Existing Data Sources in Clinical Epidemiology: Database of Community Acquired Infections Requiring Hospital Referral in Eastern Denmark (DCAIED) 2018–2021

Introduction

Globally, infectious diseases are major causes of morbidity and mortality, of which sepsis accounts for the majority of deaths.¹ Lower respiratory infections remain one of the leading causes of death in both high- and low-income countries.^1,2 Sepsis, a subset of underlying infection with the presence of organ failure, is a syndromic illness associated with high morbidity and mortality.^3,4 Estimated global sepsis cases are 48.9 million per year accounting for up to 20% of global deaths.²

A Danish single-center population-based study in 2015 suggested an incidence of 731/100,000 person-years at risk of community-acquired sepsis of any severity with an observed 30-day mortality of 14%.^5,6 Accordingly, another prospective single-center Danish study from 2021 found an incidence of sepsis (based on Sequential Organ Failure Assessment (SOFA) score of ≥2 at admission) 721/100,000 person-years and a 28-day mortality of 13.8%.⁷ However, sepsis is likely underreported in Denmark.⁸ With an aging and increasingly comorbid population presenting to the emergency department (ED), the proportion of patients suffering from infectious diseases requiring hospital referral and treatment will likely rise. To characterize the burden of disease and identify populations of interest for future prospective studies on infectious diseases and sepsis in the ED, large-scale epidemiological studies are warranted.

The purpose of this database is to establish an infrastructure for epidemiological research as well as assessment of the treatment and care of community acquired infections, sepsis and septic shock among patients with a primary contact to the ED.

Perspectives on Infection and Sepsis Definitions

Sepsis has been described since antiquity.⁹ As a syndrome rather than a disease, clinical definitions of sepsis have been proposed by combining clinical and laboratory variables. Consensus definitions have been endorsed at international conferences held in 1991, 2001, and 2016 (Table 1).^10–12 The initial sepsis definition was defined as a systemic inflammatory response syndrome (SIRS) to infection in 1991¹⁰ but revised in 2001 to incorporate organ damage thresholds but without changing the sepsis definition.¹¹ Since then, the paradigm of sepsis pathophysiology has changed. SIRS was in 2016 ascertained an outdated paradigm with poor specificity and was abandoned as a definition of sepsis.¹² Subsequently, sepsis was defined a life-threatening organ dysfunction caused by a dysregulated host response to infection.¹² In this definition SIRS was eliminated due to its low sensitivity and specificity in discriminating sepsis and noncomplicated infection. Organ dysfunction was defined as an acute increase in the total Sequential Organ Failure Assessment (SOFA) score of at least two points secondary to the infection which is associated with an in-hospital mortality of >10% (Table 2).¹² Septic shock is a subset of sepsis characterized by circulatory failure, and carries a substantially increased risk of death compared to sepsis.^12–14 To early identification of patients with suspected infection outside of critical care units and likely to develop sepsis, a new bedside tool with almost similar predictive validity for in-hospital mortality as SOFA outside the Intensive Care Unit (ICU), the quick Sequential Organ Failure Assessment (qSOFA) was introduced.³ The qSOFA score consists of three variables based on vital parameters (systolic blood pressure, respiratory frequency) and a clinical judgment of cerebral status (Glasgow Coma Score). However, qSOFA had lower validity as a measure of organ dysfunction.³ Therefore, in the 2021 updated international guidelines for the management of sepsis and septic shock, it was recommended against using qSOFA compared to SIRS, National Early Warning Score (NEWS), or Modified Early Warning Score (MEWS) as a single screening tool for sepsis or septic shock.¹⁵

Table 1 Definitions of Sepsis

Table 2 Sequential Organ Failure Assessment (SOFA) score12

Population-based epidemiological studies of community acquired infections, sepsis and septic shock are scarce and the few studies available have found differences in estimates likely reflecting differences in case definitions and quality of data.¹⁶ To determine whether an infection is present or not, is a clinical challenge. Furthermore, with the lack of a gold standard for accurate diagnosis and a heterogenous clinical presentation of sepsis, reported sepsis outcomes are prone to be biased. Table 3 provides an overview of the major epidemiological studies conducted during the past three decades.

