Abstract
Cardiovascular disease is the leading cause of death and hospitalization in both sexes in nearly all countries of Europe. The main forms of cardiovascular disease are ischaemic heart disease and stroke. Stroke by itself is the second leading cause of death in the European Union, and the annual number of cases of stroke is expected to increase within the next few decades, mainly owing to a growth in the proportion of older people.
Stroke is an expensive disease because of the large number of premature deaths, ongoing disability in survivors, and the impact on families or caregivers and on health services (treatment and rehabilitation). Therefore, there is a pressing need to make stroke prevention and treatment a priority, to reduce the growing health burden and lessen its socioeconomic impact.
The magnitude of the problem contrasts with the shortage, weak quality, and comparability of data available in most European countries. A stepwise surveillance procedure based on standardized data collection, appropriate record linkage, and validation methods was set up by the EUROCISS project (EUROpean Cardiovascular Indicators Surveillance Set), to build up comparable and reliable indicators for the surveillance of stroke at the population level.
This manual of operations is intended for health professionals and policy makers. It provides a standardized and simple model for the implementation of a population-based register, which can provide estimates of attack rate and case fatality. The manual recommends starting from a minimum data set.
Before implementing a population-based register, it is important to identify the target population under surveillance, which should preferably cover a well defined geographical and administrative area or region representative of the whole country, where population data and vital statistics (mortality and hospital discharge records at least) are routinely collected and easily available each year. All cases among residents should be recorded even if the case occurs outside the area. Validation of a sample of fatal and nonfatal events is mandatory. Eur J Cardiovasc Prev Rehabil 14 (Suppl 3):S23-S41 © 2007 The European Society of Cardiology
Introduction and rationale
Burden of disease
The most frequent forms of cardiovascular disease (CVD) are those of an atherosclerotic origin, mainly ischaemic heart disease (IHD), stroke, and heart failure (HF).
More than 1.9 million people die every year from CVD in the European Union (the EU, data refer to 25 member states: Austria, Belgium, Cyprus, Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, The Netherlands, Poland, Portugal, Slovakia, Slovenia, Spain, Sweden, and United Kingdom). Nearly half (42%) of all deaths (47% of the deaths in women and 39% of the deaths in men) are from CVD [1].
CVD clinically manifests itself in midlife and at older ages, after many years of exposure to unhealthy lifestyles (smoking habit, unhealthy diet, and physical inactivity) and risk factors (total and low-density lipoprotein cholesterol, blood pressure, and diabetes). CVD accounts for over 225000 premature deaths before the age of 65 in the EU: 7% of all men and 3% of all women die from CVD before the age of 65 [1].
Although the clinical onset is mainly acute, stroke often evolves gradually, causes substantial loss of quality of life, disability, and lifelong dependence on health services and medications. The societal costs are substantial: they include not only the costs directly related to healthcare and social services, but also those linked to (i) illness benefits and retirement; (ii) impact on families and caregivers; and (iii) loss of years of productive life.
Stroke is the second leading cause of death in the EU, accounting for 490000 deaths each year. Over one in eight women (13%) and one in 10 men (9%) die from this disease, and many more suffer from nonfatal events [1].
In most western European countries, death from stroke has declined by 30-50% since 1975; however, in the countries of Eastern Europe, stroke mortality has remained stable or has increased slightly over the same period of time [2–5]. Despite the decline in mortality in Western Europe, the annual number of cases of stroke is expected to increase within the next few decades. This increase will be mainly owing to a 30% growth in the elderly population, which will lead to an increase in the health burden of stroke and a consequent increase in economic costs [6].
In the last decade, innovations in diagnostic technologies in the cardiovascular field have facilitated diagnosis in the earlier phases of the natural course of the disease or in the presence of less severe tissue damage. The use of diagnostic technologies such as the computed tomography (CT) scan and magnetic resonance imaging (MRI) has greatly improved the accuracy of diagnoses of hospitalized cerebrovascular events, allowing delineation of the location and type of lesion.
The World Health Organization – MONItoring trends and determinants of CArdiovascular diseases (WHO–MONICA) project [7] – has demonstrated a large variation between countries in case-fatality rates (the proportion of fatalities occurring within 28 days of onset of acute stroke), ranging from 15% in the northern countries to 50% in some eastern European states. The implications of these findings are that the quality of acute stroke care varies between countries, and that an improvement in initial diagnosis, treatment, and rehabilitation programmes can reduce case-fatality rates [6].
Lifetime costs of first-ever stroke are estimated to be between 31440 euros in The Netherlands and 63000 euros in Sweden, of which hospital costs account for 45% in the first year after a stroke [8, 9]. It is estimated that hospital costs attributed to stroke will increase by 1.5%/ year [9].
Across Europe with its ageing population, there is a pressing need to cope with this increase in costs and to make stroke prevention and treatment a priority. This is essential to reduce the growing health burden and lessen its socioeconomic impact [10].
According to the Organisation for Economic Cooperation and Development (OECD), it is not inevitable that longer lifespans should lead to higher costs. This is one of the reasons why the health system should be largely oriented towards preventive action. Epidemiological studies have shown that stroke is preventable to a large extent. Different preventive strategies can be implemented to (a) reduce the occurrence and impact of stroke (through, for instance, the identification of individuals at high risk of stroke such as those with hypertension, diabetes, or the smoking habit); (b) intensify treatment in people who have already experienced a stroke or a transient ischaemic attack; or (c) improve rehabilitation. At the European level, WHO, OECD, and EUROSTAT (Statistical Office of the European Community) collect simple indicators (mortality and hospital discharge rates) and process them into tables available on their web site (www.euro.who.int/hfadb; www.oecd.org; and www.europa.eu.int/comm/eurostat). These data are rarely comparable owing to the different methodologies used and the peculiar health system of each country.
Disease register
The objectives of a stroke population-based register are to (i) evaluate the frequency, distribution, and prognosis of the disease, providing indicators such as attack rate, incidence rate, prevalence, and case fatality; (ii) compare trends in different countries; (iii) evaluate trends and changing patterns, outcomes, and treatment effectiveness; and (iv) monitor disease-prevention programmes [11].
Focusing on the general population, a stroke register might provide a comprehensive picture of stroke in the community, highlight problem areas, and suggest which population groups are at high risk and where treatment facilities are most in need of improvement. It can provide information needed to plan healthcare services and to develop and test the methods that are most useful as a basis for preventive action.
The register includes all the cases of stroke in a defined population, irrespective of whether the patients were treated at home or in hospital, and of the season of the year or time of the day that the stroke might have occurred; this would also include rapidly fatal cases that had been unable to reach medical help.
It is important that the collection of information on suspected events and on the application of diagnostic criteria should follow a standardized methodology, to enable data comparison in different areas of the same country or between different countries.
To summarize, a population-based register is intended for health professionals and policy makers. It provides the means to understand the characteristics, the burden, and the consequences of the disease in the population through the following mechanisms:
the monitoring of the occurrence of the disease (i.e. assessing population differences and trends in attack, incidence rates, and in mortality over time);
the understanding of the differences and changes in the natural disease dynamics between sexes, age groups, social classes, ethnic groups, and other categories;
the identification of vulnerable groups;
the monitoring of inpatient and outpatient case fatality;
the assessment of relationships between disease incidence, case fatality, and mortality;
the monitoring of the consequences of disease in the community, in terms of drug prescriptions and rehabilitation;
the monitoring of the use of new diagnostic tools and treatments and their impact.
