Defining cause of death in a contemporary cohort with ANCA-associated vasculitis (AAV): a comparison of electronic health record and death certificate data

Abstract

Objectives:

Patients with antineutrophil cytoplasmic antibody (ANCA)-associated vasculitis (AAV) face excess mortality compared with the general population. Mortality in clinical epidemiology research is often examined using death certificate diagnosis codes; however, the sensitivity of such codes in AAV is unknown.

Methods:

We performed a retrospective cohort study using the Mass General Brigham AAV Cohort, including patients with AAV who died between 2002 and 2019. Causes of death were determined by electronic health record (EHR) review (reference gold standard) and via cause of death diagnosis codes on death certificates. We calculated the sensitivity of death certificate diagnosis codes for AAV.

Results:

Of 684 patients in the registry, 184 died, 92 (52%) of whom had adequate EHR data available determine cause of death and 72 (40%) of whom had both EHR and death certificate data available. Death due to AAV, infection, cardiovascular disease, and cancer occurred in 8%, 29%, 5%, and 18%, respectively, when ascertained by manual review, as opposed to 0%, 11%, 25%, and 21%, as determined by death certificates. The sensitivity of AAV diagnosis codes for AAV was 16.6% (95% CI: 10.5, 22.6) among all patients with death certificate data available.

Conclusion:

In a contemporary cohort of patients with AAV, infection was the most common cause of death, while death due to AAV itself was rare. We found a high degree of discordance between causes of death determined by manual review and death certificate diagnosis codes. Mortality research on AAV should include linkage to medical records data to reduce potential bias.

Objectives

Antineutrophil cytoplasmic antibody (ANCA)-associated vasculitis (AAV) is a group of related severe systemic immune-mediated diseases that includes granulomatosis with polyangiitis (GPA), microscopic polyangiitis (MPA), and eosinophilic granulomatosis with polyangiitis (EGPA) [1]. Once an almost universally fatal diagnosis, mortality due to AAV has improved substantially over time. Nevertheless, mortality in AAV remains higher than in the general population [2–10], and contemporary studies report infection and cardiovascular disease as the leading causes of death in AAV [2–4,10–13]. To improve outcomes, we must understand the contemporary trends in mortality among people with AAV, including elucidating the factors directly contributing to death since these may be targets for intervention. The reliability of death certificate data for these tasks is unclear.

Prior studies evaluating mortality in AAV have used diagnosis codes (i.e., International Classification of Diseases (ICD)-9 and -10 codes) on death certificates to ascertain the cause of death in patients with AAV [4,9,10,14]. In one study, mortality from AAV, and its causes, was ascertained entirely using death certificate data, such that the mortality rate in people with AAV in the US was based on the inclusion of AAV as a cause on death certificates [14]. Similar methods have been used in other rheumatic diseases [15–19]. However, the sensitivity of cause of death codes on death certificates for the identification of deceased patients who had AAV, or most other rare rheumatic diseases, is not known. We sought to evaluate the causes and characteristics of death in a large, multicenter, contemporary cohort of patients with AAV and to assess the sensitivity of death certificate diagnosis codes for the diagnosis.

Methods

Study population

This study was approved by the Mass General Brigham Institutional Review Board (IRB), and all patients provided written informed consent for participation and publication. We used the 2002–2019 Mass General Brigham (MGB) AAV Cohort, a longitudinal inception cohort that includes individuals with AAV diagnosed and managed within the integrated MGB Healthcare system during this time period. We included patients from this patient who died between 1/1/2002 and 12/31/2019. Mortality was ascertained for all patients by medical records review and linkage to the National Death Index. This cohort has been previously described in detail and is assembled using both a validated algorithm and manual chart review for identification of cases and outcomes [20]. Patients with a diagnosis of eosinophilic granulomatosis with polyangiitis were excluded in this analysis.

