CODING OF CAUSES OF DEATH IN EUROPEAN COMMUNITY

Transcription

1 EUROPEAN COMMUNITY - EUROSTAT CODING OF CAUSES OF DEATH IN EUROPEAN COMMUNITY Project 96/S /EN - Lot 11 FINAL REPORT June 1998 Contractors Gérard Pavillon - INSERM, France Michel Coleman - ONS, UK Lars Age Johansson - Statistics Sweden Eric Jougla - INSERM, France Jan Kardaun - Statistics Netherlands 1

2 2

3 Acknowledgements Many thanks to all the statistical offices for answering the questionnaire. The valuable data collected in this way are the basis of this report. 3

4 INDEX INDEX...4 I. INTRODUCTION...6 I.1 OBJECTIVE...6 I.2 SCIENTIFIC BACKGROUND...7 I.3 DEATH CERTIFICATE AND CODING RULES...9 II. FUNCTIONS OF AN AUTOMATED CODING SYSTEM...12 III. DESCRIPTION OF THE US AUTOMATED CODING SYSTEM...16 III.1 ACME...18 III.2 TRANSAX...19 III.3 MICAR...19 III.4 SUPERMICAR...20 III.5 CONCLUSION...22 IV. CAUSE-OF-DEATH PROCEDURES IN EUROPEAN COMMUNITY...23 IV.1 DEATH CERTIFICATE...25 IV.2. CERTIFICATION...30 IV.3 CODING...38 IV.4 AUTOMATIC CODING SYSTEM (ACS)...45 IV.5 INTERNATIONAL COOPERATION ON CODING PROBLEMS...49 IV.6 MISCELLANEOUS...51 V. INTERVIEWS OF COUNTRIES USING ACS (SUMMARIES)...53 V.1 CATALONIA...53 V.2 ENGLAND AND WALES...55 V.3 ITALY...58 V.4 SCOTLAND

5 V.5 SWEDEN...62 V.6 THE NETHERLANDS...64 VI. RECOMMENDATIONS AND GUIDELINES FOR ACS DESIGN AND USE...66 VI.1 IMPLEMENTATION...66 VI.2 ROUTINE PROCESSING...69 VI.3 COMPARABILITY...70 VII. CONCLUSION...73 BIBLIOGRAPHY...74 ANNEX A - ICD10 RULES...76 ANNEX B - QUESTIONNAIRE...78 ANNEX C - INTERVIEWS OF COUNTRIES USING AN ACS...95 C1. CATALONIA...96 C2. ENGLAND AND WALES, SCOTLAND C3. ITALY C4. SWEDEN C5. THE NETHERLANDS C6. QUESTIONS FOR COUNTRIES USING AN AUTOMATED CAUSE CODING SYSTEM ANNEX D - DETAILLED QUESTIONNAIRE ANALYSIS

6 I. INTRODUCTION The background to this project is the need to improve comparability of cause-of-death statistics in Europe. An important aspect of this issue is the quality and homogeneity of coding procedures. One way of improving these procedures may be the adoption of Automated Coding Systems for medical cause-of-death data (ACS). This project has to be placed in a larger international context since most of the ACS, existing or in development, are based on the ACS designed by the United States. This section presents the objectives of the project and its background. It also introduces the basic document and rules for cause-of-death coding as they are stated by the World Healtrh Organisation (WHO). I.1 Objective The main objective of the project is to develop recommendations and guidelines that could be used for the application of Automated Coding Systems for causes of death (ACS) in order to achieve more comparable statistics at the European level. These recommendations and guidelines are based on the examination of existing ACS and on the analysis of requirements where ACS have not been implemented for technical or other reasons. Issues related to the use of ACS include staff problems, loss of expertise, change in trends. This study focuses on the coding process of the medical causes of death reported on the death certificate. It does not consider other aspects, which are not related to coding, that could affect the international comparability (in particular the certification process). In order to fulfil this objective, four aspects have been addressed. The functions an ACS should carry out have been reviewed, and the issues identified (section II). The system designed by the United States, the first ACS, has been studied in detail (section III). This ACS, developed in 1967, represents a lot of experience in the field of automated coding and most of the ACS presently in use are derived from or are dependent on the ACME system. 6

7 Cause-of-death coding procedures in use among Member States (MS) have been studied through a questionnaire (section IV). Information has been collected on certification practices, death certificate forms, coding procedures and ACS in use or planned. Questions also covered the important issue of international cooperation on coding problems. Countries among Member States using an ACS were interviewed with a specific grid of questions in order to obtain more information on their system design, implementation and use (section V). Section VI presents the guidelines and recommendations for ACS design and use and section VII gives the general conclusion of this study. I.2 Scientific background Four main aspects justify studies on the quality of cause-of-death statistics. European cause-of-death statistics Public health planners and epidemiologists need reliable health indicators in order to define health priorities, to allocate resources and to assess the quality of health care. Cause-of-death statistics are widely used for this purpose and are often the only data available for health status comparisons between European countries. International cause-of-death data are published annually by the WHO using a uniform list of 102 categories. Eurostat plans to publish these data at a national and regional level, according to a condensed list of 65 categories more attuned to the health needs of European countries. Comparability of cause-of-death statistics Procedures for the collection of data on cause-of-death are relatively homogeneous between countries (death certificate form, International Classification of Diseases), but in spite of these common features, quality and comparability problems persist. A number of scientific studies have assessed the general quality of national cause-of-death statistics within various countries. In only a few cases have studies focused on international comparability, and investigated possible national bias in certification of death or coding of causes of death [Jougla 97]. Some of these studies outlined comparability problems exclusively linked to differences in the 7

