WWW-publications from the WHO MONICA Project

Quality Assessment of Stroke Event Registration Data in the WHO MONICA Project

November 1998¹

Markku Mähönen², Hanna Tolonen² and Kari Kuulasmaa² for the WHO MONICA Project³

¹ Misprints in Section 8 and Tables 1 and 16  were corrected on 25 February 2000
² MONICA Data Centre, National Public Health Institute, Helsinki, Finland
³ Annex: Sites and key personnel of the WHO MONICA Project

• Copyright notice
• Document identification:
  • URL:http://www.ktl.fi/publications/monica/strokeqa/strokeqa.htm
  • URN:NBN:fi-fe19991080

This document includes the main findings of unpublished reports:

• Asplund K, Feigin V, Schädlich H, Kuulasmaa K, Rajakangas A-M, Suzuki K for the WHO MONICA Project. Stroke event registration data quality report. MONICA Memo 211A, October 1991.
• Sarti C, Mähönen M, Rajakangas A-M, Kuulasmaa K for the WHO MONICA Project. Quality assessment of stroke event data for 1982-1990. MONICA Memo 277A, June 1994.

Acknowledgements

Thanks are due to Birgitta Stegmayr, Daiva Rastenyte, Per Thorvaldsen, Kjell Asplund and Alun Evans for thoughtful comments and advice.

The MONICA Centres are funded predominantly by regional and national governments, research councils, and research charities. Coordination is the responsibility of the World Health Organization (WHO), assisted by local fund raising for congresses and workshops. WHO also supports the MONICA Data Centre (MDC) in Helsinki. Not covered by this general description is the ongoing generous support of the MDC by the National Public Health Institute of Finland, and a contribution to WHO from the National Heart, Lung, and Blood Institute, National Institutes of Health, Bethesda, Maryland, USA for support of the MDC and the Quality Control Centre for Event Registration in Dundee. The completion of the MONICA Project is generously assisted through a Concerted Action Grant from the European Community. Likewise appreciated are grants from ASTRA Hässle AB, Sweden, Hoechst AG, Germany, Hoffmann-La Roche AG, Switzerland, the Institut de Recherches Internationales Servier (IRIS), France, and Merck & Co. Inc., New Jersey, USA, to support data analysis and preparation of publications.

Contents

• 1. Introduction
• 2. Material and methods
  • 2.1 Populations
  • 2.2 Periods of events considered
  • 2.3 Age and sex
  • 2.4 Other inclusion criteria for the data
  • 2.5 Sources of information
• 3. Serial number inventory and routine data checking status
• 4. Coverage of registration
  • 4.1 Attack rates of stroke in age group 65-74
  • 4.2 Fatal events
  • 4.3 Non-fatal events
• 5. Quality of the data for diagnostic classification
• 6. Availability of data on previous stroke
• 7. Summary
• 8. Comments on individual RUAs
• References to publications
• References to internal MONICA documents
• Specific tables
  • Table 1. Reporting Unit Aggregates, MCCs, Reporting Units, calendar years considered and the number of events registered
  • Table 2.1. Number of stroke event records by year, ages 25-64
  • Table 2.2. Number of stroke event records by year, ages 65-74
  • Table 3. Serial number inventory
  • Table 4. Number of unresolved constraint violations by year
  • Table 5. Number of unresolved suspected duplicate registrations
  • Table 6. Number of stroke events and crude rates/100 000 by 5-year age group at age 60-74
  • Table 7. Comparison of fatal events in MONICA register with routine statistics
  • Table 8. Fatal events: proportions of diagnostic categories
  • Table 9. Fatal events: proportions of categories of management
  • Table 10. Non-fatal events: proportions of diagnostic categories
  • Table 11. Non-fatal events: proportions of categories of management
  • Table 12.1 Case fatality of stroke events, age 25-64
  • Table 12.2 Case fatality of stroke events, age 65-74
  • Table 13. Fatal events: proportions of unclassifiable strokes and examinations done
  • Table 14. Non-fatal events: proportions of unclassifiable strokes and examinations
  • Table 15. Proportions of categories of data on previous stroke
  • Table 16. Summary status of stroke event data.
• Appendices
  • Appendix 1. Edit specifications
  • Appendix 2. Unresolved constraint violations
• Annex: Sites and key personnel of contributing MONICA Centres

1. Introduction

The quality of the data in the event registers is of particular importance for the attainment of the goals set for the MONICA Project (1). For stroke event registers the key issues are

• the coverage of registration
• the reliability of classification

and, for the assessment of trends within populations, which is the main concern of the MONICA study,

• the stability of the methods applied for registration and the availability of data in each MONICA Collaborating Centre (MCC).

The current report addresses these issues. It follows the approach of the previous stroke events quality assessment reports (6, 7) and a publication (5).

In the specifications of the calculations for this quality assessment report the names of the data items of the Core Data Transfer Format - Stroke Events (2) have been used. The terminology used is this report follows that developed for MONICA event registration in the MONICA Manual (2), with later refinements in the collaborative publications (3, 4).

2. Material and Methods

2.1 Populations

The report considers the Reporting Unit Aggregates (RUAs) which are foreseen as potential candidates for units of analysis of the MONICA stroke event data.The RUAs, their abbreviations and Reporting Units are listed in Table 1. Different combinations of Reporting Units (RUs) may be used for analyses concerning stroke events only, and others for analyses involving both stroke event and risk factor data. Some RUAs have several versions distinguished by suffix ´a´ or ´ b´. The reason for the distinction is that some RUs of some RUAs were not included in every risk factor survey or the different RUs had different stroke registration periods. Therefore, in RUS-MOI, RUS-NOC and the RUAs of MONICA East Germany (GER-EGE,GER-HAC, GER-KMS, GER-RDM) there is an overlap of RUs included in the RUAs concerned. Altogether 25 RUAs are considered (Table 1).

2.2 Periods of events considered

Only full calendar years of registration are considered. Individual years were excluded if

• The first year of registration was clearly incomplete (1982 for HUN-BUD and HUN-PEC)
• The MCC has indicated that the first year was a pilot year (1982 for FIN-KUO and FIN-TUL; 1984 for RUS-MOC, RUS-MOIa and RUS-MOIb)

Only the years of registration which were common to each RU of the RUAs of MONICA East Germany were included.

The years considered for each RUA are shown in Table 1.

2.3 Age and sex

The quality assessment concerns the age group 25-74. Since most MCCs registered stroke events for the age group 25-64 only, separate tables were prepared for age groups 25-64 and 65-74, except where otherwise stated.