This is crucial for the following reasons:
developing health strategies and policies;
planning health services and health expenditures;
improving the allocation of resources;
evaluating the effectiveness of interventions.
To provide this, a register must be validated. Validation provides the means to carry out the following:
take into account bias from diagnostic practices and changes in coding systems;
trace the impact of new diagnostic tools and redefinition of events;
ensure data comparability within the register (i.e. different subpopulations, different time points);
ensure data comparability with other registers within and between countries.
Historical background
The WHO stroke register was the first attempt to collect data on stroke in the community in a uniform manner, from countries with different social, cultural, and environmental backgrounds. Data collection lasted from May 1971 to September 1974, and was a joint undertaking of the WHO and 15 collaborating centres in 10 countries from Asia, Africa, and Europe. About 2 million people were under surveillance, and data was obtained from 6395 new patients with stroke (3270 men and 3125 women). Fourteen of the centres covered the general population in defined geographical areas, and one centre covered an occupational group consisting mainly of men below the age of 55 years. No limitations of age and sex were set in the study areas, except for two centres in Sweden and Japan. A stroke was defined as rapidly developing clinical signs of focal (at times global) disturbance of cerebral function, lasting more than 24 h or leading to death, with no apparent cause other than that of vascular origin [12].
The WHO MONICA project [13, 14] was started in the first half of the 1980s and lasted until the first half of the 1990s. Stroke registers were established in 17 centres in 10 countries. Study populations were residents in geographically defined areas, and they included men and women aged 35-64 years, with the optional inclusion of the 65-74 years age group.
All stroke events in defined populations were ascertained and validated according to a common protocol and uniform criteria. Almost 25000 stroke events in more than 15 million person-years were analysed. Stroke was defined as ‘rapidly developing signs of focal (or global) disturbance of cerebral function, lasting more than 24h (unless interrupted by surgery or death), with no apparent nonvascular cause’. This category included patients presenting with clinical signs and symptoms suggestive of subarachnoid haemorrhage, intracerebral haemorrhage, or cerebral ischaemic infarction. This definition excluded patients with transient ischaemic attack or stroke events resulting from blood disease or brain tumours. Secondary stroke caused by trauma was also excluded. Up to six-fold differences were observed in stroke mortality. Mortality declined in eight of the 14 populations in men and in 10 of the 14 populations in women. An increase in mortality was observed in Eastern Europe. In the populations with a declining trend, about 2/3 of the change could be attributed to a decline in case fatality. In populations with increasing mortality, the rise was explained by an increase in case fatality.
Existing registers in Europe: an overview
The data collection for the international MONICA study ended in 1994-95. Some countries continued to collect data every year, whereas others did so only periodically (every 5 years).
Currently, the existing registers in Europe adopt different data collection procedures: some registers are based on the procedures used in the MONICA study and others on administrative databases, with or without record linkage; some are national, and some are regional. Different age groups are covered, the degree of validation of the diagnostic information varies, and in most registers, validation is much less intensive than in the MONICA study. The registers are used for different purposes, and have different strengths and limitations [11].
Tables 1–3 give a brief overview of the existing stroke registers in Europe. As shown in Table 1, Denmark, Finland, and Sweden have national stroke registers, which are based on record linkage between hospital registers and cause-of-death registers.
Table 2 shows regional population-based stroke registers: most of them are based on a disease-specific data collection method comparable with that of the MONICA registers.
Table 3 shows examples of registers that are based on data from healthcare institutions such as general practitioners (GPs) and hospitals. These registers are not population based, as they do not include out-of-hospital patients or those not seen by the GP. They, therefore, do not consider sudden death occurring out of hospital. These registers are not intended to assess disease occurrence, but rather to evaluate the outcome and survival of stroke patients.
It is worthwhile to mention the European Register Of Stroke (EROS), a 4-year prospective study across Europe aiming at estimating the impact of stroke. It is an attempt at understanding the factors underlying variation in the quality of care and outcome after stroke, answering unresolved issues with regard to the influence of socio-demographic, case mix, and stroke healthcare, and assessing quality factors on the variations in the health of stroke patients around Europe. The cities of London, Helsinki, Glasgow, Edinburgh, St Petersburg, Kaunas, Warsaw, Dijon, Menora, Florence, and Stockholm are participants in EROS [15].
Objectives
The purpose of the EUROCISS project is to provide a general guide and updated methods for the surveillance of stroke, for those EU countries that lack appropriate surveillance systems but wish to implement a population-based register to produce comparable and reliable indicators.
Taking into account the new developments in diagnostic criteria, treatment, and information technology in recent years, this manual provides a standardized and simple model for the implementation of a population-based register. It recommends starting from a minimum data set and following a stepwise procedure based on standardized data collection, appropriate record linkage, and validation methods. This manual is intended for investigators, health professionals, policy makers, and data collection staff interested in the surveillance of stroke.
Although in many countries, data extracted from some sources of information [mortality and hospital discharge records (HDR)] are now available thanks to the continuing process of computerization, they are rarely reliable and comparable. These data can produce reliable indicators only if they have been properly processed and validated by independent epidemiological sources.
National population-based stroke registers
NIPH, National Institute of Public Health. NBHW, National Board of Health and Welfare. Source: European J of Public Health 2003;
Regional population-based stroke registers
NIPH, National Institute of Public Health. CHU, Centre Hospitalier Universitaire. Source: European J of Public Health 2003;
Examples of healthcare services-based stroke registers in countries participating in the EUROCISS project
GP, general practitioner; HDR, hospital discharge records.
This manual represents a valid tool to build the core indicators (attack rate, incidence, and case fatality) recommended by the EUROCISS project research group for inclusion in the short list of health indicators set up by the European Community Health Indicators Monitoring (ECHIM) project. This project was launched in 2005 with the aim of implementing health monitoring in the EU [16].
Strategy for surveillance
Surveillance tools and types of registers
Surveillance is the ongoing systematic collection, analysis, interpretation, and dissemination of health information to health professionals and policy makers. Surveillance, defined as a continuous and not episodic or intermittent activity, differs from monitoring [17, 18]. Disease surveillance in a population can be done using many different data sources (Table 4).
Most countries have national databases on causes of death and on discharge diagnoses for hospitalized patients. Mortality statistics have for many years been the main tool for comparing health and disease patterns among countries: today they still remain the only source of information in some countries. They have also been used to monitor trends in cerebrovascular disease and to compare mortality among countries. Since the 1950s, the cause of death has been registered according to the International Classification of Diseases (ICD), to make data comparable. Different classifications of disease within versions and different methods of ascertainment have led to problems in comparison between different revisions of the ICD and/or similar versions among countries.
In recent years, routine statistics also include discharge diagnoses from hospitalization and, in some countries, visits to outpatient clinics that are coded according to the same international classifications as the mortality data. Stroke can be extracted for relevant populations and age groups, and these routine statistics are still very important tools for monitoring the disease.