8 coding processes [Percy78] [Kelson83] [Kelson87] [Percy89] [Jougla92] [Balkau93]. For example, a study concerning diabetes, based on samples of death certificates from nine European countries, showed wide differences between the original national coding and a central reference recoding (discrepancies in the underlying cause of death existed for 26% of all death certificates at the ICD 3 digit coding level ). A preliminary study based on an inquiry into the main European coding services pointed to a number of general differences in the procedures. It ended with a recommendation to set up a formal European Community (EC) Working Group, to which should be assigned the task of proposing methods of standardisation [Lagasse90] [Holland91]. Of all the general problems, the coding procedures should be the simplest to address in the framework of a co-operative project at the European level. An agreement on coding procedures may allow the main coding problems to be solved. The adoption of automatic coding should greatly facilitate this task. Implementation of the 10th revision of ICD Another important aspect of the project is the introduction of the 10th revision of the International Classification of Diseases (ICD10) [ICD92]. This classification offers about 12,000 4-character categories while the previous one (ICD9) has only about 6,000 [ICD77]. This dramatic increase in the number of categories is a strong reason for introducing the automated coding, since the complexity of ICD10 cannot be easily managed with manual coding. Multiple-cause coding Multiple-cause coding consists of taking into account all the causes of death mentioned on the certificate in parts I and II. An increasing number of scientific recommendations call for mortality data to be collected and analysed in terms of multiple cause of death, i.e. for the concept of a single cause of death to be replaced by that of a set of causes resulting in death. This need is becoming more urgent as chronic diseases become more widespread, and for ageing populations several causes can lead to death and it is difficult to choose one of these as the underlying cause. The coding of multiple cause using ICD10 is another powerful reason to adopt automated coding systems [WHO70] [Manton84] [Israel86] [Pavillon94] [Pavillon97]. 8

9 I.3 Death certificate and coding rules The underlying cause of death is defined by WHO as "a) the disease or injury which initiated the train of morbid events leading directly to death, or b) the circumstances of the accident or violence which produces the fatal injury" [ICD92]. In order to produce comparable cause-of-death statistics, WHO recommends the use of the international form of medical certificate of cause of death and the ICD rules to select the underlying cause. The international form of medical certificate of cause of death [ICD92] is shown below. CAUSE OF DEATH Approximate Part I interval between onset and death Disease or condition directly a) leading to death* due to (or as a consequence of) Antecedent causes b) Morbid conditions if any giving rise to the above due to (or as a consequence of) cause, stating the underlying condition last c) due to (or as a consequence of) d) Part II Other significant conditions... contributing to the death, but not related to the disease or condition causing it... * This does not mean the mode of dying, e.g. heart failure, respiratory failure. It mean the disease, injury, or complication that causes death. This death certificate includes two parts. Part I is used to give the morbid process directly leading to death. The underlying cause should be stated on the lowest used line of this part. Part II is used to mention causes contributing to the death but not related to the underlying condition. 9

10 Here are 2 examples of death certificates properly completed: Example 1 I a) Bronchopneumonia b) Chronic bronchitis II Chronic myocarditis Example 2 I a) Traumatic shock b) Multiple fractures c) Pedestrian hit by truck (traffic accident) II Chronic myocarditis The ICD rules (c.f. Annex A) give a uniform procedure to select and code the "best" underlying cause from the causes of death mentioned on the death certificate. Firstly, the General Principle applies when a death certficate is properly completed. When this principle does not apply, "...clarification of the death certificate should be sought from the certifier whenever possible" ([ICD92], Vol 2, p25). Secondly, when the General Principle cannot be applied, the selection rules help the coders with death certificates that are not properly completed. For instance, these rules are applied when more than one cause of death are mentioned on the lowest line in part I, or when the part I is filled in upside down. Secondly, the modification rules aim at selecting and coding the more informative underlying cause of death. Ill-defined causes (such as senility) can be discarded when other causes are mentioned (Rule A). They also merge, where possible, several causes into a unique code. For instance, if a "femoral hernia" gives rise to an "intestinal obstruction" in part I, the selection Rule C will lead to code "femoral hernia with obstruction" as the underlying cause. Other selection rules cope with the trivial conditions, specificity of causes, early and late stage of diseases and sequelae. Additional information and guidelines are also provided in the ICD. They help the coder to select the underlying cause on the basis of what the certifier reported on the medical cause of death certificate. The procedures defined by WHO to code causes of death are a valuable aid in producing comparable cause-of-death statistics. However these procedures are quite difficult to apply manually and they leave room for interpretation. 10

