Early Mortality in Paroxysmal Nocturnal Hemoglobinuria

Objectives: The elevated mortality risk among patients with paroxysmal nocturnal hemoglobinuria (PNH) has been suggested to derive from a high risk of thromboembolism (TE); however, the risks of coexisting cardiovascular risk factors are not well described. We studied mortality associated with PNH taking comorbidity and treatment into account. Methods: Patients with PNH (n=115) were identified in the 1977-2016 Danish National Patient Register (DNPR). For each patient with PNH, we identified 50 age- and sex-matched general population comparators. Using the Kaplan-Meier estimator and Cox regression, we compared the overall survival of patients with comparators. Cumulative incidences were used to analyze the effects of comorbidity and the causes of death. Results: One-year survival among patients and comparators was 92.2% and 99.4%, and after 10 years, it was 68.4% and 85.8%, respectively. Early mortality was associated with older age, higher levels of comorbidity, and solid malignancies prior to PNH diagnosis. The leading causes of death were infections and associated hematological diseases. Patients with early mortality were less likely to have received treatment with eculizumab and/or warfarin. Cardiovascular risk factors were evenly distributed between patients and comparators at diagnosis. Conclusion: We conclude that early mortality in PNH is associated with older age, cardiovascular comorbidity, and hematological malignancies.


Introduction
Paroxysmal nocturnal hemoglobinuria (PNH) is an acquired clonal hematopoietic stem cell disease causing non-immune hemolysis with an incidence rate of 0.35 cases per 100,000 people per year and an overall prevalence rate of 3.81 per 100,000 [1].
Manifestations of PNH are caused by a mutation in the phosphatidylinositol glycan class A and possibly in the class T gene (PIG-A and PIG-T) [2].In healthy individuals, this gene is responsible for the synthesis of glycosylphosphatidylinositol (GPI) anchors used to attach surface proteins to red blood cells (RBCs).A deficiency of GPI anchors blocks the RBCs from attaching CD55 and CD59, two essential enzymes in preventing random complement attack [2].Thus, without this protection, RBCs are susceptible to complement attacks causing intravascular hemolysis [3].
Patients with PNH can present with a wide range of disease manifestations from fully compensated asymptomatic hemolysis to severe anemia with reduced quality of life from fatigue, thromboembolism (TE), abdominal pain, smooth muscle dystonia, and death [4].Prior to the introduction of complement-inhibiting agents such as eculizumab, treatment options included supportive care, anticoagulation, and hematopoietic stem cell transplantation [5,6].With these treatment options, the median survival of PNH patients ranged from 10 to 30 years [7,8].However, from the pivotal studies of eculizumab including 195 patients, it was shown that complications and prognosis improved and the reported risk of TE was reduced from 32.6% to 3.6%, resulting in a three-year survival proportion of 97.6% [6].Similarly, patients with PNH treated with eculizumab had a reported four-to five-year survival of 95.5%-98.3%compared with 66.8%-79.7% in untreated patients [9].Death in PNH is often attributed to TE.A study from 1995 of 48 patients (for whom the cause of death was known) illustrated that TE was a contributing cause of death in 28 patients [7].Although TE is a wellrecognized complication in PNH, it is unknown how other cardiovascular risk factors such as diabetes and hypertension affect mortality [4,10].Using a nationwide matched cohort design including PNH patients and age-and sex-matched comparators from the general population, we investigated the effect of comorbidities and cardiovascular risk factors on survival among patients with PNH in Denmark during 39 years of observation.