Age was calculated in full years at the date of onset, except for mortality comparisons, when the date of death was used (Table 7). When calculating age, day 99 was interpreted as 15 and day/month 99/99 as 30/06. No age standardization was used.

2.4 Other inclusion criteria for the data

Individual records were excluded from the analysis in the following situations (except Table 2.1 and Table 2.2):

• DIACAT, SEX or PRESTR do not have valid values,
• DIACAT=4 and ICDVER=1 (8th revision of the ICD), and CLIND1-3 is not 430-434, 436,
• DIACAT=4 and ICDVER=2 (9th revision of the ICD), and CLIND1-3 is not 430-431, 433-434, 436.

Otherwise, all data available in the MDC were used in the analysis, regardless of the quality. The total number of events are shown by calendar year in Table 2.1 and Table 2.2.

2.5 Sources of information

The report is based on the data which the MDC has received from the MCCs on stroke events (Form 03), population statistics (Form A) and mortality statistics (Form C and Form E) and on other communication with the MCCs.

3. Serial number inventory and routine data checking status

Table 3 shows a summary of serial number inventory, which is based on a linkage of the stroke event data and the serial number inventory data (2) received in MDC. Its purpose is to check that the MDC database has exactly those records which it should have according to the MCC. Ideally, all entries of the last four columns should be zero. Otherwise there is a possibility that some records have been lost or duplicated from the time of the data's ascertainment.

There are quite a lot of discrepancies in the serial number inventory in LTU-KAU and RUS-NOC.

When the stroke data were received in the MDC, they were checked routinely for the constraints specified in Appendix 1. All violations of the constraints were reported to the MCC for their correction or elucidation. Data values outside the constraint limits were acceptable, but the MCC had to check that the values were not unusual because of data errors. The MCCs were only asked to revise data if they were incorrect. The current unresolved constraint violations are shown in Appendix 2 and summarized by calendar year in Table 4. There are several unconfirmed violations of the constraint DIACAT_SURVIV_3 in GER-EGE, GER-HAC, GER-KMS and GER-RDM. It is possible that these are DIACAT=9 events, in which case the constraint violations are acceptable, but the MCC was not able to check it any more. Other RUAs with unresolved constraint violations are FIN-KUO, GER-RHN, HUN-BUD, HUN-PEC and SWE-GOT.

The data are also checked periodically in the MDC for events which have been registered more than once. Table 5 shows the number of suspected duplicate registrations, which have neither been confirmed to be duplicate registrations nor separate events. There are unresolved suspected duplicate registrations in SWE-GOT.

4. Coverage of registration

4.1 Attack rates of stroke in age group 65-74

The attack rates of stroke in the age groups 60-64, 65-69 and 70-74 are shown in the Table 6. If all stroke events are registered, the rate should increase considerably with increasing age. There was a decrease with increasing age in CHN-BEI (1988-1989), GER-EGEa (1985), GER-HAC (1985-1986), GER-RDM (1984-1986), HUN-BUD (1984) and HUN-PEC (1986-1989), and only a minor increase in CHN-BEI (1985 and 1993). Clearly, the registration is incomplete in the older age groups in these RUAs.

4.2 Fatal events

Comparison with routine mortality statistics

According to MONICA Manual (2), the suspect stroke events are collected from death certificates, from hospital medical records and from community health services. All deaths with the underlying cause of death coded to 430-434 or 436 should be registered even if found not to be a stroke on review.

Data should have been submitted to the MDC for:

• every event classified in the diagnostic categories 1, 5 or 9,
• those fatal events with diagnostic category 4 and with the underlying cause of death, or (if not known) one of the three death certificate diagnoses (item CLIND of the Core Data Transfer Format - Stroke Events) coded to ICD codes 430-434 or 436.

To be certain that all stroke deaths are registered, the number of fatal events registered in each RUA with stroke as one of the death certificate diagnoses should be about the same as or greater than the number of stroke deaths reported in routine mortality statistics However, since the stroke deaths reported in the routine mortality statistics include also ICD codes 435 and 437-438, the number of MONICA registrations (fatal events with CLIND1-3 430-434 or 436) can be somewhat lower than the number of stroke deaths obtained from the routine statistics.

In some MCCs the access to death certificates or to the final ICD codes are restricted. Such MCCs are unable to check the coverage of their MONICA stroke register at the level of the individual deaths, and the MONICA register can miss deaths coded falsely as strokes in the official statistics. In such a case the MONICA register can have fewer deaths with CLIND1-3, coded 430-434 or 436 than there are official stroke deaths. However, even in such cases the total number of fatal registrations is expected to be larger than the number of stroke deaths in the official statistics.

For the comparison of the register with the routine statistics, two ratios are calculated:

• STRRATIO (STR/STR), which is the ratio of MONICA fatal registrations with CLIND1-3 430-434 or 436 to stroke deaths from routine mortality statistics;
• TOTRATIO (Total/STR), which is the ratio of total fatal registrations in the MONICA register to stroke deaths from routine mortality statistics.

Considering the above, STRRATIO below 0.9 may indicate incomplete case ascertainment. This ratio for the different RUAs is tabulated in Table 7. Note that in RUS-NOCa, RUS-NOCb and RUS-NOI it was only possible to calculate these ratios for the years 1989 and 1993 in the older age group. The ratio is below 0.9 in:

Age group 25-64

• CHN-BEI (1985-1986);
• DEN-GLO (1990);
• GER-RDMd (1989);
• GER-RHN (1984-1987);
• HUN-PEC (1984, 1986-1989);
• POL-WAR (1986, 1992-1994);
• RUS-MOC (1985, 1989);
• RUS-MOIa (1985-1986);
• RUS-MOIb (1985);
• RUS-NOCa (1987);
• RUS-NOCb (1987, 1990);
• SWE-GOT (1985).

Age group 65-74

• CHN-BEI (1984-1991, 1993);
• DEN-GLO (1982, 1984, 1990, 1991);
• GER-HACa (1985-1986);
• GER-RDMd (1984-1986);
• HUN-BUD (1983);
• HUN-PEC (1984, 1986-1989);
• RUS-NOCa (1989, 1993);
• RUS-NOCb (1989, 1993);
• RUS-NOI (1989, 1993).

TOTRATIO below 1.0 indicates incomplete coverage. This ratio for the different RUAs is tabulated in Table 7 (Reg/rout ratio: column Total/STR). The ratio is below this cut point in:

Age group 25-64

• CHN-BEI (1984-1986, 1989, 1991);
• DEN-GLO (1990-1991);
• GER-RDMd (1989);
• GER-RHN (1984-1987);
• HUN-PEC (1984, 1986-1989);
• POL-WAR (1988, 1991-1994);
• RUS-MOC (1985, 1987, 1989);
• RUS-MOIa (1985-1986, 1988-1989);
• RUS-MOIb (1985-1988, 1990-1991, 1993);
• RUS-NOCa (1987, 1991);
• RUS-NOCb (1987, 1990);
• RUS-NOI (1983, 1986-1987);
• SWE-GOT (1984-1985).