11 11

12 II. FUNCTIONS OF AN AUTOMATED CODING SYSTEM Coding of causes of death is a complex activity that is not limited to the assignment of a code. Coding of causes of death is in fact composed of four functions of which only the first is, properly speaking, a coding function. These functions will be referred to as coding, editing, selection of the underlying cause of death and classification of multiple cause of death. When carried out by a human coder, these functions are not always differentiated; but automatic coding systems (ACS) require each function to be defined precisely. Coding is the assignment of an ICD code to each condition, disease or external cause, reported on the death certificate. Coding instructions are given in the ICD volumes through the index, provisions, notes and exclusions. The ICD code assigned can depend on the deceased's sex and age. Pregnancy can also be taken into account in the assignment of an ICD code. The automation of cause-of-death coding is made difficult because of two aspects of the certification. Firstly, certifiers may report causes in a "narrative style", for instance "death of an old man probably due to an infarction" instead of "myocardial infarction" as the underlying cause. Secondly, several causes can be reported on the same line and must be differentiated. Sometimes there is a causal relationship between these causes and they should have been reported on two different lines. For instance "ischaemic heart disease due to hypertension" on the first line in part one of the certificate should have been reported as "ischaemic heart disease" on the first line and "hypertension" on the second line. Furthermore, the coding function strongly depends on the certification language and it would be quite difficult to design a general program for cause-ofdeath coding. Another complication of the coding process is that the certificate is often completed in a jargon with many abbreviations, spelling variations, and non-standard expressions. 12

13 Editing consists of testing the validity or consistency 1 of the assigned codes. A code is valid if belongs to the list of ICD codes and if it is compatible with the deceased's sex and age. The extent of editing depends on how coding has been done. If manual coding is used, code validity should be carefully checked; but if coding is automated, codes are supposed to be valid. ICD gives a list of codes limited to one sex and another list of codes that cannot be used as underlying cause of death. Additional constraints can be defined in order to ensure data validity. For instance, it is unlikely that young children commit suicide and a check should be made between suicide ICD codes and age. In this case, the problem is defining an age limit (10? 12?). Other diseases such as senility, Alzheimer diseases are much more frequent among old people and lower age limits should also be stated in these cases. Additional constraints can be put upon the assigned codes to increase the level of validity. Examples are rare diseases (e.g. typhus, rabies) according to the sex, age or other diseases of the deceased (e.g. cardiac arrest for young people, Kaposi sarcoma for a male without mention of AIDS). In these cases, it is particularly important, whenever possible, to query the certifier in order to complete or to confirm the causes of death. Selection of the underlying cause is of crucial importance for national and international quality of cause-of-death statistics since most of the statistics are published on the basis of a single cause. Although the selection of the underlying cause is governed by ICD rules, the interpretation and application of these rules leave a substantial degree of flexibility. Firstly, there is room for interpretation in the text of ICD rules and guidelines. Secondly, rules often need additional information to be applied. For instance, the General Principle requires the definition of causal relationships between two diseases, i.e. whether a disease A can cause a disease B. As another example, to apply Rule 3 it is necessary to know what causal sequences are so certain that they should be assumed even when the certifier has not written them as a sequence. The ICD provides some of the information needed to apply each of the rules, but its completeness varies widely. For instance, in the following certificate: 1 The term validity, widely employed, will be used as a synonym of the term consistency, used in the dataprocessing domain [Gardarin89]. 13

14 I II a) Septicaemia b) Acute renal failure Malignant neoplasm of prostate the General Principle will select the "acute renal failure" as the underlying cause. However, in many countries, coders would consider that the "acute renal failure" is a direct consequence of the " malignant neoplasm of prostate" and select it as the underlying cause by applying Rule 3, but some would not. The differences in interpretation reduce the comparability of statistics. The following table gives, for each coding rule, the relationship needed and the completeness of its definition in the ICD10. Coding rule Relationship Definition in ICD10 General principle Acceptable causal relationship incomplete list Rule 1 Acceptable causal relationship incomplete list Rule 2 none Rule 3 Direct consequence incomplete list Rule A Ill-defined condition complete list Rule B Trivial condition some examples Rule C Linkage complete list Rule D Specificity some examples Rule E Stage of diseases some examples Rule F Sequelae complete list Rule 2 does not use any relationship. Rules A, C and F use completely defined relationships. Relationships needed in the General Principle, Rules 1, 3, B, D and E are only partially defined. 14

15 About Rule C, it must be noted that the needed relationships are scattered in volume I, II and III and are sometimes difficult to find. Moreover Rule C is difficult to apply, especially in case of multiple linkage, and the absence of list specifying conveniently all linkages leads frequently to errors. Classification of multiple cause of death is the correct coding of causes other than the underlying cause. Coding all mentioned causes has been recommended by WHO for many years, reflecting research which has shown the analytical potential, the uses and significance of multiple cause tabulations for mortality statistics [Moriyama56][Cornfield57][Dorn64] [Israel73][Chamblee82][Goodman82][Manton84][Israel86]. Multiple cause data can be used to facilitate analyses and comparisons between countries (in particular, in case of chronic diseases) and over time, as well as to examine and adjust for artefacts in selection of the underlying cause. However, unlike underlying cause of death classification, the principles and procedures for classification of multiple cause of death are not defined in the ICD volumes. This task includes eliminating or combining codes so as to standardise the coding in order to facilitate the production of useful and comparable statistics. None of the functions of an ACS are thoroughly defined in ICD, or elsewhere. The coding stage is the best documented function. However, because coding is language dependent, this entails some differences in the codes selected. There is little documentation on the editing function though its definition is not difficult. The main issues of concern are: the selection of the underlying cause and the classification of multiple cause. The selection of the underlying cause is the most important function since most published mortality statistics are based on the underlying cause. Recommendations on the use of ACS especially need to cover these two key functions. 15

