Community-Acquired Escherichia coli Bacteremia after Age 50 and Subsequent Incidence of a Cancer Diagnosis: A Danish Population–Based Cohort Study

Bacteremia, the presence of viable bacteria in the bloodstream, is a critical medical condition, and the underlying cause is not always clear. It is widely recognized that bacteremia caused by Streptococcus gallolyticus subsp. gallolyticus may signal a prevalent colon cancer (1), and thus the diagnostic work-up should be targeted at ruling out this cancer. Several other gut microbes are observed more often in tissue from patients with colorectal cancer than in tissue from healthy controls (2). Escherichia coli, a commensal bacterium in the gastrointestinal tract, is the most frequent cause of urinary tract infection and of community-acquired bacteremia (CAB; refs. 3, 4). In addition to more benign causes, gastrointestinal and urinary tract cancers are also risk factors for E. coli CAB (5). A malignant tumor located in the gastrointestinal or genitourinary tract can disrupt the normal mucosal barrier or cause luminal obstruction, and thereby facilitate bacteremia of the commensal bacteria (6). In the existing literature, only a few case reports describe colorectal cancer presenting with E. coli bacteremia of unknown origin. The accompanying symptoms in the patients are rather nonspecific (e.g., intermittent fever, fatigue, weight loss, and lower back pain; refs. 7–10). Owing to the high incidence of E. coli CAB (approximately 16–19 cases per 100,000 persons per year; refs. 5, 11), any association with cancer would be clinically relevant.

Some studies have shown that specific E. coli phylogenetic groups may play a role in carcinogenesis. Strains belonging to phylogenetic group D and B2 are frequently isolated from colorectal cancers (12, 13). However, it remains unclear whether colonization is primary or secondary to the pathology of malignancy. Few population-based studies have examined the link between E. coli CAB and occult cancers. A recent Danish cohort study conducted by members of our group showed that patients with a discharge diagnosis of Gram-negative bacteremia had a high rate of gastrointestinal and urinary tract cancer diagnosed in the following 6 months (14). This suggests that a more thorough diagnostic examination, to exclude these cancers, also may be warranted in patients presenting with E. coli CAB.

We examined whether E. coli CAB after age 50 was associated with increased incidence of gastrointestinal, hepatobiliary, and urinary tract cancers. We also assessed the cumulative incidence of cancer according to E. coli phylogenetic groups.

Our population-based cohort study was based on data from the North Denmark Bacteremia Research Database (15) during the 1994–2013 period (covering approximately 500,000 inhabitants). We linked data using the Danish National Patient Registry (DNPR; ref. 16), the Danish Cancer Registry (DCR; ref. 17), and the Danish Civil Registration System (18). We used the civil registration number (a unique 10-digit number) to link data among the registries.

The North Denmark Bacteremia Research Database has registered prospectively all bacteremia diagnoses in the study area since 1992, including information on date of admission, date of blood culture sampling, microbial pathogen(s), polymicrobial bacteremia, clinical department, and the focus of infection (15). The focus of an infection is determined based on all microbiological and clinical evidence obtained during the relevant hospital admission, including imaging studies. The same blood culture system (BacT/Alert, bioMérieux) was used throughout the study period and has been described elsewhere (15, 19). Briefly, the blood culture system uses colorimetric technology to provide a visual growth indicator. After a cultured bottle signals positive, it is further examined for identification of the pathogen and susceptibility testing.

We identified all individuals with a registered diagnosis of a first-time E. coli CAB between January 1, 1994, and November 30, 2013. We restricted to patients older than 50 years, as the incidence of gastrointestinal or hepatobiliary tract cancer and urinary tract cancer in the younger population is low (20). We excluded patients with E. coli bacteremia occurring before 1994 to avoid including those with a recurrent episode. We also excluded patients whose first blood culture was obtained >2 days after admission, as these were considered to be hospital-acquired. We obtained information on the focus of the bacteremia from the North Denmark Bacteremia Research Database.

For a subgroup of patients diagnosed with E. coli CAB between January 1, 1995, and December 31, 2004, individual isolates had been classified previously according to phylogenetic groups (A, B1, B2, and D) using a triplex PCR method (21).

