Reference metadata describe statistical concepts and methodologies used for the collection and generation of data. They provide information on data quality and, since they are strongly content-oriented, assist users in interpreting the data. Reference metadata, unlike structural metadata, can be decoupled from the data.
National Institute of Health and Medical Research (Institut National de la Santé et de la Recherche Médicale - Inserm)
1.2. Contact organisation unit
US 10 - Centre d'épidémiologie sur les causes médicales de décès (CépiDc).
1.3. Contact name
Confidential because of GDPR
1.4. Contact person function
Confidential because of GDPR
1.5. Contact mail address
diffusion.cepidc@inserm.fr
1.6. Contact email address
Confidential because of GDPR
1.7. Contact phone number
Confidential because of GDPR
1.8. Contact fax number
Confidential because of GDPR
2.1. Metadata last certified
8 July 2025
2.2. Metadata last posted
8 July 2025
2.3. Metadata last update
8 July 2025
3.1. Data description
Data on causes of death (CoD) provide information on mortality patterns and form a major element of public health information.
CoD data refer to the underlying cause of death (UCD) which is defined by the World Health Organisation (WHO) as "the disease or injury which initiated the train of morbid events leading directly to death, or the circumstances of the accident or violence which produced the fatal injury".
The regional breakdown is based on the NUTS 2 codes from the 2023 version of the Nomenclature of Territorial Units for Statistics (NUTS), see Commission Regulation (EU) 2023/674.
3.4.1. National definition used for usual residency
Place of residence as declared on the death certificate.
According to the instructions in section 3, subsection 6 of the General Instruction on Civil Status of 11 May 1999 (Annex) given to the civil servants in charge of drawing up acts of civil status, the place of residence declared on the death certificate is "the place where one has the center of one's interests, business, and relations". This definition may differ, in certain cases of dual residence, from that of "habitual residence", but it is consistent with the one used by France to establish its death statistics.
3.4.2. Stillbirth definition and characteristics collected
The French death certificate does not apply to stillbirths. Stillbirths information is collected from the « programme médicalisé des systèmes d’information » (PMSI) by the Directorate of Research, Study, Evaluation and Statistics (DREES) Population health office depending on the Health Ministry.
All stillbirths with weight births over 500 grams or with gestational ages of at least 22 weeks are collected. The terminations of pregnancy are included from 22 weeks (gestational age) onwards.
Agregated data on stillbirths provide information on :
country/region of residence
country/region of occurrence
year and month of occurrence
sex
gestational age
mother age
cause of death
Causes of deaths are provided using the "P95" and "P964" codes from the 10th revision of the International Statistical Classification of Diseases and Related Health Problems (ICD-10).
3.5. Statistical unit
The statistical units are deceased persons and stillborns.
3.6. Statistical population
Residents and non-residents who died in France NUTS FR (see 3.7 for geographical scope).
3.6.1. Neonates of non-resident mothers
Neonates death certificates of non-resident mothers are included.
3.6.2. Non-residents
Non resident population is included.
3.6.3. Residents dying abroad
Residents dying abroad are not included.
3.7. Reference area
All French NUTS2, including Metropolitan France and overseas departments and regions (la Guadeloupe, la Martinique, la Guyane, la Réunion, Mayotte) + Saint-Martin, according to the French geographical nomenclature as of January 1st of the year of death established by the French National Institute of Statistics (Insee) : https://www.insee.fr/fr/information/6051727
Deaths occuring in other overseas areas (i.e. overseas collectivities) or abroad are not included, except for Saint-Martin, which is included in NUTS 2 (FRY1)
3.8. Coverage - Time
Yearly from 2011 (included) onwards.
3.9. Base period
Not applicable.
The unit is number.
Data refer to the calendar year (i.e. all deaths occurring during the year).
6.1. Institutional Mandate - legal acts and other agreements
French law Article L2223-42 of the "Code général des collectivités territoriales" mandates Inserm- CépiDc as in charge of the establishment of causes of death (CoD) national statistics in France.
CoD data can only be used for public health purposes :
For monitoring and alert purposes, by the State, the regional health agencies (ARS) and the national public health agency (Santé publique France)
For the production of the national cause of death statistics and for public health research, by the National Institute of Health and Medical Research (Inserm)
For the processing of data relative to health, under the conditions set out in Article L1461-3 of the Public Health Code
For inclusion into the National Health Data System (SNDS) defined in Article L1461-1 of the same Code
For the production of statistics within the framework of the Article 7 bis of the law n° 51-711 from 7 june 1951 on the obligation, the coordination, the statistical secrecy, by the National Institute of Statistics and Economic Studies (Insee), or by the statistical services of the Minister of Health (DREES). These datasets must be kept separatly from the National register for the identification of individuals data owned by the National Institute of Statistics and Economic Studies (Insee)
7.1. Confidentiality - policy
Regulation (EC) No 223/2009 on European statistics (recital 24 and Article 20(4)) of 11 March 2009 (OJ L 87, p. 164), stipulates the need to establish common principles and guidelines ensuring the confidentiality of data used for the production of European statistics and the access to those confidential data with due account for technical developments and the requirements of users in a democratic society.
