Antibiotics and healthcare facility-associated Clostridioides difficile infection: systematic review and meta-analysis 2020 update

Abstract

Background

Antibiotic use is the most important modifiable risk factor for healthcare facility-associated Clostridioides difficile infection (HCFA-CDI). Previous systematic reviews cover studies published until 31 December 2012.

Objectives

To update the evidence for associations between antibiotic classes and HCFA-CDI to 31 December 2020.

Methods

PubMed, Scopus, Web of Science Core Collection, WorldCat and Proquest Dissertations & Theses were searched for studies published since 1 January 2013. Eligible studies were those conducted among adult hospital inpatients, measured exposure to individual antibiotics or antibiotic classes, included a comparison group and measured the occurrence of HCFA-CDI as an outcome. The Newcastle–Ottawa Scale was used to appraise study quality. To assess the association between each antibiotic class and HCFA-CDI, a pooled random-effects meta-analysis was undertaken. Meta-regression and subgroup analysis was used to investigate study characteristics identified a priori as potential sources of heterogeneity.

Results

Carbapenems and third- and fourth-generation cephalosporin antibiotics remain the most strongly associated with HCFA-CDI, with cases more than twice as likely to have recent exposure to these antibiotics prior to developing HCFA-CDI. Modest associations were observed for fluoroquinolones, clindamycin and β-lactamase inhibitor combination penicillin antibiotics. Individual study effect sizes were variable and heterogeneity was observed for most antibiotic classes.

Conclusions

This review provides the most up-to-date synthesis of evidence in relation to the risk of HCFA-CDI associated with exposure to specific antibiotic classes. Studies were predominantly conducted in North America or Europe and more studies outside of these settings are needed.

Introduction

Clostridioides difficile infection (CDI) is a leading cause of healthcare facility (HCF)-associated (HCFA) infections and the most common cause of HCFA infectious diarrhoea in high-income countries.^1,C. difficile is a toxin-producing, anaerobic spore-forming bacterium that is transmitted via the faecal–oral route.² Symptoms range from mild diarrhoea to life-threatening conditions such as pseudomembranous colitis and toxic megacolon, and recurrence occurs among approximately 20% of cases following an initial episode.² CDI is associated with a protracted length of stay incurring substantial direct and indirect healthcare utilization costs.^3–5 Incidence rates of HCFA-CDI are geographically and temporally variable but have generally increased in the past 20 years.^1,6,7 Explanations for the increasing occurrence include increased testing and more sensitive diagnostic tests, increased use of broad-spectrum antibiotics, inadequate prevention measures, ageing populations and emergence of community strains.⁷

Antibiotic use is the most important modifiable risk factor for CDI. Antibiotic exposure alters the natural flora of the intestines allowing C. difficile to proliferate. Other important risk factors include advanced age, increased number of comorbidities or severe underlying disease, and duration of healthcare exposure.^8,9 A longer duration of hospitalization is correlated with advanced age and severity of underlying illness, and increases the probability of C. difficile acquisition from the environment and exposure to antibiotics. Although almost all antibiotics have been associated with CDI, clindamycin, cephalosporins, carbapenems and fluoroquinolones are the most frequently associated with this infection.^10,11

Antibiotic stewardship reduces unnecessary antibiotic use in hospitals¹² and is crucial to controlling CDI in HCFs alongside surveillance, isolation precautions, hand hygiene and environmental cleaning.^8,13 Limiting antibiotic use effectively reduces the incidence of CDI in HCFs; ¹⁴ however, CDI remains a significant problem and has been named by the US CDC as an urgent threat to public health.¹⁵

It is therefore important to monitor changes in risk for HCFA-CDI associated with changes in antibiotic usage and changing C. difficile epidemiology. The objective of this systematic review was to update the evidence for epidemiological associations between specific antibiotic classes and HCFA-CDI. This review updates earlier reviews that covered studies conducted up to 2012,^10,11 to include studies undertaken between 2013 and 2020.

Methods

The study has been conducted in accordance with the PRISMA statement.¹⁶ The objectives, inclusion criteria and methods for analysis were specified in advance and registered with PROSPERO (CRD42020181817).

Eligibility criteria

Studies were eligible for inclusion in the review if they were conducted among hospital inpatients, measured any exposure to individual antibiotics or antibiotic classes prior to development of the outcome, included a comparison group and measured the occurrence of HCFA-CDI as an outcome. Study designs eligible for inclusion consisted of case–control studies, cohort studies, analytical cross-sectional studies and randomized controlled trials (if they reported the risk associated with the exposure).

