American Journal of Epidemiology Advance Access originally published online on December 7, 2006
American Journal of Epidemiology 2007 165(5):591-596; doi:10.1093/aje/kwk042
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
PRACTICE OF EPIDEMIOLOGY |
Defining Cause of Death in Stroke Patients
The Brain Attack Surveillance in Corpus Christi Project
1 Stroke Program, University of Michigan Medical School, Ann Arbor, MI
2 Stroke Program, King Fahad Medical City, Riyadh, Kingdom of Saudi Arabia
3 Department of Epidemiology, University of Michigan School of Public Health, Ann Arbor, MI
4 Division of General Medicine, The University of Texas Medical School at Houston, Houston, TX
5 Division of General Medicine, Department of Internal Medicine, University of Michigan Medical School, Ann Arbor, MI
6 Veterans Affairs Center for Practice Management and Outcomes Research, Ann Arbor, MI
7 Institute for Social Research, University of Michigan, Ann Arbor, MI
8 Department of Health Management and Policy, University of Michigan, Ann Arbor, MI
Reprint requests to Dr. Lewis B. Morgenstern, TC 1920/0316, 1500 East Medical Center Drive, University of Michigan, Ann Arbor, MI 48109 (e-mail: LMorgens{at}umich.edu).
Received for publication May 22, 2006. Accepted for publication July 27, 2006.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
Stroke mortality is an important national health statistic and represents a frequent endpoint for epidemiologic studies. Several methods have been used to determine cause of death after stroke, but their agreement and reliability are unknown. Two hundred consecutive deaths of transient ischemic attack or ischemic stroke patients were identified (January 2000–September 2001) from an ongoing population-based stroke surveillance study in Texas, The Brain Attack Surveillance in Corpus Christi Project. Two neurologists independently recorded the cause of death based on two methods: 1) determining the underlying cause of death as defined by the World Health Organization, and 2) determining whether the death was stroke related. Kappa statistics with 95% confidence intervals were calculated by comparing agreement between methods within reviewers and between reviewers within methods. Agreement between the two cause-of-death-determination methods for each neurologist was 0.41 (95% confidence interval (CI): 0.31, 0.51) and 0.47 (95% CI: 0.38, 0.58), respectively. Agreement between neurologists for the underlying-cause-of-death method was 0.46 (95% CI: 0.32, 0.60); for the stroke-related method, it was 0.63 (95% CI: 0.52, 0.75). Accurate, reliable determinations of cause of death after stroke/transient ischemic attack are not currently feasible. More research is needed to identify a reliable process for coding cause of death from stroke.
cause of death; cerebrovascular accident; mortality
Abbreviations: CI, confidence interval; DNR, do not resuscitate; ICD-10, International Statistical Classification of Diseases and Related Health Problems, Tenth Revision; TIA, transient ischemic attack
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
Stroke mortality data are critical for monitoring disease trends and planning public health interventions. Stroke mortality is also an important outcome measure in stroke epidemiology studies and clinical trials. National health statistics rely on the underlying cause of death, defined by the World Health Organization as "the disease or injury which initiated the train of morbid events leading directly to death or the circumstances of the accident or violence which produced the fatal injury" (1). The focus on underlying cause of death is dictated by a public health perspective in which preventing the precipitating event should have the greatest impact on outcome. Although standardizing designations of underlying cause of death by the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) has been valuable, use of the ICD-10 nonetheless has challenges. Inaccuracies in the coding of deaths using the ICD-10 are recognized by the World Health Organization and are attributed to misunderstanding the rules, among other reasons (1). The rules for coding cause of death are complex and are outlined by the Centers for Disease Control and Prevention (Atlanta, Georgia) in a 241-page document (2). Unfortunately, few physicians receive formal training in determining underlying cause of death, contributing to errors (3).
Perhaps a more clinically relevant determination than the underlying cause of death after stroke is to ask whether the death would have occurred in the absence of the stroke. This type of "stroke-related" method has been used in previous stroke and cardiovascular epidemiology projects and represents a different frame of reference from the underlying cause of death (4, 5). To our knowledge, differences between these two methods of assigning cause of death have not been investigated previously. Furthermore, the interrater reliability of these methods has implications for their usefulness. We therefore sought to compare the results of cause-of-death determination between the "underlying-cause-of-death" and "stroke-related" methods and to assess their interrater reliability for transient ischemic attack (TIA) and ischemic stroke patients within a population-based stroke surveillance project. Secondarily, we compared the underlying cause of death assigned by the neurologist with the cause of death noted on the death certificate.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
Patient population
The Brain Attack Surveillance in Corpus Christi (BASIC) Project is an ongoing, population-based stroke surveillance study conducted in Nueces County, Texas. Detailed methods have been published previously (6, 7). This county is 56 percent Mexican American and 38 percent non-Hispanic White. This urban community is geographically isolated, with the largest nearby cities of San Antonio and Houston approximately 150 miles (241 km) away, and it serves as the regional referral center for the sparsely populated surrounding counties. This isolation allows for complete case capture of all cerebrovascular events. Both passive and active surveillance have been used to identify all nontraumatic cases of cerebrovascular disease in Nueces County residents older than age 44 years since January 1, 2000. All cases identified by trained abstractors are validated by study neurologists.
