Measurement and prediction of inpatient case manager workload in a tertiary hospital setting
Author(s)Stuck, Jason Edward
Leaders for Global Operations Program.
Retsef Levi and David Simchi-Levi.
MetadataShow full item record
A patient's care needs often extend past discharge from an acute hospital setting. At Massachusetts General Hospital (MGH), inpatient case managers, acting in a discharge planning capacity, help develop and coordinate the execution of plans, specifically tailored for a patient, to ensure these care needs are met. Case managers, and case management leadership, must confront multiple sources of workload variability across different time and scale perspectives. Case managers are assigned a relatively invariant number of cases by floor. Inter-floor workload variability exists because the "typical" case on one floor may require more or less work than the "typical" case on another floor. Inter-case variability is also present; for a given case manager, the concept of a "typical" case has limited value. Some cases require essentially no work from a discharge planning case manager, while other cases can consume many hours, either on a single day, or spread across multiple days. The case characteristics determining the amount of work required of a case manager are not solely, or even primarily, clinical. Instead, discharge disposition, insurance considerations, patient preferences, and a wide array of psycho-social factors, as well as complex interactions among case characteristics, drive the workload for any single case. Finally, the total amount of work required, across all assigned cases, can vary dramatically from day to day. In any discussion of case manager workload, variability, in all of its dimensions, is a fundamental characteristic. From an operational improvement standpoint workload variability has to be fully considered, understood, and accommodated. The current static staffing scheme, based on the number of beds a case manager is responsible for, does not adequately address the observed variability in daily workload. Therefore, the ultimate objective of our work is to develop a candidate staffing scheme and staffing guidelines incorporating requisite dynamic element to address variability in a case manager's daily workload and/or reduce observed upside variability. Since the requisite understanding of workload variability will always prove elusive without a meaningful way to measure workload, in the first, necessary step for our work we develop a method of measuring the amount of work performed by a case manager, for a given case or on a given day. Though the scale for our work metric requires more refined calibration, it allows one to say with a high degree of certainty that "this case required more work than that case" or "this day represented a higher workload for a case manager than that day". The source of the score for a case or day is the work documented in case manager notes. We develop an automated scoring procedure to retrospectively score cases based on the text of case manager notes. At the heart of our text-analytical engine is an augmented bag-of-words approach that preserves the relevant context for a case manager note. Using a regression tree to operate on our text feature vector for a case note results in validation set scoring with an R2 of 0.98 at the case and day level. In validating our scoring methodology case managers were asked to rank a group of cases in order of increasing workload. This ordinal ranking was compared to the ranking derived from our work score and yielded a value for Kendall's coefficient of concordance, W, of 0.98, indicating exceptional agreement. Results using our score provide further indirect support for the validity of our scoring methodology. For example, the top decile of patients by work score accounted for 40% of the total work scored. This is in line with case manager reports that a relatively small number of patients require a disproportionately large amount of case manager time. Our validated work score is then used as a response variable for explanatory and predictive modeling of case manager workload. The predictor variables are derived from a phased framework we developed over the course of our work. That is, distinct phases can be identified on a discharge planning plane as a patient progresses to ultimate discharge. For the majority of cases it is possible to identify, unambiguously, which phase a case is in. Counts of the number of cases in each phase at 04:00 form our predictor variables in projecting the amount of case manager workload required for the upcoming day. Each phase is associated with both a characteristic amount of work and, as importantly, whether a given case will require any case manager work on a given day. This allows us to introduce the concept of an active census or active caseload. It is this concept that allows us to capture a key, under-considered source of variability - whether a case will require any work of a case manager on a given day. Using a regression-based model, the work for a case manager can currently be predicted with an R2 of 0.51 and a case can be predicted as active with an R2 of 0.66. With classification based on a boosted tree, a day can be correctly predicted as high, medium, or low workload with an accuracy of 81%. Two class misclassification error rates (high-as-low or low-as-high) of 7% can currently be achieved. Finally, in a synthesis of all of our work, we present the outline for a dynamic case assignment scheme based on pooling and balancing the number of cases in each phase between case managers within a pool. This can help attenuate the magnitude of high workload days and reduce upside variability.
Thesis: S.M. in Engineering Systems, Massachusetts Institute of Technology, School of Engineering, Institute for Data, Systems, and Society, 2016. In conjunction with the Leaders for Global Operations Program at MIT.Thesis: M.B.A., Massachusetts Institute of Technology, Sloan School of Management, 2016. In conjunction with the Leaders for Global Operations Program at MIT.Cataloged from PDF version of thesis.Includes bibliographical references (pages 197-205).
DepartmentMassachusetts Institute of Technology. Institute for Data, Systems, and Society.; Sloan School of Management.; Massachusetts Institute of Technology. Engineering Systems Division.; Leaders for Global Operations Program.
Massachusetts Institute of Technology
Institute for Data, Systems, and Society., Sloan School of Management., Engineering Systems Division., Leaders for Global Operations Program.