Advertisement

Long term sepsis readmission, mortality and cause of death following Gram negative bloodstream infection: a propensity matched observational linkage study

  • John F. McNamara
    Correspondence
    Corresponding author: John F McNamara, Postal Address: The Prince Charles Hospital, 627 Rode Road, Chermside, 4032, Brisbane, QLD, Australia, Telephone +61 07 3139 4414.
    Affiliations
    University of Queensland Centre for Clinical Research, Building 71/918 Royal Brisbane & Women's Hospital Campus, Herston, 4029, Brisbane, QLD, Australia

    The Prince Charles Hospital, 627 Rode Road, Chermside, 4032, Brisbane, QLD, Australia
    Search for articles by this author
  • Patrick N.A. Harris
    Affiliations
    University of Queensland Centre for Clinical Research, Building 71/918 Royal Brisbane & Women's Hospital Campus, Herston, 4029, Brisbane, QLD, Australia

    Central Microbiology, Pathology Queensland, Royal Brisbane and Women's Hospital, Cnr Butterfield St and Bowen Bridge Rd, Herston, 4029, QLD, Australia
    Search for articles by this author
  • Mark D. Chatfield
    Affiliations
    University of Queensland Centre for Clinical Research, Building 71/918 Royal Brisbane & Women's Hospital Campus, Herston, 4029, Brisbane, QLD, Australia
    Search for articles by this author
  • David L. Paterson
    Affiliations
    University of Queensland Centre for Clinical Research, Building 71/918 Royal Brisbane & Women's Hospital Campus, Herston, 4029, Brisbane, QLD, Australia

    Department of Infectious Diseases, Royal Brisbane and Women's Hospital, Cnr Butterfield St and Bowen Bridge Rd, Herston, 4029, QLD, Australia
    Search for articles by this author
Open AccessPublished:October 27, 2021DOI:https://doi.org/10.1016/j.ijid.2021.10.047

      Highlights

      • GN-BSI patients have poor long-term mortality in comparison to a matched cohort
      • The underlying cause of death was due to infection only within the first 90 days
      • Sepsis was the contributory cause of death for one year after GN-BSI

      Abstract

      Objectives

      Understand the long-term mortality, risk of readmission for sepsis and cause of death following a gram-negative bloodstream infection (GN-BSI).

      Methods

      This was a propensity-matched study using data linkage of Queensland hospital data, Australia. GN-BSIs were collected from 2005 to 2010 and matched 1:1 to hospital admissions without BSI for age, gender, year of culture collection, frequency of admissions in the prior year and Charlson-Deyo Comorbidity score and each comorbidity within the Charlson-Deyo score. Readmissions for sepsis, mortality and causes of death were evaluated.

      Results

      Cases of GN-BSI were propensity-matched 1:1 to culture-negative hospital admissions (n = 14016). Readmissions for sepsis were higher in the GN-BSI cohort from 91 to 365 days (P < 0.001) and in the four subsequent years (P < 0.001). The five-year survival in the GN-BSI cohort was 52% versus 65% in the culture-negative cases (P < 0.001). Infection was only a common underlying cause of death within the first 90 days. Sepsis was the most common contributing cause of death (CCOD) for the two years following index culture in the GN-BSI cohort.

      Conclusions

      Compared to a similarly vulnerable group of hospital attendees, GN-BSI had higher mortality and demonstrated a persistent long-term risk of readmission for sepsis and sepsis as a CCOD.

      Key words

      Introduction

      Establishing a causal relationship between acute infection and the long-term outcome requires an appreciation of a complex interplay between individual demographics, pre-illness comorbidity, pre-illness health trajectory and risk factors for infection. Ideally individuals with severe infection would be compared to a suitable group without infection with a similar degree of vulnerability. (
      • Shankar-Hari M
      • Rubenfeld GD
      Understanding Long-Term Outcomes Following Sepsis: Implications and Challenges.
      )
      GN-BSI represents a severe disseminated infection that is often but not always associated with sepsis syndrome. (
      • Huerta LE
      • Rice TW.
      Pathologic Difference between Sepsis and Bloodstream Infections.
      ;
      • McNamara JF
      • et al.
      Evaluation of quick sequential organ failure assessment and systemic inflammatory response syndrome in patients with gram negative bloodstream infection.
      ) It provides the opportunity to assess specificity in the context of proven infection. It also represents an opportunity to study the dose effect of the severity of infection on long-term outcomes by stratifying for the presence of sepsis.
      We hypothesised that GN-BSI would have an independent association with persistent risk of sepsis and mortality in the long term. We evaluated long-term readmission for sepsis, five-year mortality, the underlying cause of death (UCOD) and the most common contributing cause of death (CCOD) in an observational linkage study of GN-BSI propensity matched to blood culture-negative hospital admission.

