Infective Endocarditis Hospitalizations and Outcomes in Patients With End‐Stage Kidney Disease: A Nationwide Data‐Linkage Study

Background We investigated the clinical features, microbiology, and short‐ and long‐term outcomes of incident infective endocarditis (IE) hospitalizations in patients with end‐stage kidney disease (ESKD) requiring dialysis or with a kidney transplant over 25 years in Scotland. Methods and Results In this retrospective, population‐based cohort study linking national hospitalization and mortality data, we identified patients with a history of ESKD and hospitalized with IE in Scotland between January 1, 1990 and December 31, 2014. From January 1, 2008, individual IE hospitalizations were additionally linked to national microbiology data. Multivariable logistic regression, adjusting for patient demographics and comorbidities, evaluated the association between ESKD and all‐cause death at 1 and 3 years. Of 7638 incident IE hospitalizations between 1990 and 2014, 2.8% (216/7638) occurred in 210 patients with ESKD and 97.2% (7422/7638) occurred in 7303 patients without ESKD. Positive findings from blood cultures were identified in 42% (950/2267) of incident IE hospitalizations from 2008. Staphylococcus aureus was isolated in 25.9% (21/81) and 12.8% (280/2186) of patients with and without ESKD, respectively (P=0.002). ESKD was associated with an increased odds of death at 1 (44.9% versus 31.4%; adjusted odds ratio [aOR], 2.47, 95% CI, 1.85–3.30;, P<0.001) and 3 years (63.9% versus 42.8%; aOR, 3.77; 95% CI, 2.79–5.12; P<0.001). Conclusions IE is associated with a poor prognosis in patients with ESKD, especially in the longer term. Compared with patients without ESKD, patients with ESKD were twice as likely to die within 1 year, and 3 times as likely to die within 3 years of IE hospitalization.

I nfectious diseases are the second commonest cause of death after cardiovascular disease in patients with end-stage kidney disease (ESKD) requiring dialysis or with a kidney transplant. 1,2 The incidence of infective endocarditis (IE) is ≈50-to 70-fold higher in those with ESKD compared with the general population, partly attributable to the use of arterio-venous grafts and indwelling catheters in patients undergoing dialysis, and long-term immunosuppression in renal transplant recipients. [2][3][4] The limited data availablemostly from small or single-center studies with short follow-up-suggest that the prognosis of IE in patients with ESKD is poor. 2,3 With the increasing global burden of ESKD, 5 contemporary population-based studies detailing the burden and outcomes of IE in patients with ESKD would be invaluable for planning healthcare provision for this at-risk group. The aim of this nationwide data linkage Gallacher et al Infective Endocarditis in End-Stage Kidney Disease study was to investigate the clinical features, microbiology and long-term outcomes of incident IE hospitalizations in patients with and without ESKD over the past 25 years in Scotland.

Study Design
We conducted a retrospective, population-based cohort study linking national hospitalization, microbiology, and mortality data sets in Scotland (Data S1). 4

Study Population
Using International Classification of Diseases (ICD) coding and a 5-year look-back period (Tables S1 and  S2), incident IE hospitalizations were identified from national inpatient records in those aged ≥20 years admitted to any Scottish hospital between January 1, 1990 and December 31, 2014. To optimize specificity and sensitivity, we included only hospitalizations with a diagnostic code for IE appearing in the first 2 (of 6) positions of the national inpatient record. We extracted demographic data (age, sex, and deprivation status) and selected comorbidities (history of stroke, heart failure, myocardial infarction, cardiac device, and previous cardiac valvular surgery) (Data S1). Patients with ESKD were identified by searching linked inpatient records before incident IE hospitalization for relevant ICD codes (Table S1) appearing in any of the 6 available diagnostic positions. Additionally, between January 1, 2008 and December 31, 2014, incident IE hospitalizations were linked to blood culture data obtained from national microbiology records (Data S1).

Determination of Social Deprivation Status and Comorbidities
Social deprivation status was determined according to the Scottish Index of Multiple Deprivation (SIMD) (Data S1). 6 SIMD is a geographical-based measure of deprivation. It is measured in quintiles, where the first and fifth quintiles are the most and least deprived, respectively. Every patient was assigned an SIMD quintile based on their individual SIMD rank at the time of incident IE hospitalization-determined by social factors related to residential address (zip code).
Comorbidities (history of myocardial infarction, stroke, heart failure hospitalization, implanted cardiac device, and prior valvular heart surgery) were defined by identifying relevant ICD codes attributed to inpatient records of hospitalizations and procedures during the 5 years preceding incident IE hospitalization (Data S1). 4

Study Outcomes
Outcomes included stroke, heart failure, and subsequent valvular heart surgery at 1 and 3 years; and allcause death at 30 days, 1 year, and 3 years.

