Lots to learn about automated medical rostering
HosPortal has had a long-standing aversion to try to fully automate rostering. We know this is all the trend in the USA, but we have been concerned that the complexities of the rosters in any real hospital mean that having a system to manage it also requires hiring a quantitative analyst to manage the system!Since we were asked to do preference-based rostering we have looked quite hard at rostering rules and now have a system that can build rosters. But we still believe that trying to build software that can do unintermediated rostering is a waste of development effort for most doctor rosters.
State of the art of rostering
Many of the rostering challenges are summarised by the 'nurse rostering problem' that is tackled by academic research. A summary of this research (albeit a little dated) is in this excellent overview by Burke et al. from the Journal of Scheduling 7(6):441-499.For us to make sense of this overview of 30 years' of academic research and more recent published papers we have had to brush up our familiarity with such concepts as heuristics, hill-climbing and memetic algorithms, pareto-simulated annealing and tabu search.The implied conclusions of Burke et al. is interesting: there is nothing out there, theoretically or practically, that can actually solve the real-life rostering problems reliably.Our conclusion is unchanged: setting the rostering rules for any hospital environment such that a computer can generate a result without supervision is not worth the effort.
HosPortal's approach to automated rostering
Our system aims to balance the challenges to ultimately end up with a tool that will work in all hospital environments. At its core, it provides the tools to allow an administrator to easily and quickly manage users' preferences, the hospital's needs, and the working rules attached to each on-call roster.These criteria can be set in a couple of minutes.Our auto-filling algorithm quickly generates rosters that meet all the defined criteria, and then leaves the administrator to do the hard bits that humans are best placed to do, such as working out who to allocate to shifts that no one wants, and which of the roster constraints should be broken in this case.We would be delighted to talk about this in detail. Get in touch for a demo.