Abstract
Objectives:
Mortality due to rare diseases, which are substantial sources of premature mortality, is underreported in mortality studies. The objective of this study was to determine the completeness of reporting systemic lupus erythematosus (SLE) as a cause of death.
Methods:
In 2017, we linked data on a Swedish population-based cohort (the Swedish Lupus Linkage, 2001-2013) comprising people with SLE (n = 8560) and their matched general population comparators (n = 37 717) to data from the Cause of Death Register. We reviewed death records of deceased people from the cohort (n = 5110) and extracted data on patient demographic characteristics and causes of death. We estimated odds ratios (ORs) and 95% confidence intervals (CIs) for not reporting SLE as a cause of death by using multivariable-adjusted logistic regression models.
Results:
Of 1802 deaths among SLE patients in the study, 1071 (59%) did not have SLE reported on their death records. Most SLE decedents were aged 75-84 at death (n = 584, 32%), female (n = 1462, 81%), and born in Nordic countries (n = 1730, 96%). Decedents aged ≥85 at death were more likely to have SLE not reported on their death records than were decedents aged <50 (OR = 2.34; 95% CI, 1.48-3.68). Having renal failure listed as a cause of death decreased the likelihood of SLE not being reported on the death record (OR = 0.54; 95% CI, 0.40-0.73), whereas having cancer listed as a cause of death increased this likelihood (OR = 2.39; 95% CI, 1.85-3.07).
Conclusions:
SLE was greatly underreported as a cause of mortality on death records of SLE patients, particularly in older decedents and those with cancer, thereby underestimating the true burden of this disease. Public health resources need to focus on improving the recording of rare diseases in order to enhance the epidemiological utility of mortality data.
Keywords
Death certificates serve as legal documents and public health records, and they are often assumed to be accurate. Mortality statistics derived from death records are used to estimate the burden of disease in populations. 1 Mortality rates are important indicators of the health and well-being of a population, providing information about the nature and efficacy of health care delivery systems. 1,2 Mortality statistics provide data on demographic characteristics that are used to estimate population projections for policy planning purposes. In addition, data on cause of death can help policy planners plot a country’s trajectory through an epidemiological transition. 1,3,4 Cause-specific mortality data are also important for elucidating etiology in epidemiological studies. For example, the causal association between exposure to asbestos and high death rates was initially recognized in people who worked in industries with elevated concentrations of asbestos particles. 1,5 These people not only had a higher risk of premature mortality, but they also had a higher risk of dying from lung-related diseases than did the general population. 1,5
In many high-income countries, such as the United States and Sweden, cardiovascular disease (CVD) has been the leading cause of death for decades, followed by cancer. 6,7 Cause-of-death rankings indicate the mortality burden of one cause relative to other causes. However, mortality due to rare diseases that are substantial sources of premature mortality is often ignored in these mortality reports. 8 Rare diseases is a term that describes diseases such as systemic lupus erythematosus (SLE); they are considered rare because of their low estimated prevalence and incidence and, to a degree, their high morbidity and chronically debilitating disease courses. 8 In Sweden, prevalence estimates of SLE range from 46 to 85 per 100 000 population, depending on geographic region, case definition, and data source. 9
The limitations of mortality data sources in estimating deaths due to rare diseases must be scrutinized because, even in the widely researched field of cancer epidemiology, for example, it has long been acknowledged that deaths due to rare cancers (eg, cancers of the buccal cavity) are underreported compared with more common cancers (eg, breast cancer). 10,11 Underreporting of SLE on the death certificates of people with this disease may underestimate the true burden of disease, which jeopardizes the veracity of epidemiological studies. 12-14 These studies estimate that SLE is not recorded on the death certificates of SLE decedents approximately 40% to 70% of the time. Differences in the degree of underreporting of SLE on death certificates by sociodemographic characteristics and multiple causes of death could result in underestimates in certain population subgroups. 13
Major reasons that may contribute to the underreporting of SLE on death certificates include the following: proximate causes of death masking more distal causes of deaths, insufficient training in filling out death certificates, and disparate views of how comorbidities affect the ultimate cause of death. 15 Other factors implicated in underreporting may include variations in reporting proximate and distal causes on death certificates among clinicians, as well as imprecision or inconsistency in the criteria for SLE case definition. 15 Also, it is unknown whether trends in mortality rates using data from death certificates reflect true changes in the rate and not simply changes in disease recording practices on death certificates. 16 For these reasons, it is crucial to probe issues related to the validity of SLE reporting on death certificates.
