Lack of recording of systemic lupus erythematosus in the death certificates of lupus patients

Abstract

Objective. To determine to what extent the diagnosis of systemic lupus erythematosus (SLE) in deceased lupus patients is under-reported in death certificates, and the patient characteristics associated with such an occurrence.

Methods. The death certificates of 76 of the 81 deceased SLE patients from two US lupus cohorts (LUMINA for Lupus in Minorities: Nature vs Nurture and CLU for Carolina Lupus Study), including 570 and 265 patients, respectively, were obtained from the Offices of Vital Statistics of the states where the patients died (Alabama, Georgia, North Carolina, South Carolina, Tennessee and Texas). Both cohorts included patients with SLE as per the American College of Rheumatology criteria, aged ≥16 yr, and disease duration at enrolment of ≤5 yr. The median duration of follow-up in each cohort at the time of these analyses ranged from 38.1 to 53.0 months. Standard univariable analyses were performed comparing patients with SLE recorded anywhere in the death certificate and those without it. A multivariable logistic regression model was performed to identify the variables independently associated with not recording SLE in death certificates.

Results. In 30 (40%) death certificates, SLE was not recorded anywhere in the death certificate. In univariable analyses, older age was associated with lack of recording of SLE in death certificates [mean age (standard deviation) 50.9 (15.6) years and 39.1 (18.6) yr among those for whom SLE was omitted and included on the death certificates, respectively, P = 0.005]. Patients without health insurance, those dying of a cardiovascular event and those of Caucasian ethnicity were also more likely to be in the non-recorded group. In the multivariable analysis, variables independently associated with not recording SLE as cause of death were older age [odds ratio = (95% confidence interval) 1.043 (1.005–1.083 per yr increase); P = 0.023] and lack of health insurance [4.649 (1.152–18.768); P = 0.031].

Conclusions. A high proportion of SLE diagnoses are not recorded in death certificates. Older patients and those without health insurance are more prone to have SLE not recorded. These findings do have implications for the assessment of the impact of this disease in epidemiological studies conducted using vital statistics records.

Vital statistics records, such as death certificates, constitute an important source of information for epidemiological studies. However, physicians completing death certificates may be unaware of patients' underlying conditions, which may result in an under-reporting that jeopardizes the value of the data. This seems to be particularly important in multisystemic autoimmune diseases which are not generally felt to be a direct cause of death; physicians may attribute the cause of death to complications such as heart failure, sepsis or renal failure, omitting the underlying process in an important proportion of these patients.

Whether systemic lupus erythematosus (SLE) is recorded in the death certificate, as an immediate or underlying cause or as a significant condition present at the time of death, has not been studied to date. SLE patients can die from disease-related causes (e.g. renal failure due to lupus nephritis, or central nervous system involvement due to lupus) but a clear decline in the frequency of these disease-related events as cause of death in these patients has occurred during the last few decades [1]. Conversely, infections or acute vascular arterial events (myocardial infarction, stroke, etc.) may be felt to be unrelated to the disease itself, when in fact the disease or the treatments used for it predispose to them [2–4]. If lupus is not recorded in the death certificate, SLE will be under-ascertained in epidemiological studies based on these records, providing inaccurate data. We have now examined the degree of SLE under-reporting using data from two separate United States (US) lupus cohorts.

