Abstract
Antibiotics use is associated with higher colorectal cancer risk, but little is known regarding any potential effects on survival.
We conducted a nationwide cohort study, using complete-population data from Swedish national registers between 2005 and 2020, to investigate prediagnostic prescription antibiotics use in relation to survival in colorectal cancer patients.
We identified 36,061 stage I–III and 11,242 stage IV colorectal cancer cases diagnosed between 2010 and 2019. For stage I–III, any antibiotics use (binary yes/no variable) was not associated with overall or cancer-specific survival. Compared with no use, moderate antibiotics use (total 11–60 days) was associated with slightly better cancer-specific survival [adjusted HR (aHR) = 0.93; 95% confidence interval (CI), 0.86–0.99)], whereas very high use (>180 days) was associated with worse survival [overall survival (OS) aHR = 1.42; 95% CI, 1.26–1.60, cancer-specific survival aHR = 1.31; 95% CI, 1.10–1.55]. In analyses by different antibiotic types, although not statistically significant, worse survival outcomes were generally observed across several antibiotics, particularly macrolides and/or lincosamides. In stage IV colorectal cancer, inverse relationships between antibiotics use and survival were noted.
Overall, our findings do not support any substantial detrimental effects of prediagnostic prescription antibiotics use on cancer-specific survival after colorectal cancer diagnosis, with the possible exception of very high use in stage I–III colorectal cancer. Further investigation is warranted to confirm and understand these results.
Although the study findings require confirmation, physicians probably do not need to factor in prediagnostic prescription antibiotics use in prognosticating patients with colorectal cancer.
Introduction
Antibiotics use has recently been recognized as a novel, plausible and modifiable risk factor for development of colorectal cancer (1–3), the second leading cause of cancer death world-wide (4). The predominant hypothesis posits that antibiotics use disrupts the composition and function of the gut microbiome, creating an environment conducive to carcinogenesis in the epithelium (5, 6). Three recent large epidemiologic studies, including our previous Swedish nationwide study (7–9), have consistently found that antibiotics use was associated with a higher risk of proximal colon cancer but a lower risk of rectal cancer. The findings indirectly support the role of gut microbiota in colorectal cancer development, but also underscore the apparent complexity of the relationship between antibiotics and colorectal carcinogenesis (6, 10, 11).
Clinical studies suggest that, through modulation of intestinal bacteria, antibiotics use during cancer therapy may negatively impact treatment outcomes, particularly when immunotherapy is used (12, 13). A similar effect for earlier use of antibiotics might be plausible, given reported long-term alterations to the gut microbiome (14). Speculatively, exposure to antibiotics might also result in an intestinal environment favorable for specific carcinogenic pathways or tumor aggressiveness, which could, in turn, influence survival. However, little is known about the potential effects of prediagnostic antibiotics on the survival of patients with colorectal cancer. The few studies on the subject report inconsistent results and have several limitations including small study sizes, short follow-up, and lack of data on important confounders (15, 16). Furthermore, they have focused primarily on advanced stage cancers and have not assessed curatively treated patients.
The aim of this study was to investigate prediagnostic antibiotics use in relation to overall and cancer-specific survival in colorectal cancer using Swedish national registry data, stratified for stages I–III treated with a curative intent and stage IV.
Materials and Methods
Study design and population
This is a nationwide population-based cohort study using data from Swedish population registers. In summary, primary colorectal cancer cases were selected, and clinical data were retrieved, from the Swedish Colorectal Cancer Register (17). Information on cause of death (until December 31, 2020) was extracted from the Swedish Cause of Death Register (18). Data on antibiotics use was extracted from the Swedish Prescribed Drug Register (19). Additional data on comorbidities, clinical characteristics, socioeconomic status and family history of cancer were obtained from other national registers including the Swedish National Patient Register (20), the Swedish Cancer Register (21), the Longitudinal Integration Database for Health Insurance and Labor Market Studies (LISA by Swedish acronym; ref. 22), the Total Population Register (23) and the Swedish Multi-generational Register (24).
Exposure variables
Antibiotics exposure data from the Swedish Prescribed Drug Register included all dispensed prescriptions of antibiotics from July 1, 2005 (start of the register). On the basis of the defined daily dose (DDD) reported in the register, cumulative antibiotics exposure prior to colorectal cancer diagnosis was categorized as no recorded use (reference category), low use (1–10 days, corresponding to 1–2 typical prescriptions), moderate use (11–60 days), high use (61–180 days) and very high use (>180 days). This categorization was based on reporting in previous studies, clinical relevance and to capture the full exposure spectrum. Categories were also created for no use vs. any use (binary) and for time-standardized usage (days/year, DDD divided by the number of observation years prior to colorectal cancer diagnosis and categorized as no use, 1–3, 4–7, 8–14, and ≥15 days per year based on the distribution of the data, clinical relevance and the average antibiotics usage in outpatient care in Sweden; ref. 25). Antibiotics use within one year prior to colorectal cancer diagnosis was excluded, to account for potential use due to undiagnosed but clinically manifest colorectal cancer. All antibiotics under Anatomical Therapeutic Chemical (ATC) codes J01 and J04 (anti-infective agents for systemic use), A07AA (intestinal anti-infectives with antibiotic effects) and P01AB (antiprotozoal agents with antibiotic effects) were included. Detailed classifications of antibiotics and the defined daily dose for each antibiotic can be found in Supplementary Tables S1 and S2, respectively.
Endpoint variables
All primary colorectal cancer cases (C18.0, C18.2–18.9, C19, and C20) diagnosed between January 1, 2010, and December 31, 2019, were identified and followed until December 31, 2020. The year 2010 was selected to ensure an exposure observation period of at least 4.5 years. Cases were classified as proximal colon cancer (caecum, ascending colon, hepatic flexure, transverse colon, or splenic flexure), distal colon cancer (descending or sigmoid colon) and rectal cancer, as reported in the Swedish Colorectal Cancer Register. Overall survival (OS) was defined as the time from the date of diagnosis to the date of death from any cause. Cancer-specific survival was defined as the time from the date of diagnosis to the date of death from colorectal cancer, defined as the underlying cause of death reported as any International Classification of Diseases (ICD) code for colorectal cancer (C18.0, C18.2–18.9, C19, and C20). Patients who had died from any cancer other than colorectal cancer (n = 1,946) were excluded from the study population to control for potential confounding or competing risk due to other primary cancers (Fig. 1). Remaining patients were censored at the end of follow-up, at the date of emigration when applicable or, for cancer-specific survival, at the date of death from any cause other than colorectal cancer. The main analysis focused on stage I–III colorectal cancer treated with a curative intent, defined as having undergone resectional surgery with a curative intent. Stage IV colorectal cancer was assessed separately as a secondary analysis. Tumor staging was based on stage at diagnosis as reported in the Swedish Colorectal Cancer Register.
Flow chart of colorectal cancer case selection. The incident primary colorectal cancer cases were selected from the Swedish Colorectal Cancer Register and the Swedish Cancer Register, and their clinical data were retrieved. To extract antibiotics exposure data, outcome variables and other relevant variables for analysis, the study population was further linked to multiple registers, including the Swedish Prescribed Drug Register, the Swedish Cause of Death Register, the Total Population Register, the Longitudinal Integration Database for Health Insurance and Labor Market Studies (LISA by Swedish acronym), the Swedish National Patient Register and the Swedish Multi-generational Register.
Flow chart of colorectal cancer case selection. The incident primary colorectal cancer cases were selected from the Swedish Colorectal Cancer Register and the Swedish Cancer Register, and their clinical data were retrieved. To extract antibiotics exposure data, outcome variables and other relevant variables for analysis, the study population was further linked to multiple registers, including the Swedish Prescribed Drug Register, the Swedish Cause of Death Register, the Total Population Register, the Longitudinal Integration Database for Health Insurance and Labor Market Studies (LISA by Swedish acronym), the Swedish National Patient Register and the Swedish Multi-generational Register.