Many countries also have health interview surveys/health examination surveys. These surveys are primarily used for monitoring disease prevalence (including cerebrovascular disease), prevalence of risk factors (health behaviour, social network, and environmental risk factors), and disease consequences (disability, reduced physical function, and unemployment). They are described in detail in the manuals of operation of CVD surveys. Few countries have an established disease-specific stroke register, which ensures a more precise and valid monitoring of this disease. A population-based register is usually formed through linkage of various sources of information (mortality data, hospital discharge, and records of GPs), and covers a defined population (entire municipalities, regions, or the whole country) and a specific age group (35-74 years; 35-64 years; or all ages).
A population-based register should be used for the surveillance of stroke morbidity and mortality, as it considers both fatal and nonfatal events occurring in hospital and out of hospital; therefore, it provides estimates of key indicators such as attack rate and case fatality. Incidence can be assessed if information on first event is available. If survival rates are available, prevalence can be assessed as well.
Methods for surveillance of cerebrovascular disease in the population
GP, general practitioner; HDR, hospital discharge records.
Case finding and validation procedures depend on data collection methods, the healthcare and financing system, and diagnostic criteria applied in the definition of events. The accuracy of rates produced is related to the completeness and quality control of the data collected for the numerator (death and hospital discharge registers) and the denominator (census or population register). Completeness also depends on tracing the patients treated outside hospitals (nursing home, clinic, and others) and outside the area of surveillance. The routine recording of diagnoses might be a problem for the registration of stroke: a large proportion of ‘new stroke diagnoses’ are merely sequelae of an old stroke. This problem increases with ageing.
The definition of the event must take into account both the ICD codes reported in the hospital discharge diagnoses (main or secondary) or in the causes of death (underlying or secondary) and the duration of the event. Stroke can occur more than once; therefore, it is necessary to consider both the first and the recurrent events. In this context, deaths occurring within 28 days are usually considered to reflect the same event [17] (see the definition of recurrent events in the definition of events section).
A personal identification number (PIN) is a strong tool in the linkage procedures between hospital discharge diagnoses, records of GPs, and death certificates; alternatively, multiple variables (e.g. name, date and place of birth, sex, and residence) can be used for record linkage.
Specific stroke population-based register
The strength of this register lies in the possibility of validating each single event according to standardized diagnostic criteria and collecting disease-specific clinical and paraclinical data [19]. The weakness lies in the fact that data collection is expensive, and this kind of register can usually be maintained only for a limited period of time, in a defined population of reasonable size. Another limitation is that a local or regional register might not be representative of the whole country.
Identification of events can be obtained by ‘hot pursuit’ or ‘cold pursuit’. ‘Hot pursuit’ means identifying patient admissions to hospital, usually within 1 or 2 days of event onset, and acquiring relevant information by visiting the ward or interviewing the patient. Information bias is minimized by the hot pursuit approach, as information is collected immediately after the event. The process is comparatively demanding in terms of resources. ‘Cold pursuit’ implies the use of routine and delayed procedures to obtain data, such as by consulting hospital discharge and death records. The process is easier and less expensive than hot pursuit. The number of cases studied is typically smaller because discharge diagnoses are more precise and specific than those on admission; however, there is a possibility of missing important information. Both methods are used to identify suspected events, which are subsequently validated using specific diagnostic criteria.
The specific stroke register is important as it collects data on fatal and nonfatal events; actually, official mortality statistics provide only a limited and sometimes biased picture of stroke in the population. A large proportion of stroke victims are left with permanent disability; the economic and human consequences of stroke extend far beyond what emerges from routine mortality data. The specific stroke register, which allows the assessment of incidence and prevalence, reflects the impact of stroke in the community better than mortality does. Monitoring nonfatal stroke is associated with a number of problems, the most important being the completeness of case finding, especially in areas where many stroke patients are not treated in hospital. An extensive review of stroke incidence registers showed that few of them provide reliable data [20]. Indeed, it has been claimed that most of the differences in stroke mortality and incidence rates reported to exist between populations are attributable to methodological bias.
A specific stroke register provides standardized and reliable epidemiological data for public health initiatives aimed at preventing the disease. It has been used in the WHO MONICA project, in which uniform criteria for recording cerebrovascular accidents (CVA) have been applied [14].
Population-based register based on routine administrative data
Identification of events is based on the linkage of mortality data and the HDR. The register, which is based on routine administrative data, has existed for many years in the northern countries, where all individuals are identified by the PIN, which allows for record linkage between different information sources. This register is economical, covers the whole country, covers all age groups, and collects large numbers of events. The main objective of administrative databases is to produce relevant statistics to plan health services and healthcare expenditure and to give internationally comparable data on mortality, causes of death, and hospital admissions. The register, which is based on routine administrative data, is not primarily planned for research purposes, but is increasingly used in epidemiological research. Its strength lies in the fact that it covers the whole country, and the completeness is close to 100%. The weakness lies in the fact that data are not standardized to the same degree as in the disease-specific data collection, and that the available clinical and paraclinical data are limited. If used in research, this register needs to be carefully validated. Stroke registers based on administrative data such as HDR and deaths have been employed in Denmark and Finland, to obtain national rates of stroke incidence, mortality, and case fatality [21, 22]. A similar approach is being investigated for use in Sweden.
Studies on the feasibility of combining data from routine hospital discharge and cause-of-death registers have been carried out in Finland: over 90% of the hospitalized acute stroke events (first and recurrent) included in the Finland MONICA stroke register were found in the HDR with one of the stroke diagnoses. The missing events were mainly explained by errors in the PIN (leading to unsuccessful record linkage) and the different practice of defining an event as a hospitalized one when death occurred in the emergency room (leading to exclusion from the HDR) [21].
In the past, hospitalizations for rehabilitation purposes were often coded using an ICD code for acute stroke. With the ICD-9 version, a separate code for acute events and sequelae was introduced. The definition of stroke death also differs between the specific stroke register and the mortality register: in the specific stroke register, the death is very strictly defined as a death occurring within 28 days of the onset of the event; in contrast, deaths occurring after 28 days of the onset of symptoms are often coded as stroke deaths in the mortality register [21].
In studies assessing trends in stroke subtypes, the change in the use of neuroimaging examinations and autopsy frequency should be reported.
General practitioner register
A great majority of health problems are managed in primary care, and do not go further into other levels of the healthcare system. This is true especially for those less serious problems that do not require hospitalization. The fact that primary healthcare is generally the first and most frequently used health service makes general practice a rich source of information. This further emphasizes the need for monitoring health in primary care settings, to have a full picture of the health status of populations. This is particularly necessary for stroke, which occurs especially among the elderly. In some countries, patients with stroke are treated at home even during the acute phase: this makes the GP register a valid source of information for monitoring stroke. Monitoring health in primary care should, however, not be seen in isolation from other sources of information about health.
Essentially, there are two models for collecting morbidity data in primary care. One is based on episodes of care, recording data on all doctor-patient interactions and gathering information on consultation rates and patterns of clinical management. The other focusses on specific disorders, using a limited number of standardized case definitions and attempting to assess the burden of disease attributable to those disorders in the population in question. The first model is exemplified by the English General Practice Research Database programme [23, 24], and the use of International Classification of Primary Care (ICPC) codes [25]; whereas, the second one is illustrated by the Morbidity Sentinel Stations programme that is now operational in several European countries [26–28].
Target population
A population-based stroke register might cover an entire country; if this is not feasible, the population under surveillance would typically be residents of a defined region in the country. The target population should preferably cover a well-defined geographical and administrative area or region for which population data and vital statistics are routinely collected and are easily available each year. Both urban and rural areas should be monitored: differences often exist with regard to exposure to risk factors, treatment of predisposing disease, and access to facilities.