According to Article L1461-1 of the Public Health Code, the persons responsible for processing personal health data from the National Health Data System, as well as those implementing it or authorised to access the personal data resulting from it, shall be subject to professional secrecy under the conditions and under the penalties laid down in Article 226-13 of the Penal Code which states that the disclosure of secret information by a person who is in possession of it either by virtue of his or her status or profession, or by virtue of an office or temporary assignment, is punishable by one year's imprisonment and a fine of 15,000 euros.
The French Council of State decree referred to in Article L1461-7 of the Public Health Code draws up the list of State services, public establishments or bodies entrusted with a public service mission authorised to process personal data from the National Health Data System (SNDS) for the purposes of their missions.
7.2. Confidentiality - data treatment
All age groups showing a total mortality of less than 5 cases are considered as confidential. Therefore, any 'confidential' agegroup is grouped with another one to have higher numbers. In practice, this problem mainly occurs for young ages so, either the ages from 0 to 14 years old, or the ages from 0 to 14 and 15 to 24 years old are grouped. The age groups considered as confidential show then the value ':' and the age group '0-14y' (and '15-24y' if needed) shows the sum of all ages before 15 years old (or between 15 and 24 years old). In addition, special measures for ensuring confidentiality may be taken for small countries. For neonatal figures, no breakdown by parity is displayed to ensure confidentiality.
8.1. Release calendar
Following Regulation (EC) No 328/2011, Inserm-CépiDc has to send data to Eurostat at the latest 24 months after the end of the year of death. National releases follow as soon as possible.
Data are released on the website of CepiDc, freely accessible.
Annual.
10.1. Dissemination format - News release
The dissemination of the 2023 cause of death data, on July 8th 2024, was coupled with two publications and a news release. This mirror the approach taken in 2020, 2021 and 2022, where the cause of death data was also disclosed with a new release and two accompanied publications.
All the relevant news releases and publications can be accessed on the CepiDc website.
For 2021, data dissemination included two publications:
Fouillet A, Ghosn W, Rivera C, Clanché F, Coudin É. Grandes causes de mortalité en France en 2021 et tendances récentes. Bull Épidémiol Hebd. 2023;(26):554-69.
Cadillac M, Clanché F, Coudin E, Fouillet A, Riviera C. Les grandes causes de décès en France en 2021 : une année encore fortement impactée par le Covid-19. Études et Résultats. 2023;(1288)
For 2020, data dissemination included two publications:
Fouillet A, Ghosn W, Naouri D, Coudin E. Covid-19 : troisième cause de décès en France en 2020, quand les autres grandes causes baissent. Bull Épidémiol Hebd. 022;(Cov_16):2-15.
Naouri, D., Fouillet, A., Ghosn, W., Coudin, E. (2022, décembre). Covid-19 : troisième cause de décès en France en 2020, quand les autres grandes causes de décès baissent. DREES, Études et Résultats, 1250.
and Rey, 2016 : "Death certificate data in France: Production process and main types of analyses", La revue de medecine interne, doi : J.revmed.01 January 2016
And on annual production reports available on the website.
11.1. Quality assurance
The causes of death data are based on European regulation and European Statistics Code of Practice, which defines scope, definitions of variables and characteristics of the data, and on the WHO guidelines. France follows these clear and detailed guidelines for recording and coding the causes of deaths using the ICD-10 classification published by the WHO.
France also performs a review stage after an initial coding, especially for the most complex cases and the most frequent errors. In order to ensure cause of death coding quality, checks on specific types of coded death certificates are performed regularly. Those manual checks are performed though the Iris coding software user interface.
inconsistencies between age or sex and cause of death
applying explicit checks on certain causes of death (violent deaths, maternal deaths, neonatal deaths, COVID-19, ...)
addressing technical issues identified in the automatic coding process (automatic coding could be incorrect)
ensuring coding quality :
for sensitive causes of death (infant mortality, AIDS, ...) for human coding of complex or new coding rules
Random quality checks are also made to assess the quality of automatic coding and the consistency of manual coding.
For deaths in 2018 and 2019, produced in a context of production catch-up, all usual checks are not performed in the same way as usual. The production report details which types of checks amongst those listed above were performed, entirely, partially or let apart.
In the assessment of coding quality for the years of 2018, 2019, 2021, 2022 and 2023, a thorough evaluation was conducted employing a method that integrates automated rule-based coding, AI coding, and assisted manual coding. The evaluation for 2018 and 2019 are extensively documented in Zambetta et al 2023. This approach goes beyond the conventional coding campaigns, based on automated rule-based coding, and assisted manual coding, offering a more exhaustive analysis.
11.2. Quality management - assessment
Quality assesment is conducted within the INSEE quality approach framework.