Since 1943 all new cancer cases in Denmark have been registered in the DCR (17). The cancers are classified currently according to International Classification of Diseases (ICD), Tenth Revision (ICD-10) codes. We linked persons with E. coli bacteremia to the DCR to identify all incident primary cancers. We obtained information on gastrointestinal or hepatobiliary tract cancers (esophagus, stomach, small intestine, colon, rectum, anus, liver, gallbladder, and pancreas cancer), urinary tract cancers (bladder, kidney, renal pelvis, ureter, and prostate cancer), as well as breast and lung cancer (the latter two cancers serving purpose as “negative outcome”). Cancer stage for solid tumors (at the time of cancer diagnosis or within the first 4 months after diagnosis) is classified according to the Tumor Node Metastasis system. We ascertained cancer stage and classified cancers as “localized,” “advanced” (regional or distant spread), and “unknown stage” (Supplementary Table S1). Finally, we excluded patients diagnosed with gastrointestinal, hepatobiliary, or urinary tract cancers before bacteremia or before 1994 (22).

We linked all patients with E. coli bacteremia to the DNPR, to ascertain information on their hospital history dating back to 1977 (16). In the DNPR, the main condition prompting a hospital admission is registered as the primary diagnosis, and other conditions or diseases are recorded as secondary diagnoses [coded by physicians according to ICD-8 (during 1977–1993) and ICD-10 (since 1994)].

We characterized patients according to their Charlson Comorbidity Index (CCI) score (23), and presence/absence of specific chronic diseases, including liver disease, chronic pancreatitis, chronic inflammatory bowel disease, diabetes, chronic kidney disease, and benign prostate hyperplasia. See Supplementary Materials for Appendix with ICD codes.

The prevalence of patient characteristics, including distributions and frequencies of gender, age categories, and medical history, is shown in Table 1. We computed median follow-up time and median age at inclusion (with interquartile range, IQR).

We followed all patients for occurrence of gastrointestinal, hepatobiliary tract, and urinary tract cancer from the E. coli CAB diagnosis date until date of death, emigration, or November 30, 2013, whichever came first. We calculated cumulative incidence of a cancer diagnosis (as a percentage) at 1 and 5 years among patients with E. coli CAB. The cumulative incidence function accounted for death as a competing risk (24). We compared cancer incidence observed among patients with E. coli CAB with that expected in the entire Danish population; expected numbers were estimated based on national cancer incidence rates by age, sex, and calendar year. Standardized incidence ratios (SIR) were calculated as a measure of relative incidence, assuming that the observed number of cases in a specific category followed a Poisson distribution (using Byar's approximation; ref. 25). We computed SIRs for 0–<1 year and ≥1 year following diagnosis of E. coli CAB. We computed confidence intervals (CI) for SIRs and used exact 95% CIs when the observed number was less than ten. SIRs were calculated for gastrointestinal, hepatobiliary, and urinary tract cancers. To evaluate whether observed associations were specific to cancers arising from organs that potentially harbor E. coli, we estimated SIRs of breast and lung cancer as “negative outcome” comparison.

Prevalence ratios of localized and advanced cancer were calculated by dividing the proportion of E. coli CAB cases to the proportion of patients from the general population with same cancer stage. This analysis aimed to examine whether E. coli CAB was an early sign of cancer or a marker of disseminated disease, e.g., if prevalence ratio of localized cancer was higher among E. coli CAB, then the episode could have been the first signal of cancer. By contrast if prevalence ratio of advanced cancer was higher among E. coli CAB, then the bacteremia could have been a marker of occult cancer.

We calculated the prevalence of the phylogenetic groups (A, B1, B2, and D) in a subcohort of 1,136 patients with E. coli CAB with and without subsequent cancer diagnosis, and computed the cumulative cancer incidence by phylogenetic groups.

Statistical analyses were performed using SAS 9.4 (SAS Institute Inc.).

The Danish Data Protection Board approved the study (record number 2008–58–0028). Danish registry data generally are available for research purposes, and informed consent or permission from a health research ethics committee is not needed.

We identified 3,899 individuals with a recorded diagnosis of E. coli CAB during the study period; after exclusion, the final cohort included 2,735 patients. Inclusion and exclusion criteria are shown in a flowchart in Fig. 1.

The cohort of 2,735 patients included 61% females, and median age was 78 years (IQR, 68–84 years). Patients were followed for a total of 10,092 years, with median follow-up of 2.3 years (IQR, 0.4–5.6 years). More than half of patients (59%) had a CCI score of 1 or higher. Among men, 22% had benign prostatic hyperplasia. Diabetes was registered for 14% of all patients, kidney disease for 6% of patients, and chronic liver disease, chronic pancreatitis, and inflammatory bowel disease each were registered for 1% of patients (Table 1). Although patients with previous cancers located in the gastrointestinal, hepatobiliary, or urinary tract were excluded, the cohort included 35 patients with a previous cancer diagnosis at another site.