Age group 65-74

• CHN-BEI (1984-1991, 1993);
• DEN-GLO (1982-1985, 1987, 1990-1991);
• FIN-KUO (1991);
• FIN-TUL (1991);
• GER-HACa (1985-1986);
• GER-RDMd (1984-1986);
• HUN-BUD (1983);
• HUN-PEC (1984, 1986-1989);
• RUS-NOCa (1989, 1993);
• RUS-NOCb (1989, 1993);
• RUS-NOI (1989, 1993).

The numbers are small and in many RUAs the occasional years having ratios straying below 0.9 (STR/STR) or 1.0 (Total/STR) have only a few "missing" fatal events, which sometimes have acceptable explanations. The findings are commented on for individual RUAs in Section 8.

The proportion of DIACAT=4 and DIACAT=9 among all registered fatal events

If the registration of fatal events follows the MONICA rules, there should be some fatal events registered as suspect strokes (from death certificates with stroke as the cause of death) but finally classified as DIACAT=4 (not stroke). There should also be some DIACAT=9 (unclassifiable) events, since it would be very unusual that data on all  stroke deaths in the routine mortality statistics are so complete that they allow the classification as definite strokes or not strokes. A low proportion of DIACAT=4 and DIACAT=9 may indicate incomplete registration of suspect fatal strokes. Table 8 shows the proportions of the diagnostic categories in fatal events. There are no DIACAT=4 events in RUS-NOCa and only a few in RUS-NOCb, RUS-NOI and SWE-GOT. There are a few DIACAT=9 events in RUS-NOCa, RUS-NOCb and RUS-NOI. In SWE-GOT the proportion of DIACAT=9 events is notable. In the older age group there are no DIACAT=4 or DIACAT=9 events in RUS-NOCa, RUS-NOCb and RUS-NOI (see Section 8).

The proportion of non-hospitalized events among all fatal events

It is to be expected that some stroke deaths occur outside hospital. If the proportion of fatal events coded as hospitalized is very high it may indicate incomplete registration of out-of-hospital stroke deaths. The proportions of different categories of management are shown in Table 9. In some MCCs the proportion of out-of-hospital deaths is quite small; however, this probably more reflects the access to medical care, the organization of acute medical care and emergency services, and also different definitions of a "hospital" than the coverage of registration (5). This proportion is below 5% in the age group 25-64 in DEN-GLO (1985, 1991), HUN-PEC (1988-1989), and for age group 65-74 in DEN-GLO (1990), FIN-KUO (1991), FIN-NKA (1983-1984, 1986-1987, 1989), GER-RDM (1984-1985), and HUN-PEC (1988-1989). See Section 8 for specific comments on each RUA.

4.3 Non-fatal events

The assessment of the coverage of the registration of nonfatal events can be done only indirectly using the MONICA data, since there are generally no external data sources which the MONICA register could be compared with.

According to the MONICA protocol, suspected stroke events should be collected from emergency medical wards, from hospital medical records and from community health services. All clinically diagnosed strokes should have been registered. Data should have been submitted to the MDC for

• every event classified as DIACAT 1, 5 or 9; and
• on non-fatal events with DIACAT=4 and having stroke as one of the clinical diagnoses (items CLIND1-3).

Three indicators of the coverage of nonfatal events have been used:

• the proportion of diagnostic categories other than definite stroke (Table 10);
• the proportion of events managed outside hospital (Table 11);
• case fatality (Tables 12.1 and 12.2).

These indicators have not proved very useful, however. Non-adherence to the MONICA protocol is a reason for the low proportion of DIACAT=4 events in many MCCs. Only definite strokes have been registered in many MCCs, not nonfatal suspect strokes, even with clinical stroke diagnoses if they were not classified as stroke events. Considering non-hospitalized events, the proportion managed outside hospital is more dependent on the organization of medical care and perhaps on the definition of the term "hospital" than on the coverage of registration. If the case fatality is high, it may indicate incomplete case ascertainment of non-fatal events; however, it may also reflect true differences in the severity of strokes between the RUAs. These issues were also addressed in the MONICA collaborative publication on evaluation of case ascertainment in the WHO MONICA Stroke Study (5). See Section 8 for specific comments on each RUA.

Over 97% of the registered non-fatal events were definite (DIACAT 1+5) strokes during one or several years in the age group 25-64 in all RUAs but FIN-TUL and SWE-NSW, and in the age group 65-74 in all RUAs except FIN-TUL and SWE-NSW.

The proportion of nonhospitalized nonfatal events in age group 25-64 was below 1% in DEN-GLO (1988); FIN-NKA (1982,1986-1989,1991), HUN-BUD (1985), HUN-PEC (1986-1989), ITA-FRI (1984-1993), POL-WAR (1984-1988, 1990-1994), RUS-MOC (1987-1991, 1993), RUS-MOIa (1987,1989-1990, 1992-1993), RUS-MOIb (1987) and SWE-GOT (1984, 1986, 1988-1994), and in age group 65-74 in FIN-NKA (1985-1989,1991), GER-EGE (1987-1988), GER-HAC (1985-1986), GER-KMSb (1987-1988), HUN-BUD (1983-1985, 1987-1989), and HUN-PEC (1984, 1986, 1989).

Crude case fatality (DIACAT 1+5) was over 40% in age group 25-64 in GER-RDM (1988), HUN-BUD (1984-1985, 1989), POL-WAR (1984-1986), RUS-MOC (1991-1993), RUS-MOIa (1990-1993), RUS-MOIb (1987,1991-1993), RUS-NOCa (1989), RUS-NOI (1990) and YUG-NOS (1983, 1985, 1988-1989, 1994), and in age group 65-74 in CHN-BEI (1984-1986, 1991), GER-EGEa (1985-1986), GER-EGEb (1985-1986, 1990), GER-HAC (1985-1986), GER-KMS (1985-1986), GER-RDMd (1984-1985), HUN-BUD (1983-1989), HUN-PEC (1984, 1986-1987), RUS-NOCa (1990) and RUS-NOI (1987).

5. Quality of the data for diagnostic classification

According to the MONICA rules, the classification of suspect stroke events is based on clinical signs. As no data are collected about these findings, there are no direct quality indicators in the data which could be used in the assessment of the validity of the diagnostic classification.

