Description and quality assessment of MORGAM data

MORGAM logo

Follow-up data

Matti Niemelä1, Sangita Kulathinal1,2, Bijoy Joseph1,2, Olli Saarela1 and Kari Kuulasmaa1 for the MORGAM Project3

1 Department of Health Promotion and Chronic Disease Prevention, National Public Health Institute, Helsinki, Finland
2 Since January 2007 at Indic Society for Education and Development (INSEED), Nashik, India
3 See Annex for the sites and key personnel of contributing MORGAM Centres

Valid HTML 4.01!
© National Institute for Health and Welfare and the MORGAM Project investigators
Last updated: 4 July 2007
For more information, please contact Kari Kuulasmaa (firstname.lastname@thl.fi)

Contents

1. Introduction

After the baseline examination, the MORGAM cohorts were followed up for all deaths and optionally for non-fatal coronary, stroke and thromboembolic events. The follow-up procedure for each cohort is described in document "Description of MORGAM Cohorts". The purpose of this part of the Description and quality assessment of MORGAM data is to assess the follow-up procedures and the quality of the follow-up data.

The follow-up period varies between the cohorts, and the MPCs can extend the follow-up period when a longer follow-up becomes possible during the course of the study. Consequently, different publications from the Project may consider different follow-up periods, and this document will be updated after the extension of the follow-up period of a cohort. The current follow-up periods for the different study end-points can be identified in sub-sections "End of follow-up periods" below.

2. Approach to the description and quality assessment

The data will be described and the quality assessed separately for the different study end-points: deaths, coronary events, stroke events and venous thromboembolic events. The assessment of deaths will consider general aspects of the mortality follow-up, whereas any disease-specific aspects of deaths will be considered under the sections for the three disease-specific end-points. For each of the four end-point categories, we will consider:

The last two of these are most challenging.

2.1 Assessment of follow-up procedures and coverage

The assessment of follow-up procedures and coverage comprise:

The key components of good coverage of follow-up are good notification of end-point event and good notification of loss to follow-up.

By notification of events we mean the procedure through which the end-point events of the cohort members during the follow-up come to the attention of the study team of the MPC. If they do not come to their attention, the persons concerned will become misclassified as non-cases, which will bias the results of the data analyses. A good notification of events may yield possible events which, after validation, are found not to be end-point events, but it does not miss true events.

A person is lost to follow-up on a certain date (before the end of the study) if his end-point status can be ascertained until that date, but not thereafter. If loss to follow-up remains unnoticed, the person will become misclassified as a non-case even if he or she has an end-point event after the loss to follow-up, which will bias the results of the study. The percentage of loss-to-follow-up is also relevant for the quality of follow-up, but it is considered to be far less important than the availability of  good notification of loss-to follow-up, and it could anyway be assessed only for the cohorts where a good notification of loss-to-follow-up was in place. Although the percentage of loss-to-follow-up will be reported, it will not be considered as a major quality indicator.

As an additional indicator of the coverage of the follow-up, mortality in the cohorts was compared with the mortality in the general population from which the cohort was sampled. For this purpose, the official annual mortality statistics and population sizes of the MORGAM Reporting Units were collected for the follow-up periods. These data were used to estimate the number of deaths in cohorts, and the estimates were compared with the number of deaths identified in the follow-up. The details of the estimation procedure are described in Annex F.1.

Four scores were defined to summarize the quality of the follow-up procedures and coverage:

1. Event notification score:

Two definitions of the score are given based on the source of notification of the events.

I: Event notification using event registers or medical records
Event notification score = 2 if sources of notification cover the whole country or more;
1 if sources of notification cover the Reporting Unit and possibly surrounding areas but not the whole country;
0 if sources of notification cover an area less than the Reporting Unit area.

 

II. Event notification by letter or telephone or clinical event questionnaire
Event notification score = 2 if, in case the first contact fails, satisfactory attempts are made to contact the person and/or relatives and/or his/her general practitioner. If these contacts fail, the person's vital status and emigration from the area is checked satisfactorily from other sources;
1 if intermediate between 2 and 0.
0 if only one attempt is made to contact the participant, or in case of no contact, the person's vital status is not checked from other sources;

2. Loss-to-follow-up notification score:

Loss-to-follow-up notification score = 2 if good notification of loss to follow-up or Event notification score = 2;
  1 if no good notification of loss to follow-up;

3. Mortality comparison score:

Let R be the ratio of the observed (O) number of deaths in the cohort to the expected (E) number of deaths calculated from the mortality statistics of the Reporting Units. The Mortality comparison score is defined as:

Mortality comparison score = 2 if R ≥ 70%,
1 if 55% ≤ R < 70%,
0 if  R < 55%.