A predetermined convenience sample of the first 200 deaths beginning on January 2000 among patients already identified by the surveillance and validation process to have had a TIA or ischemic stroke is the subject of the current analysis. Standard clinical definitions of ischemic stroke and TIA were used (8). Although TIA itself cannot cause death, the risk of both cardiac ischemia and a recurrent cerebral ischemic event is very high in TIA patients (9, 10). The distribution of causes of death in TIA and stroke patients is similar (4, 11), and mortality analyses often combine data for TIA and ischemic stroke patients (12). Demographics and information on risk factors for stroke, including hypertension, coronary heart disease, high cholesterol, atrial fibrillation, diabetes, previous stroke or TIA, and current smoking, were obtained from chart review. This project was approved by the institutional review boards of the appropriate universities, the Corpus Christi/Nueces County Health District, and all Corpus Christi hospitals.
Deaths were identified from 1) routine active and passive surveillance of all in-hospital and emergency department deaths, 2) Texas Department of Health data for the state of Texas, 3) the Social Security Death Index, and 4) the Nueces County coroner. One year was allowed to pass before we accessed data from sources 2 and 3 to ensure complete capture of deaths. Since Nueces County is not an immigrant community, it is exceptionally unlikely that residents emigrate to another country to die. For almost all deaths in Nueces County, a death certificate is filed within 8 months of death. The Texas Department of Health has an aggressive policy to assure physician and funeral home compliance with this policy. The cause of death from the death certificates was ascertained from the ICD-10 code for the underlying cause of death. ICD-10 codes I60–69 were designated stroke, and all other codes were designated "other."
Source documentation from the initial TIA or ischemic stroke hospitalization, and subsequent hospitalizations, was obtained for assessment by reviewers. If the patient died in the nursing home, nursing home records were obtained when possible. When insufficient information was available to determine cause of death based on a prereview of charts, a physician questionnaire was sent to the patient's primary care physician. Cause of death was not coded if the neurologist found insufficient data on which to base a designation.
Underlying-cause-of-death method
All available medical record information, and nursing home notes for those who died in a nursing home, was examined to determine the underlying cause of death, defined by the Centers for Disease Control and Prevention and the World Health Organization as "the disease or injury which initiated the train of events leading directly to death" (13, p. 473; 2). When more than one event was determined to have contributed to the underlying cause of death, the degree to which the conditions contributed was considered to identify the "main disease or injury" that ultimately resulted in death (13). Two board-certified neurologists with stroke fellowship training reviewed the Centers for Disease Control and Prevention's instructions for classifying underlying cause of death (2) and completed an online tutorial (14) prior to independently reviewing charts and assigning an underlying cause of death. Underlying cause of death was considered stroke if the neurologist determined that the cause was ICD-10 codes I60–I69, and other if the cause was attributable to another ICD-10 diagnosis.
Stroke-related method
Stroke was determined as the cause of death in the "stroke-related" method when it was thought that the death would not have occurred in the absence of the stroke. Examples include 1) stroke-induced brain herniation or respiratory arrest from brainstem stroke, 2) aspiration or aspiration pneumonia due to dysphagia caused by stroke, 3) an immobilization-related event (such as pulmonary embolism from deep venous thrombosis, or sepsis from a decubitus ulcer) including those occurring in the nonacute phase after stroke, or 4) withdrawal of care because of stroke. Deep venous thrombosis and aspiration pneumonia that occurred after stroke-related symptoms had resolved were coded as non-stroke-related causes of death. After discussing these rules in detail and coding 10 sample charts together, the same study neurologists independently designated a cause of death based on this method as stroke or other.
Do-not-resuscitate (DNR) determination
Because DNR orders are associated with mortality from stroke (15), any reduction in aggressiveness of care, including withdrawal of support, DNR orders, or deferral of tests or treatments, was assessed based on source documentation. The timing of the limitation in care, whether it occurred prior to 24 hours or after 24 hours from stroke hospital presentation, was also determined.