Variables
Other variables were considered on the basis of their relevance as potential confounders or for subgroup analyses, as well as based on availability from the national registers. They included main demographic factors (sex, age, and calendar year of diagnosis, county of residence, place of birth, and marital status), socioeconomic factors (level of education and disposable income) at or prior to colorectal cancer diagnosis, Charlson's comorbidity index (from 2005 up to colorectal cancer diagnosis and calculated using an algorithm adapted for register-based medical research in Sweden; ref. 26), family history of colorectal cancer (number of first-degree relatives with a registered colorectal cancer diagnosis) and cancer therapy variables (including surgical details such as operation type and preoperative chemotherapy and radiotherapy).
Statistical analysis plan
Overall and cancer-specific survival differences among the exposure groups were illustrated by Kaplan–Meier plots, including log-rank tests.
Cox proportional-hazards regression models were used to calculate overall and cancer-specific survival estimates, reported as hazard ratio (HR) with 95% confidence interval (CI). The proportional-hazard assumption was verified by Schoenfeld residuals’ test. Variables that violated the proportional-hazard (PH) assumption were controlled for by stratification in the adjusted models (27).
In addition to univariable analysis (model 0, no adjustments), we performed three multivariable regression models for stage I–III colorectal cancer. Model 1 included main demographic and socioeconomic factors (sex, age at diagnosis, county, place of birth, marital status, level of education, and disposable income, as listed in Table 1), as well as family history of colorectal cancer (0, 1, or ≥2 first-degree relatives), calendar year of diagnosis (continuous variable, coded 1–10) and prediagnostic Charlson comorbidity index (continuous variable, ranging 0–14), treating these variables as potential confounders. Model 2 additionally included tumor site (proximal colon, distal colon, rectum, unspecified colorectum) and stage (I, II, III). This was done in a separate step given the potential of these factors to act as confounders and/or mediators of an association between prediagnostic antibiotics use and survival. Model 2 was considered the main adjusted model. Model 3 added therapy-related variables with a potential mediating role, including operation type (elective or emergency), preoperative chemotherapy (yes/no) and preoperative radiotherapy (yes/no). As the covariates each had <1% missing values in the main analysis, we performed complete-case analyses. The model fit was evaluated graphically by Cox-Snell residuals. The hazard functions for the preferred adjusted model (model 2) supported its use for downstream analyses (Supplementary Fig. S1), and the results derived from the preferred model 2 were then presented accordingly. In the preferred model 2, we identified variables that did not satisfy the PH assumption, namely age at diagnosis (continuous), prediagnostic Charlson Comorbidity Index (CCI; continuous), family history of colorectal cancer (0/1/≥2 first-degree relatives), tumor site (proximal colon, distal colon, rectum, unspecified), and tumor stage (stage I–III), and we considered these variables by stratification in the Cox model. For continuous variables not satisfying the PH assumption, we employed binary variable types for age at diagnosis (</≥50 years) and prediagnostic CCI (no/any CCI) in the stratified Cox regression (27). Cut points for binary variables were arbitrary based on clinical relevance (28) and simplicity. To account for residual confounding within each binary stratum (29), we additionally adjusted for the continuous variables age at diagnosis and CCI in the stratified Cox model. Prespecified subgroup analyses included sex, age at diagnosis (early-onset defined as diagnosis <50 years of age) and tumor site. We also ran prespecified analyses for antibiotics classes. Post hoc analyses included antibiotics grouped by typical indications in Sweden (Supplementary Table S1), as well as specific antibiotic drugs based on commonness of the prescriptions, as well as known or suspected effects on the gastrointestinal microbiome (or, for methenamine hippurate, lack thereof).
Characteristics of patients with stage I–III colorectal cancer.
. | Prediagnostic antibiotics usea . | ||
---|---|---|---|
Patient characteristics . | No use (n = 8,876) . | Any use (n = 27,185) . | |
Age at diagnosis, mean (SD) | 69 (11.6) | 71 (11.1) | |
Age at diagnosis, N (%) | |||
<50 years (early onset) | 540 (6.1) | 1,213 (4.5) | |
≥50 years | 8,336 (93.9) | 25,972 (95.5) | |
Sex, N (%) | |||
Men | 5,229 (58.9) | 13,699 (50.4) | |
Women | 3,647 (41.1) | 13,486 (49.6) | |
Place of birth, N (%) | |||
Sweden | 7,589 (85.5) | 23,822 (87.6) | |
Nordic countries | 453 (5.1) | 1,310 (4.8) | |
Rest of Europe | 521 (5.9) | 1,371 (5) | |
Non-European countries | 313 (3.5) | 682 (2.5) | |
County of residence, N (%) | |||
Region Stockholm | 1,240 (14) | 5,004 (18.4) | |
Region Skåne | 1,197 (13.5) | 3,606 (13.3) | |
Region Västra Götaland | 1,334 (15) | 4,835 (17.8) | |
Other regions | 5,105 (57.5) | 13,740 (50.5) | |
Main demographic and socioeconomic variables | Marital status, N (%) | ||
Married/registered partner | 4,787 (53.9) | 14,627 (53.8) | |
Unmarried | 1,340 (15.1) | 3,069 (11.3) | |
Divorced/separated | 1,364 (15.4) | 4,581 (16.9) | |
Widowed | 1,385 (15.6) | 4,908 (18.1) | |
Education, N (%) | |||
Compulsory education, 9 years | 3,129 (35.3) | 8,983 (33) | |
Secondary education | 3,588 (40.4) | 11,181 (41.1) | |
Postsecondary education | 2,015 (22.7) | 6,781 (24.9) | |
Unknown | 144 (1.6) | 240 (0.9) | |
Disposable income, N (%) | |||
First quartile (lowest) | 2,343 (26.4) | 6,243 (23) | |
Second quartile | 2,038 (23) | 6,658 (24.5) | |
Third quartile | 2,169 (24.4) | 7,146 (26.3) | |
Fourth quartile | 2,326 (26.2) | 7,138 (26.3) | |
Family history of colorectal cancer, N (%) | |||
0 first-degree relative | 7,546 (85) | 23,070 (84.9) | |
1 first-degree relative | 1,177 (13.3) | 3,651 (13.4) | |
≥2 first-degree relatives | 153 (1.7) | 464 (1.7) | |
Prediagnostic CCIb, mean (SD) | 0.6 (1.2) | 1 (1.6) | |
Prediagnostic CCIb, N (%) | |||
No reported comorbidities | 6,315 (71.1) | 15,029 (55.3) | |
Any comorbidity (CCIb 1–14) | 2,561 (28.9) | 12,156 (44.7) | |
Calendar year of diagnosis, N (%) | |||
2010–2013 | 4,374 (49.3) | 8,711 (32) | |
2014–2019 | 4,502 (50.7) | 18,474 (68) | |