See Annex F.2 for the discussion on the limits 55% and 70%.

4. Coverage score:

The Coverage score summarizing the above three scores is defined as:

  Loss-to-follow-up notification score
2 1
Event notification score Event notification score
2 1 0 2 1 0
Mortality comparison score 2 2 2 1 - 1 1
1 2 1 1 - 1 0
0 1 1 1 - 0 0

2.2 Assessment of end-point diagnoses

The approach for the assessment of the diagnoses differs for the different end-point events. Therefore, the approach is described together with the assessment of each end-point below.

3. Mortality follow-up

In this section, the follow-up of deaths is considered to the extent that is covered by the data specified in the Data transfer format: follow-up data (Form 25). This part of the follow-up is compulsory for all cohorts. Issues related specifically to the follow-up of coronary, stroke and thromboembolic events and their diagnosis are considered in later sections.

3.1 Availability and distributions of data items

Here are hyperlinks to the distributions of the data items of the Data transfer format: follow-up data (Form 25):

From the point of view of data analysis, the most important ones of these data items are EXDATE, which indicates the end of mortality follow-up for each subject, and the DEATHDU, which specifies the underlying cause of death. For EXDATE, MORGAM has no option for missing data. If the date is not known exactly by the MPC, they are asked to provide their best estimate. Therefore, this item is practically always available. DEATHDU is missing very occasionally in some cohorts, and more often in:

3.2 Reasons for exit from the study

According to the MORGAM Manual the follow-up of a person continues until the earliest one of the following events occurs:

Table F.1 gives the percentage of the different reasons of exit. The reason is mostly "end of the study" or "death". Loss to follow-up is an occasional reason in the Cohorts where the MPC had notification of loss to follow-up (see Table F.3). For these, "moved away" was used in the Cohorts which were followed up using event registers or medical records, and "refusal" was used in the cohorts where the subjects were contacted periodically. Reason "other" was used in some of the cohorts for subjects, for which the record linkage for the follow-up failed (POL-WAR). In these cases the follow-up was coded to end on the day of the baseline examination. For some other cohorts it was apparently used in situations where the reason could not be specified.

3.3 End-of-follow-up period for death

Table F.2 gives the dates, reported by the MPCs, when the follow-up was ended in the case the person did not die, did not refuse from participation or was not lost to follow-up earlier. The table also gives the three most frequent exit dates in these cases. These dates are consistent with the reported end-of follow-up dates, and confirm that nearly all cohorts ended the follow-up at a fixed date. The exceptions are:

To avoid bias in the results, it is important that the follow-up period is not extended from the general upper limit for some subject because their later death becomes to the attention of the MPC. To check for this, the last two columns of Table F.2 give the number of exit dates that are later than the most frequent one, and latest exit date of the cohort. The findings suggest that the good practise was followed in all cohorts.

3.4 Follow-up procedures and coverage

The procedures used for notification of deaths and notification of loss to follow-up are given in Table F.3, the percentage of loss-to-follow-up for these cohorts is available in Table F.4. More details of these procedures are given in the Description of MORGAM Cohorts. Table F.5 reports the comparison of  the expected and observed number of deaths in the Cohorts.

The scores to summarize the quality of the follow-up procedures and coverage are shown in Table F.6.

Most RUAs used national death register for the follow-up and hence it covered the whole country. For many RUAs, notification of loss to follow-up was not required due to the high coverage of the death register. The percentage lost to follow-up was small in all Cohorts: less than 2% in nearly all cohorts and even the maximum was only 6.6%. The ratio of observed to expected deaths varies from 39% to 123%. There are large differences between the observed and expected number of deaths in some RUAs. This may be due to poor coverage of follow-up, but other potential reasons are the possibility that the Cohort is healthier than the general population. For some Cohorts the annual population mortality data were not available for the last years of follow-up, and therefore the estimated mortality may be an overestimate.

Comments on individual RUAs on the follow-up procedure and coverage:

3.5 Diagnosis of cause of death

According to the MORGAM Manual, whenever possible, final official codes should be used. If these are not available the death diagnoses should be derived from the latest available source, such as the death certificate. Table F.7, computed from the MORGAM follow-up data,  gives the percentage of the final official codes and death certificates as the source of the underlying cause of death. For the PRIME cohorts (FRA-LIL/STR/TOU and UNK-BEL), Table F.7 suggest incorrectly that the underlying cause of death was derived from death certificates. In these cohorts, the PRIME "Deaths" Medical Committee assigned the underlying cause of death using all available information from the death certificate, from the autopsy report and from the general practitioner's, hospital and emergency team's notes.

