Multinational Comparisons of Stroke Epidemiology
Evaluation of Case Ascertainment in the WHO MONICA Stroke Study
Background and Purpose As part of the WHO MONICA Project (World Health Organization Monitoring Trends and Determinants in Cardiovascular Disease), mortality and incidence rates of acute stroke in 14 centers covering 21 populations from 11 countries were compared.
Methods In this report, coverage and quality of the MONICA stroke registers were evaluated on five key indicators using data submitted to the MONICA Data Center.
Results A low ratio of MONICA stroke register to routine statistics of stroke mortality and a low proportion of nonfatal out-of-hospital events were the most common biases; they indicate that identifications of fatal cases and/or case finding of nonfatal events occurring outside the hospital were inadequate in many MONICA centers. In 10 populations, the data quality analyses suggested that clarification of possible biases would be needed before these populations can be included in a comparative study. Data from the remaining 11 populations meet the data quality standards for multinational comparisons with respect to case ascertainment.
Conclusions These results show that multinational comparisons of stroke incidence involve considerable problems in developing and maintaining appropriate standards of data quality. However, after considerable efforts to ensure quality, comparisons of stroke data within the MONICA Project are possible among a large number of the MONICA populations. Our observations also indicate that results from multinational comparisons of stroke mortality based on routine statistics must be interpreted with caution.
Most comparisons of stroke epidemiology among different populations are based on routine mortality data. Cause-specific mortality statistics for stroke are problematic because they may include serious biases and they are not validated in a uniform manner. Moreover, they provide only a limited picture of the occurrence of stroke in the population. A large proportion of stroke victims are left with permanent disability, and the human and economic consequences of stroke extend far beyond what emerges from routine mortality data. Stroke incidence and prevalence data better reflect the impact of stroke in the community. However, population-based monitoring of stroke events involves a number of problems, the most important being the completeness of case ascertainment; this is especially true for population areas where many stroke patients are not diagnosed or treated in the hospital. An extensive review of stroke incidence studies showed that few provide reliable data.1 Indeed, it has been claimed that much of the reported variation in stroke mortality and stroke incidence rates between populations may be attributed to methodological problems.2
In the WHO MONICA Project (World Health Organization Monitoring Trends and Determinants in Cardiovascular Disease), secular trends in myocardial infarction and stroke rates are being measured in relation to changes in population levels of cardiovascular risk factors such as hypercholesterolemia, hypertension, and cigarette smoking.3 In the years 1982 through 1986, 19 of the 39 MONICA Collaborating Centers (MCCs) entered the stroke component of the project. Some of these are subdivided into two or several populations. Stroke data are available for 21 of the MONICA populations. The aim of the stroke component of the MONICA study is to monitor both fatal and nonfatal stroke events in each population over a 10-year period and to relate changes in event rates with changes in cardiovascular risk factors in the populations.
The study protocol (MONICA Manual) emphasizes the need for uniformity of diagnostic criteria and data collection in the registration of stroke events in the population.4 However, registration is performed in different cultural, administrative, economic, and health service settings. Local health service routines, as well as human and financial resources, vary considerably among the MCCs. It is therefore inevitable that there are variations in the application of the study protocol. All data submitted to the MONICA Data Center are subjected to checks for consistency and completeness. To ascertain uniformity in coding, sets of test cases have been distributed to all MONICA stroke centers, and feedback has been provided. The test case histories include relevant clinical information on actual patients, and they are collected from all participating centers.4 In this report, the quality of the MONICA Project stroke data in the first years of registration has been assessed with special emphasis on case ascertainment for hospitalized and nonhospitalized stroke cases, fatal and nonfatal.
Subjects and Methods
The 21 populations considered in this report are described in Table 1⇓. A few other MONICA centers began stroke registration but they later withdrew, mainly because of lack of funding and difficulties in keeping up with requirements for complete case ascertainment, including access to routine death certificates.
The age range covered by the stroke registers varies among the MCCs. For this report, we analyzed data for the 35- to 64-year range because this age group was covered by all centers and is a core requirement for the MONICA Project. We used stroke data for full years from the start of continuous registration of stroke events (usually 1982 through 1985, Table 1⇑) through the end of 1987 (up to 1986 in one population).
