Lucile Packard Children’s hospital has been pioneering the introduction of target based care for their surgical patients. The overarching goal is to assign each patient that walks through the door to their respective cohort. This would allow the hospital to provide the target length of stay for the patient as well as a standardized clinical pathway. The problem right now, however, is that the process of creating cohorts is manual and time consuming. The goal of this project was to automate the creation of cohorts by gathering a centralized dataset which allows for the filtering of outlier patients from their cohorts to ensure that the set targets are accurate.
Hospital construction of building 500P and renovation of building 300P at Stanford Health Care will lead to shifting capacities in different hospital units over the next five years. The current workflow does not support University Medicine to operationalize six teams across the two future buildings. Given the patient care constraints for each team, it is important to consider the spatial distribution of patients located in different hospital units assigned to any given team. Maximizing the co-location of patients in the same units would allow for more time spent with patients for each team and less time walking in between them.
The objective of this project is to understand the current state of patient co-location in the hospital and to predict future trends given the changing unit capacities. We aim to develop a simulation model that determines future patient volume and unit occupancy and their impact on the co-location of Medicine patients.
Multiple factors and constraints impact how surgeons are scheduled in multiple OR blocks and clinics. These include subspecialty coverage across clinics and fixed clinic days in a specific location. Current methods of scheduling are not quantitative and rarely result in optimal scheduling. The goal of this project is to use a mathematical program to optimize certain conditions, such as maximizing clinic coverage and minimizing driving time, while respecting relevant constraints, such as fixed sessions at a certain clinic.
Planning to provide the highest quality patient care requires three complex decisions: how to estimate patient demand, how to allocate physical capacity, and how to allocate staff. The first decision, the level of patient demand, depends on forecasting based on historical data and strategic intiatives. The second decision, how to allocate physical capacity a given level of demand, depends on a careful analysis of hospital priorities and bottlenecks. The third decision, how to staff, depends on patient needs and staff competencies. The goal of this project is to understand, from the point of view of these decisions, how to best re-allocate and staff APU and SSU capacity to accomodate higher sumemrtime surgical volume.
Scheduled operating room cases are sometimes ‘bumped’ in order to accommodate urgent surgical cases. Numerous factors weigh on the selection of which service should be bumped and how to reschedule the bumped case. The urgency of the situation may make it difficult to decide quickly and fairly. The current method to select the case to be bumped is based on a manual process and a paper log. The log is used to tracks which service was bumped last and which service is due to be bumped next. For this project, we created and implemented a tool that will provide a more efficient, equitable, and automated method to help determine which case to bump when an emergency case arrives.
Accurate preparation of surgical supplies, instrumentation, and equipment is crucial to ensure that each surgery begins and ends on time and with minimal waste. Currently, the process is driven by preference cards that are not representative of what is used in the OR. This results in three significant sources of avoidable costs: 1) delays caused by nurses struggling to find missing equipment immediately before or during a case 2) wasted, unused supplies 3) supplies that are used but not billed for. The goal of our project is to automate the maintenance of preference cards by using historical usage data; to recommend updates to the existing preference cards; and thus, minimize delays, wasted supplies, and improper billing.
Hospital quality of care, nurse retention, and nurse satisfaction depend on the quality of the hospital’s decision-making processes. These decisions include: how to assign nurses to patients, which patients are at risk of decline, when to transfer patients between units, and how to bill for services provided. To assist with these decisions, hospitals currently rely on several closely related, but independent tools. These tools share several shortcomings: the data input is manual; the output requires manual manipulation; and only a small fraction of all relevant data are considered. We propose a single system to replace these tools. The central component of the system is a tool to estimate patient acuity and nurse workload from EMR data. In the proposed system the data input is automated; the output does not require further calculation; and all relevant data in the EMR are considered. Such a system can support the quality of care, nurse retention, and nurse satisfaction.
Planning a Major Hospital Expansion
Lucile Packard Children’s Hospital Stanford is opening a major expansion in December, 2018. To plan the use of the new space effectively, one must forecast demand, allocate capacity to match this demand, and understand the operational implications of this match. The goal of this project is to design tools to meet to plan for surgical services after the expansion. These include a forecasting tool to estimate demand; a simulation tool to test how much capacity will be necessary to accommodate the forecast demand; and an optimization tool to find the best reassignment of surgeons to operating rooms and clinics in the new operational model.
Diabetes and Continuous Glucose Monitors
The patients of the Endocrinology medical service use numerous remote care monitoring devices (e.g., insulin pumps and glucometers). The goal of this project is to design a comprehensive system for patients to learn to use these devices; to deliver the data to healthcare providers; and to put a system in place to analyze the data and track the outcomes of the work.