It is important that all patients among those residing in the area are recorded even if the event occurs outside the area (completeness). In the same way, all patients treated at hospitals within the area but with residence outside the area must be excluded. If this is not possible, it is important to give an estimate of the magnitude of the loss of cases and establish whether it could be changing and interfering with the validity of the observed trends in the rates over a period of years.
It is also important to consider to what extent an area is representative of the whole country (representativeness): it could be representative according to the CVD mortality rate, the distribution of risk factors (socioeconomic status and health behaviour), and the distribution of health services (specialized hospital and GP). In some countries, it might be better to start the implementation of a register with the high-risk areas.
The population to be monitored should be selected to produce estimates of disease rates that are sufficiently robust from a statistical point of view, so that trends can be established and data comparability ensured. In general, it is necessary to select more than one area representative of socioeconomic or ethnic differences, to have a comprehensive picture for the whole country, and a coordinating body between the areas is recommended to ensure comparability.
The target population should be selected taking the following parameters into account:
Age
The age range covered by the MONICA project was 35-64 years. As reported in the final report, the EUROCISS research group suggests the wider age range of 35-74 years, or even up to 84 years of age when possible, considering that more than half of the stroke events occur in patients above 65 years of age. The age groups recommended from the EUROCISS project to present morbidity and mortality are decennia, in particular the age ranges 35-44, 45-54, 55-64, 65-74, and 75-84. If routine administrative data are used, all ages will automatically be included, but for patients aged 85 and above, the validity of the diagnostic information tends to be less reliable. Age-standardized rates (35-74 and 35-84) are recommended using the European standard population as reference.
Sex
Stroke is an important cause of death and disability in men and women, and the population should include both sexes. No major sex differences have been evidenced in stroke presentation or management: mortality and quality of life at 6 months are similar in women and men.
Population size
To be eligible to participate in a stroke population-based register, a minimum of 300 stroke events/year in the population aged 45-74 years is necessary. The size of the population under surveillance is determined by the number of fatal and nonfatal events and the event rate in the age group concerned. The minimum of 300 events (fatal and nonfatal) has been established to detect a decrease in mortality trend of 2% in event rate/year. This means that the population under surveillance could range between approximately 1200000 (all ages) in low-incidence countries like Italy and approximately 400000 (all ages) in a high-incidence country like Finland, basing the calculation on the attack rate among women, which is usually lower than that among men (1998-99 Register data). If more areas are enrolled, it would be desirable that the same number of 300 total events is considered for each single area.
Patient eligibility
Individuals are considered eligible for inclusion in a stroke population-based register only if they are resident in the area under surveillance, meet the selected age criteria, and had a stroke event within the defined time period.
Data sources
To monitor stroke in the general population, at least the following sources of information should be available: mortality records with death certificates and HDR with clinical information. A special stroke register would typically include several sources of information.
Some events occur suddenly, and the patients might not have been able to reach the hospital; some nonfatal cases might not have been referred to hospitals for treatment, particularly if they involved very old individuals. Therefore, additional sources are usually needed to obtain complete information on all fatal and nonfatal events: clinical pathology laboratories (autopsy register), nursing homes, clinics, emergency or ambulance services, GPs, and radiology units (Table 5).
Sources of information
GP, general practitioner; HDR, hospital discharge records.
Death certificate
The death certificate provides complete data on fatal events. It is collected in a systematic and continuous way in all the EU countries. Mortality statistics are easily accessible in all countries, but are usually available in a detailed and complete form after 2-4 years.
The format of the death certificate varies from country to country; however, it generally includes personal identification data, date and place of death (i.e. municipality, nursing home, hospital, or other), and causes of death (underlying, immediate, and contributory). Causes of death are coded according to the ICD. Problems of temporal and geographic comparisons derive from the different versions of the ICD adopted over time (seventh, eighth, ninth, and tenth revisions) and from the different coding practices in each country. Furthermore, diagnostic criteria for coding death certificates are not defined at the international level, and the ICD nosologic and nosographic versions are updated every 10 years by the WHO. Some countries code only the underlying cause of death.
The reliability of mortality data depends on the completeness and accuracy of the vital registration system as well as on the registration and coding of causes of death. When the proportion of deaths coded as ‘unknown cause of death’ is higher than 5%, cause-specific mortality data should be used with caution. The accuracy of the recorded causes of death depends on the autopsy rate. This rate varies largely between countries and over time. In some countries, the autopsy rate has declined in recent years, which is a problem for the use of mortality statistics in disease surveillance.
Hospital discharge records
HDR give the number of hospitalizations for stroke, which are absolutely necessary to monitor CVD. Hospital discharge data are available in most EU countries; however, in some countries, these are only aggregated tables without detailed information on age and sex distribution and without haemorrhagic and ischaemic stroke as separate diagnostic categories.
HDR include personal identification data, admission date, type of hospitalization (urgent, ordinary, or transfer to other structure), and discharge diagnoses. Hospital discharge diagnoses are coded by the ICD codes (currently ICD-9 or ICD-10). In some countries, only a limited number of diagnoses is coded.
Clinical information and medical care reported in hospital documents are important for the validation of events.
Problems in the assessment of a specific stroke event can arise when an acute event is followed by a period of rehabilitation or a transfer to other wards, and the event could thus be counted more than once (sequelae). Therefore it is important to perform record linkage between different sources in order to identify the same event. HDR do not include emergency rooms, and private hospitals or nursing homes are only included in some countries.
Discharge diagnoses are not validated on a routine basis, and validation studies are necessary in all countries to check the diagnostic quality and data comparability. The validity of a hospital discharge diagnosis might vary on the basis of patient characteristics, geographical region, and type of hospital or clinic.
Moreover, hospital admission policies may vary over time and place; the registration of the most severe cases dying shortly after the arrival at the hospital differs between hospitals, depending on the administrative procedures connected with hospital admissions. HDR might also include patients not resident in the area under surveillance. The adoption of new diagnostic techniques such as MRI and CT scan might cause major changes in event rates estimated from HDR. Therefore, these techniques should be taken into account when interpreting trends.
A further problem might arise from the use of the Diagnosis-Related Group (DRG) system. In some countries, the financing of healthcare services is based on the DRG tariff system, which is built on equal-resources criteria and aggregates events into major diagnostic categories. For this reason, DRGs might be useful for acute events, but are not reliable for chronic phase of stroke (sequelae) which require a long hospital stay and rehabilitation. Countries using the DRG system are Denmark, Finland, France, Germany, Italy, Norway, Portugal, Spain, and Sweden. To assess the occurrence of stroke, HDR from all hospital departments should be used. If this is not possible, however, at least the following departments must be taken into consideration:
intensive care (an intensive care unit, including any type of acute medical unit);
medical (a general medical ward, including a geriatric unit);
neurological/neurosurgical (a general neurological ward);
rehabilitation (a specialized rehabilitation unit, except a rehabilitation stroke unit);
stroke (acute and rehabilitation stroke units);
other (other units such as radiology).