Patients and methods

The death certificates of all deceased lupus patients from two large US lupus cohorts (LUMINA, Lupus in Minorities, Nature vs Nurture and CLU, Carolina Lupus Study) were obtained from the Offices of Vital Statistics of the states where the patients died (Alabama, Georgia, North Carolina, South Carolina, Tennessee and Texas). Both cohorts included SLE patients as per the American College of Rheumatology criteria, aged ≥16 yr, and disease duration at entry into the cohort of ≤5 yr; the main characteristics of these cohorts have been previously published [5, 6]. At the time these analyses were performed, LUMINA included 570 patients (90% women). One hundred and ten patients (19%) were Hispanics from Texas, 90 (16%) Hispanics from Puerto Rico, 208 (37%) African Americans and 162 (28%) Caucasians. These patients were recruited at three different institutions, the University of Alabama at Birmingham (UAB), the University of Texas Health Science Center at Houston (UTH) and the University of Puerto Rico Medical Sciences Campus (UPR). Mean (s.d.) age at enrolment was 37.0 (12.6) yr. The mean time of follow-up from enrolment to the last visit was 53.0 (40.6) months; the mean disease duration (from the time of diagnosis to the time of enrolment) was 17.3 (16.1) months. For this set of analyses the patients from UPR were not included, as no deaths had occurred at the time of these analyses. CLU includes 265 patients (91% women), recruited in 60 counties of North and South Carolina. Sixty per cent of CLU patients are African American and 34% Caucasian. The mean age at enrolment was 40.0 (14.8) yr; the mean total follow-up time and disease duration at enrolment were 38.1 (10.2) and 14.8 (10.5) months, respectively.

All death certificates from patients known to be deceased in both cohorts were abstracted and information regarding patient demographics and immediate, underlying or contributing causes of death, as noted, was tabulated. The National Death Index was not searched for the purpose of this study; yet the patients status (alive or deceased) was known for nearly 85% of patients in both cohorts. Based on the information in the death certificates, the causes of death were categorized into five groups: infectious processes, cardiovascular disease (including acute coronary or cerebrovascular events, heart failure and thromboembolic events), malignancy, SLE itself, and others or unknown. We also recorded where the death had occurred (at a tertiary or community hospital or at home). In the US any physician whether or not involved in the care of the patient prior to his or her demise can complete the death certificate; however, the characteristics of this physician (seniority, specialty) are not included in the death certificate and thus this information is not available. Socioeconomic status variables (income, education and health insurance) were obtained from the cohort databases and included in the analyses. Income data were available only for the LUMINA patients; thus this variable was not included in the multivariable analyses.

Deceased patients in whom SLE was or was not recorded anywhere in the death certificate were compared with respect to different socioeconomic–demographic variables and attributable causes of death using standard statistical tests (χ² for categorical variables and Student's t-test for continuous variables). A multivariable logistic regression model was then performed in order to identify those factors independently associated with not recording SLE anywhere in the death certificate including cohort assignment as a covariate. In a second model the specific institution affiliation (UAB or UTH for LUMINA patients or CLU) rather than cohort assignment was entered to exclude its influence in the final results. In a third model, place of death was added as a covariate. In all analyses a P ≤ 0.05 was set as the level of statistical significance.

The study was conducted with the approval of the ethics committee at the participating institutions.

Results

At the time these analyses were performed there were 81 deceased patients (58 or 72% from LUMINA and 23 or 28% from CLU). Five patients' death certificates were not located at the states' offices where they had reportedly died and were excluded, leaving 76 for analysis. Mean age (s.d.) at the time of death was 43.9 (18.3) yr. Sixty-six (87%) of the deceased patients were women, 14 (18%) Hispanic (from Texas), 45 (59%) African American and 17 (21%), Caucasian [one additional patient (2%) was from another ethnicity]. Other demographic data are presented in Table 1. The leading attributable cause of death was cardiovascular (33%), followed by infection (23%) and SLE itself (23%). In 17% of the cases other or unknown causes were recorded in the death certificate. SLE was recorded as the immediate, underlying or contributing cause of death in 46 (60%) cases and omitted entirely in 30 (40%). In only 8 of the 46 patients in whom SLE was recorded, was SLE recorded as the immediate cause of death. If these cases are excluded, the degree of under-reporting is even higher (44%).

Univariable analyses

As shown in Table 1, patients in whom SLE was not recorded anywhere in the death certificate were older than patients in whom SLE was recorded [50.9 (15.6) yr vs 39.1 (18.6) yr; P = 0.005]. The group that did not have SLE listed in the death certificate tended to have an over-representation of Caucasian patients (33 vs 13%), of male gender (20 vs 9%), of those in whom cardiovascular disease was considered as the attributable cause of death (38 vs 30%) and of patients dying at home (60 vs 40%), but these differences were not statistically significant. This group of patients was also less likely to have health insurance (64 vs 52%). The level of education was comparable for both groups as was the proportion of patients with income below the poverty line, although these data were only available for the LUMINA patients.