Intensive glucose control in patients with type 1 diabetes (T1D) is essential to prevent the development of vascular complications. However, in the United States only ~25% of children and adolescents with T1D meet long-term glucose control goals. Despite the high incidence and significant morbidity and mortality associated with T1D, few quantitative tools to personalize care models are available and deteriorating glucose control is often detected only retrospectively. A typical patient may check their blood glucose 4-10 times/day and receive feedback from their care team once every 3 months based on a single lab value (HbA1c) that represents glucose control over the past 3 months. Continuous glucose monitors (CGMs), which record blood glucose levels once every 5 minutes, offer an opportunity to use high-frequency data to identify and predict deteriorating glucose control.
The Imaging services Department at Lucile Packard Children’s Hospital provides services to a patient population with highly variable needs. The variability at each step of the process can impact patient experience and the operations of the department. The process can potentially be improved through redesigned work flows and changes in operating processes.
Our objective is to maximize the Imaging Services’ machine utilization while also improving patient experience by reducing wait time through the minimization of scan delays. Opportunities include emphasis on proactive thinking in patient preparation, optimizing operating hours for increased efficiency, and improving data capture for monitoring performance metrics.
The operating room is a major cost and revenue center. More effective operating room management and scheduling can provide significant benefits. The post-anesthesia care unit (PACU), where patients go to recover after their surgical procedures, is at times a bottleneck. If the PACU reaches capacity, operating room procedures are delayed or canceled. We present a data-driven approach to determine the order of procedures in the operating rooms that minimizes the likelihood of PACU holds. Specifically, we use a machine learning approach to estimate the required PACU time for each type of surgical procedure; we develop and solve integer programming models to schedule procedures in the operating rooms so as to minimize maximum PACU occupancy; and we use discrete event simulation to compare our optimized schedule to the existing schedule.
High-frequency collection of patients’ physiological waveforms at the bedside provides a dense information stream we may use to identify and predict clinical events. Due to data size and complexity, new techniques are required to efficiently suppress noise and extract relevant information. We are constructing an adaptable machine learning toolset to process these waveforms to support applications including crisis event prediction, readmission risk identification, and long-term patient trend analysis. The purpose of these tools is to support sharing methodology between the wide range of relevant problems.
Population Health, Policy, and Spending
The 2020 presidential elections are inching closer, and more than a few candidates are proposing or considering a universal health insurance system. Schulman and Milstein provide a “back of the envelope” estimation for the financial impact of a “Medicare for All” approach to health insurance on US hospitals. This is a tremendous estimation of the impact of such a policy on hospital revenue on an aggregate level. However, we aim to develop a model to inform policy makers and voters about the potential impact of such a policy on deeper levels. We will populate this model using available data on Medicare and private insurer reimbursement. Currently, our model measures the aggregate impact on hospitals in the United States. Future iterations will allow users to explore more granular analysis, such as individual hospitals. This model will be made available to the public to do their own analysis using a user-friendly UI powered by RShiny. The UI will display creative visualizations of the impact of “Medicare for All.”
The Stanford Supply Chain Department is responsible for non-labor spend, contract management, and item and implant procurement for the Stanford hospital system including Stanford Health Care (Adult), ValleyCare, and Lucile Packard Children’s Hospital. The organization participates in a number of cost savings initiatives with physicians in order to reduce spend, ranging from vendor consolidation, contract negotiation, and value-based selection of new cost-effective medical technologies, such as for Spine surgeries. To successfully drive these initiatives, Supply Chain depends on reliable and accessible data to perform the relevant analytics with its available tools. However, these systems have not been thoroughly investigated for the construction of a standardized playbook for cost savings: one that generalizes to all of the enterprise’s departments and units. Our team was tasked with generating a comprehensive data mapping of all the databases, analytical tools, and external benchmarking data services that Stanford uses. This data mapping will help with different goals such as establishing a source of truth for item price, increasing PO accuracy, easily finding and tracking spend per item/department, and being able to present surgeons with an accurate cost-per-case after every surgery. By interviewing over 50 internal and external employees and stakeholders we were able to generate a detailed data mapping and a comprehensive documentation of the many analytical tools and databases used regularly by Supply Chain.
Early identification of non-accidental trauma (NAT) is critical as up to 30% of victims, by some estimates, will be re-injured without adequate protection conferred by early identification. Unfortunately, the identification of NAT is often inconsistent and difficult owing to the sensitive nature of the screening process. This had led to the development of screening protocols by some children’s hospitals to decrease bias and improve NAT detection. Students will work with the clinician and administrative leaders to design a system to improve our screening and reporting of NAT at LPCH.