16 III. DESCRIPTION OF THE US AUTOMATED CODING SYSTEM The US automated coding system was the first automated coding system used in mortality. It was developed in 1967 by the National Centre for Health Statistics (NCHS) and has been used for 30 years in the United States and in other countries. This system is based on invaluable experience of automated coding, accumulated through years of fine tuning. The full US system comprises four components: SUPERMICAR translates the full text of the conditions reported on the death certificate into standardised expressions MICAR translates standardised expressions into ICD codes and performs editing ACME selects the underlying cause TRANSAX produces the codes for multiple cause analyses The functional sequence of these components is represented in the following figure. 16

17 Causes reported on the death certificate SUPERMICAR Standardized expressions MICAR ICD codes ACME TRANSAX Underlying cause of death Multiple cause of death All the four components can be used separately providing that the input data are correctly formatted. In the US, ACME is always used, but depending on the American State, MICAR is used to code 72% of the records and SUPERMICAR to code 16% (1995 mortality file). The four components are detailed in the following sections. These functions were designed in the following order: ACME, TRANSAX, MICAR and SUPERMICAR. The presentation follows this chronological order in order to show the reasons and issues which led to the design of the whole system. 17

18 III.1 ACME The first experience of automated coding of mortality data which was reported on, was made by the NCHS of the United States in 1967 [Chamblee79]. This covered a feasibility research project on the automated selection of the underlying cause. The hypothesis tested was that the underlying cause of death could be automatically selected for the majority of death records through the application of only four ICD rules. These rules were the General Rule, Rules 1 and 2 and a part of Rule 7 (Linkage) as they were stated in ICD8 [ICD68]. A set of 15,000 death certificates was manually coded to the seventh revision of the International Classification of diseases (ICD7). For each certificate, the underlying cause was then selected by a program implementing the 4 rules and the cause chosen was compared to the underlying cause manually selected. The study revealed an 83% level of agreement between the two methods. This level was considered sufficient to show that the design of a comprehensive automated system was feasible. Next, the NCHS decided that the automated selection of the underlying cause needed to be demonstrated with respect to accuracy and reliability, before a final decision on its adoption could be made. A new study was undertaken in January 1968 [NCHS69a]. The new program, identified as ACME (Automated Classification of Medical Entities), carried out most of the selection and modification rules of the ICD8. The system was then designed to either select the underlying cause of death or to recognise situations where the record must be rejected for a manual processing. The relationships needed for the application of rules (c.f. section II) were stored on the computer in the form of "decision tables". For example, to apply the General Principle, the computer needed to know whether the disease X can or cannot cause the disease Y. This information was stored in a table mentioning all the possible pairings of diseases A and B for which A can cause B. ACME comprises 6 decision tables including the table of valid codes, causal relationships (General Principle and Rule 1), direct sequel (Rule 3) and 3 other tables needed by modification rules. These tables were built incrementally from the actual data processed. A new version of ACME was assessed using a sample of 18,539 death certificates comprising actual data from January This test showed a 96% level of agreement between the manual and ACME selection of the underlying cause of death. Of the 4% disagreements, 1.5% was death 18

19 certificates intentionally rejected by ACME and 2.5% was due to manual coding errors[nchs69b] [NCHS69c]. On the US mortality files, ACME processes 98.5% of the records with an error rate of 0.5%. III.2 TRANSAX Since the selection of the underlying cause by ACME requires keying of all the entities (diseases, accidents or injuries) reported on the death certificate, it is possible to produce multiple cause statistics. For ACME purposes, conditions reported on the death certificate are coded "as they are reported". For instance, if coma and diabetes are reported separately, they must be coded separately (780.0 and with ICD9) for the selection of the underlying cause. However for statistical purposes, it is much more interesting to combine the two entities in the single code (diabetic coma) if the coma is due to the diabetes. The representation of codes with separated entities is called "Entity axis" and the merging of codes "Record axis". The conversion from Entity to Record axis is called "TRANSlation of AXis" and it is carried out by the program TRANSAX [NCHS81] [NCHS86]. Furthermore, the purpose of TRANSAX is to remove contradictions, duplications and to produce complete and accurate codification within the framework of the ICD structure. III.3 MICAR MICAR (Mortality Medical Indexing Classification and Retrieval) simplifies the data entry for the coding of medical entities [NCHS90][Hart96][NCHS96a][NCHS96b][NCHS96c]. The input to ACME and TRANSAX are the same ICD codes and at first, the task of coding was done manually by nosologists. They had to apply complex guidelines particular to the coding task and specified in hundred of pages of documentation to code multiple cause of death. Generally, the training of a nosologist requires attending a several months course and a year of experience to become proficient. The MICAR system was designed to reduce the resources needed for such training. The data entry personnel need only understand the basic terminology and concepts of multiple cause coding; training is reduced to approximately one month. In addition, this system improves accuracy and validity in coding by avoiding human errors. MICAR is a complex program made of 2 modules (MICAR100 and MICAR200). 19