For each event identified by the local stroke register, a core data record form is submitted to the MONICA Data Center in Helsinki, Finland. Suspected stroke events that are registered include all cases with a clinical diagnosis of acute stroke (International Classification of Diseases, 9th Revision [ICD-9] codes 430 through 434 and 436) after treatment in the hospital or on an outpatient basis and all fatal cases in the study population with ICD-9 codes 430 through 436 given on the death certificate. Each center should also perform a preliminary screening of cases with selected nonstroke diagnoses to identify commonly occurring types of misclassification and include such diagnoses in those regularly screened. On the basis of all the available information, the local MONICA team then classifies the event as “definite stroke,” “definite stroke associated with definite myocardial infarction,” “not stroke,” or “insufficient data” by the MONICA criteria.4 Stroke subtype is coded based on findings at brain imaging and autopsy.4 Core data for all registered cases are submitted to the MONICA Data Center irrespective of the final diagnostic category.
Stroke events fulfilling the following minimum criteria are included in the present data quality analyses: onset occurring within the specified years, onset occurring between 35 and 64 years of age, clinical and/or death certificate diagnosis of ICD code 430-434 and 436, and clinical and/or death certificate diagnosis other than ICD code 430-434 and 436 but classified by the local MONICA team as acute stroke on the basis of clinical presentation.
Procedures for case ascertainment, verification of stroke diagnosis, and classification of events into different stroke subtypes depend on whether the outcome is fatal and where the patient is managed. Therefore, within each MCC, data quality may be different for fatal in-hospital, fatal out-of-hospital, nonfatal in-hospital, and nonfatal out-of-hospital cases. Consequently, where appropriate, stratification according to survival and place of management has been applied.
Five key indicators of data quality were identified: the ratio of fatal cases in the MONICA register to stroke deaths in routine mortality statistics, the proportion of fatal cases that are not hospitalized, the 28-day case fatality, the proportion of nonfatal cases that are not hospitalized, and the proportion of fatal cases autopsied or examined by a physician before death.
(I) Fatal events in the MONICA Project stroke study were defined as death within the first 28 days after onset. For each year, the number of fatal cases in the MONICA register was compared with the number of deaths from stroke in the routine mortality statistics. The ratio between the number of deaths identified by the MONICA register and the number of deaths according to routine statistics should approach or exceed 1.0 if all fatal stroke events were identified. A ratio a little lower than 1.0 can be explained by the fact that data from the routine mortality statistics are based on a wider range of cerebrovascular diagnoses than the MONICA registers (ICD codes 430-438 versus ICD codes 430-434 and 436). In addition, some stroke deaths occurring more than 28 days after onset may be included as nonfatal cases in the MONICA register but may appear as fatal acute stroke (ICD 430-434 and 436) in the routine mortality statistics. An arbitrary stroke register–to–routine mortality ratio of less than 0.75 was used as an indicator of possible underreporting of stroke deaths in the register.
(II) The proportion of fatal cases that occurred outside the hospital in relation to all stroke deaths was used to estimate the completeness of data on fatal out-of-hospital events, since not all stroke patients reach the hospital alive. This proportion varies from country to country, depending on the local practice for stroke care and the facilities available. In some countries (eg, Denmark, Finland, and Sweden), traditionally a high proportion of stroke patients are admitted to the hospital.5 6 7 8 9 Assuming that the natural history of acute stroke does not differ drastically among the MONICA populations, the proportion of fatal stroke cases not admitted to the hospital is unlikely to be <10%.
(III) The 28-day case fatality rate was used as an indicator of the completeness of registration of nonfatal stroke cases. There is no generally accepted medical or surgical treatment that changes the natural course of acute stroke itself, although death due to secondary complications occurring after the first week (eg, pulmonary embolism, cardiac events, and bronchopneumonia) may be prevented. Incomplete coverage of nonfatal events will lead to an overestimation of case fatality. Therefore, very high case fatality may indicate inadequate ascertainment of nonfatal events.
(IV) A very low proportion of surviving stroke patients cared for outside the hospital is likely to reflect less-than-optimal case ascertainment. In general, this proportion was used to estimate the completeness of nonfatal, nonhospitalized stroke events. In recognition that in some countries all patients with suspected stroke are sent to the hospital for diagnostic evaluation and treatment, details of this practice were required for populations with few nonfatal cases receiving care outside the hospital.
(V) In the MONICA stroke study, the decision of whether a suspected nonfatal case should be classified as a stroke is based entirely on the clinical presentation.4 For fatal events, necropsy findings are accepted if the clinical presentation is unknown. In general, a physician should examine the patient to verify focal neurological symptoms and/or signs of presumed cerebrovascular origin. The proportion of fatal cases examined by a physician before death or subjected to autopsy was used to estimate the accuracy of the assignment of diagnostic category (definite stroke, not stroke, or insufficient data) in fatal cases.