Baseline demographics, disease characteristics, and treatments used at the time of diagnosis are ascertained using structured data from the Research Patient Data Registry (RPDR) and by manual review of the electronic health record (EHR). Initial treatment strategies were categorized as “rituximab (RTX)-based” if RTX was used with or without cyclophosphamide (CYC), “CYC-based” if CYC was used without RTX, or “other” if neither RTX nor CYC was used. This study was approved by the MGB Institutional Review Board, protocol number 2016P000633.

Ascertainment of causes and characteristics of death

For all patients, causes of death were determined using two methods. First, medical records were reviewed for all patients independently by two rheumatologists (GK and ZSW) for the most proximate cause of death that could be identified. This was determined by manual review of all available data, structured and unstructured, available in the EHR that was relevant to cause of death. Causes of death were categorized as AAV (i.e., death directly due to active AAV), cardiovascular disease (CVD) (i.e., documented fatal myocardial infarction or cerebrovascular accident), infection, end-stage renal disease (ESRD), cancer, or other. All differences were resolved by consensus. Patients in whom it was felt that there was inadequate information available in the EHR to determine cause of death were excluded from the chart review analysis.

Second, where death certificates data could be obtained, death certificate cause of death ICD-10 codes were extracted. In the United States, death certificates are signed by the provider who pronounces the death, and this provider indicates on the death certificate what the cause(s) of death were on multiple lines, with the primary cause of death (factor immediately leading to death) being the first line on the death certificate. The diagnoses on each line are classified according to ICD-10 diagnosis codes. We classified death certificates as having coded for AAV if an ICD-10 diagnosis code for MPA (M31.7) or GPA (M31.3X) was present in any line on the death certificate. Other causes of death according to death certificate ICD-10 codes were classified into specific categories (CVD, infection, ESRD, cancer, and other) using a validated schema, as has been reported previously [2,21].

For patients in whom adequate EHR data were present to determine a cause of death, we also reviewed medical records to determine the disease activity at the time of death and location of death. Disease status was categorized as “induction” if within 6 months of initial treatment following diagnosis of AAV, “flare” if disease had been active within the 6 months prior to death but not within 6 months of initial treatment, or “remission” if there was no evidence of active disease in the 6 months prior to death. Patients were classified as being on immunosuppression if they were on glucocorticoids at or above 5 mg or equivalent or conventional synthetic disease-modifying antirheumatic drugs for AAV (e.g., cyclophosphamide, methotrexate, azathioprine, or mycophenolate mofetil) leading up to the time of death, or if they had received rituximab or had undetectable B cells within 6 months prior to death.

Statistical analysis

Mean and standard deviation (SD) or median and interquartile range (IQR) were reported for normally and non-normally distributed continuous variables, respectively, and n (%) was reported for categorical variables. Sensitivities for AAV ICD-10 codes, calculated as the proportion of patients in whom AAV ICD-10 codes were present among all patients with AAV in the group of interest, were calculated along with Wald 95% confidence intervals (CI). Odds ratios with 95% confidence intervals were calculated to compare the odds of AAV ICD-10 code presence in each subgroup compared to the remainder of the cohort. Chi-square and Fisher’s Exact tests were used as appropriate to compare the frequency of AAV coding in two-group comparisons. Comparison of continuous variables between patients in whom AAV was or was not coded were performed using two-sample t-tests. Statistical analysis was performed using SAS version 9.4.

Results

Study population and demographics

Of 684 patients in the registry, 184 (27%) died during the follow-up period, and of these, 92 (50%) had adequate information available in the EHR to determine cause of death by manual review. Demographics and disease features are shown in Table 1. Deceased patients were older than the overall cohort (mean [SD] 72.1 [12.2] years vs. 59.6 [17.7]), and they tended to be diagnosed with AAV earlier in the follow-up period (53% vs. 35% diagnosed in 2002–2009, 19% vs. 36% diagnosed in 2014–2019). Deceased patients were also more likely to have been treated with a CYC-based treatment strategy than the overall cohort (52% vs 36%), which corresponds with the shifts in treatment patterns that occurred during the study period. Among the 92 patients with adequate information available in the EHR to determine cause of death, the frequency of CYC-based treatment strategy was more similar to the overall cohort (29% vs. 36%).