      Methods

      Setting

      This study was a data extract from a larger public health application undertaken in Queensland, Australia, spanning 2000 to 2015. During this period, results from blood cultures collected and processed through the AUSLAB pathology database were used to collect relevant cases for the linkage process. The original public health application consisted of multiple classes of pathogens comprising 56,972 BSIs and 113,789 culture-negative cases. Cases were collected from 2000 to 2010, with an additional five-year follow-up data collection for re-admission and death completed in 2015.
      Follow-up admission data included data collected from Queensland public and private hospitals. (

      Queensland Department of Health. Queensland Hospital Admitted Patient Data Collection (QHAPDC) Manual. 2019.

      ) Queensland had 122 public and 109 private hospitals by the end of the study period. (

      Australian Inst. of Health and Welfare. Admitted patient care 2017-2018. Canberra: 2018.

      ) This network of hospitals provides universal healthcare access, free hospital care in public hospitals and private healthcare for those with private health insurance for an estimated population of 4.51 million in 2010. Overseas migration from Queensland was low during the study period (1%–1.3%). Interstate migration resulted in a net gain in the Queensland population throughout the study period. The median age of individuals migrating out of Queensland (overseas or interstate) was < 30 years. (

      Australian Bureau of Statistics. Migration, Australia, 2015-16. 2017 2017. https://www.abs.gov.au/AUSSTATS/[email protected]/Lookup/34 (accessed May 25, 2021).

      ,

      Australian Bureau of Statistics. Migration 3412.0. 2007.

      )
      Cases were linked to both the Queensland Hospital Admitted Patient Data collection (QHAPDC) and the Queensland Government's registry of deaths. (

      Queensland Department of Health. Queensland Hospital Admitted Patient Data Collection (QHAPDC) Manual. 2019.

      ) The linkage was performed by the Queensland Health Statistical Services Branch linkage unit. They utilised a sequential process of deterministic and rule-based linkage followed by probabilistic linkage for those records unable to be matched with the aforementioned process. (

      Queensland Department of Health. Queensland Data Linkage Framework. vol. 1. 2014.

      ) This linkage dataset served as the pool for cases to be evaluated for propensity matching.
      From this dataset, all cases of GN-BSI recorded on the AUSLAB pathology database during the years 2005 to 2010 in the state of Queensland, Australia, were included for propensity matching. This time frame was chosen as the 3rd edition ICD-10-AM had been updated to standardise sepsis terminology and was implemented in Australian hospitals in 2004. (

      Independent Hospital Pricing Authority. Chronicle of The International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, Australian Modification (ICD-10-AM). 2019.

      ) Admission and survival data were censored at five years for this study.
      This study is reported according to the RECORD and STROBE statements. (
      • Benchimol EI
      • et al.
      The REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) Statement.
      ;
      • von Elm E
      • et al.
      The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) Statement: Guidelines for Reporting Observational Studies.
      )

      Inclusion criteria: GN-BSIs and culture-negative hospital admissions

      GN-BSIs included at least one positive blood culture for one of the 10 most common gram-negative pathogens in Australia considered not to have a chronic carriage state (Salmonella sp. were excluded). (

      Coombs G et al. Sepsis Outcome Programs 2016 Report. Aust Comm Qual Saf Heal Care 2016.

      ) These organisms were Escherichia coli, Klebsiella pneumoniae, Pseudomonas aeruginosa, Enterobacter cloacae, Klebsiella oxytoca, Proteus mirabilis, Serratia marcescens, Klebsiella (formerly Enterobacter) aerogenes, Citrobacter freundii and Morganella morganii. Included cases were 20 years or older, had no positive blood cultures in the prior year, an admission associated with culture collection and complete gender data.
      The pool of negative culture cases consisted of individuals with a negative blood culture during the same time frame as GN-BSI cases (2005-2010), no positive blood cultures in the prior year, an admission associated with blood culture collection and complete gender data.
      All comorbidities listed in the Charlson-Deyo Comorbidity Score were established from ICD-10-AM coding to derive the Charlson-Deyo Comorbidity Score as validated by Quan and colleagues for use with ICD-10-AM coding. (
      • Charlson ME
      • et al.
      A new method of classifying prognostic comorbidity in longitudinal studies: development and validation.
      ;
      • Deyo RA
      • et al.
      Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases.
      ;
      • Gasparini A.
      comorbidity: An R package for computing comorbidity scores.
      ;
      • Quan H
      • et al.
      Coding Algorithms for Defining Comorbidities in ICD-9-CM and ICD-10 Administrative Data.
      ) If sepsis was recorded during an admission commencing seven days prior or seven days after positive blood culture collection, then the diagnosis of sepsis was assumed for this study to be associated with the blood culture collection. The presence of neutropenia was assessed in the GN-BSI cohort. For this study, neutropenia was defined as an absolute neutrophil count of ≤ 0.5 ✕ 109/L, determined by the lowest neutrophil count in the 48 hours following blood culture collection. (
      • Lingaratnam S
      • et al.
      Introduction to the Australian consensus guidelines for the management of neutropenic fever in adult cancer patients, 2010/2011.
      ) The ICD-10-AM coding list for sepsis is provided in the supplementary section.