20 The first module turns the text of the medical causes of death keyed in by the operator into entity reference numbers using a dictionary. The MICAR dictionary comprises a list of causes called entities. "Myocardial infarction", "arm fracture" are examples of entities. The text of each medical cause reported on the death certificate must be standardised by the operator, i.e. it must match exactly with an entity in the MICAR dictionary. For example, if the certifier reports "fracture of arm", the operator must key in "arm fracture" dropping the "of"; if the certifier reports "cancer of lung and liver", the operator has to key in the two entities "lung cancer" and "liver cancer". Each entry in the MICAR dictionary is associated with a number called the entity reference number. The use of an intermediate entity reference number instead of an ICD code makes the MICAR first module independent of the ICD revisions. The operator can enter either full text or an abbreviation or an entity reference number directly. The MICAR dictionary includes more than 100,000 medical terms. The second module transforms the entity reference numbers into ICD codes by using specific coding rules. An ICD code is generated from one or several entity reference numbers depending on their positional relationships. For instance, if the two entity reference numbers corresponding to "haemorrhage" and "stomach ulcer" are adjacent, "haemorrhage" will be coded as "gastric ulcer with haemorrhage". Because of the complexity of the design and testing, MICAR was implemented only recently by NCHS. In 1992, NCHS was in the process of testing and certifying MICAR using data from During this test, MICAR had a throughput rate of 85%, with a multiple cause error rate of 0.7% and an underlying cause of death error rate of 0.4%. With these results, MICAR was certified and NCHS began to use it on 1990 mortality data. Performance of MICAR depends on the type of medical causes of death treated. Currently, MICAR codes approximately 88% of the death certificates with 0.3% error in code assignment. MICAR was used to process 72% of the 1995 US mortality files. An interactive PC version of MICAR is available. III.4 SUPERMICAR SUPERMICAR is the last module of the US software for automated coding of causes of death [Hart96]. It is designed to facilitate the entry of literal causes of death. As mentioned in the previous 20

21 section, MICAR only accepts standardised expressions for causes of death. This standardisation is done by the operator who needs special training. The objective of SUPERMICAR is to allow the keying of causes of death as they are reported by the certifier. For instance, if the death certificate mentions "cancer of liver and lung", SUPERMICAR will change this expression into the two standardised expressions "liver cancer" and "lung cancer" acceptable by MICAR. The same would occur with the entry "pneumonia with liver cancer" which will be turned into "pneumonia" and "liver cancer". Due to its recent implementation, only a small amount of performance data is available on SUPERMICAR. This program was used to process 16% of the 1995 US mortality file. Presently, it processes successfully 75% of the records with an error rate of 1.2%. SUPERMICAR has also been prototyped in Scotland (c.f. section V.4). 21

22 III.5 Conclusion The ICD8 version of the NCHS ACS has been used by several other countries. It was first implemented by Canada and Brazil [NCHS81]. In the early eighties, a new system was developed for ICD9 on the basis of the ICD8 version. Parts of this system are now used by England and Wales, Scotland, Sweden, Australia and Israel. The ICD10 version is being developed and should be available for testing in The US system, especially ACME, is widely used. The ACME decision tables are a reference for the selection of the underlying cause of death, because they establish the relationships needed for the application of ICD rules. Even when these tables are modified by the countries using ACME, the differences in codes assigned can be clearly explained from the differences in the decision tables. Because of the differences in languages, the adpatation of SUPERMICAR and MICAR is not so easy. While these programs can be easily adapted for English speaking countries, it seems more difficult for other languages. NCHS continues to enhance the ACS with the ultimate goal of creating an electronic certificate, thus allowing the certifier to directly enter the causes of death, with a minimum of training. 22

23 IV. CAUSE-OF-DEATH PROCEDURES IN EUROPEAN COMMUNITY In the context of this project, it was essential to study the cause-of-death procedures in EC countries. These procedures have been investigated on the basis of a questionnaire sent to all Member States (Annex B). This questionnaire is divided into six parts. IV.1 Death certificate. This part describes which types of forms are in use among Member States, and what additional information is captured from death certificates. IV.2 Certification. Questions in this part concern mainly certifiers' qualifications, training and querying procedures. IV.3 Coding. This part reviews used and planned revisions of the ICD. It also examines who are the coders, and where and how the coding is performed. IV.4 Automatic coding systems. This part begins with a brief explanation of the different functions performed by an automatic coding system (ACS). This explanation is based on the US system. The questions which follow focus on the function(s) implemented by each country. IV.5 International cooperation on coding problems. Questions in this part show what topics are considered most important for international collaboration, and which countries are already involved in such collaborations. IV.6 Miscellaneous. This part includes some important questions not covered by previous sections. 23

24 In April 1997, 23 questionnaires were sent to 15 EC countries and to 4 other European countries. In some countries, several questionnaires were sent because there were more than one office responsible for mortality statistics production. The following table lists the countries according to the number of questionnaires. Country N. quest Country N. quest Belgium 2 Portugal 1 Denmark 1 Finland 1 Germany 1 Sweden 1 Greece 1 UK: England and Wales, 3 Scotland, Northern Ireland Spain / Catalonia 2 France 1 Irish Republic 1 Iceland 1 Italy 1 Liechtenstein 1 Luxembourg 1 Norway 1 The Netherlands 1 Switzerland 1 Austria 1 In January 1998, 21 completed questionnaires were received: all the countries answered except Liechtenstein and only one questionnaire was received from Belgium. The detail of responses by questionnaire is given in Annex D. The next sections summarise the answers question by question. Since two countries sent back more than one questionnaire (Spain and UK), we refer to questionnaire or to "mortality statistical offices" (offices in short) rather than countries in the following. 24

