Reducing the impact of no-shows in healthcare by data-driven patient scheduling

Patients who miss their appointments, referred to as "no-shows" or "did not attend" (DNA), have detrimental effect on healthcare delivery performance. They waste valuable scarce resources, thus affecting the efficiency of the healthcare provider. This is even more important in the current Covid19 pandemic with a higher number of patients in the waiting lists, and potentially reduced number of available healthcare staff. Electronic healthcare records enable the analysis of historical patients' admission data to predict no attendance of each patient.
In this project Key research questions are (1) How can state-of-the-art data analytics be used to predict the probability of a patient attending the appointment and(2) How can this information be used to make the patient schedule more resilient and efficient.
We will collaborate with the University Hospital of Coventry and Warwickshire. They will be involved in (1) data collection, (2) discussion about the real-world factors potentially useful for identification of DNA patients and about their current scheduling practice to mitigate the effect of DNA, and (3) the evaluation of the developed decision support tool.
Past research identified depending on the type of facility and practice [1, 2], a patient's non-attendance to scheduled appointments may affect productivity, consume resources, prolong the waiting time for an examination and reduce customer satisfaction.
The novelties of the proposed research include
- the investigation of state-of-the-art data analytics tools for predicting no-show patients. Ideally, these would not only predict the probability of DNA, but additionally provide a confidence score.
- a tight integration with the patient scheduling system that exploits the predictions and recommends time slots for patients and overbooking limits that lead to a resilient schedule.
- the explicit consideration of multiple objectives such as a patient's time to get an appointment, a patient's waiting time at the ward, lost doctor's time, doctor's overtime, and unused time slots.

The PhD project will include the following main steps.
1. Collect historical admission data and prepare them for data mining, including handling of imbalance, which most DNA datasets exhibit (majority of patients show up).
2. Investigate and narrow down factors relevant to the prediction of no-show patients, using techniques such as meta-heuristic search. Possible factors include patient's age, condition and perceived urgency, employment, lead time, transport/parking facilities, etc. A suitable set of factors decreases model complexity and training time, and avoids overfitting. The identified factors will then be used in state-of-the-art machine learning techniques such as Random Forest, Support-Vector Machine, Artificial Neural-Network to predict a patient's DNA probability.
3. Build a simulation model. This will be needed for demonstrating the effectiveness of the developed methodology, but also for simulation-based optimisation.
4. Develop a scheduling approach based on priority rules, which would suggest the most suitable time slot from a schedule resilience perspective. For example, "risky" patients could be offered time slots such that they are spread across the day and across the week, and perhaps not early on the day. The scheduling of patients with historical record of being late will be scheduled in a similar manner.
5. Develop a prototype tool that assists a healthcare manager in scheduling appointments. .

References:
1. Collins J, Santamaria N, Clayton L. Why outpatients fail to attend their scheduled appointments: a prospective comparison of differences between attenders and non-attenders. Aust Health Rev. 2003;26:52-63.
2. Moore CG, Wilson-Witherspoon P, Probst JC. Time and money: effects of no-shows at a family practice residency clinic. Fam Med. 2001;33:522-7