Tumor site, N (%) | |||
Colorectum (unspecified tumor site) | 166 (1.8) | 519 (1.9) | |
Proximal colon | 3,238 (36.5) | 11,450 (42.1) | |
Distal colon | 2,240 (25.2) | 7,070 (26.0) | |
Rectum | 3,232 (36.4) | 8,146 (30.0) | |
Stage at diagnosis, N (%) | |||
Tumor characteristics and mortality outcomes | Stage I | 1,948 (21.9) | 6,810 (25.1) |
Stage II | 3,356 (37.8) | 10,182 (37.5) | |
Stage III | 3,572 (40.2) | 10,193 (37.5) | |
Mortality outcome, N (%) | |||
Alive at last date of follow up | 6,453 (72.7) | 20,439 (75.2) | |
Death due to colorectal cancer | 1,357 (15.3) | 3,394 (12.5) | |
Death from other causes | 1,032 (11.6) | 3,314 (12.2) | |
Emigrated | 34 (0.4) | 38 (0.1) | |
Surgical operation type, N (%) | |||
Elective surgery | 8,023 (90.4) | 25,038 (92.1) | |
Emergency surgery | 850 (9.6) | 2,139 (7.9) | |
Missing | 3 (<0.1) | 8 (<0.1) | |
Preoperative radiotherapy, N (%) | |||
Therapy-related variables | Yes | 2,249 (25.3) | 4,894 (18.0) |
No | 6,606 (74.4) | 22,211 (81.7) | |
Missing | 21 (0.2) | 80 (0.3) | |
Preoperative chemotherapy, N (%) | |||
Yes | 818 (9.2) | 1,529 (5.6) | |
No | 8,039 (90.6) | 25,576 (94.1) | |
Missing | 19 (0.2) | 83 (0.3) |
. | Prediagnostic antibiotics usea . | ||
---|---|---|---|
Patient characteristics . | No use (n = 8,876) . | Any use (n = 27,185) . | |
Age at diagnosis, mean (SD) | 69 (11.6) | 71 (11.1) | |
Age at diagnosis, N (%) | |||
<50 years (early onset) | 540 (6.1) | 1,213 (4.5) | |
≥50 years | 8,336 (93.9) | 25,972 (95.5) | |
Sex, N (%) | |||
Men | 5,229 (58.9) | 13,699 (50.4) | |
Women | 3,647 (41.1) | 13,486 (49.6) | |
Place of birth, N (%) | |||
Sweden | 7,589 (85.5) | 23,822 (87.6) | |
Nordic countries | 453 (5.1) | 1,310 (4.8) | |
Rest of Europe | 521 (5.9) | 1,371 (5) | |
Non-European countries | 313 (3.5) | 682 (2.5) | |
County of residence, N (%) | |||
Region Stockholm | 1,240 (14) | 5,004 (18.4) | |
Region Skåne | 1,197 (13.5) | 3,606 (13.3) | |
Region Västra Götaland | 1,334 (15) | 4,835 (17.8) | |
Other regions | 5,105 (57.5) | 13,740 (50.5) | |
Main demographic and socioeconomic variables | Marital status, N (%) | ||
Married/registered partner | 4,787 (53.9) | 14,627 (53.8) | |
Unmarried | 1,340 (15.1) | 3,069 (11.3) | |
Divorced/separated | 1,364 (15.4) | 4,581 (16.9) | |
Widowed | 1,385 (15.6) | 4,908 (18.1) | |
Education, N (%) | |||
Compulsory education, 9 years | 3,129 (35.3) | 8,983 (33) | |
Secondary education | 3,588 (40.4) | 11,181 (41.1) | |
Postsecondary education | 2,015 (22.7) | 6,781 (24.9) | |
Unknown | 144 (1.6) | 240 (0.9) | |
Disposable income, N (%) | |||
First quartile (lowest) | 2,343 (26.4) | 6,243 (23) | |
Second quartile | 2,038 (23) | 6,658 (24.5) | |
Third quartile | 2,169 (24.4) | 7,146 (26.3) | |
Fourth quartile | 2,326 (26.2) | 7,138 (26.3) | |
Family history of colorectal cancer, N (%) | |||
0 first-degree relative | 7,546 (85) | 23,070 (84.9) | |
1 first-degree relative | 1,177 (13.3) | 3,651 (13.4) | |
≥2 first-degree relatives | 153 (1.7) | 464 (1.7) | |
Prediagnostic CCIb, mean (SD) | 0.6 (1.2) | 1 (1.6) | |
Prediagnostic CCIb, N (%) | |||
No reported comorbidities | 6,315 (71.1) | 15,029 (55.3) | |
Any comorbidity (CCIb 1–14) | 2,561 (28.9) | 12,156 (44.7) | |
Calendar year of diagnosis, N (%) | |||
2010–2013 | 4,374 (49.3) | 8,711 (32) | |
2014–2019 | 4,502 (50.7) | 18,474 (68) | |
Tumor site, N (%) | |||
Colorectum (unspecified tumor site) | 166 (1.8) | 519 (1.9) | |
Proximal colon | 3,238 (36.5) | 11,450 (42.1) | |
Distal colon | 2,240 (25.2) | 7,070 (26.0) | |
Rectum | 3,232 (36.4) | 8,146 (30.0) | |
Stage at diagnosis, N (%) | |||
Tumor characteristics and mortality outcomes | Stage I | 1,948 (21.9) | 6,810 (25.1) |
Stage II | 3,356 (37.8) | 10,182 (37.5) | |
Stage III | 3,572 (40.2) | 10,193 (37.5) | |
Mortality outcome, N (%) | |||
Alive at last date of follow up | 6,453 (72.7) | 20,439 (75.2) | |
Death due to colorectal cancer | 1,357 (15.3) | 3,394 (12.5) | |
Death from other causes | 1,032 (11.6) | 3,314 (12.2) | |
Emigrated | 34 (0.4) | 38 (0.1) | |
Surgical operation type, N (%) | |||
Elective surgery | 8,023 (90.4) | 25,038 (92.1) | |
Emergency surgery | 850 (9.6) | 2,139 (7.9) | |
Missing | 3 (<0.1) | 8 (<0.1) | |
Preoperative radiotherapy, N (%) | |||
Therapy-related variables | Yes | 2,249 (25.3) | 4,894 (18.0) |
No | 6,606 (74.4) | 22,211 (81.7) | |
Missing | 21 (0.2) | 80 (0.3) | |
Preoperative chemotherapy, N (%) | |||
Yes | 818 (9.2) | 1,529 (5.6) | |
No | 8,039 (90.6) | 25,576 (94.1) | |
Missing | 19 (0.2) | 83 (0.3) |
Abbreviations: N, number; SD, standard deviation.
aDefined as dispensed antibiotics prescriptions. Antibiotics use during one year before colorectal cancer diagnosis was excluded to account for potential use due to undiagnosed but clinically manifest colorectal cancer.
bCharlson comorbidity index (CCI) calculated from the Swedish Patient Register based on the algorithm developed by Ludvigsson and colleagues (26) for medical register–based research in Swedish settings. CCI was calculated using data from 2005 up to the date of colorectal cancer diagnosis.
We conducted several sensitivity analyses to assess the robustness of our main findings. To address potential time-window bias and time-varying exposure, we conducted analyses limiting the antibiotics exposure to a 4.5-year period prior to diagnosis, to ensure an equally long exposure period for the full study population. We also applied a fixed 10-year prediagnostic period for cases diagnosed after July 1, 2015, to understand the potential varying effects of antibiotics depending on different observation windows of exposure. In both analyses, the time period used to calculate the Charlson comorbidity index corresponded to that of the antibiotics exposure. We further ran the analyses using the time-standardized antibiotics variable (days of antibiotics use per year).
To account for hospital-level variance, we performed a multilevel mixed-effect analysis, assuming a Weibull distribution and including operating hospitals as a second-level variable. We also included analyses of patients with colon cancer stratified by type of surgical operation (elective or emergency). Emergency surgery, typically only necessary in colon cancer, is overrepresented in patients presenting with an inflammatory phenotype, which generally has a poor prognosis (30), and which might, speculatively, more often be treated with antibiotics in the prediagnostic phase. Finally, we performed analyses excluding cases who died within three months after surgery to account for fatal perioperative complications.