Most patients in the study cohort (n = 2,504, 92%) had mono-bacterial E. coli CAB. Among all patients, 1,680 (61%) had concurrent urinary tract infection, 306 (11%) had hepatobiliary tract infection, 114 (5%) had abdominal infection (mainly peritonitis), 50 (2%) had other foci recorded, and 585 patients (21%) had unknown focus of their bacteremia. A minority of patients (n = 231, 8%) had polymicrobial bacteremia, most frequently with other enterobacterales, enterococci, streptococci, or anaerobic bacteria (Supplementary Table S2).

Absolute cancer incidence, calculated as a cumulative incidence at 1 year, was 3.0%, with gastrointestinal and hepatobiliary cancers accounting for 1.9% and urinary tract cancers for 1.0%, and the corresponding cumulative cancer incidences at 5 years were 5.5%, 3.8%, and 1.7% respectively (Table 2).

During the full follow-up period, 173 gastrointestinal, hepatobiliary, and urinary tract cancers were diagnosed. As we only expected 90 cancers (based on age and gender distribution), this translated into a SIR of 1.91 (95% CI, 1.64–2.22). Overall cancer incidence was similar for gender, each age group, and focus of infection. Patients with mono-bacterial E. coli bacteremia had an overall cancer SIR of 1.87 (95% CI, 1.59–2.19), whereas patients with poly-bacterial bacteremia (including E. coli) had a cancer SIR of 2.31 (95% CI, 1.41–3.57).

The cancer SIRs for all individual gastrointestinal cancers were approximately 2- to 4-fold increased. Although the SIR of kidney cancer was 3.6-fold increased, we found no evidence of a higher than expected occurrence of bladder or prostate cancer (Table 3). SIRs for renal pelvis and ureter cancer were based on only a few observations and thus had limited precision.

Estimation of SIRs by follow-up time showed an increase mainly within the first year after a bacteremia diagnosis. During this follow-up period, individual SIRs varied between a 3-fold and a 10-fold increase (Table 3). One or more years after the E. coli CAB, the SIRs for several cancers reached 1. However, an excess of colorectal and pancreas cancer diagnoses was evident beyond 1 year of follow-up (Table 3).

In order to examine if patients were already under suspicion for cancer, we looked for recent hospital contacts (within 90 days) before the E. coli bacteremia episode. A total of 879 patients had one or more hospital contact (hospitalization, outpatient contact, emergency room contact) during this preceding period. The disease categories listed as primary discharge diagnosis were heterogeneous and included both diseases of circulatory organs, respiratory organs, urinary tract, gastrointestinal tract, endocrine and metabolic diseases, disease of bone, muscles, and connective tissue, diseases of eye, and lesions etc. The SIR of cancer was almost similar for patients with a recent contact [SIR = 1.86 (95% CI, 1.37–2.48)] and those without a recent contact [SIR = 1.93 (95% CI: 1.61–2.30)].

We found that E. coli CAB was not associated with breast cancer diagnosis (observed number = 22). The 0–<1-year SIR was not calculated due to few numbers; the ≥1-year was 0.98 (95% CI, 0.60–1.52). In contrast, E. coli CAB was associated with a higher occurrence of lung cancer diagnosis (observed number n = 43). The 0–<1-year SIR of lung cancer was 1.72 (95% CI, 0.79–3.28), and the ≥1-year SIR was 1.66 (95% CI, 1.15–2.31). Notably, 15 patients were diagnosed with lung cancer more than 5 years after the bacteremia, and as only 8 were expected, this translated into a SIR of 1.82 (95% CI, 1.02–3.00).

Of the 173 gastrointestinal, hepatobiliary, and urinary tract cancers diagnosed among E. coli CAB patients, cancer stage was known for 116 (67%) patients. Among these patients, the majority of gastrointestinal and hepatobiliary tract cancers were diagnosed at an advanced/metastatic stage (52 of 81 cancers, i.e., prevalence = 64%), whereas urinary tract cancers were less often advanced (9 of 35 cancers, i.e., prevalence = 26%). Compared with the prevalence of cancer stages in the general cancer population, we found that the prevalence ratio was 1.11 (95% CI, 0.94–1.31) for advanced/metastatic gastrointestinal and hepatobiliary tract cancers and 0.94 (95% CI, 0.54–1.66) for advanced/metastatic urinary tract cancers.