Table F.7 also gives the source of the other death diagnoses, when available. In some cases the sum of the alternatives is less than 100%, indicating that the cause of death was not available at all. For the underlying cause of death, column "NNN" of  the table on irrelevant codes of the underlying cause of death gives the number of missing data for the diagnosis. The missing data is described in section Availability and distributions of data items above.

The ICD version used for coding the death diagnoses is given in Table F.7 along with the period of using each version. ICD- 8 was in use in Denmark up to year 1993 and in Finland up to 1986. Otherwise ICD-9 was used until it was changed to ICD-10 in the late 1990s. In Augsburg, Germany and Italy, ICD-9 was used still in the 2000s. The periods of the ICD versions understandably follow the periods generally used in the country. However, in GER-AUG, the MPC had coded the deaths to ICD-9 throughout the follow-up period, using information from the death certificates, even though the official statistics in Germany changed to ICD-10 already in 1998.

The results of necropsy give valid data for the diagnostic classification of fatal events. The cause of death remains often uncertain. Table F.7 gives the percentage of deaths in the cohorts that are known to have been necropsied. The percentage varies between 0% and 56%. Necropsies are rare in many countries, but also many MPCs have not provided the data on whether or not necropsy was performed.

4. Follow-up of coronary events

This section considers the aspects of the follow-up of coronary events beyond what was already considered in the section Mortality follow-up above. The follow-up of fatal coronary events is compulsory for all cohorts, and the follow-up of non-fatal events is highly recommended. There is flexibility on the extent of the end-point categories of non-fatal events. The data specific to the follow-up of coronary events are described in two data transfer formats:

  1. Data Transfer Format: Coronary Events Inventory (Form 27). These data should be provided for every member of the cohorts;
  2. Data Transfer Format: Coronary Events (Form 22). These data should be provided for:

4.1 Types of non-fatal coronary events followed up

The extent to which the follow-up of non-fatal coronary events was done depends on each MPC. Alternative schemes providing data on non-fatal coronary events included:

The types of non-fatal coronary events followed up in each Cohort are given in Table F.8a. The main deviations from the majority of the cohorts were in:

FRA-LIL, FRA-STR, FRA-TOU and UNK-BEL did not have the category "possible MI", but this is largely compensated by their category "unstable angina". All other Cohorts followed up also possible MIs. For some of the cohorts it is not possible to separate between definite and possible MIs. Some cohorts have included all unstable angina pectoris, but the borderline between unstable MI and possible MI varies largely depending on the diagnostic criteria that were used for the cohort. In MONICA classification most cases of unstable angina are apparently included in the category  "possible MI", whereas in the PRIME classification, most of the possible MIs go to the category "unstable angina". Cardiac revascularizations were recorded for most cohorts, silent MI in Rome only and stable angina in the PRIME (France and UNK-BEL) and Scottish (UNK-EDI, UNK-GLA and UNK-SHH) cohorts. Details of the  diagnostic criteria used in the Cohorts are described in the Description of MORGAM Cohorts, and evaluated further in section Diagnosis of coronary events below. Recurrent MI was followed up systematically in DEN-GLO, FIN-ATB and the PRIME (France and UNK-BEL).

The number of events observed during the follow-up are heavily dependent on the extent of the non-fatal follow-up.

4.2 Availability and distributions of data items (Forms 22 and 27)

Here are hyperlinks to the distributions of the data items of the Data Transfer Format: Coronary Events Inventory (Form 27) and Data Transfer Format: Coronary Events (Form 22):

In FRA-LIL/STR/TOU and UNK-BEL, the main clinical diagnosis (CLIND1) was not the diagnosis from the health care provider, but it was assigned by the PRIME Medical Committee using all available information.

4.3 Reasons for exit from the follow-up of non-fatal coronary events

As the main rule, a person exits from the follow-up of non-fatal coronary events at the same time as he/she exits from the mortality follow-up. The latter is considered in section Reasons for exit from the study above. In some situations for some cohorts the exit from the follow-up of non-fatal coronary events can take place earlier than the exit from mortality follow-up:

Table F.9 gives the number of exits occurred due to above reasons. In most cases, the reason of exit (EXREASC) is coded as irrelevant, which means that the end of follow-up of non-fatal coronary events is the same as that of mortality follow-up. Cohorts where the follow-up of non-fatal coronary events ends earlier than for deaths for some of the subjects are:

We also checked that whenever the Description of MORGAM Cohorts suggests that the follow-up of non-fatal coronary events ends earlier than for deaths, this is consistent with Table F.9.