For the analyses of stage IV colorectal cancer, we used multivariable Cox proportional-hazards regression models following the same general analysis plan as for stage I–III. Model fitness is illustrated in graphs of Cox–Snell residuals (Supplementary Fig. S2), and details of covariates, subgroup analyses and sensitivity analyses are included in the results figures and tables. We also conducted a Pearson χ2 test for antibiotics use by stage at diagnosis across all stages, I–IV.
All statistical tests were two-sided and were performed using Stata/MP 16.1 (Stata Corp., College Station). P < 0.05 was considered statistically significant. For analyses by different antibiotics types and indications, to ensure a rigorous evaluation of the findings while considering the potential impact of multiple comparisons, a more stringent threshold of statistical significance (0.005) was also applied (31) and both 95% CI and 99.5% CI were presented.
Ethical approval
The study was approved by the Regional Ethical Review Board in Umeå, Sweden (Dnr: 2017/338–31 and 2020/02312) and was conducted in accordance with the Declaration of Helsinki.
Data availability
Data can be requested from the Swedish National Board of Health and Welfare and respective national registries with relevant ethical approval.
Results
The main analysis included 36,061 patients with stage I–III incident colorectal cancer treated with a curative intent and 11,242 patients with stage IV incident colorectal cancer (Fig. 1). The median follow-up time after colorectal cancer diagnosis was 3.9 years (interquartile range: 1.4–5.9 years).
Background characteristics, stage I–III
Of the stage I–III patients, 75% had used prescription antibiotics prior to colorectal cancer diagnosis (after excluding use during the one-year period prior to diagnosis; Table 1). Compared with no antibiotics use, participants who had used antibiotics were, on average, two years older at colorectal cancer diagnosis (mean 71 vs. 69 years), more often women (49.6% vs. 41.1%), more likely to have comorbidities (44.7% vs. 28.9%), more likely to have proximal colon cancer (42.1% vs. 36.5%) and less likely to have rectal cancer (30% vs. 36.4%).
Main survival analyses, stage I–III
No clear dose–response association was observed between prediagnostic antibiotics use and survival outcomes in stage I–III colorectal cancer. Very high use (>180 days total over the study period), compared with no use, had worse survival in both Kaplan–Meier survival analyses (P < 0.001; Supplementary Fig. S3) and multivariable adjusted Cox regression analyses, specifically the preferred model 2 (HR for OS = 1.42; 95% CI, 1.26–1.60, HR for cancer-specific survival = 1.31; 95% CI, 1.10–1.55; Fig. 2), whereas moderate use (11–60 days), was associated with modestly better cancer-specific survival (HR = 0.93; 95% CI, 0.86–0.99). Any versus no antibiotics use (binary variable) was not clearly associated with overall or cancer-specific survival in stage I–III colorectal cancer (HR for OS = 1.01; 95% CI, 0.96–1.06, HR for cancer-specific survival = 0.95; 95% CI, 0.89–1.02; Fig. 2). Patterns of associations were more or less consistent across Cox models (Supplementary Fig. S4), and the results from the preferred model 2 were subsequently presented.
Antibiotics use in relation to overall and cancer-specific survival in stage I–III colorectal cancer. Prediagnostic antibiotics use was categorized as no use, any use, low use (1–10 days), moderate use (11–60 days), high use (61–180 days) and very high use (>180 days), on the basis of defined daily doses. Antibiotics use during one year before colorectal cancer diagnosis was excluded to account for potential use due to undiagnosed but clinically manifest colorectal cancer. Hazard ratios (HR) and 95% confidence intervals (CI) were calculated from the Cox proportional hazards models, stratified by tumor site (proximal colon/distal colon/rectum/unspecified colorectum categories in the analysis for total colorectal cancer), tumor stage (stage I–III), age at diagnosis (</≥50 years), prediagnostic Charlson comorbidity index (no/any comorbidity), family history of colorectal cancer (0/1/ ≥2 first-degree relatives), and adjusted for: age and calendar year of diagnosis and prediagnostic Charlson comorbidity index as continuous variables, and sex, place of birth, county of residence, marital status, level of education and income status as categorical variables, defined as presented in Table 1.
Antibiotics use in relation to overall and cancer-specific survival in stage I–III colorectal cancer. Prediagnostic antibiotics use was categorized as no use, any use, low use (1–10 days), moderate use (11–60 days), high use (61–180 days) and very high use (>180 days), on the basis of defined daily doses. Antibiotics use during one year before colorectal cancer diagnosis was excluded to account for potential use due to undiagnosed but clinically manifest colorectal cancer. Hazard ratios (HR) and 95% confidence intervals (CI) were calculated from the Cox proportional hazards models, stratified by tumor site (proximal colon/distal colon/rectum/unspecified colorectum categories in the analysis for total colorectal cancer), tumor stage (stage I–III), age at diagnosis (</≥50 years), prediagnostic Charlson comorbidity index (no/any comorbidity), family history of colorectal cancer (0/1/ ≥2 first-degree relatives), and adjusted for: age and calendar year of diagnosis and prediagnostic Charlson comorbidity index as continuous variables, and sex, place of birth, county of residence, marital status, level of education and income status as categorical variables, defined as presented in Table 1.
Subgroup analyses, stage I–III
Patterns of associations were also largely consistent across anatomical tumor sites (Fig. 2), with the exception of a null relationship between very high antibiotics use and cancer-specific survival in proximal colon cancer. The somewhat better cancer-specific survival was observed for moderate antibiotics use in patients with distal colon cancer (HR = 0.86; 95% CI, 0.74–0.99).
Subgroup analyses by sex and age at diagnosis are presented in Table 2. Here, the associations between very high antibiotics use and poorer survival outcomes were observed primarily in men (HR for OS = 1.64; 95% CI, 1.41–1.91, HR for cancer-specific survival = 1.66; 95% CI, 1.33–2.07). Analyses of early-onset colorectal cancer were limited by low numbers of events (113 and 57 cancer-specific deaths among users and nonusers of antibiotics, respectively). Although the results did not differ substantially from those of patients ≥50 years of age at diagnosis, HR magnitudes were greater for very high antibiotics use and an inverse risk relationship was observed for high but not moderate antibiotics use.
Antibiotics use in relation to overall and cancer-specific survival in stage I–III colorectal cancer by sex and age at diagnosis.
