Validation and algorithmic audit of a deep learning system for the detection of proximal femoral fractures in patients in the emergency department: a diagnostic accuracy study.

BACKGROUND: Proximal femoral fractures are an important clinical and public health issue associated with substantial morbidity and early mortality. Artificial intelligence might offer improved diagnostic accuracy for these fractures, but typical approaches to testing of artificial intelligence models can underestimate the risks of artificial intelligence-based diagnostic systems. METHODS: We present a preclinical evaluation of a deep learning model intended to detect proximal femoral fractures in frontal x-ray films in emergency department patients, trained on films from the Royal Adelaide Hospital (Adelaide, SA, Australia). This evaluation included a reader study comparing the performance of the model against five radiologists (three musculoskeletal specialists and two general radiologists) on a dataset of 200 fracture cases and 200 non-fractures (also from the Royal Adelaide Hospital), an external validation study using a dataset obtained from Stanford University Medical Center, CA, USA, and an algorithmic audit to detect any unusual or unexpected model behaviour. FINDINGS: In the reader study, the area under the receiver operating characteristic curve (AUC) for the performance of the deep learning model was 0·994 (95% CI 0·988-0·999) compared with an AUC of 0·969 (0·960-0·978) for the five radiologists. This strong model performance was maintained on external validation, with an AUC of 0·980 (0·931-1·000). However, the preclinical evaluation identified barriers to safe deployment, including a substantial shift in the model operating point on external validation and an increased error rate on cases with abnormal bones (eg, Paget's disease). INTERPRETATION: The model outperformed the radiologists tested and maintained performance on external validation, but showed several unexpected limitations during further testing. Thorough preclinical evaluation of artificial intelligence models, including algorithmic auditing, can reveal unexpected and potentially harmful behaviour even in high-performance artificial intelligence systems, which can inform future clinical testing and deployment decisions. FUNDING: None.

The effects of the COVID pandemic on the federal budget outlook.

We examine the impact of COVID-19 on the federal budget outlook. We find substantial but temporary effects on spending and revenues, with more moderate but permanent effects on the long-term projections. We project that the debt-to-GDP ratio, currently 98%, will rise to 190% in 2050 under current law, compared to a CBO pre-COVID projection of 180%. Sharply lower interest rates projected for the next dozen years help moderate future debt accumulation. Under a "current policy" projection that allows temporary tax provisions-such as those in the Tax Cut and Jobs Act of 2017-to be made permanent, the debt-to-GDP ratio would rise to 222% by 2050 and would continuing rising thereafter. The long-term projections are sensitive to interest rates. We discuss several aspects of these results, including how the current episode compares to past debt changes, the role of historically low interest rates, and the role of recent Federal Reserve Board policies and actions. Because of the macro-stabilization effects of fiscal tightening, and because low interest rates create "breathing room" for fiscal policy, we do not see the large, short-run debt accumulation resulting from the current pandemic as necessitating any immediate offsetting response. But the long-term projections show that significant fiscal imbalances remain and will eventually require attention.

Deep learning predicts hip fracture using confounding patient and healthcare variables.

Hip fractures are a leading cause of death and disability among older adults. Hip fractures are also the most commonly missed diagnosis on pelvic radiographs, and delayed diagnosis leads to higher cost and worse outcomes. Computer-aided diagnosis (CAD) algorithms have shown promise for helping radiologists detect fractures, but the image features underpinning their predictions are notoriously difficult to understand. In this study, we trained deep-learning models on 17,587 radiographs to classify fracture, 5 patient traits, and 14 hospital process variables. All 20 variables could be individually predicted from a radiograph, with the best performances on scanner model (AUC = 1.00), scanner brand (AUC = 0.98), and whether the order was marked "priority" (AUC = 0.79). Fracture was predicted moderately well from the image (AUC = 0.78) and better when combining image features with patient data (AUC = 0.86, DeLong paired AUC comparison, p = 2e-9) or patient data plus hospital process features (AUC = 0.91, p = 1e-21). Fracture prediction on a test set that balanced fracture risk across patient variables was significantly lower than a random test set (AUC = 0.67, DeLong unpaired AUC comparison, p = 0.003); and on a test set with fracture risk balanced across patient and hospital process variables, the model performed randomly (AUC = 0.52, 95% CI 0.46-0.58), indicating that these variables were the main source of the model's fracture predictions. A single model that directly combines image features, patient, and hospital process data outperforms a Naive Bayes ensemble of an image-only model prediction, patient, and hospital process data. If CAD algorithms are inexplicably leveraging patient and process variables in their predictions, it is unclear how radiologists should interpret their predictions in the context of other known patient data. Further research is needed to illuminate deep-learning decision processes so that computers and clinicians can effectively cooperate.