Following an evaluation process by the Official Statistics Authority, French cause of death statistics produced by the Inserm-CépiDc was given a quality label by the november 14, 2017 notice for a duration of 5 years. At the end of 2022, the CépiDc did not ask for a renewal of the label as the structural delay for transmitting to Eurostat and disseminating annual CoD data was not solved yet. As of 2024, the dissemination delay problem is now resolved. In 2025, the CépiDc asked for a renewal of the label of quality to the Official Statistics Authority. The process is on going as this text is written (25 June 2025).
12.1. Relevance - User Needs
The main users of COD data are Research institutes, Universities, Public Government agencies. The main request from cause of death data users and stakeholders is the access to more recent data.
12.2. Relevance - User Satisfaction
Not available.
12.3. Completeness
1
12.3.1. Data completeness - rate
1. For mandatory variables: 1
2. For voluntary variables: 1
3. For additional variables:
External CoD : 1
Place of occurrence for external CoD : 1
Activity for external CoD : 0
13.1. Accuracy - overall
The main issue limiting the accuracy of the cause of death statistics is data coverage : each year, 98% of civil state registered deaths in France match a death certificate received at Inserm-CépiDc. For the other 2% medical death certificates are missing. These remaining 2% are identified by the impossibility of indirect record linkage between civil state mortality data (collected by INSEE) and the medical death certificate data. They are at the end added to the CoD data with "unspecified cause of mortality". Missing data is non random and overrepresents suspicious deaths for which the death certificate can be blocked at various stages of the process, as described in Transmission of death certificates to CepiDc-Inserm related to suspicious deaths, in France, since 2000 (doi.org/10.1016/j.respe.2017.11.006), or some specific small geographical areas, which failed to provide data. However, the accuracy is rather good overall, and at the NUTS 2 level.
13.2. Sampling error
Not applicable. Data collection and processing are based on exhaustive administrative data.
13.2.1. Sampling error - indicators
Not applicable.
13.3. Non-sampling error
Some due to under-coverage
In 2022 and 2023, the final stage of the full synchronization of deaths registered by the INSEE and death certificates received by the CépiDc was not performed, as it would lead to undercoverage. Consequentely, there is a small discrepancy between the total deaths in demo_magec (2022: 675 271, 2023: 639 426) and the total number of deaths reported in the data (2022: 675 413, 2023: 639 441).
Title : "Transmission of death certificates to CepiDc-Inserm related to suspicious deaths, in France, since 2000" Creator : Elsa Richaud-Eyraud, Claire Rondet and Grégoire Rey Publisher : Revue d'Épidémiologie et de Santé Publique, Volume 66, Issue 2, March 2018, Pages 125-133 Created : 2017 Language : fr Link : doi.org/10.1016/j.respe.2017.11.006
13.3.1.1. Over-coverage - rate
Duplicates of certificates (which can happen for e-certificates) are removed during the synchronisation process of CoD deaths with deaths registered to INSEE.
Stillbirths and neonatal births do not come from the same source (death certificates vs « programme médicalisé des systèmes d’information ») and those two sources are not linked allowing one to check and correct for double counts. Hence, there is a small risk of overestimating perinatal mortality due to double counting in each source of deaths/stillbirths at day 0.
13.3.1.2. Common units - proportion
Not applicable. Data collection is from administrative sources.
13.3.2. Measurement error
Not applicable.
13.3.3. Non response error
Only partial non-response.
13.3.3.1. Unit non-response - rate
Not applicable, deaths with no death certificates are added to the final data set with unknown cause of death.
13.3.3.2. Item non-response - rate
Total non-response (not receiving causes of death for a person deceased in France) is added to the dataset with COD coded as R99. Information on age, gender and region of death is available. CepiDc-Inserm does not receive around 2% of deaths certificates.
13.3.4. Processing error
In accordance with WHO recommendations, the physician's description of the cause of death is not questioned when compiling the statistical database. Inserm-CépiDc uses an external service provider to receive, digitize and standardize paper death certificates. Standardization also applies to electronic certificates that are not automatically coded with IRIS/MUSE. During this data entry phase, errors are limited by the use of a transcription software. The service provider sends both the scanned images and the digitized text to Inserm-CépiDc. During the coding phase, the IRIS/MUSE coding software is used both for batch coding and as an interface for manual coding (assisted coding). If the text on a death certificate appears to be inconsistent, the coding team checks the data entry using the scanned image.
For deaths in 2018 and 2019, data were produced in a context of production catch-up: certificates that were not automatically coded by the rule-based coding system (IRIS/MUSE, 62%) were either coded by deep learning algorithms trained on 2011-2017 and 2020 data (33%), or coded manually (3%). This may result in some errors for specific categories. Those are detailed in Zambetta et al. (2023) also reported in the appendix of the present metadata page.
For deaths in 2021 on, AI coding is integrated as a third mode of coding in regular production. 62,3% of death certificates in 2023 are automatically batch coded up to the UCOD determination by IRIS/Muse, 20,8% by AI and the remaining by the coding experts (via assisted coding with IRIS/Muse, 14,8%). Manual coding is targeted to certificates the more complex to code. 2% of the certificate in 2023 were missing.