Table 1.

Patient demographics.

	Entire Cohort (n=684)	All Deceased (n=184)	Deceased with Death Information Available in EHR (n=92)
Mean age at diagnosis, mean (SD)	59.6 (17.7)	72.1 (12.2)	73.3 (10.3)
Mean age at death, years (SD)	---	78.1 (10.6)	78.0 (9.2)
Year of diagnosis
2002–2005	102 (15)	50 (27)	26 (10)
2006–2009	140 (20)	47 (26)	22 (19)
2010–2013	197 (29)	51 (28)	24 (29)
2014–2017	168 (25)	30 (16)	15 (28)
2018–2019	77 (11)	6 (3)	5 (14)
Years from diagnosis to death, median [IQR]	---	4.8 [1.7, 7.6]	3.6 [1.0, 7.5]
Sex
Male	280 (41)	80 (43)	37 (40)
Race
White	591 (86)	148 (80)	74 (89)
Asian	11 (2)	3 (2)	2 (2)
Black	14 (2)	5 (3)	2 (2)
Other	17 (2)	5 (3)	2 (2)
Unknown	37 (5)	17 (9)	9 (4)
ANCA
PR3-ANCA+	215 (31)	38 (21)	20 (35)
MPO-ANCA+	469 (69)	146 (79)	72 (65)
Baseline BVAS/WG, mean (SD)	5.0 (2.2)	4.8 (1.9)	5.0 (2.0)
Any renal involvement at baseline	442 (65)	132 (72)	66 (62)
ESRD at any time	52 (8)	28 (15)	21 (5)
Initial treatment strategy
RTX-Based	352 (51)	69 (38)	38 (57)
CYC-Based	243 (36)	96 (52)	43 (29)
Other	89 (13)	19 (10)	11 (4)

SD: standard deviation,

Causes and characteristics of death by chart review and death certificates

Ninety-two patients who died during the follow-up period had adequate information available in the EHR to determine cause of death. Causes of death, disease status at the time of death, and location of death as determined by chart review are shown in Table 2 and Supplementary Table 1. Based on chart review, death due to AAV occurred in 7 (8%): three patients had progressive interstitial lung disease attributed to AAV, two had refractory diffuse alveolar hemorrhage, one had central nervous system vasculitis due to AAV, and one had multiorgan failure from active AAV. Deaths from other causes were more common than death from active vasculitis; there were 27 (29%) deaths from infection and 17 (18%) from malignancy. Five (5%) deaths were due to CVD, and 35 (38%) deaths were categorized as “other,” and this included surgical complications, valvular heart disease, dementia, and cardiac arrest of unknown etiology. The most common sites of cancer that led to deaths included lung (n=4), pancreatic (n=2), acute myeloblastic leukemia (n=2), and colon (n=2). One patient each died of Merkel cell carcinoma, prostate adenocarcinoma, angiosarcoma, hepatocellular carcinoma, cholangiocarcinoma, urothelial cell carcinoma, and endometrial carcinoma.

Table 2.

Sensitivity of ICD-codes on death certificates for causes and characteristics of death determined by review of medical records.