Programmed cell death in wheat (Triticum aestivum L.) endosperm cells is affected by drought stress.

Drought frequently occurs during wheat (Triticum aestivum L.) grain filling. The objectives of this study were (i) to investigate the effect of post-anthesis drought on programmed cell death (PCD) in wheat endosperm cells and (ii) to examine the role of ethylene (ETH) receptors and abscisic acid (ABA) in regulating wheat endosperm PCD. Two winter wheat cultivars ('Xindong 18' and 'Xindong 22') were used in this study. Grain samples were collected from normal and drought stressed plants at 5-day intervals between 5 and 35 days post-anthesis. The samples were then compared with respect to cell viability, nuclear morphometry, cell ultrastructure, DNA integrity, nucleic acid content, and nuclease activity. Analysis was also conducted about gene transcripts related to PCD, ETH receptors, and ABA biosynthesis and degradation. Drought stress reduced cell viability, accelerated nuclear deformation, and increased mitochondrial dissolution. The activity of nucleic acid hydrolase was greater, and the nucleic acid concentrations were less in the drought treatments than in the control. As a result, the peak in DNA fragmentation occurred earlier in the drought treatment. Drought stress significantly increased the expression of four genes related to ABA (nced1, nced2, ao1, ao2). In contrast, drought significantly reduced the expression of four genes related to ETH receptors (ers1, ers2 etr1, etr2) and one gene related to PCD (dad1). In summary, the results indicated that drought stress caused PCD to occur earlier in the endosperm of winter wheat.

Effects of reclamation years on composition and diversity of soil bacterial communities in Northwest China.

The objective of this study was to evaluate bacterial community structure and diversity in soil aggregate fractions when salinized farmland was reclaimed after >27 years of abandonment and then farmed again for 1, 5, 10, and 15 years. Illumina MiSeq high-throughput sequencing was performed to characterize the soil bacterial communities in 5 aggregate size classes in each treatment. The results indicated that reclamation significantly increased macro-aggregation (>0.25 mm), as well as soil organic C, available N, and available P. The 10-year field had the largest proportion (93.9%) of soil in the macro-aggregate size classes (i.e., >0.25 mm) and the highest soil electrical conductivity. The 5 most dominant phyla in the soil samples were Proteobacteria, Actinobacteria, Gemmatimonadetes, Acidobacteria, and Bacteroidetes. The phylogenetic diversity, Chao1, and Shannon indices increased after the abandoned land was reclaimed for farming, reaching maximums in the 15-year field. Among aggregate size classes, the 1-0.25 mm aggregates generally had the highest phylogenetic diversity, Chao1, and Shannon indices. Soil organic C and soil electrical conductivity were the main environmental factors affecting the soil bacterial communities. The composition and structure of the bacterial communities also varied significantly depending on soil aggregate size and time since reclamation.

Duration of continuous cropping with straw return affects the composition and structure of soil bacterial communities in cotton fields.

Salinized land in the China's Xinjiang Region is being reclaimed for continuous cotton production. The specific objectives of this field study were (i) to compare bacterial composition and diversity in unfarmed (i.e., unreclaimed) and continuously (5, 10, 15, and 20 years) cropped soils and (ii) to explore correlations between soil properties and the bacterial communities identified by Illumina MiSeq sequencing. The results showed that bacterial species richness and diversity increased for 10-15 years and then declined when salinized land was reclaimed for cotton production. Proteobacteria and Firmicutes were the dominant phyla in unfarmed soil. Continuous cropping reduced the abundance of Firmicutes but increased that of Chloroflexi, Acidobacteria, and Actinobacteria. Cluster analyses showed that the greatest similarities in bacterial communities were between the 5- and 10-year treatments and between the 15- and 20-year treatments. Soil pH, electrical conductivity, alkali-hydrolyzable N, and available P were significantly correlated with bacterial community distribution. Overall, cotton production improved soil physicochemical properties and altered the structure and composition of soil bacterial communities compared with unfarmed soil. These positive effects began to decrease after 10-15 years of continuous cotton production.

HOW THE GROWING GAP IN LIFE EXPECTANCY MAY AFFECT RETIREMENT BENEFITS AND REFORMS.