For 1,136 patients with E. coli CAB (42% of the cohort), information was available on the phylogenetic group of the infecting strain (21). Among these, 758 (67%) belonged to phylogenetic group B2, 181 (16%) to group D, 152 (13%) to group A, and 45 (4%) to group B1. For the strains from 85 patients subsequently diagnosed with cancer, the distribution of phylogenetic groups was similar. We could not show clear differences between cumulative cancer incidence according to phylogenetic groups. However, patients with strains in phylogenetic group D had the highest point estimate of 8.8 stemming mainly from a high incidence of gastrointestinal and hepatobiliary tract cancers. The overall cumulative cancer incidence was 5.9 for phylogenetic group A, 6.7 for group B1, and 7.5 for group B2. Cumulative incidences for urinary tract cancer were in similar across the four phylogenetic groups (Table 4).

In this population-based study, we found a higher than expected incidence rate of gastrointestinal, hepatobiliary, and urinary tract cancer in patients with a first-time E. coli CAB.

There is a strong link between bacteremia caused by S. gallolyticus subsp. gallolyticus and prevalent colon cancer (1), and patients presenting with this type of bacteremia should be offered colonoscopy to rule out colorectal cancer (and echocardiography to rule out infectious endocarditis). However, a similar association could also apply to other gut microbes. A few case reports describe patients presenting with E. coli bacteremia of unknown focus and unspecific symptoms who are then diagnosed with occult colorectal cancer (7–10). The altered micromilieu surrounding a solid tumor favors colonization of bacteria (26), and a selection of bacteria with genotoxic properties may even play a role in the initiation and promotion of colorectal cancer (26). In fact, certain strains of E. coli have been shown to be more prevalent in tissue from patients with colorectal cancer than in tissue from healthy controls (2). It thus seems plausible that bacteremia of commensal bacteria can be facilitated by impaired mucosa or luminal obstruction with a subsequent increase of mucosal permeability (6). Whereas the incidence of bacteremia with S. gallolyticus is low, E. coli is the most frequent pathogen in CAB—accordingly any association with cancer is clinically important.

A recent cohort study by members of our group found an elevated rate of gastrointestinal and urinary tract cancer during the first 6 months after a diagnosis of Gram-negative bacteremia (14). In addition, an increased incidence of kidney cancer persisted one or more years after the bacteremia diagnosis. This earlier study was not restricted to CAB and lacked information on the specific pathogen (14). In the current study, we found similarly increased cancer incidence within the first year after E. coli CAB. Moreover, the relative incidences of colorectal and pancreatic cancer remained increased beyond 1 year of follow-up. The data lacked sufficient power to clearly confirm an increased incidence of kidney cancer more than 1 year following the bacteremia diagnosis.

The reasons for the persistently increased risk more than 1 year after the bacteremia are not entirely clear. We know that progression from a colonic polyp to colorectal adenocarcinoma may take as long as 10 to 20 years, i.e., the tumor may have been present in some patients at the time of bacteremia.

In Denmark, screening for colorectal cancer was introduced in 2014, i.e., there was no standardized screening for the cancers of interest in this study during our follow-up period.

We included estimation of breast and lung cancer occurrence as “negative outcome” comparisons. As there was no association with breast cancer, we were somewhat puzzled to find an association with lung cancer. Although E. coli is a rare cause of pulmonary infection, it has been documented as a pathogen in pulmonary infection in patients with lung cancer (27). Bronchial obstruction by a lung tumor could facilitate E. coli infection.

Alternatively, shared risk factors for infection and cancer may play a role (e.g., lifestyle factors). The observed elevated SIRs for several other smoking-related cancers (e.g., esophageal cancer) among E. coli CAB patients could be mediated through smoking. Unfortunately, we did not have information on smoking and alcohol use. We also noted a high prevalence of diabetes in our cohort, known as a risk factor for both infection and cancer (28). It is however unlikely that this fully explains the association with for instance colorectal cancer.

We found that the majority of gastrointestinal and hepatobiliary tract cancers in patients with E. coli CAB were diagnosed at an advanced/metastatic stage, whereas urinary tract cancers were less often advanced stage. However, the cancer stage distributions were similar among E. coli patients and the general population, i.e., there were no indication of earlier cancer diagnosis among patients presenting with E. coli CAB.

Although E. coli is a commensal inhabitant of the lower gastrointestinal tract, pathogenic strains are thought to play a role in predisposing individuals to colorectal cancer. We examined the distribution of E. coli phylogenetic types among patients with and without subsequent cancer because there are indications that E. coli virulence factors may be linked with colorectal cancer.

In this context, the phylogenetic groups B2 and D are considered to be pathogenic strains; group B2 was identified in bacteremia (29) and in uropathogenic strains (30), whereas phylogenetic group D was reported to be more prevalent in colorectal cancers than healthy controls (12, 13).

