To the Editor
The recent paper by Dr. Moolgavakar (Air Pollution and Daily Mortality in Two U.S. Counties: Season-Specific Analyses and Exposure-Response Relationships, Vol. 15, issue 9) examining the relationship between air pollution and daily mortality was most interesting. I believe that this paper, along with others that utilize mortality data to determine the power of associations using statistical methods, needs to step back and scrutinize the validity of the data. The author states that he “obtained daily counts of total mortality” and then “extracted the daily counts of deaths due to diseases of the circulatory system.” Following this, he proceeds to analyze all the circulatory system deaths in one broad group to increase the power of the season-specific analyses.
Mortality data are almost always dependent on the accuracy of death certificates. The certificate must state the cause of death in accord with established codes and other pertinent contributing pathology that played a role in death. This sounds simple but in reality, how are death certificates subject to quality assurance? My personal experience over a 53-year career in pathology leads me to conclude that the error rate can be significant. The major reasons are usually dependent on: (1) Where did the individual die, in a hospital, at home, or elsewhere? (2) Was the physician who signed the certificate one who knew the deceased or not? (3) If not, what evidence was available to make an accurate diagnosis? (4) If the death occurred in a hospital, was the patient's chart peer-reviewed to validate the cause of death? (5) Once death certificates are received by the National Center for Health Statistics, how do they determine the accuracy of the information? (6) Many physicians, unfortunately, are fearful of potential litigation and give as a cause of death the least disturbing diagnosis to the family. Because death due to cardiovascular disease is listed as the leading cause of death in the United States, families are not perturbed if this is the diagnosis.
The national death rate hovers around 2.2 million a year. With a current autopsy rate between 8 and 10% based on estimates, death certificates are largely based on clinical information and the best judgment of the physician at the time of death. Of these who have postmortem examinations, the largest number is usually forensic in nature, carried out only by certified pathologist or medical examiners. Other physicians selected by local coroners are also employed. Since teaching hospitals are usually the place for autopsies to occur, the rate drops off significantly in smaller community hospitals. We look to Europe for autopsy statistics, where the rate is higher. It would be interesting for statisticians to study cardiovascular mortality in other countries compared to the United States.
The error rates between death certificate data and autopsy data as to cause of death ranges from 13 to 25%, depending on the institution reporting. Today, in the United States, there are hospitals that carry out practically no autopsies, and medical school centers where the rate is 10%. When, some time ago, I asked a highly placed official in the Centers for Disease Control about these issues, the answer I received was, “Death certificates are the best we have to determine why people die.”
It is with a sense of skepticism that I believe that the raw data used for mortality correlations in the United States today can be statistically manipulated to yield factual results.
Notes
BernardM.WagnerMDEmeritus Research Professor of Pathology New York University Medical Center