Table 3 Major Epidemiological Studies Reporting Sepsis Outcomes and Trends

Materials and Methods

Aim of the Database

The purpose of the database is to conduct epidemiological research and to examine the treatment and care of community acquired infections, sepsis and septic shock.

Study Population and Setting

The database comprises all adult ED visits of patients (age ≥18 years) of the Capital and Zealand Regions of Denmark from January 1, 2018 to January 1, 2022. The EDs provide 24-h acute emergency care for 2.6 million inhabitants with approximately 803,983 annual adult contacts and 257,238 annual admissions.^39,40 All adults living in the catchment area are eligible for inclusion at the time of presentation to the EDs. In Denmark, all residents have access to free health care. Every resident is assigned a unique 10-digit personal civil registration number (CPR-number). This unique CPR-number enables accurate linkage between the Danish national registers⁴¹ and true population-based studies are hereby possible.⁴¹ Data on municipality of residence, migration-, and vital status, and date of birth are retrieved from The Danish Civil Registration System (DCRS) and linked to other registries and databases using the unique CPR-number.⁴¹

The Danish National Patient Registry (DNPR) cover all unique inpatient and outpatient clinical contacts at hospitals in Denmark collecting data regarding dates of admission, discharge, admitting departments and all primary and secondary discharge diagnoses (ICD-10 code system)⁴² from hospitals (except psychiatric departments and hospitals).⁴³ Information regarding the population of persons living in the hospitals catchment area is available from Statistics Denmark website.⁴⁴

The advantages and use of Danish registries in epidemiological studies as well as details of the Danish health care system has been described elsewhere.⁴⁵ Patients in Denmark, who are hospitalized acutely, are admitted through the ED of a public hospital, either by their own initiative, referred by general practitioners, or directly by the emergency ambulance service. Patients experiencing more severe infections requiring acute treatment are often treated initially in the ED, including administration of antibiotics. Patients with more severe organ failure (respiratory failure, circulatory failure, etc) are treated in ICU.

Infection, Sepsis and Septic Shock Definitions

In this cohort, suspected infection is adapted and modified from the US Centers for Disease Control and Prevention and defined clinically as a patient admitted to an ED who has any specimens obtained for culture from any anatomic site (irrespective of the result) and the administration of any antimicrobial agent starting within 24 h after arrival to the ED (Table 4).⁴⁶ Accordingly, the sepsis case definition is adapted and modified from the US Centers for Disease Control and Prevention, and The Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3).^12,46 Septic shock is defined by the need for vasopressor to maintain mean arterial pressure ≥65 mmHg or a serum lactate ≥18 mg/L (2 mmol/L).⁴⁷ See Table 4 for complete definitions.

Table 4 Definitions Used in the Database of Community: Acquired Infections in Eastern Denmark 2018–2022

Data Collection and Variables

Database Population

Since March 18, 2017 all patients records from the EDs and hospitals in the Capital and Zealand Region of Denmark are registered and available as electronic medical records (EMR) in Sundhedsplatforme provided by EPIC©.⁴⁸ By electronic screening and the unique CPR-number it is possible to identify and retrieve information on all patients' unique variables including demographic variables (municipality of residence, migration, and vital status, and date of birth), vital parameters, laboratory values, primary and secondary diagnoses, medication status at admission, and medication administered during hospitalization. As of January 1, 2022, 2,241,652 adult ED cases have been registered during the period of inclusion. Of these 451,825 are unique ED encounters of which suspected infection was registered in 60,316. Sepsis were present in 28,472 and 8027 were defined as septic shock.

Informatics

By joint venture cooperation between authors JGH/TS and the Capital Region of Denmark with EPIC/Health Data (Danish: Sundhedsdata) an electronic data warehouse has been established based on a user-friendly Microsoft Azure cloud computing platform (Danish: Forskerplatformen) in which EMR data – from Danish registries (DCRS and DNPR), laboratorial (registered with the use of the International System of Nomenclature, Properties, and Units (the NPU system) – and microbiological results (Laboratory information system at the Department of Clinical Microbiology, Herlev), medication (categorized according to the Anatomical Therapeutic Chemical Classification System, ATC codes) and vitals (triage and early warning score (EWS) data) are merged and made accessible for data and biostatistical analyses. Continuous update of data is possible after obtaining proper approvals.⁴⁹

Variables

The database includes several descriptive variables, clinical and time dependent variables, in addition to outcome indicators (Table 5).

