Actionable Variables Research Articles

MOREOVER: multiomics MR-guided radiotherapy optimization in locally advanced rectal cancer

BackgroundComplete response prediction in locally advanced rectal cancer (LARC) patients is generally focused on the radiomics analysis of staging MRI. Until now, omics information extracted from gut microbiota and circulating tumor DNA (ctDNA) have not been integrated in composite biomarkers-based models, thereby omitting valuable information from the decision-making process. In this study, we aim to integrate radiomics with gut microbiota and ctDNA-based genomics tracking during neoadjuvant chemoradiotherapy (nCRT).MethodsThe main hypothesis of the MOREOVER study is that the incorporation of composite biomarkers with radiomics-based models used in the THUNDER-2 trial will improve the pathological complete response (pCR) predictive power of such models, paving the way for more accurate and comprehensive personalized treatment approaches. This is due to the inclusion of actionable omics variables that may disclose previously unknown correlations with radiomics. Aims of this study are: - to generate longitudinal microbiome data linked to disease resistance to nCRT and postulate future therapeutic strategies in terms of both type of treatment and timing, such as fecal microbiota transplant in non-responding patients. - to describe the genomics pattern and ctDNA data evolution throughout the nCRT treatment in order to support the prediction outcome and identify new risk-category stratification agents. - to mine and combine collected data through integrated multi-omics approaches (radiomics, metagenomics, metabolomics, metatranscriptomics, human genomics, ctDNA) in order to increase the performance of the radiomics-based response predictive model for LARC patients undergoing nCRT on MR-Linac.Experimental designThe objective of the MOREOVER project is to enrich the phase II THUNDER-2 trial (NCT04815694) with gut microbiota and ctDNA omics information, by exploring the possibility to enhance predictive performance of the developed model. Longitudinal ctDNA genomics, microbiome and genomics data will be analyzed on 7 timepoints: prior to nCRT, during nCRT on a weekly basis and prior to surgery. Specific modelling will be performed for data harvested, according to the TRIPOD statements.DiscussionWe expect to find differences in fecal microbiome, ctDNA and radiomics profiles between the two groups of patients (pCR and not pCR). In addition, we expect to find a variability in the stability of the considered omics features over time. The identified profiles will be inserted into dedicated modelling solutions to set up a multiomics decision support system able to achieve personalized treatments.

Enhancing high-school dropout identification: a collaborative approach integrating human and machine insights

Despite its proven success in various fields such as engineering, business, and healthcare, human–machine collaboration in education remains relatively unexplored. This study aims to highlight the advantages of human–machine collaboration for improving the efficiency and accuracy of decision-making processes in educational settings. High school dropout prediction serves as a case study for examining human–machine collaboration’s efficacy. Unlike previous research prioritizing high accuracy with immutable predictors, this study seeks to bridge gaps by identifying actionable factors for dropout prediction through a framework of human–machine collaboration. Utilizing a large dataset from the High School Longitudinal Study of 2009 (HSLS:09), two machine learning models were developed to predict 9th-grade students’ high school dropout history. Results indicated that the Random Forest algorithm outperformed the deep learning algorithm. Model explainability revealed the significance of actionable variables such as students’ GPA in the 9th grade, sense of school belonging, self-efficacy in mathematics and science, and immutable variables like socioeconomic status in predicting high school dropout history. The study concludes with discussions on the practical implications of human–machine partnerships for enhancing student success.

Enhancing life cycle assessment framework to support product ecodesign through index decomposition analysis

In ecodesign, life cycle assessment (LCA) has not only been used to evaluate a product's environmental sustainability but also serves as a tool for designers to identify areas for sustainability improvement. Conventionally, this is achieved by ranking all contributing LCA variables to identify high-impact areas. However, interpreting and translating this information into actionable design variables can be challenging. An LCA variable can be influenced by multiple design variables, each of which might affect the design's environmental sustainability to varying degrees. Presently, there is a lack of tools capable of analysing the influence of each design variable. To bridge this gap, this study proposes an LCA disaggregation methodology that uses Index Decomposition Analysis (IDA) to assess the environmental sustainability significance of the considered design variables. Using the methodology, designers can make informed decisions that prioritise design effort on design variable with the greatest potential for enhancing the product's environmental sustainability. The proposed methodology is demonstrated through a case study involving the design of a desk fan cover, considering material choice, geometry, and manufacturing process as key design variables.