      Data sources

      The AUSLAB pathology database records all blood cultures collected in Queensland public hospitals. The QHAPDC meets the legislative requirement for reporting hospital admission and separations in Queensland for public and private hospitals. (

      Queensland Department of Health. Queensland Hospital Admitted Patient Data Collection (QHAPDC) Manual. 2019.

      ) It records the principal admission diagnosis and records up to 20 additional diagnoses according to the ICD-10-AM. Mortality data was derived from the Registry of deaths which reports all deaths in the state of Queensland. Principal cause of death is recorded as the underlying cause of death (UCOD) along with multiple contributory causes of death in ranked order, referred to as other cause of death. The UCOD recorded on the death certificate is defined as ‘the disease or injury which initiated the train of morbid events leading directly to death, or the circumstances of the accident or violence which produced the fatal injury’. When more than one condition is entered on the death certificate, coders select the underlying cause using the coding rules of the ICD, recording the UCOD and multiple other causes of death sequentially in rank (

      Australian Bureau of Statistics. Cause of Death Certification Australia. Information Paper. 2008.

      ). The first ranked other cause of death is referred to in this study as the contributing cause of death (CCOD).

      Approach

      Previous articles have identified a number of challenges in defining the long-term outcome following sepsis. (
      • Shankar-Hari M
      • et al.
      Evidence for a causal link between sepsis and long-term mortality: a systematic review of epidemiologic studies.
      ;
      • Shankar-Hari M
      • Rubenfeld GD.
      Understanding Long-Term Outcomes Following Sepsis: Implications and Challenges.
      ) We attempted to address these challenges by propensity matching by age (in 5 year strata), gender, year of culture collection, pre-illness trajectory, Charlson-Deyo comorbidity score and individual comorbidities listed within the Charlson-Deyo score. The individual comorbidities included in the generation of the propensity score were: acute myocardial infarction, congestive heart failure, peripheral vascular disease, cerebrovascular disease, dementia, chronic obstructive pulmonary disease, rheumatoid disease, peptic ulcer disease, mild liver disease, moderate or severe liver disease, diabetes without complication, diabetes with complication, hemiplegia or paraplegia, renal disease, cancer (in the Charlson-Deyo comorbidity score lymphoma and leukaemia are included in cancer rather than the separate categories used in the original Charlson comorbidity score), metastatic cancer or acquired immune deficiency syndrome. Pre-illness trajectory was measured by the number of admissions (0, 1 to 5, 6 to 10, 11 to 20, 21 to 30 or > 30) in the prior year, beginning 30 days prior to blood culture collection
      Propensity matching was conducted using nearest neighbour 1:1 propensity matching as it has been shown to reduce bias, and is simple to implement and understand. (
      • Rubin D.
      Matching to Remove Bias in Observational Studies.
      ;
      • Stuart EA.
      Matching methods for causal inference: A review and a look forward.
      ) Matching was performed without replacement and set to a caliper distance of up to 0.2 standard deviations.

      Readmission for sepsis

      Readmission was assessed using ICD-10-AM admission coding for sepsis (see supplemental data for ICD-10-AM codes). Readmission was evaluated with landmark analysis from 91 to 365 days and in each subsequent year up to five years.

      Mortality assessment

      Mortality was reported with Kaplan Meier survival curves with the log-rank test. Time zero was defined as the date of index culture collection. A number of sensitivity analyses were also undertaken for known risk factors for mortality in patients with GN-BSI. The mortality assessment was stratified for the presence or absence of the sepsis syndrome to examine the potential dose effect of severity of infection. In addition, the presence of neutropenia was evaluated within the first 48 hours following culture collection for the GN-BSI cohort. Patients without a white cell count collected during this time were assumed not to be neutropenic.

      UCOD and CCOD

      The most common UCOD were described for GN-BSI and culture-negative cases. The UCOD were grouped into the time frames 0 to 90, 91 to 365 days and each subsequent year until five years in a landmark analysis. The most common CCOD were also evaluated in the same time frames.

      Statistical analysis

      All analyses were undertaken using the statistical software program R studio version 1.1.463 with multiple software packages. (
      • Gasparini A.
      comorbidity: An R package for computing comorbidity scores.
      ;
      • Ho DE
      • et al.
      MatchIt : Nonparametric Preprocessing for.
      ;

      Kassambara A et al. Package “survminer” Type Package Title Drawing Survival Curves using “ggplot2.” 2020.

      ;

      Wickham H. ggplot2: Elegant Graphics for Data Analysis. 2016.

