Data Availability

Several data sources are used for international comparisons

Data for international public health comparisons may be retrieved from several data sources. Among them:

• The databases of Eurostat.
• The WHO Health for All database (WHO-HFA database).
• The WHO European mortality database (WHO-MDB database).
• The OECD Health database.
• The database of EUCAN/GLOBOCAN (for cancer).
• European indicator projects, for example:
• The MINDFUL mental health indicator database.
• EUROTHINE - Tackling Health Inequalities in Europe; providing data on health inequalities.
• EUDIP/EUCID - European Diabetes Indicators Project and the European Core Indicators in Diabetes project; providing data on diabetes.

ECHI makes use of international databases. In cases where no regular/adequate data are available in these databases ECHI uses either project results as interim solution and/or national level data.

Availability of data for ECHI shortlist indicators

As shown by ECHIM project’s final report (Kilpeläinen et al., 2008), the availability of ECHI shortlist indicators varies between 60% and 100%, for a set of 18 countries. For quite a few countries, the actual availability turns out to be better when consulting national experts, than is apparent from the international databases of WHO, OECD and Eurostat. This implies that for several topics, countries do have data available, but these have not yet been sent or incorporated into the international databases.

The best available data are not always the best for international comparisons

Sometimes the preferred data for national trends are not the best data for comparisons with other countries. For example, the chronic disease prevalence; for comparability reasons EHIS is preferable, but for validity reasons a register based estimate, as done within the Eurostat morbidity strand, is preferable. The latter therefore will be more informative regarding the actual national situation.

Issues of data comparability

Common comparability issues

Data from different countries on the same indicator may have comparability problems for a variety of reasons. In many cases these problems are specific for a certain data source type:

Causes of death

Although the International Classification of Disease (ICD) is very precise, regional differences in medical practice and training may lead to differences in coding, especially in cases of multi-morbidity. This source of bias is decreasing and ICD-based data on causes of death are nowadays considered among the best comparable within the EU.

Another issue regarding ICD is that the change from ICD-9 to ICD-10 did not take place simultaneously in all countries. This may cause temporary differences (sometimes shown as trend breaks) or differences between specific countries.

Infant and perinatal mortality

There are variations among countries in what circumstances premature infants are reported as live births. Parameters such as gestational age, birth weight and plurality (e.g. twins, triplets) affecting perinatal and infant mortality rates. The interpretation of data is limited when different countries are using different limitation of gestational age and or birth weight to define live birth.

In addition, other factors might influence the variability of pregnancy outcomes e.g., the mother’s age, mother’s alcohol consumption, use of folic acid, whether screening for congenital anomalies widely used within a country etc. (Buitendijk et al., 2003; Lack et al., 2003). The PERISTAT project (Lack et al., 2003) has recommended a number of improvements in this politically sensitive area.

Disease-specific morbidity

In order to produce comparable data, the mechanism of data collection has to be included in the indicator definition. For example, the indicator ‘prevalence of diabetes’ can theoretically be derived from hospital discharge statistics, from primary care contacts, from population surveys (question ‘do you have diabetes?’), or from Health Examination Surveys in which blood glucose is measured and thus also previously unknown diabetes is recorded. These four data sources will produce different figures, as they basically measure different things.

Another problematic group of indicators is in the area of mental health and related conditions (e.g. depression, Alzheimer). Here, the samplings frame of a survey and the response behaviour (how to get mentally ill persons to respond to a questionnaire?), are crucial for the outcome. The ideal instrument for this area is a full-coverage disease register, as is used in many countries for certain cancers and for some communicable diseases.

Items derived from Health Interview Surveys

These normally include perceived and functional health, a range of health determinants, and issues of health services utilization. Problems may include difficulties in translating the same meaning into different languages, as well as cultural differences in interpreting linguistically similar expressions. Apart from that, a major issue is to arrange for representative and similar sampling frames, e.g. in terms of age groups and inclusion of institutionalized persons and minorities.

Data derived from medical registers

Hospital-based data, but especially data from primary care or outpatient registries, may lack precise comparability due to national differences in the organization of the health delivery system. Other sources of bias include different classification systems and coding rules, differences in admission and discharge practices, and different patient populations.

General

Comparability may be hampered by differences in the age composition of populations. For mortality, it is common practice to solve this problem by direct standardization. This is possible because the basic data for mortality are always age structured. For many other health issues the exact age is either unavailable, or the sample sizes are too small to make meaningful calculations.

Issues of data quality

Do the data measure what they intend to measure?

Quality aspects play a role in comparability. Aspects of data quality include:

• ‘validity’: this refers to the question of whether the instrument we use is adequate to measure the phenomenon we want to monitor. For instance, the ‘burden of disability’ can be assessed by measuring the number of people receiving disability benefits, but this tells us nothing about the burden of the disability in terms of functional capacity.
• ‘reliability’: this may relate to issues like sampling frame and size, statistical significance, representativeness, population background of the measurement, sensitivity to seasonal fluctuations, etc. For example do the trends and differences shown reflect real differences, and are they reproducible?