Autopsy register
Not all countries perform autopsy on suspected or sudden deaths on a routine basis. Autopsy is performed on violent deaths or on deaths occurring in hospital when the clinical diagnosis has not been determined. The first type is performed by a forensic medicine specialist; the second by a pathologist of the hospital where the death occurred. Data from this register refer, therefore, to a low percentage of deaths, but provide a more valid diagnosis to complement the information reported on the death certificate.
Nursing home and clinic
The nursing home and clinic mainly provide data on older patients, who sometimes get care from these institutions without being admitted to hospital. Therefore, information on events occurring in the nursing home can be critical, especially if the register covers elderly patients. In some countries, rehabilitation after an acute event is provided by the rehabilitation clinic, which might give information on patients who have received the acute care outside the region.
Emergency and ambulance services
Data provided by emergency and ambulance services are useful to integrate information for register implementation, as patients dying suddenly are not able to reach the hospital. These services are able to provide data otherwise not obtainable, such as CT scan or MRI records during the acute phase of the event, blood pressure measurement, blood glucose, peripheral oxygen saturation, body temperature and fluid balance, level of consciousness (fully conscious, somnolent, semicomatose, and comatose), and muscular deficit at the time of event occurrence, in paucisymptomatic patients referring to emergency services. The need of very urgent medical treatment often makes information partial, but the integration of these data with those from other sources of information contributes to the implementation of the register.
General practitioner register
In some countries a GP register can be useful when dealing with events not necessarily requiring hospitalization. This is particularly important for the elderly population.
Radiology unit
The role of the radiology unit (CT scan or MRI) is a support in the identification of nonhospitalized events, in the diagnosis of stroke type (haemorrhagic or ischaemic), and in treatment.
Methods
Definition of events - subtypes
Three major stroke subgroups have been identified as follows: ischaemic stroke, intracerebral haemorrhage, and subarachnoid haemorrhage (Table 6). For the definition of subtype, neuroimaging is necessary. It should be noted that each type differs with respect to survival and long-term disability.
General major symptoms
Symptoms should be of a presumed vascular origin and should include one or more of the following definite focal or global disturbances of the cerebral function:
unilateral or bilateral motor impairment (including lack of coordination);
unilateral or bilateral sensory impairment;
aphasia/dysphasia (nonfluent speech);
hemianopia (half-sided impairment of visual fields);
forced gaze (conjugate deviation);
apraxia of acute onset;
ataxia of acute onset;
perception deficit of acute onset.
Other symptoms Other symptoms that might be present but are not adequate for stroke diagnosis (often resulting from other diseases or abnormalities such as dehydration, cardiac failure, infections, dementia, and malnutrition) are as follows:
dizziness, vertigo;
localized headache;
blurred vision of both eyes;
diplopia;
dysarthria (slurred speech);
impaired cognitive function (including confusion);
impaired consciousness;
seizures;
dysphagia.
Subarachnoid haemorrhage For subarachnoid haemorrhage, at least one of the following must be present in addition to the general major symptoms:
recent subarachnoid haemorrhage, aneurysm, or arteriovenous malformation (necropsy/autopsy);
blood in the fissura sylvii or between the frontal lobes or in the basal cistern or in the cerebral ventricles (CT or MRI);
bloodstained cerebrospinal fluid (>2000 red blood cells/mm3), aneurysm, or an arteriovenous malformation (angiography);
bloodstained cerebrospinal fluid (>2000 red blood cells/mm3), also xanthochromic and intracerebral haemorrhage (necropsy or CT scan).
Strokelike symptoms A broad range of other diseases can cause similar symptoms, for example, HIV/AIDS, tuberculosis, syphilis, and intracerebral cancer. These diseases are known to cause focal neurologic disturbances and thereby mimic a stroke. Attention to the development of symptoms is an important factor to consider, to avoid other diseases being misinterpreted as vascular disease and leading to ineffective preventive strategies.
Onset and survival
Stroke events are classified as ‘first ever’ or ‘recurrent’, with ‘nonfatal’ and ‘fatal’ outcomes:
Subtypes of stroke
Modified from WHO STEPS Stroke Manual V2.1. ICD, the International Classification of Diseases.
‘first ever’ stroke event refers to people who have never had a stroke before;
‘recurrent’ stroke event refers to a new episode: for the symptoms to be counted as a recurrent event, general stroke criteria must be met and either:
- The day of onset is day one (1);
- A new stroke occurring after 28 days is a new event. If a patient, however, experiences further acute symptoms suggestive of stroke within 28 days of the onset of the first episode and in the same carotid or vertebral artery territory, this second episode is not counted as a new stroke event. Equally, if a patient experiences further acute symptoms suggestive of stroke after 28 days of the onset of a first episode, this second episode is counted as a new stroke event.
A ‘nonfatal’ stroke event refers to surviving for at least 28 days after the onset of the stroke symptoms.
A ‘fatal’ stroke event refers to a stroke causing death within 28 days of symptom onset.
It should be noted that each event is registered separately.
Indicators
Attack rate
Attack rate is the total number of new cases and recurrences/100 000 target population over 1 year. It is calculated using either the main cause of hospitalization or, in cases of out-of-hospital deaths, the underlying or contributory causes of death. It should be noted that in the case of stroke, the hospital discharge could sometimes be quite distant from the onset of stroke event. Therefore, a hospital discharge register alone is not always an accurate source of information. Ideally, the inpatient inventory should be checked at the end of each year, to identify patients who are hospitalized for stroke but have not yet been discharged [20].
Incidence rate
Incidence is the number of new cases/100 000 target population over 1 year [20].
Case fatality
Case fatality is the proportion of events that are fatal by the 28th day.
The EUROCISS project recommends, for cerebrovascular events, a 7-day and 28-day case fatality. All inpatient and outpatient fatal and nonfatal events are to be considered as the denominator.
Data collection methods
The different types of registers described in the section ‘Surveillance methods and types of registers’ use different data collection methods. Registers with disease-specific data collection can be divided into registers based on routine administrative data using record linkage, disease-specific registers using hot and cold pursuit, and GP registers.
Population-based register based on routine administrative data
In recent years, the development of computerized record linkage has made it possible to overcome obstacles in linking the administrative database. Record linkage methods can be summarized into three broad categories: manual, deterministic, and probabilistic.
Manual matching is the oldest, most time-consuming, and most expensive method. In general, it is not a feasible option when large databases are involved.
Deterministic linkage matches records from two data sets (or two records from different locations in a single data set), either by using a unique variable (e.g. the PIN or hospital chart number) or by verifying the full agreement of a set of common variables (e.g. name, sex, and birth date).
Probabilistic linkage [29] is used to identify and link records from one data set to corresponding records in another data set (or two records from different locations in a single data set), on the basis of a calculated statistical probability for a set of relevant variables (e.g. name, sex, and date of birth). This type of record linkage links records with a specified high probability of matching. The method requires detailed prior knowledge about various measures of the relative importance of specific identifier values in both the files that are to be linked.
The main limitations of record linkage are the difficulties in the following:
obtaining administrative files for research purposes: mortality data files are usually available at the National Institute of Statistics, whereas hospital discharge data are available at the Ministry of Health. These kinds of data are anonymous and therefore do not allow record linkage. Nominal files are available at the regional level or at the sanitary units;
combining data: missing events are mainly explained by errors in the PIN or in name and they can lead to unsuccessful record linkage;
defining and obtaining a minimal data set (for mortality: the PIN, family and first name, date and place of birth, sex, residence, date and place of death, and underlying and secondary causes of death. For hospital discharge diagnosis, the same variables should be considered together with admission date and hospital discharge diagnoses);
obtaining necessary funds for processing large administrative files.