Studies were excluded if they investigated risk factors for severe disease, relapse or recurrence, asymptomatic colonization, or were specific studies of paediatric populations. Studies that did not report exposure to specific antibiotics or antibiotic classes were also excluded, as were studies examining community-associated (CA) infection or that did not adequately exclude CA infection. Case reports, case series and descriptive cross-sectional studies (i.e. those without comparison groups) were excluded. The review was limited to English-language publications; however, non-English articles were included in searches and their abstracts and full texts were assessed for eligibility.

Information sources and searching

PubMed, Scopus and Web of Science Core Collection were searched on 21 April 2020 for studies published since 1 January 2013. WorldCat (www.worldcat.org) and Proquest Dissertations & Theses (PQDT; www.proquest.com) were used to search for dissertations and theses. Final follow-up searches to 31 December 2020 were conducted on 11 February 2021.

The search strategy used was consistent with the previous review.¹⁰ The full search strategy for each database is presented in the Supplementary data, available at JAC Online. Search results were exported to Endnote X9.1, which was used to identify and remove duplicates prior to importing to Covidence (www.covidence.org) for title and abstract screening.

Study selection

Titles and abstracts were screened to eliminate irrelevant studies. Full-text articles were then inspected for eligibility. Studies were classified as confirmed HCFA-CDI if they included a clear definition of HCFA acquisition of CDI, or probable HCFA-CDI on the basis of information provided in the report, e.g. evidence that the minimum length of stay until onset of symptoms was >48 h. Studies that had been excluded from the previous review because they did not use an explicit definition of HCFA acquisition were re-evaluated and subsequently included in this review if there was sufficient evidence that HCFA-CDI was probable.

Data collection

Data were extracted from each study using a pro forma template that included the study characteristics (citation, country, setting, study period based on earliest timepoint, study population, study design, HCFA-CDI case and non-case definitions, antibiotic name, exposure period and timing, comparison group), number of subjects in exposure categories for cases and non-cases, effect estimates (OR, relative risk) and 95% CIs.

Antibiotic exposures were categorized into their main classes with additional subgroup categorization for cephalosporins (first, second, third or fourth generation) and penicillins [ampicillin-like drugs (aminopenicillins); β-lactamase inhibitor combinations; broad-spectrum (antipseudomonal) penicillins; or penicillin G-like drugs (natural penicillins)] using the MSD Manual.¹⁷ The unexposed comparator groups were categorized as: no antibiotics; unexposed to antibiotic of interest; or a reference antibiotic. The exposure period was classified in two ways. First, whether antibiotic administration before CDI diagnosis was recorded by the study and second, the period of time over which antibiotic administration was measured and categorized as: during index admission; up to 1 month prior to CDI diagnosis; up to 1–2 months prior; up to 2–3 months prior; preoperative prophylaxis; during admission—prior to ICU; or during ICU admission.

The non-CDI case groups were recorded as asymptomatic or symptomatic (patients had diarrhoea but tested negative for C. difficile). Study setting was classified according to geographical region (North America, Latin America, Europe, East Asia & Pacific) and patient population [general adult inpatients and specific clinical subgroups (HCFA-pneumonia; HCFA-diarrhoea; antibiotic treatment; ICU patients; COPD patients; C. difficile-colonized patients; haematology/oncology patients; surgical—gastrointestinal; surgical—non-gastrointestinal; or type 2 diabetes)].

Assessment of methodological quality

The Newcastle–Ottawa Scale for the assessment of quality (NOS) was used to assess study quality.¹⁸ The NOS comprises nine items across the domains of selection, comparability and exposure (case–control studies) or outcome (cohort studies). Each accomplished item receives 1 point and studies are classified as high quality (score 7–9), moderate quality (score 4–6) or poor quality (score 0–3). Important confounding domains considered in the study for assessment of comparability were identified from previous studies and included age, comorbidities, severity of underlying disease (1 point) and HCF exposure or other treatment (1 point).⁹ Poor quality studies were not excluded from the review; instead, overall study quality and level of comparability were explored as potential sources of heterogeneity.

Analysis

Data were entered into Excel for coding and exported to Stata version 14 for analysis. A quantitative descriptive analysis summarizing the characteristics of included studies was undertaken, presented as counts and percentages.

