Leadership Auditorium, 2501 Student Center
Real-time prediction of unplanned 30-day hospital readmissions
Hospital readmissions are frequent and costly. It has been estimated that unplanned readmissions account for $17.4 billion in Medicare expenditures annually. Since the fiscal year 2013, the Hospital Readmissions Reduction Program (HRRP) has been established to financially penalize hospitals with excessive readmissions after initial admissions for particular conditions and procedures. In recent years, numerous hospital readmission predictive models have been reported and most of them rely on attributes that are only available near or post-discharge of the current encounter, such as the length of stay, discharge disposition, diagnosis codes. By incorporating these attributes, it is impossible to perform real-time readmission prediction during an inpatient encounter. However, early prediction of readmission can help deliver timely interventions to reduce the readmission risk. In this work, a machine learning predictive model has been built based on the Health Facts data. Because this model focuses more on patients’ medical history and requires less information from the current encounter, it allows real-time readmission prediction. I will discuss more details during the seminar.
Leadership Auditorium, 2501 Student Center
Medical Calculators: Prevalence, and Barriers to Use
Medical calculators synthesize measurable evidence and help introduce new medical guidelines and standards. Some medical calculators can fulfill the role of CDS for Meaningful Use purposes. However, there are barriers for clinicians to use medical calculators in practice. Objectives of this study were to determine whether lack of EHR integration would be a barrier to use of medical calculators and understand factors that may limit use and perceived usefulness of calculators
A survey about medical calculators as they relate to clinical efficiency, perceived usefulness, and barriers to effective use was conducted at a medium-sized academic medical center. 819 physicians were invited to participate in an online survey with a 13% response rate. Results were statistically analyzed to highlight factors related to use or non-use of medical calculators.
We found a negative correlation between use of medical calculators and years of experience (p<0.001), with decreasing calculator use as experience goes up. Barriers to using medical calculators by non-users and users of medical calculators show that necessity and integration are significantly different with p<0.001 and p=0.037, respectively. 46.7% of non-users reported necessity as a barrier compared to 7.7% of users. Integration was reported as a barrier for 43.6% of users, but only 13.3% of non-users. 61% of users indicated that calculators made them more efficient, and 70% reported that unavailability of normally used calculators make them less efficient. 60% of users indicated that they are somewhat or very likely to use newly published medical calculators.
The results highlight that medical calculators are important for care delivery by both users and non-users. For non-users, they are seen as having a potentially positive impact on patient care, but unnecessary as part of clinical practice. For medical calculator users, calculators are an important part of regular workflow for efficiency improvement. Clinicians with fewer years of experience show an eagerness to consume newly published calculators, making these kinds of CDS a potentially useful way to disseminate new medical evidence. The survey results suggest that when medical calculators can be automated and integrated into the EHR as part of everyday workflow then efficiency and adoption may increase.
222 Naka Hall
Application of Deep Learning in Predicting Phenotypes
Genomic selection (GS) can use single-nucleotide polymorphism (SNPs) markers to predict breeding values (BV) for enhancing quantitative traits in breeding populations. GS has been proved to increase breeding efficiency in both plant and animal breeding. However, existing statistical and machine-learning methods require imputation to missing values in genotypes, which leads to poor generalization and computation inefficiency. Here, we propose a deep-learning model using convolutional neural networks (CNN) to predict the Genomic Estimated Breeding Value (GEBV) and also to investigate contributions of genomic SNPs to GEBV using a saliency map approach.Comparing with traditional statistical models including rr-BLUP, Bayesian ridge regression, Bayesian LASSO and Bayes A on a Glycine Max (soybean) Nested Association Mapping (NAM) dataset and a simulation dataset, our model can better handle the missing values in genotypes and is more efficient in accurately predicting breeding values. Our model also has a great potential in interpreting phenotype-genotype associations over the entire genome. The model is available at https://github.com/kateyliu/DL_gwas.
Leadership Auditorium, 2501 Student Center
Volumetric Analysis of Adipose Tissue
Body Condition Score is the veterinary equivalent of BMI in humans, in which veterinarians attempt to assess adiposity of an animal and make appropriate recommendations. However, this measure of adiposity is fairly subjective and quite variable depending on the species being analyzed. Thus, a more quantifiable and objective measure of adiposity, through the utilization of initially CT scans and subsequently through radiographs would be beneficial. CT scans were taken from the University of Missouri Veterinary Health Center PACS from patients who had received thoracic CT scans, a full body CT scan, as well as a thoracic and abdominal CT scan. OTSU thresholding and an inversion algorithm were utilized to select only the patient, while a threshold algorithm was then applied to isolate out the adipose tissue from the remainder of the tissue. Percent adiposity from each provided region were then compared and presented in contrast to the allocated body condition score from the consensus of clinicians. Results showed that the allocated body condition score had a wide range of adiposities associated with it, including adiposity ranges that overlapped across multiple body condition scores, which provides evidence that the subjective measure of body condition score may be worse at assessing adiposity when compared to volumetric analysis of adiposity.
Leadership Auditorium, 2501 MU Student Center
Investigating Genome Compositional Features of Apis and other Hymenopteran Species
Initial analysis of the honey bee (Apis mellifera) genome in 2006 revealed several interesting features compared to other metazoan genome sequences available at that time: a low but heterogeneous GC content, an overabundance of CpG dinucleotides and a lack of repetitive elements. The average GC content of the honey bee genome is only 33%, but GC content is highly heterogeneous, ranging from 11% to 67%, with a bimodal distribution. Furthermore, unlike genes in most other metazoans, honey bee genes are overly abundant in regions of low GC content (<30%). It is unclear whether any of these genome features are related to the evolution of eusociality and we lack satisfactory explanations for them more generally. Since publication of the A. mellifera genome, genomes of several other hymenopteran insects, including additional Apis species, have become available. In this study, we compare Apis genome compositional characteristics with those of other hymenopteran insects. Comparing genome composition and organization among species with different levels of social complexity may lead to insight into genomic structural changes associated with the evolution of eusociality. We used a recursive segmentation procedure to partition genomic sequences into GC compositional domains, maximizing the difference in GC content between adjacent subsequences. We compared the distributions of GC contents in GC compositional domains among 21 hymenopteran genomes ranging in social complexity from solitary to complex eusocial. We also analyzed one eusocial and four solitary outgroups representing diverse insect taxa. Bimodal distribution of GC content within the GC compositional domains was a characteristic of the complex eusocial bees (Apis and Melipona), but not solitary or simple eusocial bees. The Apis genomes had larger ranges in GC content compared to the other species. Genes were biased to lower GC content regions in all bees, with the strongest bias in the complex eusocial bees, and the weakest bias in the solitary bees, while gene distribution tended to show little or no bias to low GC content regions in the ant genomes and non-hymenopteran insect outgroups. Further investigation of these preliminary data will provide insight into whether genomic compositional features unique to Apis are associated with the evolution of eusociality.
1 2 3 … 21 Next »