Statistical Analysis
Clinical characteristics and outcomes were summarized according to ESKD status. Groupwise comparisons were performed using Chi-square tests, as appropriate. Multivariable logistic regression (adjusted for age, sex, social deprivation status; history of stroke, heart failure and myocardial infarction) evaluated the association between ESKD and all-cause death at 1 and 3 years in all patients hospitalized with IE. In a within-exposure analysis restricted to those with ESKD, multivariable logistic regression was used to determine factors associated with all-cause death at 1 and 3 years. Age, sex, and outcome data were complete for all IE hospitalizations. Data on social deprivation status were missing in 0.6% of all IE hospitalizations; these records were excluded from our analysis. Statistical analysis was performed in R, Version 3.5.1 (Vienna, Austria).

Ethical Considerations and Access to Study Data
The study was approved by the National Health Service Public Benefit and Privacy Panel (reference: 1516-0116). Patient consent was not sought as the analysis used fully anonymized data. Individual-level data are available via application to the National Health Service e-Data Research and Innovation Scotland team, which is part of Public Health Scotland. Access to original study data are restricted to approved members of the research team and will not be made publicly available. However, these individual-level data are available via application to the National Health Service e-Data Research and Innovation Scotland team, which is part of Public Health Scotland. Source analysis code will be made available upon request to the corresponding author.  Figure B).

DISCUSSION
To date, this is one of the largest nationwide data linkage studies to compare the clinical characteristics, microbiology, and long-term outcomes of IE in patients with and without ESKD. One fifth of patients with ESKD died within 30 days of their incident IE hospitalization, half died within 1 year, and two thirds within 3 years. Compared with those without ESKD, patients with ESKD were twice as likely to die within 1 year of IE hospitalization and >3 times as likely to die within 3 years. In patients with ESKD, older age was independently associated with a poorer prognosis at both 1 and 3 years, whilst the lowest level of deprivation was associated with better outcomes at 3 years only.
The few studies that have defined the microbiology of IE in patients with ESKD report Staphylococcus aureus infection rates of ≈50% to 80% 7,8 -substantially higher

Gallacher et al
Infective Endocarditis in End-Stage Kidney Disease than in the current study. However, these types of cohort study often rely on cases of IE being identified by clinicians, increasing the risk of selection bias and thus, the proportion of patients with positive microbiology. In contrast, our study-which used individual patient-level blood culture data-was free from selection bias as all IE hospitalizations were identified from routine diagnostic coding, the accuracy of which was recently reported as ≈94% for cardiovascular diagnoses. 9 Whilst short-term outcomes of IE are similar between those with and without ESKD, patients with ESKD do worse in the longer term. Our data are consistent with a contemporary Danish study 3 and compare favorably with an American study in patients with ESKD performed ≈20 years ago, 10 which described 1-and 3-year mortality rates of 61.6% and 81.7%, respectively. In addition, the in-hospital mortality described in a recently published singlecenter study from Taiwan was approximately double the figure we report at 30 days, despite similar Staphylococcus aureus infection rates. 11 Bhatia and colleagues 12 previously demonstrated a comparable increased risk of death in patients with and without ESKD. Although well-powered, their analysis was restricted to in-hospital death only. In this regard, the long-term outcomes we report are more relevant in the contemporary era, given ≈80% of patients with ESKD survive the initial IE hospitalization.
There are some limitations to consider. Our analysis was under-powered to stratify outcomes by renal replacement therapy (hemodialysis, peritoneal dialysis or renal transplant) or microbiological etiology, or to include these variables in the multivariable regression analyses. As we used routine administrative ICD codes to identify IE hospitalizations, our study was free from selection bias but subject to case ascertainment bias. To limit the impact of this, the study cohort was restricted to hospitalizations with a diagnostic code for IE in the first 2 (of 6) positions.
Overall, this comprehensive nationwide analysis over a 25-year period highlights the poor prognosis associated with IE in patients with ESKD compared with patients without ESKD, especially in the longer term. Our results underscore the importance of a multidisciplinary approach to the clinical management of this complex and vulnerable patient group.