In this study, we provide a population-based assessment of the completeness of SLE reporting in the Cause of Death Register and patterns of completeness in Sweden. We investigated factors associated with the reporting of SLE as a cause of death, evaluated patterns of accuracy, and examined the extent of the underreporting of SLE mortality.
Methods
Study Population
The study population comprised decedents from the Swedish Lupus Linkage (SLINK) cohort, described in detail elsewhere. 17,18 Briefly, Swedish national registers were used to identify people with prevalent SLE and matched (1:5) on birth year, sex, and county of residence to general-population (non-SLE) comparators (2001-2013). SLE was defined as ≥2 physician visits listing an International Classification of Diseases (ICD) code for SLE in the National Patient Register (a national database that houses all inpatient and specialist outpatient care in Sweden) (ICD, 8th Revision [ICD-8] 734.1; ICD, 9th Revision [ICD-9] 710.0; ICD, 10th Revision [ICD-10] M32) 19 –21 with ≥1 visit at a department or specialist typically known to diagnose, treat, or manage SLE (eg, rheumatology, dermatology, nephrology, internal medicine, or pediatrics). We included all prevalent SLE cases from 2001 onward identified in the SLINK cohort in the primary analysis. As a sensitivity analysis, we also examined newly identified and incident SLE patients from 2003 onward by allowing a minimum washout of 2 years since the outpatient component of the National Patient Register became available, to assess mortality among new cases. We obtained information on vital status for all people in the cohort through the end of 2013 via linkage to the Cause of Death Register, 22 which captures data on the underlying and contributory causes of death, along with data on other demographic characteristics.
Outcomes
We extracted data from the Cause of Death Register for decedents diagnosed with SLE using the National Patient Register (classified as SLE deaths; n = 1802) and for decedents from the group of general-population comparators without SLE in SLINK (non-SLE deaths; n = 3308). Data extracted from the Cause of Death Register included demographic characteristics (age at death, sex, region, place of death, and year of death) and underlying and contributory causes of death. We identified any mention of SLE and/or other causes of death, such as CVD or renal failure, by using ICD-10 codes. Then, we further grouped SLE decedents into those for whom SLE was reported as a cause of death and those for whom SLE was not reported.
Statistical Analysis
We derived descriptive statistics of SLE and non-SLE deaths. Among SLE deaths only, we examined differences between decedents with and without SLE reported in the Cause of Death Register by demographic factors. We built univariate logistic regression models with SLE not reported as the outcome to identify potential variables associated with SLE not being reported as a cause of death among lupus decedents in SLINK. We estimated odds ratios (ORs) and 95% confidence intervals (CIs) as measures of association. In constructing the final model, we used multivariable logistic regression, including the variables that were found to be significant in bivariate tests.
We examined SLE deaths recorded in SLINK and non-SLE deaths recorded in the SLINK general-population comparator. We evaluated whether SLE was reported as a cause of death by calculating the comparability ratio, defined as the ratio of SLE deaths as defined by the Cause of Death Register divided by SLE deaths as determined in the SLINK cohort. 23 This measure estimates the extent to which death records underestimate SLE mortality in mortality statistics. Finally, we examined the most common causes of death among decedents in SLINK whose death records did not report SLE.
