Machine learning modeling practices to support the principles of AI and ethics in nutrition research.

BACKGROUND: Nutrition research is relying more on artificial intelligence and machine learning models to understand, diagnose, predict, and explain data. While artificial intelligence and machine learning models provide powerful modeling tools, failure to use careful and well-thought-out modeling processes can lead to misleading conclusions and concerns surrounding ethics and bias. METHODS: Based on our experience as reviewers and journal editors in nutrition and obesity, we identified the most frequently omitted best practices from statistical modeling and how these same practices extend to machine learning models. We next addressed areas required for implementation of machine learning that are not included in commercial software packages. RESULTS: Here, we provide a tutorial on best artificial intelligence and machine learning modeling practices that can reduce potential ethical problems with a checklist and guiding principles to aid nutrition researchers in developing, evaluating, and implementing artificial intelligence and machine learning models in nutrition research. CONCLUSION: The quality of AI/ML modeling in nutrition research requires iterative and tailored processes to mitigate against potential ethical problems or to predict conclusions that are free of bias.

Deep Vision for Breast Cancer Classification and Segmentation.

(1) Background: Female breast cancer diagnoses odds have increased from 11:1 in 1975 to 8:1 today. Mammography false positive rates (FPR) are associated with overdiagnoses and overtreatment, while false negative rates (FNR) increase morbidity and mortality. (2) Methods: Deep vision supervised learning classifies 299 × 299 pixel de-noised mammography images as negative or non-negative using models built on 55,890 pre-processed training images and applied to 15,364 unseen test images. A small image representation from the fitted training model is returned to evaluate the portion of the loss function gradient with respect to the image that maximizes the classification probability. This gradient is then re-mapped back to the original images, highlighting the areas of the original image that are most influential for classification (perhaps masses or boundary areas). (3) Results: initial classification results were 97% accurate, 99% specific, and 83% sensitive. Gradient techniques for unsupervised region of interest mapping identified areas most associated with the classification results clearly on positive mammograms and might be used to support clinician analysis. (4) Conclusions: deep vision techniques hold promise for addressing the overdiagnoses and treatment, underdiagnoses, and automated region of interest identification on mammography.

Healthcare market segmentation and data mining: A systematic review.

Providing insight into healthcare consumers' behaviors and attitudes is critical information in an environment where healthcare delivery is moving rapidly towards patient-centered care that is premised upon individuals becoming more active participants in managing their health. A systematic review of the literature concerning healthcare market segmentation and data mining identified several areas for future health marketing research. Common themes included: (a) reliance on survey data, (b) clustering methods, (c) limited classification modeling after clustering, and (d) detailed analysis of clusters by demographic data. Opportunities exist to expand health-marketing research to leverage patient level data with advanced data mining methods.

Reducing cost drivers in total joint arthroplasty: understanding patient readmission risk and supply cost.

Introduction: Understanding and planning for the factors that impact supply cost and unplanned readmission risk for total joint arthroplasty (TJA) patients is helpful for hospitals at financial risk under bundled payments. Readmission and operating room supply costs are two of the biggest expenses. Methods: Logistic and linear regressions are used to measure the impacts of TJA patient attributes on readmission risk and supply costs, respectively. Results: Patients' health market segment and the number/type of comorbidity impacts 30/90-day readmission rates. Surgeon implant preference and type of surgery impact supply costs. Discharge location and two of the five health market segments increase the odds of 30-day readmission. Arrhythmia and lymphoma are the primary comorbidities that impact the odds of readmission at 90 days. Conclusions: Preoperatively identifying TJA patients likely to have large supply costs and higher readmission risk allows hospitals to invest in low-cost interventions to reduce risk and improve healthcare value.

Evaluating the Relationship between Productivity and Quality in Emergency Departments.

BACKGROUND: In the United States, emergency departments (EDs) are constantly pressured to improve operational efficiency and quality in order to gain financial benefits and maintain a positive reputation. OBJECTIVES: The first objective is to evaluate how efficiently EDs transform their input resources into quality outputs. The second objective is to investigate the relationship between the efficiency and quality performance of EDs and the factors affecting this relationship. METHODS: Using two data sources, we develop a data envelopment analysis (DEA) model to evaluate the relative efficiency of EDs. Based on the DEA result, we performed multinomial logistic regression to investigate the relationship between ED efficiency and quality performance. RESULTS: The DEA results indicated that the main source of inefficiencies was working hours of technicians. The multinomial logistic regression result indicated that the number of electrocardiograms and X-ray procedures conducted in the ED and the length of stay were significantly associated with the trade-offs between relative efficiency and quality. Structural ED characteristics did not influence the relationship between efficiency and quality. CONCLUSIONS: Depending on the structural and operational characteristics of EDs, different factors can affect the relationship between efficiency and quality.