To assess the association between each antibiotic class and HCFA-CDI, a pooled random-effects analysis using the DerSimonian–Laird inverse variance approach was used.^19,20 The meta-analysis was restricted to studies that used a comparison group that was unexposed to the antibiotic of interest, as this was the comparison used in most studies. As most studies reported ORs, the most fully adjusted OR was used for synthesis. Where studies reported results for multiple antibiotics within each class, without providing the relevant overall association for that class, effect estimates were combined by taking the weighted average of the log ORs, with inverse variance weights. Heterogeneity of pooled effects was assessed using the I² statistic, which describes the percentage of variability in the effect estimates that is due to heterogeneity rather than chance, and its chi-squared statistic for evidence of heterogeneity (P < 0.10).

A subgroup analysis was used to assess study characteristics identified a priori based on previous findings as sources of heterogeneity. Specific characteristics investigated included setting (geographical region; time period when the study was conducted; and patient population), exposure measurement (time period prior to CDI diagnosis; and measurement after onset) and methodology (study design; HCFA definition; non-case comparison group; study quality; and confounder adjustment). Only antibiotic classes with data from at least 10 studies were included in the subgroup analysis. The subgroup pooled associations and their corresponding I² statistics and tau-squared statistics (for between-study variance) were examined. The overall associations between study characteristics and between study variation were investigated using random-effects meta-regression; variables associated with between-study variation at P ≤ 0.10 in a bivariate analysis were included in multivariable models.

Funnel plot analysis was used to assess bias due to missing results (publication bias).²¹

Results

The PRISMA diagram is presented in Figure 1. Of 804 titles and abstracts of non-duplicate articles screened, 197 full-text articles were assessed for eligibility and 23 were included in the review, along with 16 studies from the previous review covering studies until the end of 2012 (total of 39). Of the 23 studies identified from the latest searches, only 11 studies reported a clear definition of HCF acquisition, with a further 12 deemed probable HCFA-CDI based on information available in the article. Two papers previously excluded from the earlier review (consisting of 14 studies) were identified as probable HCFA-CDI and subsequently added to this update.

Figure 1.

PRISMA flow diagram of study selection. This figure appears in colour in the online version of JAC and in black and white in the print version of JAC.

Four studies of patients undergoing solid organ transplants^22–25 and five stem cell transplant studies^26–30 were excluded because of the long follow-up times after discharge precluding establishment of HCF acquisition. For example, in the study by Cusini et al.,²² kidney transplant patients were followed for 2 years, of which only 56% of CDI cases were deemed nosocomial.

Description of studies

The characteristics of the included studies identified are summarized in Table 1. Full details of each study identified are provided in Table S1. Most studies were either cohort or case–control studies, where cohort studies were the predominant design in articles published since 2013, in contrast to case–control studies being predominant in earlier articles. Compared with the earlier review, a more diverse range of patient groups were investigated in the recent studies and from a wider range of geographical regions. Of the 18 cohort studies, 16 were clinical cohort studies, examining specific inpatient subgroups. Most studies at both timepoints used patients not diagnosed with CDI as the non-case group, with 23% using symptomatic non-cases. Most studies (87%) used non-exposure to the antibiotic of interest as the unexposed comparator.

Table 1.

Characteristics of studies examining associations between antibiotics and HCFA-CDI included in the 2020 systematic review update