Data sources and patients
Denmark provides tax-financed universal national healthcare, and the study population derives from the entire Danish population through routinely registered national health data.The unique and permanent civil registration number (CRN) allows individual-level linkage between all Danish registries [11,12].For this study, we linked data from the Danish National Patient Register (DNPR), the Danish National Prescription Registry, and the Danish Register of Causes of Death [13,14].The DNPR comprises all inpatient hospital contacts since 1977 and associated diagnoses using the International Classification of Diseases (ICD).Since 1994, hospital outpatient specialist clinic visits and emergency room contacts have also been included using the ICD-10, whereas ICD-8 was used before 1994.Private hospitals are included in the DNPR, but none of them are engaged in the management of hematological disorders or associated complications [12,15].The diagnosis codes of cardiovascular disorders, hematological cancers, and hemolytic disorders within this registry have previously been validated, all with high positive predictive values [13,16].The DNPR contains limited information regarding hospital treatments, but the registration is incomplete.
The present study derives from the Danish Hemolysis Cohort, a cohort comprising all patients with hemolytic disorders identified from the above-described registries [13,14,17,18].In the Danish Hemolysis Cohort, we identified all patients diagnosed with PNH in Denmark between January 1, 1977, and December 31, 2016.To increase the reliability of the diagnosis, we excluded patients only registered with a PNH diagnosis from surgical departments or when PNH was only registered as a post-mortem diagnosis [13].
We used the DNPR to identify hospital-registered comorbidity at baseline such as cardiovascular comorbidity, cardiovascular risk factors, and associated hematological disorders (e.g., aplastic anemia (AA) or myelodysplastic syndrome (MDS)).A detailed list of the diagnosis codes is shown in the Appendices.Identification of comorbidities and risk factors was enhanced by the individual-level data on prescription drugs available from the Danish National Prescription Registry since 1994 [19].We used the Anatomical Therapeutic Chemical (ATC) codes to identify drugs associated with specific comorbidities, e.g., insulin to identify diabetes or digoxin to identify cardiac arrhythmia (Appendices).
The Danish Register of Causes of Death holds information on immediate and underlying causes of death registered by the doctors who issued a death certificate in Denmark.This register is updated annually, and we included causes of death from 1977 to 2016 [20].We used the recommended main underlying cause of death to assess cause-specific mortality among patients with PNH and comparators.

Comparators and follow-up
For each patient with PNH, we identified up to 50 age-and sex-matched comparators from the general population without PNH, each allotted an index date corresponding to the date of diagnosis of the index PNH patient (Figure 1).Information regarding comorbidity, prescription drugs, causes of death, etc. were obtained for both patients and comparators.Patients and comparators were followed from inclusion to the first of emigration, death, or end observation time on December 31, 2016.

FIGURE 1: Flowchart of cohorts
During the analysis, it was found that one PNH patient did not have any registered cause of death.This person and the associated comparators were excluded (red box).Also, 50 comparators were excluded because no matched PNH patient was found (red box).The PNH cohort was subdivided into two groups: short survival (PNH patients who died within two years of diagnosis) and long survival (PNH patients who lived longer).The blue boxes illustrate the final groups.

Statistics
Proportions with 95% confidence intervals (CIs) and medians with interquartile range (IQR) were used to assess age, sex, comorbidity (such as AA, MDS, and other hematological malignancies), venous TE, cardiovascular comorbidity and risk factors, diabetes, and the use of estrogens.The primary endpoints were overall mortality and cause-specific mortality.To describe factors associated with early mortality, we described patients who died within two years from diagnosis and patients who lived longer, separately.Overall survival was assessed using the Kaplan-Meier estimator, and the difference in survival was tested using a log-rank test.Cause of death was estimated under competing risk with nonparametric cumulative mortality proportions [21][22][23].Causes of death were grouped into six main categories based on previous studies and conditions associated with PNH: TE, hemorrhage, infection, cardiovascular disease, associated hematological disease, and other causes of death [4,7,24].Differences in survival between groups were assessed with Cox proportional hazard regression estimating unadjusted and adjusted hazard ratios.Hazard ratios were adjusted for date of diagnosis, sex, age, cardiovascular comorbidity and risk factors, diabetes, use of estrogens, preexisting hematological disease, and TE.Model assumptions were assured with Schöenfeld residuals and comparing the Kaplan-Meier curve with the univariate predicted curves.Goodness of fit was assessed with Nelson-Aalen/Cox-Snell residuals.
The statistical software package Stata (StataCorp, College Station, TX) was used for data management and analyses [25].