Autopsy gives valid data on fatal events. Therefore, if the autopsy rate is high the classification of fatal events is reliable. If there are changes in the autopsy rate over time within RUA then the reliability of the classification may vary.

The proportion of autopsies and/or CT-scan done in fatal events is shown in Table 13. There are big differences between the MCCs in the frequency of autopsies. The increased availability of CT scans is probably the explanation for the decline in autopsy rates in some RUAs. In the age group 25-64 the proportion of fatal events with either autopsy or CT scan (or both) was over 60% for all years in DEN-GLO, FIN-KUO, FIN-TUL, HUN-BUD, HUN-PEC, ITA-FRI, LTU-KAU, RUS-NOCa and SWE-NSW.

The proportion of fatal events with either autopsy or CT scan (or both) increased in most MCCs. Both of these procedures increase the accuracy of the clinical diagnoses and also the MONICA classification, either directly (autopsy) or indirectly (CT). More accurate classification may have an impact on the trends in fatal events, especially in those RUAs where the frequency of autopsies and/or CT scans was low at the beginning of the study period. This concerns RUAs CHN-BEI and GER-EGE.

In fatal events without autopsy, the data items on investigations gives indirect evidence of the quality of the data on which the classification is based. Since the classification is based on clinical symptoms and signs, the patients should have been examined by a physician, and the classification is then made using the clinical notes.

In some MCCs a sizeable proportion of fatal events was not seen by a physician (Table 13). In age group 25-64 this proportion was over 10% in GER-EGEa (1988), GER-EGEb (1988);GER-HAC (1985-1987), GER-KMS (1988), GER-RDM (1986-1987, 1989), GER-RHN (1984-1986), HUN-PEC (1986), LTU-KAU (1986-1988, 1993), POL-WAR (1984-1992), RUS-MOC (1987-1989, 1992-1993), RUS-MOIa (1987-1989, 1992-1993), RUS-MOIb (1987-1993), RUS-NOCa (1988), RUS-NOCb (1988-1990), RUS-NOI (1988), SWE-GOT (1987, 1992-1994), and in age group 65-74 in GER-HAC (1985-1986), GER-RDM (1985-1986), HUN-BUD (1988) and HUN-PEC (1987). For such events it is especially important to get information about the data available to the MCC and about the procedures used to validate these events. (See Section 8 for specific comments on each RUA.)

In nonfatal events the classification is based on clinical findings and the evaluation of the findings by a physician. All non-fatal events were seen by a physician (Table 14), the few exceptions below 100% being LTU-KAU in 1988 (99%), POL-WAR in 1989 (97%) and 1993 (99%) in the age group 25-64.

According to the MONICA classification rules, the stroke diagnosis is based on clinical signs only. However, the powerful new imaging possibilities and especially CT scanning may have had an impact on the classification. The proportion of CT scans and other diagnostic examinations for non-fatal events is shown in Table 14. During the study period there was a dramatic rise in the proportion of CT scans. In the age group 25-64 this proportion was over 50% in all study years in DEN-GLO, FIN-KUO, GER-RHN, ITA-FRI, SWE-GOT and SWE-NSW. In the age group 65-74 CT scanning was less frequent.

Even though the frequency of autopsies is a quality indicator of the reliability of the classification, it is important to note that MONICA protocol does not set any criteria for the frequency of autopsy. The classification of strokes is a clinical one in the MONICA stroke study, based on clinical signs. Nevertheless, in non-fatal events the increased use of CT scans is a potential source of bias.

6. Availability of data on previous stroke

The data item on previous stroke is important when the first ever strokes need to be identified. This concerns, for example, the calculation of incidence rates. The proportions of different categories of data on previous stroke are given in Table 15.

In the age group 25-64 years the proportion of insufficient data on previous stroke was over 30 % in:

• GER-RHN (1984)
• HUN-BUD (1983-1984)
• HUN-PEC (1984)
• POL-WAR (1985)
• RUS-NOCa (1988-1989)
• RUS-NOCb (1988)
• RUS-NOI (1989)

In addition to these, the proportion of insufficient data was over 10% in the age group 25-64 years in:

• CHN-BEI (1984-1985)
• GER-RDMd (1987-1988)
• HUN-PEC (1989)
• ITA-FRI (1985)
• POL-WAR (1984, 1986, 1989-1994)
• RUS-MOC (1985-1986, 1989)
• RUS-MOIa (1985-1986, 1988-1989)
• RUS-MOIb (1985-1986, 1988-1989)
• RUS-NOCa (1987, 1990-1993)
• RUS-NOCb (1987, 1989-1990, 1992-1993)
• RUS-NOI (1983-1986, 1988-1993)

In the age group 65-74 years the proportion of insufficient data was over 30 % in:

• HUN-BUD (1983-1984)
• HUN-PEC (1984)

In addition to these, the proportion of insufficient data was over 10% in the age group 65-74 years in:

• CHN-BEI (1984)
• DEN-GLO (1984)
• GER-EGEa (1985)
• GER-EGEb (1985)
• GER-HAC (1985-1986)
• GER-KMSb (1985)
• GER-RDMd (1985)
• HUN-PEC (1984, 1987, 1989)
• RUS-NOCa (1987-1988)
• RUS-NOCb (1988)
• RUS-NOI (1984-1985)

7. Summary

Table 16 gives an overall summary of the annual data quality for each RUA. Data for the RUAs with years coded "2" have sufficient quality for the calculation of attack rates and case fatality. Code "1" indicates that there is a deviation from the assessment criteria for good quality, and  a decision on the use of the data needs to be made separately for each publication. Score "0" is used for cases where the data are not sufficiently complete for inclusion in analyses. Code "2*" means that the year considered does not satisfy the criteria set but there is an acceptable explanation for the deviant pattern (see Section 8). It is important that those who use the data read the comments carefully, especially concerning RUAs coded "1" or "2*".

In the age group 25-64 years there have been several problems in the data for RUS-NOC and RUS-NOI (see Section 8), particularly, several questions about the completeness of registration. The MCC has worked hard with the data and several data amendments have been possible; however, there are still  data quality questions, especially in RUS-NOCa and RUS-NOCb. A separate decision for each publication should be made for RUS-MOC, RUS-MOIa and RUS-MOIb: the coverage is incomplete to some extent; there were also several years for which the proportion of fatal events unseen by a physician was over 10%. In POL-WAR the case-fatality was over 40% throughout several years; also, the proportion of fatal events unseen by a physician was over 10% for several years (see Section 8). There were also several years in GER-HAC, GER-KMS, GER-RDM, GER-RHN, SWE-GOT and LTU-KAU when over 10% of fatal events were unseen by a physician. The availability of data on such events, and therefore the diagnostic classification, varies, but mostly they end up as unclassifiable, which should be taken into account when analysing the data (see Section 8).