	Number of studies (%)
	2013 reviewa	2020 update	Total
Number of studies	16	23	39
Region
North America	11 (68.7)	9 (39.1)	20 (51.3)
Latin America	0 (0.00)	3 (13.0)	3 (7.7)
Europe	4 (25.0)	7 (30.4)	11 (28.2)
East Asia & Pacific	1 (6.2)	4 (17.4)	5 (12.8)
Study population
All inpatients	11 (68.7)	6 (26.1)	17 (43.6)
HCFA-pneumonia	1 (6.2)	2 (8.7)	3 (7.7)
HCFA-diarrhoea	0 (0.0)	2 (8.7)	2 (5.1)
Antibiotic treated	4 (25.0)	4 (17.4)	8 (20.5)
ICU patients	0 (0.0)	2 (8.7)	2 (5.1)
COPD inpatients	0 (0.0)	1 (4.3)	1 (2.6)
C. difficile colonization	0 (0.0)	1 (4.3)	1 (2.6)
Haematology/oncology patients	0 (0.0)	1 (4.3)	1 (2.6)
Surgical	0 (0.0)	3 (13.0)	3 (7.7)
Type 2 diabetes inpatients	0 (0.0)	1 (4.3)	1 (2.6)
Outbreak investigation
No	13 (81.2)	22 (95.6)	35 (89.7)
Yes	3 (18.7)	1 (4.3)	4 (10.3)
Study design
Cohort	2 (12.5)	16 (69.6)	18 (46.1)
Case–control	13 (81.2)	5 (21.7)	18 (46.1)
Nested case–control	1 (6.2)	0 (0.0)	1 (2.6)
Case–cohort	0 (0.0)	2 (8.7)	2 (5.1)
Non-case groupb
Symptomatic	2 (12.5)	7 (30.4)	9 (23.1)
Non-CDI	13 (81.2)	14 (60.9)	27 (69.2)
Two control groups	1 (6.2)	2 (8.7)	3 (7.7)
Unexposed group
No antibiotics	0 (0.0)	1 (4.3)	1 (2.6)
Unexposed to specific antibiotic	16 (100)	18 (78.3)	34 (87.2)
Reference antibiotic	0 (0.0)	4 (17.4)	4 (10.3)
Quality (NOS score)
Low (0–3)	0 (0.0)	1 (4.3)	1 (2.6)
Moderate (4–6)	3 (18.7)	10 (43.5)	13 (33.3)
High (7–9)	13 (81.2)	12 (52.2)	25 (64.1)

a

Includes two studies identified but not included in the 2013 review.

b

Symptomatic refers to patients with diarrhoea who tested negative for C. difficile. Non-CDI refers to all patients that were not diagnosed with CDI. May include patients without diarrhoea and also patients with diarrhoea plus a negative C. difficile test.

Quality appraisal

Most studies were graded as medium or high quality (Table 1). Full details of the NOS quality appraisal score for each study are summarized in Table S2. Selection and confounding were important sources of error, with only 17 (43%) studies attaining the maximum score for selection and 15 (38%) studies reaching the maximum score for comparability; 86% of case–control designs scored 3/3 for exposure measurement and 50% of cohort designs scored 3/3 for outcome measurement.

Study results and pooled effects

Studies that used an exposure comparison group consisting of those not exposed to the antibiotic measured were included in meta-analyses (n = 34).

For non-β-lactam antibiotic classes (Figures 2 and 3), the strongest evidence for an association was seen for quinolones (fluoroquinolones) and lincosamides (clindamycin). Overall, quinolones were associated with 34% increased odds of HCFA-CDI (OR = 1.34, 95% CI = 1.13–1.60), although individual ORs ranged from 0.15 to 15.30. Excluding the study with an extremely small outlier association (0.15) made little change to the pooled result (OR = 1.37, 95% CI = 1.15–1.63; I² = 85.4%). Lincosamides were associated with 56% increased odds of HCFA-CDI (OR = 1.56, 95% CI = 1.13–2.14), with individual study ORs ranging from 0.39 to 9.10. Weak positive pooled associations were observed for aminoglycosides, macrolides and trimethoprim/sulphonamides and a wide range in individual study effect sizes was found. There was strong heterogeneity observed for each of these classes, with ORs distributed in both positive and negative directions. Associations were found also for antibiotics used to treat CDI, particularly vancomycin (glycopeptide class) with a pooled OR of 1.91 (95% CI = 1.32–2.78).

Figure 2.

Forest plot for non-β-lactam antibiotic classes: quinolones, aminoglycosides, macrolides and lincosamides. This figure appears in colour in the online version of JAC and in black and white in the print version of JAC.

Figure 3.

Forest plot for non-β-lactam antibiotic classes: tetracyclines, trimethoprim/sulphonamides and miscellaneous antibiotics. This figure appears in colour in the online version of JAC and in black and white in the print version of JAC.

Figure 4 displays the results for β-lactam classes. The strongest association was seen for carbapenems (OR = 2.55, 95% CI = 1.83–3.55), with penicillins and cephalosporins associated with 33% and 79% increased odds of HCFA-CDI, respectively. There was substantial heterogeneity for all three main classes. For carbapenems, all ORs except one were in the positive direction, ranging from 0.79 to 14.13. For penicillins and cephalosporins there were a number of ORs in opposite directions.

Figure 4.

Forest plot for β-lactam antibiotic classes. ‘Other’ subgroup refers to a circumstance where data on general β-lactam exposure were reported. This figure appears in colour in the online version of JAC and in black and white in the print version of JAC.

