Background: Almost 400,000 deaths are registered each year in Thailand. Their value for public health policy and planning is greatly diminished by incomplete registration of deaths and by concerns about the quality of cause-of-death information. This arises from misclassification of specified causes of death, particularly in hospitals, as well as from extensive use of ill-defined and vague codes to attribute the underlying cause of death. Detailed investigations of a sample of deaths in and out of hospital were carried out to identify misclassification of causes and thus derive a best estimate of national mortality patterns by age, sex, and cause of death.
Methods: A nationally representative sample of 11,984 deaths in 2005 was selected, and verbal autopsy interviews were conducted for almost 10,000 deaths. Verbal autopsy procedures were validated against 2,558 cases for which medical record review was possible. Misclassification matrices for leading causes of death, including ill-defined causes, were developed separately for deaths inside and outside of hospitals and proportionate mortality distributions constructed. Estimates of mortality undercount were derived from "capture-recapture" methods applied to the 2005-06 Survey of Population Change. Proportionate mortality distributions were applied to this mortality "envelope" and ill-defined causes redistributed according to Global Burden of Disease methods to yield final estimates of mortality levels and patterns in 2005.
Results: Estimated life expectancy in Thailand in 2005 was 68.5 years for males and 75.6 years for females, two years lower than vital registration data suggest. Upon correction, stroke is the leading cause of death in Thailand (10.7%), followed by ischemic heart disease (7.8%) and HIV/AIDS (7.4%). Other leading causes are road traffic accidents (males) and diabetes mellitus (females). In many cases, estimated mortality is at least twice what is estimated in vital registration. Leading causes of death have remained stable since 1999, with the exception of a large decline in HIV/AIDS mortality.
Conclusions: Field research into the accuracy of cause-of-death data can result in substantially different patterns of mortality than suggested by routine death registration. Misclassification errors are likely to have very significant implications for health policy debates. Routine incorporation of validated verbal autopsy methods could significantly improve cause-of-death data quality in Thailand.