Scottish Morbidity Record 01 (SMR01)
The Scottish Morbidity Record 01 (SMR01) is an episode-based hospitalization record, indexed by community health index (CHI) number, relating to all inpatient and day case hospitalizations from non-obstetric and non-psychiatric specialties. 13 A record is generated when a patient completes an episode of inpatient or day case care. Up to six diagnoses are recorded using the International Classification of Diseases (ICD) classification, whilst interventions or procedures are recorded using the OPCS-4 classification. SMR01 data are considered amongst the best routinely-collected healthcare data worldwide in terms of granularity, population coverage and linkage capabilities. Across all SMR01 records, estimated completion and accuracy rates are 99% and 89%, respectively. 14 For SMR01 records relating to cardiovascular diagnoses, the accuracy rate is 94.2%. 9

National Records of Scotland (NRS) Death Records
The NRS register is indexed by CHI number and records all deaths in Scotland, with ~55,000 deaths registered annually. Available data from individual patients include demographics, date of death, and primary and secondary causes of death. 15 These data were linked to patients identified as having a record of an incident IE hospitalization during the study period. Of note, it is a statutory requirement that any death occurring in Scotland, or out-with Scotland but within the United Kingdom, is registered on the NRS death register within 8 days of death.
Although patients who emigrate to other countries will be lost to follow-up, the Scottish population is historically very stable, with a low rate (<0.5%) of overseas emigration each year. 16 Data S1.

Electronic Communication of Surveillance in Scotland (ECOSS)
The Scottish microbiology surveillance registry or 'Electronic Communication of Surveillance in Scotland' (ECOSS), as it is termed by NHS National Services Scotland, was used to identify positive blood culture results from microbiology laboratories within NHS Scotland health boards pertaining to incident IE hospitalizations in SMR01 between 01/01/2008 and 12/31/2014. Causative organisms were defined as those identified ≤90 days on either side of the incident IE hospitalization date. Polymicrobial status was defined when >1 causative organism was identified on the same culture date. If >1 causative organism was identified on different dates ≤90 days on either side of the incident IE hospitalization date, then the organism identified closest to the incident IE hospitalization date was assigned as the causative organism.
Although data were first recorded in ECOSS from 2007, near-complete data are available from 2008 and so ECOSS records were linked to SMR01 in the present study from this year onwards. 17,18 ECOSS is maintained by NHS National Services Scotland on behalf of Health Protection Scotland. NHS National Services Scotland monitors the completeness and accuracy of ECOSS data through its 'Data Monitoring and Support Service'. 17 Further, NHS National Services Scotland routinely informs data users of any problems affecting the accuracy of these data. More information on ECOSS is available from https://www.hps.scot.nhs.uk/data/.

Identification of study participants
Incident hospitalizations with infective endocarditis (IE), end-stage kidney disease (ESKD) status and comorbidities were defined from SMR01 in patients aged ≥20 years admitted to any Scottish hospital between 01/01/1990 and 12/31/2014 using International Classification of Diseases (ICD) codes (Table S1). To optimize specificity and sensitivity, we included only hospitalizations with a diagnostic code for IE appearing in the first 2 (of 6) positions of the SMR01 record. We extracted demographic data (age, sex and deprivation status [see next section]) and selected comorbidities (history of stroke, heart failure, myocardial infarction, cardiac device and previous cardiac valvular surgery) from SMR01 using records of hospitalizations and procedures during the 5 years preceding hospital admission (a 5-year 'lookback' period). Patients with ESKD were identified by searching linked inpatient records prior to hospitalization with IE for relevant ICD codes (Table S1) appearing in any of the 6 available diagnostic positions.

Definition of deprivation status: the Scottish Index of Multiple Deprivation (SIMD)
The Scottish Index of Multiple Deprivation (SIMD) is a geographical-based measure of deprivation. SIMD identifies small geographical regions (where each region is determined by zip code and corresponds to ~750 residents) of material deprivation based on information derived from seven domains (income; employment; health; education, skills and training; geographic access to services; crime; and housing). 6 Each domain is weighted according to its relative importance and provides a score which is then summed with the other domains. The total score for each geographical region enables the areas to be ranked. SIMD scores and ranks (quintiles/deciles) have been used extensively in published epidemiological research from Scotland. 6 In this study, all patients were assigned a SIMD quintile based on their individual SIMD rank at the time of incident IE hospitalization. Table S2 illustrates the look-back period for the years 2000-2010 in three exemplar patients (patients A, B and C). The total incident count for each year is shown in the final column. Where a patient is hospitalized with an episode of IE, a '1' appears in the 'Admission' column. If no IE event has occurred in the 5-years prior (light grey shading), then the event is considered an incident event and a '1' will also appear in the 'Incident' column (dark grey shading).