In GER-RHN, HUN-BUD and HUN-PEC the coverage is clearly incomplete for age group 25-64. The data from the East German RUAs are only for 5-6 years which should be considered when planning the analyses.

In the older age group, data from RUAs DEN-GLO, FIN-KUO, FIN-NKA, FIN-TUL and SWE-NSW are admissible. The data indicate that registration of fatal events in RUS-NOCa and RUS-NOCb are incomplete (low and fluctuating case fatality, and low STR/STR and total/STR ratio) but the assessment cannot be completed, since routine mortality statistics data are missing for almost all years. The data from RUS-NOI are more consistent. The data from GER-EGEa are for only two years and for GER-EGEb and GER-KMSb for three years and cannot be used in the trend analyses.

The validity of the diagnostic classification of non-fatal events cannot be assessed from the data. However, the symptoms and clinical findings are very typical. For fatal events the high proportion of events unseen by a physician is explained by the completeness of registration of suspected stroke events - paradoxically, good coverage can lead to less reliable classification. Most of these events were registered from death certificates and there was very little information available to the MONICA team to validate these events. This is reflected in the proportion of DIACAT=9 events.

The proportion of CT scans increased markedly during the study period which may introduce a bias in the trends. CT findings have a notable impact on the death certificate diagnoses and, thus, also on the MONICA event registration which depends on the death certificates. In many MCCs the frequency of autopsies decreased along with the increase in CT scans which to some extent may 'cancel out' the biases introduced by increased reliance on CT scans.

In non-fatal events, the use of CT scans increased dramatically in several MCCs. According to the MONICA rules the classification of events into diagnostic categories is based on clinical findings, but the results of CT scan are important for clinical diagnoses and, so, indirectly guides the allocation of events to different MONICA categories (and perhaps directly, too, since the results of CT scan are known to the MONICA team). Those preparing collaborative publications should report the frequency of autopsy and/or CT scans in fatal events and the increased use of CT scans in nonfatal events. It is difficult to evaluate the bias introduced by the increased use of CT, but this should be attempted.

8. Comments on individual RUAs

CHN-BEI

ICD-version used: 9
Nonfatal case finding method: Cold pursuit
Access to death certificates: Full

Comments:

• Data have been received in the MDC for ten full years, from 1984 to 1993.
• According to the MCC, the registration of fatal events is probably incomplete in 1984-1986 and it is recommended that these years are not used in trend analyses.
• Because the registration of fatal events is most probably incomplete in the older age group (65-74 years) it is recommended that the data for this age group are not used.
• The routine mortality statistics have been poorly developed previously. The MONICA study has been a key factor in developing it. Death certificate registration has not been a legal responsibility in China before 1990. Death certificates are often written by administrators and are not very reliable.
• In fatal events in the age group 25-64 years, the STR/STR ratio and total/STR ratios are below one in 1984-1986, 1989 and 1991 which may indicate an incomplete case ascertainment. On the other hand, the proportion of DIACAT 4 is substantial, as well as the proportion of fatal non-hospitalized events.
• In fatal events in the age group 65-74 years, the STR/STR ratio is below 0.90 and total/STR ratio is below one in all years but 1992 which may indicate an incomplete case ascertainment. On the other hand, the proportion of DIACAT 4 is sizeable, as well as the proportion of fatal non-hospitalized events.
• The event rates in the age group 70-74 are lower than in the age group 65-69 in the years 1988-1989. This indicates an incomplete case ascertainment in the age group 70-74 in 1988 and in 1989.
• In nonfatal events, the proportion of DIACAT 4 (not stroke) events is negligible.
• In the age group 65-74, case fatality is over 40% in 1984-87 and 1991 which may indicate an incomplete case ascertainment of non-fatal events.
• In fatal events, the proportion of events without necropsy and not examined by a physician varies over time.
• Serial number inventory shows some discrepancies.
• Quite a lot of stroke event forms (form 03) do not have the corresponding serial number inventory form (form 07).

DEN-GLO

ICD-version used: 8
Nonfatal case finding method: Cold pursuit
Access to death certificates: Full

Comments:

• STR/STR ratio is <0.9 for the year 1990 may indicate an incomplete registration of fatal events. TOTRATIO is <1.0 in 1990-1991 (11 and 3 missing events). All death certificates with stroke as the underlying cause of death (written diagnosis) have been looked at, however, the reason for the discrepancies most probably is that the ICD codes are assigned by central authorities after the death certificates have been seen by the MONICA team, and these do not always correspond with the diagnoses written in death certificates.
• There are only a few DIACAT 5 events.
• There are only a few fatal events coded as MANAGE=4 (medically unattended) which may indicate that the registration of acute non-hospitalized stroke deaths is incomplete. The reason for the low proportion is in the organization of acute medical care and emergency medical services in an urban area.
• The proportion of nonfatal non-hospitalized stroke events is quite low. Also, there are no nonfatal DIACAT 4 events and only a few DIACAT 9 events. This may indicate an incomplete registration of nonfatal events. However, crude case-fatality is quite low which speaks against incomplete nonfatal case ascertainment. The reasons for the discrepancies are found in the organization of the health care system - all patients suspected of having a stroke are admitted to a hospital.
• In the older age group, STR/STR ratio was <0.9 in 1982,1984 and 1990-1991 which may indicate an incomplete case ascertainment. Also, TOTRATIO was below one in 1982-1984 (11, 3 and 9 missing events) and in 1990-1991 (6 and 5 missing events). For an explanation, see the first comment considering the younger age group.
• The event rates in the older age group rises with age, which also speaks against incomplete case ascertainment.
• The proportion of necropsies done in fatal events, and CT-scan done both in fatal and in nonfatal events is quite high during the whole time period.
• In fatal events, the proportion of events without necropsy and not examined by a physician varies over time.

FIN-KUO

ICD-version used: 8/9 (During 1983-1986 ICD 9, from 1987 ICD 9 )
Nonfatal case finding method: Hot pursuit
Access to death certificates: Full

Comments:

• TOTRATIO was below one in the age group 65-74 years in 1991 (one missing event); STR/STR ratio was >0.9.