Stochastic multi-objective auto-optimization for resource allocation decision-making in fixed-input health systems.

The management of hospitals within fixed-input health systems such as the U.S. Military Health System (MHS) can be challenging due to the large number of hospitals, as well as the uncertainty in input resources and achievable outputs. This paper introduces a stochastic multi-objective auto-optimization model (SMAOM) for resource allocation decision-making in fixed-input health systems. The model can automatically identify where to re-allocate system input resources at the hospital level in order to optimize overall system performance, while considering uncertainty in the model parameters. The model is applied to 128 hospitals in the three services (Air Force, Army, and Navy) in the MHS using hospital-level data from 2009 - 2013. The results are compared to the traditional input-oriented variable returns-to-scale Data Envelopment Analysis (DEA) model. The application of SMAOM to the MHS increases the expected system-wide technical efficiency by 18 % over the DEA model while also accounting for uncertainty of health system inputs and outputs. The developed method is useful for decision-makers in the Defense Health Agency (DHA), who have a strategic level objective of integrating clinical and business processes through better sharing of resources across the MHS and through system-wide standardization across the services. It is also less sensitive to data outliers or sampling errors than traditional DEA methods.

Evaluating the Impact of Hospital Efficiency on Wellness in the Military Health System.

Like all health care delivery systems, the U.S. Department of Defense Military Health System (MHS) strives to achieve top preventative care and population health outcomes for its members while operating at an efficient level and containing costs. The objective of this study is to understand the overall efficiency performance of military hospitals and investigate the relationship between efficiency and wellness. This study uses data envelopment analysis and stochastic frontier analysis to compare the efficiency of 128 military treatment facilities from the Army, Navy, and Air Force during the period of 2011 to 2013. Fixed effects panel regression is used to determine the association between the hospital efficiency and wellness scores. The results indicate that data envelopment analysis and stochastic frontier analysis efficiency scores are congruent in direction. Both results indicate that the majority of the MHS hospitals and clinics can potentially improve their productive efficiency by managing their input resources better. When comparing the performance of the three military branches of service, Army hospitals as a group outperformed their Navy and Air Force counterparts; thus, best practices from the Army should be shared across service components. The findings also suggest no statistically significant, positive association between efficiency and wellness over time in the MHS.

The Impact of a Pay-for-Performance Program on Central Line-Associated Blood Stream Infections in Pennsylvania.

Healthcare associated infections have significantly contributed to the rising cost of hospital care in the United States. The implementation of pay-for-performance (P4P) programs has been one approach to improve quality at a reduced cost. We quantify the impact of Highmark's Quality Blue (QB) hospital P4P program on central line-associated blood stream infections (CLABSI) in Pennsylvania. The impact of years of participation in QB on CLABSI is also evaluated. Data from 149 Pennsylvania hospitals on CLABSI from 2008-2013 are used. Negative binomial regression and fixed effects panel regression are performed. Hospitals participating in QB have 0.727 times the CLABSI as those hospitals that do not participate. Hospitals participating for four or more years have on average 3.13 fewer CLABSI per year compared to those participating for less than four years. Highmark's P4P program has shown improved outcomes with regards to CLABSI, but further research is needed to determine if QB is cost effective.

A Mixed-Methods Research Framework for Healthcare Process Improvement.

UNLABELLED: The healthcare system in the United States is spiraling out of control due to ever-increasing costs without significant improvements in quality, access to care, satisfaction, and efficiency. Efficient workflow is paramount to improving healthcare value while maintaining the utmost standards of patient care and provider satisfaction in high stress environments. This article provides healthcare managers and quality engineers with a practical healthcare process improvement framework to assess, measure and improve clinical workflow processes. DESIGN AND METHODS: The proposed mixed-methods research framework integrates qualitative and quantitative tools to foster the improvement of processes and workflow in a systematic way. The framework consists of three distinct phases: 1) stakeholder analysis, 2a) survey design, 2b) time-motion study, and 3) process improvement. RESULTS: The proposed framework is applied to the pediatric intensive care unit of the Penn State Hershey Children's Hospital. The implementation of this methodology led to identification and categorization of different workflow tasks and activities into both value-added and non-value added in an effort to provide more valuable and higher quality patient care. CONCLUSIONS: Based upon the lessons learned from the case study, the three-phase methodology provides a better, broader, leaner, and holistic assessment of clinical workflow. The proposed framework can be implemented in various healthcare settings to support continuous improvement efforts in which complexity is a daily element that impacts workflow. PRACTICAL IMPLICATIONS: We proffer a general methodology for process improvement in a healthcare setting, providing decision makers and stakeholders with a useful framework to help their organizations improve efficiency.