Table 5 Variables Included in Database of Community Acquired Infections and Sepsis in Eastern Denmark (DCAISED)

Descriptive variables include baseline demographics: age, gender, municipality of residence, migration, and vital status. Inpatient and outpatient diagnosis as well as discharge diagnoses and a main diagnosis from the index hospital contact. Clinical time-dependent variables based on patient data upon ED arrival includes time and date of arrival, triage level and time of assessment by ED personal and physician, time of microbiological requisition for culture and time of positive culture, as well as time of laboratory to analyze requisitions, and initiation of empiric antimicrobial treatment. Clinical variables include vital values and signs at presentation to the ED and during the course of hospitalization; systolic and diastolic blood pressure, heart rate, temperature, oxygen saturation and respiratory frequency, as well as clinical judgment of consciousness in terms of AVPU (“alert”, “voice responsive”, “pain responsive”, “unresponsive”) scale or GCS (Glasgow Coma Score). Laboratory variables for the entire hospitalization are included and consist, but are not limited to arterial blood gas variables (including lactate), hematological variables (eg hemoglobin, leucocytes, thrombocytes), inflammatory markers and acute phase proteins (eg C-reactive protein, procalcitonin) and organ-specific variables (eg creatinine, bilirubin). The laboratory variables and results are registered including the unique identification codes of the laboratory and the specific requisitioner. Microbiological variables include culture results from blood, urine, lower respiratory tract, cerebrospinal fluid (CSF) and other anatomical sites (eg pleural fluid, ascites, swabs from abscess cavities and skin lesions). Culture results include bacterial/fungal species and antimicrobial susceptibility patterns used to determine rates of appropriate therapy. In addition, specific polymerase chain (PCR) results are included (eg influenza virus and SARS-CoV-2 from the respiratory tract) as well as serological tests.

Additional variables include medication administered upon inclusion and during the course of hospitalization. Special emphasis is on antimicrobial treatment (including antibacterials, antifungals, and antivirals), treatment with intravenous fluids and volumes administered, as well as vasopressor agents (ATC codes) (Table 5).

Computational covariates include Charlson Comorbidity Index (CCI), SOFA score, number and site of organ failure. CCI as a proxy for comorbid illness, are included for each enrolled patient upon index contact and computed based on diagnoses up to the previous 10 years before index contact.⁵⁰ The SOFA score is computed from six variables, each representing an organ system (eg pulmonal or circulatory) for each included patient upon index contact. Each organ system is assigned a point value from 0 (normal) up to 4 (severe degree of dysfunction/failure) and ranges from 0 to 24 (Table 2).^12,51 The individual organ systems and presence of degree of dysfunction/failure are based on laboratory variables, vasopressor administration and vital values as described above.

Outcome indicators include occurrence of infections, foci, severity of infections (localized, versus systemic) mortality proportions at 30, 90, and 365 days from index date. Number of organ failures based on SOFA score and ICU care during hospitalization. In case a patient has multiple ED events with the event of interest within the study period, only the first event is included for the outcome indicators.

Algorithm development and validation of definitions used for this cohort is planned as future objectives for investigation.

Completeness, Missing Data, and Data Validity

All patient data stored in the EMR software Sundhedsplatformen in the clinical setting is available in the database. The strength of this database is the detailed clinical data from all EDs and inpatient wards from all hospitals in the Capital and Zealand Region of Denmark. These two regions together account for half of the Danish population. The EMR data is independently recorded and based on data from the public Danish health care system (free tax-paid for all Danish residents), which is almost complete^41,43,45 whereby the risk of selection bias and loss to follow-up is regarded as minimal. Moreover, the catchment area is well-defined in this part of Denmark bordered by the ocean due to the geographical characteristics (Figure 1).

Figure 1 (A) Overview of the five regions of Denmark (Total population ~ 5.8 million). (1) North Denmark Region (population ~0.6 million), (2) Central Denmark Region (population ~1.3 million), (3) Region of Southern (population ~1.2 million) Denmark, (4) Zealand Region (population ~0.8 million) and (5) Capital Region of Denmark (population ~1.8 million). The DCAIED database comprise all adult ED visits of patients (age ≥18 years) in whom an infection is suspected, and the background population are inhabitants living in Regions 4 and 5 (eastern part of Denmark). The EDs provide all together 24-h acute emergency care for 2.6 million inhabitants. Red dots represent EDs in the Zealand region. (B) Overview of the Capital Region of Denmark. EDs are represented by red (24-h acute emergency care) and blue dots (day-time acute emergency care). The municipalities highlighted by green color, and the EDs situated within, are served by the Department of Clinical Microbiology, Herlev.