Year
IRIS/Muse
(%)
Assisted expert coding (%)
AI (%)
[Missing certificates]
2018
61.7
3.0
32.8
2.5
2019
62.5
3.1
32.0
2.5
2020
53.7
44.1
2.2
2021
61.4
14.0
22.5
2.0
2022
61.4
12.9
23.7
2.1
2023
62.3
14.8
20.8
2.0
13.3.5. Model assumption error
Not applicable.
14.1. Timeliness
Year
Number of months between the end of the reference year and the publication at national level
2011
32
2012
31
2013
25
2014
25
2015
26
2016
30
2017
47
2018
50 for provisional data +9 for final data
2019
38 for provisional data +9 for final data
2020
24
2021
24
2022
21
2023
19
14.1.1. Time lag - first result
For 2018 and 2019 only – provisional data were disseminated in December 2022.
14.1.2. Time lag - final result
For 2018 and 2019 data only – Final data were sent to Eurostat in September 2023.
14.2. Punctuality
From 2020 onwards, data are delivered on time with respect to the announced calendar.
14.2.1. Punctuality - delivery and publication
Reference year
Time between the end of the reference year (year of death) and the delivery of final data to Eurostat
2011
24+8
2012
24+7
2013
24+1
2014
24+1
2015
24+2
2016
24+6
2017
24+23
2018
24+24 for provisional data; 24+33 for final data
2019
24+12 for provisional data; 24+33 for final data
2020
24+0
2021
24+0
2022
21
2023
18
15.1. Comparability - geographical
The cause of death coding is centralized at Inserm-CépiDc. With the exception of Martinique and French Guiana, each region has a coverage rate of over 95%.
Région
2015
2016
2017
2018
2019
2020
2021
2022
2023
Auvergne-Rhône-Alpes
3,7
2,0
3,4
2,6
2,5
2,2
2,1
1,9
2,0
Bourgogne-Franche-Comté
2,0
4,1
2,2
2,3
2,0
1,3
1,3
3,4
2,5
Bretagne
0,6
0,9
0,9
0,9
0,9
1,0
0,8
0,7
0,8
Centre-Val de Loire
1,4
1,0
1,3
1,4
2,1
1,4
1,3
1,2
1,3
Corse
5,3
4,5
3,2
4,6
5,6
5,1
3,3
3,6
2,5
Grand Est
1,6
1,6
2,0
2,1
1,7
1,7
2,2
2,2
1,3
Guadeloupe
7,3
6,3
14,2
12,1
2,8
3,6
3,3
5,5
4,9
Guyane
7,7
4,2
7,0
19,6
5,9
6,7
7,3
6,8
10,5
Hauts-de-France
1,4
1,6
2,5
2,4
2,5
2,3
2,9
1,6
2,0
Ile-de-France
6,5
5,0
6,0
4,2
4,5
4,4
3,2
3,3
3,0
La Réunion
4,3
0,5
1,7
5,8
2,1
1,5
4,7
8,8
2,2
Martinique
1,7
2,1
1,6
7,6
1,0
3,7
5,1
11,0
8,1
Mayotte
8,3
10,2
19,4
29,3
17,0
16,2
8,7
4,1
3,2
Normandie
1,8
1,8
3,7
2,5
2,1
1,6
1,2
1,3
1,7
Nouvelle-Aquitaine
1,5
1,1
1,1
1,3
1,5
1,4
1,6
1,7
0,9
Occitanie
3,6
3,0
2,2
3,0
2,7
2,3
2,0
1,7
1,8
Pays de la Loire
1,1
1,7
2,3
1,5
3,2
1,1
1,0
1,1
1,7
Provence-Alpes-Côte d'Azur
1,8
6,0
2,7
2,1
2,1
2,4
1,7
1,8
3,4
France
2,7
2,7
2,8
2,5
2,5
2,2
2,0
2,1
2,0
Some regions (Île-de-France) do not send a countable part of their death certificates on a regular basis. This concerns in particular suspicious deaths in the event of a forensic investigation. The total number of deaths for these kinds of causes is therefore subject to variable underestimation over the territory and over time. The medico-legal institute of Paris transmitted medical causes for deaths subject to autopsy occurring from 2018 onwards.
Death certificates can be paper-back or electronic – the rate of electronic death certificates increased in 2020 from 20% at the beginning of the year to 29% in December 2020, around 32% in 2021, 35% in 2022 and 41% in 2023. The certificate model remains consistent, whether is electronically filled or paper, without risk of lack of transmission of data to CepiDc-Inserm for e-certification.
Since 2018, a new model of certificate was introduced. This new model of death certificate allows certifiers to provide additional information on manner of death especially for external causes of death, which enables to identify more accurately suicides for instance. E-certification was adapted to this new model in January 2018. Its use has been more progressive for paper-back certificates. The old model continues to be used especially by certifiers who do not certify deaths very often. The rates of dissemination can differ on a geographical basis.