Perioperative services at Lucile Packard Children’s Hospital provides surgical services to a patient population with highly variable needs. Many of the departments in perioperative services are interconnected and interdependent. In order for perioperative services to perform as efficiently as possible, they must be equipped with the right analytical tools. The objective of this project is to facilitate patient access and efficient resource use through data-driven decisionmaking in perioperative services.
Effective management of operating room resources relies on accurate predictions of surgical case durations. This prediction problem is known to be particularly difficult in pediatric hospitals due to the extreme variation in pediatric patient populations. We pursue two supervised learning approaches: (1) We directly predict the surgical case durations using features derived from electronic medical records and from hospital operational information. For this regression problem, we propose a novel metric for measuring accuracy of predictions which captures key issues relevant to hospital operations. We evaluate several prediction models; some are automated (they do not require input from surgeons) while others are semi-automated (they do require input from surgeons). We see that many of our automated methods generally outperform currently used algorithms and our semi-automated methods can outperform surgeons by a substantial margin. (2) We consider a classification problem in which each prediction provided by a surgeon is predicted to be correct, an overestimate, or an underestimate. This classification mechanism builds on the metric mentioned above and could potentially be useful for detecting human errors. Both supervised learning approaches give insights into the feature engineering process while creating the basis for decision support tools.
Acute Kidney Injury (AKI)
At LPCH about 25% of Intensive Care Unit patients and 5% of Acute Care Unit patients experience an acute kidney injury (AKI) during their stay, a condition associated with chronic kidney disease (CKD) in the months following discharge. The goals of this project are: (1) to better understand what risk factors are associated with the development of CKD after discharge for AKI survivors through retrospective data analysis, and (2) to develop and implement a process for identifying high-risk AKI survivor patients and enrolling them into a AKI survivor clinic post-discharge.
Crisis Event Prediction
Hospitalized patients often exhibit predictive indicators 24-48hr prior to experiencing a life-threatening event. Detecting and filtering these predictors can enable early identification of patient decline, allowing preventative measures to be taken. Current algorithms utilize select clinical variables for predictive modeling, but have generally excluded congenital heart disease due to the heterogeneity of both patient and disease. The goal of this project is to improve on previous methods by incorporating the physiological waveforms collected by bedside monitors. We accomplish this by leveraging deep learning techniques to extract information from these dense data.
Predictive Model of Patient Flow
Patients who receive major cardiovasculare surgery, e.g., unifocalization, require a long recovery stay in the hospital. In order to accommodate more such patients without disrupting patient flow, it is necessary to map their path through LPCH, the resources they need, the other patients who also need those resources, and to create a model to support decisions on how to most efficiently allocate these resources. We will use historical data, expert interviews, and mathematical optimization to build such a model.
The LPCH Cardiovascular (CV) Surgery Unit is one of a select few facilities where families can turn to have complex pediatric cardiac surgeries. However, surgical scheduling has not yet been optimized to allow for more families to take advantage of our world class care. Our project to improve CV access–by optimizing the proxy of block utilization–led us to develop three resources: (1) Detailed process flows of the Pulmonary Artery Reconstruction (PAR) and general CV surgery scheduling processes, capturing relevant decision and outcome variables, (2) Identification of strategies to increase utilization of current OR blocks and (3) A prototype, Excel-based automated scheduling tool, featuring beta versions of ideal functionalities along with the TAP Lists. We analyzed short run-time procedures and suggest that schedulers use our analyses to increase access and OR block utilization. Scheduling more of these types of procedures on days with remaining OR block time would improve access and avoid bumps from delays. With the expansion of the department, continued data analyses and iterations of the tool are critical for further optimization.
Central line associated bloodstream infections (CLABSIs) are hospital-acquired complications that contribute to increased patient morbidity and longer lengths of stay. Lucille Packard Children’s Hospital (LPCH) has previously reported 2.2 CLABSIs per 1000 line-days, which is higher than the national average of 1.5 CLABSIs per 1000 line-days. One of the leading strategies for reducing CLABSI rates is improving compliance with central line bundle elements – a standard of care for dressing changes, antiseptic washes, sterile line accesses etc. Our objective was to use data to reduce CLABSIs by changing institutional practices and individual behaviors.
Lucile Packard Children’s Hospital (LPCH) is home to the premier pediatric heart center on the west coast, working with some of the most challenging cases. LPCH is working to design a new operational model for the Heart Center Clinic. The goals are: 1) to increase capacity by 15% with existing MD/APP staff; and 2) to provide high quality care while actively managing clinic bandwidth, including temporary closure of specific providers to new patients, and proactive addition of new clinics when they lack sufficient capacity.