4.4 End-of-follow-up period for coronary events

In most cohorts, the date of exit from the follow-up of non-fatal coronary events is always the same as from the mortality follow-up for every subject. This can be seen as blank rows in Table F.10, which shows the dates of exit from the follow-up of non-fatal coronary events. The end of mortality follow-up period was assessed in section End-of-follow-up period for death above.

The end of follow-up of non-fatal coronary events differs from that of mortality follow-up only for the cohorts specified in section Reasons for exit from the follow-up of non-fatal coronary events above. For these cohorts, it is relevant to check that:

These checks show no discrepancies.

4.5 Follow-up procedures and coverage

The procedures used for notification of deaths was given in the section on mortality follow-up procedures and coverage above. The procedures used for notification of non-fatal coronary events are given in Table F.11a. More details of the procedures are given in the Description of MORGAM Cohorts. Table F.11b shows the percentage of different sources of notification of fatal and non-fatal coronary events as indicated in the data. In the cases where the event was identified through several sources, it is assigned to the left-most relevant column in Table F.11b. Most MPCs used linkage to a national or regional hospital discharge register for the follow-up or/and linkage to a local or regional coronary event register maintained by the MPC. For four cohorts, there is inconsistency between Table F.11a and Table F.11b:

The procedures used for notification of loss to follow-up of non-fatal coronary events are also given in Table F.11a, and the percentage of loss-to-follow-up of fatal and non-fatal coronary events is available in Table F.12. In general, as expected, the same data sources were used for notification of loss to follow-up of non-fatal events as for deaths (cf. Table F.3). Compared to the mortality follow-up, the percentage of loss to follow-up increased for the cohorts where the geographical coverage of non-fatal notification was smaller (GER-AUG, ITA-BRI/PAM and SWE-NSW). Nevertheless, the percentage remained reasonably low (2-4%) in these cohorts. Note that reason "other" of exit from the follow-up of non-fatal coronary events was not considered as loss to follow-up because this code was used when the person reached the possible upper age limit of the follow-up of non-fatal events.

Table F.13 reports the comparison of  the expected and observed number of coronary deaths in the Cohorts. The ratio of observed to expected numbers are mostly similar to the numbers for all deaths in Table F.5 with, as expected, more variation when the numbers are small. There are clear systematic differences in:

According to the MORGAM data requirements, a coronary event record should be provided for each death with CHD, sudden death, unattended death or cardiac arrest in the death certificate or final death diagnoses. This is checked in Section Record inventory and routine data checking.

The scores (see Assessment of follow-up procedures and coverage above) to summarize the quality of the follow-up procedures and coverage of fatal and non-fatal coronary events are shown in Table F.14. The coverage score is less than two for the following cohorts:

4.6 Diagnosis of coronary events

We will focus here on the diagnosis of acute coronary events: myocardial infarctions, unstable angina pectoris and other coronary deaths.  The diagnosis is made difficult by (a) the fact that the severity of a non-fatal coronary events is very heterogeneous, where the diagnosis of the milder forms is often uncertain or they may not come to medical attention at all, and  (b) for fatal events the diagnostic information is often very parse. Most of the coronary deaths take place without medical presence, and in most countries necropsy is rarely performed to the uncertain cases. Because of these uncertainties, MORGAM has defined several diagnostic categories for coronary events, reflecting the characteristics of the events and the certainty of the diagnostic information.

The initial recommendation in MORGAM was that coronary events should be classified using the WHO MONICA procedure (see  MONICA Manual, Part IV, Section 1, Subsections 2.2 and 2.3). However, it was recognized that some of the cohorts had already diagnosed their coronary events using some other standardized procedure. Also the use of clinical diagnoses was considered acceptable provided that validation studies are available to support the good quality of the clinical diagnoses in the region. During the course of the study, strict MONICA validation has become partially outdated due to the replacement of cardiac enzymes with more specific and sensitive non-enzymatic markers as routine diagnostic tests.