Model-informed drug development of autologous CAR-T cell therapy: Strategies to optimize CAR-T cell exposure leveraging cell kinetic/dynamic modeling.

Autologous Chimeric antigen receptor (CAR-T) cell therapy has been highly successful in the treatment of aggressive hematological malignancies and is also being evaluated for the treatment of solid tumors as well as other therapeutic areas. A challenge, however, is that up to 60% of patients do not sustain a long-term response. Low CAR-T cell exposure has been suggested as an underlying factor for a poor prognosis. CAR-T cell therapy is a novel therapeutic modality with unique kinetic and dynamic properties. Importantly, "clear" dose-exposure relationships do not seem to exist for any of the currently approved CAR-T cell products. In other words, dose increases have not led to a commensurate increase in the measurable in vivo frequency of transferred CAR-T cells. Therefore, alternative approaches beyond dose titration are needed to optimize CAR-T cell exposure. In this paper, we provide examples of actionable variables - design elements in CAR-T cell discovery, development, and clinical practice, which can be modified to optimize autologous CAR-T cell exposure. Most of these actionable variables can be assessed throughout the various stages of discovery and development as part of a well-informed research and development program. Model-informed drug development approaches can enable such study and program design choices from discovery through to clinical practice and can be an important contributor to cell therapy effectiveness and efficiency.

Rational or Emotional User

Understanding why users continue or discontinue using specific technology is vital for its providers. Existing literature has explored the reasons for continuance and discontinuance by taking into account both rational and emotional factors. However, one question remains unanswered: Why do some users depend more on rational factors for decision-making, while others rely more on emotional factors? This study addresses this question by integrating cognitive decision-making styles and the fear of making incorrect decisions into traditional continued usage models. Data were gathered from 285 TV users and analyzed using structural equation modeling. The results demonstrate that the introduced constructs moderated the direct effects of rational and emotional factors. The contribution of the study lies in incorporating the underlying cognitive processes of decision-making. It presents two actionable variables that managers can utilize to categorize their customers and enhance the effectiveness of their use continuance strategies.

Affective domain predictors of PharmD student pursuit of and successful matching for postgraduate pharmacy residency training.

A study was conducted to identify significant associations between affective domain (AD) features identified using the Birkman Method assessment and students' likelihood to pursue and to successfully match for postgraduate residency training (PGRT), while controlling for demographic and academic variables known to impact PGRT match rates. A retrospective analysis of 3 graduating classes of PharmD students from 2 colleges of pharmacy was performed. Data points such as PGRT match results, PGRT pursuit, student demographics, academic performance information, and AD data from the Birkman assessment were analyzed. Regression analysis was used to identify statistically significant associations between demographic, academic performance, and AD variables with both pursuit of PGRT and successful matching for PGRT. Data from 503 students were evaluated, with 211 (42%) pursuing PGRT. A variety of AD variables were significantly associated with pursuing PGRT and matching for PGRT. Two groups of actionable variables emerged: (1) variables associated with a lesser likelihood of pursuing PGRT but a greater likelihood of successful matching, and (2) variables associated with a greater likelihood of pursuing PGRT and a lesser likelihood of successful matching. Early identification of students' AD features along with specific interventions to promote PGRT pursuit in those less likely to pursue but more likely to match, as well as interventions to promote successful matching in students most likely to pursue but less likely to match for PGRT, is a strategy for possibly optimizing PharmD student career path planning and PGRT match success that merits further evaluation.