Results
Of 1802 deaths among SLE patients in the study, 1071 (59%) did not have SLE on their death records (Table 1). For the 731 decedents for whom SLE was recorded, SLE was listed as the underlying cause of death in 250 (34%) death records and the contributory cause in 481 (66%) death records. For 2 deaths, SLE was erroneously recorded as a cause of death (decedent did not have SLE). The median age at death among all SLE decedents was 74; the median age of SLE decedents whose death records did not report SLE was 75, and among those for whom SLE was reported, the median age was 73. The median age at death among non-SLE decedents in SLINK was 79. A total of 1265 (38%) non-SLE decedents in SLINK were aged 75-84, and 2607 (79%) were female. Compared with all SLE decedents, non-SLE decedents in SLINK were more likely to be aged ≥85 at death (28% vs 17%). SLE decedents whose death records did not report SLE and non-SLE decedents in SLINK had, on average, 4 causes of death listed in the Cause of Death Register. SLE decedents whose death records reported SLE had an average of 6.3 causes of death listed. Most SLE decedents (n = 1730, 96%) were born in Nordic countries (Table 1).
Characteristics of decedents in the Swedish Lupus Linkage (SLINK) population, Sweden, 2001-2013a
Abbreviations: IQR, interquartile range; SLE, systemic lupus erythematosus.
b Non-SLE deaths refer to decedents from the general population comparators group in SLINK. The general population comparators group in SLINK comprises people without SLE who were matched 5:1 to SLE cases on birth year, sex, and county of residence at the time each case was diagnosed.
c SLE deaths refer to decedents who were diagnosed with SLE through International Classification of Diseases–coded visits in the National Patient Register before death.
d All values in parentheses are column percentages unless otherwise noted as SDs or IQRs.
e For SLE deaths in SLINK, this refers to the year during which they received their diagnosis using information from the National Patient Register. For non-SLE deaths, this refers to the year they were sampled from the general population and individually matched to a case with SLE in SLINK.
f Nordic countries include Sweden, Norway, Denmark, Finland, and Iceland.
Increasing age was associated with SLE not being reported as a cause of death (Table 2). Decedents aged ≥85 were more than 2 times as likely to have SLE not reported as those aged <50 (OR = 2.34; 95% CI, 1.48-3.68). Those living in southern Sweden had 37% lower odds of having SLE not reported as a cause of death compared with decedents in Stockholm (OR = 0.63; 95% CI, 0.44-0.89). Decedents who had renal failure listed as a cause of death were less likely than those without renal failure listed as a cause of death to have SLE not reported (OR = 0.54; 95% CI, 0.40-0.73). SLE decedents who had cancer listed as a cause of death were more than twice as likely as those who did not have cancer listed as a cause of death to have SLE not reported in the Cause of Death Register (OR = 2.39; 95% CI, 1.85-3.07).
Multivariate analyses of characteristics associated with not having systemic lupus erythematosus reported in death records in the Swedish Lupus Linkage (SLINK) population, Sweden, 2001-2013a
b Nordic countries include Sweden, Norway, Denmark, Finland, and Iceland.
The estimate of the overall comparability ratio, which compares SLE deaths defined by the death record, a proxy for the death certificate, with SLE deaths as determined in SLINK, was 0.41 (95% CI, 0.38-0.43) (Table 3). Mortality was underestimated by 68% among those aged ≥85 and by 49% among those aged <50. The comparability ratio did not vary by year of diagnosis, sex, or marital status. Mortality in Stockholm was underestimated by 65% compared with about 56% to 59% in other parts of the country. Among SLE deaths with cancer listed as a cause of death, the underestimation of SLE mortality was 73% compared with 44% for those with renal failure.
Comparability ratioa of systemic lupus erythematosus (SLE) deaths as defined by the Cause of Death Register to SLE deaths as determined in the Swedish Lupus Linkage (SLINK) population, Sweden, 2001-2013b
a Calculated as the ratio of SLE deaths as defined by the Cause of Death Register, divided by SLE deaths as determined in the SLINK cohort. This measure is an estimation of the extent to which the death certificate underestimates SLE mortality in mortality statistics.
