deleterious is unclear, and likely beyond the typical target value for most clinical indications. Findings suggest that clinicians should remain judicious and thoughtful in the use of supplemental oxygen therapy in critically ill children.
Recent decision rules for the management of febrile infants support the identification of infants at higher risk of serious bacterial infections (SBIs) without the performance of routine lumbar ...puncture. We derive and validate a model to identify febrile infants ≤60 days of age at low risk for SBIs using supervised machine learning approaches.
We conducted a secondary analysis of a multicenter prospective study performed between December 2008 and May 2013 of febrile infants. Our outcome was SBI, (culture-positive urinary tract infection, bacteremia, and/or bacterial meningitis). We developed and validated 4 supervised learning models: logistic regression, random forest, support vector machine, and a single-hidden layer neural network.
A total of 1470 patients were included (1014 >28 days old). One hundred thirty-eight (9.3%) had SBIs (122 urinary tract infections, 20 bacteremia, and 8 meningitis; 11 with concurrent SBIs). Using 4 features (urinalysis, white blood cell count, absolute neutrophil count, and procalcitonin), we demonstrated with the random forest model the highest specificity (74.9, 95% confidence interval: 71.5%-78.2%) with a sensitivity of 98.6% (95% confidence interval: 92.2%-100.0%) in the validation cohort. One patient with bacteremia was misclassified. Among 1240 patients who received a lumbar puncture, this model could have prevented 849 (68.5%) such procedures.
We derived and internally validated a supervised learning model for the risk-stratification of febrile infants. Although computationally complex, lacking parameter cutoffs, and in need of external validation, this strategy may allow for reductions in unnecessary procedures, hospitalizations, and antibiotics while maintaining excellent sensitivity.
PDF