Cost-effectiveness of sofosbuvir-based treatments for chronic hepatitis C in the US.

BACKGROUND: The standard care of treatment of interferon plus ribavirin (plus protease inhibitor for genotype 1) are effective in 50 % to 70 % of patients with CHC. Several new treatments including Harvoni, Olysio + Sovaldi, Viekira Pak, Sofosbuvir-based regimens characterized with potent inhibitors have been approved by the Food and Drug Administration (FDA) providing more options for CHC patients. Trials have shown that the new treatments increased the rate to 80% to 95%, though with a substantial increase in cost. In particular, current market pricing of a 12-week course of sofosbuvir is approximately US$84,000. We determine the cost-effectiveness of new treatments in comparison with the standard care of treatments. METHODS: A Markov simulation model of CHC disease progression is used to evaluate the cost-effectiveness of different treatment strategies based on genotype. The model calculates the expected lifetime medical costs and quality adjusted life years (QALYs) of hypothetical cohorts of identical patients receiving certain treatments. For genotype 1, we compare: (1) peginterferon + ribavirin + telaprevir for 12 weeks, followed by 12 or 24 weeks treatment of peginterferon + ribavirin dependent on HCV RNA level at week 12; (2) Harvoni treatment, 12 weeks; (3) Olysio + Sovaldi, 12 weeks for patients without cirrhosis, 24 weeks for patients with cirrhosis; (4) Viekira Pak + ribavirin, 12 weeks for patients without cirrhosis, 24 weeks for patients with cirrhosis; (5) sofosbuvir + peginterferon + ribavirin, 12 weeks for patients with or without cirrhosis. For genotypes 2 and 3, treatment strategies include: (1) peginterferon + ribavirin, 24 weeks for treatment-naïve patients; (2) sofosbuvir + ribavirin, 12 weeks for patients with genotype 2, 24 weeks for genotype 3; (3) peginterferon + ribavirin as initial treatment, 24 weeks for patients with genotype 2/3, follow-up treatment with sofosbuvir + ribavirin for 12/16 weeks are performed on non-responders and relapsers. RESULTS: Viekira Pak is cost-effective for genotype 1 patients without cirrhosis, whereas Harvoni is cost-effective for genotype 1 patients with cirrhosis. Sofosbuvir-based treatments for genotype 1 in general are not cost-effective due to its substantial high costs. Two-phase treatments with 12-week and 16-week follow-ups are cost-effective for genotype 3 patients and for genotype 2 patients with cirrhosis. The results were shown to be robust over a broad range of parameter values through sensitivity analysis. CONCLUSIONS: For genotype 1, sofosbuvir-based treatments are not cost-effective compared to Viekira Pak and Harvoni, although a 30% reduction in sofosbuvir price would change this result. Sofosbuvir + ribavirin are cost-effective as second-phase treatments following peginterferon + ribavirin initial treatment for genotypes 2 and 3. However, there is limited data on sofosbuvir-involved treatment, and the results obtained in this study must be interpreted within the model assumptions.

Evaluating MEDEVAC Force Structure Requirements Using an Updated Army Scenario, Total Army Analysis Admission Data, Monte Carlo Simulation, and Theater Structure.

In this study, we re-evaluate air ambulance requirements (rules of allocation) and planning considerations based on an Army-approved, Theater Army Analysis scenario. A previous study using workload only estimated a requirement of 0.4 to 0.6 aircraft per admission, a significant bolus over existence-based rules. In this updated study, we estimate requirements for Phase III (major combat operations) using a simulation grounded in previously published work and Phase IV (stability operations) based on four rules of allocation: unit existence rules, workload factors, theater structure (geography), and manual input. This study improves upon previous work by including the new air ambulance mission requirements of Department of Defense 51001.1, Roles and Functions of the Services, by expanding the analysis over two phases, and by considering unit rotation requirements known as Army Force Generation based on Department of Defense policy. The recommendations of this study are intended to inform future planning factors and already provided decision support to the Army Aviation Branch in determining force structure requirements.

Association of food environment and food retailers with obesity in US adults.

The food environment has been shown to be a factor affecting the obesity rate. We studied the association of density of food retailer type with obesity rate in U.S. adults in local regions controlling for socioeconomic factors. Parametric nonlinear regression was used on publically available data (year=2009) at the county level. We used the results of this association to estimate the impact of the addition of a new food retailer type in a geographic region. Obesity rate increased in supercenters (0.25-0.28%) and convenience stores (0.05%) and decreased in grocery stores (0.08%) and specialized food stores (0.27-0.36%). The marginal measures estimated in this work could be useful in identifying regions where interventions based on food retailer type would be most effective.