Statistical methods
Frequencies and percentages were calculated for demographic variables and vascular risk factors. The proportion of deaths determined to be stroke related and the proportion with stroke as the underlying cause of death were calculated by the two neurologists. Two-by-two tables were generated to assess the percent agreement between methods within reviewers, the percent agreement between reviewers within methods, and the percent agreement between each neurologist and the death certificate regarding underlying cause of death. Unweighted kappa statistics with 95 percent confidence intervals were calculated for each comparison by using SAS 9.1 software (SAS Institute, Inc., Cary, North Carolina). Kappa statistics were repeated for patients with TIA and ischemic stroke separately. They were also repeated for early and late deaths separately, based on a median split of time to death (55 days).
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
Of the 200 patients identified, one, upon further review, was determined not to have had a cerebral ischemic event, leaving data on 199 for analysis. Cause of death could be determined for 186 (93 percent). For the 13 remaining patients, there was insufficient information for either neurologist to designate a cause of death by using either method. The mean age at time of death was 80 years. Eighty-seven (44 percent) were male, 103 (52 percent) were non-Hispanic White, 83 (42 percent) were Mexican American, 1 (0.5 percent) was Asian, and 12 (6 percent) were African American.
One hundred seventy-seven (89 percent) had had an ischemic stroke and 22 (11 percent) a TIA. At the time of the surveillance for ischemic stroke or TIA, 128 (64 percent) patients had a history of hypertension; 70 (35 percent), diabetes; 44 (22 percent), atrial fibrillation; 101 (51 percent), coronary disease; 83 (42 percent), prior stroke or TIA; and 16 (8 percent), current smoking. Location of death was as follows: 122 (61 percent), inpatient; 6 (3 percent), outpatient or emergency room; 40 (20 percent), nursing home; and 31 (16 percent), home. Regarding the undetermined causes, 4 patients died in a nursing home, 7 at home, and 2 in the hospital. Of the 186 for whom a cause could be determined, 23 (12 percent) were never designated DNR, 49 (26 percent) were designated DNR within or prior to the first 24 hours of TIA/ischemic stroke presentation, and 114 (61 percent) were designated DNR after 24 hours.
Stroke was determined by the two neurologists to be the underlying cause of death in 27 percent and 32 percent of cases, while it was determined that the deaths were stroke related in 59 percent and 60 percent of cases. On the basis of the death certificate, 40 percent of deaths were due to stroke.
There was 68 percent and 72 percent agreement between the underlying-cause-of-death and stroke-related cause designations for the two neurologists (table 1), with an interrater agreement measured by kappas of 0.41 (95 percent confidence interval (CI): 0.31, 0.51) and 0.47 (95 percent CI: 0.38, 0.58). There were many instances in which the stroke-related cause was stroke when the underlying cause was not stroke (neurologist 1: 59 (32 percent), neurologist 2: 52 (28 percent)). There were no cases, identified by either neurologist, for whom the underlying cause was stroke when the stroke-related cause was not stroke.
|
Agreement between neurologists was 77 percent for the underlying-cause-of-death method and 82 percent for the stroke-related method (table 2), with kappas of 0.46 (95 percent CI: 0.32, 0.60) and 0.63 (95 percent CI: 0.52, 0.75), respectively. Agreement between the death-certificate cause of death and the underlying cause of death determined by the neurologists was 69 percent and 68 percent (table 3), with kappas of 0.32 (95 percent CI: 0.19, 0.46) and 0.31 (95 percent CI: 0.18, 0.45), respectively. Kappa statistics by stroke type (ischemic stroke or TIA) and time of death (early or late) are found in table 4.
|
|
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
The use of underlying cause of death for national health statistics may significantly underestimate the impact of stroke on mortality. Compared with use of the underlying-cause-of-death method, determination of stroke-related deaths increased the deaths attributable to stroke by 100 percent. In general, the determination of underlying cause may underestimate stroke as the cause because of coding to a more general category such as atherosclerosis or circulatory disorder, or an ill-defined cause. Stroke-related deaths are coded with an underlying cause unrelated to stroke in specific circumstances when stroke is along the causal pathway. For instance, although a patient who died from an ischemic stroke related to atrial fibrillation clearly had suffered stroke-related death, on the basis of the underlying-cause-of-death instructions, the death would be ascribed to atrial fibrillation. Although conceptually correct that targeting the underlying cause of death will have the most substantial public health benefit, exclusively assessing underlying-cause-of-death statistics undervalues the importance of stroke's impact on mortality. Stroke's burden to society is also dramatically underestimated by exclusive use of the underlying-cause-of-death method. A more full appreciation for stroke-related mortality may influence policy makers in prioritizing resources and setting research agendas.