FIN-NKA

ICD-version used: 8/9 (During 1982-1986 ICD 9, from 1987 ICD 9)
Nonfatal case finding method: Hot pursuit
Access to death certificates: Full

Comments:

• The proportion of other diagnostic categories than definite strokes in nonfatal events is low during several years.
• The proportion of non-hospitalized fatal events is low, but all deaths with a stroke diagnosis in the routine mortality statistics are registered (both TOTRATIO and STR/STR ratio are over 1.0 during all years). The explanations are the local health care organization and the interpretation of health care centre with facilities for in-patient treatment as a hospital.
• A few discrepancies with the serial number inventory.

FIN-TUL

ICD-version used: 8/9 (During 1983-1986 ICD 9, from 1987 ICD 9)
Nonfatal case finding method: Hot pursuit
Access to death certificates: Full

Comments:

• In the age group 65-74 years, TOTRATIO was below one in 1991 (2 missing events); STR/STR ratio was >0.9.
• A few discrepancies with the serial number inventory.

GER-EGE, GER-HAC, GER-KMS and GER-RDM

ICD-version used: 9
Nonfatal case finding method: Mixed
Access to death certificates: Full

Comments:

• The MCC has data for the defined RUAs for five to six years, and in the older age group only for two to three years.
• In the age group 25-64 years, the STR/STR ratio and TOTRATIO are low in 1989 in GER-RDMd which may indicate an incomplete case ascertainment. The numbers are small, however (three missing events out of 39 routine mortality statistics deaths).
• In fatal events, the proportion of events without necropsy and not examined by a physician is sizeable and varies over time. These events were registered on the basis of diagnoses obtained from death certificates. These death certificates were obtained very late and therefore no additional information was available for these events. This quite high proportion is the reason why the summary score is 2* for several years.
• Case fatality is over 40% in GER-RDMd in 1988 in the age group 25-64 years. This is clearly higher than during the other years and may indicate an incomplete case ascertainment of nonfatal events. This is the reason for summary score 1. In the older age group (65-74 years) case fatality is quite high which may indicate an incomplete case ascertainment of nonfatal events (the reason for summary score1). The numbers are small, however, and there may be random fluctuation in the case fatality figures.
• Data on the age group 65-74 are clearly incomplete in GER-HACa and in GER-RDMd.
• A big proportion of strokes are coded as 437 in routine mortality statistics.
• After the reunification of Germany there were changes in death certificate coding procedures which explain changes in the routine mortality statistics, however, data after 1990 are not considered in this report (only RU19 has data from 1991 onwards).
• There are a few discrepancies in the serial number inventory and several unresolved constraint violations (mainly DIACAT_SURVIV violations); the MCC was unable to correct them in the final years of the Project.
• There are several DIACAT_SURVIV violations: 33 in GER-EGE; 11 in GER-HAC; 33 in GER-KMSb. These are most probably DIACAT=9 events (in which case the coding is unusual but correct); the MCC was unable to check this in the final years of the Project.

GER-RHN

ICD-version used: 9
Nonfatal case finding method: Hot pursuit
Access to death certificates: Restricted

Comments:

• The MCC has data for only four years.
• The coverage of fatal events is clearly incomplete in 1984 and may be incomplete in 1985-1987.
• The proportion of fatal events not seen by a physician is over 10% over all years.
• There are a few unresolved constraint violations.
• The MCC stopped the event registration early and was unable to amend the data thereafter.

HUN-BUD and HUN-PEC

ICD-version used: 9
Nonfatal case finding method: Hot pursuit
Access to death certificates: Full

Comments:

• The coverage of fatal events is incomplete in HUN-PEC in 1986-1989 and possibly in 1984. In the older age group the coverage is probably incomplete in HUN-PEC.
• The case fatality is over 40% in HUN-BUD in 1984, 1985 and 1989 which may indicate incomplete case ascertainment of nonfatal events; in the older age group, the case fatality is quite high in HUN-BUD which clearly indicates incomplete case ascertainment of nonfatal events. The MCC has notified the MDC that there are problems in the case ascertainment; there are also other hospitals treating patients than those from which data have been collected.
• There are quite a lot of discrepancies in the serial number inventory and some unresolved constraint violations.
• The MCC stopped event registration early and was unable to amend the event registration thereafter.

ITA-FRI

ICD-version used: 9
Nonfatal case finding method: Cold pursuit
Access to death certificates: Full

Comments:

• The proportion of nonfatal events not hospitalized is zero. Case fatality>30% might indicate an incomplete nonfatal case ascertainment; case fatality has been declining and is clearly below 30% at the end of the study period. According to the MCC, all stroke patients are admitted to hospital which explains why there are no nonfatal non-hospitalized events. Also, the proportion of DIACAT 4 events is sizeable which speaks against case ascertainment bias.
• The proportion of necropsies done is quite stable, between 21%-33% of all fatal events.
• The proportion of CT examinations in nonfatal events is quite high and stable, between 70%-80%.
• 4 missing stroke event forms (FORM07 with status=1 and no FORM03).
• The proportion of events without necropsy and not examined by a physician is between 1-9%.

LTU-KAU

ICD-version used: 9
Nonfatal case finding method: Cold pursuit
Access to death certificates: Full

Comments:

• In nonfatal events the proportion of DIACAT 4 events is low which may indicate an incomplete case ascertainment. The proportion of non-hospitalized nonfatal events is sizeable, however, and crude case-fatality is not high, which speaks against incomplete nonfatal case ascertainment.
• In fatal events, the proportion of events without necropsy and not examined by a physician is substantial and varies over time. The explanation is that in out-of hospital deaths without necropsy and without any previous information (like ambulatory cards or hospital medical records or the family doctor's records) the MONICA team is assured that the patient was not seen by a doctor. The codes used in such situations are: 'MANAGE=4', 'DIACAT=9', and (EXAM1 or EXAM2)=2 or 9. The frequency of autopsies is very high, and between 70-80% of fatal events had either autopsy or CT.
• Quite a lot of discrepancies in the serial number inventory.

