Misclassification of deaths due to ill-defined UCD remains a problem in national vital statistics. Prominent among the ill-defined UCD is heart failure, which was listed as the UCD in over 3% of all vital records from 1999 to 2010. On a national scale, this represents approximately 63,273 deaths in 2010 among United States decedents ≥55 years old. Deaths attributed to heart failure redistributed by coarsened exact matching were predominantly assigned to CHD, with variations in redistribution proportions by state, sex, and race. Similar redistribution patterns were seen for heart failure deaths classified by a physician reviewer panel following ARIC’s Community Surveillance protocol. These approaches reclassified 22% to 37% of deaths coded as heart failure to CHD deaths and classified more heart failure deaths as CHD in Mississippi and among males. CHD mortality increased by 0.7% to 13.7% among the four states after redistributing heart failure deaths, suggesting that without redistributing heart failure deaths, CHD mortality based on vital statistics is underestimated by 0.7% to 13.7% in the four states included.
As shown in our analysis and those of others [5, 6, 12], heart failure deaths are mainly redistributed to CHD in economically advanced, industrialized settings. Stevens et al. used coarsened exact matching and showed that heart failure deaths were primarily redistributed to CHD, COPD, diabetes, and hypertensive heart and kidney disease . Using a regression-based approach, Ahern et al. also redistributed most heart failure deaths to CHD  although all heart failure deaths were redistributed to CHD among males, and 87% of the heart failure deaths were redistributed to CHD and 13% to COPD among females, which was not the case in the present study. An advantage of coarsened exact matching in this regard is that no a priori assumptions about the UCD target groups are necessary, and a heart failure death may be redistributed to several underlying causes of death. Stevens et al. reported that the overall percent increase in mortality for CHD after coarsened exact matching did not vary over time between 1999 to 2004 in the United States . To this we add the hypothesized observation of state-level variation in the percent increase in CHD due to redistribution. Furthermore, the percent increase in CHD mortality after redistribution was higher in males compared to females from 1990 to 2002, whereas from 2002 to 2010 females had a higher percent increase. The larger decline in CHD mortality from 2001 to 2003 among males compared to females in our analysis may contribute to this pattern.
Misclassification of the UCD due to ill-defined codes has been reported to vary by location [2, 17], the certifying physician , sex, race , and also with older age of the decedent [2, 5]. Murray et al. reported that in-hospital deaths and deaths in counties with more cardiologists per capita were less likely assigned to heart failure or general atherosclerosis compared to CHD . Misclassification of heart failure could also be due to differences in the training of nosologists by state and the use of commercial firms to perform the coding.
There is plausibility for heart failure deaths to be redistributed to CHD where the predominant underlying mechanism of heart failure is ischemic cardiomyopathy [19, 20], and the majority of deaths coded as heart failure or other ill-defined cardiovascular UCD are redistributed to CHD when adjudicated by an expert panel . We compared the statistical approach to redistributing heart failure deaths to a standardized and rigorous method of reclassification by a panel of physicians, based on reviews of medical records and surveys of the decedent’s physician and next of kin. The statistical approach redistributed a higher percentage of heart failure deaths to CHD than the panel of physicians, but the latter identified some records as unclassifiable. Both methods resulted in similar redistribution patterns and yielded comparable mortality estimates that reduce the impact of misclassification of the UCD in the medical certification of deaths.
Our study is limited by the lack of autopsy data to determine the true UCD, given that very few autopsies are conducted in the United States. Also to be considered, the method we used assigns deaths to plausible UCD and records could therefore be redistributed to different UCD to account for the uncertainty of the true UCD. Our analysis is based on vital records that may have inaccurate demographic information and cause of death coding; however, redistribution of heart failure offers a way to improve the misclassification of the UCD and comparability of mortality statistics. Lastly, this analysis only included four states in the United States to allow for comparison with an established community surveillance program. Even within this constrained framework, we were able to detect geographic variations in the redistribution of heart failure deaths both from vital records and the ARIC study. We also were able to compare temporal trends in CHD mortality by race and sex to previous studies that used redistribution of deaths due to ill-defined causes based on vital records prior to 2008 [2, 5, 6, 12].
Efforts to improve vital statistics are not new. Other studies have shown feasible ways to calibrate mortality to account for ill-defined UCD [2, 6, 12, 21], but these methods are not commonly used or cited. Considering that cause-specific mortality estimates for heart failure as the UCD are available on CDC Wonder, there is a clear need to inform public health researchers of the potential for misclassification of the UCD due to ill-defined causes such as heart failure when evaluating mortality statistics, as well as a need for practical tools to calibrate such data. Emphasis needs to be on improving the quality of medical records and the training of physicians on death certification and reporting multiple cause of death information to provide a more accurate estimation of the burden of disease , especially since diseases and conditions frequently co-exist. Despite the possibilities of inaccuracy and misclassification, the ICD system is used around the world to monitor diseases and for public health policies.