β-Lactamase inhibitor combination penicillin antibiotics were the most frequently reported penicillin subclass (Figure 5). ORs ranged from 0.47 to 17.40 (three studies had OR < 1) with a pooled OR of 1.43 (95% CI = 1.16–1.77). There was little evidence of an association for aminopenicillins, and few studies reported data for broad-spectrum or penicillinase-resistant penicillins.

Figure 5.

Forest plot for penicillin subclasses. This figure appears in colour in the online version of JAC and in black and white in the print version of JAC.

For cephalosporins, third- and fourth-generation classes were associated with a doubling of the odds of HCFA-CDI, second-generation cephalosporins with a 58% increased odds of HCFA-CDI and no evidence for an association with first-generation cephalosporins (Figure 6). The individual study findings for fourth-generation cephalosporins were the most homogeneous, with all effect sizes in a positive direction and ranging from 1.1 to 3.24.

Figure 6.

Forest plot for cephalosporin subclasses. This figure appears in colour in the online version of JAC and in black and white in the print version of JAC.

Meta-regression and subgroup analyses

Ten antibiotic classes with a minimum of 10 studies and where more than 50% of the variation was due to heterogeneity were included in the subgroup analysis and meta-regression. The results of subgroup analyses are presented in Figures S1 to S3. Although several sources of heterogeneity were identified for eight antibiotic classes, there were no sources common to all classes (Table 2). The most common sources were geographical region, exposure measurement period and measurement of exposure after onset of symptoms.

Table 2.

Meta-regression analyses

Factor	Bivariate associations				Multivariable associations
	OR (95% CI)	P value	Residual I2	Adjusted R2	OR (95% CI)	P value	Residual I2	Adjusted R2
Combination penicillins (n = 15)
Regiona							63.2	82.8
Europe	0.77 (0.37–1.63)	0.02	84.8	43.3	0.65 (0.34–1.24)	0.02
Latin America/East Asia & Pacific	3.68 (1.49–9.08)				2.98 (1.35–6.57)
Exposure measured after onsetb
Yes	—	—	—	—	—	—
Unsure	0.50 (0.26–0.96)	0.04	70.0	67.6	0.51 (0.33–0.80)	0.01
First-generation cephalosporins (n = 10)
Exposure measurement periodc	1.20 (0.96–1.50)	0.10	46.7	100.0	—	—	—	—
Second-generation cephalosporins (n = 13)
Exposure measurement periodc	1.73 (1.02–2.93)	0.04	59.4	48.0	1.42 (0.85–2.39)	0.16	55.7	64.1
Confirmed vs. probable HCFA definition	3.05 (1.14–8.17)	0.03	74.4	46.0	2.36 (0.81–6.88)	0.10
Third-generation cephalosporins (n = 13)
Exposure measurement periodc	1.59 (1.13–2.25)	0.01	54.9	76.9	0.85 (0.42–1.74)	0.61	12.3	84.6
Clinical subgroup vs. general inpatient population	0.46 (0.22–0.96)	0.04	80.8	25.5	0.42 (0.18–0.97)	0.04
Case–control vs cohort study design	2.51 (1.30–4.82)	0.01	69.1	51.8	3.57 (0.86–14.79)	0.07
Confirmed vs. probable HCFA definition	2.38 (1.02–5.58)	0.05	80.3	25.2	0.51 (0.15–1.70)	0.23
Aminoglycosides (n = 22)
Regiona							0.00	93.8
Europe	1.28 (0.77–2.11)	0.005	39.6	63.9	1.11 (0.73–1.68)	0.001
Latin America/East Asia & Pacific	3.42 (1.72–6.79)				3.49 (1.94–6.27)
Exposure measured after onsetb
Yes	1.85 (0.92–3.74)	0.03	48.0	42.9	2.08 (1.29–3.35)	0.001
Unsure	0.57 (0.31–1.04)				0.69 (0.45–1.05)
Lincosamides (n = 22)
High vs. low/medium study quality	0.48 (0.20–1.15)	0.09	86.7	11.6	—	—	—	—
Macrolides (n = 22)
Clinical subgroup vs. general inpatient population	0.49 (0.23–1.05)	0.07	94.3	17.8	0.51 (0.28–0.92)	0.03	69.4	67.2
Exposure measured after onsetb
Yes	3.40 (1.04–11.15)	0.05	90.4	27.0	3.35 (1.44–8.25)	0.03
Unsure	0.57 (0.22–1.49)				0.99 (0.46–2.15)
Symptomatic vs. asymptomatic non-cases	2.88 (1.24–6.71)	0.02	92.6	24.6	2.63 (1.36–5.09)	0.01
Trimethoprim/sulphonamides (n = 15)
Regiona
Europe	1.22 (0.56–2.68)	0.05	61.2	55.9	—	—	—	—
Latin America/East Asia & Pacific	4.23 (1.36–13.17)				—	—	—	—