Simulating Soil Water Status of Irrigated Fields: The Effects of Soil Data and Root Water Uptake Distribution

<b>Highlights</b> <list list-type=bullet><list-item> The effects of variability in two key inputs to irrigation scheduling models were investigated. </list-item><list-item> Commonly used soil textural data underestimated sand particles, resulting in larger errors in simulated water content. </list-item><list-item> Sensor-based root water uptake distribution led to better model performance than constant and linear patterns. </list-item><list-item> Irrigation timing and amount variables were more impacted by variability in soil data than water uptake distributions. </list-item></list> <b>Abstract</b>. With recent advances in web-based irrigation scheduling tools and mobile applications and the possibility of using more complex modeling approaches, it is important to evaluate the effects of variable input data on the output of these tools and models. Two types of input data that are highly variable across irrigated fields and soil profiles are soil textural data and root water uptake distribution (RWUD). In this study, root zone soil textural data from two sources of commonly used, freely available web soil survey (WSS) and time-consuming, labor-intensive in-situ sampling (ISS) were used in combination with three RWUDs (constant, linear, and sensor-based) to simulate volumetric water content (θv) at four soil layers in six irrigated fields, using the HYDRUS model. The percentage of sand particles based on WSS was about half of the measured amount on average, resulting in a considerable difference in estimated hydraulic properties and soil water thresholds. Sensor data revealed that RWUDs were highly nonuniform, with more than 60% of water extraction occurring from the top 30 cm of the root zone. Among the six combinations of two sources of soil data and three RWUDs, ISS-sensor resulted in the smallest errors in simulated θv, and WSS-constant yielded the largest errors. Simulated θv data were translated to actionable end-user variables of irrigation trigger (IT) and soil water depletion (SWD), which determine the timing and the amount of irrigation applications, respectively. Relying on WSS resulted in irrigation trigger being called about four times more than when measured soil data were used. The average SWD based on WSS was 157 mm, about two times larger than the average SWD based on ISS (68 mm).

The Value and Interpretation of Race and Ethnicity Data in the Era of Global Migration: A Change Is in Order.

ABSTRACT.Human migration and travel are leading to increasingly diverse populations throughout the world. Data collection practices need to adapt to these changes to expand our understanding of health disparities and to optimize the efforts to address health equity, particularly during public health emergencies such as the current COVID-19 pandemic. Race and ethnicity classifications in the United States have failed to evolve since the 1970s despite an increasingly diverse population. Current commonly collected categories are inadequate to accurately describe the economic, educational, and sociopolitical circumstances of different groups. Further, these categories lend little practical information to inform health policy. More predictive and actionable variables should be routinely collected to improve appropriateness and timeliness of health interventions. The immediate adoption of the collection of primary/preferred language and country of birth/origin by public health organizations, health systems, and clinical providers would be a concrete and valuable first step.

Cardiac risk stratification in cancer patients: A longitudinal patient-patient network analysis.

