Hospital cause-of-death statistics: what should we make of them?

is the preferred standard for generating cause-of-death statistics. Most deaths whose cause is certified occur in hospitals. Although not all hospitals have the same diagnos -tic tools, it seems reasonable to expect hospital-based certifying physicians to correctly identify patients’ underlying causes of death, since hospitals usually have established clinical protocols for monitoring disease progression. After all, if physicians in hospitals cannot correctly ascertain their patients’ cause of death, who can?Cause-of-death statistics from hospitals are routinely amalgamated – along with mortality statistics from other sources – to constitute the es-sential statistics on the health of a population. Such statistics are widely used by governments, researchers, do-nors and global development agencies, often uncritically, on the assumption that they reliably capture a country’s epidemiological profile. They are used to periodically review health priorities, set research agendas and monitor progress towards national and global health and development goals. We take for granted that such data are correct. But are they?That the answer is a resounding no is only a part of the problem. More wor -rying perhaps is the fact that custodians of national mortality data systems fail to grasp the importance of periodically assessing the accuracy of hospital cause-of-death data. Physicians in hospitals may lack the time, incentives, diagnostic facilities or training to correctly certify causes of death and seldom understand that their diagnoses guide national health priorities. Medical records’ de-partments, which code death certificates and compile the data into cause-of-death tables by age and sex, are often understaffed, lack rigorous statistical protocols for checking data quality, and may not appreciate the epidemiological and statistical importance of their work. Thus, hospital cause-of-death data, even from advanced tertiary hospitals, can be expected to contain errors. But how common are these errors and are they serious enough to make the data unsuitable for informing public policy and research?Remarkably, the quality of hospital-based mortality data has seldom been investigated. In a recent review, Ram-patige et al.