Adjusted R², percentage of between-study variance explained; residual I², percentage of residual variation due to heterogeneity.

a

Reference category: North America.

b

Reference category: no.

c

Per-interval increase (short/mid/long).

Studies conducted in Latin America and East Asia & Pacific regions reported stronger associations for combination penicillins, aminoglycosides and trimethoprim/sulphonamides, approximately 3–4 times higher than those reported by North American studies. Measurement of the antibiotic exposure after onset of symptoms was also a source of heterogeneity for aminoglycosides, with associations twice that of studies that collected exposure information preceding onset of symptoms. A longer window of antibiotic exposure measurement was associated with all three cephalosporin subclasses in unadjusted analyses and associations were attenuated in multivariable models. Other sources of heterogeneity for cephalosporins included definition of HCF acquisition (second and third generation), study population (third generation) and study design (third generation). Studies using patient subgroups tended to have weaker associations for a number of antibiotic classes, and this was a strong source of heterogeneity for associations with macrolides and third-generation cephalosporins, with effect sizes half of those seen in studies of general inpatients.

Other findings

Five studies were not included in the meta-analysis due to differences in antibiotic exposure measurement; two Canadian studies used no antibiotic exposure as the reference category,^31,32 one study examined duration and dosage of antibiotic exposure³³ and two UK studies used historical controls to examine the impact of changes to antibiotic formulary changes.^34,35 The main results are summarized below.

A case–cohort study of adult inpatients of a large tertiary hospital in Canada measured the risk of HCFA-CDI associated with exposure to various antibiotic classes compared with patients with no antibiotic exposure in the 5 days prior to developing symptoms.³¹ After adjusting for age, gender, hospital exposure and infection pressure, the incidence of CDI was 2–3 times higher for penicillins, cephalosporins, fluoroquinolones and carbapenems. The second Canadian study compared the risk of CDI for various antibiotics according to whether CDI developed during or following cessation of exposure compared with no exposure, taking into account competing events of death or discharge.³² They found the risk of CDI was greatest in the period following exposure to penicillins, quinolones, macrolides, aminoglycosides or clindamycin.

The cohort study of ICU patients with HCFA pneumonia by Li et al.³³ reported the mean duration of treatment and total dose per patient to be higher in patients who developed CDI for broad-spectrum cephalosporins, carbapenems and oxacephems, although not all of the differences were statistically significant.

Two UK clinical cohort studies used historical controls to examine the impact of changes to antibiotic formulary changes. In patients with severe HCFA pneumonia, none of the patients treated with dual β-lactam therapy of amoxicillin plus temocillin (n = 98) developed CDI, compared with 7.4% (7/94) of patients who had previously been treated with piperacillin/tazobactam (a β-lactamase inhibitor combination).³⁴ In a study comparing gentamicin (aminoglycoside) with cefuroxime (second-generation cephalosporin) for total hip and knee replacement surgery prophylaxis, there were no CDI cases in the gentamicin group (n = 2101) compared with 11 cases (0.2%) in the cefuroxime group (n = 6094).³⁵

Non-English language studies

Thirteen non-English language articles were identified in the current review, of which only one study was eligible for inclusion in the review.³⁶ The case–control study of immunosuppressed patients reported that exposure to fluoroquinolones was higher in CDI cases (36%) than in controls (28%) and exposure to second- or third-generation cephalosporins was lower in cases (6%) than in controls (14%). Details of the non-English language studies are summarized in Table S3.

Funnel plot symmetry

Figure S4 displays the funnel plots for the studies included in the meta-analyses of each main antibiotic class. Although there was not strong evidence for bias due to non-reporting of small studies, several of the plots reflect asymmetry associated with the heterogeneity previously addressed.