Cardiovascular disease is a leading cause of death in general population and the second leading cause of mortality and morbidity in cancer survivors after recurrent malignancy in the United States. The growing awareness of cancer therapy-related cardiac dysfunction (CTRCD) has led to an emerging field of cardio-oncology; yet, there is limited knowledge on how to predict which patients will experience adverse cardiac outcomes. We aimed to perform unbiased cardiac risk stratification for cancer patients using our large-scale, institutional electronic medical records. We built a large longitudinal (up to 22 years' follow-up from March 1997 to January 2019) cardio-oncology cohort having 4,632 cancer patients in Cleveland Clinic with 5 diagnosed cardiac outcomes: atrial fibrillation, coronary artery disease, heart failure, myocardial infarction, and stroke. The entire population includes 84% white Americans and 11% black Americans, and 59% females versus 41% males, with median age of 63 (interquartile range [IQR]: 54 to 71) years old. We utilized a topology-based K-means clustering approach for unbiased patient-patient network analyses of data from general demographics, echocardiogram (over 25,000), lab testing, and cardiac factors (cardiac). We performed hazard ratio (HR) and Kaplan-Meier analyses to identify clinically actionable variables. All confounding factors were adjusted by Cox regression models. We performed random-split and time-split training-test validation for our model. We identified 4 clinically relevant subgroups that are significantly correlated with incidence of cardiac outcomes and mortality. Among the 4 subgroups, subgroup I (n = 625) has the highest risk of de novo CTRCD (28%) with an HR of 3.05 (95% confidence interval (CI) 2.51 to 3.72). Patients in subgroup IV (n = 1,250) had the worst survival probability (HR 4.32, 95% CI 3.82 to 4.88). From longitudinal patient-patient network analyses, the patients in subgroup I had a higher percentage of de novo CTRCD and a worse mortality within 5 years after the initiation of cancer therapies compared to long-time exposure (6 to 20 years). Using clinical variable network analyses, we identified that serum levels of NT-proB-type Natriuretic Peptide (NT-proBNP) and Troponin T are significantly correlated with patient's mortality (NT-proBNP > 900 pg/mL versus NT-proBNP = 0 to 125 pg/mL, HR = 2.95, 95% CI 2.28 to 3.82, p < 0.001; Troponin T > 0.05 μg/L versus Troponin T ≤ 0.01 μg/L, HR = 2.08, 95% CI 1.83 to 2.34, p < 0.001). Study limitations include lack of independent cardio-oncology cohorts from different healthcare systems to evaluate the generalizability of the models. Meanwhile, the confounding factors, such as multiple medication usages, may influence the findings. In this study, we demonstrated that the patient-patient network clustering methodology is clinically intuitive, and it allows more rapid identification of cancer survivors that are at greater risk of cardiac dysfunction. We believed that this study holds great promise for identifying novel cardiac risk subgroups and clinically actionable variables for the development of precision cardio-oncology.

Factors Associated With Inpatient Endoscopy Delay and its Impact on Hospital Length-of-Stay and 30-Day Readmission

Inpatient endoscopy delay (IED) negatively impacts the delivery of high-quality care. We aimed to identify factors associated with IED and evaluate its effect on hospital length-of-stay (LOS) and readmission. This was a retrospective analysis of all inpatient endoscopies performed between November 2017 and November 2019 at a tertiary care center. IED was defined as the number of days elapsed between anticipated versus actual procedure day. Data were extracted from the endoscopy documentation software and via electronic chart review. Multivariate logistic regressions were modeled to determine variables associated with IED and hospital readmission. A total of 4239 inpatients (mean age, 58.3 years; 50.3% women) underwent endoscopic procedures during the study period of which 819 patients (19.3%) experienced a delay. IED resulted in a median prolonged LOS of 2 days (interquartile range, 1-2 days). Patients with IED were less likely to have an etiology identified on endoscopy (odds ratio [OR], 0.73; 95% confidence interval [CI], 0.63-0.86; P < .001). The 2 most common causes for delays were poor bowel preparation (n= 218; 27%) and lack of endoscopy personnel/unit availability (n= 197; 24.4%). Independent predictors of IED included: older age (OR, 1.1; 95% CI, 1.01-1.03; P= .03), female sex (OR, 1.20; 95% CI, 1.03-1.40; P= .02), use of antithrombotics (OR, 1.30; 95% CI, 1.08-1.57; P= .006), opioids (OR, 1.23; 95% CI, 1.04-1.44; P= .012), being on contact isolation (OR, 1.38; 95% CI, 1.09-1.75; P= .008), and colonoscopy (OR, 1.50; 95% CI, 1.27-1.77; P < .001). Conversely, inpatients admitted to a dedicated GI medicine service were less likely to have IED (OR, 0.79; 95% CI, 0.65-0.96; P= .02). IED was the only independent predictor of 30-day readmission (OR, 1.22; 95% CI, 1.02-1.47; P= .03). IED occurred frequently, unfavorably prolonged LOS, and was an independent risk factor for 30-day readmission. We provide a comprehensive analysis of actionable variables associated with IED that can be targeted to improve inpatient endoscopy delivery.