      ) Propensity matching was conducted according to nearest neighbour matching using the Matchit package as described by Ho and colleagues. (
      • Ho DE
      • et al.
      MatchIt : Nonparametric Preprocessing for.
      ) Greedy rather than optimal matching was used in the selection of treated units. Matching was undertaken without replacement. A caliper distance of up to 0.2 standard deviations was set with the aim of optimising matching at baseline without excluding a large number of GN-BSI cases.
      Assessment of proportions pre- and post-match was described according to percentage and the standardised difference of means (for non-binary variables) for relevant covariates with the acceptable difference set at 0.1. Differences in the proportion of the outcome of cause of death was evaluated with Pearson's Chi-Squared test where applicable. (
      • Stuart EA.
      Matching methods for causal inference: A review and a look forward.
      )

      Results

      A pool of 13,250 culture-negative cases was used to generate suitable 1:1 matching for 7,061 GN-BSI cases. Matching was performed at a caliper distance of 0.1 and 0.2 standard deviations. Fair matching was achieved at 0.2 standard deviations with the exclusion of a small number of GN-BSI cases (53 cases). In total, 7,008 GN-BSI cases were successfully matched, as shown in Figure 1. Pre- and post-match distribution of covariates is shown in Table 1 and 2 respectively. The distribution of propensity scores and the standardised difference of means for non-binary variables pre/post-matching is shown in the supplemental section. The matched cohorts had a median age-strata of 60–64 years of age, a male predominance of 53% and a mean Charlson-Deyo Comorbidity Score of 1.9 (see Table 2). No admissions in the prior year was the most common admission category (48%), and one to five admissions was the second most common (45%; see Table 2). The mean length of stay for the admission associated with the index culture was 11 days for GN-BSI cases and eight days for culture-negative cases. In the GN-BSI group, 42% had associated coding for sepsis acutely associated with the positive blood culture. In total, 5% of the culture-negative cases had sepsis coding associated with their negative blood culture.
      Figure 1
      Figure 1Linkage and patient flow diagram for measuring long-term mortality and cause of death following GN-BSI.
      Table 1Distribution of covariates prior to nearest neighbour propensity matching.
      Pre-match
      Culture-negativeGN-BSIOverall
      (n = 13,250)(n = 7,061)(n = 20,311)
      Male7,652 (57.8%)3,702 (52.4%)11,354 (55.9%)
      Age strata (years, mean)60 to 6460 to 6460 to 64
      Charlson-Deyo Comorbidity Score (mean)1.721.8
      Comorbidities
      Acute myocardial infarction1,161 (8.8%)431 (6.1%)1,592 (7.8%)
      Congestive heart failure1,521 (11.5%)669 (9.5%)2,190 (10.8%)
      Peripheral vascular disease589 (4.4%)301 (4.3%)890 (4.4%)
      Cerebrovascular disease891 (6.7%)313 (4.4%)1,204 (5.9%)
      Dementia791 (6.0%)362 (5.1%)1,153 (5.7%)
      Chronic obstructive pulmonary disease1,463 (11.0%)477 (6.8%)1,940 (9.6%)
      Rheumatoid disease151 (1.1%)61 (0.9%)212 (1.0%)
      Peptic ulcer disease133 (1.0%)65 (0.9%)198 (1.0%)
      Mild liver disease245 (1.8%)283 (4.0%)528 (2.6%)
      Moderate or severe liver disease109 (0.8%)168 (2.4%)277 (1.4%)
      Diabetes without complication866 (6.5%)546 (7.7%)1,412 (7.0%)
      Diabetes with complication1,856 (14.0%)1,107 (15.7%)2,963 (14.6%)
      Hemiplegia or paraplegia510 (3.8%)245 (3.5%)755 (3.7%)
      Renal disease1,314 (9.9%)946 (13.4%)2,260 (11.1%)
      Cancer1,849 (14.0%)1,453 (20.6%)3,302 (16.3%)
      Metastatic cancer794 (6.0%)613 (8.7%)1,407 (6.9%)
      Acquired immunodeficiency syndrome17 (0.1%)11 (0.2%)28 (0.1%)
      Admissions in year prior to blood culture collection
      No admission7,099 (53.6%)3,350 (47.4%)10,449 (51.4%)
      1 to 5 admissions5,640 (42.6%)3,152 (44.6%)8,792 (43.3%)
      6 to 10 admissions293 (2.2%)291 (4.1%)584 (2.9%)
      11 to 20 admissions145 (1.1%)163 (2.3%)308 (1.5%)
      21 to 30 admissions27 (0.2%)48 (0.7%)75 (0.4%)
      > 30 admissions46 (0.3%)57 (0.8%)103 (0.5%)
      Note. Abbreviation: gram-negative bloodstream infection, GN-BSI.
      Table 2Distribution of covariates post 1:1 propensity matching.
      Post-match
      Culture-negativeGN-BSIOverall
      (n = 7,008)(n = 7,008)(n = 14,016)
      Male3,632 (51.8%)3,673 (52.4%)7,423 (53%)
      Age strata (years, mean)60 to 6460 to 6460 to 64
      Charlson-Deyo Comorbidity Score (mean)1.91.91.9
      Comorbidities
      Acute myocardial infarction411 (5.9%)431 (6.2%)849 (6.1%)
      Congestive heart failure665 (9.5%)668 (9.5%)1,344 (9.6%)
      Peripheral vascular disease298 (4.3%)299 (4.3%)597 (4.3%)
      Cerebrovascular disease318 (4.5%)313 (4.5%)597 (4.3%)
      Dementia377 (5.4%)362 (5.2%)749 (5.3%)
      Chronic obstructive pulmonary disease464 (6.6%)474 (6.8%)917 (6.5%)
      Rheumatoid disease51 (0.7%)60 (0.9%)111 (0.8%)
      Peptic ulcer disease60 (0.9%)65 (0.9%)133 (0.9%)
      Mild liver disease219 (3.1%)252 (3.6%)477 (3.4%)
      Moderate or severe liver disease103 (1.5%)131 (1.9%)237 (1.7%)
      Diabetes without complication509 (7.3%)541 (7.7%)1,075 (7.7%)
      Diabetes with complication1,097 (15.7%)1,099 (15.7%)2,169 (15.5%)
      Hemiplegia or paraplegia247 (3.5%)243 (3.5%)466 (3.3%)
      Renal disease892 (12.7%)933 (13.3%)1,822 (13.0%)
      Cancer1,353 (19.3%)1,435 (20.5%)2,768 (19.7%)
      Metastatic cancer566 (8.1%)600 (8.6%)1,152 (8.2%)
      Acquired immunodeficiency syndrome10 (0.1%)11 (0.2%)20 (0.1%)
      Admissions in year prior to blood culture collection
      No admission3,365 (48.0%)3,341 (47.7%)6,769 (48.3%)
      1 to 5 admissions3,177 (45.3%)3,125 (44.6%)6,227 (44.4%)
      6 to 10 admissions252 (3.6%)285 (4.1%)551 (3.9%)
      11 to 20 admissions141 (2.0%)160 (2.3%)299 (2.1%)
      21 to 30 admissions27 (0.4%)41 (0.6%)68 (0.5%)
      > 30 admissions46 (0.7%)56 (0.8%)102 (0.7%)
      Note. Abbreviation: gram-negative bloodstream infection, GN-BSI.