POL-WAR

ICD-version used: 9
Nonfatal case finding method: Cold pursuit
Access to death certificates: Restricted

Comments:

• After preliminary review of all hospital records only cerebrovascular events lasting more than 24 hours (definite events) were registered from the ICD-9 codes 430-431, 433-343 and 436.
• Out-of hospital deaths were registered from death certificates.
• The STR/STR ratio is below 0.90 in 1986 and 1992-1994 which may indicate an incomplete registration of fatal events. TOTRATIO is below one in 1988 and 1991-1994 which also may indicate an incomplete registration of fatal events. The explanation for the low ratios is that only deaths with ICD-9 codes 430-431, 433, 434 and 436 were registered. The MCC did not, however, register fatal events with ICD-9 codes 432, 435, 437 and 438. For fatal hospitalized cases the number of hospitalizations with these diagnoses (432, 435, 437-438) were 15-20/year: 16 in 1986, 15 in 1992, 23 in 1993 and 20 in 1994. For fatal non-hospitalized cases the respective numbers were 3-4/year. There were also death certificates in which the codes were for definite stroke but the written diagnosis was totally different (0-6/year). Taking all this into account, the coverage of stroke event registration is complete (see Table 7).
• For nonfatal hospitalized cases, there were between 25-43 hospitalizations/year with diagnoses 432, 435, 437-438 during the study period.
• In fatal events, the proportion of events without necropsy and not examined by a physician is substantial and varies over time. These are registered from death certificates. The MONICA team has no possibility of obtaining any more information from these events than the information obtained from death certificates (cause of death, dates of birth and death, name and sex). Also, the frequency of necropsies is quite low in such out-of-hospital deaths (5%). Therefore the proportion of DIACAT 9 events is sizeable. These events are registered on the basis of diagnoses in the death certificate but not validated.
• Almost all nonfatal events are hospitalized, which may indicate incomplete case ascertainment. According to the evaluation the MCC has done in the area, all suspect stroke events are hospitalized. The evaluation was done writing a letter to general practitioners residing in the MONICA study area once a year and asking them if they treat suspect stroke patients. The response rate has been 50-80%. No GP reported such cases. Other stroke registers in Warsaw operating in 1990-1991 based on hospital sources and GP registers showed that less than 5% of all stroke events (both fatal and nonfatal) were managed as outpatients only.
• The crude case-fatality is quite high. The reason may be that several of the DIACAT 9 events may be in fact false positives, but in the absence of information it has not been possible to validate these events.

RUS-MOC

ICD-version used: 9
Nonfatal case finding method: Cold pursuit
Access to death certificates: Full

Comments:

• Because of administrative changes in Moscow (population data cannot be calculated for the final year) the MCC will not be able to collect data for the final year of registration, and the stroke event registration covers years 1985-1993 only.
• The number of registered nonfatal events is decreasing while the number of fatal events is stable or increasing which arises suspicion of incomplete case ascertainment. The same methods are applied over the whole time period, however: 53 hospitals and 19 out-patient clinics are covered. The register team does not have access to some medical institutions; these are, however, inaccessible for ordinary people and very expensive.
• According to the MCC the number of hospitalizations for stroke is decreasing and the number of out-of-hospital stroke deaths is increasing. According to the MCC the reason is the reorganization of primary care in Moscow, people do not go to hospitals as often as previously.
• In fatal events, the STR/STR ratio is below 0.90 in 1985 and in 1989. The TOTRATIO is below one in 1985, in 1987 and in 1989. A few fatal events may be missing especially in 1989. The explanation is medical institutions from which no information is available to the register team.
• The number of registered nonfatal events declines from 142 (in 1986) to 48 (in 1992). This may indicate an incomplete case ascertainment especially in 1992 and possibly in 1993. Also, the crude case-fatality is clearly higher in 1922-1993 than during the previous years.
• It seems probable that more nonfatal strokes remained unrecognized in 1992-1993 compared with previous years, because people did not go to hospital.
• The proportion of fatal events without necropsy and not examined by a physician is notable and varies over time. These events were registered on the basis of diagnoses obtained from death certificates and no other information was available for the MONICA team. Therefore the proportion of DIACAT 9 events is quite high.
• The MCC did not have resources to correct the data in the last years of the Project.

RUS-MOI

ICD-version used: 9
Nonfatal case finding method: Cold pursuit
Access to death certificates: Full

Comments:

• See the general comments for RUS-MOC.
• In fatal events, the STR/STR ratio is below 0.90 in 1985 and in 1986 in RU 02, and in 1985 in RU 03. The total/STR ratio is below one in 1985-1988, 1990-1991 and 1993. Thus a few fatal events are missing especially in 1985.
• The number of registered nonfatal events declines from 295 (in 1986) to 155 (in 1993). This may indicate an incomplete case ascertainment especially in 1992 and possible in 1993. Also, the crude case-fatality is clearly higher in 1922-1993 than during the previous years.
• The MCC did not have resources to correct the data in the last years of the Project.

RUS-NOC

ICD-version used: 9
Nonfatal case finding method: Hot pursuit
Access to death certificates: Full

Comments:

• The MCC has worked hard with the data problems, and the situation improved towards the end of the study; however, there are still questions about the completeness of registration, particularly, unclear year-to-year fluctuations in the number of fatal events and fluctuations in case fatality especially in the older age group. According to the MCC, due to the big social changes in Russia there has been a decline in the standard of health care; also, there has been substantial changes in alcohol consumption; these factors may explain some of the the fluctuation in case fatality.
• Nonfatal events are missing in 1994.
• Fatal events are missing in 1994 in RUS-NOCb.
• For RUS-NOCb the comparison with routine mortality statistics cannot be done for the years 1991-1993; for financial reasons, after 1990 routine mortality statistics data were collected only for RU 1 (i.e. RUS-NOCa).
• Routine mortality statistics are missing for 1994.
• The comparison with routine mortality statistics cannot be done in the older age group but for the years 1989 and 1993 since routine mortality statistics data are missing; for these years the coverage of fatal events is clearly incomplete.
• In 1987 a significant proportion of fatal cases in official statistics were coded 437 which has an impact on STR/STR ratios and TOTRATIOs.
• Previously, the MCC did not send DIACAT 4 or 9 events to the MDC, even though they were registered. Now the MCC has sent fatal DIACAT 4 and 9 events to the MDC but it was not possible to add nonfatal DIACAT 4 and 9 events to the database sent to the MDC.
• In fatal events there are no DIACAT 4 or DIACAT 9 events in the older age group.
• In fatal events the proportion of management not known is high and rising over time in the age group 25-64; in the older age group, management was known almost always. The reason for this discrepancy is that a substantial amount of  fatal events have been added later on in the age group 25-64 to complete the registration.
• Almost all nonfatal events are DIACAT 1 events, which indicates that other events were not registered and may be a sign of incomplete case ascertainment; on the other hand, the registration covers also out-of-hospital events since quite a big proportion were non-hospitalized events.
• In fatal events the proportion of autopsies done is high.
• Case fatality is unusually high in 1989 in RUS-NOCa which indicates that case ascertainment of nonfatal events is incomplete.
• Several discrepancies with the serial number inventory.