Ethics and data sharing
This study was approved by the Region of Southern Denmark (case number 17/10885).Registry-based research without direct contact with patients is exempt from Ethics Committee approval in Denmark.Danish law does not allow sharing of patient data publicly.For legal reasons, individual-level patient data therefore cannot be shared, and table output containing information from less than three persons must be anonymized.The reporting of results was guided using the REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) statement [26].

Overview
During the study period , the Danish population increased from 5.1 to 5.7 million persons.In this period, 115 patients were registered with PNH.We identified 5,723 age-and sex-matched comparators from the general population, giving a mean match rate of 49.8:1.The patients and comparators contributed 1,364 and 85,080 person-years, respectively.At diagnosis, the patients with PNH had a median age of 48.2 (IQR: 44.5-51.9),and 50.4% were females (Table 1 1).After the first year, overall survival was 92.2% (95% CI) in patients with PNH and 99.4% (95% CI) among comparators.After 10 years, 68.4% (95% CI) of patients with PNH were still alive versus 85.8% (95% CI) of the comparators (Figure 2A).
The adjusted hazard ratio (HR) for death in patients with PNH was 10.5-fold (95% CI: 6.3-17.3) the risk among comparators.
Mortality in patients was particularly elevated during the first two years after diagnosis but approached that of the general population gradually thereafter (Figure 2A).We did not observe the difference in survival for patients with PNH diagnosed before and after 2006 when eculizumab became available (Figure 2B).However, only 11 patients were registered with exposure to eculizumab either as monotherapy or in combination with warfarin (Table 2).The limited number of eculizumab-exposed patients did not allow for a specific analysis of the effects of eculizumab treatment on survival.In patients with PNH who died within two years from diagnosis, associated hematological disease (48.4% (95% CI: 21.6-70:9)) or infections (23.6% (95% CI: 5.9-47.9))were the most frequent cause of death (Table 2).Among patients and comparators with late mortality, cardiovascular disease was an equally frequently registered cause of death, whereas death from associated hematological diseases was significantly more frequent among patients (16.5% (95% CI: 7.6-28.3))than comparators (1% (95% CI: 0.6-1.5))(Table 2).Furthermore, patients with early mortality after PNH diagnosis were often untreated with warfarin or eculizumab (94.4% (95% CI: 72.7-99.9))compared to patients who lived longer (63.9% (95% CI: 53.5-73.4))(Table 2, Figure 2F).The same trend is suggested from the Kaplan-Meier plot comparing patients with PNH who received treatment (eculizumab and/or warfarin) to the patients with PNH who did not receive any such treatment (Figure 2F).However, a post hoc analysis indicated that the difference in survival between treated and untreated patients with PNH was only significant within the first five years following diagnosis (p=0.01 at five years, p=0.10 at 10 years).