Discussion

This systematic review updates the findings from two previous reviews through to 2020. Twenty-one studies published since 2013 were included in this review, in addition to 16 studies identified in earlier reviews.^10,11 The main findings indicate that carbapenems and third- and fourth-generation cephalosporin antibiotics remain the most strongly associated with HCFA-CDI, with cases more than twice as likely to have recent exposure to these antibiotics prior to developing CDI. Modest associations (OR > 1.5) were observed for quinolones (predominantly fluoroquinolones), lincosamides (namely clindamycin), second-generation cephalosporins and β-lactamase inhibitor combination penicillin antibiotics. Individual study effect sizes were variable and heterogeneity was a common problem observed for most antibiotic classes. However, the risk of CDI for a given antibiotic will depend on the local prevalence of strains that are resistant to the particular antibiotic and a certain degree of heterogeneity is therefore to be expected.⁸

Since 2013, there have been more studies conducted in a wider range of countries, including three from Latin America (covering Argentina, Mexico and Brazil)^37,38 and four from East Asia & Pacific countries (China, Korea, Malaysia and Taiwan).^33,39–41 The studies from Latin America and East Asia & Pacific regions reported stronger associations for a number of antibiotic classes with HCFA-CDI. With only a small number of studies from these regions, conclusions regarding such differences are limited and further studies are needed.

In addition, a more diverse range of patient populations have been studied since 2013, with an increase in focused studies on clinical patient groups, such as surgical^35,42,43 and ICU patients,⁴⁴ or patients with particular conditions, such as HCFA-pneumonia,^33,34,45 compared with earlier studies that predominantly studied general inpatients. Studies using patient subgroups tended to have weaker associations. A potential explanation is that studies of patient subgroups had shorter periods of antibiotic exposure measurement. For example, in studies assessing third-generation cephalosporins, three out of seven studies of general inpatients measured exposure up to 3 months prior to CDI, whereas no study that was restricted to patient subgroups used a 3 month exposure window. Studies that used longer windows of exposure measurement had stronger effect sizes, which could reflect a protracted time at risk for CDI following antibiotic treatment. Previous studies have shown that the risk of CDI declines between 1 and 2 months following cessation of antibiotic treatment.^31,46 Alternatively, it could mean that these studies simply picked up more exposed individuals, resulting in differential misclassification if this were not consistent for cases and non-cases.

Another exposure measurement source of heterogeneity was the inadequate description or failure to limit recording of antibiotic exposure to a period preceding onset or diagnosis of CDI, producing biased effect estimates. This could explain the associations found for antibiotics used to treat CDI such as vancomycin and metronidazole (glycopeptide and nitroimidazole classes, respectively), although this could not be established in this review, and both are able to incite CDI.⁴⁷ Future studies should clearly report the parameters of exposure measurement, including all sources of information on exposure, as it was often unclear, particularly in studies with a longer exposure window, whether in-hospital prescription only was recorded, or whether prescription in the community setting was included. Information on dose–response relationships was generally lacking and studies investigating the risk of HCFA-CDI associated with the timing and duration of antibiotic exposure are needed.

Other methodological differences in studies, such as the use of a clear definition of HCF acquisition, the choice of non-case comparison group, and group comparability on confounding factors, were not consistent sources of heterogeneity, but had varying influences on study results. A variety of diagnostic laboratory testing methods that have varying sensitivity and specificity were used by studies included in the review. With increasing computing power, cohort studies utilizing routinely collected patient data have increased in recent years—investigators should clearly report surveillance definitions, criteria and laboratory methods for diagnosis of CDI in their institutions, and how these align with relevant current guidelines.^8,48 Studies are subject to diagnostic suspicion bias if patients exposed to antibiotics are more likely to be tested for C. difficile.

This review provides the most up-to-date synthesis of evidence in relation to the risk of HCFA-CDI associated with exposure to specific antibiotic classes. However, the review is limited by the availability of a single reviewer to select, extract and critically appraise the studies. Since the previous review covering studies published up to the end of 2012, there has been a marked increase in the overall number of studies eligible for inclusion. As found previously, studies were variable in methodological quality and the quality of reporting results. Studies from the Latin America and East Asia & Pacific regions have increased, although studies were predominantly conducted in North America or Europe, and more studies outside of these settings are needed.

Funding

This study did not receive any external funding.

Transparency declarations

All authors have no conflicts to declare in relation to this work.

Supplementary data

Tables S1 to S3, Figures S1 to S4, literature search information and included studies are available as Supplementary data at JAC Online.