RUS-NOI

ICD-version used: 9
Nonfatal case finding method: Hot pursuit
Access to death certificates: Full

Comments:

• See also the comments for RUS-NOC, especially concerning fluctuations in case fatality and registration of DIACAT 4 and DIACAT 9 events.
• The data from RUS-NOI are more consistent than those from RUS-NOC.
• The TOTRATIO is below one in 1983 and 1986-1987.
• Nonfatal events are missing in 1994.
• Routine mortality statistics are missing for 1994.
• The comparison with routine mortality statistics cannot be done in the older age group except for the years 1989 and 1993 since routine mortality statistics data are missing; for these years the coverage of fatal events is clearly incomplete.
• In 1987 the proportion of fatal cases in official mortality statistics coded 437 was quite high (34 cases) which explains the lower STR/STR ratio and TOTRATIO.
• Previously, all registered fatal events were DIACAT 1 or 5 events, there are no DIACAT 4 or DIACAT 9 events, which indicated that the registration is incomplete and the classification did not follow the MONICA rules. The MCC has completed the registration of fatal events and there are now also DIACAT 4 and DIACAT 9 events but the numbers are small.
• In fatal events there are no DIACAT 4 or DIACAT 9 events in the older age group.
• Almost all nonfatal events are DIACAT 1 events, which indicates that other events were not registered and may be a sign of incomplete case ascertainment; on the other hand, the registration covers also out-of-hospital events since quite a big proportion were non-hospitalized events.
• In fatal events the proportion of autopsies done is high.
• Case fatality fluctuates and is higher in 1985 and 1990 which may indicate incomplete nonfatal case ascertainment during these years.
• A few discrepancies with the serial number inventory.

SWE-GOT

ICD-version used: 8/9 (ICD 8 until 1986, from 1987 ICD 9)
Nonfatal case finding method: Hot pursuit
Access to death certificates: Full

Comments:

• In fatal events, the STR/STR ratio is below 0.90 in 1985, and total/STR ratio is below one in 1984 and 1985 which may indicate an incomplete case ascertainment. According to the MCC, all death certificates having stroke as one of the clinical diagnoses have been reviewed, however, and DIACAT 4 events are not sent to the MDC.
• In fatal events, there were more DIACAT 9 events after 1991; the MCC succeeded to get the necessary information afterwards and now the proportion is clearly lower than previously but still somewhat greater than before 1991. Also, the proportion of fatal events with no necropsy and unseen by a physician before death is sizeable after 1991. These events are registered on the basis of diagnoses obtained from death certificates. The reason is a new law in Sweden regarding autopsies, a consequence of which was that much fewer cases went to autopsy and death certificate was written by GP. About 70% of out-of hospital deaths were medically unattended during 1992-1994.
• Almost all nonfatal events have been hospitalized. This may indicate an incomplete case ascertainment. The explanation is that in Sweden all suspect strokes are hospitalized. Special case-finding attempts have not yielded any nonfatal strokes entirely on an outpatient basis.
• There are no nonfatal DIACAT 4 events because they are not sent to the MDC. According to the MCC all suspect strokes have been assessed but those considered non-strokes have not been registered. The methods for case ascertainment have been similar over time. Validation study has been done in the 70s and there have not been resources for repeat studies.
• Quite a lot of discrepancies in serial number inventory.
• Several unresolved constraint violations.
• 11 suspected duplicates.

SWE-NSW

ICD-version used: 8, from 1987 9
Nonfatal case finding method: Cold pursuit
Access to death certificates: Full

Comments:

• In nonfatal events the proportion of events that are not hospitalized is quite low. The proportion of DIACAT 4 events is sizeable, however, which speaks against biased case ascertainment. Also, crude case fatality is low which speaks against incomplete case ascertainment. There are no missing fatal events, either;  STR/STR ratio and TOTRATIO are over one over all years.
• A few discrepancies with the serial number inventory.

YUG-NOS

ICD-version used: 9
Nonfatal case finding method: Hot pursuit
Access to death certificates: Full

Comments:

• Data from hospitals are collected using cold pursuit, and data outside hospitals are collected using cold pursuit.
• All of the registered nonfatal events are DIACAT 1 events. The reason is that clinical stroke diagnoses are given at discharge from the hospital at the same time as the MONICA registration is completed. Only events fulfilling MONICA criteria for definite stroke are given clinical stroke diagnosis.
• In nonfatal events, DIACAT 5 is missing. According to the MCC, all such events are included in the register, however.
• Completeness of registration is ensured by a team consisting of an epidemiologist, statistician and a community-health nurse.
• Community health centres and GPs treat also strokes but only 7% of strokes are treated by GPs exclusively, and nearly 93% are seen by a neurologist. Stroke diagnosis is given only to those fulfilling the MONICA criteria for stroke. This explains why there are no false positives.
• Somewhat higher case fatality in 1983, 1985, 1988 and 1989 and 1994 is explained by relatively small numbers and some more recurrent strokes during those years.

References to publications

1. Tunstall-Pedoe H for the WHO MONICA Project. The World Health Organization MONICA Project (Monitoring Trends and Determinants in Cardiovascular Disease): A major international collaboration. J Clin Epidemiol 1988;41:105-14.
2. WHO MONICA Project. MONICA Manual. Part IV: Event registration. Section 2: Stroke event registration data component. (November 1990). Available from: URL:http://www.ktl.fi/publications/monica/manual/part4/iv-2.htm, URN:NBN:fi-fe19981155.
3. Tunstall-Pedoe H, Kuulasmaa K, Amouyel P,Arveiler D, Rajakangas A-M, Pajak A for the WHO MONICA Project. Myocardial infarction and coronary deaths in the World Health Organisation MONICA Project. Registration procedures, event rates and case-fatality rates in 38 populations from 21 countries in four continents. Circulation 1994;90:583-612.
4. Thorvaldsen P, Asplund K, Kuulasmaa K, Rajakangas A-M, Schroll M for the WHO MONICA Project. Stroke incidence, case fatality, and mortality in the WHO MONICA Project. Stroke 1995;26: 361-367.
5. Asplund K, Bonita R, Kuulasmaa K, Rajakangas A-M, Feigin V, Schädlich H, Suzuki K, Thorvaldsen P, Tuomilehto J for the WHO MONICA Project. Multinational comparisons of stroke epidemiology. Evaluation of case ascertainment in the WHO MONICA stroke study. Stroke 1995; 26: 355-360.

References to internal MONICA documents

6. Asplund K, Feigin V, Schaedlich H, Kuulasmaa K, Rajakangas A-M, Suzuki K for the WHO MONICA Project. Stroke event registration data quality report. MONICA Memo 211A, October 1991.
7. Sarti C, Mähönen M, Rajakangas A-M, Kuulasmaa K for the WHO MONICA Project. Quality assessment of stroke event data for 1982-1990. MONICA Memo 277A, June 1994.