Older Americans have experienced dramatic gains in life expectancy in recent decades, but an emerging literature reveals that these gains are accumulating mostly to those at the top of the income distribution. We explore how growing inequality in life expectancy affects lifetime benefits from Social Security, Medicare, and other programs and how this phenomenon interacts with possible program reforms. We first project that life expectancy at age 50 for males in the two highest income quintiles will rise by 7 to 8 years between the 1930 and 1960 birth cohorts, but that the two lowest income quintiles will experience little to no increase over that time period. This divergence in life expectancy will cause the gap between average lifetime program benefits received by men in the highest and lowest quintiles to widen by $130,000 (in $2009) over this period. Finally we simulate the effect of Social Security reforms such as raising the normal retirement age and changing the benefit formula to see whether they mitigate or enhance the reduced progressivity resulting from the widening gap in life expectancy.

Effects of an Individual Development Account Program on Retirement Saving: Follow-up Evidence From a Randomized Experiment.

We examine the 10-year follow-up effects on retirement saving of an individual development account (IDA) program using data from a randomized experiment that ran from 1998 to 2003 in Tulsa, Oklahoma. The IDA program included financial education, encouragement to save, and matching funds for several qualified uses of the saving, including contributions to retirement accounts. The results indicate that as of 2009, 6 years after the program ended, the IDA program had no impact on the propensity to hold a retirement account, the account balance, or the sufficiency of retirement balances to meet retirement expenses.

Global pattern of soil carbon losses due to the conversion of forests to agricultural land.

Several reviews have analyzed the factors that affect the change in soil organic C (SOC) when forest is converted to agricultural land; however, the effects of forest type and cultivation stage on these changes have generally been overlooked. We collated observations from 453 paired or chronosequential sites where forests have been converted to agricultural land and then assessed the effects of forest type, cultivation stage, climate factors, and soil properties on the change in the SOC stock and the SOC turnover rate constant (k). The percent decrease in SOC stocks and the turnover rate constants both varied significantly according to forest type and cultivation stage. The largest decrease in SOC stocks was observed in temperate regions (52% decrease), followed by tropical regions (41% decrease) and boreal regions (31% decrease). Climate and soil factors affected the decrease in SOC stocks. The SOC turnover rate constant after the conversion of forests to agricultural land increased with the mean annual precipitation and temperature. To our knowledge, this is the first time that original forest type was considered when evaluating changes in SOC after being converted to agricultural land. The differences between forest types should be considered when calculating global changes in SOC stocks.

Raising household saving: does financial education work?

This article highlights the prevalence and economic outcomes of financial illiteracy among American households, and reviews previous research that examines how improving financial literacy affects household saving. Analysis of the research literature suggests that previous financial literacy efforts have yielded mixed results. Evidence suggests that interventions provided for employees in the workplace have helped increase household saving, but estimates of the magnitude of the impact vary widely. For financial education initiatives targeted to other groups, the evidence is much more ambiguous, suggesting a need for more econometrically rigorous evaluations.

The accumulation of organic carbon in mineral soils by afforestation of abandoned farmland.

The afforestation of abandoned farmland significantly influences soil organic carbon (OC). However, the dynamics between OC inputs after afforestation and the original OC are not well understood. To learn more about soil OC dynamics after afforestation of farmland, we measured the soil OC content in paired forest and farmland plots in Shaanxi Province, China. The forest plots had been established on farmland 18, 24, 48, 100, and 200 yr previously. The natural (13)C abundance of soil organic matter was also analyzed to distinguish between crop- and forest-derived C in the afforested soils. We observed a nonlinear accumulation of total OC in the 0-80 cm depth of the mineral soil across time. Total soil OC accumulated more rapidly under forest stands aged 18 to 48 yr than under forest stands aged 100 or 200 yrs. The rate of OC accumulation was also greater in the 0-10 cm depth than in the 10-80 cm depth. Forest-derived OC in afforested soils also accumulated nonlinearly across time, with the greatest increase in the 0-20 cm depth. Forest-derived OC in afforest soils accounted for 52-86% of the total OC in the 0-10 cm depth, 36-61% of the total OC in the 10-20 cm depth, and 11-50% of the total OC in the 20-80 cm depth. Crop-derived OC concentrations in the 0-20 cm depth decreased slightly after afforestation, but there was no change in crop-derived OC concentrations in the 20-80 cm depth. The results of our study support the claim that afforestation of farmland can sequester atmospheric CO(2) by increasing soil OC stocks. Changes in the OC stocks of mineral soils after afforestation appear to be influenced mainly by the input of forest-derived C rather than by the loss of original OC.

Attachment of Listeria monocytogenes to an austenitic stainless steel after welding and accelerated corrosion treatments.