	Overall Deceased	Death Certificate Available	AAV ICD Code Present on Death Certificate	Sensitivity, % [95% CI]	OR for AAV ICD Code PresenceΩ [95% CI]
Adequate EHR data available to determine cause of death	92	72	15	20.8 [11.2, 30.4]	---
Cause of death per chart review *
AAV	7 (8)	5 (7)	1	20.0 [0.0, 55.1]	0.95 [0.09, 9.67]
CVD	5 (5)	3 (4)	0	0 [--]	0
Infection	27 (29)	23 (32)	8	34.8 [15.3, 54.2]	3.20ϕ [0.99, 10.34]
ESRD	1 (1)	0 (0)	---	---	---
Cancer	17 (18)	16 (22)	2	12.5 [0.0, 28.7]	0.47 [0.09, 2.36]
Other**	35 (38)	25 (35)	4	16.0 [1.6, 30.4]	0.62 [0.18, 2.21]
Disease status at death
Induction	23 (25)	18 (25)	5	27.8 [7.1, 48.5]	1.69 [0.49, 5.84]
Flare	6 (7)	6 (8)	1	16.7 [0.0, 46.5]	0.74 [0.08, 6.89]
Remission	63 (68)	48 (67)	9	18.7 [7.7, 29.8]	0.69 [0.21, 2.24]
On immunosuppression at time of death	68 (74)	52 (72)	13	25.0 [13.2, 36.8]	3.00 [0.61, 14.71]
Location of death
Inpatient	56 (61)	43 (60)	10	23.3 [10.6, 35.9]	1.45 [0.44, 4.81]
Rehabilitation/SNF	6 (7)	6 (8)	1	16.7 [0.0, 46.5]	0.74 [0.08, 6.89]
Hospice	23 (25)	17 (24)	3	17.6 [0.0, 35.8]	0.77 [0.19, 3.12]
Home	5 (5)	5 (7)	0	0 [--]	0
Unknown	2 (2)	1 (1)	1	100 [--]	Undefined

^ΩOdds ratios reflect odds of AAV being coded in the group divided by odds of AAV being coded in the remainder of the group.

^*Causes of death were considered mutually exclusive.

^**“Other” causes of death included surgical complications, valvular heart disease, dementia, and cardiac arrest of unknown etiology.

^ϕIndicates p < 0.05 by Chi-square test. EHR: electronic health record; AAV: antineutrophil cytoplasmic antibody; CVD: cardiovascular disease; ESRD: end-stage renal disease; ICD: International Classification of Diseases; CI: confidence interval.

In contrast to cause of death observed when assessed by two physician reviewers, we observed different trends when cause of death was determined by death certificates. Based on death certificates, 36 (25%) deaths attributed to CVD, 16 (11%) to infection, and 30 (21%) to malignancy; none were attributed to vasculitis (Table 3).

Table 3.

Sensitivity of ICD-10 codes on death certificates for ANCA-associated vasculitis by disease characteristics.

	Overall Deceased	Death Certificate Available	AAV ICD Code Present on Death Certificate	Sensitivity, % [95% CI]	OR for AAV ICD Code PresenceΩ [95% CI]
N (%)	184	145*	24 (17)	16.6 [10.5, 22.6]	---
Cause of death per death certificate
AAV	---	0 (0)	0	0	0
CVD	---	36 (25)	3	8.3 [0.0, 17.4]	0.38 [0.11, 1.36]
Infection	---	16 (11)	1	6.3 [0.0, 18.1]	0.31 [0.04, 2.44]
ESRD	---	0 (0)	0	0	0
Cancer	---	30 (21)	1	3.3 [0.0, 9.8]	0.14 [0.02, 1.07]ϕ
Other	---	63 (43)	19	30.2 [18.8, 41.5]	6.65 [2.32, 19.05]ϕ

AAV: ANCA-associated vasculitis; ESRD: end-stage renal disease; CVD: cardiovascular disease; ICD: International Classification of Diseases; CI: confidence interval.

^*Includes patients with adequate electronic health record data available to determine cause of death (n=72) and those without (n=73).

^ΩOdds ratios reflect odds of AAV being coded in the group divided by odds of AAV being coded in the remainder of the cohort.

^ϕIndicates p < 0.05 by Chi-square or Fisher’s Exact test.

