Lessons from History for Designing and Validating Epidemiological Surveillance in Uncounted Populations

Abstract

BACKGROUND: Due to scanty individual health data in low- and middle-income countries (LMICs), health planners often use imperfect data sources. Frequent national-level data are considered essential, even if their depth and quality are questionable. However, quality in-depth data from local sentinel populations may be better than scanty national data, if such local data can be considered as nationally representative. The difficulty is the lack of any theoretical or empirical basis for demonstrating that local data are representative where data on the wider population are unavailable. Thus these issues can only be explored empirically in a complete individual dataset at national and local levels, relating to a LMIC population profile. METHODS AND FINDINGS: Swedish national data for 1925 were used, characterised by relatively high mortality, a low proportion of older people and substantial mortality due to infectious causes. Demographic and socioeconomic characteristics of Sweden then and LMICs now are very similar. Rates of livebirths, stillbirths, infant and cause-specific mortality were calculated at national and county levels. Results for six million people in 24 counties showed that most counties had overall mortality rates within 10% of the national level. Other rates by county were mostly within 20% of national levels. Maternal mortality represented too rare an event to give stable results at the county level. CONCLUSIONS: After excluding obviously outlying counties (capital city, island, remote areas), any one of the remaining 80% closely reflected the national situation in terms of key demographic and mortality parameters, each county representing approximately 5% of the national population. We conclude that this scenario would probably translate directly to about 40 LMICs with populations under 10 million, and to individual states or provinces within about 40 larger LMICs. Unsubstantiated claims that local sub-national population data are "unrepresentative" or "only local" should not therefore predominate over likely representativity.