The vital and laboratory parameters are registered prospectively for routine documentation and not for research purposes. We argue that laboratory parameters are missing in a systematic, nonrandom fashion as biochemical and microbiological samples are retrieved only when a health-care worker deemed it appropriate based on a clinical judgment.

Currently, microbiological data are available for North Zealand Hospitals and Herlev and Gentofte Hospital which are served by the same Department of Clinical Microbiology at Herlev Hospital (see Figure 1 for geographical location). It is planned to establish future cooperation’s with the other clinical microbiological departments in the regions. Electronic medical records are well implemented in all Danish Hospitals and EDs. Initiatives to expand the present database by a similar data-infrastructure application from the western part of Denmark (Figure 1) are relevant for research purposes on a national and international level. Presently, the technological solution for this infrastructure has yet to be established, as the software applications and companies hosting the electronic medical records across the regions in Denmark are different. Validation of data completeness and variables in the database is ongoing.

Permissions and Access for Other Researchers

The data retrieval is approved by the Danish Health and Medicines Authority (DHMA) (record number: 3-3013- 2976/1) and the legal department of the Capital Region of Denmark (record number: P-2019-626). We have sought a waived informed consent since the collection of data does not carry any risk for the involved patients. Permission from an ethical committee and patient consent is therefore not required for this type of database in Denmark. All data are handled according to national and regional laws and in compliance with relevant data protection and privacy regulations. Data are made available in an anonymized form to accomplish the DHMA regulations.

Suggestions for studies and collaboration can be directed to the corresponding author JGH at [email protected] and to author TS at [email protected]. Studies and ideas for collaboration will be evaluated according to feasibility, timeliness, and clinical relevance, among others, together with relevant representatives from the emergency departments, and hospitals in the Capital and Zealand Region of Denmark. Data on the individual-level will not be made publicly available in accordance with Danish law. Multidisciplinary collaborations with other Danish and international researchers are encouraged.

Perspectives on Infection and Sepsis Research and Reported Outcomes

Population-level epidemiological estimates of the incidence of infections and sepsis are largely based on administrative databases (hospital discharge codes, etc) (Table 3). These data are often advantageous in the context of large-scale epidemiologic research, as the data acquisition is less resource demanding. Studies based on administrative databases coherently describe an increasing incidence of sepsis during the past 30 years^{18,19,23–25,28–31,33,35,36,38,52} and declining mortality rates.^{19,24,25,29,31,33,35,36,38,52} However, administrative data is susceptible to bias by different thresholds and variability in diagnosing and coding practices of infectious diseases, sepsis and organ failures. Despite of similar trends in outcome, Gaieski et al³¹ found great variations in incidence and mortality estimates, based on four different administrative coding definitions.^18,19,25,26 Moreover, changing coding practices over time caused by changing definitions of sepsis, increased focus (eg Surviving Sepsis Campaign),⁵³ as well as reimbursement motivations in some countries all add to possible discordance in the reported estimates and challenges comparability of outcomes.

In contrast, clinical databases, as a data source for epidemiological studies, commonly include patients in a prospective protocolized manner, based on consistent and well-defined clinical inclusion criteria. The latter contrasts to administrative definitions based on less precise discharge codes. However, due to the critical nature of severe infections and presence of organ failure, most studies based on clinical databases have only included patients in the ICU setting and epidemiological estimates therefore are prone to selection bias. Studies based on this selection, prior to the consensus definition of 2016, report incidences of severe sepsis treated in the ICU of 46 to 95 per 100,000 population^21,22,24,27 and mortality estimates between 28% and 48%.^{21,22,24,27,32} In the large epidemiological study by Kaukonen et al³² the proportion of sepsis admissions (relative to the total ICU admissions) increased from 7.2% to 11% from 2000–2012, whereas mortality rates decreased from 35.0% to 18.4%. In the study of similar size by Harrison et al²⁴ hospital mortality for admissions with severe sepsis decreased from 48.3% in 1996 to 44.7% in 2004, but the total number of deaths increased with an estimated increase in incidence of severe sepsis of 46 in 1996 to 66 in 2003 per 100,000 and an associated number of hospital deaths per 100,000 population rising from 23 to 30. Due to the heterogeneity in epidemiological estimates and uncertainty about the accuracy of the reported trends, Rhee et al⁴ explored trends in incidence of sepsis and related mortality or discharge to hospice, based on clinical sepsis surveillance definition (sepsis-3) vs ICD-9-CM codes (claims data) and found outcomes based on the clinical surveillance definition stable between 2009–2014. Authors also stated that clinical data likely provide more objective estimates than claims-based data for sepsis surveillance.⁴

