Fifty Years of Death Certificates: The Framingham Heart Study

Since the Framingham Heart Study enrolled its first patient in 1948, we have learned much about the epidemiology and pathophysiology of the United States' deadliest malady: coronary heart disease. No single medical research effort has contributed more to our increased awareness and understanding of coronary artery disease than the Framingham Heart Study. The hundreds of scientific reports during the past 50 years that have been based on Framingham data have changed not only the way we think about our patients with coronary disease but also how we actually care for them. In this issue, Lloyd-Jones and colleagues [1] suggest that the Framingham Heart Study has also influenced our consideration of heart disease when determining why an individual patient has died.

The causes of death listed on death certificates are not always accurate [2-4]. When a person dies after a long, well-characterized illness, the cause of death on the death certificate is likely to be more accurate than when a death is relatively sudden and unobserved [4, 5]. Similarly, in the absence of adequate information, the more narrowly characterized the cause of death on the certificate, the more likely it is to be in error. Thus, more miscodings are likely when the cause of death is listed as acute myocardial infarction or chronic ischemic heart disease than when both of these causes of death are coded as coronary heart disease [6]. A physician may not be entirely sure whether a new acute myocardial infarction has occurred when a person dies suddenly outside of the hospital, but he or she may be relatively sure that the death was due to coronary heart disease when the person had a history of the disease. When a patient dies unexpectedly and has no history of coronary heart disease, however, assigning even a general cause of death becomes more difficult. Many causes may be reasonably excluded, depending on the person's medical history, but the uncertainty remains.

In addition to the inherent errors caused by inadequate information, physicians may not pay enough attention to entering the cause of death on death certificates. The certificate is a legal document and must be completed within a few days of a person's death. The effort to be as accurate as possible can be considerable, and such factors as hospital billing practices and concern expressed by the patient's family about the societal unacceptability of some causes of death may influence diagnosis. Coding of the underlying cause of death depends on the specific medical terms entered on the certificates and the sequence in which they are entered; these can make appreciable differences in the statistics for coronary heart disease [6].

The extent of misclassification must be determined because death certificate data are used to generate the national (and, indeed, international) cause-of-death statistics on which public health decisions are based. Lloyd-Jones and colleagues [1] attempted to quantify the errors. They compared the cause of death on the death certificate with coding by a physician panel for 2683 participants in the Framingham Heart Study. A panel of three physicians carefully reviewed all available data, including results of autopsies, and interviewed family members when appropriate. Independently of the cause listed on the death certificate, the panel classified deaths into six underlying causes: coronary heart disease, stroke, other cardiovascular disease, cancer, other, and unknown. Sudden death, or death within 1 hour of symptom onset, was generally attributed to coronary heart disease.

Among the 2683 deaths, death certificates classified the underlying cause as coronary heart disease for 35% of cases, whereas the physician panel ascribed coronary heart disease as the cause for 28% of the deaths. The physician panel agreed with only 635 (two thirds) of the causes of death listed on the death certificate (positive predictive value, 67.4%). The positive predictive value for coronary heart disease was better for men than for women at all ages. For the other causes of death, the death certificate and the panel agreed more frequently with respect to the total number of deaths assigned. Except for cancer, however, the positive predictive values (as derived from concordance between the codings on specific cases) were even lower than for coronary heart disease. For stroke, the positive predictive value was only 59%. In addition, only 7 causes of death were unknown according to the death certificate, whereas 242 were unknown on the basis of coding by the physician panel. Lloyd-Jones and colleagues conclude that coronary heart disease is over-represented as a cause of death on death certificates. They estimate this excess to be as high as 24% overall and much higher in persons dying at older ages. When deaths due to unknown cause are excluded from the analyses, the excess is approximately 8%.

The authors correctly point out several limitations of their study. First, there is no gold standard. Although the physician panel paid considerable attention to classifying deaths, the lack of important information cannot be overcome. The physician panel coded many deaths as due to unknown cause; this indicates that the panel avoided making wild guesses in the absence of sufficient data. Even so, we do not know how many of those deaths whose causes were classified as unknown were classified as such on the basis of relevant hospital or other medical records and autopsy data. Even when such information is available to the physician signing the death certificate, many cases are complicated; it is difficult to assign a death to just one cause, particularly in elderly patients [7].

Second, although the Framingham Heart Study has been spectacularly successful in developing our concepts of risk factors and heart disease prevention, it is a study of a single community that has distinctive characteristics. Whenever we generalize results obtained by studying a single community, we need to do so cautiously. A further limitation not mentioned by the authors is that when the data are broken down by age and sex, some of the numbers are relatively small. These small numbers are apt to be uncertain, and reliance on the actual percentages of predictive value or specificity would be wrong.

Despite misclassification of death on the death certificate, death certificate data remain important. They have been helpful in identifying risk factors and high-risk population groups; these findings have been confirmed over time and have correlated well with our knowledge of risk factors and trends. Time trends for validated deaths from coronary heart disease have been consistent with trends based on death certificate tabulations [8-10]. Of particular importance, such data have allowed us to see changes in causes of death over time. Despite the errors in death certificate coding, the Framingham Heart Study data do not show differential coding over time. Although not proof that the national time trends are valid, these data are consistent with such an interpretation.

We can, and should, do better in our coding of death certificates. Physicians must pay more attention to these valuable records, not only because doing so would improve epidemiologic studies but because physicians owe the families of deceased patients the best possible understanding of why death occurred. Better understanding of cause of death also cannot help but improve disease prevention and medical care; in this context, it is regrettable that autopsy is done for only 10% of deaths [11].

Even if the findings from the Framingham Heart Study can be extended to the rest of the United States and incidence of coronary heart disease as the cause of death is considerably overstated, coronary heart disease remains a major health problem. Despite a more than 50% decline in the age-adjusted mortality rate between 1963 and 1996 [12], it is the most common cause of death in the coding done by the panel of Framingham physicians and remains the most common cause of death according to national death certificate data. In the 50 years since the Framingham Study began, we have witnessed increasing containment of what was once called an epidemic. However, coronary heart disease remains a huge problem, and we must not lessen our efforts to prevent and treat this disease.

• Copyright ©2004 by the American College of Physicians