At the time of death, 68% of patients were in remission, while 25% were in the induction phase (within six months of initiation of treatment) and 7% had a disease flare in the six months prior to death. Causes of death generally differed among disease activity states (Figure 1). Death due to AAV occurred both during induction (n=4, 57%) and flare (n=3, 43%). Deaths due to CVD, ESRD, and malignancy occurred almost entirely during remission (n=5, 100%; n=1, 100%; and n=16, 94%, respectively). Deaths due to infection occurred in all three disease states, although most occurred during remission (n=17, 63%). Sixty-eight (74%) patients were on immunosuppression at the time of death. Most deaths available for review occurred in the inpatient setting (61%).

Figure 1.

Causes of death by disease state.

Sensitivity of AAV diagnosis codes on death certificates to identify deceased patients with AAV

Among the 145 patients with death certificates available, the sensitivity of AAV ICD codes on the death certificate for identifying patients with AAV was 16.6% (95% CI 10.5%, 22.6%). An AAV diagnosis code was more likely to be included in the death certificate when the primary cause of death according to the death certificate was classified as “other” (OR [95% CI] 6.65 [2.32, 19.05], p<0.001). Conversely, an AAV diagnosis code was less likely to be included in the death certificate when the primary cause of death according to the death certificate was cancer (OR 0.14 [0.02, 1.07], p=0.029) (Table 3, Supplementary Table 2).

We repeated these analyses among the 72 patients who had adequate EHR information to determine cause of death as well as death certificates available (Table 2). The overall sensitivity of AAV diagnosis codes to identify deceased patients with AAV was 20.8% (95% CI 11.2%, 30.4%). An AAV diagnosis code was more likely to be included in the death certificate when the death was determined by chart review to be due to infection compared with other causes of death (OR 3.20 [0.99, 10.34], p=0.046). Sensitivities did not differ by other death characteristics (Table 2, Supplementary Table 1).

Conclusion

In this multicenter retrospective cohort study, we made several novel and important observations regarding the factors contributing to death in patients with AAV. First, we found that death in patients with AAV was rarely due to AAV itself and was most often due to infection or cancer. Second, when death certificate data are used to assess the cause of death, this methodology overestimates the contributions of CVD and underestimates the importance of infection. Third, most deaths are occurring during the remission phase of disease as opposed to during remission induction or flares. In addition, we found that the sensitivity of death certificate diagnosis codes for the diagnosis of AAV was quite low. Collectively, our findings highlight an important limitation to research aimed at characterizing death in patients with AAV by using data derived only from death certificates.

Consistent with prior literature describing contemporary AAV cohorts, we found that death due to uncontrolled AAV itself was uncommon when compared to other causes of death; rather, the most proximate causes of death in patients with AAV were infection, cancer, and CVD [2,8,9,11]. Prior studies have shown that deaths due to vasculitis itself occur almost exclusively during the first year after diagnosis, yet death due to infection during this time period is approximately as common if not more common [8,9,11,22]. After the first year, infection remains one of the most common causes of death, particularly during the first five years after diagnosis, but CVD and malignancy also contribute to excess mortality observed in patients with AAV [2,8,9,11]. When manually assessing the cause as opposed to relying on death certificates, we found that infection was a far more common cause of death than CVD. While CVD remains an important cause of morbidity in AAV, these findings speak to the importance of ongoing efforts to optimize management in ways that reduce infection risk, especially during remission. Reassuringly, these findings highlight the tremendous progress in the management of AAV and ability to control disease well, albeit with serious complications related to treatment.

In our study, malignancy accounted for 18% of all deaths as determined by manual review of the medical records. Numerous prior studies have similarly shown an increased risk of malignancy in patients with AAV [2,23–25]. At least some of this association appears to be driven by treatments that increase the risk for malignancy, as the risk of malignancy in patients with AAV treated with rituximab is lower than in patients treated with cyclophosphamide-based regimens [23–25]. In our cohort, the most common malignancy was lung cancer (4/17, 24%), followed by colon cancer, prostate cancer, and acute myeloblastic leukemia (2/17, 12% each). A 2015 meta-analysis of observational data identified non-melanoma skin cancer, leukemia, lymphoma, lung cancer, and bladder cancer as types of malignancy that were associated with a statistically significant increase in risk in patients with AAV [23]. Other recent AAV cohorts have also reported lung cancer as the most common cause of cancer, and a recent nationwide study in Korea showed that patients with AAV were at higher risk for lung, bladder, and hematologic malignancies [24–27]. The types of malignancy that most commonly occur in patients with AAV and the mechanisms underlying this increased risk require further study.