. | . | . | Overall survival . | Cancer-specific survivalb . | ||
---|---|---|---|---|---|---|
. | Prediagnostic antibiotics usea . | No. . | Events (%) . | Adjusted HRc (95% CI) . | Events (%) . | Adjusted HRc (95% CI) . |
No use | 5,229 | 1,442 (27.6) | 1.00 (Ref.) | 791 (15.1) | 1.00 (Ref.) | |
Any use | 13,699 | 3,560 (26.0) | 0.97 (0.91–1.03) | 1,740 (12.7) | 0.94 (0.86–1.03) | |
Men | Low use | 2,248 | 640 (28.5) | 0.99 (0.90–1.09) | 313 (13.9) | 0.96 (0.84–1.09) |
Moderate use | 8,301 | 2,085 (25.1) | 0.94 (0.88–1.01) | 1,036 (12.5) | 0.92 (0.83–1.01) | |
High use | 2,601 | 637 (24.5) | 0.94 (0.85–1.04) | 295 (11.3) | 0.93 (0.81–1.07) | |
Very high use | 549 | 198 (36.1) | 1.64 (1.41–1.91) | 96 (17.5) | 1.66 (1.33–2.07) | |
No use | 3,647 | 947 (26.0) | 1.00 (Ref.) | 566 (15.5) | 1.00 (Ref.) | |
Any use | 13,486 | 3,148 (23.3) | 1.07 (0.99–1.15) | 1,654 (12.3) | 0.95 (0.86–1.05) | |
Women | Low use | 2,070 | 548 (26.5) | 1.07 (0.96–1.19) | 282 (13.6) | 0.94 (0.82–1.09) |
Moderate use | 7,812 | 1,781 (22.8) | 1.02 (0.94–1.11) | 961 (12.3) | 0.93 (0.83–1.03) | |
High use | 3,001 | 680 (22.7) | 1.21 (1.09–1.35) | 348 (11.6) | 1.04 (0.90–1.20) | |
Very high use | 603 | 139 (23.1) | 1.20 (0.99–1.44) | 63 (10.4) | 0.97 (0.74–1.27) | |
No use | 540 | 61 (11.3) | 1.00 (Ref.) | 57 (10.6) | 1.00 (Ref.) | |
Any use | 1,213 | 122 (10.1) | 1.01 (0.73–1.41) | 113 (9.3) | 1.02 (0.72–1.43) | |
Age <50 years (Early onset) | Low use | 212 | 16 (7.5) | 0.68 (0.38–1.24) | 15 (7.1) | 0.68 (0.37–1.27) |
Moderate use | 746 | 87 (11.7) | 1.16 (0.82–1.64) | 80 (10.7) | 1.16 (0.80–1.66) | |
High use | 220 | 11 (5.0) | 0.51 (0.25–1.02) | 11 (5.0) | 0.55 (0.27–1.11) | |
Very high use | 35 | 8 (22.9) | 2.57 (1.18–5.63) | 7 (20.0) | 2.55 (1.11–5.85) | |
No use | 8,336 | 2,328 (27.9) | 1.00 (Ref.) | 1,300 (15.6) | 1.00 (Ref.) | |
Any use | 25,972 | 6,586 (25.4) | 1.01 (0.96–1.06) | 3,281 (12.6) | 0.94 (0.88–1.01) | |
Age ≥50 years | Low use | 4,106 | 1,172 (28.5) | 1.03 (0.96–1.10) | 580 (14.1) | 0.96 (0.87–1.06) |
Moderate use | 15,367 | 3,779 (24.6) | 0.97 (0.92–1.02) | 1,917 (12.5) | 0.91 (0.85–0.98) | |
High use | 5,382 | 1,306 (24.3) | 1.07 (1.00–1.15) | 632 (11.7) | 1.00 (0.90–1.10) | |
Very high use | 1,117 | 329 (29.5) | 1.39 (1.24–1.57) | 152 (13.6) | 1.27 (1.07–1.51) |
. | . | . | Overall survival . | Cancer-specific survivalb . | ||
---|---|---|---|---|---|---|
. | Prediagnostic antibiotics usea . | No. . | Events (%) . | Adjusted HRc (95% CI) . | Events (%) . | Adjusted HRc (95% CI) . |
No use | 5,229 | 1,442 (27.6) | 1.00 (Ref.) | 791 (15.1) | 1.00 (Ref.) | |
Any use | 13,699 | 3,560 (26.0) | 0.97 (0.91–1.03) | 1,740 (12.7) | 0.94 (0.86–1.03) | |
Men | Low use | 2,248 | 640 (28.5) | 0.99 (0.90–1.09) | 313 (13.9) | 0.96 (0.84–1.09) |
Moderate use | 8,301 | 2,085 (25.1) | 0.94 (0.88–1.01) | 1,036 (12.5) | 0.92 (0.83–1.01) | |
High use | 2,601 | 637 (24.5) | 0.94 (0.85–1.04) | 295 (11.3) | 0.93 (0.81–1.07) | |
Very high use | 549 | 198 (36.1) | 1.64 (1.41–1.91) | 96 (17.5) | 1.66 (1.33–2.07) | |
No use | 3,647 | 947 (26.0) | 1.00 (Ref.) | 566 (15.5) | 1.00 (Ref.) | |
Any use | 13,486 | 3,148 (23.3) | 1.07 (0.99–1.15) | 1,654 (12.3) | 0.95 (0.86–1.05) | |
Women | Low use | 2,070 | 548 (26.5) | 1.07 (0.96–1.19) | 282 (13.6) | 0.94 (0.82–1.09) |
Moderate use | 7,812 | 1,781 (22.8) | 1.02 (0.94–1.11) | 961 (12.3) | 0.93 (0.83–1.03) | |
High use | 3,001 | 680 (22.7) | 1.21 (1.09–1.35) | 348 (11.6) | 1.04 (0.90–1.20) | |
Very high use | 603 | 139 (23.1) | 1.20 (0.99–1.44) | 63 (10.4) | 0.97 (0.74–1.27) | |
No use | 540 | 61 (11.3) | 1.00 (Ref.) | 57 (10.6) | 1.00 (Ref.) | |
Any use | 1,213 | 122 (10.1) | 1.01 (0.73–1.41) | 113 (9.3) | 1.02 (0.72–1.43) | |
Age <50 years (Early onset) | Low use | 212 | 16 (7.5) | 0.68 (0.38–1.24) | 15 (7.1) | 0.68 (0.37–1.27) |
Moderate use | 746 | 87 (11.7) | 1.16 (0.82–1.64) | 80 (10.7) | 1.16 (0.80–1.66) | |
High use | 220 | 11 (5.0) | 0.51 (0.25–1.02) | 11 (5.0) | 0.55 (0.27–1.11) | |
Very high use | 35 | 8 (22.9) | 2.57 (1.18–5.63) | 7 (20.0) | 2.55 (1.11–5.85) | |
No use | 8,336 | 2,328 (27.9) | 1.00 (Ref.) | 1,300 (15.6) | 1.00 (Ref.) | |
Any use | 25,972 | 6,586 (25.4) | 1.01 (0.96–1.06) | 3,281 (12.6) | 0.94 (0.88–1.01) | |
Age ≥50 years | Low use | 4,106 | 1,172 (28.5) | 1.03 (0.96–1.10) | 580 (14.1) | 0.96 (0.87–1.06) |
Moderate use | 15,367 | 3,779 (24.6) | 0.97 (0.92–1.02) | 1,917 (12.5) | 0.91 (0.85–0.98) | |
High use | 5,382 | 1,306 (24.3) | 1.07 (1.00–1.15) | 632 (11.7) | 1.00 (0.90–1.10) | |
Very high use | 1,117 | 329 (29.5) | 1.39 (1.24–1.57) | 152 (13.6) | 1.27 (1.07–1.51) |
aPrediagnostic antibiotics use was categorized as no use, any use, low use (1–10 days), moderate use (11–60 days), high use (61–180 days) and very high use (>180 days), based on defined daily doses. Antibiotics prescribed during one year before colorectal cancer diagnosis were excluded to account for potential use due to undiagnosed but clinically manifest colorectal cancer.
bDeath due to colorectal cancer.
cHazard ratios (HR) and 95% confidence intervals (CI) were estimated from the Cox proportional hazards models stratified by sex (men/women), age at diagnosis (</≥50 years), prediagnostic Charlson comorbidity index (no/any comorbidity), tumor site (proximal colon/distal colon/rectum/unspecified colorectum), tumor stage (stage I–III), family history of colorectal cancer (0/1/≥2 first-degree relatives), and adjusted for: age and calendar year of diagnosis, and prediagnostic Charlson comorbidity index as continuous variables, and sex (for subgroup analyses by age at diagnosis), place of birth, county, marital status, level of education and income status as categorical variables, defined as presented in Table 1.