25 IV.1 Death certificate Objective This part of the questionnaire describes the medical death certificate in use in the statistical offices of the EC in 1997: differences with the WHO model, number of lines, additional information captured, different types of certificates used and status (public, secret). All these characteristics may influence the comparability of the statistics between countries. Results Death certificates are relatively homogeneous among EC countries and they are based on the international form recommended by WHO. This is the most important point for the comparability. Additional information available on the death certificate (accidental death, pregnancy...) differs between countries and common guidelines should be provided by WHO on this topic. Apart from the "general certificate", different specific forms are used in some offices, especially for infant deaths. Detailled answers a. Do you use a death certificate based on the international form recommended by WHO in the International Classification of Diseases (ICD)? yes: 19 (91%) - no: 2 (9%) Most of the offices use the international form. Belgium, one of the 2 offices using a special form, will use the international model in

26 b. Since when have you been using the current format of death certificate (year)? Year N (%) before (29%) (37%) since (24%) other 1 (5%) Not specified 1 (5%) Total 21 (100%) More than half of the offices have modified the death certificate since The "other" category relates to Germany where the different Länder started with the current death certificate at different dates. c. How many lines does your certificate provide in part I? Number of lines N (%) less than 3 lines 4 (19%) 3 lines 14 (67%) 4 lines 3 (14%) Total 21 (100%) Most of the certificates offer 3 lines as recommended in ICD9. d. If the number of lines is less than 4, do you intend to provide 4 lines in part I? yes : 7 (39%) - no: 10 (55%) - Other: 1 (6%) Among the 18 offices with a death certificate providing less than 4 lines, one third intend to provide 4 lines as recommended in the ICD10. These offices will implement this certificate before

27 e. Does the certificate capture additional information? Additional information N (%) accidental death 14 (67%) occupational accident 4 (19%) pregnancy 6 (29%) autopsy 13 (62%) surgery 5 (24%) occupation 9 (43%) smoking 0 (0%) Multiple response possible (percentage total > 100) This table shows the additional pieces of information captured by the death certificate. "accidental death" and "autopsy" are asked in two third of the certificates. "Smoking" is never asked. f. Do you have more than one form of certificate? yes: 15 (71%) - no: 6 (29%) More than 70% of the questionnaires mention several forms of death certificates. These forms are specified in question g. 27

28 g. If yes, what special certificates are there: N (%) neonatal death 5 (33%) stillbirth 12 (80%) late foetal death 4 (27%) other 6 (40%) Multiple response possible (percentage total > 100) Percentages relate to the 15 questionnaires with "yes" in question f. Most of the special certificates are for stillbirth registration. The "other" category includes special forms for faeto-infant death, external death, death registered by coroners and death under 1 year. h. If no, do you have additional parts included in the adult certificate for: N (%) neonatal death 4 (67%) stillbirth 3 (50%) late foetal death 3 (50%) Multiple response possible (percentage total > 100) Percentages relate to the 6 questionnaires with "no" to question g. In 4 offices, the general certificate is also used for neonatal death. 28

29 i. Is the medical death certificate a: N (%) public record 0 (0%) secret record 14 (67%) other 7 (33%) Total 21 (100%) The death certificate is never a public record. The death certificate is secret in two thirds of the cases, but usually researchers have access to this document. j. Is the deceased's name available with the medical causes of death? yes: 14 (66%) - no: 6 (29%) - other: 1 (5%) The death certificate is often anonymous. The case "other" corresponds to Germany where the situation depends on the Land concerned. 29

30 IV.2. Certification Objective This part of the questionnaire examines who certifies deaths and how. It investigates ways of improving the certification (training, query, additional information). Questions are also asked on the homogeneity of certification between regions. It is important to study these aspects to know the quality of the basic information used for coding. Results Training practices for certification vary markedly among Member States, from a set of examples to a video. In the same way, there is no common approach to sending an enquiry to the certifying physician when death certificates are incomplete or ambiguous. Common agreement on these aspects would be needed to increase quality and comparability of the data (e.g. definition of the cases that need to be queried). It would also be important to adopt a common definition of minima gestation period and weight for stillbirths and late foetal deaths. Detailled answers a. Who can certify a death? Certifier yes N (%) no N (%) qualified medical practitioner 21 (100%) 0 (0%)) coroner or legal officer 8 (38%) 13 (62%) other 3 (14%) 18 (86%) Multiple response possible (percentage total > 100) A qualified medical practitioner can always certify a death. In some cases a coroner, a legal officer or the police can certify a death too. 30

31 b. Please specify what categories of deaths are certified by each of the certifiers mentioned in question IIa: Certifier natural deaths N (%) non natural deaths N (%) qualified medical practitioner 20 (100%) 15 (75%) coroner, legal officer or other 2 (10%) 8 (40%) Multiple response possible (percentage total > 100) (One questionnaire did not answer this question and percentages are computed out of 20 questionnaires). Two broad categories are mentioned in the questionnaires: natural death, due to a known disease and non natural death including violent deaths and death with unknown cause. Qualified medical practitioners can always certify natural death and in 75% of the cases non natural deaths. Certifiers other than qualified medical practitioners certify mainly non natural deaths. 31