Record linkage studies, nonetheless, provide evidence of the statistics that could become available with greater integration of administrative databases.
The national stroke registers in the northern countries use record linkage between hospital discharge registers and causes-of-death registers as the basis for the register. The linkage as such is easy because of the PIN given to every citizen in the country. The linkage however, has, to be followed by many specific definitions: how to handle primary and secondary diagnoses, underlying and contributory causes of death, transfer between hospitals with differences in the diagnoses between the admitting hospital and the hospital to which the patient has been transferred, date of attack, first time events and recurrences. Practical methods of approaching these problems have been suggested from the work carried out in Finland [21, 22].
It is usually difficult to detect the incident cases (first events): hospitalization records within the previous 5-7 years are reviewed to check for disease; if no hospital admission for stroke is found, then the stroke event identified is considered a first event. Further problems can arise when estimating trends: for example, the changes in the use of neuroimaging examinations and autopsy frequency can lead to an overestimation of the number of events or make the interpretation of stroke subtypes difficult.
Specific stroke population-based registers
This kind of register uses hot and/or cold pursuit methods for data collection [30].
Hot pursuit This method of detecting events involves identifying patients with acute illness in hospital, by interviewing them directly. The problem with this method is that the data collection technique is very difficult to standardize (e.g. descriptions of symptoms might vary with the observer). Periods of staff shortages or holidays can lead to loss of cases that cannot be recovered, and a large team is needed to search the wards for cases. Some information might, however, be more complete than that obtainable from case notes. The notification of events should be instituted on a routine basis, checking admission registers on the wards.
Although the extreme forms of hot pursuit involve getting the information from the patient who is acutely ill, an alternative is to use the hot pursuit method to identify the patients of interest and to mark their notes or list them for review later. An efficient and reliable routine is needed for picking up the case notes at an identifiable point in their processing.
One benefit of the hot pursuit method is that information on the diagnosis is collected soon after admission. This has its limitations, however, as initial diagnosis can sometimes be superseded by subsequent tests and other more detailed investigations.
Residents hospitalized outside the area will always have to be registered by cold pursuit, weeks or months later.
Cold pursuit Use of discharge diagnoses rather than hospital admissions is a more simple system of identifying events for the study. Its advantage is that it can be done months or years after the event. Its limitation is that the information in the case notes might not be complete and the notes themselves might not be accessible.
Once the event has been identified, validation is required. Medical notes should then be obtained, to extract the necessary information. When a register is launched for the first time, a plan for the future evaluation of trends is recommended. This can be achieved by continuous surveillance as part of a broader health information system or by annual registers compiled at 5-10-year intervals. The minimum recommended period of observation is one complete calendar year because of possible seasonal variations.
Combined approach A mix of hot and cold pursuit ensures the most complete identification of stroke events. Some of the patients must have been identified as soon as possible after symptom onset, with the possibility of direct examination, whereas the remaining events are based on routine data.
It is difficult to check up on a hot pursuit system several months later, but discharge lists can be used as a backup method to ensure that the hot pursuit method had detected all the diagnosed cases. Residents hospitalized outside the area and other late-detected patients mean that a proportion of the events will always have to be registered by cold pursuit, weeks or months later.
Quality control
Quality control of the registers is extremely important for valid monitoring and comparison between regions and countries. The quality of the register depends on the following:
completeness of coverage (sequence of events) and completeness of information;
internal validity;
external validity (representativeness).
The surveillance of stroke is complicated by the fact that a number of patients are not admitted to hospital, particularly in older age. The identification of patients in older populations outside hospital is essential for a precise determination of occurrence. These events are a combination of milder or more severe strokes than those admitted to hospital; consequently, their inclusion influences incidence as well as case fatality.
Completeness of coverage and completeness of information
Completeness of coverage means that all the stroke patients in the target population are included; that is, events are covered, which occur independently inside or outside the region. The register also has to cover events whenever they occur, irrespective of the time of day/night or winter/summer, as well as events occurring outside hospital (e.g. sudden death among patients who never reach the hospital). Completeness of information means that all relevant information should have been registered (e.g. place of treatment, date of admission, date of discharge, PIN, sex, hospital discharge diagnostic codes, intervention/procedure codes, department/ward, and date of birth).
The most important source of systematic bias in estimating incidence is related to the coverage of event registration. The registration system must attempt to identify all possible cases of the disease that have come to the attention of the existing medical and medicolegal sources. The completeness of event identification (acute-care hospital, primary healthcare, and nursing home) and the completeness and availability of information, obtainable for each event recording and diagnosis, depend on the existing standards of medical care: if the medical care system misses or misdiagnoses cases, a register cannot remedy the omission.
When the event is defined (codes and duration), it might be possible to identify duplicate coding and to take out information for quality control purposes. Duplicate codes might include events transferred from one ward to another, for example, for rehabilitation. In some cases, the duration of the admission is very short (>2 days) either because of transferral difficulties or because of misclassification of the diagnosis. These events can also be picked up for validation.
Patients not admitted to general hospitals are a problem for registration if the system is based only on hospital records. Another source of potential loss of identification is private practice: private physicians and hospitals might be less cooperative than those in the public system. In private hospitals, the staff might be more sensitive to criticism and more anxious to show how they register medical documents. GP patient records are usually inadequate for full registration because the patients are frequently looked after at home.
The identification of fatal events is in some ways less difficult than that of nonfatal events. Although survivors might be lost in the totality of inhabitants of the surveillance area, death is unequivocal. The registration of the causes of death might, however, not be correct and will need to be validated. 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 might indicate incomplete registration of out-of-hospital stroke deaths. High case fatality can also indicate loss of nonfatal cases.
The identification of potential events can be based on many different data sources. This might involve a considerable amount of record linkage, which is facilitated if the PIN is adopted.
Another problem relates to medical records, the quality of which might be variable: younger patients might have had no other illness episodes, and the records might be restricted to the relevant stroke event. In older patients, the identification of the event is more complicated owing to the existence of comorbidities.
Internal validity
The most important question regarding validity concerns the diagnostic information. The diagnostic criteria for the event definition are valid if they measure the stroke they claim to measure. Validation preferably evaluates the sensitivity, specificity, and predictive value of the registered diagnosis compared with a golden standard [19].
Validation studies of routine statistics have been carried out over the years with heterogeneous results that were due to differences in methodology, or which reflected true differences between countries in the validity of the routinely collected data. Some studies comparing community registers with national statistics and data from the MONICA project have been carried out. These findings stress the importance of validating routine mortality and hospital statistics against the national register, to determine whether and how they can be used to reflect the true incidence and mortality [31]. Particular attention in this type of validation should be given to secondary discharge diagnosis or causes of death, especially to the diagnostic codes, to detect potentially hidden cardiovascular diagnosis. Consistency of the coding with the diagnosis, and consistency of coding/comparability of the information over different areas of the country and over time represent other problems for validation.
If it is not possible to validate all the diagnoses included in the disease register or in the mortality routine statistics, the objective for validation should be to evaluate a sample of events. The sample should be distributed across a full year, to ensure that potential seasonal or other time-related variations of diagnostic patterns are traced. The sample could include a feasible fraction of the 365 annual days (working and weekend days). For example in n days/month, all consecutive hospital admissions and deaths with eligible ICD codes can be validated.