Analyses by antibiotics types, stage I–III colorectal cancer
In analyses of any versus no use of antibiotics classes, several associations with worse OS were observed, which were strongest for the broad-spectrum beta lactam class and the macrolides and/or lincosamides class, as well as for the grouping of antibiotics with effects on anaerobic bacteria (Supplementary Fig. S5). For cancer-specific survival, the risk relationship remained for macrolides and/or lincosamides, although it was not significant at the more stringent 99.5% confidence level. No antibiotics classes were clearly associated with better cancer-specific survival. In an attempt to gain further insight into the risk relationships observed in the main analyses we, therefore, conducted post hoc analyses based on typical indications for use of various antibiotics drugs in Sweden as well as specific commonly prescribed antibiotics drugs (Supplementary Fig. S5). For most indications, antibiotics use was associated with worse OS, and the results for specific antibiotics drugs generally followed the patterns of the antibiotics classes to which they belonged. However, better OS was observed for antibiotics used for respiratory and some skin and soft tissue infections, and especially for phenoxymethylpenicillin in specific. These inverse associations held also for cancer-specific survival, although with wider CIs (Supplementary Fig. S5). Excluding very high users had no material effects on the results for antibiotics classes, typical indications and specific antibiotics drugs.
Sensitivity analyses, stage I–III
Using fixed exposure time windows of 4.5 years and 10 years prior to diagnosis generally yielded similar distributions of study characteristics, except for the exposure status (Supplementary Table S3). Analyses using fixed exposure time windows and time-standardized prediagnostic exposures showed broadly similar patterns of association with survival as in the main analyses, though with some attenuated HRs and loss of statistical significance (Table 3). Results from other sensitivity analyses were more or less similar as in the main analyses (Supplementary Fig. S6).
Antibiotics use in relation to overall and cancer-specific survival in stage I–III colorectal cancer using fixed exposure time windows and time-standardized exposure.
. | . | . | Overall survival . | Cancer-specific survivalb . | ||
---|---|---|---|---|---|---|
. | Prediagnostic antibiotics usea . | No. . | Events (%) . | Adjusted HRc (95% CI) . | Events (%) . | Adjusted HRc (95% CI) . |
No use | 14,188 | 3,199 (22.5) | 1.00 (Ref.) | 1,816 (12.8) | 1.00 (Ref.) | |
Any use | 21,873 | 5,898 (27.0) | 1.04 (0.99–1.09) | 2,935 (13.4) | 1.00 (0.94–1.06) | |
4.5-year exposure periodd | Low use | 4,673 | 1,255 (26.9) | 1.03 (0.97–1.10) | 656 (14.0) | 1.01 (0.92–1.10) |
Moderate use | 14,231 | 3,643 (25.6) | 1.00 (0.96–1.05) | 1,837 (12.9) | 0.97 (0.91–1.04) | |
High use | 2,501 | 802 (32.1) | 1.18 (1.09–1.28) | 361 (14.4) | 1.09 (0.97–1.22) | |
Very high use | 468 | 198 (42.3) | 1.63 (1.41–1.89) | 81 (17.3) | 1.47 (1.17–1.84) | |
No use | 3,561 | 455 (12.8) | 1.00 (Ref.) | 291 (8.2) | 1.00 (Ref.) | |
Any use | 14,133 | 2,051 (14.5) | 0.97 (0.87–1.08) | 1,171 (8.3) | 0.96 (0.84–1.09) | |
10-year exposure periode | Low use | 1,798 | 250 (13.9) | 1.00 (0.86–1.17) | 160 (8.9) | 1.07 (0.88–1.30) |
Moderate use | 8,505 | 1,145 (13.5) | 0.92 (0.83–1.03) | 680 (8.0) | 0.92 (0.80–1.06) | |
High use | 3,142 | 493 (15.7) | 1.01 (0.88–1.15) | 256 (8.1) | 0.95 (0.80–1.13) | |
Very high use | 688 | 163 (23.7) | 1.42 (1.18–1.71) | 75 (10.9) | 1.24 (0.95–1.61) | |
No use | 8,876 | 2,389 (26.9) | 1.00 (Ref.) | 1,357 (15.3) | 1.00 (Ref.) | |
Any use | 27,185 | 6,708 (24.7) | 1.01 (0.96–1.06) | 3,394 (12.5) | 0.95 (0.88–1.01) | |
Time-standardized exposuref | 1–3 days per year | 13,187 | 2,894 (22.0) | 0.97 (0.92–1.03) | 1,527 (11.6) | 0.91 (0.84–0.98) |
4–7 days per year | 7,849 | 1,895 (24.1) | 0.96 (0.91–1.02) | 980 (12.5) | 0.92 (0.85–1.00) | |
8–14 days per year | 3,961 | 1,139 (28.8) | 1.03 (0.96–1.11) | 554 (14.0) | 0.99 (0.90–1.10) | |
≥15 days per year | 2,188 | 780 (35.7) | 1.21 (1.12–1.32) | 333 (15.2) | 1.07 (0.94–1.21) |
. | . | . | Overall survival . | Cancer-specific survivalb . | ||
---|---|---|---|---|---|---|
. | Prediagnostic antibiotics usea . | No. . | Events (%) . | Adjusted HRc (95% CI) . | Events (%) . | Adjusted HRc (95% CI) . |
No use | 14,188 | 3,199 (22.5) | 1.00 (Ref.) | 1,816 (12.8) | 1.00 (Ref.) | |
Any use | 21,873 | 5,898 (27.0) | 1.04 (0.99–1.09) | 2,935 (13.4) | 1.00 (0.94–1.06) | |
4.5-year exposure periodd | Low use | 4,673 | 1,255 (26.9) | 1.03 (0.97–1.10) | 656 (14.0) | 1.01 (0.92–1.10) |
Moderate use | 14,231 | 3,643 (25.6) | 1.00 (0.96–1.05) | 1,837 (12.9) | 0.97 (0.91–1.04) | |
High use | 2,501 | 802 (32.1) | 1.18 (1.09–1.28) | 361 (14.4) | 1.09 (0.97–1.22) | |
Very high use | 468 | 198 (42.3) | 1.63 (1.41–1.89) | 81 (17.3) | 1.47 (1.17–1.84) | |
No use | 3,561 | 455 (12.8) | 1.00 (Ref.) | 291 (8.2) | 1.00 (Ref.) | |
Any use | 14,133 | 2,051 (14.5) | 0.97 (0.87–1.08) | 1,171 (8.3) | 0.96 (0.84–1.09) | |
10-year exposure periode | Low use | 1,798 | 250 (13.9) | 1.00 (0.86–1.17) | 160 (8.9) | 1.07 (0.88–1.30) |
Moderate use | 8,505 | 1,145 (13.5) | 0.92 (0.83–1.03) | 680 (8.0) | 0.92 (0.80–1.06) | |
High use | 3,142 | 493 (15.7) | 1.01 (0.88–1.15) | 256 (8.1) | 0.95 (0.80–1.13) | |
Very high use | 688 | 163 (23.7) | 1.42 (1.18–1.71) | 75 (10.9) | 1.24 (0.95–1.61) | |
No use | 8,876 | 2,389 (26.9) | 1.00 (Ref.) | 1,357 (15.3) | 1.00 (Ref.) | |
Any use | 27,185 | 6,708 (24.7) | 1.01 (0.96–1.06) | 3,394 (12.5) | 0.95 (0.88–1.01) | |
Time-standardized exposuref | 1–3 days per year | 13,187 | 2,894 (22.0) | 0.97 (0.92–1.03) | 1,527 (11.6) | 0.91 (0.84–0.98) |
4–7 days per year | 7,849 | 1,895 (24.1) | 0.96 (0.91–1.02) | 980 (12.5) | 0.92 (0.85–1.00) | |
8–14 days per year | 3,961 | 1,139 (28.8) | 1.03 (0.96–1.11) | 554 (14.0) | 0.99 (0.90–1.10) | |
≥15 days per year | 2,188 | 780 (35.7) | 1.21 (1.12–1.32) | 333 (15.2) | 1.07 (0.94–1.21) |
aPrediagnostic antibiotics use was categorized as no use, any use, low use (1–10 days), moderate use (11–60 days), high use (61–180 days) and very high use (>180 days), based on defined daily doses.
