Foodborne illness afflicts 48 million people annually in the U.S. alone. Over 128,000 are hospitalized and 3,000 die from the infection. While preventable with proper food safety practices, the traditional restaurant inspection process has limited impact given the predictability and low frequency of inspections, and the dynamic nature of the kitchen environment. Despite this reality, the inspection process has remained largely unchanged for decades. CDC has even identified food safety as one of seven ”winnable battles”; however, progress to date has been limited. In this work, we demonstrate significant improvements in food safety by marrying AI and the standard inspection process. We apply machine learning to Twitter data, develop a system that automatically detects venues likely to pose a public health hazard, and demonstrate its efficacy in the Las Vegas metropolitan area in a double-blind experiment conducted over three months in collaboration with Nevada’s health department. By contrast, previous research in this domain has been limited to indirect correlative validation using only aggregate statistics. We show that adaptive inspection process is 64 percent more effective at identifying problematic venues than the current state of the art. If fully deployed, our approach could prevent over 9,000 cases of foodborne illness and 557 hospitalizations annually in Las Vegas alone. Additionally, adaptive inspections result in unexpected benefits, including the identification of venues lacking permits, contagious kitchen staff, and fewer customer complaints filed with the Las Vegas health department.

Purpose Osteoporosis and osteopenia are extremely common and can lead to fragility fractures. The purpose of this study was to determine whether a computer learning system could classify whether a hand radiograph demonstrated osteoporosis based on the second metacarpal cortical percentage. Methods We used the second metacarpal cortical percentage as the osteoporosis predictor. A total of 4,000 posteroanterior (PA) radiographs of the hand were standardized through laterality correction, vertical alignment correction, segmentation, proxy osteoporosis predictor, and full pipeline. Laterality was classified using a LeNet convolutional neural network (CNN). Vertical alignment classification used 2,000 PA x-rays to determine vertical alignment of the second metacarpal. We employed segmentation to determine which pixels belong to the second metacarpal from 1,000 PA x-rays using the FSN-8 CNN. The full pipeline was tested on 265 previously unseen PA x-rays. Results Laterality classification accuracy was 99.62%, with a specificity of 100% and sensitivity of 99.3%. Rotation of the hand within 10 of vertical was accurate in 93.2% of films. Segmentation was 94.8% accurate. Proxy osteoporosis predictor was 88.4% accurate. Full pipeline accuracy was 93.9%. In the testing data set, the CNN had a sensitivity of 82.4% and specificity of 95.7%. In the balanced data set, 6 of 39 osteoporotic films were classified as nonosteoporotic; sensitivity was 82.4% and specificity, 94.3%. Conclusions We have created a series of CNN that can accurately identify osteoporosis from non-osteoporosis. Furthermore, our CNN is able to make adjustments to images based on laterality and vertical alignment. Clinical relevance Convolutional neural network and computer learning can be used as an adjunct to dual-energy x-ray absorptiometry scans or to screen and make appropriate referrals for further workup in patients with suspected osteoporosis.

Background Determining discharge disposition after total joint arthroplasty (TJA) has been a challenge. Advances in machine learning (ML) have produced computer models that learn by example to generate predictions on future events. We hypothesized a trained ML algorithm’s diagnostic accuracy will be better than that of current predictive tools to predict discharge disposition after primary TJA. Methods This study was a retrospective cohort study from a single, tertiary referral center for primary TJA. We trained and validated an artificial neural network (ANN) based on 4368 distinct surgical encounters between 1/1/2013 and 6/28/2016. The ANN’s ability to identify discharge disposition was then tested on 1452 distinct surgical encounters between 1/3/17 and 11/30/17. Results The area under the curve and accuracy achieved during model validation were 0.973 and 91.7%, respectively, with 25% of patients being discharged to skilled nursing facilities (SNFs). Within our testing data set, 6.7% of patients went to SNFs. The performance in the testing set included an area under the curve of 0.804, accuracy of 61.3%, sensitivity of 28.9%, and specificity of 93.8%. Conclusions This is the first prediction tool using an electronic medical record–integrated ANN to predict discharge disposition after TJA based on locally generated data. Dramatically reduced numbers of patients discharged to SNFs due to implementation of a bundled payment model lead to poor recall in the testing model. This model serves as a proof of concept for developing an ML prediction tool using a relatively small data set and subsequent integration into the electronic medical record.