15.1.1. Asymmetry for mirror flow statistics - coefficient
Not applicable.
15.2. Comparability - over time
The 10th revision of the International statistical classification of diseases and related health problems (ICD-10) has been implemented by Inserm-CépiDc from reference year 2000 onwards for cause of death statistics production. Except for very specific causes of death, there has been no major changes enough to warrant the designation of a break in series since.
Breaks in time comparability include e-certification deployment since 2007, 2017 new model of death certificate (at stake in 2018), and coverage of causes of deaths when forensic investigations occur as detailed above.
On electronic death certification, introduced in 2007,
title : "Evolution of the electronic death certification in France from 2011 to 2018" creator : Anne Fouillet, Dominique Pigeon, Isabelle Carton, Aude Robert, Isabelle Pontais, Céline Caserio-Schönemann and Grégoire Rey publisher : Bulletin épidémiologique hebdomadaire N° 29-30 - 12 novembre 2019 created : 2019 language : fr link : beh.santepubliquefrance.fr/beh/2019/29-30/pdf/2019_29-30_2.pdf
15.2.1. Length of comparable time series
From 2011 on.
15.3. Coherence - cross domain
Checks are done with death records (INSEE) every week, with indirect record linkage every month and at the end of the coding of a given year, and final data includes demographic information also for missing death certificates
For 2020 data due to the pandemic, checks were done comparing CoD death certificates with mention of COVID on information feedback from the information systems of hospitals and medico-social local units. COVID dead countings differentials were within acceptable range (less than 15% of differential), see
Fouillet A, Ghosn W, Naouri D, Coudin E. Covid-19 : troisième cause de décès en France en 2020, quand les autres grandes causes baissent. Bull Épidémiol Hebd. 022;(Cov_16):2-15.
Naouri, D., Fouillet, A., Ghosn, W., Coudin, E. (2022, décembre). Covid-19 : troisième cause de décès en France en 2020, quand les autres grandes causes de décès baissent. DREES, Études et Résultats, 1250.
15.3.1. Coherence - sub annual and annual statistics
Data are interpretable by month of death.
15.3.2. Coherence - National Accounts
Not applicable.
15.4. Coherence - internal
The coding team follows the same rules to code the causes of death.
Incoherence between cause of death and age or sex are identified and corrected.
Cause of death statistics production costs approximately 2 million euros annually. This bugdet does not cover the cost of the data collection process and its transmission to Inserm-CépiDc.
The cost and burden of the data collection is reduced by using automatic coding systems and by the increasing coverage of electronic death certification.
17.1. Data revision - policy
There is no data revision and no data revision policy. 2018 and 2019 are exceptions.
17.2. Data revision - practice
Not applicable.
17.2.1. Data revision - average size
Not applicable.
18.1. Source data
French death certificates comply with the WHO standards. There are two different templates the neonatal death certificate for deaths occuring on the day of birth up to the 27th day of life and the general (adult) model of death certificate for deaths occuring from the 28th day of life (included) onwards. From 2018 on (and slightly 2017) a new templace of the adult death certificate has been disseminated (but has not totally replaced the old ones). The new template offers more details on death circumstances, which has an impact in evaluating external causes of death.
18.2. Frequency of data collection
Daily.
18.3. Data collection
Data from administrative sources, collected either electronically or on paper.
18.3.1. Certification
Regarding training, not applicable because there is no survey conducted on death certification training or post mortem examination.
Regarding queries related to medical part of the death certificate, approximately 2000 to 3000 problematic death certificates have been queried for each reference year, with an estimated 50% response rate. For reference year 2017, only death certificates presenting a blank medical part have been queried. The proportion of deaths where the underlying cause is changed as a result of the query is unknown.