32 c. Estimate percentage of deaths certified by each type of certifiers mentioned in question IIa: Certifier 100% N (%) 90-99% N (%) <90% N (%) Total qualified medical practitioner 10 (47%) 6 (29%) 5 (24%) 21 (100%) 47% of the questionnaires report that qualified medical practitioners certify 100% of the death certificates, 29% certify between 90 and 99% and 24% certify under 90% of the death certificates. Certifier 0% N (%) 1-10% N (%) >10% N (%) Total coroner, legal officer or other 12 (57%) 4 (19%) 5 (24%) 21 (100%) In 19% of the cases, coroner, legal officer or others certifiers certify 1 to 10% of the death certificates and in 24% they certify more that 10%. d. Do you have training materials available for certifiers such as: N (%) set of examples 13 (62%) guidelines 13 (62%) video 2 (10%) other 5 (24%) Multiple response possible (percentage total > 100) The more used training materials are set of examples and guidelines. Other materials include training courses for physicians at university or seminaries. 32

33 e. Do you send query letters to certifiers in order to obtain further information on incomplete or ambiguous deaths certificates? yes: 18 (85%) - no: 2 (10%) - unknown: 1 (5%) Query letters to certifiers are widely used by the offices. The case "unknown" concerns Germany where some Länder send query letters and others not. f. If yes, what is the estimated percentage of death certificates affected? % of death certificates N (%) <5% 10 (55%) 5-10% 2 (11%) >10% 3 (17%) Unknown 3 (17%) Total 21 (100%) When query letters are sent (18 questionnaires), in more than half of the casesthe queries relate to less than 5% of the death certificates. g. Have you defined a particular procedure to send query letters (e.g. list of typical cases)? yes: 11 (52%) - no: 10 (48%) More than half of the questionnaires mention a procedure for query letters. Two lists of typical cases were received. h. Do you have a provision to allow further information about the death to be reported if it becomes available after certification (e.g. histology)? yes: 11 (52%) - no: 9 (43%) - unspecified: 1 (55%) In 52% of the cases, it is possible to take into account further information when available after certification. 33

34 i. For stillbirth or late foetal deaths, for what minima gestation/weight is registration required? Minimum gestation (weeks) N (%) 22 2 (10%) 23 1 (5%) 24 5 (24%) 26 (180 days) 2 (10%) 28 3 (14%) unspecified 8 (37%) Total 21 (100%) When specified, the minimum duration of gestation for registration varies from 22 to 28 weeks. Minimum weight (gr.) N (%) (29%) (10%) unspecified 13 (62%) Total 21 (100%) In most cases, 500 gr. is the minimum weight for registration. 34

35 j. Language(s) used for certification? Number of languages N (%) 1 17 (81%) 2 3 (14%) 3 1 (5%) Total 21 (100%) In the majority of cases, 1 language is used for certification. Yet, in 4 cases, 2 or 3 languages can be used. k. How long are the certificates archived (under any form)? Time of archive (year) N (%) (24%) (24%) >50 1 (5%) forever 8 (37%) unspecified 2 (10%) Total 21 (100%) More than one third of the offices archives the death certificates forever. l. Do you use an electronic certification system (certification made directly by the certifier to the statistical office via electronic means)? yes: 1 (5%) - no: 20 (95%) Electronic certification is only used by 1 office (Sweden) in forensic institutes. 35

36 m. What is the estimated percentage of death certificates typewritten? % of typewritten certificates N (%) 0 3 (14%) <5 5 (24%) (28%) >50 2 (10%) unspecified 5 (24%) Total 21 (100%) The percentage varies markedly from 0 to 98%. 28% of the questionnaires mention 5% to 49% of typewritten certificates. n. Have you undertaken studies on the differences in quality of certification between regions of your country? yes: 4 (19%) - no: 17 (81%) In a quarter of the cases, quality of certification is studied between regions. o. If yes, what method did you use? The 4 questionnaires which answer "yes" to question n described the method used. Methods consist of sample recoding, counting incomplete certificates and the level of ill-defined causes. p. Do you estimate that the certification quality is homogeneous between the different regions of your country? yes: 11 (52%) - no: 9 (43%) - unknown: 1 (5%) In more than half of the questionnaires, certification between regions is considered as homogeneous. 36

37 37

38 IV.3 Coding Objective This part analyses the ICD revision used for coding because differences between revisions make coding comparisons more difficult. It also studies whether cause-of-death are coded centrally or regionally. Some questions are about the staff involved in cause-of-death coding. This is an important aspect relating to the quality of coding even where ACS are implemented. The last question concerns the issue of ill-defined causes. Results Most of the offices code using ICD9. The date of implementation of ICD10 varies according to the country. The use of different revisions will lead to problems of comparability. Most of the offices are now routinely coding causes other than the underlying cause. This will help comparability studies based on multiple cause analyses. However, the number of causes coded varies (only 7 offices code all the causes of death) and common procedures should be adopted. The rate of illdefined causes is both an indicator of the quality of the causes of death statistics and an important factor in international comparability. This problem will need to be tackled by Member States. Detailled answers a. Which revision of the ICD do you currently use for coding? ICD revision N (%) ICD8 0 (0%) ICD9 15 (71%) ICD10 6 (29%) Total 21 (100%) No one is still using ICD8 and 6 countries are using ICD10. The 6 ICD10 users are Denmark (since 1994), Switzerland (since 1995), The Netherlands, Finland and Iceland (since 1996) and Sweden (since 1997). For information, ICD8 was published in English in 1968 [ICD68], ICD9 in 1977 [ICD77] and ICD10 in 1992 [ICD92]. 38