Category: Other Introduction/Purpose: Each year approximately 30-40% of people over the age of 65 fall. Approximately one half of these falls result in an injury with the estimated annual direct medical costs of $30 billion. Pain, mobility issues, neuropathy and post-operative weight bearing limitations make foot and ankle patients particularly vulnerable to falls. Current approaches to determine at risk patients are cumbersome and time consuming requiring performance testing and “hands on” clinical assessment. The efficiency of obtaining PRO, such as PROMIS, in the clinical arena has been well documented. The purpose of this study is determine if patient reported outcomes (PROMIS) can identify orthopaedic and specifically foot and ankle patients at risk to fall. Methods: Prospective patient reported outcomes (PROMIS CAT physical function, pain interference and depression and CMS fall risk assessment questions) and patient demographics were collected for all patients at each clinic visit from an academic orthopaedic multi-specialty practice between January 2015 and November 2017. Standardized yes/no validated self-reported fall risk questions include: “Have you fallen in the last year?” and “Do you feel you are at risk of falling?” Histograms, t-tests, confidence intervals and effect size were used to determine the fall risk “YES” patients were different than the “NO” for ALL orthopaedic patients and specifically foot and ankle patients. Logistic Regression was used to determine if age, gender, height, weight, and PROMIS scales predicted self-reported falls risk. Results: 94,761 orthopaedic patients comprising 315,273 visits (44% male, mean age 53.7+/-17 years) and 13,720 foot/ankle patients comprising 33,480 visits (37% male, mean age 52.7+/-16.1 years) had complete data for analysis. Table 1 provides the means/SD/p-values/effect sizes for patient self-identifying at risk to fall stratified by PROMIS PF/ PI/Dep t-scores. Although all PROMIS scores demonstrated significant impairment between patients at risk designation (yes/no), PROMIS PF had the largest effect size for ALL Ortho and FOOT AND ANKLE patients (0.8 and 0.7 respectively). Patients who are at risk to fall have PROMIS PF t-scores >1.5 lower than the United States normative population while the patients not at risk are less <1 SD. In the adjusted regression models gender and PROMIS PF had the largest coefficients. Conclusion: Falls are a major threat to quality of life and independence yet prevention/treatment strategies are difficult to implement across a health system. There is also a tremendous societal cost with orthopaedic surgeons often the recipient of these debilitated patients. PROMIS assessments are part of the AOFAS OFAR initiative to track patient recovery with treatment and can additional be used to fulfill a quality indicator requirement by CMS. This study demonstrates these assessments (PROMIS threshold values) can also be linked to self-report falls risk (yes/no) and may identify patients at risk with no face to face time requi...

Foodborne illness afflicts 48 million people annually in the U.S. alone. Over 128,000 are hospitalized and 3,000 die from the infection. While preventable with proper food safety practices, the traditional restaurant inspection process has limited impact given the predictability and low frequency of inspections, and the dynamic nature of the kitchen environment. Despite this reality, the inspection process has remained largely unchanged for decades. We apply machine learning to Twitter data and develop a system that automatically detects venues likely to pose a public health hazard. Health professionals subsequently inspect individual flagged venues in a double blind experiment spanning the entire Las Vegas metropolitan area over three months. By contrast, previous research in this domain has been limited to indirect correlative validation using only aggregate statistics. We show that adaptive inspection process is 63% more effective at identifying problematic venues than the current state of the art. The live deployment shows that if every inspection in Las Vegas became adaptive, we can prevent over 9,000 cases of foodborne illness and 557 hospitalizations annually. Additionally, adaptive inspections result in unexpected benefits, including the identification of venues lacking permits, contagious kitchen staff, and fewer customer complaints filed with the Las Vegas health department.