Hospital outliers: Impact on length of stay and long chains as mitigation strategy

Hospital wards specialise on particular types of patients (e.g. orthopaedics, haematology, geriatrics). However, the number of patients each specialised ward can treat on any given day is limited by its number of beds. With the number of hospital admissions fluctuating over time, it routinely happens that patients have to be allocated to wards which are not specialised on the patients' clinical needs - those patients are called outliers. Estimates of the prevalence of outlying vary between 5% and 10% of all hospital patients in the UK.
The dissertation will consist of two parts: In Part 1, we will carry out a thorough econometric analysis to investigate the claim that outliers have a longer hospital length of stay. Our analysis will be based on historic health episode data from Addenbrooke's Hospital and 83 German hospitals. In Part 2, we will apply long chain theory, a classic operations management concept, to the outlier problem. We will describe the implementation of the long chain theory to manage outliers and empirically evaluate its effectiveness.
PART 1: Estimate the effect of being an outlier on length of stay
As highlighted by a recent qualitative study, it is widely believed that outliers receive poorer care: the communication between nurses on the outlying wards and medical staff on the fully occupied home ward of the patient is often problematic; nurses on inappropriate wards have often not the right expertise to care for outliers; and the environment on inappropriate wards may be unsuitable for outlying patients' needs. Poorer care may have an adverse effect on patients' recovery and therewith their length of stay. Existing empirical research does not provide a convincing answer on whether outliers have a longer length of stay.
Estimating outlier effects is complicated: First, outliers may not be randomly chosen. Instead, doctors may well choose healthier patients as outliers if the ward is too busy to receive new patients. This would lead to an underestimation of the outlying effect. Second, relatively few patients are outliers during their entire stay in the hospital. Patients may move between their home ward and outlier wards. This poses the question, how one can adequately measure the degree of outlying for patients over time? To account for these problems, we propose to combine a survival model with an instrumental variable approach to establish a causal link between becoming an outlier and length of stay.
PART 2: Exploring long chain theory to better manage outliers
A hospital can be viewed as a system with n specialised servers (wards) caring for n different types of customers (patients). Long chains is a popular theory in operations management to efficiently coordinate the flow of those customers who cannot be catered for by the server most appropriate to them because this server is busy, and therefore have to be dealt with by another server. Applying long chain theory is likely to help (i) reduce the number of outliers and (ii) concentrate a clinical speciality's outliers on a single other ward ("deputy server") fostering economics-of-scale effects for those outliers' care.
A literature review indicated that there exists no off-the-shelf long chain design readily applicable to outlier management. We will therefore develop a suitable long chain design taking into account the two key features of the outlier problem: (i) the decision on what ward to admit a new patient must be made before knowing how many other patients of the different types will arrive later during the day and (ii) it is strongly preferred to assign patients to the most appropriate ward rather than the deputy ward or any other ward.
After describing the application of the long chain theory to outliers, we will empirically evaluate the effectiveness of our long chain design using our large hospital episode dataset. Our literature review indicates a clear lack of empirical evaluation studies for long chain theory.