Comprehensive meta-analysis of observational and randomized controlled trials (RCTs) delineate great heterogenicity in studies included and questions the decline in mortality rates, as described by administrative data.^54–57 In the study by Luhr et al⁵⁶ 44 RCTs were included and analyzed in the usual care arm during the period 2002–2016. They found a declining trend in 28-day mortality in severe sepsis and septic shock during the years 1991–2013, however, the trend was nonsignificant when controlling for severity of illness. In the study by Vincent et al exploring mortality outcome among ICU RCTs, significant heterogeneity was observed.⁵⁷ Similarly, de Grooth et al⁵⁴ found significant heterogeneity in the control groups of mortality rates, based on 65 septic shock RCTs and a nonsignificant decline in mortality rates 2006–2018. Lately, Bauer et al⁵⁵ found a pooled septic shock estimate of 34.7% based on 170 studies, and mortality rates from RCTs were below previous prospective and retrospective cohort studies. A statistically significant decrease of 30-day septic shock mortality was found between 2009 and 2011, but not after 2011. Whether significant epidemiological trends in outcomes of sepsis and septic shock exist is thus still a controversy and population-based estimates are of high relevance.

The number of patients presenting to EDs is growing in many countries with crowding issues as a consequence.⁵⁸ In Denmark crowding has become a problem in the ICUs and internal medicine departments and it is expected to be a concern in Danish EDs in the near future.⁵⁸ With an increasing population presenting to the ED, of which a substantial proportion are elderly comorbid, and often critically ill patients, the acute medical personnel and health care decision-makers seem to face a major challenge. As sepsis and septic shock are common and severe clinical entities, and likely underreported in Denmark,⁸ epidemiological characterization of community acquired infections requiring hospital referral, sepsis and septic shock is of importance to identity the burden of disease and characterize the population of interest in the ED.

Furthermore, the present database extends the period before and during the COVID-19 pandemic. Register-based studies have delineated a decrease in hospitalization among non-COVID-19 disease groups during lockdown periods due to governmental mitigation strategies, whereas overall mortality rate ratios were higher between lockdown periods for non-COVID-19 respiratory diseases, cancer, pneumonia, and sepsis.⁵⁹ Authors highlight the study limitations in clinical details on disease severity as well as comorbidities of people admitted to hospital in analyses of mortality. Due to the granular individual-level clinical data and longitudinal surveillance in the present database, it is possible to further delineate and address, potential confounding factors by examining pre- and post-pandemic outcome rates for specific conditions and community-acquired infectious diseases within the present cohort.

Recently, perpetual observational studies have been proposed as a study design to continuously collect clinical and demographical data for infectious disease entities.⁶⁰ In our setting, the retrospectively available detailed clinical data results in a similar database based on routinely collected data without the need to enroll patients and thereby avoiding bias.

In all respects, Denmark is a perfect setting for this kind of population-based studies. The available detailed clinical data from the Capital and Zealand Region of Denmark, enables large-scale, precise, and detailed epidemiological research on these clinical conditions.

Conclusion

The Database of Community Acquired infections in Eastern Denmark is a new database established by data from EMR records provided by the software Sundhedsplatformen including all adult ED hospital contacts in the Eastern part of Denmark with the purpose of conducting large-scale epidemiological research on community-acquired infections requiring hospital referral. The database includes a large cohort, with granular data, which gives detailed and precise research opportunities on different aspects of severe community-acquired infections and sepsis. It is planned to expand, improve, and validate the data completeness.