The reliability of data from vital statistics depends on the accuracy of death certificate data. However, many studies have documented the potential inaccuracies of death certificates for stroke and other diseases (16–20). It has been postulated that the inaccuracies are largely due to insufficient knowledge of the individual's medical condition rather than to a deficiency in the coding system (18). Yet, others have suggested that inaccuracies are due to insufficient training of physicians on cause-of-death determination (3). The current study shows that despite appropriate education on determining the underlying cause of death, and a detailed review of all available information, the underlying cause of death has an interrater reliability typically considered to be in the low "fair" range (21, 22) for patients with ischemic stroke and TIA, even when assigned by physicians with similar training backgrounds and with stroke expertise. The fair interrater reliability of underlying-cause-of-death designation and the "slight" (21) agreement with the death certificate calls into question the usefulness of underlying causes of death for national health statistics. Interestingly, although there was much disagreement with regard to the cause of death of individual cases by the two reviewers, the proportions of the total deaths determined to be from stroke, regardless of the method used, were quite similar between the two neurologists. However, this is certainly not sufficient to support the overall reliability of cause-of-death statistics in a population.
The results of the current study are troubling because many epidemiologic and genetic studies attempt to relate factors to stroke mortality (23–25). But, what is stroke mortality? Differing definitions of stroke as a cause of death in the literature make comparisons across studies challenging (4, 26, 27). This problem is further compounded by the low interrater reliability of cause-of-death determination identified in this study, despite the method used. If the coding of stroke as the cause of death is inaccurate, what should be done? Research studies in which stroke mortality is calculated should detail the process by which cause of death is determined. This process should be tested for interrater reliability, and these results should be reported. Disclosure of the method used will help facilitate comparisons across studies, because there is currently no standard definition used to report stroke mortality. When studies use vital statistics to calculate stroke mortality, caution should be used regarding how these data are interpreted. When appropriate, alternative outcomes to stroke mortality, such as new stroke or all-cause mortality, should also be considered. On a more global level, the rules for determining cause of death in stroke patients need to be revisited. Perhaps a more simplified protocol could be established that would have higher reproducibility. Regardless, more widespread teaching is needed related to coding underlying cause of death for death certificates in an attempt to standardize the process (28). Determining whether a death was stroke related has a higher interrater reliability (typically considered "moderate" (21)) than underlying-cause-of-death determinations and may be a more relevant outcome to stroke research goals.
Limitations to this study exist. Although the neurologists in the current study had a high chance of agreement because of similar training backgrounds and identical training in the process of coding underlying cause of death, it is possible that some of the disagreements were specific to these neurologists and that other physicians would have a higher level of agreement. Differences could also have stemmed from idiosyncrasies or ambiguities in the medical records available from the hospital systems specific to the Corpus Christi community. This possibility seems unlikely. Both physicians agreed that a cause of death could be designated for 93 percent of the cases. The other 7 percent of cases for whom disagreement should have been most likely because of insufficient data were excluded from interrater reliability testing. The current study's interrater reliability estimates may therefore be overestimates. The results of this study are based on a predetermined convenience sample of 200 cases, which may have resulted in a lack of precision.
Accurate, reliable determinations of cause of death after stroke/TIA are not currently feasible. More research is needed to identify a reliable process for coding cause of death from stroke.
|
ACKNOWLEDGMENTS |
|---|
Conflict of interest: none declared.
|
References |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONReferences |
|---|
- World Health Organization. Health statistics and health information systems. Available at (http://www.who.int/healthinfo/statistics/mortdata/en/index.html). Accessed September 26, 2005.
- Instructions for classifying the underlying cause-of-death, ICD-10, 2005 Centers for Disease Control and Prevention/National Center for Health Statistics. (http://www.cdc.gov/nchs/about/major/dvs/im.htm). Accessed June 27, 2005.
- Lakkireddy DR, Gowda MS, Murray CW, et al. (2004) Death certificate completion: how well are physicians trained and are cardiovascular causes overstated? Am J Med 117:492–8.[CrossRef][ISI][Medline]
- Hartmann A, Rundek T, Mast H, et al. (2001) Mortality and causes of death after first ischemic stroke: the Northern Manhattan Stroke Study. Neurology 57:2000–5.
[Abstract/Free Full Text] - Longstreth WT Jr, Bernick C, Fitzpatrick A, et al. (2001) Frequency and predictors of stroke death in 5,888 participants in the Cardiovascular Health Study. Neurology 56:368–75.