Table on certification (Percentage)
Years
All doctors (certifiers) trained in the certification
All doctors (certifiers - pathologists or others doctors) trained in the post-mortem examination (autopsies)
Certificates filled by persons who attended a course on certification or post-mortem examination
Death certificates that are queried (only queries related to medical part of the death certificate should be included)
Replies received for queries sent
Deaths where the underlying cause is changed as a result of the query
Death certificates with incorrect sequence
2011 to 2023
Not applicable
Not applicable
Not applicable
Not available
Not available
Not available
Not available
18.3.2. Automated Coding
Data year
Use of any form of automated coding
System used (IRIS, MICAR, ACME, STYX, MIKADO, others)
2011
Yes
Partly Styx, Iris, MICAR, ACME tables-Y2011S2, partly manual, with systematic review of deaths
2012
Yes
Partly Iris 4.0.52, MICAR, ACME tables-Y2012S1, partly manual, with systematic review of deaths
2013
Yes
Partly Iris 4.3.0, MICAR, ACME tables-Y2013S1, partly manual, with systematic review of deaths
Final data : Partly Iris 5.5.0, Muse 2.6 (specV2018SR10), partly with deep learning, partly manual
2019
Yes
Final data : Partly Iris, 5.6.0, Muse 2.7.1 (specV2019SR10), partly with deep learning, partly manual
2020
Yes
Partly Iris 5.7.0, Muse 2.8 and partly manual
2021
Yes
Partly Iris 5.8.1, Muse 2.9, partly with deep learning, partly manual
2022
Yes
Partly Iris 5.8.1, Muse 2.9, partly with deep learning, partly manual
2023
Yes
Partly Iris 5.8.1, Muse 2.9, partly with deep learning, partly manual
18.3.3. Underlying cause of death
Data year
Only manual selection of underlying cause
Manual with ACME decision tables (if yes, version of ACME)
ACS utilising ACME decision tables (if yes, version of ACME)
Own system (ACS without ACME)
Comments
2011
No
No
Yes (ACME tables-Y2011S2)
No
Styx, Iris
2012
No
No
Yes (ACME tables-Y2012S1)
No
Iris 4.0.52
2013
No
No
Yes (ACME tables-Y2013S1)
No
Iris 4.3.0
2014
No
No
Yes (ACME tables-Y2014S2)
No
Iris 4.3.0
2015
No
No
Yes (ACME tables-Y2015S1)
No
Iris 4.5.6
2016
No
No
No
Yes (Muse specV2016SR10)
Iris 5.4.0
2017
No
No
No
Yes (Muse specV2018SR10)
Iris 5.5.4, no change of code from WHO so the most up-to-date tables at the time of the coding have been used (2018 instead of 2017)
2018
No
No
No
Yes (Muse specV2018SR10 and deep learning)
Iris 5.5.0, Muse 2.6
2019
No
No
No
Yes (Muse specV2019SR10 and deep learning)
Iris 5.6.0, Muse 2.7.1
2020
No
No
No
Yes (Muse)
Iris 5.7.0, Muse 2.8
2021
No
No
No
Yes (Muse specV2021SR30 and deep learning)
Iris 5.8.1, Muse 2.9
2022
No
No
No
Yes (Muse specV2021SR30 and deep learning)
Iris 5.8.1, Muse 2.9
2023
No
No
No
Yes (Muse specV2021SR30 and deep learning)
Iris 5.8.1, Muse 2.9
18.3.4. Availability of multiple cause
Data year
Information stored in the national CoD database, UC (Underlying cause) or MC (Multiple cause)
2011 to 2023
UC + MC
18.3.5. Stillbirths and Neonatal certificates
French death certificate does not apply to stillbirths. Stillbirths information is collected from the « programme médicalisé des systèmes d’information » (PMSI) by the Directorate of Research, Study, Evaluation and Statistics (DREES) Population health office depending on the Health Ministry.
All stillbirths with weight births over 500 grams or with gestational ages of at least 22 weeks are collected. The terminations of pregnancy are included from 22 weeks (gestational age) onwards.
18.4. Data validation
Inconsistencies between the cause of death and other information on the death certificate (age, sex, manner of death) are detected with alerts during coding using the Iris software, so coders can check the original death certificate and correct it. Consistency checks are also run on specific categories (children under 15, pathologies of particular interest for public health, maternal deaths, external causes) and randomly. Coverage validation is done by indirect record linkage with population registers (Insee).
18.4.1. Coding
Description of coding procedure (central level, distributed among other bodies, etc.) :
The production process is centralised at Inserm-CépiDc. The coding procedure relies on the automated coding system Iris. 58 % of 2017 death certificates and 63,6% for 2023 were coded automatically by Iris. Rejected certificates are either manually revised by trained coders, and experts if they remain not yet coded, either by coded by AI (for 2018, 2019 and from 2021 on).
For deaths in 2018 and 2019, coding was also performed by predictive deep learning algorithms trained on past data. From 2021 on, coding strategy relies on the three modes of coding - automatically by the rule-based system IRIS/MUSE (62,3% in 2023), manually and assisted by IRIS/MUSE (14,8%) and predictive deep learning (20,8%). Deep learning algorithms are used both to predict CoD (multiple causes and UC) and to target certificates with low confidence levels in predictions to send them to manual coding. Manual coding therefore focuses on deaths of special interest for public health and research (maternal deaths, children, AIDS/HIV, research databases) plus those for which AI algorithms performance is not so good (other external causes, blood diseases, musculoskeletal system diseases, accidental poisoning...)
See Zambetta et al (2023) for details also reported as appendix to this metadata web page, with applications/quantification for 2018 and 2019 data.
Description of the procedures to detect errors (i.e. errors such as potential inconsistency in the death certificate or error due to mistake when filling the deaths certificates) :
Inconsistencies between the cause of death and other information on the death certificate (age, sex, manner of death) are detected with alerts during coding using the IRIS/MUSE software, so the coders can check the original death certificate and correct it. A consistency check is also run for the detection of inconsistencies at the end of a given sample coding.