External validity
It is not essential that the whole country should be covered by a surveillance system, but it is essential that the registration system of events should be complete with regard to events occurring in the target population. It is important to know how representative the register is for the whole country according to the CVD mortality rate, the distribution of age and sex, health determinants (socio-economic status and health behaviour), and the distribution of health service (specialized hospital and GP).
For the population chosen, there must be good demographic data subject to at least annual revision; inaccuracies might become apparent years after the period being studied because of the results of a decennial national census. A careful description of the population characteristics can help describe how representative the target population is for the whole country.
Methods to evaluate the diagnostic quality
Using the diagnostic criteria, it is possible to evaluate whether the diagnostic tools used to establish the application of valid methods are different for hot pursuit and cold pursuit. Validation of the diagnostic information recorded in the register can include the examination either of all the events or of random samples. The relevant register data must be checked periodically by sampling, as it is usually not feasible to check all the data [31]. Validation has to be carried out by an epidemiological team not involved in the patient's treatment. For local registers with a limited number of cases, it might be possible to validate each single event, whereas national registers (for practical reasons) can only validate data on the basis of random samples of suspected cases recorded during a selected period or during some days each month. One selection method consists of choosing some days each month and evaluating all the events that have occurred in those days, extracted either from HDR or mortality records, applying diagnostic criteria. In this way, seasonal variation can be traced. The most important phase is the evaluation of the diagnostic information, although other information in the register also needs to be included in the validation.
To produce valid indicators, conditio sine qua non is to allow access to the relevant medical records and to the routine raw data of health statistics. In some cases it is possible to validate a register by linking the register to an independent data source, for example, a high quality register for a small area within the region.
Criteria for the validation of acute cerebrovascular events
This manual of operations does not aim to improve existing stroke definitions or to formulate new ones, but only to suggest a definition that already exists and to ensure comparability. According to the WHO criteria, stroke is defined as ‘rapidly developing clinical signs of focal (or global) disturbance of cerebral function lasting more than 24 h (except in cases of sudden death or if the development of symptoms is interrupted by a surgical intervention), with no apparent cause other than a vascular origin’ [19, 32]. Global clinical signs are accepted only in cases of subarachnoid haemorrhage or in patients with deep coma. Brain lesions detected by CT scan but not accompanied by acute focal signs are not accepted as stroke; neither are extradural and subdural haemorrhages. Stroke cases with concomitant brain cancer, trauma, or severe blood disorders are also excluded [19]. Therefore, the key features of the clinical definition are as follows:
sudden onset,
neurological deficit,
lasting 24 h or longer,
of presumed vascular origin.
Table 7 below provides an example of some of the diagnoses that should be considered for stroke registration.
Focal and global signs that should be considered for stroke
A stroke case is recorded as fatal if death occurs within the first 28 days.
Ethical issues
The Helsinki Declaration requires that biomedical research with human participants must conform to generally accepted scientific principles.
The ‘Recommendation n. R (97)5 of the committee of ministers to the EU member states on the protection of medical data’ [33] gives the guidelines as to how medical data can be registered, stored, and used in a way that ensures the rights and the fundamental freedoms of the individual, in particular the right to privacy (adopted by the Committee of Ministers on 13 February 1997 at the 584th meeting of the Ministers’ Deputies).
In the following passages, the most important recommendations are presented.
‘Medical data should be collected and processed only by health-care professionals, or by individuals or bodies working on behalf of health-care professionals. Individuals or bodies working on behalf of health-care professionals who collect and process medical data should be subject to the same rules of confidentiality incumbent on health-care professionals, or to comparable rules of confidentiality.’
Therefore, it is essential that a neurological or stroke physician (or study nurse) with proven experience in the field of cerebrovascular should be involved in the coordination of the stroke register.
‘Medical data shall be collected and processed fairly and lawfully and only for specified purposes.’
‘Medical data may be collected and processed:
if provided for by law for:
public health reasons; or
subject to Principle 4.8 (processing of genetic data for the purpose of a judicial procedure or a criminal investigation should be the subject of a specific law offering appropriate safeguards), the prevention of a real danger or the suppression of a specific criminal offence; or
another important public interest; or
if permitted by law:
for preventive medical purposes or for diagnostic or for therapeutic purposes with regard to the data subject or a relative in the genetic line; or
to safeguard the vital interests of the data subject or of a third person; or
for the fulfilment of specific contractual obligations; or
to establish, exercise or defend a legal claim; or
if the data subject or his/her legal representative or an authority or any person or body provided for by law has given his/her consent for one or more purposes, and in so far as domestic law does not provide otherwise.’
Whenever possible, medical data used for scientific research purposes should be anonymous. Professional and scientific organisations as well as public authorities should promote the development of techniques and procedures securing anonymity. However, if such anonymisation would make a scientific research project impossible, and the project is to be carried out for legitimate purposes, it could be carried out with personal data on condition that:
the data subject has given his/her informed consent for one or more research purposes; or
when the data subject is a legally incapacitated person incapable of free decision, and domestic law does not permit the data subject to act on his/her own behalf, his/her legal representative or an authority, or any person or body provided for by law, has given his/her consent in the framework of a research project related to the medical condition or illness of the data subject; or
disclosure of data for the purpose of a defined scientific research project concerning an important public interest has been authorised by the body or bodies designated by domestic law, but only if:
the data subject has not expressly opposed disclosure; and
despite reasonable efforts, it would be impracticable to contact the data subject to seek his consent; and
the interests of the research project justify the authorisation; or
the scientific research is provided for by law and constitutes a necessary measure for public health reasons.
Record linkage between mortality and HDR is possible in countries that have adopted a PIN on a national level. Other nominal data (such as name, sex, and date and place of birth) are usually available at a regional level. Record linkage is important to match admissions and discharges or admissions and deaths, thus avoiding double counting, which can occur when, for example, the same patient transferred to another ward (e.g. from neurology to neurosurgery and then to rehabilitation) is registered in the HDR more than once.
Moreover, the identification of the patient is essential for event validation, as it is necessary to collect and examine the history and clinical documentation and to assess case fatality at different intervals (6 months, 1 year). Before starting any study, it is recommended to seek approval from the local ethics committee.
Economic consideration
Stroke is a costly disease because of the large number of premature deaths, ongoing disability in survivors, and impact on families or caregivers and on health services (treatment and rehabilitation).
Stroke is estimated to cost the EU economy over h34 billion a year: around 1/5 of the overall cost of CVD. Of the total cost of stroke in the EU, around 62% is due to direct healthcare costs, 18% to productivity losses, and 20% to the informal care of people with stroke [1]. Cost considerations are essential before implementing a population-based register.
Without a valid surveillance system, it is not possible to plan and evaluate health services for populations, implement interventions for primary prevention, and identify ‘vulnerable subgroups’ in terms of burden of disease, such as the elderly, the young, the poor, and the unemployed. Surveillance and evaluation mean a systematic way of learning from experience. This learning should be used to improve current activities and promote better planning, by careful selection of alternatives and allocation of resources for future action. The economic benefit of a good surveillance system clearly exceeds the cost of the registers.