To explore how the MCCs performed case ascertainment and diagnostic assessment of suspected stroke events, a questionnaire was distributed to all centers. After a preliminary assessment of data quality on the basis of the information provided, MCCs, where relevant, were asked to supply further information on steps taken to validate data that did not comply with the quality criteria. For each key indicator, the current status of data quality and how the data could be used for between-population comparisons was assigned a score as follows: (A) requires no further validation, (B) requires some validation, and (C) requires extensive validation.
All events in the MONICA stroke registers are classified by place of management of the stroke event (in-hospital or out-of-hospital) and outcome (fatal or nonfatal at 28 days after onset). The proportion of cases with missing data regarding place of management exceeded 1% in 4 of the 21 populations. Information permitting classification of survival at 28 days after onset was available for virtually all (>99%) of the events, except in one center where problems with confidentiality made it difficult to perform follow-up. In all populations, it was possible to accurately classify stroke cases by place of management and survival in 94% or more of the events.
The Figure⇓ shows the ratio of the number of fatal stroke events in the MONICA registers to the number of stroke deaths in routine mortality statistics. Events with unknown survival status at 28 days in the MONICA registers were excluded from the analysis. The ratio was less than 1.0 in 10 of the 21 populations. A MONICA stroke register–to–routine mortality statistics ratio below 0.75 was found in only 1 population; this could not be readily explained other than by underreporting of fatal cases in the local MONICA registers.
For all populations, death certificates were used to identify stroke deaths occurring outside the hospital. Additional sources (such as necropsy reports, medicolegal information, information from general practitioners, emergency services, and/or obituaries) were used in 20 of the 21 populations. Out-of-hospital fatal strokes as a proportion of all fatal strokes varied from 0% to 37% (calculated from data given in Table 2⇓). In two MONICA populations, fewer than 10% of all stroke deaths were reported as occurring outside the hospital, suggesting underreporting of such patients.
The proportion of nonfatal, hospitalized stroke events of all stroke events varied from 54% to 85% across the populations (Table 2⇑). The proportion of nonfatal stroke cases diagnosed and treated out of the hospital varied between 0% and 16% and was 5% or less in 13 populations. In 4 populations, the proportion of nonfatal out-of-hospital stroke events was less than 1%. Case fatality at 28 days ranged from 12% to 47% (Table 2⇑). Case fatality above 40% was reported in 2 populations, suggesting that not all nonfatal strokes were included in the local stroke register. The proportion of nonfatal cases examined by a neurologist or physician was almost 100% in all populations. For fatal cases, the proportion examined by a physician or subjected to autopsy ranged from 31% to 100%.
Four of the five quality assessment criteria and the description of the scoring system for each indicator are presented in Table 3⇓. These criteria are not direct measurements of performance. Fulfillment of a criterion (A) does not exclude the possibility of a bias. Similarly, not fulfilling a criterion (B or C) may only represent a value deviating from the majority of the populations; validation may demonstrate that there is no bias. The quality assessment suggested that at least some validation of case ascertainment was required in 17 of the 21 populations. As indicated in Table 3⇓, many MONICA centers have actually evaluated their aberrant data and found them to be accurate. Thus, data quality was considered to be adequate for 11 populations. For 7 populations, some further validation was needed, and for 3 populations there were such problems with data quality that inclusion of these populations in multinational comparisons on the basis of case ascertainment was considered inappropriate.
This study reports for the first time an assessment of the quality of data, in particular the completeness of case ascertainment, in a large number of populations being investigated for the occurrence of stroke. Our findings show that in the MONICA Project stroke study, data quality was adequate (or deviant data were found to be correct by local validation) for 11 of the 21 MONICA populations. Direct multinational comparisons of the occurrence of stroke is possible among these populations. In another 7 populations, further validation is needed before their data are useful for multinational comparisons. In 3 populations, extensive validation is required. Variation among the MONICA centers with respect to local applications, healthcare systems, and resources has a major impact on the nature and quality of stroke register data.