      Readmission for sepsis

      There were 9,541 versus 9,082 all-cause readmissions in the 91 to 365 days following index culture collection in the GN-BSI and culture-negative cases, respectively. In the four subsequent years, there were 22,204 (GN-BSI) versus 24,879 (culture-negative) all-cause readmissions. Readmission for sepsis in the 91 to 365 day period was higher in the GN-BSI cohort at 1.7% (165/9,541) versus 0.3% (28/9,082) for the culture-negative cohort (OR 5.7, 95% CI 3.8–8.5, P < 0.001). Longer-term readmission for sepsis remained higher in the GN-BSI cohort in the four following years up to five years, 1.4% (319/22,204) versus 0.5% (124/24,879) for the culture-negative cohort (OR for readmission 2.9, 95% CI 2.4–3.6, P < 0.001). The histogram of readmission for sepsis is shown in Figure 2. The number of sepsis readmissions for GN-BSI by year of index culture collection was 66, 75, 81, 62, 78 and 122 (2005–2010, respectively; i.e., total readmission for sepsis 484 = 165 readmitted from 91 to 365 days and 319 in four subsequent years). Readmission for sepsis for culture-negative cases by year of index culture collection was 18, 27, 18, 29, 26 and 34 (2005–2010, respectively).
      Figure 2
      Figure 2Histogram describing the frequency of readmissions for sepsis from 90 days to five years following index culture collection for GN-BSI and culture-negative cases.
      Culture-negative cases (blue) overlaid with GN-BSI cases (white).
      Note. Abbreviation: gram-negative bloodstream infection, GN-BSI.

      Long-term mortality

      The five-year survival was 52% versus 65% for GN-BSI and culture-negative cohorts (P < 0.0001), respectively, as shown in Figure 3A. When patients with GN-BSI were stratified according to acute coding for sepsis or its absence, there was no long-term difference in mortality of the GN-BSI cases, as shown in Figure 3B. The majority of the divergence in mortality between the two groups occurred prior to 365 days.
      Figure 3
      Figure 3A). Kaplan Meier survival curve with log rank test of propensity-matched GN-BSI and culture-negative cases up to five years following index culture collection. B) Kaplan Meier survival curve with log rank test comparing GN-BSI and culture-negative cases coded with and without sepsis.
      Note. Abbreviation: gram-negative bloodstream infection, GN-BSI.
      Figure 3
      Figure 3A). Kaplan Meier survival curve with log rank test of propensity-matched GN-BSI and culture-negative cases up to five years following index culture collection. B) Kaplan Meier survival curve with log rank test comparing GN-BSI and culture-negative cases coded with and without sepsis.
      Note. Abbreviation: gram-negative bloodstream infection, GN-BSI.
      The presence of neutropenia associated with a GN-BSI culture gave a five-year survival proportion of 28% versus 53% for GN-BSI with and without neutropenia, respectively, (P < 0.0001) as shown in the supplemental section.