AAV diagnosis codes were included as a cause of death in a minority of the death certificates, regardless of cause of death, location of death, or disease status (e.g., remission or induction) at the time of death. Even among the five patients in whom the cause of death by manual review was determined to be AAV itself and who had death certificate data available, only one had an AAV diagnosis code on the death certificate. These findings indicate that death certificate data underestimate—likely to a substantial degree—the impact of AAV on mortality in the US as well as the ability to describe mortality trends among people with AAV in the US using only death certificate data. Whether this applies to the use of mortality data outside of the US requires further study.

Death certificate diagnosis codes are commonly used in epidemiology research to advance our understanding of factors driving mortality in specific diseases. However, data from other rheumatic diseases including systemic lupus erythematosus and rheumatoid arthritis have shown that these diagnoses are often not included as causes of death in death certificates [28–30]. Indeed, our findings suggest that without linkage to medical records data, death certificates likely identify only a small proportion of overall deaths among people with AAV. In the 2020 study by Steinberg et al, for example, United States death certificate data indicated that between 1999 and 2017, 64% of deaths in AAV were due to AAV itself [14]. This contrasts with other contemporary studies in which data derived from medical records show that death due to AAV itself is rare, including our own study [2,8,9,11]. This contrast suggests that patients in the Steinberg et al. study were more likely to have AAV included in death certificate causes of death when AAV was more proximate to the cause of death. Our findings suggest that the study by Steinberg et al., which highlighted AAV as an important cause of death and improvements in mortality, may have underestimated mortality and associated temporal or racial trends in patients with AAV who died from infection, malignancy, and other factors.

Our study has several strengths. This is a large, multicenter study including both community-based practices and academic tertiary care centers. Deaths were identified through a data linkage between our MGB AAV Cohort and the CDC National Death Index. Further, we had data from both death certificates and medical records with long-term follow-up data for most patients, allowing for comparisons between the information derived from each data source. Our study also has important limitations. As this was a retrospective study, there were missing data. In particular, EHR-determined cause of death could not be ascertained in 50%, which is likely because patients received care in other centers, relocated, discontinued follow-up with specialty providers as they aged, transitioned to hospice care near the end of life, or were lost to follow-up for other reasons. In addition, few patients had autopsies or other information to provide definitive causes of death. This may have led to some misclassification of causes of death; in particular, since cardiac arrest of unknown etiology was classified as “other” cause of death, our study may underestimate deaths due to CVD. Centers included in this study were from the greater Boston, MA area so findings may not be generalizable to populations from different regions or countries. Finally, despite our overall large sample size for a rare disease, death was overall rare, so we were likely underpowered to detect differences in some of our subgroup analyses.

In summary, our study highlights the substantial role that infection has on mortality in patients with AAV, especially during remission. Treatment regimens that reduce the degree of immunosuppression both during remission induction and maintenance are likely to improve mortality due to infection in AAV. We also found important areas of discordance between medical records and death certificate data in the context of AAV, with death certificates underestimating the mortality among people with AAV and providing different perspectives on important drivers of death. Whenever possible, research on mortality in AAV should include linkage to medical records data in order to reduce potential bias resulting from the low sensitivity of AAV diagnosis codes in death certificates. Ongoing studies to reduce the impact of infection on mortality risk may improve survival for people living with AAV.

Key Messages:

• Death in patients with ANCA-associated vasculitis (AAV) more often results from infection than active vasculitis.

• Causes of death as determined by medical records review and death certificates are often discordant.

• The sensitivity of diagnosis codes on death certificates for identifying AAV is low.