Predictive analytics is increasingly used to guide decision-making in many applications. In practice, we often have limited data on the true predictive task of interest, and must instead rely on more abundant data on a closely-related proxy predictive task. For example, Medicare uses abundant patient readmissions data (proxy) to evaluate a hospital's quality of care rather than the relatively sparse patient mortality rates (true outcome of interest); alternatively, hospitals often rely on medical risk scores trained on a different patient population (proxy) rather than their own patient population (true cohort of interest) to assign interventions. However, not accounting for the bias in the proxy can lead to sub-optimal decisions. Using real datasets, we find that this bias can often be captured by a sparse function of the features. Thus, we propose a novel two-step estimator that uses techniques from high-dimensional statistics to efficiently combine a large amount of proxy data and a small amount of true data. We prove upper bounds on the error of our proposed estimator and lower bounds on several heuristics commonly used by data scientists. In particular, our proposed estimator can achieve the same accuracy with exponentially less true data (in the number of features d). Our proof relies on a new tail inequality on the convergence of LASSO for approximately sparse vectors. Finally, we demonstrate the effectiveness of our approach on training personalized patient risk scores from electronic medical record data; we achieve significantly better predictive accuracy as well as managerial insights into the nature of the bias in the proxy data.
Title: Do Customer Emotions Affect Worker Productivity? An Empirical Study of Emotional Load in Online Customer Contact Centers
Authors: Daniel Altman, Marcello Olivares, and
Galit B. Yom-Tov
Research in Operations Management focused mainly on system-level load, ignoring the fact that service agents and customers express a variety of emotions that may impact service processes and outcomes. We introduce the concept of emotional load - the emotional demands that customer behaviors impose on service agents - to analyze how customer emotions affect service agent productivity. Theories in psychology literature generate ambiguous predictions about the effect of customer emotion on employee productivity: some theories predict that emotions expressed by customers reduce productivity (e.g. by increasing the service time required to serve an angry customer) whereas other theories predict a positive effect (e.g. by “spicing up” the monotonous routine of the agent). We aim to shed light on which theory is more dominant in reality and discuss practical opportunities that arise from measuring emotional load, and how it can be used to enhance productivity. We measure emotional load of agents using a sentiment analysis tool which assesses how positive/negative customer emotion is in chat-based-service interactions, and link it to agent productivity. This is challenging because of a simultaneity estimation problem: it is likely that customer emotions and employee productivity affect one another. Our identification strategy uses exogenous shocks to agent productivity that have been studied in the service operations literature, which effectively disentangle the causal effects between customer emotions and agent response time (RT), our key measure of productivity. The analyses show that emotional load created by negative customer emotions increases agent RT. Emotional load and agent RT reciprocally effect each other, with long agent RT increasing negative customer emotions, suggesting a vicious cycle of emotional load that can hamper system efficiency. We discuss ways to improve productivity by prioritization of tasks by real-time monitoring of emotional load in customer messages.
Title: Big Data for Health Management - Early Diagnosis, Intervention and Prevention
Health is the most important demand for humans. Long and healthy life is one of the primary research subjects in human health research. However, it is difficult to accurately access health status at a very early stage, with the aim of determining appropriate interventions to maintain good health and wellbeing. Therefore, it is essential to optimise human health management polices and assess the risk factors associated with health status. Human health management is the process and means for health risk factors monitoring, prognostics, intervention and control based on our knowledge on human health and prevention using non-clinical and clinical linkage data. Some symptoms that could indicate potential advanced disease or chronic disease can often be ignored or missed. This will lead to serious delay in clinical diagnosis and timely treatment intervention. Subsequently, it will increase the medical treatment costs as well as increasing the patient’s physical, mental and financial burden. Our study aims to develop a systematic approach which integrates statistical and artificial intelligent health big data modelling into optimal health management decision-making with mobile application. By developing statistical modelling method for health big data on early diagnosis, prevention and intervention, we are developing a multi stage delay-time model to investigate risk factors and predict heath status at an earlier stage of disease/illness progression using linked clinical and non-clinical data. In this talk, we will our recent research outcomes and discuss the challenges for the future study.