A common problem is a low number of fatal cases in the MONICA stroke register compared with the number of stroke deaths in routine mortality statistics. The instructions in the MONICA Manual establish that all cases with a clinical diagnosis of stroke must be recorded. If an event does not fulfill the MONICA criteria for acute stroke, it is kept in the local register but recorded as “not stroke.” If this procedure is not followed and the number of fatal events not fulfilling the MONICA criteria is substantial, a low register–to–routine statistics ratio results. The most likely explanation for the low ratio of register to routine statistics is therefore that not all fatal stroke cases have been identified by the stroke register. Previous investigators have pointed out that the time from onset to death as selected to define case fatality has a marked influence on fatality rates.10 A 28-day limit for defining events is used in the MONICA Project. It is possible that some patients dying from the sequelae of stroke (infection, pulmonary embolism, cardiac events, etc) later than 28 days after onset have been registered as nonfatal cases in the MONICA register, while acute stroke may appear as the underlying cause of death on the death certificate.
For many MONICA populations, there is good agreement between routine stroke mortality statistics and fatal cases in the stroke register. This is more likely a result of “false-positive” and “false-negative” stroke cases counterbalancing each other in the routine mortality statistics than of identical coding in routine statistics and the MONICA register. In detailed validation studies performed in Swedish9 and Finnish11 MONICA populations, 10% to 13% false-positive and 14% to 17% false-negative cases were found in the routine vital statistics. In Poland, the death certificate diagnoses were found to agree with the final diagnoses in the MONICA stroke register in only 64% of the stroke cases.12 As reported in the accompanying article (this issue of Stroke), official mortality statistics are in good agreement with mortality in the MONICA registers in some countries in western and northern Europe, whereas the official statistics greatly underestimate or overestimate stroke mortality in many other areas. A substantial disagreement between antemortem and postmortem diagnoses has been reported in several countries,13 14 15 16 17 and a decline in autopsy rate from previously high levels is a possible threat to the quality of stroke registers.16
The MONICA stroke study,4 like many other population-based stroke registers1 5 6 7 16 17 18 19 20 and the previous WHO stroke registers,21 accepts only documented focal neurological symptoms or signs lasting for at least 24 hours. Routine hospital discharge diagnoses as such are not acceptable as a basis for estimating stroke incidence rates. The proportion of “false-positive” stroke diagnoses at discharge may be as high as one third of all diagnoses of stroke.9 Diffuse neurological or other symptoms (dizziness, sudden general weakness, blurring of vision, etc) often are labeled as cerebrovascular events. A diagnosis of stroke is often lacking in the medical records of patients in whom the neurological disease is but one of many serious illnesses, especially after a very long hospital stay or when myocardial infarction and stroke occur at the same time. Therefore, it is essential that an independent evaluation using strict entry criteria is performed before giving an event a stroke diagnosis in a community-based stroke register.
Our assessment indicates that, for the majority of populations covered by the stroke component of the MONICA study, data on nonfatal hospitalized cases of stroke are of satisfactory quality. Nonfatal hospitalized cases as a proportion of all strokes varied considerably, from 54% to 87%. This suggests a systematic ascertainment bias in some populations. More detailed explorations of one MONICA stroke register showed that 4% of stroke cases are not detected if only patients hospitalized with ICD codes 430-438 or deaths with these ICD codes are included in the case ascertainment, a proportion similar to that observed in other stroke registers.9 The proportion of nonfatal stroke cases treated out of the hospital was low in most populations. In some countries,5 7 9 20 this may be a true reflection of a standard medical practice to hospitalize patients with acute stroke, especially patients below the age of 65 years. A small proportion of MONICA centers reported no nonfatal out-of-hospital cases, and further evidence is needed before these registers can be accepted as population based rather than hospital based.
Further quality assurance is required in half of the centers to make sure that case ascertainment is complete. Once the validation is completed, the MONICA Project stroke study, with a combined population of over 4.5 million men and women aged 35 to 64 years, will provide the largest database yet available for international comparisons of stroke incidence and case fatality.
Sites and Key Personnel of Contributing MONICA Centers
China. Beijing Heart, Lung, and Blood Vessel Research Institute, Beijing: Wu Zhausu and Wu Yingkai.
Denmark. Glostrup Population Studies, Glostrup University Hospital, Glostrup: M. Schroll, H. Kirkby, H. Schnack, G. Vincents, and P. Thorvaldsen.
Finland. National Public Health Institute, Helsinki: E. Kaarsalo, E.V. Narva, T. Nuottimäki, P. Puska, K. Salmi, V. Salomaa, C. Sarti, J. Sivenius, J. Torppa, and J. Tuomilehto.