CVD was the most common cause of death among decedents without SLE listed as a cause of death (Table 4). Heart failure was the most prevalent cause of death, comprising 243 of 1071 (23%) SLE deaths for which SLE was not reported. Other top causes of death in this group were pneumonia (11%), secondary malignant neoplasms (8%), malignant neoplasms of the bronchus and lung (8%), chronic obstructive pulmonary disease (8%), and myocarditis (7%). Neoplasms were associated with earlier age at death (ie, deaths occurring in those aged late 60s to early 70s), and CVD was associated with later deaths (in those aged late 70s and early 80s); the median age at death was slightly higher among non-SLE decedents in SLINK (ie, the general-population comparator). Among those with CVD, the difference in median age between SLE decedents (SLE not reported) and non-SLE decedents, as previously defined, was approximately 2-5 years.
The 10 most common causes of death among decedents with systemic lupus erythematosus (SLE) in the Swedish Lupus Linkage (SLINK) population for whom SLE was not reported as a cause of death in the Cause of Death Register, and number of non-SLE deathsa for each cause of death in SLINK, Sweden, 2001-2013b
Abbreviations: ICD-10, International Classification of Diseases, 10th Revision; IQR, interquartile range.
a Non-SLE deaths refer to decedents from the general population comparators group in SLINK. The general population comparators group in SLINK comprises people without SLE who were matched 5:1 to SLE cases on birth year, sex, and county of residence at the time each case was diagnosed.
c International Classification of Diseases, 10th Revision. 21
When we limited our sensitivity analyses to incident SLE cases, we found similar results as in the prevalent cohort. However, most findings were not significant. In the inception cohort, we identified 346 deaths among SLE patients, 218 (63%) of which did not have SLE reported as a cause of death. The median age at death was 74 among all SLE decedents; this number did not vary between those with SLE reported or not reported in their death records. In the multivariable models, age, sex, country of birth, and marital status were associated with SLE not being reported in the Cause of Death Register in the inception cohort; however, these associations were not significant at the α = 0.05 level. SLE decedents with cancer listed as a cause of death were nearly 5 times as likely as SLE decedents without cancer listed as a cause of death to have SLE not reported as a cause of death (OR = 4.86; 95% CI, 2.51-9.41). The comparability ratio estimate was 0.37 (95% CI, 0.32-0.42) in the inception cohort, indicating an underestimation of SLE mortality by 63% when death record data were used. In an analysis by sex, mortality was underestimated by 61% among females and 71% among males. By region, mortality was underestimated by 50% in southern Sweden and by more than 70% in the western part of Sweden. Mortality was underestimated by 56% among those with infections listed on their death record and by 83% among those with cancer. Much like the prevalent cohort, top causes of death were CVD and neoplasms, and the median ages at death related to these conditions were similar to those observed in the prevalent cohort (data not shown).
Discussion
We found that among approximately 3 in 5 deceased SLE patients, death records did not report SLE. One cohort study in the United States published in 2005 found that among 40% of deceased SLE patients, SLE was not reported on the death certificate. This level of underreporting may vary by sociodemographic factors and is particularly more frequent among patients who are older or uninsured. 12,14 A second study presented in 2017 as a conference abstract that linked the population-based Georgia Lupus Registry to the local death registry found that SLE was recorded on only 25% of SLE patient death certificates. 14
Our study and others 12,14 showed that older age best predicted underreporting of SLE as a cause of death. As a result, life expectancy for those with SLE estimated using only the death data may appear artificially lower compared with their actual life expectancy—a troubling error because tremendous strides have been made in recent years to increase survival among people with SLE. 24 –26 Our findings indicated no difference in age at death between those for whom SLE was not reported in the Cause of Death Register and the general population. This finding may explain why the proportion of underreporting on death certificates was high among older decedents, who may die of diseases perceived to be related to normal aging (eg, CVD) instead of dying of complications from accrual of SLE damage. The mean age at death for SLE decedents in Sweden in our population-based data was in the early 70s, which was higher than the mean age at death in other countries (approximately mid-60s). 9,27,28
Our study showed that relying on death records alone to estimate the burden of mortality of rare conditions such as SLE could lead to erroneous estimates. A natural corollary is to ask whether the absence of SLE on death records is a result of physicians being unaware of the accelerated mortality among people with SLE. 8
Death records are rich resources for estimating the burden of prevalent chronic diseases but may not be appropriate for understanding the epidemiology of rare diseases. Most rare diseases are either underdiagnosed or categorized by coding systems under the nebulous title “other.” 8 For example, a 2009 study of discrepancies between hospitalization records and mortality data in Sweden found that the recording of malignant neoplasms and ischemic heart disease had the highest accuracy of death certificate recording (90% and 87%, respectively) and “other conditions” had low accuracy (58%). 29 Several studies showed that death certificates tend to capture data on CVD mortality more accurately than the data on mortality from other conditions, especially for out-of-hospital deaths. 23
To solve the problem of underreporting of rare diseases, systems-level changes need to be made (by clinicians, hospitals, medical examiners, and public health officials) to mitigate errors found on death certificates. Training clinicians in death certificate recording and emphasizing error checks by vital records registrars can improve clinicians’ awareness of how rare diseases can cause a chain of events that lead to death. Incorporating autopsy results whenever possible could also potentially improve the accuracy of death records.