bDeath due to colorectal cancer.
cHazard ratios (HR) and 95% confidence intervals (CI) were estimated from the Cox proportional hazards models stratified by sex (men/women), age at diagnosis (</≥50 years), prediagnostic Charlson comorbidity index (no/any comorbidity), tumor site (proximal colon/distal colon/rectum/unspecified colorectum), tumor stage (stage I–III), family history of colorectal cancer (0/1/≥2 first-degree relatives), and adjusted for: age and calendar year of diagnosis and prediagnostic Charlson comorbidity index as continuous variables, and sex (for subgroup analyses by age at diagnosis), place of birth, county, marital status, level of education, and income status as categorical variables, defined as presented in Table 1.
dAnalyses using a fixed exposure period of 4.5 years prior to colorectal cancer diagnosis in the full study population.
eAnalyses using a fixed exposure period of 10 years prior to colorectal cancer diagnosis for cases diagnosed after July 1, 2015.
fAnalyses using time-standardized prediagnostic antibiotics variable (days of antibiotics use per year).
Stage IV colorectal cancer
Characteristics of the 11,242 patients with stage IV colorectal cancer are presented in the Supplementary Table S4. Prediagnostic antibiotics use (70%), as well as apparent differences in characteristics between users and nonusers of antibiotics during the study period, were similar to those observed for stage I–III colorectal cancer in Table 1.
In the survival analyses for stage IV colorectal cancer (Fig. 3), any use compared with no use of prescription antibiotics was associated with modestly better OS (HR = 0.94; 95% CI, 0.89–0.98) and cancer-specific survival (HR = 0.93; 95% CI, 0.89–0.98). These inverse relationships were due primarily to the moderate and high use categories. Similar findings were observed in distal colon and rectal cancer, whereas associations were close to null for proximal colon cancer. There were no material sex differences in associations (Supplementary Table S5), and apparent similar patterns of association in the sensitivity analyses (Supplementary Fig. S6). Several antibiotics classes, groups of antibiotics based on typical indications in Sweden and specific common antibiotics drugs showed associations with better overall and/or cancer-specific survival in stage IV colorectal cancer, and none were clearly associated with worse survival (Supplementary Fig. S7). Further post hoc analysis, cross tabulating antibiotics use by stage I to IV colorectal cancer, showed a trend of higher usage in earlier stages, particularly stage I and II, and lower usage in stage IV (χ2P < 0.001; Supplementary Table S6).
Antibiotics use in relation to overall and cancer-specific survival among stage IV colorectal cancer patients. Prediagnostic antibiotics use was categorized as no use, any use, low use (1–10 days), moderate use (11–60 days), high use (61–180 days) and very high use (>180 days), on the basis of defined daily doses. Antibiotics use during one year before colorectal cancer diagnosis was excluded to account for potential use due to undiagnosed but clinically manifest colorectal cancer. Hazard ratios (HR) and 95% confidence intervals (CI) were estimated from the Cox proportional hazards models, stratified by tumor site (proximal colon/distal colon/rectum/unspecified colorectum categories in the analyses for total colorectal cancer), sex(men/women), age at diagnosis (</ ≥50 years), prediagnostic Charlson comorbidity index (no/any comorbidity), income status (1st/2nd/3rd/4th quartile), and adjusted for: age and calendar year of diagnosis and prediagnostic Charlson comorbidity index as continuous variables, place of birth, county, marital status, level of education, family history of colorectal cancer, calendar year at diagnosis, palliative chemotherapy and surgical therapy as categorical variables, defined as presented in Supplementary Table S4. Given the relatively high proportion of missing data for oncologic therapies, we included a separate category for missing values in the multivariable model.
Antibiotics use in relation to overall and cancer-specific survival among stage IV colorectal cancer patients. Prediagnostic antibiotics use was categorized as no use, any use, low use (1–10 days), moderate use (11–60 days), high use (61–180 days) and very high use (>180 days), on the basis of defined daily doses. Antibiotics use during one year before colorectal cancer diagnosis was excluded to account for potential use due to undiagnosed but clinically manifest colorectal cancer. Hazard ratios (HR) and 95% confidence intervals (CI) were estimated from the Cox proportional hazards models, stratified by tumor site (proximal colon/distal colon/rectum/unspecified colorectum categories in the analyses for total colorectal cancer), sex(men/women), age at diagnosis (</ ≥50 years), prediagnostic Charlson comorbidity index (no/any comorbidity), income status (1st/2nd/3rd/4th quartile), and adjusted for: age and calendar year of diagnosis and prediagnostic Charlson comorbidity index as continuous variables, place of birth, county, marital status, level of education, family history of colorectal cancer, calendar year at diagnosis, palliative chemotherapy and surgical therapy as categorical variables, defined as presented in Supplementary Table S4. Given the relatively high proportion of missing data for oncologic therapies, we included a separate category for missing values in the multivariable model.
Discussion
In this nationwide cohort of 49,249 colorectal cancer cases, no clear links were found between prediagnostic prescription antibiotics use and poorer survival outcomes, except for very high use (> 180 days over the study period) in stage I–III patients, particularly in men. In contrast, moderate antibiotics use (11–60 days) was associated with modestly better cancer-specific survival in stage I–III colorectal cancer. For antibiotics classes, groups of antibiotics based on typical indications in Sweden, and specific common antibiotics drugs, poorer survival outcomes or null associations were most common. In patients with stage IV colorectal cancer, several inverse associations were observed between prediagnostic antibiotics use and better overall and/or cancer-specific survival.
Overall, our findings do not fully support the hypothesis that use of antibiotics prior to colorectal cancer diagnosis negatively impacts prognosis. Potential adverse effects of antibiotics use have been reported for survival outcomes in both metastatic colorectal cancer (15) and other types of cancer (32), although not consistently (16, 33). Previous studies used binary yes/no exposure variables for antibiotics use, which may explain some of the discrepancies in results. They also assessed antibiotics use after diagnosis, which may in part reflect infections due to cancer-related factors such as disease burden, therapy or malnutrition. The use of detailed prediagnostic exposure data in the present study, excluding antibiotics use during the year prior to diagnosis, eliminated these issues. Our findings, therefore, highlight the possibility of differing effects of antibiotics depending on the timing of exposure in relation to colorectal cancer diagnosis.
Of particular interest in clinical oncology is a putative role for antibiotics in reducing the effect of cancer immunotherapy, which has been observed in multiple studies (12, 13). Because very few patients in our study would have been eligible for immunotherapy (currently only a treatment option in patients with metastatic disease demonstrating microsatellite instability, or defect mismatch repair (34)), our results do not contradict previous reports.
The inverse associations between antibiotics use and survival outcomes in some of our analyses, as well as in some previous reports (16, 33), might be explained by detection bias or health care–seeking behavior. Antibiotics users may represent a group of patients under more health care surveillance, or who more actively seek and receive health care (including cancer screening). This could, potentially, result in an earlier colorectal cancer diagnosis and better survival outcomes (35). Indeed, in our study, patients who had used antibiotics were diagnosed at lower stages. However, adjusting for CCI, stage at diagnosis and socioeconomic factors accounted for this confounding to some extent. Furthermore, observations of better survival among patients who had used any versus no use antibiotics were most apparent in stage IV patients. In this group, any residual confounding due to a high burden of preexisting medical conditions or frailty would be expected to bias toward shorter rather than longer survival, both directly and through more conservative oncologic therapy. Finally, although health care–seeking behavior and use of prescription antibiotics are both more common in women than men (36, 37), the inverse associations between antibiotics use and survival outcomes in our study were generally similar in both sexes.