Table F.15 lists the source of validation for acute coronary events by RUA. Most of the MPCs used MONICA validation, but most of them only for a part of the events. More details of the diagnostic procedures are given in the Description of MORGAM Cohorts. Some centres used MONICA validation for events found in a local or regional coronary event register maintained by the MPC, and used e.g. a clinical diagnosis for events that took place outside the registration area or at age higher than the upper age limit of the event registration. Table F.16 gives the percentages of the source of validation used, as indicated in the event data. For one cohort, there is inconsistency between Table F.15 and Table F.16:

Perhaps a dominating aspect concerning the diagnosis of fatal events is the availability of information on which the diagnosis can be based. When MONICA diagnostic criteria are used, this is reflected in the percentage of the diagnostic category "unclassifiable". A high percentage of "unclassifiable" indicates much uncertainty in the diagnosis, and there is nothing that the MPC can do to remove this uncertainty. The percentage of unclassifiable events in the MORGAM cohorts can be used as an indicator only when MONICA classification was used for nearly all events. However, even when MONICA classification was not used extensively in for MORGAM, the percentage of unclassifiable events in the MONICA event registration that took place in the same population can often be used as an indicator (see  Table 7 of Quality Assessment of Coronary Event Registration Data in the WHO MONICA Project).

Furthermore, when the diagnosis of fatal events was based on the official death diagnoses or death certificates, the comparison of the official mortality with the MONICA validated mortality is a useful indicator of the quality of the routine diagnoses in the populations which took part in the MONICA Project. Such a comparison can be seen in Figure 5 of reference [1].

A Diagnosis Score based on the above aspects was defined for each cohort as:

Diagnosis score = 2 if no concern in the diagnosis of coronary events was identifiable
1.5 if small concern was identifiable, such as:
  • routine diagnosis was used for remarkable percentage of the events but there are validation studies suggesting the diagnosis is reasonable for epidemiologic studies, or
  • MONICA validation was used for nearly all events, but there is a moderate percentage of unclassifiable deaths.
1 if small concern was identifiable on several quality aspects or there was moderate concern on some aspect, such as a remarkable percentage of unclassifiable deaths in the population.
0.5 if there is moderate concern about various aspects of the diagnosis.
0 if the diagnosis of coronary events is considered clearly unsuitable for MORGAM.

The Diagnosis scores are given in Table F.14. The scoring gives emphasis on a wide definition of a coronary event, which includes both definite and possible non-fatal MIs. This principle is consistent with the finding [ref?] that persons with even mild events have a high risk of developing a more severe coronary event later. This principle differs from that adopted in the WHO MONICA Project, where only definite non-fatal MIs were considered in the main analyses. In MONICA the objective was to use a reliable definition for assessing time trends in the population and comparisons between populations, whereas in MORGAM it is more important to classify each event into coronary/non-coronary with as few miss-classifications as possible. There may be situations in MORGAM where a more narrow definition will be used. In such cases the scoring of Table F.14 is not necessarily good.

The findings on which the scores of Table F.14 were based for each RUA are given here:

4.7 Coronary Event Score

To summarize the reliability of the end-point data, a simple Coronary Event Score was defined as the mean of the Coverage Score and the Diagnosis Score. The Coronary Event Score for each Cohort is shown in Table F.14.

5. Follow-up of stroke events

This section considers the aspects of the follow-up of stroke events beyond what was already considered in the section Mortality follow-up above. The follow-up of fatal strokes is compulsory for all cohorts, and the follow-up of non-fatal events is highly recommended. The data specific to the follow-up of stroke events are described in two data transfer formats:

  1. Data Transfer Format: Stroke Events Inventory (Form 28). These data should be provided for every member of the cohorts;
  2. Data Transfer Format: Stroke Events (Form 23). These data should be provided for:

5.1 Types of non-fatal stroke events followed up

The types of non-fatal stroke events followed up in each Cohort are given in Table F.8b:

Details of diagnostic criteria used in the Cohorts are described in the Description of MORGAM Cohorts and evaluated further in section Diagnosis of stroke events below.

5.2 Availability and distributions of data items (Forms 23 and 28)

Here are hyperlinks to the distributions of the data items of the Data Transfer Format: Stroke Events Inventory (Form 28) and Data Transfer Format: Stroke Events (Form 23):

In DEN-GLO, most of SESOUR* and DGSOUR* items were coded as 9 (insufficient data) for all members of the Cohort. There is always code 1 (yes) for some of the sources. Apparently code 9 in this cohort  means that the source was not used, but there is no confirmation of this from the MPC.
In FRA-LIL/STR/TOU and UNK-BEL, the main clinical diagnosis (CLIND1) was not the diagnosis from the health care provider, but it was assigned by the PRIME Medical Committee using all available information.