Checks on cause of death coding quality are performed regularly. The manual inconsistency checks performed via the IRIS/MUSE coding software user interface concerns:
inconsistencies between age or sex and cause of death
applying explicit checks on certain causes of death (violent deaths, maternal deaths, neonatal deaths, COVID-19, AIDS...)
addressing automated coding accuracy
Random quality checks are also considered to assess quality of automated coding quality and consistency of manual coding.
Description of the measures taken in order to solve detected errors:
Errors detected are corrected manually by expert coders using the Iris software.
In the case of death certificates presenting a blank medical part, an e-mail or a paper mail may be sent to the certifier, but most of these queries remain unanswered.
Coding performed by a certifier:
no
Estimation of the percentage of autopsy from which information is available for coding:
Unknown.
Description of double coding exercises and rate of codification errors for underlying cause of death:
Unknown
18.4.2. Unspecified CoD code
ICD codes for the underlying cause (% of the Total)
Deaths due to senility (for ICD10 : R54 code, for ICD9 : 797 code)
Deaths due to exposure to unspecified factor (for ICD10 : X59 code, ICD9 : 928.9 code)
2011
8.29
3.59
0.76
1.41
2012
9.41
4.47
0.71
1.47
2013
10.14
5.28
0.69
1.44
2014
9.05
4.06
0.68
1.50
2015
9.25
4.02
0.71
1.63
2016
9.35
4.27
0.64
1.61
2017
9.86
4.60
0.65
1.63
2018
10.22
4.69
0.64
1.51
2019
11.01
5.34
0.69
1.57
2020
10.15
4.90
0.67
1.48
2021
10.46
4.85
0.73
1.53
2022
11.22
5.12
0.83
1.65
2023
10.98
4.90
0.86
1.70
18.4.3. Unknown country or region
Unknown country/region (%) for residents and non-residents who died in the country
Year
Residents
Non-residents
Unknown residency (NUTS2)
Unknown occurrence (NUTS2)
Unknown residency (country)
Unknown residency (NUTS2)
Unknown occurrence (NUTS2)
2011
0
0
13.05
100
10.62
2012
0
0
24.04
100
0
2013
0
0
26.38
100
0
2014
0
0
28.38
100
0
2015
0
0
0
100
0
2016
0
0
12.35
100
9.47
2017
0
0
6.66
98.98
3.97
2018
0
0
3.0
100
0
2019
0
0
1.8
100
0
2020
0
0
3.4
100
0
2021
0
0
2.1
97.8
0
2022
0
0
0
100
0
2023
0
0
0
98.5
0
Before 2023, deaths certificates received by France from overseas collectivities, which are not part of French NUTS except Saint-Martin (Saint-Barthélémy, Wallis-and-Futuna, French Polynesia, Saint-Pierre-and-Miquelon, French Southern and Antarctic Lands and the Clipperton Island) were encoded as "UN99". Starting from 2023, they are encoded as any other foreign countries (ISO2-code + 99).
18.4.4. Validation of the coverage
By indirect record linkage with the population registers.
18.5. Data compilation
No operation is performed.
18.5.1. Imputation - rate
Not applicable, no imputation is made. For deep learning coding of provisional data in 2018 and 2019 see dedicated report.
Data on causes of death (CoD) provide information on mortality patterns and form a major element of public health information.
CoD data refer to the underlying cause of death (UCD) which is defined by the World Health Organisation (WHO) as "the disease or injury which initiated the train of morbid events leading directly to death, or the circumstances of the accident or violence which produced the fatal injury".
The statistical units are deceased persons and stillborns.
Residents and non-residents who died in France NUTS FR (see 3.7 for geographical scope).
All French NUTS2, including Metropolitan France and overseas departments and regions (la Guadeloupe, la Martinique, la Guyane, la Réunion, Mayotte) + Saint-Martin, according to the French geographical nomenclature as of January 1st of the year of death established by the French National Institute of Statistics (Insee) : https://www.insee.fr/fr/information/6051727
Deaths occuring in other overseas areas (i.e. overseas collectivities) or abroad are not included, except for Saint-Martin, which is included in NUTS 2 (FRY1)
Data refer to the calendar year (i.e. all deaths occurring during the year).
The main issue limiting the accuracy of the cause of death statistics is data coverage : each year, 98% of civil state registered deaths in France match a death certificate received at Inserm-CépiDc. For the other 2% medical death certificates are missing. These remaining 2% are identified by the impossibility of indirect record linkage between civil state mortality data (collected by INSEE) and the medical death certificate data. They are at the end added to the CoD data with "unspecified cause of mortality". Missing data is non random and overrepresents suspicious deaths for which the death certificate can be blocked at various stages of the process, as described in Transmission of death certificates to CepiDc-Inserm related to suspicious deaths, in France, since 2000 (doi.org/10.1016/j.respe.2017.11.006), or some specific small geographical areas, which failed to provide data. However, the accuracy is rather good overall, and at the NUTS 2 level.
The unit is number.
No operation is performed.