Modeling Patient No-Show History and Predicting Future Appointment Behavior at the Veterans Administration's Outpatient Mental Health Clinics: NIRMO-2.

No-shows are detrimental to both patients' health and health care systems. Literature documents no-show rates ranging from 10% in primary care clinics to over 60% in mental health clinics. Our model predicts the probability that a mental health clinic outpatient appointment will not be completed and identifies actionable variables associated with lowering the probability of no-show. We were granted access to de-identified administrative data from the Veterans Administration Corporate Data Warehouse related to appointments at 13 Veterans Administration Medical Centers. Our modeling data set included 1,206,271 unique appointment records scheduled to occur between January 1, 2013 and February 28, 2017. The training set included 846,668 appointment records scheduled between January 1, 2013 and December 31, 2015. The testing set included 359,603 appointment records scheduled between January 1, 2016 and February 28, 2017. The dependent binary variable was whether the appointment was completed or not. Independent variables were categorized into seven clusters: patient's demographics, appointment characteristics, patient's attendance history, alcohol use screening score, medications and medication possession ratios, prior diagnoses, and past utilization of Veterans Health Administration services. We used a forward stepwise selection, based on the likelihood ratio, to choose the variables in the model. The predictive model was built using the SAS HPLOGISTIC procedure. The best indicator of whether someone will miss an appointment is their historical attendance behavior. The top three variables associated with higher probabilities of a no-show were: the no-show rate over the previous 2 years before the current appointment, the no-show probability derived from the Markov model, and the age of the appointment. The top three variables that decrease the chance of no-showing were: the appointment was a new consult, the appointment was an overbook, and the patient had multiple appointments on the same day. The average of the areas under the receiver operating characteristic curves was 0.7577 for the training dataset, and 0.7513 for the test set. The National Initiative to Reduce Missed Opportunities-2 confirmed findings that previous patient attendance is one of the key predictors of a future attendance and provides an additional layer of complexity for analyzing the effect of a patient's past behavior on future attendance. The National Initiative to Reduce Missed Opportunities-2 establishes that appointment attendance is related to medication adherence, particularly for medications used for treatment of mood disorders or to block the effects of opioids. However, there is no way to confirm whether a patient is actually taking medications as prescribed. Thus, a low medication possession ratio is an informative, albeit not a perfect, measure. It is our intention to further explore how diagnosis and medications can be better captured and used in predictive modeling of no-shows. Our findings on the effects of different factors on no-show rates can be used to predict individual no-show probabilities, and to identify patients who are high risk for missing appointments. The ability to predict a patient's risk of missing an appointment would allow for both advanced interventions to decrease no-shows and for more efficient scheduling.

A Machine Learning Approach to Management of Heart Failure Populations

BackgroundHeart failure is a prevalent, costly disease for which new value-based payment models demand optimized population management strategies. ObjectivesThis study sought to generate a strategy for managing populations of patients with heart failure by leveraging large clinical datasets and machine learning. MethodsGeisinger electronic health record data were used to train machine learning models to predict 1-year all-cause mortality in 26,971 patients with heart failure who underwent 276,819 clinical episodes. There were 26 clinical variables (demographics, laboratory test results, medications), 90 diagnostic codes, 41 electrocardiogram measurements and patterns, 44 echocardiographic measurements, and 8 evidence-based “care gaps”: flu vaccine, blood pressure of <130/80 mm Hg, A1c of <8%, cardiac resynchronization therapy, and active medications (active angiotensin-converting enzyme inhibitor/angiotensin II receptor blocker/angiotensin receptor-neprilysin inhibitor, aldosterone receptor antagonist, hydralazine, and evidence-based beta-blocker) were collected. Care gaps represented actionable variables for which associations with all-cause mortality were modeled from retrospective data and then used to predict the benefit of prospective interventions in 13,238 currently living patients. ResultsMachine learning models achieved areas under the receiver-operating characteristic curve (AUCs) of 0.74 to 0.77 in a split-by-year training/test scheme, with the nonlinear XGBoost model (AUC: 0.77) outperforming linear logistic regression (AUC: 0.74). Out of 13,238 currently living patients, 2,844 were predicted to die within a year, and closing all care gaps was predicted to save 231 of these lives. Prioritizing patients for intervention by using the predicted reduction in 1-year mortality risk outperformed all other priority rankings (e.g., random selection or Seattle Heart Failure risk score). ConclusionsMachine learning can be used to priority-rank patients most likely to benefit from interventions to optimize evidence-based therapies. This approach may prove useful for optimizing heart failure population health management teams within value-based payment models.