5.3 Reasons for exit from the follow-up of non-fatal stroke events

As the main rule, a person exits from the follow-up of non-fatal stroke events at the same time as he/she exits from the mortality follow-up. The latter is considered in section Reasons for exit from the study above. In some situations for some cohorts the exit from the follow-up of non-fatal stroke events can take place earlier than the exit from mortality follow-up:

Table F.17 gives the number of exits occurred due to above reasons. In most cases, the reason of exit (EXREASS) is coded as irrelevant, which means that the end of follow-up of non-fatal stroke events is the same as that of mortality follow-up. Cohorts where the follow-up of non-fatal stroke events ends earlier than for deaths for some of the subjects are:

We also checked that whenever the Description of MORGAM Cohorts suggests that the follow-up of non-fatal stroke events ends earlier than for deaths, this is consistent with Table F.17.

5.4 End-of-follow-up period for stroke events

In most cohorts, the date of exit from the follow-up of non-fatal stroke events is always the same as from the mortality follow-up for every subject. This can be seen as blank rows in Table F.18, which shows the dates of exit from the follow-up of non-fatal stroke events. The end of mortality follow-up period was assessed in section End-of-follow-up period for death above.

The end of follow-up of non-fatal stroke events differs from that of mortality follow-up only for the cohorts specified in section Reasons for exit from the follow-up of non-fatal stroke events above. For these cohorts, it is relevant to check that:

These checks show no discrepancies.

5.5 Follow-up procedures and coverage

The procedures used for notification of deaths was given in the section on mortality follow-up procedures and coverage above. The procedures used for notification of non-fatal stroke events are given in Table F.19a. More details of the procedures are given in the Description of MORGAM Cohorts. Table F.19b shows the percentage of different sources of notification of fatal and non-fatal stroke events as indicated in the data. In the cases where the event was identified through several sources, it is assigned to the left-most relevant column in Table F.19b. Most MPCs used linkage to a national or regional hospital discharge register for the follow-up or/and linkage to a local or regional stroke event register maintained by the MPC. There is no inconsistencies between Table F.19a and Table F.19b.

The procedures used for notification of loss to follow-up of non-fatal stroke events are also given in Table F.19a, and the percentage of loss-to-follow-up of fatal and non-fatal stroke events is available in Table F.20. Compared to the mortality follow-up, the percentage of loss to follow-up increased for cohorts where the geographical coverage of non-fatal notification was smaller (ITA-BRI/PAM and SWE-NSW). The percentage remained reasonable low (2-4%) in these cohorts. Note that reason "other" of exit from the follow-up of non-fatal stroke events was not considered as loss to follow-up because this code was used when the person reached the possible upper age limit of the follow-up of non-fatal events.

Table F.21 reports the comparison of the expected and observed number of stroke deaths in the Cohorts. The ratio of observed to expected numbers are mostly similar to the numbers for all deaths in Table F.5 with, as expected, more variation when the numbers are small. Like for coronary deaths there are clear systematic difference in:

According to the MORGAM data requirements, a stroke event form should be provided for each death with stroke in the death certificate or final death diagnoses. This is checked in Section Record inventory and routine data checking.

The scores (see Assessment of follow-up procedures and coverage above) to summarize the quality of the follow-up procedures and coverage of fatal and non-fatal stroke events are shown in Table F.22. The coverage score is less than two for the following cohorts:

It is striking that the ratio of stroke mortality in the cohort to stroke mortality in the population  (and hence the stroke mortality comparison score) is generally often much lower or higher than the ratio for all deaths (see Table F.21 and Table F.5). It is unclear to which extent this is due to bad coverage of stroke. A potential explanation for a general tendency for lower stroke mortality ratios is that persons with a high risk of stroke are likely to be non-respondents in the baseline survey. However, the number of stroke events is generally low, suggesting that randomness explains much of the variation.

5.6 Diagnosis of stroke events

Stroke is diagnosed in MORGAM in two levels. The first level is based on clinical signs, and determines whether or not the suspect event was a stroke. The second level determines the subtype of stroke and requires special diagnostic procedures. These two levels of diagnosis are considered here separately.

Stroke or not a stroke?

The recommended diagnostic classification of stroke events in MORGAM follows the WHO MONICA procedure (see MONICA Manual, Part IV, Section 2, Subsections 2.2):

Stroke is defined as rapidly developed clinical signs of focal (or, in selected instances, global) disturbance of cerebral function lasting more than 24 hours (unless interrupted by surgery or death), with no apparent cause other than a vascular origin: it includes patients presenting clinical signs and symptoms suggestive of subarachnoid haemorrhage, intracerebral haemorrhage or cerebral infarction. It does not include transient cerebral ischaemia or stroke events in cases of blood disease (e.g. leukemia, polycythaemia vera), brain tumour or brain metastases. Secondary stroke caused by trauma should also be excluded.