E. coli strains of the B2 phylogroup produce several virulence factors, including pili adhesins which enable colonization of the urinary tract (31). Some have speculated if these adhesins also could favor a colonization of a malignant tumor in the colon. Cyclomodulins also seem especially interesting because of genotoxic effects and an ability to modulate cellular differentiation, apoptosis, and proliferation, and thereby be potential inducers of cancer (31). The prevalence of cyclomodulin-producing B2 strain of E. coli was higher in late stage colon cancer than early stage colon cancer, suggesting a possible link with development and prognosis of colon cancer (12). The polyketide synthase gene complex (pks) products cause epithelial DNA damage; pks is also responsible for producing the genotoxin colibactin, which has been shown to induce inflammation-associated colorectal cancer in mice (32). E. coli associated with colonic mucosa biopsies from patients with colorectal cancer more often showed expression of the pks island than biopsy-associated strains from healthy individuals (31).

Overall, we found a similar distribution of phylogenetic groups among patients with and without subsequent cancer and no clear differences in cumulative cancer incidence across phylogenetic groups. We were not able to show that phylogenetic group was associated with cancer incidence.

Because of Denmark's universal health care coverage, the risk of selection bias in our data was minimal. We used data from a bacteremia database with complete information on all E. coli CAB in one region of Denmark; therefore, E. coli diagnoses had both high sensitivity and specificity. We had complete follow-up for cancer and mortality, and information on phylogenetic groups among patients both with and without a subsequent cancer diagnosis.

Some of the patients, diagnosed with cancer within the first year following bacteremia, may have had other symptoms suspicious of cancer, and some may already have been enrolled in diagnostic work-up for cancer. However, as we restricted the study to CAB, the bacteremia was prevalent at the time of hospital admission and not related to interventions performed during the admission. Although we treated death as competing risk, our reported absolute cancer incidence could be attenuated if the patients dying from the bacteremia episode in fact had occult cancer that was never diagnosed.

In conclusion, E. coli CAB may sometimes be the first clinical presentation leading on to the diagnosis of a gastrointestinal, hepatobiliary tract, or urinary tract cancer. We found that gastrointestinal and hepatobiliary cancers were diagnosed at a late stage in our cohort, whereas urinary tract cancers were more often diagnosed at early stage. We did not find that phylogenetic group was clearly linked with cancer incidence among patients with E. coli CAB.

Based on the high incidence of E. coli CAB, our finding of a higher occurrence of cancer among affected patients is of clinical significance. As it is unlikely that it would be cost-efficient to perform endoscopic examination or CT scans in any patient presenting with E. coli CAB, we need to identify the subgroup of patients where a thorough diagnostic examination to rule out an incident cancer would be valuable.

K.K. Søgaard reports grants from Region Nord Forskningsfond, Harboe fonden, Th. Maigaards Eftf. Fru Lily Benthine Lunds Fond af 1. juni 1998, and Trigon Fonden tidl. Civilingeniør Bent Bøgh og hustru Inge Bøghs Fond during the conduct of the study. H.T. Sørensen reports grants from various (the Department of Clinical Epidemiology is involved in studies with funding from various companies as research grants to and administered by Aarhus University; none of these studies are related to the current study) outside the submitted work. No potential conflicts of interest were disclosed by the other authors.

K.K. Søgaard: Conceptualization, funding acquisition, visualization, methodology, writing–original draft, project administration, writing–review and editing. K. Veres: Software, formal analysis, methodology, writing–review and editing. C.M.J.E. Vandenbroucke-Grauls: Conceptualization, writing–review and editing. J.P. Vandenbroucke: Conceptualization, methodology, writing–review and editing. H.T. Sørensen: Conceptualization, resources, data curation, funding acquisition, methodology, writing–review and editing. H.C. Schønheyder: conceptualization, resources, data curation, funding acquisition, writing–review and editing.

This work was supported by Region Nord Forskningsfond (K.K. Søgaard), Harboe fonden (K.K. Søgaard), Th. Maigaards Eftf. Fru Lily Benthine Lunds Fond af 1. juni 1998 (K.K. Søgaard), and Trigon Fonden tidl. Civilingeniør Bent Bøgh og hustru Inge Bøghs Fond (K.K. Søgaard). The sponsors did not have influence on the design or conduct of the study or the collection, management, analysis, and interpretation of the data, nor did they influence preparation, review, approval of the article, or the decision to submit the article for publication.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.