Publications Partners White Paper Tools and networks for the federation, analysis and use of data in forensic medicine

Forensic Medicine Services


CHU Jean Verdier Bondy


CHU Hôpital Roger Salengro Lille


CHU de Nancy


CHU de Dijon

CHU Tours

CHU Boulogne sur mer

CHU Hôpital de Rangueil Toulouse

Centre Hospitalier Sud Francilien

AP-HP Hôpital Raymond-Poincaré Garches

CHI Hopital Intercommunal Créteil

CHU Hôpital de la Timone Marseille

Are you a medico-legal unit, would you like to join the project?

Other partners

White Book

The origin and founding principles of ORFéAD

Multicenter research in forensic medicine. ORFéAD, for Tools and network for
federation, the analysis and use of data in forensic medicine, covers a set of tools to
provision of a network of practitioners and researchers. ORFéAD was born out of the need to acquire
means to promote multicentre research in forensic medicine. The network was born in
December 2016, marking the start of a pilot phase and assessment of the feasibility and
the acceptability of ORFéAD’s principles. ORFéAD mobilized 5 medical centers when it left
legal. Quickly, he believed to reach 12 crosses. Other centers have expressed their wish
join the network; we preferred to focus on putting the first ones into production
tools first, then open up the network more widely. The pilot phase has been declared
validated by all participating centers in January 2019. The start of production has been validated

ORFéAD’s initial principle: collect and federate data from current practice,
with minimum constraints for practitioners. That is to say, data sharing must be able to
do without additional or specific collection, requiring their own human resources. The
doctor, or any other person identified within the participating center, should not have to
devote significant time.

Structure the data, acquire a common repository. The chosen scheme is based on
production of an object common to all forensic experts: the drafting of a medical certificate, in the form
digital. We use these certificates to extract data that is the same for
all centers. This makes it possible to acquire a common repository, and to structure a base
monocentric by center, and multicentric for all participants. Thus, at a minimum, for
the centers not having a structure of their data for research, ORFéAD their
gives access to their own data, in the form of blocks of text, and in the form of variables

Provide access to data processing, but not disseminate the data. The
data dissemination is a major issue in the age of open data, data security,
possible overlaps between databases and efficient hardware and software
and increasing access to exploit that data. ORFéAD provides participants with its
dedicated interface, Spe3dLab, which provides access to a whole set of data processing operations,
without ever having to export them. The data therefore remains in a secure environment, in a
environment that meets regulatory research requirements. Tools
processing range from simple filters for inclusion criteria, to model building
artificial intelligence, through descriptive and analytical epidemiology, data mining.

Transparency and control over data quality and data processing. In one
secondary data use context – the data used and the variables extracted or
created are created from documents established in an initial context other than research – the
question of data quality is central. ORFéAD has a process for estimating the
quality of its data, which makes it possible to document each variable using conventional indicators
(sensitivity, specificity, precision). The researcher can then decide on the use of such and such

variable, depending on what he wants to do with it. Likewise, all the treatments made available are
documented and proven. The researcher knows exactly what the platform is running.
Compliance with current regulations. ORFéAD must comply with the regulations,
specific to the GDPR. ORFéAD, like most healthcare data warehouses, is based on
CNIL MR 004 reference methodology. The procedures are in progress.

ORFéAD, a suite of tools and a network for research, training and observation of
violence in France

ORFéAD is based on several software modules, designed specifically for research:
Spe3dLab for data creation and analysis; an automatic data extraction module
from textual documents; a module for uploading textual documents; an engine of
pseudonymization of textual documents; a data qualification module

ORFéAD’s primary vocation is to serve the forensic community, but aims
the gradual opening up of its resources to other communities of researchers, above all around
of themes determined jointly. This is the meaning of the connections with
epidemiologists (social epidemiology in particular), and researchers in HSS, in particular at
Toulouse (MSHS-T, PUD-T)

The structure of ORFéAD is that of an open cohort, whose data are constantly
enrichable, depending on the need for more or less targeted studies. This can be done in 3 ways
distinct: i) the creation of new variables by natural language processing and various
AI methods; ii) coding by the practitioner of a new variable, via the upload module of the
document; iii) matching with other databases. Indeed, we have designed ORFéAD and the
data circuit in such a way that matches are possible, on a case-by-case basis, according to the

By opening up its network to more and more legal medicine centers,

ORFéAD also aims to provide an important source and holds a privileged place in terms of
continuous observation of situations of violence

ORFéAD, a pilot study on the determinants of total incapacity for work

Seven centers shared their data for the pilot feasibility phase. The documents
federated medical officers had to respect the following criteria: i) acts of willful violence; ii)
concerning persons aged at least 10 years; iii) delay between the facts and the consultation of
less than 31 days; iv) excluded sexual violence. A total of 10,000 people were included
in this first pilot study

First data to be available, example of quality

Two types of data for the pilot study: data extracted directly from documents,
data created from natural language analysis

40 variables common to the 7 centers, for 10,000 people

These variables concerned: age, sex of victims, examination center, examination time,
different types of traumatic lesions observed, a set of symptoms and repercussions
psychological (appetite and sleep disorders, intrusive symptoms, pain, fear and anxiety, etc.),
or characterizing the circumstances of the violence (use of a weapon, aggressors
multiple …)
The quality of these created variables was evaluated according to a process specific to ORFéAD, by
two reviewers blinded each other. Most of the variables show sensitivity and
specificity of plus 0.89, respectively 0.92. A minority of more complex variables and
requiring another approach, exhibited sensitivities or specificities of the order of 0.60-0.70,
which made them unsuitable for statistical use.

Collaborations and publications

Guez S, Laugier V, Saas C, Lefèvre T The IA, the forensic scientist and the magistrate: forensic treatment
interpersonal violence. In: Science and meaning of artificial intelligence, Julia G. Themes and
comments, Dalloz 2020

2020, Lefèvre T and the Drop It research team, pre-final report of the Drop It project for the
Law and Justice Mission. 162p

51st Francophone International Congress of Forensic Medicine. ORFéAD-ML: back
of experience. A network and a tool dedicated to forensic medicine research. Dijon, 1-3
july 2019

51st Francophone International Congress of Forensic Medicine. Study the determinants of
total incapacity for work. Preliminary results of a multicenter study – a study
ORFéAD-ML. Dijon, July 1-3, 2019

Outlook and roadmap

ORFéAD is in its production start-up phase, and therefore operating under
permanent; several partners partially support its operating costs; others
sources are being researched
Several modules are planned in the short term, including a template creation interface
for the integration of new types of documents