IBS Patients' Treatment Expectancy and Motivation Impacts Quality of the Therapeutic Alliance With Provider: Results of the IBS Outcome Study.

In the absence of a satisfactory medical or dietary treatment, the quality of the therapeutic alliance between irritable bowel syndrome (IBS) patients and their provider is deemed critical to managing refractory IBS. Surprisingly, little research has been conducted on the nature of the therapeutic alliance, factors that influence it, or practical strategies to improve it. This study sought to identify actionable variables that impact therapeutic alliance in patients with refractory IBS. Subjects included a total of 436 Rome III-diagnosed IBS patients (80% female, mean age=41.39 y) who completed a battery of clinical measures at the beginning of the acute treatment phase of an National Institutes of Health (NIH) behavioral trial. Pretreatment candidate predictor variables were organized into 4 categories: sociodemographic, extraintestinal, interpersonal, clinical (eg, symptom severity, pain intensity), cognitive (eg, treatment motivation, expectancy for improvement). Alliance was assessed by patient and clinician-rated measures of the Working Alliance Inventory after first treatment session. Patient reports of alliance were most strongly and consistently predicted by patient access to interpersonal support [β=0.16; 95% confidence interval (CI)=0.07-0.25], motivation for IBS symptom improvement (β=0.12; 95% CI=0.02-0.21), and expectancy of IBS symptom improvement (β=0.35; 95% CI=0.25-0.44). Therapist ratings of alliance also were predicted by patient expectancy of IBS symptom improvement (β=0.16; 95% CI=0.05-0.26). When managing IBS, a focus on dynamic factors of treatment motivation, social support, and treatment expectancy may be useful in improving the quality of the therapeutic alliance between patient and clinical gastroenterologist.

LEAP nets for system identification and application to power systems

Using neural network modeling, we address the problem of system identification for continuous multivariate systems, whose structures vary around an operating point. Structural changes in the system are of combinatorial nature, and some of them may be very rare; they may be actionable for the purpose of controlling the system. Although our ultimate goal is both system identification and control, we only address the problem of identification in this paper. We propose and study a novel neural network architecture called LEAP net, for Latent Encoding of Atypical Perturbation. Our method maps system structure changes to neural net structure changes, using structural actionable variables. We demonstrate empirically that LEAP nets can be trained with a natural observational distribution, very concentrated around a “reference” operating point of the system, and yet generalize to rare (or unseen) structural changes. We validate the generalization properties of LEAP nets theoretically in particular cases. We apply our technique to power transmission grids, in which high voltage lines are disconnected and re-connected with one-another from time to time, either accidentally or willfully. We discuss extensions of our approach to actionable variables, which are continuous (instead of discrete, in the case of our application) and make connections between our problem setting, transfer learning, causal inference, and reinforcement learning.