A population-based register might be expensive, and to produce meaningful data it needs to be in operation for at least 1 year, but preferably for more. The importance of a valid and efficient stroke register, however, justifies the high implementation costs and the consequent need to find adequate financing.
The register based on record linkage between administrative databases is the most cost-effective, but this register depends on the data quality of the HDR and the cause-of-death register and also on the possibility of a valid record linkage. In addition, methods need further evaluation and implementation. Notably, if the hospital discharge and mortality registers are available for record linkage, the costs for the linkage and dissemination of results will be low. The main costs for using this methodology for the assessment of stroke incidence in a defined population concern the need to perform regular validations of the diagnostic information. It is recommended to include a basic epidemiologic team in the cost. Sometimes access to data produces separate costs.
The register based on disease-specific data collection is more expensive, especially if hot pursuit is used. Besides the cost mentioned above, this type of register also needs funding for the detailed prospective data collection and for the validation of diagnostic information. Data collection includes the identification of patients, reading medical records, making enquiries of additional data sources, and filing and validation of the data. This means that a team of epidemiologists, nurses, medical doctors, and information technologists, which is dedicated to this work full time, is needed. To give an example, resources needed to run the MONICA project in northern Sweden for the stroke registration included one nurse working full time (i.e. 40 h/week), one medical secretary working 25% of full time, and 1 internist working 5% of full time [34]. It should be recognized that this type of register usually collects information that permits analyses of research questions beyond the monitoring of stroke incidence, mortality, and case fatality. This might concern either the role of risk factors for disease occurrence or the role of treatment for survival in stroke patients. In the northern countries, registers based on disease-specific data collection have for several years complemented national administrative registers in providing a comprehensive picture of the burden of stroke in the population.
Implementation: a stepwise procedure
This section describes the procedures needed to implement a stroke register, taking into account the recommendations reported in this manual of operations.
Step 1. Define the target population and routine data
Select a geographical administrative area with a population big enough to provide stable estimates. This means that a stable population in a representative area of the country with 300 fatal and nonfatal stroke events/year in the age range 45-74 years should be chosen.
Characterize the population from a demographic point of view through a detailed description of the characteristics of the population under surveillance, in particular, demographic characteristics (age and sex distribution), socio-cultural characteristics (educational level, occupation, social group, unemployment rate, migration, and immigrants with or without citizenship), characteristics of the healthcare system (specialized hospital, GP, and rehabilitation clinic), macro areas, and micro areas (urban and rural). Disease frequency is often different in the macro areas of the country; a description of the differences in mortality and risk factors allows for the selection of those areas for inclusion in the surveillance system. Within the population-based surveillance study, the phenomenon of immigration plays an important role; therefore, immigrants originating from European and extra-European countries, who are resident in the study area, must be enrolled. Geographical or administrative borders of the surveillance areas must be clearly defined.
Analyse existing hospital discharge and mortality data: events among nonresidents occurring in the study area or admitted to hospitals in the study area do not qualify. Events concerning residents occurring out of the area do qualify. Efforts must be made to find them or to estimate the potential loss and whether or not this could change and interfere with the validity of the observed trends in rates over a period of years.
Identify problems with these data: coverage, the ICD version, the ICD codes, procedures, DRGs, unit of analysis (number of events or discharges and/or number of patients), the PIN, coherence with previous studies, and others. Data files are often available in detailed forms at the regional level.
When a register is launched for the first time, a plan for future follow-ups to measure trends is recommended. This can be achieved by continuous surveillance as part of a broader health information system or by annual registration, repeated at 5-10-year intervals.
Step 2. Perform a pilot study and validate routine data
Before starting a stroke register or the large-scale use of linked administrative data, a pilot study on available hospital discharge and mortality data in a small area is recommended, to study the feasibility and to estimate internal validity.
Validation studies on available data include the following:
estimation of coverage: comparison of different routine data sets (electronic or manual), the number of patients treated in the area and out of it, hospital/mortality ratios, age and sex ratios, and principal vs. secondary and/or procedure diagnoses;
validation of discharge diagnoses according to a standard method (including revision and abstraction of medical records) in a random sample or in all cases (including a check of other related diagnoses);
validation of mortality causes according to a standard method in a random sample or in all cases;
analysis of the demography and the representativeness of the area in comparison with the region or country;
selection of the age range of interest (35-74 or 35-84).
Step 3. Carry out record linkage using administrative data
In the northern countries, where every citizen has a PIN included in the national registers of hospital discharges and deaths, record linkage for the identification of stroke events is efficient and reliable. In countries that have not adopted the PIN, it might be much more difficult to perform this step. Files have to be organized in the same format and with the same variables (family name, name, date of birth, residence, and place of birth).
The following preparatory tasks are recommended:
explore the feasibility of record linkage within hospital records - using either the probabilistic or deterministic approach, or using the PIN (within the same hospital, among hospitals of the area, and among hospitals on a regional or national level). When hospital records are collected at the regional or national level, it is possible to collect events that occur out of hospital;
explore the feasibility of record linkage between hospital records and the mortality register (probabilistic or deterministic approach or using the PIN);
explore the feasibility of linkage with other sources of information (e.g. GP or drug-reimbursement register). Not all GPs are organized in networking, with computerized documentation of patient history; when they are, the definitions of events rarely use the same diagnostic criteria.
Step 4. Set up a stroke register
After performing steps 2 and 3, it is possible to set up a population-based stroke register following (A) record linkage between administrative registers or (B) specific stroke register.
A. Register using routine administrative data based on record linkage:
when the linkage procedure between hospital discharge and mortality records is feasible, it is important to define the event, the duration, how to handle transfers between hospitals with differences in diagnoses between the admitting hospital and the hospital to which the patient is transferred, and how to define first time events, recurrent events, and fatal and nonfatal events. A linkage system and a control for duplicate records should be set up;
validation of diagnostic information is recommended in a random sample of sufficient size of the identified events, with the estimation of sensitivity, specificity, and positive predictive value of the defined events;
target population data by age and sex are needed to estimate incidence, recurrence, attack rate, case fatality, and mortality rates;
periodic validations should be performed.
B. These are the steps for setting up a specific stroke register:
set up a pilot population-based register, with a proven standardized protocol, for stroke and evaluate the pilot study results (coverage, completeness of information, and diagnostic validity);
on the basis of the results of the pilot study, set up, if feasible, a full-scale register and decide whether to use hot or cold pursuit;
if feasible, then design the full-scale register (target population, data collection methods, and validation procedures).
These are the steps to set up a full-scale register:
select one or more populations representative for the region or the country;
for each selected population, set up a population-based register with an approved standardized protocol for stroke;
write a detailed protocol for the data collection including validation procedures for each single case;
evaluate the coverage, the representativeness, and completeness of information;
use the results from the register to validate the administrative data.
Step 5. Disseminate results
Set up a strategy for the analysis of data and for the dissemination of results.
Indicators of attack rate, incidence, case fatality, and others (defined in EUROCISS phase I) should be published yearly, for example, on a web site, according to sex, age, and other relevant characteristics. All indicators should be age-adjusted according to men and women standard European population.
Use data for research. This is very important, to ensure a high quality of the register over time. A high quality register can be the basis for good research.
Footnotes
Acknowledgements
This project has received financial support from the European Commission, the Directorate General for Health and Consumer Protection – Grant Agreement No. 2003118, and from the CUORE project of the Italian Ministry of Health.