39 b. In the past, which revision of the ICD have you used for coding? All ICD9 users were using ICD8 and all ICD10 users were using ICD9 except Denmark and Switzerland that directly jumped from ICD8 to ICD10. c. If you are not currently using ICD10, when do you plan to do so? Year of implementation N (%) (20%) (60%) (20%) Total 15 (100%) The 15 offices which do not already use ICD10, plan to use it and mainly in d. Are causes of death coded: N (%) centrally and regionally 2 (10%) centrally only 18 (85%) regionally only 1 (5%) Total 21 (100%) In most cases, causes of death are coded centrally. Spain and Austria code both centrally and regionally. Germany codes regionally. 39

40 e. If causes of death are coded regionally, how many regional offices are involved? Germany has 16 regional offices involved, Spain 17 and Austria 2. f. Number of medical coders coding causes of death? Number of coders N (%) <1 1 (5%) (56%) (24%) >10 1 (5%) Unspecified 2 (10%) Total 21 (100%) More than half of the offices have 1 to 4 coders and one quarter 5 to 10. g. Do these coders perform work other than the cause-of-death coding? yes: 12 (57%) - no: 8 (38%) - unknown: 1 (5%) More than half of the coders code also information other than causes of death or do other work. 40

41 h. If yes, percentage of total coders time devoted to causes of death coding? Percentage of time N (%) (17%) (8%) (17%) (50%) Unspecified 1 (8%) Total 12 (100%) When coders perform work other than cause-of-death coding ("yes" to question g), half of them devote 75% or more of their time to cause-of-death coding. i. Do you code causes other than the underlying cause of death? yes: 15 (71%) - no: 6 (29%) More than two third of the statistical offices code multiple cause of death. It must be noted that, among the 6 offices that code the only underlying cause, 3 offices code 2 causes (see question k), this may be because they code the external cause and the nature of injury for violent deaths. j. Do you code all the causes mentioned on the death certificate? yes: 7 (33%) - no: 14 (67%) One third of the statistical offices code all causes of death reported on the death certificate. 41

42 k. If not, what is the maximum number of causes that you code (the underlying cause included)? Maximum number of causes N (%) 1 3 (21%) 2 5 (36%) 3 0 (0%) 4 4 (29%) 5 2 (14%) Total 14 (100%) Among the 14 statistical offices which do not code all the causes of death, more than half code 1 or 2 causes and less than half 4 or 5 causes. l. Do all your coders have the same level of competence, or do you give the "difficult" certificates to specialist coders (nosologists)? The formulation of this question was ambiguous and responses cannot be easily interpreted (does the answer "yes" concern the first part or the second part of the question?) m. Do you have a medical consultant(s)? yes: 17 (81%) - no: 3 (14%) - Unspecified: 1 (5%) More than 80% of the statistical offices have a medical consultant. n. Do you verify the coding, e.g. through independent recoding? yes: 7 (33%) - no: 14 (67%) In two third of the cases, there is no independent verification of the coding. 42

43 o. After what full-time period of training would you consider a new coder qualified to begin to work independently? Training period (months) N (%) (19%) (33%) (14%) (19%) >12 2 (10%) Unspecified 1 (5%) Total 21 (100%) The training period to qualify a new coder varies markedly depending on the office. One third of the offices consider that a 4 to 6 months period is necessary. p. In your country, what is the main reason(s) that deaths are certified as due to ill-defined causes? Main reason N (%) poorly completed certificates 19 (90%) insufficient information available to the certifier 16 (76%) legal investigation for violent death 4 (19%) medical confidentiality 1 (5%) other 9 (43%) Multiple response possible (percentage total > 100) The main reasons for ill-defined causes are: the certifier does not properly complete the certificate (90%) and the certifier does not know exactly the cause of death (76%). Other reasons given: old age, coding rules, lack of autopsies and deaths occurring abroad. 43

44 44

45 IV.4 Automatic Coding System (ACS) Objective This part analyses the precise functions carried out by ACS and the type of ACS used. Results Six offices use an ACS and 4 of these also use a computerised system for coding multiple cause of death. A number of other offices plan to adopt an ACS. So it is extremely important to adopt common guidelines, in particular a consensus on the decision tables. This is a necessary precondition to improving the comparability of coded data. Detailled answers a. Do you use a computerised system for translating causes of death into ICD codes (i.e. a system where the ICD code is assigned by the program)? yes: 6 (29%) - no: 15 (71%) 6 statistical offices use a computer program to code causes of death. One of them (Italy) use a coding program at an experimental stage. b. If yes, do you use: System used N (%) MICAR or SUPERMICAR 2 (33%) Own system 2 (33%) MICAR/SUPERMICAR and own system 2 (33%) Total 21 (100%) Among the 6 offices using an automated system for coding causes of death, 2 are exclusively using the US system MICAR or SUPERMICAR (Scotland and Italy), 2 are using their own system (Denmark and Sweden) and 2 their own system plus a part of the US system (Catalonia and England and Wales). 45