Multivariable analyses

In the final multivariable model, age (older) [odds ratio (OR) (95% confidence interval, CI) = 1.043 (1.005–1.083) per yr increase; P = 0.023] and lack of health insurance [OR (95% CI) = 4.649 (1.152–18.768); P = 0.031] were independently associated with not recording SLE in the death certificates (Table 2). The results remained essentially unchanged whether institution (in the case of the LUMINA patients) (UAB or UTH or CLU cohort) or place where death occurred were entered as covariates in the models; in this last model, however, lack of health insurance did not reach statistical significance but the sample size was smaller, given that the information on place of death was missing in five additional patients (data not shown).

Discussion

Our data show that in 40% of deceased SLE patients, SLE is not recorded in the death certificate and that the level of under-reporting may vary according to specific socioeconomic and demographic variables, being more frequent in older patients and in those without health insurance. The fact that the level of under-recording was comparable in both cohorts and across different states suggests that this is not a local phenomenon, but probably one that occurs across the entire nation, and probably in other countries around the world.

If SLE is not recorded in death certificates, epidemiological studies using such data [3, 7–9] will be inaccurate. A good example of this is the elegant work recently published by Ward on mortality in SLE based on data from the Multiple Causes of Death files from the National Center for Health [10], which derives information from official death certificates. In this study a higher mortality rate was observed in White patients with lower education level compared with Whites with higher education levels, but the opposite pattern was observed in the African Americans. As an explanation for his findings, Ward hypothesized that SLE might be under-diagnosed or under-reported in death certificates of African Americans with low education levels. However, this supposition is not supported by our data; we did observe, however, a significant over-representation of patients without insurance among those in whom SLE was not recorded, which may explain the data from Ward's study since lack of insurance is more frequent in minority population groups such as African Americans [10]. Our study also shows a trend towards not recording SLE in patients with a clinical picture not perceived as characteristic of lupus; for example in older White men with cardiovascular disease.

Mortality in SLE has declined over the last few decades while the pattern of the causes of death in these patients has also changed with infections and cardiovascular problems being the leading causes of death in SLE [2, 4, 11, 12]. SLE by itself, or the treatments used for it, predispose to both [13, 14]. Not recording SLE in patients dying from infectious or cardiovascular disorders, because these processes are felt to be unrelated to SLE, precludes recognizing the full impact of SLE.

Our study is, however, not without limitations. First, despite including patients from two large cohorts, the total number of deceased patients included is relatively small and thus definitive conclusions cannot be made. Second, we did not search the National Death Index in order to ascertain the status of patients lost to follow-up; given that this is not a study of mortality per se, the omission of a few patients among the deceased is unlikely to have affected the results obtained. Finally, our study only includes the recording pattern of some states in the US; however, the data were comparable by cohort and institution and all major US ethnic groups were included.

In conclusion, SLE is frequently not included anywhere in the death certificates of deceased lupus patients. These findings have important implications for epidemiological studies of SLE based on death certificates and for correctly assessing the impact of the disease. Specific interventions aimed at educating physicians in how to accurately record data in death certificates and its importance are needed.

This work was supported by grants from the National Institute of Arthritis and Musculoskeletal and Skin Diseases no R01-AR42503 and General Clinical Research Centers nos M01-RR02558 (UTH) and M01-RR00032 (UAB), the Division of Intramural Research, NIEHS/NIH/DHHS and The Mary Kirkland Scholars Award Program (UAB) and a PANLAR Fellowship Program (UAB). The authors would like to acknowledge Drs William J. Koopman, Christine Parks and Matthew Longnecker for their most helpful comments and suggestions and Mrs Ella Henderson, A.A. for her expert assistance in the preparation of this manuscript.

The authors have declared no conflicts of interest.

References

Author notes

The University of Alabama at Birmingham, Birmingham, AL, 1Case Western Reserve University, Cleveland, OH, 2The University of Texas Health Science Center at Houston, Houston, TX and 3National Institute of Environmental Health Sciences, Durham, NC, USA.