Predicting solutions of large-scale optimization problems via machine learning: A case study in blood supply chain management

Practical constrained optimization models are often large, and solving them in a reasonable time is a challenge in many applications. Further, many industries have limited access to professional commercial optimization solvers or computational power for use in their day-to-day operational decisions. In this paper, we propose a novel approach to deal with the issue of solving large operational stochastic optimization problems (SOPs) by using machine learning models. We assume that decision makers have access to facilities to optimally solve their large-scale optimization model for some initial and limited period and for some test instances. This might be through a collaborative project with research institutes or through short-term use of high-performance computing facilities. We propose that longer term support can be provided by utilizing the solutions (i.e., the optimal value of the actionable decision variables) of the stochastic optimization model from this initial period to train a machine learning model to learn optimal operational decisions in the future. In this study, the proposed approach is employed to make decisions on transshipment of blood units in a network of hospitals. We compare the decisions learned by several machine learning models with the optimal results obtained if the hospitals had access to commercial optimization solvers and computational power, and with the hospital network’s current empirical heuristic policy. The results show that using a trained neural network model reduces the average daily cost by about 29% compared with current policy, while the exact optimal policy reduces the average daily cost by 37%. Although optimization models cannot be fully replaced by machine learning, our proposed approach while not guaranteed to be optimal can improve operational decisions when optimization models are computationally expensive and infeasible for daily operational decisions in organizations such as not-for-profit and small and medium-sized enterprises.

Precision medicine in pancreatic cancer: treating every patient as an exception

Patients with pancreatic cancer have not benefited from recent improvements in overall survival brought about by precision medicine in other malignancies. This failure is not due to a dearth of precision-medicine research in pancreatic ductal adenocarcinoma (PDAC), the main type of pancreatic cancer. In fact, the stalled progress in precision therapies for this type of cancer is due to the absence of agents that are able to target the common genetic alterations in PDAC. Several studies have attempted to phenotypically stratify PDAC at the transcriptional level. However, the value of such classifications will only be revealed through prospective studies and, crucially, only after development of new treatment options for this disease. Therefore, it is essential to learn from breakthrough discoveries in other cancer types that could benefit subpopulations of patients with PDAC and convert them from ordinary to exceptional responders. Identifying these exceptional patients will help to bring PDAC in line with other cancer types in terms of availability of precision therapies. Thus, the true challenge to precision medicine for PDAC might be the poor consensus on which genetic and phenotypic alterations across the spectrum of this disease are actionable; not the absence of actionable variables themselves. To reach consensus, knowledge and tools must be developed and disseminated for individuals who provide pancreatic cancer care, to enable the real-time identification of exceptional patients, more precise subgroup classifications, and effective disease management strategies; all informed by immediate feedback from clinical outcome data.

PHP256 - MACHINE LEARNING TO IDENTIFY REAL-WORLD PATIENT CHARACTERISTICS ASSOCIATED WITH LENGTH OF HOSPITALIZATION

Despite substantial morbidity, mortality, and economic burden due to length of hospitalization (LOH) for sepsis patients, little is known about actionable variables to minimize its impact. This study aimed to identify unique patient clusters and assess feature’s influence on LOH. Patients (≥18 years) with sepsis discharge diagnoses (ICD-9 or 10 codes) from 2012 to 2016 were identified in Premier, a U.S.-based hospital database. Characteristics included sociodemographic, sepsis class, organ failure, infection, hospital size and hospital ward location. Unsupervised k-means clustering identified unique patient clusters, (validated by depicting the first two principal components and with clustering labels) followed by partial least-squares regression (PLSR) to identify feature importance associated with LOH in each cluster. Feature importance was estimated from significant cross-validated regression components. 390,497 patients were included, mean age 65 years (18-89), 50% male. Mean LOH was 10 days (SD: 11) with discharge diagnoses of sepsis (58%), severe sepsis (19%), and septic shock (23%). Three unique clusters were identified based on calculated withinness values. A total of three, four, and three significant components extracted from PLSR analysis on patient clusters one to three, respectively. In all the clusters the severity of sepsis and then organ failure have the highest influence in LOH. The importance ranking of the other features are different for each class: in class one, the hospital size (staying in hospitals bigger than 200 beds ) and receiving care in South Atlantic, in class two and three having infection, and then admission type as urgent and trauma center (comparing to emergency), are the most positively influential factors in increasing the length of hospitalization. Results suggest that unique sepsis patient clusters exist. Further exploration will provide greater precision and delivery of personalized care with need for validation across patient populations and care settings to identify actionable insights for LOH reduction.