For further details, see the specific instructions for item DGNCAT of Form 23 in the MORGAM Manual.

Table F.23 lists the source of validation for stroke events used in each RUA. More details of the diagnostic procedures are given in the Description of MORGAM Cohorts. Most of the MPCs used the MONICA validation. Some centres used MONICA validation for events found in a local or regional stroke event register maintained by the MPC, and used e.g. the clinical diagnosis for events that took place outside the registration area or at age higher than the upper age limit of the event registration. Table F.24 gives the percentages of source of validation used. There is inconsistency between Table F.23 and Table F.24 for:

When MONICA diagnostic criteria are used, the percentage of diagnostic category "unclassifiable" reflects the availability of information on which the diagnosis was based. A high percentage of "unclassifiable" indicates uncertainty in the diagnosis.

When the diagnostic classification is based on routine diagnosis and this indicates a stroke, then the MORGAM diagnostic category always becomes "unclassifiable". There is some variation between the MPCs in whether or not to consider other cerebrovascular disease (437 of ICD-9, I67 of ICD-10) or late effects of cerebrovascular disease (438 of ICD-9, I69 of ICD-10) as a stroke (see Description of MORGAM Cohorts). If the population took part in the MONICA stroke registration, comparison of the official mortality with the MONICA validated mortality gives some indication of the quality of the routine diagnoses. Such a comparison can be seen in Figure 2 of reference [5].

A Diagnosis Score was defined for each cohort as:

Diagnosis score = 2 if no concern in the diagnosis of stroke events was identifiable
1.5 if small concern was identifiable, such as:
  • routine diagnosis was used for remarkable percentage of the events but there are validation studies suggesting the diagnosis is reasonable for epidemiologic studies, or
  • MONICA validation was used for nearly all events, but there is a moderate percentage of unclassifiable deaths.
1 if small concern was identifiable on several quality aspects or there was moderate concern on some aspect, such as a remarkable percentage of unclassifiable deaths in the population.
0.5 if there is moderate concern about various aspects of the diagnosis.
0 if the diagnosis of stroke events is considered clearly unsuitable for MORGAM.

The findings on which the scores of Table F.22 were based for each RUA are given here:

Subtype of stroke

Stroke subtyping is essential for phenotypic characterization in the MORGAM genetic component. MORGAM data identifies the three main subtypes, which are subarachnoid haemorrhage (SAH), intracerebral haemorrhage (ICH) and cerebral infarction (CI), and has a separate data item (OTYPE) for other subtypes of stroke. The criteria of the subtypes of stroke events presume that diagnostic procedures or necropsy findings provide an unambiguous diagnostic information. In addition to clinical and necropsy findings, diagnostic procedures taken into consideration are computerized axial tomography (CT), magnetic resonance imaging (MRI) and cereprospinal fluid examination (for SAH and ICH) (see the Data Transfer Format: Stroke Events of  MORGAM Manual). If routine diagnosis is used and no further information is available, all subtypes of stroke are coded as "insufficient data".

Compared to MONICA, the criteria for subtypes of stroke were updated for MORGAM, and some items for further specification of type of stroke were added.  Further classification of cerebral infarction to its subtypes is still impossible in MORGAM because data on the location and the size of the infarction are not collected.

The assessment of the diagnosis of stroke subtypes will be completed later.

5.7 Stroke summary scores

Stroke Event Score

To summarize the reliability of the end-point data, a simple Stroke Event Score was defined for general definition of stroke as the mean of the Coverage Score and the Diagnosis Score. The Stroke Event Score for each Cohort is shown in Table F.22.

6. Follow-up of thromboembolic events

This section considers the aspects of the follow-up of thromboembolic events beyond what was already considered in the section Mortality follow-up above. The follow-up for non-fatal venous thromboembolic events is optional for the cohorts which take part in the genetic substudy, and should be done only if the coverage of non-fatal hospitalized events is good. The data specific to the follow-up of thromboembolic events are specified in two data transfer formats:

The relevant ICD codes are

6.1 Types of non-fatal thromboembolic events followed up

Thromboembolic events were followed up in DEN-GLO, FIN-ATB and FIN-EAS/WES as listed in Table 8b. Recurrent thromboembolic events were followed up systematically in DEN-GLO and FIN-ATB. Details of diagnostic criteria used in the Cohorts are described in the Description of MORGAM Cohorts and evaluated further in section End-point diagnosis below.

