David Scheinker

Bounded extremum seeking with discontinuous dithers

The analysis of discontinuous extremum seeking (ES) controllers, e.g.?those applicable to digital systems, has historically been more complicated than that of continuous controllers. We establish a simple and general extension of a recently developed bounded form of ES to a general class of oscillatory functions, including functions discontinuous with respect to time, such as triangle or square waves with dead time. We establish our main results by combining a novel idea for oscillatory control with an extension of functional analytic techniques originally utilized by Kurzweil, Jarnik, Sussmann, and Liu in the late 80s and early 90s and recently studied by Durr et?al. We demonstrate the value of the result with an application to inverter switching control.

Detecting Inaccurate Predictions of Pediatric Surgical Durations

Accurate predictions of surgical case lengths areuseful for patient scheduling in hospitals. In pediatric hospitals, this prediction problem is particularly difficult. Predictions aretypically provided by highly trained medical staff, but thesepredictions are not necessarily accurate. We present a noveldecision support tool that detects when expert predictions areinaccurate so that these predictions can be re-evaluated. We explore several different algorithms. We provide methodologicalinsights and suggest directions of future work.

A quality improvement initiative to optimize dosing of surgical antimicrobial prophylaxis

The risk of surgical site infections is reduced with appropriate timing and dosing of preoperative antimicrobials. Based on evolving national guidelines, we increased the preoperative dose of cefazolin from 25 to 30 mg·kg−1. This quality improvement project describes an improvement initiative to develop standard work processes to ensure appropriate dosing.

Improving predictions of pediatric surgical durations with supervised learning

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.

A Practical Approach to Machine Learning for Clinical Decision Support

Machine learning has produced effective clinical decision support tools. The impact of such work is limited by the difficulty of implementing such tools outside the institution where they were designed. The recent wide-spread adoption of Electronic Medical Record systems (EMRs) makes possible the development and application of tools across institutions. We describe three machine learning projects to develop generalizable, EMR-based clinical decision support tools at the cardiac care units of Lucile Packard Children’s Hospital Stanford: false alarm suppression, detection of critical events, and automated identification and detection of drug-drug interactions. These projects utilize flexible statistical and deep learning frameworks to enable automated, patient-specific care. We focus on the practical challenges of implementing such methodology and describe our progress on producing tools useful for our institution.

Analytical Approaches to Operating Room Management

In recent decades, healthcare has become increasingly expensive, creating pressure on healthcare providers to cut costs while maintaining or improving quality. Operations research can play an important role in supporting such efforts. A key challenge faced by hospital planners is scheduling and management of operating rooms, as operating rooms typically provide highly specialized care, require significant resources, and contribute significantly to a hospital’s bottom line. We describe recent work on hospital operating room management at Lucile Packard Children’s Hospital Stanford. We describe preliminary outcomes of three projects aimed at improving the efficiency of the hospital’s operating rooms: machine learning to improve surgical case length estimation; queuing analysis to improve operational efficiency; and integer programming to schedule cases to reduce surgical delays.