Aeromedical evacuation planning using geospatial decision-support.

In this study, we proffer an algorithmic, geospatial-based decision-support methodology that assists military decision-makers in determining which aeromedical evacuation (MEDEVAC) assets to launch after receiving an injury location, given knowledge only of terrain, aircraft location, and aircraft capabilities. The objective is for military medical planners to use this decision-support tool (1) to improve real-time situational awareness by visualization of MEDEVAC coverage, showing which areas can be reached within established timelines; (2) to support medical planning by visualizing the impact of changes in the medical footprint to the MEDEVAC coverage; and (3) to support decision-making by providing a time-sorted list of MEDEVAC asset packages to select from, given the location of the patients. This same geospatial-based decision tool can be used for proper emplacement of evacuation assets such that the theater is covered within a truly representative 1-hour response time. We conclude with a discussion of applicability of this tool in medical force structure planning.

Analyzing the future of army aeromedical evacuation units and equipment: a mixed methods, requirements-based approach.

We utilize a mixed methods approach to provide three new, separate analyses as part of the development of the next aeromedical evacuation (MEDEVAC) platform of the Future of Vertical Lift (FVL) program. The research questions follow: RQ1) What are the optimal capabilities of a FVL MEDEVAC platform given an Afghanistan-like scenario and parameters associated with the treatment/ground evacuation capabilities in that theater?; RQ2) What are the MEDEVAC trade-off considerations associated with different aircraft engines operating under variable conditions?; RQ3) How does the additional weight of weaponizing the current MEDEVAC fleet affect range, coverage radius, and response time? We address RQ1 using discrete-event simulation based partially on qualitative assessments from the field, while RQ2 and RQ3 are based on deterministic analysis. Our results confirm previous findings that travel speeds in excess of 250 knots and ranges in excess of 300 nautical miles are advisable for the FVL platform design, thereby reducing the medical footprint in stability operations. We recommend a specific course of action regarding a potential engine bridging strategy based on deterministic analysis of endurance and altitude, and we suggest that the weaponization of the FVL MEDEVAC aircraft will have an adverse effect on coverage capability.

The Future of Vertical Lift: initial insights for aircraft capability and medical planning.

The U.S. Army continues to evaluate capabilities associated with the Future of Vertical Lift (FVL) program-a futures program (with a time horizon of 15 years and beyond) intended to replace the current helicopter fleet. As part of the FVL study, we investigated required capabilities for future aeromedical evacuation platforms. This study presents two significant capability findings associated with the future aeromedical evacuation platform and one doctrinal finding associated with medical planning for future brigade operations. The three results follow: (1) Given simplifying assumptions and constraints for a scenario where a future brigade is operating in a 300 x 300 km2, the zero-risk aircraft ground speed required for the FVL platform is 350 nautical miles per hour (knots); (2) Given these same assumptions and constraints with the future brigade projecting power in a circle of radius 150 km, the zero-risk ground speed required for the FVL platform is 260 knots; and (3) Given uncertain casualty locations associated with future brigade stability and support operations, colocating aeromedical evacuation assets and surgical elements mathematically optimizes the 60-minute set covering problem.

Financial performance monitoring of the technical efficiency of critical access hospitals: a data envelopment analysis and logistic regression modeling approach.

From 1980 to 1999, rural designated hospitals closed at a disproportionally high rate. In response to this emergent threat to healthcare access in rural settings, the Balanced Budget Act of 1997 made provisions for the creation of a new rural hospital--the critical access hospital (CAH). The conversion to CAH and the associated cost-based reimbursement scheme significantly slowed the closure rate of rural hospitals. This work investigates which methods can ensure the long-term viability of small hospitals. This article uses a two-step design to focus on a hypothesized relationship between technical efficiency of CAHs and a recently developed set of financial monitors for these entities. The goal is to identify the financial performance measures associated with efficiency. The first step uses data envelopment analysis (DEA) to differentiate efficient from inefficient facilities within a data set of 183 CAHs. Determining DEA efficiency is an a priori categorization of hospitals in the data set as efficient or inefficient. In the second step, DEA efficiency is the categorical dependent variable (efficient = 0, inefficient = 1) in the subsequent binary logistic regression (LR) model. A set of six financial monitors selected from the array of 20 measures were the LR independent variables. We use a binary LR to test the null hypothesis that recently developed CAH financial indicators had no predictive value for categorizing a CAH as efficient or inefficient, (i.e., there is no relationship between DEA efficiency and fiscal performance).