      Underlying Cause of Death

      During the first 90 days following index culture, the most common causes of death in GN-BSI and culture-negative cases were gastrointestinal malignancy and respiratory tract infection, respectively (see Figure 4, panel A). Urinary tract infection was the fourth most common UCOD for GN-BSI (equal with acute myocardial infarction). Beyond 90 days, infection did not feature as a common UCOD for either group in the five subsequent years.
      Figure 4
      Figure 4Landmark analysis of the most common UCOD for GN-BSI and culture-negative cases.
      A) 0 to 90 days, B) 91 to 365 days, C) year 2, D) year 3, E) year 4 and F) year 5 post-index culture
      Note. Abbreviation: chronic obstructive pulmonary disease, COPD.
      Figure 4
      Figure 4Landmark analysis of the most common UCOD for GN-BSI and culture-negative cases.
      A) 0 to 90 days, B) 91 to 365 days, C) year 2, D) year 3, E) year 4 and F) year 5 post-index culture
      Note. Abbreviation: chronic obstructive pulmonary disease, COPD.
      During the period 91 to 365 days and up to four years after index culture collection, gastrointestinal malignancy remained the most common cause of death for GN-BSI cases (see Figure 4). For culture-negative cases, respiratory tract malignancy was the most common cause of death for the first year, then ischaemic heart disease in the second year and cerebrovascular disease for years three and four. By the fifth year following culture collection, ischaemic heart disease and acute myocardial infarction were the most common UCOD for GN-BSI and culture-negative cases, respectively (see Figure 4, panel F).
      Given the high frequency of death from gastrointestinal malignancy, evaluation of coding for gastrointestinal malignancy prior to blood culture collection was also completed. In the GN-BSI cohort, there were 405 diagnoses of gastrointestinal malignancy occurring prior to index culture collection and 180 diagnoses of gastrointestinal malignancy in the culture-negative cases (P < 0.001).

      Contributing Cause of Death

      Sepsis was the most common CCOD in the GN-BSI cohort at each time frame until two years post culture collection, as shown in Figure 5, panel A for days 0 to 90, panel B for days 91 to 365 and panels C through to F for years two, three, four and five, respectively.
      Figure 5
      Figure 5Landmark analysis of the most common CCOD for GN-BSI and culture-negative cases.
      A) 0 to 90 days, B) 91 to 365 days, C) year 2, D) year 3, E) year 4 and F) year 5 post-index culture
      Figure 5
      Figure 5Landmark analysis of the most common CCOD for GN-BSI and culture-negative cases.
      A) 0 to 90 days, B) 91 to 365 days, C) year 2, D) year 3, E) year 4 and F) year 5 post-index culture
      Risk ratios for sepsis as a CCOD are described in the landmark analysis in Table 3. Despite sepsis remaining the most common CCOD for the GN-BSI cohort at two years (see Figure 5, panel C), landmark analysis showed that the degree of difference between the GN-BSI cohort and the culture-negative cohort for sepsis as a CCOD at two years was not statistically significant (see Table 3).
      Table 3Landmark analysis of OR for sepsis as the CCOD for the GN-BSI cohort compared to the culture-negative cohort.
      Odds ratio95% Confidence intervalP-value
      0 to 90 days3.62.7 to 4.7< 0.001
      91 to 365 days3.41.9 to 6.3< 0.001
      Year 21.40.8 to 2.40.2
      Year 31.20.6 to 2.30.6
      Year 41.30.6 to 2.90.5
      Year 51.80.7 to 4.50.2