An effect of antibiotics use in improving survival in colorectal cancer may have biological plausibility. For example, some antibiotics, such as tetracycline and fluoroquinolone have reported antineoplastic effects (38, 39). However, the relatively short-term antibiotics usage associated with better survival in our analyses seems unlikely to slow tumor progression enough to influence prognosis. Furthermore, in contrast to the results for overall antibiotics use, several antibiotics classes, antibiotics grouped by typical indications in Sweden and specific commonly prescribed antibiotics drugs were associated with worse survival outcomes in stage I–III colorectal cancer. Better overall and cancer-specific survival in stage I–III disease was seen only for antibiotics typically used for respiratory and some skin and soft-tissue infections, and for phenoxymethylpenicillin in specific, whereas among the stage IV patients, inverse associations were observed for several types of antibiotics.
Dysbiosis of the gut microbiome could have adverse effects on long-term oncologic outcomes after colorectal cancer surgery (40, 41). For instance, colonization by collagenase-producing microbes may induce collagen degradation in the gut tissue contributing to anastomotic leakage and subsequent infection (42), which might in turn influence long-term survival among stage I–III patients (43). Supporting this hypothesis, we observed a robust association between use of macrolides and/or lincosamides, and particularly clindamycin (typically for complicated skin and soft tissue and deep-seated infections), and worse cancer-specific survival in stage I–III colorectal cancer. This class of antibiotics has long-term negative effects on total bacterial diversity in the gut (6). However, in general, the microbiome hypothesis for a detrimental role of antibiotics use is weaker for survival compared with incidence of colorectal cancer, which is reflected in our results. The positive associations for methenamine hippurate (a urinary antiseptic with no known microbiome effects), although not statistically significant at the stricter 99.5% confidence level, suggest the involvement of other mechanisms. Despite the greater fermentation and higher microbial products and biofilm formation in the proximal colon compared with the rest of the colorectum (44–46), we observed no distinct tumor site–specific associations between antibiotics use and colorectal cancer survival. In addition, whereas the anaerobic bacteria such as Fusobacteria and Bacteroides have been implicated in colorectal carcinogenesis (47–49), antibiotics with antianaerobic effects were not associated with survival in stage I–III colorectal cancer in our study, and similar inverse associations were observed for both antianaerobic and antiaerobic agents in stage IV patients.
Our study has several strengths. In addition to the very large sample size, it is the first study to address prediagnostic antibiotics use and survival in patients with colorectal cancer. Using high-quality Swedish national registries, including the complete and reliable Swedish Cause of Death Register (18), we were able to calculate cancer-specific survival outcomes in a nationwide cohort. We were also able to define a relatively homogeneous population of stage I–III patients treated with a curative intent for the main analyses, in addition to conducting analyses on stage IV patients separately. Within the stage IV group, patient-related factors such as preexisting medical conditions and frailty weigh heavily in therapeutic decision making and prognosis, entailing a greater risk of residual confounding compared to stage I–III. Limiting the main analyses to stage I–III helped account for this but also may have attenuated any potential mediating effects by stage, that is, effects of antibiotics on tumor progression. Investigating the use of antibiotics prior to colorectal cancer diagnosis avoided bias due to infections caused by cancer-related factors, such as disease burden, cancer therapy or malnutrition, as well as selection bias, which can occur when the exposure classification depends on the survival follow-up time (50).
Use of the Prescribed Drug Register provided essentially complete, detailed national data on dispensed antibiotics prescriptions. Given the generally high compliance to antibiotics drugs in Sweden (51), these data are probably also highly reflective of actual use. However, we were not able to include antibiotics prescribed before 2005, when the register was established, or antibiotics administered during inpatient care. In Sweden, the majority of antibiotics use is prescription derived (88.6%) in both specialist and primary health care (25), but differences in the types of antibiotics use in hospital compared with outpatient care might still have influenced our results.
Our study also included several important covariates including the CCI adapted for the Swedish register research setting (26). We acknowledge that use of the national patient register to calculate CCI cannot fully capture comorbidity, as the register does not include diagnoses made in primary health care. However, it allowed us to account for all comorbidity requiring at least one visit in secondary care. Bias could also be introduced due to unmeasured cofounding, such as lifestyle-related factors including physical activity and smoking (52), though socioeconomic factors were considered as a proxy, at least in part. We did not directly adjust for participation in colorectal cancer screening programs, which might affect results (53, 54). However, full-scale national screening had not been implemented during the study period, and we incorporated adjustments for county of residence prior to or at diagnosis and calendar year of diagnosis in the statistical model to help account for the potential effects of local screening programs in the Stockholm-Gotland healthcare region, which was implemented in 2008 (54). It could be argued that better surveillance and consequent earlier detection of colorectal cancer in hereditary syndromes such as Lynch syndrome and familial adenomatous polyposis, could impact survival outcomes. However, although we lacked data on hereditary syndromes, they constitute less than 5% of the total colorectal cancer population (55, 56), and we included family history of colorectal cancer in the main model to help account for this issue. We were also able to assess different classes of antibiotics and antibiotics grouped according to typical indications for prescribing in Sweden during the study period. However, antibiotics may be prescribed for other indications, and indications for antibiotics vary around the world. Generalizability of these results may, therefore, be limited to areas/regions/countries with similar antibiotics prescription patterns.
We performed sensitivity analyses to assess the robustness of the associations. For instance, we tested excluding deaths within 90 days after surgery to account for fatal perioperative complications and used fixed exposure time windows prior to diagnosis as well as time-standardized antibiotics use (i.e., days of antibiotics use per observation year) to consider possible influence of exposure variability with time, without material effects on the main findings.
In conclusion, the results of this Swedish nation-wide cohort study do not support any substantial detrimental effects of prediagnostic prescription antibiotics use on cancer-specific survival after colorectal cancer diagnosis, with the exception of very high use in stage I–III patients. Although any causal interpretation should be made with care due to the possibility of residual confounding, these findings are intriguing and warrant further investigation.
Authors' Disclosures
Å. Gylfe reports other support from Eurocine Vaccines outside the submitted work. S. Harlid reports grants from Cancerforskningsfonden Norrland during the conduct of the study. B. Van Guelpen reports grants from Region Västerbotten and grants from the Knut and Alice Wallenberg Foundation during the conduct of the study. No disclosures were reported by the other authors.
Disclaimer
The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Authors' Contributions
S.S.M. Lu: Conceptualization, data curation, formal analysis, investigation, visualization, methodology, writing–original draft. M. Rutegård: Conceptualization, supervision, investigation, methodology, writing–review and editing. M. Ahmed: Formal analysis, investigation, writing–review and editing. C. Häggström: Supervision, investigation, methodology, writing–review and editing. Å. Gylfe: Supervision, investigation, writing–review and editing. S. Harlid: Conceptualization, data curation, formal analysis, supervision, investigation, methodology, project administration, writing–review and editing. B. Van Guelpen: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, investigation, methodology, project administration, writing–review and editing.
Acknowledgments
S. Harlid received grants from the Lions Cancer Research Foundation/Cancerforskningsfonden Norrland, Umeå University (grant nos. LP 17–2154 and LP 21–2275), and B. Van Guelpen received grants from the Region Västerbotten (grant no. RV-932777) and the Knut and Alice Wallenberg Foundation.
The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.
Note: Supplementary data for this article are available at Cancer Epidemiology, Biomarkers & Prevention Online (http://cebp.aacrjournals.org/).