6.2 Availability and distributions of data items (Forms 26 and 29)

Here are hyperlinks to the distributions of the data items of the Data Transfer Format: Venous Thromboembolic Events Inventory (Form 29) and the Data Transfer Format: Venous Thromboembolic Events (Form 26):

6.3 Reasons for exit from the follow-up of non-fatal thromboembolic events

Table F.25 gives the number of exits due to different reasons by cohort for each RUA. The reason is always coded as "irrelevant", which means that the end of follow-up of non-fatal thromboembolic events was the same as that of mortality follow-up.

6.4 End-of-follow-up period for thromboembolic events

Table F.2 and Table F.26 give the end-of-follow-up dates for fatal and non-fatal thromboembolic events by RUA, respectively. Table F.26 also gives the percentages of three most frequently occurring dates among such individuals. The end of follow-up of non-fatal thromboembolic events is the same as that of mortality follow-up in all cohorts.

6.5 Follow-up procedures and coverage

Table F.3 and Table F.27a give the source of notification used for the follow-up of fatal and non-fatal thromboembolic events by RUA. Table F.27a also gives the procedures used for the notification of loss to follow-up from the follow-up of thromboembolic events. Table F.27b shows the percentage of different sources of notification of fatal and non-fatal thromboembolic events as indicated in the data. In the cases where the event was identified through several sources, it is assigned to the left-most relevant column in Table F.27b. The percentages of loss to follow-up by consecutive three year period for non-fatal thromboembolic events are given in Table F.28. As we do not have available the numbers of thromboembolic deaths in the population,  we are unable to calculate the Mortality comparison score and hence the Coverage score for thromboembolic events.

The follow-up procedures for thromboembolic events are identical with the follow-up procedures for coronary and stroke events in the MPCs that have provided thromboembolic events. The coverage score is high (2) for coronary and stroke deaths in all Cohorts except in Cohort 02 of DEN-GLO.

6.6 Diagnosis of thromboembolic events

Table F.29 lists the source of validation for thromboembolic events by RUA. Table F.30 gives the percentages of source of validation used.

Table F.30 gives the ICD-version used for codes of clinical diagnoses by cohort for each RUA.

The diagnosis of thromboembolic events is based on routine diagnosis in all Cohorts. There are no validation studies available assessing the suitability of the routine diagnosis of thromboembolic events for epidemiologic studies.

References

  1. 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 Organization MONICA Project. Registration procedures, event rates and case fatality in 38 populations from 21 countries in 4 continents. Circulation. 1994;90:583-612.
  2. Madsen M, Davidsen M, Rasmussen S, Abildstrom SZ, Osler M. The validity of the diagnosis of acute myocardial infarction in routine statistics: a comparison of mortality and hospital discharge data with the Danish MONICA registry. J Clin Epidemiol. 2003 Feb;56(2):124-30.
  3. Rapola JM, Virtamo J, Korhonen P, Haapakoski J, Hartman AM, Edwards BK, Heinonen OP. Validity of diagnoses of major coronary events in national registers of hospital diagnoses and deaths in Finland. Eur J Epidemiol. 1997;13(2):133-8.
  4. Pajunen P, Koukkunen H, Ketonen M, Jerkkola T, Immonen-Räihä P, Kärjä-Koskenkari P, Mähönen M, Niemelä M, Kuulasmaa K, Palomäki P, Mustonen J, Lehtonen A, Arstila M, Vuorenmaa T, Lehto S, Miettinen H, Torppa J, Tuomilehto J, Kesäniemi YA, Pyörälä K, Salomaa V. The validity of the Finnish hospital discharge register and causes of death register data on coronary heart disease. Eur J Cardiovasc Prev Rehabil. 2005;12:132-137.
  5. 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.
  6. Leppala JM, Virtamo J, Heinonen OP. Validation of stroke diagnosis in the National Hospital Discharge Register and the Register of Causes of Death in Finland. Eur J Epidemiol. 1999;15(2):155-60.
  7. Tolonen H, Salomaa V, Torppa J, Sivenius J, Inmmonen-Räihä P, Lehtonen A and for the FINSTROKE register. The Validation of the Finnish Discharge Register and Causes of Death Register data on stroke diagnoses. Eur J Cardiovasc Prev Rehabil. 2006; in press.

Updates to this document

Date Update
2007-07-04 First published version.