Germany. Department of Clinical and Social Medicine of the University Medical Clinic, Heidelberg: E. Nüssel and E. Ostör-Lamm; Centre for Epidemiology and Health Research, Berlin: W. Barth, L. Heinemann, H. Schaedlich, and D. Eisenblätter.
Italy. Institute of Cardiology, Regional Hospital, Udine: G.A. Feruglio.
Lithuania. Kaunas Medical Academy, Institute of Cardiology, Kaunas: J. Bluzhas and D. Rastinyte.
Poland. National Institute of Cardiology, Department of Cardiovascular Epidemiology and Prevention, Warsaw: S.L. Rywick, M. Polakowska, G. Broda, and B. Jasinski.
Russian Federation. National Research Centre of Preventive Medicine, Moscow: T. Varlamova, E. Surkov, and S. Krassilov; Institute of Internal Medicine, Academy of Medical Sciences, Novosibirsk: Y. Nikitin and V. Feigin.
Sweden. Preventive Cardiology Unit, Östra Hospital, Göteborg: L. Wilhelmsen and P. Harmsen; Department of Medicine, Kalix Hospital, Kalix: F. Huhtasaari and V. Lundberg; Department of Medicine, University Hospital, Umeå: P.O. Wester, K. Asplund, and B. Stegmayr.
Yugoslavia. Novi Sad Health Centre, Novi Sad: M. Planojevic and D. Jacovljevic.
MONICA Management Center. World Health Organization, Geneva, Switzerland: I. Gyarfas, Z. Pisa, S.R.A. Dodu, S. Böthig, I. Martin, M.J. Watson, and M. Hill.
MONICA Stroke Advisory Group. K. Asplund, R. Bonita, D. Eisenblätter, S. Hatano, M. Schroll, H. Tunstall-Pedoe, J. Tuomilehto, P.O. Wester, and W. Zhaosu.
MONICA Data Center. National Public Health Institute, Helsinki, Finland: K. Kuulasmaa, J. Tuomilehto, V. Moltchanov, A.-M. Rajakangas, E. Ruokokoski, and J. Torppa.
MONICA Quality Control Center for Event Registration. University of Dundee, Dundee, Scotland: H. Tunstall-Pedoe, K. Barret, and C. Brown.
- Received July 28, 1994.
- Revision received October 10, 1994.
- Accepted December 19, 1994.
- Copyright © 1995 by American Heart Association
Tunstall-Pedoe H, for the WHO MONICA Project Principle Investigators. The World Health Organization MONICA Project (Monitoring of Trends and Determinants in Cardiovascular Disease): a major international collaboration. J Clin Epidemiol. 1988;41: 105-114.
Tuomilehto J, Sarti C, Narva EV, Salmi K, Sivenius J, Kaarsalo E, Salomaa V, Torppa J. The FINMONICA Stroke Register: community-based stroke registration and analysis of stroke incidence in Finland, 1983-1985. Am J Epidemiol. 1992;135: 1259-1270.
Terént A. Increasing incidence of stroke among Swedish women. Stroke. 1988;19:598-603.
Iso H, Jacobs D, Goldman L. Accuracy of death certificate diagnosis of intracranial hemorrhage and nonhemorrhagic stroke. Am J Epidemiol. 1990;132:993-998.
Sarti C, Tuomilehto J, Sivenius J, Kaarsalo E, Narva EV, Salmi K, Salomaa V, Torppa J. Stroke mortality and case-fatality rates in three geographic areas of Finland from 1983 to 1986. Stroke. 1993;24:1140-1147.
Bamford J, Sandercock P, Dennis M, Burn J, Warlow C. A prospective study of acute cerebrovascular disease in the community: the Oxfordshire Community Stroke Project. 1981-86, I: methodology, demography and incident cases of first-ever stroke. J Neurol Neurosurg Psychiatry. 1988;51:1373-1380.
Bonita R, Beaglehole R, North J. Event, incidence, and case fatality rates of cerebrovascular disease in Auckland, NZ. Am J Epidemiol. 1984;120:236-243.
Harmsen P, Tsipogianni A, Wilhelmsen L. Stroke incidence rates were unchanged, while fatality rates declined, during 1971-1987 in Göteborg, Sweden. Stroke. 1992;23:1410-1415.
World Health Organization. Cerebrovascular diseases: prevention, treatment, and rehabilitation. Report from a WHO Meeting. Geneva, Switzerland: World Health Organ Tech Rep Ser; 1971.