Discussion
Using a nationwide cohort with long and complete follow-up, we confirm that mortality is higher among patients with PNH than in the general population.Mortality is particularly increased during the first two years after PNH diagnosis and was associated with older age, preexisting solid cancer, and a higher prevalence of comorbidities at the time of diagnosis.Death due to associated hematological disease and infections prevailed within the first two years from diagnosis, accounting for 48.4% and 23.6% of the fatalities, respectively.Death due to cardiovascular disease was a frequent cause of death both among PNH patients and comparators, but TE was not registered as a cause of death among patients with PNH.
Our results comply with earlier notice that associated hematological diseases, including AA and malignancy, and infections are the leading causes of death in PNH [24].However, the absence of TE as a reported cause of death is in contrast to earlier reports [7,10,24].A French retrospective study including 460 patients with PNH, diagnosed before eculizumab became available, found that infections (24%) and Budd-Chiari syndrome (21%) were the most frequent causes of death [24].Another study conducted before the introduction of eculizumab included 80 patients with PNH diagnosed over a period of 30 years and obtained causes of death for 48 deceased patients of whom 28 (35%) died from either venous TE or hemorrhage [7].A study conducted from 1970 to 2013 that included 56 patients with PNH and retrospective data from medical records reported that TE episodes and cancer constituted the main causes of death, each accounting for 8.9% [27].Taken together, these studies indicate that TE is a frequent cause of death, although we did not find this in our data.The algorithm for deriving the main cause of death in Danish registries defines the most probable main cause, and TE may well have been a contributing cause of death in our patients.Further, the main cause of death derives from the death certificate, which is susceptible to errors.This may be augmented by the decline in the autopsy rate in Denmark [20].Together, our causes of death results should be interpreted with caution and are best suited to highlight the differences within the cohort rather than describing expected causes of death in future cohorts.
The two-year cutoff for early versus late mortality was justified by the change in slope on the survival curve for patients with PNH at two years (Figure 2A).This phenomenon was taken as an indication of a change in risk among the patients and was used to divide them into two potentially different subgroups.The distinction between early and later mortality yielded some differences in patients' characteristics, treatment, and causes of death, but a causal relationship cannot be established.Especially as the age at death is comparable in the two groups, a possibility persists that the predominant difference between the groups could be defined by delayed diagnosis in the one group, whereby the groups represent a surveillance bias and different steps in disease progression.
Our data shows that patients with early mortality were less likely to have been treated with warfarin and eculizumab.However, this observation could reflect both undertreatment and confounding from comorbidity or advanced age leading to death before treatment was initiated.In fact, patients with early mortality were characterized by increased age, previous solid cancer, cardiovascular comorbidity, and higher prevalence of comorbidities at diagnosis.Note that older age in patients with early mortality and death a year after diagnosis could often suggest that the diagnoses of PNH in these cases were delayed.
All of these characteristics are also associated with an increased risk of thrombosis.Our study could therefore suggest, but not conclude, that these high-risk patients in particular might benefit from antithrombotic treatment.
This difference between the groups could also be due to diagnoses before eculizumab became available in 2007, when the overall mortality would be expected to be higher.Further, we collected our data over a period of 39 years, which introduces the risk of cohort effects with increased survival among both patients and comparators.Supportive care has improved over time and would be expected to increase survival; however, our data do not deliver firm evidence of this.The comparators increased survival slightly before and after 2007, but the patients with PNH have overlaying survival curves before and after (Figure 2B).
Our inclusion of comorbid conditions and cardiovascular risk factors as covariates was based on hospital diagnoses from routine registrations.The registry-based data do not include granular results from medical files such as body mass index and smoking history.This shortcoming in data may disproportionately affect comparators that are not routinely followed in the hospital and may therefore have incomplete data regarding, e.g., obesity or hypertension.Noteworthy, most of our included cardiovascular risk factors were evenly distributed among patients and comparators, indicating that this bias is unlikely to play a major role.
Our patients with PNH were captured using hospital diagnosis registrations from non-surgical departments and did not include post-mortal diagnosis.We excluded patients captured only through such registrations because our previous validation study found that a hemolytic anemia diagnosis in these settings is often not valid [13].This study also validated the hemolysis diagnosis registered in the DNPR and found that the diagnosis of acquired hemolytic anemia (PNH included) had a positive predictive value (PPV) of 83.4% (95% CI: 77-89).PNH, as an independent diagnosis, had a PPV of 80% (95% CI: 28-88); however, this was based on only five registrations [13].Because of inclusion criteria, our study design could potentially lead to the exclusion of the most severely ill patients presenting with a rapidly fatal outcome such as splanchnic TE, and therefore, survival may be overestimated.
Considering the rarity of the PNH condition, our study includes a relatively high number of patients diagnosed during a 39-year period.The number of patients with PNH in this period may be higher since patients with a small clone in conjunction with, e.g., AA are not completely registered.In general, AA and not a small PNH clone would often dominate the clinical course of such patients, and the effects of the incomplete inclusion of AA-PNH patients on our results are unknown.
Despite the nationwide accrual of patients with PNH, a matched comparison group, and complete follow-up, our study has limitations.We based inclusion on register data, and therefore, we lack granular laboratory data such as PNH clone size and LDH, as well as patient-specific factors such as symptoms, BMI, and smoking habits.Therefore, we cannot disentangle PNH severity and lifestyle factors that may influence prognosis.
Another inherited limitation of the register data is that we did not have the granularity to inform on the etiology of lethal infections, as only discharge diagnosis is registered in the DNPR or the Danish Register of Causes of Death.Further, the limited number of patients with PNH who died from infections did not allow for detailed analysis of the specific infections associated with mortality [28][29][30].
Eculizumab exposure represents a singular limitation to our study as there is no mandatory registration of this treatment.Without complete capture of eculizumab exposure, a bias in the relative estimate between the treated and untreated groups may decrease the difference, biasing the untreated group estimate toward that of the eculizumab-treated group.We therefore focused on the chronological disjunction of patients diagnosed before or after the introduction of eculizumab in Denmark, rather than registered exposure.This partitioning will also bear a risk of bias, as patients diagnosed before but surviving at least until the time of introduction may also have been exposed to eculizumab and are selected for longer survival.
Overall, the very limited number of patients registered with eculizumab treatment does not allow for conclusions regarding the isolated effect of eculizumab on survival.

