Optimizing the management of patients in hospital – ClinMine

An increasing number of hospital data is nowadays available. In France, each hospital has data Medicalization Program Information Systems (PMSI) and most of hospitals rely on software to manage their drug prescriptions or test results, including biology. Many studies are emerging to exploit these data, expressing the growing need for data analysis. Often, statistical analysis is a first approach, however, more and more learning and data mining approaches are implemented to deal with these large volumes of data.
In addition, the concept of trajectory of patients becomes more and more important for the business of care. HPST Act (Hospital, Patient, Health Planning) passed in France in July 2009 imposed gradually creating coordinated paths within a country. These trajectories may involve, depending on the conditions, several hospitals, the medical city, health and social institutions... A better understanding of these trajectories would able to focus on trajectories associated with a given disease (associated with stroke, for example) or to identify opportunities for a patient to be recruited in a clinical trial. All these possibilities contribute to improving the care of the patient.
In this context, CLINMINE aims to develop innovative methods to analyze, from the data available in hospitals, types of trajectories. A generic approach is proposed, and different 'case-studies' will be studied in order to validate the approach. The final product will be a functional software platform products based on open licenses.

The project has several innovative characteristics:
- CLINMINE proposes to hybridize computational methods, from the field of optimization and statistics. Indeed, advanced methods of data mining have to be designed and developed for analyzing incomplete data describing patients (PMSI data type, for example) and be able to classify the patient trajectory. These methods must take into account a large number of factors (medical, diagnostic, treatment etc.) and their potential combinations. The selection of factors of interest in building the statistical model for classification represents a high combinatorial challenge and the combinatorial complexity becomes too large to be explored exhaustively. To perform an effective search, combinatorial optimization algorithms (such as metaheuristics) will be developed for their strong ability to explore and manipulate a large number of potential solutions. An innovative aspect deals with the cooperation that we will develop between these optimization methods and statistical approaches to improve the efficiency of the search.
- CLINMINE aims to represent the patient’s trajectories taking into account their fine temporal aspects: placement in time of stays, duration of stay, and by deduction, frequency, frequency change, interruption, proximity in time, etc. For this, we propose to model the evolution over time of the trajectory of a patient by a stochastic process, and analyze a set of trajectories by factorial methods.
- CLINMINE aims to produce a wide dissemination platform. This platform will illustrate the various capabilities of the processing chains of information offered through three major applications related to clinical trials, patients’ trajectories and neurovascular disease.