Accurate cause of death data are critical to developing an effective health policy agenda. Just as clinicians must accurately diagnose patients before prescribing treatment, policymakers must assess population-level disease burden in order to prioritize health interventions . Despite this, the availability of high-quality, cause-specific mortality data is limited [2–6]. Even for countries with high vital registration coverage, the quality of the data is often substandard [2, 4–7]. Using the heart failure cause of death code as a case study, this article proposes a method to improve the quality of vital registration mortality data to provide policymakers with more accurate diagnostic datasets from which to prioritize health interventions.
In mortality data analysis, the underlying cause of death is of greatest importance [8, 9]. The World Health Organization (WHO) defines the underlying cause of death as "the disease or injury which initiated the train of morbid events leading directly to death, or the circumstance of the accident or violence which produced the fatal injuries" . The underlying cause must satisfy two criteria fundamental to this definition. Specifically, the underlying cause must be the primary or initial cause such that it represents a logical stage of intervention to prevent death, while retaining its status as a disease or injury to which death can be definitively assigned. One of the most significant barriers to accurate cause of death data is the widespread use of nonspecific cause of death codes, such as those for heart failure. These ambiguous or vague codes were first described as "garbage codes" in the Global Burden of Disease (GBD) 1990 study, in reference to ill-defined cardiovascular, malignancy, and injury codes . A recent paper by Naghavi et al stresses that garbage codes negatively impact the public health utility of cause of death data . This problem stems from the fact that the International Classification of Diseases (ICD), the coding scheme used in most countries with vital registration systems, contains many codes that signify signs, symptoms, and conditions, or intermediate and immediate causes of death . These codes are often misrecorded as an underlying cause of death on death certificates or in mortality datasets for a number of reasons [11–15]. Physicians do not always have access to accurate or complete medical records from which to determine the correct underlying cause of death [6, 15, 16]. They may be inadequately trained to complete death certificates , or, in some cases, may intentionally miscode deaths in the face of political, financial, or social pressure (i.e., HIV-related stigma). In developing nations, these realities are compounded by the fact that only a small percentage of death certificates are completed by physicians .
Heart failure is the end stage of many cardiac and noncardiac pathological processes, from ischemic heart disease and the range of cardiomyopathies to respiratory disease and severe anemia. As such, heart failure is not an underlying cause of death according to the WHO definition, but rather an intermediate cause of death with a diverse range of possible underlying causes of death. The use of heart failure as an underlying cause of death code leads to a vague and sometimes inaccurate representation of the population-level causes of death [2, 18]. Because prevention, detection, and treatment efforts for severe anemia and ischemic heart disease are different, it is essential for policymakers to know the true etiology of heart failure-attributed deaths. [Please note that for the rest of the paper, the term "heart failure-attributed deaths" refers solely to deaths in which heart failure was attributed as the underlying cause of death and not deaths in which heart failure is coded as a complication of the underlying cause in Part I of the death certificate or contributory causes in Part II of the death certificate.] In fact, Naghavi et al defines garbage codes, of which heart failure is most frequent, as "all deaths assigned to codes that should be redistributed to enhance the validity of public health analysis" . A process called heart failure redistribution, in which heart failure-attributed deaths in existing mortality data are reallocated to the correct underlying causes of death, can accomplish this.
Though the problem of miscoded mortality data is appreciated in the literature [2–6, 10, 12, 14–20], there is a paucity of redistribution methods described in the literature. Two methods used in the GBD literature are outlined below. GBD 1990 and 2001 redistributed aggregated groups of ill-defined cardiovascular codes (heart failure, atherosclerosis, essential hypertension, etc.) to one underlying cause of death, ischemic heart disease [2, 4, 18]. Less specific garbage codes, such as "other ill-defined, and unspecified causes of mortality (ICD-10, R99)," have been redistributed proportionally across all-cause mortality .
Building on these previous approaches, this paper describes in detail a novel method to carry out heart failure redistribution. This method was employed, but not described, in Naghavi's recent work . The method redistributes from a single garbage code - heart failure - to multiple underlying causes of death while accounting for the fact that clinicians miscode heart failure deaths at rates dis proportionate to the relative prevalence of its plausible underlying causes. It is important to note here that this method does not redistribute heart failure-attributed deaths proportionally across the underlying causes, a method that has been used in the past. We then compare cause of death patterns on a pre- and post-redistribution mortality dataset to show the effect of heart failure redistribution on age-adjusted death rates and cause of death rankings for the leading causes of death in 2005.