      Discussion

      This observational linkage study found a persistent association with mortality and GN-BSI following propensity matching for the effect of age, gender, year of culture, pre-illness health trajectory and baseline comorbidity. In addition, GN-BSI cases were more likely to be readmitted for sepsis and for sepsis to be a CCOD in the following years. This finding is consistent with studies of sepsis suggesting that a significant proportion of sepsis survivors die acutely within two years as a consequence of sepsis. (
      • Prescott HC
      • et al.
      Late mortality after sepsis: propensity matched cohort study.
      ) Consistent with previous studies, we found GN-BSI associated with neutropenia had a poor survival rate. (
      • Gustinetti G
      • Mikulska M.
      Bloodstream infections in neutropenic cancer patients: A practical update.
      )
      We were not able to demonstrate whether increasing severity of GN-BSI (with or without sepsis) modified subsequent mortality as might be expected. There are two possible explanations for this: GN-BSI prognosticates poorly independently of the presence of sepsis or, alternately, clinical coding from the medical record did not accurately reflect the incidence of sepsis associated with GN-BSI or culture negative cases. Previous studies of coding for sepsis in Australian hospitals have identified high specificity but poor sensitivity in the accuracy of sepsis coding to reflect clinical status during admission. (
      • Ibrahim I
      • et al.
      Accuracy of International classification of diseases, 10th revision codes for identifying severe sepsis in patients admitted from the emergency department.
      ) Under-coding of sepsis may explain the absence of a measurable difference in mortality in GN-BSI with and without sepsis in our study.
      The GN-BSI cohort had a significantly higher proportion of gastrointestinal malignancy despite propensity matching for cancer and metastatic cancer prior to index culture collection. The high proportion of gastrointestinal malignancy in this context suggests gastrointestinal malignancy is a risk factor for GN-BSI. If gastrointestinal malignancy case detection rates drop, such as has likely occurred in the COVID-19 pandemic, GN-BSI as a marker of occult gastrointestinal malignancy should be considered. (
      • Søgaard KK
      • et al.
      Gram-negative bacteremia as a clinical marker of occult malignancy.
      ;
      • Williams E
      • et al.
      The impact of the COVID-19 pandemic on colorectal cancer diagnosis and management: A Binational Colorectal Cancer Audit study.
      )
      During the acute period following index blood culture collection, urinary tract infection and sepsis were common UCOD in patients with GN-BSI. Beyond 90 days, infection was not a common UCOD in the following years for either group.
      Sepsis was the most common CCOD in GN-BSI for one year following the index culture. Cause of death studies commonly evaluate only the underlying cause, which identifies the disease or injury that initiated the events leading to death. This practice may significantly underestimate the contribution of sepsis to mortality. UCOD statistics may be biased towards single events like malignancy. (
      • Heather CS
      • et al.
      Do death certificates accurately record deaths due to bloodstream infection?.
      )
      Wang and colleagues identified a higher risk of readmission for infection in the year following sepsis (
      • Wang T
      • et al.
      Subsequent Infections in Survivors of Sepsis.
      ) consistent with our findings. Persistent vulnerability to infection following severe infection has been hypothesised to occur as a direct consequence of the primary infection. (
      • Yende S
      • et al.
      Long-term Host Immune Response Trajectories among Hospitalized Patients with Sepsis.
      )
      The strengths of this study include the use of a definitive disseminated infection, a large number of cases from multiple sites accessible through the use of a single pathology database for public hospitals in Queensland enabling extensive matching and the long duration of follow-up. In addition, the linkage of the QHAPDC for comorbidity/admission data, the use of death registry coding to interrogate the underlying and the contributory cause of death provide valuable insights into the long-term outcome following GN-BSI.
      The limitations of this study include the use of ICD-10-AM administrative coding data from databases not designed for research purposes. Our study was limited to using temporal association with sepsis for inference regarding the severity. We were unable to evaluate the nature or duration of treatment for the GN-BSI and its effect on subsequent morbidity and mortality. We were also unable to account for the low rate of migration that occurred out of Queensland during the study (

      Australian Bureau of Statistics. Migration, Australia, 2015-16. 2017 2017. https://www.abs.gov.au/AUSSTATS/[email protected]/Lookup/34 (accessed May 25, 2021).

      ), the loss to follow-up or the original source of the infection.
      GN-BSI is associated with a persistent risk of readmission for sepsis and mortality compared to hospital attendees of similar vulnerability. Sepsis is a significant CCOD in patients with GN-BSI for a prolonged period of time.

      Conflict of interest

      The authors declare no conflict of interest.

      Funding Source

      This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

      Ethical Approval Statement

      This study was approved by the Metro North Hospital and Health District Human Research Ethics Committee (HREC/17/QRBW/215).

      References

      1. Australian Bureau of Statistics. Migration, Australia, 2015-16. 2017 2017. https://www.abs.gov.au/AUSSTATS/[email protected]/Lookup/34 (accessed May 25, 2021).

      2. Australian Bureau of Statistics. Cause of Death Certification Australia. Information Paper. 2008.

      3. Australian Bureau of Statistics. Migration 3412.0. 2007.

      4. Australian Inst. of Health and Welfare. Admitted patient care 2017-2018. Canberra: 2018.

        • Benchimol EI
        • et al.
        The REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) Statement.
        PLOS Med. 2015; 12e1001885https://doi.org/10.1371/journal.pmed.1001885
        • Charlson ME
        • et al.
        A new method of classifying prognostic comorbidity in longitudinal studies: development and validation.
        J Chronic Dis. 1987; 40: 373-383
      5. Coombs G et al. Sepsis Outcome Programs 2016 Report. Aust Comm Qual Saf Heal Care 2016.