Do life events always lead to change in purchase? The mediating role of change in consumer innovativeness, the variety seeking tendency, and price consciousness

Consumers can experience major life events (e.g., the birth of a child or retirement) that have key strategic managerial implications for firms in terms of creating new market segments as well as preventing defection from existing brands. Therefore, in the context of a consumer-based strategy, this study investigates whether life events change actual purchase behavior as well as how three important and managerially actionable consumer difference variables (consumer innovativeness, the variety seeking tendency, and price consciousness) mediate this link. Purchase records of an individual-level consumer panel, combined with demographic and survey data over a three-year period, indicate that there is a direct link between life events and changes in share of wallet (SOW) and share of units (SOU) for the preferred brand in a product category. Further, experiencing a life event leads to an increase in consumer innovativeness that, in turn, decreases SOW/SOU for the preferred brand for some purchases, while at the same time leading to a decrease in consumers’ variety seeking tendency and price consciousness for others that, in turn, lead to an increase in SOW/SOU. These findings suggest a paradox of change that can both enhance and inhibit purchasing new brands. Finally, the effects of first-time life events (i.e., significant markers in the course of life that occur for the first time) are stronger than the effects of repeated life events, and life events with a positive economic impact have a strong negative effect on change in price consciousness.

Associates of Mortality and Hospitalization in Hemodialysis: Potentially Actionable Laboratory Variables and Vascular Access

To determine the most significant potentially actionable clinical variables associated with mortality and hospitalization risk in hemodialysis (HD) patients. Cohort study. Adult maintenance HD patients in the Fresenius Medical Care, North America database as of January 1, 2004, with baseline information from October 1, 2003, to December 31, 2003, comprising approximately 26% of the US HD population. Case-mix (age, sex, race, diabetes, vintage, and body surface area), vascular access, and laboratory (albumin, equilibrated Kt/V, hemoglobin, calcium, phosphorus, creatinine, bicarbonate, biointact parathyroid hormone, transferrin saturation, and white blood cell count) variables. 1-year mortality and hospitalization risk from January 1 to December 31, 2004. Cox proportional hazards models for death and hospitalization. The cohort (N = 78,420) had a mean age of 61.4 +/- 15.0 years, 47% were women, 49% were white, 41% were black race (10% defined as "other"), and 52% had diabetes. The top 5 actionable variables were the same for mortality and hospitalization. Final case-mix plus laboratory-adjusted hazard ratios for these top 5 actionable variables indicate 177% increased risk of death and 67% increased risk of hospitalization per 1-g/dL decrease in albumin level, 39% and 45% greater risk with catheters compared with fistulas, 18% and 9% greater risk per 1-mg/dL greater phosphorus level, 11% and 9% lower risk per 1-g/dL greater hemoglobin level, and 5% and 2% greater risk per 0.1-unit decrease in equilibrated Kt/V, respectively (all P < 0.0001). Observational cross-sectional study with limited comorbidity adjustment (for diabetes). The same variables are associated with both mortality and hospitalization in HD patients. The top 5 potentially actionable variables are readily identifiable, with albumin level and catheter use the most prominent, and all 5 are appropriate targets for improvement.

A Feature-Based Approach to Market Segmentation via Overlapping K-Centroids Clustering

Nonhierarchical partitioning techniques are used widely in many marketing applications, particularly in the clustering of consumers, as opposed to brands. These techniques can be extremely sensitive to the presence of outliers, which might result in misinterpretations of the segments, and subsequently to inferring incorrect relationships of segments to independently defined, actionable variables. The authors propose a general approach to market segmentation based on the concept of overlapping clusters (Shepard and Arabie 1979), wherein each pattern of overlap can be interpreted as a distinct partition. Both K-means and K-medians clustering procedures are special cases of the proposed approach. The suggested procedure can handle relatively large data sets (e.g., 2000 entities), is easily programmable, and hence can be gainfully employed in marketing research.