French death certificates comply with the WHO standards. There are two different templates the neonatal death certificate for deaths occuring on the day of birth up to the 27th day of life and the general (adult) model of death certificate for deaths occuring from the 28th day of life (included) onwards. From 2018 on (and slightly 2017) a new templace of the adult death certificate has been disseminated (but has not totally replaced the old ones). The new template offers more details on death circumstances, which has an impact in evaluating external causes of death.
Annual.
Year
Number of months between the end of the reference year and the publication at national level
2011
32
2012
31
2013
25
2014
25
2015
26
2016
30
2017
47
2018
50 for provisional data +9 for final data
2019
38 for provisional data +9 for final data
2020
24
2021
24
2022
21
2023
19
The cause of death coding is centralized at Inserm-CépiDc. With the exception of Martinique and French Guiana, each region has a coverage rate of over 95%.
Région
2015
2016
2017
2018
2019
2020
2021
2022
2023
Auvergne-Rhône-Alpes
3,7
2,0
3,4
2,6
2,5
2,2
2,1
1,9
2,0
Bourgogne-Franche-Comté
2,0
4,1
2,2
2,3
2,0
1,3
1,3
3,4
2,5
Bretagne
0,6
0,9
0,9
0,9
0,9
1,0
0,8
0,7
0,8
Centre-Val de Loire
1,4
1,0
1,3
1,4
2,1
1,4
1,3
1,2
1,3
Corse
5,3
4,5
3,2
4,6
5,6
5,1
3,3
3,6
2,5
Grand Est
1,6
1,6
2,0
2,1
1,7
1,7
2,2
2,2
1,3
Guadeloupe
7,3
6,3
14,2
12,1
2,8
3,6
3,3
5,5
4,9
Guyane
7,7
4,2
7,0
19,6
5,9
6,7
7,3
6,8
10,5
Hauts-de-France
1,4
1,6
2,5
2,4
2,5
2,3
2,9
1,6
2,0
Ile-de-France
6,5
5,0
6,0
4,2
4,5
4,4
3,2
3,3
3,0
La Réunion
4,3
0,5
1,7
5,8
2,1
1,5
4,7
8,8
2,2
Martinique
1,7
2,1
1,6
7,6
1,0
3,7
5,1
11,0
8,1
Mayotte
8,3
10,2
19,4
29,3
17,0
16,2
8,7
4,1
3,2
Normandie
1,8
1,8
3,7
2,5
2,1
1,6
1,2
1,3
1,7
Nouvelle-Aquitaine
1,5
1,1
1,1
1,3
1,5
1,4
1,6
1,7
0,9
Occitanie
3,6
3,0
2,2
3,0
2,7
2,3
2,0
1,7
1,8
Pays de la Loire
1,1
1,7
2,3
1,5
3,2
1,1
1,0
1,1
1,7
Provence-Alpes-Côte d'Azur
1,8
6,0
2,7
2,1
2,1
2,4
1,7
1,8
3,4
France
2,7
2,7
2,8
2,5
2,5
2,2
2,0
2,1
2,0
Some regions (Île-de-France) do not send a countable part of their death certificates on a regular basis. This concerns in particular suspicious deaths in the event of a forensic investigation. The total number of deaths for these kinds of causes is therefore subject to variable underestimation over the territory and over time. The medico-legal institute of Paris transmitted medical causes for deaths subject to autopsy occurring from 2018 onwards.
Death certificates can be paper-back or electronic – the rate of electronic death certificates increased in 2020 from 20% at the beginning of the year to 29% in December 2020, around 32% in 2021, 35% in 2022 and 41% in 2023. The certificate model remains consistent, whether is electronically filled or paper, without risk of lack of transmission of data to CepiDc-Inserm for e-certification.
Since 2018, a new model of certificate was introduced. This new model of death certificate allows certifiers to provide additional information on manner of death especially for external causes of death, which enables to identify more accurately suicides for instance. E-certification was adapted to this new model in January 2018. Its use has been more progressive for paper-back certificates. The old model continues to be used especially by certifiers who do not certify deaths very often. The rates of dissemination can differ on a geographical basis.
The 10th revision of the International statistical classification of diseases and related health problems (ICD-10) has been implemented by Inserm-CépiDc from reference year 2000 onwards for cause of death statistics production. Except for very specific causes of death, there has been no major changes enough to warrant the designation of a break in series since.
Breaks in time comparability include e-certification deployment since 2007, 2017 new model of death certificate (at stake in 2018), and coverage of causes of deaths when forensic investigations occur as detailed above.
On electronic death certification, introduced in 2007,
title : "Evolution of the electronic death certification in France from 2011 to 2018" creator : Anne Fouillet, Dominique Pigeon, Isabelle Carton, Aude Robert, Isabelle Pontais, Céline Caserio-Schönemann and Grégoire Rey publisher : Bulletin épidémiologique hebdomadaire N° 29-30 - 12 novembre 2019 created : 2019 language : fr link : beh.santepubliquefrance.fr/beh/2019/29-30/pdf/2019_29-30_2.pdf