A complementary hypothesis is that the underreporting of SLE in death records may reflect the perception that SLE does not contribute to the sequence of events leading to death. This issue is complex and has implications for epidemiological inquiry into the natural history of SLE. First, our study found that the leading cause of death among decedents for whom SLE was not reported on death records was CVD. This finding is concerning because the most plausible underlying cause of CVD in patients with lupus is often related to SLE. 30,31 The lack of attribution of the underlying cause of CVD in SLE decedents may lead to inadequate attention being paid to the contribution of SLE to premature CVD death.
Second, and more importantly, we found that the number of causes of death was lower among decedents for whom SLE was not reported in the death record than among those for whom SLE was reported. The multi-systemic damage associated with SLE includes extensive cardiovascular, hematologic, respiratory, renal, and neuropsychiatric manifestations. 32 Failing to record these conditions could impede the deepening of our understanding of SLE etiology and obscure novel areas for intervention; specifically, measures such as differential cause-specific survival, mortality rates, and premature mortality would be greatly underestimated. Our study also showed that SLE mortality among those with cancer (73%) and to a smaller extent among those with renal failure (44%) was underestimated. This latter finding is not surprising because one of the most severe manifestations of SLE is lupus nephritis. 24 However, studies show that SLE disease damage (in addition to immunosuppressive therapy used to treat lupus) is associated with elevated cancer risk. 33 Yet the contribution of lupus to cancer secondary to systemic inflammation has received inadequate attention compared with lupus mortality related to direct organ involvement.
Limitations
Our study had several limitations. First, we assumed that our register-based definition of SLE did not lead to misclassification. Previous work on the SLINK cohort demonstrated some misclassification, but certainly not enough to explain the findings in our study. 18 Second, available evidence on the conditions associated with a death event may be inaccurate, substandard, or inaccessible. 23
Conclusions
To our knowledge, our study is the largest demonstration of the accuracy of death registers in reporting SLE as a cause of death using population-based sources. Accurate death statistics are important to appreciate the burden of rare diseases. This study showed that SLE is greatly underreported in death records. As a consequence, the burden of SLE may be considerably underestimated when studies use death data. Alternative measures that enable recording of rare diseases would enhance the epidemiological utility of mortality data. 34 These measures could include the routine recording of the history of all chronic conditions, as well as those that are contributory causes of death, or enhanced and routine linkage of mortality data with disease registries and ascertainment across hospital and community-based services.
Footnotes
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
Dr Chaichian receives salary support via the John & Marcia Goldman Foundation and the LuCIN Lupus Clinical Investigators Network and has received previous salary support through the Genomics Institute of the Novartis Research Foundation. Dr Simard is supported by a career development award from the National Institute of Arthritis, Musculoskeletal and Skin Diseases, National Institutes of Health (NIH NIAMS K01-AR066878). Dr Falasinnu received grants from the National Heart, Lung, and Blood Institute during the conduct of the study.