Austenitic stainless steels, widely used in food processing, undergo microstructural changes during welding, resulting in three distinctive zones: weld metal, heat-affected zone, and base metal. This research was conducted to determine the attachment of Listeria monocytogenes in these three zones before and after exposure to a corrosive environment. All experiments were done with tungsten inert gas welding of type 304 stainless steel. The four welding treatments were large or small beads with high or low heat. After welding, all surfaces were polished to an equivalent surface finish. A 10-microl droplet of an L. monocytogenes suspension was placed on the test surfaces. After 3 h at 23 degrees C, the surfaces were washed and prepared for scanning electron microscopy, which was used to determine attachment of L. monocytogenes by counting cells remaining on each test surface. In general, bacteria were randomly distributed on each surface type. However, differences in surface area of inoculum due to differences in interfacial energy (as manifested by the contact angle) were apparent and required normalization of bacterial count data. There were no differences (P > 0.05) in numbers of bacteria on the three surface zones. However, after exposure to the corrosive medium, numbers of bacteria on the three zones were higher (P < 0.05) than those on the corresponding zones of noncorroded surfaces. For the corroded surfaces, bacterial counts on the base metal were lower (P < 0.05) than those on heat-affected and weld zones.

Novel metal clusters isolated from blood are lethal to cancer cells.

Unfolding and subsequent aggregation of proteins is a common phenomenon that is linked to many human disorders. Misfolded hemoglobin is generally manifested in various autoimmune, infectious and inherited diseases. We isolated micrometer and submicrometer particles, termed proteons, from human and animal blood. Proteons lack nucleic acids but contain two major polypeptide populations with homology to the hemoglobin alpha-chain. Proteons form by reversible seeded aggregation of proteins around proteon nucleating centers (PNCs). PNCs are comprised of 1- to 2-nm metallic nanoclusters containing 40-300 atoms. Each milliliter of human blood contained approximately 7 x 10(13) PNCs and approximately 3 x 10(8) proteons. Exposure of isolated blood plasma to elevated temperatures increased the number of proteons. When an aliquot of this heated plasma was introduced into untreated plasma that was subsequently heated, the number of proteons further increased, reaching a maximum after a total of three such iterations. Small concentrations of PNCs were lethal to cultured cancer cells, whereas noncancerous cells were much less affected.

Metallic nanoparticle production utilizing a supercritical carbon dioxide flow process.

Metallic nanoparticles of palladium and silver ranging in size from 1 to 15 nm were produced entirely within carbon dioxide by spraying a carbon dioxide carrier solution containing CO2-soluble metal precursors into a CO2 receiving solution containing a reducing agent (NaBH(OAc)3 or H2) and fluorocarbon thiol stabilizing ligands. The process uses the benign solvent CO2 while also allowing for the production of nanoparticles with a limited number of chemical components. Particles were characterized by transmission electron microscopy (TEM) and energy dispersive spectroscopy (EDS).

Interaction of xenobiotics with estrogen receptors alpha and beta and a putative plasma sex hormone-binding globulin from channel catfish (Ictalurus punctatus).

Estrogens are important regulators of physiological functions. Although environmental contaminants (xenoestrogens) which interfere with estrogen signaling are of increasing concern, there is only limited information about their ability to interact with estrogen-binding proteins (SHBG) or receptors (ER). Recombinant ERalpha and beta were obtained after transient transfection of COS-7 cells with channel catfish ER cDNA. Plasma from adult female channel catfish was the source of SHBG. Tritiated estradiol (3H-E2) was used in standard radioligand-binding assays to characterize the binding properties of channel catfish SHBG (ccfSHBG) and to estimate the inhibition constants for various estrogenic compounds. Binding of 3H-E2 to ccfSHBG was saturable and of high affinity with a Kd (+/-SE) of 1.9+/-0.14 nM and a Bmax of 14.3+/-2.4 pmol/mg protein ( n = 3 assays). Additionally, ccfSHBG displayed binding specificity for androgens and estrogens. Endosulfan, 4-nonylphenol, and 4-octylphenol displaced 3H-E2 binding to ccfSHBG albeit only at very high concentrations, whereas dieldrin and atrazine showed little displacement activity even at the highest concentrations used. The synthetic estrogen ethynylestradiol had higher affinity than E2 for ccfSHBG. This finding differs from results with human and rainbow trout SHBG. The alkylphenolic compounds (4-octylphenol and 4-nonylphenol) displayed some ability to displace 3H-E2 binding from ERalpha and beta at high concentrations, but dieldrin and atrazine had little binding activity for both ER subtypes and endosulfan for ERbeta. The xenobiotics tested generally showed equivalent or greater affinity for ERalpha than ERbeta, whereas natural estrogens had much greater affinity for ERbeta than ERalpha. These observations suggest that results of studies using fish tissue ER extracts must be interpreted with caution, since both ER subtypes may be present, and that the binding of xenoestrogens to SHBG must be taken into account for proper assessment of endocrine disruption caused by environmental contaminants.