Title:Optimal Scheduling of Proactive Care with Patient Degradation
Healthcare is a limited resource environment where scarce capacity is often reserved for the most severe patients. However, there has been a growing interest in the use of preventive care to treat patients early on before they deteriorate. On one hand, providing care for patients when they are less critical could mean that fewer resources are needed to return them to a healthy, stable state. On the other hand, utilizing limited capacity for patients who may never need care in the future takes the capacity away from other more critical patients who need it now. To understand this tension, we propose a multi-server queueing model with two patient classes: moderate and urgent. A moderate patient who does not receive treatment may recover and leave or may deteriorate and become an urgent patient. In this setting, we characterize how moderate and urgent patients should be prioritized for care when proactive care for moderate patients is an option. The analysis replies on several interesting applications of optimal control theory. This is joint work with Yue Hu and Carri Chan.
Title:Design of Incentive Programs for Optimal Medication Adherence
Premature cessation of antibiotic therapy (non-adherence) is common in long treatment regimens and can severely compromise health outcomes. In this work, we investigate the problem of designing a schedule of incentive payments to induce socially-optimal treatment adherence levels with heterogeneous patient preferences for treatment adherence that are unobservable to a health provider. Unlike past contract-theoretic models, a unique challenge in this problem is that any prior commitment that a patient makes to a given level of treatment adherence typically cannot be enforced and contracted upon in practice. Consequently, we had to develop new analyses to handle this problem feature. Furthermore, motivated by pragmatic issues of implementing such incentive payments in resource-poor clinics serving a primarily low-income population, we investigate an extension with an additional constraint on the shape of the incentive payment schedule. We show that the optimal payment schedule can be constructed through the solution of a single convex optimization problem in the base case and a sequence of convex optimization problems in the extension. We conduct a numerical study using representative data in the context of the tuberculosis epidemic in India. Our study shows that using either the base case or the extension incentive schedules to encourage treatment adherence would be very cost-effective, although costs may vary widely across incentive schedule functional forms.
Title:Managing Appointment-based Healthcare Services with Strategic Walk-in Patients
In addition to serving patients with scheduled appointments, outpatient care providers often set aside some time to see walk-in patients, who arrive without making an appointment in advance. Facing these two channels of accessing care, patients make choices based on their health conditions and the utilities of these two options. Patients with acute symptoms likely choose to walk in for care right away. Those with less urgent conditions may choose between making an appointment and walk-in based on the trade-off of waiting in two different time scales—appointment delay (days to wait for the appointment) vs. in-clinic waiting (in minutes). A critical question faced by outpatient care providers is how much daily capacity to offer for each of these two care channels, in anticipation for patient strategic choice behavior. We develop a stylized single-server queueing model to shed light on this important practical problem. For any given capacity decision of the provider, we fully characterize patient equilibrium behavior, and then we seek to optimize the provider’s capacity decision, to which patient choices are endogenized. We consider the settings where appointment-seeking patients may have exact or expected appointment delay information; we also compare the model with strategic patients and the model in which no patients are (allowed to be) strategic (e.g., walk-in capacity is reserved for patients with acute conditions via a nurse-triage process). Our results inform the design of appointment scheduling systems for outpatient care providers. Time permitting, we will discuss how daily capacity should be planned for scheduled and walk-in patients depending the practice environment (characterized by our model parameters); what delay information should be provided to appointment-seeking patients; and, how walk-in capacity should be used. This is joint work with Willem van Jaarsveld, Shan Wang and Guanlian Xiao.
Title:Invite Your Friend and You’ll Move Up in Line: Optimal Design of Referral Priority Programs
This paper studies the optimal design of referral priority programs, in which customers on a waitlist can jump the line by inviting their friends to also join the waitlist. Recent years have witnessed a growing presence of referral priority programs as a novel customer acquisition strategy. Multiple variants of this scheme are seen in practice. Some firms give full-priority to referring customers (who can skip over all the non-referring ones), whereas others assign only partial priority (i.e., referring customers can overtake some non-referring ones but not all of them). Yet, some run a first-in-first out (FIFO) scheme without any priority as a referral incentive, while others strategically delay customers by an extra amount to create a referral incentive. The existence of these variations naturally raises the question of what is the optimal referral priority mechanism in general. We formulate an analytical model that integrates queuing theory into a mechanism design framework. The objective of the firm is to maximize the system throughput (i.e., customer acquisition) by inducing the optimal joining and referral behavior. We completely characterize the structure of the optimal mechanism. We show that FIFO is optimal when either the base market size or the referral cost is large. When the referral cost is not too large, partial priority is the optimal structure if the base market size is intermediate and above a certain threshold value; if the base market size is at the threshold value, full priority is optimal; if the base market size is small and below the threshold value, strategic delay is the optimal structure. Our paper rationalizes common referral schemes such as full priority, partial priority, FIFO and strategic delay. We provide prescriptive guidance for launching the optimal referral priority program.