        • Deyo RA
        • et al.
        Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases.
        J Clin Epidemiol. 1992; 45: 613-619https://doi.org/10.1016/0895-4356(92)90133-8
        • von Elm E
        • et al.
        The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) Statement: Guidelines for Reporting Observational Studies.
        J Clin Epidemiol. 2008; 61https://doi.org/10.1016/J.JCLINEPI.2007.11.008
        • Gasparini A.
        comorbidity: An R package for computing comorbidity scores.
        J Open Source Softw. 2018; 3: 648https://doi.org/10.21105/joss.00648
        • Gustinetti G
        • Mikulska M.
        Bloodstream infections in neutropenic cancer patients: A practical update.
        Virulence. 2016; 7: 280-297https://doi.org/10.1080/21505594.2016.1156821
        • Heather CS
        • et al.
        Do death certificates accurately record deaths due to bloodstream infection?.
        Med J Aust. 2014; 201: 518https://doi.org/10.5694/mja14.00694
        • Ho DE
        • et al.
        MatchIt : Nonparametric Preprocessing for.
        J Stat Softw. 2011; 42: 1-28https://doi.org/10.18637/jss.v042.i08
        • Huerta LE
        • Rice TW.
        Pathologic Difference between Sepsis and Bloodstream Infections.
        J Appl Lab Med. 2019; 3: 654-663https://doi.org/10.1373/jalm.2018.026245
        • Ibrahim I
        • et al.
        Accuracy of International classification of diseases, 10th revision codes for identifying severe sepsis in patients admitted from the emergency department.
        Crit Care Resusc. 2012; 14: 112-118
      6. Independent Hospital Pricing Authority. Chronicle of The International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, Australian Modification (ICD-10-AM). 2019.

      7. Kassambara A et al. Package “survminer” Type Package Title Drawing Survival Curves using “ggplot2.” 2020.

        • Lingaratnam S
        • et al.
        Introduction to the Australian consensus guidelines for the management of neutropenic fever in adult cancer patients, 2010/2011.
        Intern Med J. 2011; 41: 75-81https://doi.org/10.1111/j.1445-5994.2010.02338.x
        • McNamara JF
        • et al.
        Evaluation of quick sequential organ failure assessment and systemic inflammatory response syndrome in patients with gram negative bloodstream infection.
        Infect Dis Heal. 2020; 25: 151-157https://doi.org/10.1016/j.idh.2020.01.003
        • Prescott HC
        • et al.
        Late mortality after sepsis: propensity matched cohort study.
        BMJ. 2016; 353: i2375https://doi.org/10.1136/bmj.i2375
        • Quan H
        • et al.
        Coding Algorithms for Defining Comorbidities in ICD-9-CM and ICD-10 Administrative Data.
        Med Care. 2005; 43: 1130-1139https://doi.org/10.1097/01.mlr.0000182534.19832.83
      8. Queensland Department of Health. Queensland Data Linkage Framework. vol. 1. 2014.

      9. Queensland Department of Health. Queensland Hospital Admitted Patient Data Collection (QHAPDC) Manual. 2019.

        • Rubin D.
        Matching to Remove Bias in Observational Studies.
        Biometrics. 1973; 29: 159-183
        • Shankar-Hari M
        • et al.
        Evidence for a causal link between sepsis and long-term mortality: a systematic review of epidemiologic studies.
        Crit Care. 2016; (Apr 13;20101 2016;13:101)
        • Shankar-Hari M
        • Rubenfeld GD
        Understanding Long-Term Outcomes Following Sepsis: Implications and Challenges.
        Curr Infect Dis Rep. 2016; 18: 37
        • Shankar-Hari M
        • Rubenfeld GD.
        Understanding Long-Term Outcomes Following Sepsis: Implications and Challenges.
        Curr Infect Dis Rep. 2016; 18https://doi.org/10.1007/s11908-016-0544-7
        • Søgaard KK
        • et al.
        Gram-negative bacteremia as a clinical marker of occult malignancy.
        J Infect. 2017; 74: 153-162https://doi.org/10.1016/j.jinf.2016.09.011
        • Stuart EA.
        Matching methods for causal inference: A review and a look forward.
        Stat Sci. 2010; 25: 1-21https://doi.org/10.1214/09-STS313
        • Wang T
        • et al.
        Subsequent Infections in Survivors of Sepsis.
        J Intensive Care Med. 2014; 29: 87-95https://doi.org/10.1177/0885066612467162
      10. Wickham H. ggplot2: Elegant Graphics for Data Analysis. 2016.

        • Williams E
        • et al.
        The impact of the COVID-19 pandemic on colorectal cancer diagnosis and management: A Binational Colorectal Cancer Audit study.
        ANZ J Surg. 2021; https://doi.org/10.1111/ans.17071
        • Yende S
        • et al.
        Long-term Host Immune Response Trajectories among Hospitalized Patients with Sepsis.
        JAMA Netw Open. 2019; 2: 1-15https://doi.org/10.1001/jamanetworkopen.2019.8686