Conclusions
Our study indicates that age, previous solid cancer, cardiovascular comorbidity, and comorbidity at the time of diagnosis are associated with early mortality after PNH diagnosis.In addition, mortality due to hematological disease and infections prevail as frequent causes of death among patients with early mortality.
The increased mortality in older patients could suggest that older patients remain undiagnosed for longer periods and therefore succumb to PNH-related complications shortly after diagnosis and that undertreatment is a potential risk in these high-risk patients.The rarity of the disease calls for international collaboration to investigate rare events and complications associated with PNH, particularly when subgroups such as elderly patients are considered.

Variable name
Left: PNH cohort, right: comparators PNH: paroxysmal nocturnal hemoglobinuria Mortality and causes of death were assessed for early versus late mortality (<2 years versus ≥2 years) after the index date.

FIGURE 2 :
FIGURE 2: Survival among patients with PNH (A) Overall survival for patients with PNH diagnosed from 1977 to 2016 and age-and sex-matched comparators from the general population.(B) Overall survival for PNH and comparators, subdivided by year of diagnosis, with eculizumab being released in 2007.(C) Subdivided by age at diagnosis of PNH, with the older age groups starting at 60 years.(D) Subdivided by thrombosis prior to PNH diagnosis.(E) Subdivided by diagnosis of cancer prior to diagnosis of PNH.(F) Subdivided based on exposure to eculizumab and/or warfarin during follow-up.p-values in the individual graphs are given with one or, when applicable, three degrees of freedom.p-values with one degree of freedom are in B to F testing the two subgroups of PNH, whereas three degrees of freedom p-values test all four groups.PNH: paroxysmal nocturnal hemoglobinuria

TABLE 1 : Basic description of the 115 patients with PNH and 5,723 age-and sex-matched general population comparators
Results marked with CI are estimated as proportions presented with 95% confidence interval.Results marked with IQR are estimated as medians presented with 75th and 25th percentiles.Due to current rules on the protection of patient anonymity from Statistics Denmark, all columns with less than three persons must be anonymized.

TABLE 2 : Cumulative causes of death and treatment in 115 patients with PNH and 5,723 age-and sex-matched general population comparators
Results marked with CI are estimated as proportions presented with 95% confidence interval.Due to current rules on the protection of patient anonymity from the Danish Statistic, all cells with one or two persons must be anonymized as "<3." PNH: paroxysmal nocturnal hemoglobinuria, BMT: bone marrow transplant, TE: thromboembolism, AA: aplastic anemia, CI: confidence interval

TABLE 4 : ICD-8 and ICD-10 codes used in Table 1
ICD: International Classification of Diseases, AA: aplastic anemia, MDS: myelodysplastic syndrome, TE: thromboembolism, PNH: paroxysmal nocturnal hemoglobinuria Table5shows a detailed list of diagnosis codes used in Table2 and Table 3.

TABLE 5 : ICD-8 and ICD-10 codes used in Table 2 and Table 3
ICD: International Classification of Diseases, AA: aplastic anemia, TE: thromboembolism