[Abstract/Free Full Text] - Morgenstern LB, Smith MA, Lisabeth LD, et al. (2004) Excess stroke in Mexican Americans compared with non-Hispanic Whites: The Brain Attack Surveillance in Corpus Christi Project. Am J Epidemiol 160:376–83.
[Abstract/Free Full Text] - Piriyawat P, Smajsova M, Smith MA, et al. (2002) Comparison of active and passive surveillance for cerebrovascular disease: The Brain Attack Surveillance in Corpus Christi (BASIC) Project. Am J Epidemiol 156:1062–9.
[Abstract/Free Full Text] - Asplund K, Tuomilehto J, Stegmayr B, et al. (1988) Diagnostic criteria and quality control of the registration of stroke events in the MONICA project. Acta Med Scand Suppl 728:26–39.[Medline]
- Johnston S, Gress D, Browner W, et al. (2000) Short-term prognosis after emergency-department diagnosis of TIA. JAMA 284:2901–6.
[Abstract/Free Full Text] - Elkins J, Sidney S, Gress D, et al. (2002) Electrocardiographic findings predict short-term cardiac morbidity after transient ischemic attack. Arch Neurol 59:1437–41.
[Abstract/Free Full Text] - Dennis M, Bamford J, Sandercock P, et al. (1990) Prognosis of transient ischemic attacks in the Oxfordshire Community Stroke Project. Stroke 21:848–53.
[Abstract/Free Full Text] - Touze E, Varenne O, Chatellier G, et al. (2005) Risk of myocardial infarction and vascular death after transient ischemic attack and ischemic stroke: a systematic review and meta-analysis. Stroke 36:2748–55.
[Abstract/Free Full Text] - Lindahl BI, Glattre E, Lahti R, et al. (1990) The WHO principles for registering causes of death: suggestions for improvement. J Clin Epidemiol 43:467–74.[CrossRef][ISI][Medline]
- Tutorial from the National Association of Medical Examiners (NAME). (http://www.thename.org/index.php?option=com_content&task=section&id=11&Itemid=36).
- Alexandrov AV, Bladin CF, Meslin EM, et al. (1995) Do-not-resuscitate orders in acute stroke. Neurology 45:634–40.
[Abstract/Free Full Text] - Sehdev AES and Hutchins GM. (2001) Problems with proper completion and accuracy of the cause-of-death statement. Arch Intern Med 161:277–84.
[Abstract/Free Full Text] - Iso H, Jacobs DR Jr, Goldman L. (1990) Accuracy of death certificate diagnosis of intracranial hemorrhage and nonhemorrhagic stroke. The Minnesota Heart Survey. Am J Epidemiol 132:993–8.
[Abstract/Free Full Text] - Corwin LE, Wolf PA, Kannel WB, et al. (1992) Accuracy of death certification of stroke: the Framingham Study. Stroke 13:818–21.
- Reggio A, Failla G, Patti F. (1995) Reliability of death certificates in the study of stroke mortality. A retrospective study in a Sicilian municipality. Ital J Neurol Sci 16:567–70.[CrossRef][ISI][Medline]
- Sington JD and Cottrell BJ. (2002) Analysis of the sensitivity of death certificates in 440 hospital deaths: a comparison with necropsy findings. J Clin Pathol 55:499–502.
[Abstract/Free Full Text] - Landis JR and Koch GG. (1977) The measurement of observer agreement for categorical data. Biometrics 33:159–74.[CrossRef][ISI][Medline]
- Altman DG. (1991) Practical statistics for medical research. (Chapman and Hall, London, United Kingdom).
- Walldius G, Aastveit AH, Jungner I. (2006) Stroke mortality and the apoB/apoA-I ratio: results of the AMORIS prospective study. J Intern Med 259:259–66.[CrossRef][ISI][Medline]
- Mitchell P, Wang JJ, Wong TY, et al. (2005) Retinal microvascular signs and risk of stroke and stroke mortality. Neurology 65:1005–9.
[Abstract/Free Full Text] - Paultre F and Mosca L. (2005) Association of blood pressure indices and stroke mortality in isolated systolic hypertension. Stroke 36:1288–90.
[Abstract/Free Full Text] - Sarti C, Rastenyte D, Cepaitis Z, et al. (2000) International trends in mortality from stroke, 1968 to 1994. Stroke 31:1588–601.
[Abstract/Free Full Text] - Vernino S, Brown RD Jr, Sejvar JJ, et al. (2003) Cause-specific mortality after first cerebral infarction: a population-based study. Stroke 34:1828–32.
[Abstract/Free Full Text] - Messite J and Stellman SD. (1996) Accuracy of death certificate completion: the need for formalized physician training. JAMA 275:794–6.[Abstract]
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||