Section: New Results

Modeling and optimizing patient pathways in hospital

Participants : James Leifer, Michel Sorine.

External scientific collaboration with:

- Niccolo Curatolo, Directeur des opérations, Hôpitaux universitaires Paris-Sud, Assistance publique-Hôpitaux de Paris (AP-HP);

- Dr Maurice Raphaël, Chef de service, Urgences adultes, Hôpital Bicêtre, AP-HP;

- Dr Christophe Vincent-Cassy, Responsable des systèmes informatiques des urgences, AP-HP;

- Lucie Gaillardot-Roussel, Ingénieur en organisation, AP-HP;

- Dr Paul Jarvis, Senior consultant doctor in emergency medicine, Calderdale and Huddersfield National Health Service Foundation Trust, UK.

In 2014, we began a case study of the emergency department (ED) at Bicêtre Hospital, a large ED handling 50,000 patient visits per year, which is amongst the top 10 by volume and by annual volume growth for EDs in the Paris region.

Rather than presume the appropriateness of a predetermined scientific formalism, our strategy was to allow the application to frame a series of questions in order to lead us to experiment with several potential scientific tools at the present “low risk, high uncertainty” phase of investigation:

- Top-down modeling: Can we capture the expert knowledge of doctors and nurses as to the pathways followed by their patients by transforming this knowledge into a series of “use case” rules borrowed from the techniques of software specification? Can these rules by transformed into an executable model using business process modeling languages and tools (Orc, YAWL, ...) for simulating the complex parallel composition of man-machine processes in a hospital setting?

- Bottom-up modeling: How can the hospital be instrumented for cheaply and accurately capturing its real activity (movement of people and machines, delays, errors, ...) and tuning the parameters of the model? Can we intercept HL7 messages (a standardized electronic message format for medical data) and/or access raw time-stamped database entries to use machine learning techniques (particularly process mining) to extract from the running hospital the graphs representing the actual sequence of care events in order to get rapid feedback about the most heavily used and most often delayed path segments?

- Underlying cost semantics: Can we formalize in process calculi (for example, a variation of pi calculus) the “micro internal economy” of costs exchanged inside the hospital to quantify the economic performance of each patient pathway?

- Offline experimentation and optimization: Can potential optimization to the model be explored offline in a sort of “serious game” to allow non-intrusive experimentation with different strategies for eliminating bottlenecks, increasing flow rates, decreasing costs, etc.?

- Data visualization for medical personnel: Given that the medical personnel themselves are best suited to fixing the daily frictional time losses that most are resigned to accept as “part of the job”, how can the model be presented in a visually lucid manner to render the previously “invisible” aspects of the hospital's organization visible?

- Online real-time control: Can the feedback loop be completed and the model be used to directly provide real-time visual feedback to the hospital personnel to enable them to measure their systemic progress (or systemic unintended consequences) of their localized optimizations?