Patient and Strain Characteristics Associated With Clostridium difficile Transmission and Adverse Outcomes

Combining epidemiological, clinical, and genomic data for Clostridium difficile cases offers fresh insight into characteristics associated with transmission and outcome. C. difficile lineages may have different reservoirs and modes of transmission, and recent strain acquisition is associated with poorer outcomes.

the metropolitan area of Leeds, UK (population 750 000). It comprises 3 large teaching hospitals and numerous smaller community care providers. During this study, Leeds Teaching Hospitals' infection control policy stipulated that any patient with ≥1 episode of unexplained diarrhea should be isolated and tested for CDI [13]. Patients were source-isolated from the day of onset of diarrhea to being either >48 hours symptom-free or discharged from hospital. Standard CDI treatment during this period was oral metronidazole for non-severe cases and oral vancomycin for severe cases (total white cell count >15.0x10 9 /L, serum creatinine >50% above baseline, or evidence of colitis).
Consecutive diarrheal samples submitted for CDI testing and confirmed as toxin-positive by a cell cytotoxicity assay were cultured from a single colony, ribotyped [4], and sequenced using Illumina technology [9]; the samples included any repeat positives >7 days later from the same patient. Samples were compared using ribotype and differences in single nucleotide variants (SNVs) obtained from maximum likelihood phylogenies (Supplementary Methods). Sequences generated during this study can be found on the NCBI short read archive under BioProject PRJNA317528 (http://www.ncbi.nlm.nih.gov/ bioproject/PRJNA317528).
CDI cases were defined as the first culture-positive C. difficile sample from each cytotoxin-positive patient, plus repeat culture-positive samples >10 SNVs distinct from all prior samples (extremely unlikely to arise through mutation during the study) [8,9,14]. For each case, data were collected on sex; age; all inpatient episodes in the 6 months before and after each sample, including all ward-stays, day-case procedures, dialysis, and chemotherapy attendances; time between stool sampling and source isolation; home postcode district; general practice location; severity biomarkers at diagnosis (summarized as severity score 0-3; one point each for age >65 years, peak creatinine >176μmol/L, and peripheral blood white cell count >20 000 cells/µL) [15]; recurrence (occurring in the next 120 days); and all-cause 30-day mortality. Recurrent infection was defined as a repeat cytotoxin-positive sample >14 days after the first positive sample (reliable information was not available about symptom resolution between samples). Consensus definitions were used for community-associated (CA), indeterminate (I), community-onset healthcare facility-associated (CO-HCFA), and hospital-onset healthcare facility-associated (HO-HCFA) CDI (Supplementary Figure S1) [3].
For each new sequenced case (n = 640), we considered all previous case strains (including repeat positives within individuals) as potential transmission sources. Using rates of C. difficile evolution and within-host diversity, direct transmission was considered plausible where a prior "donor" isolate was within 0-2 SNVs of a "recipient" [9]. Cases within 3-10 SNVs were likely to share a common source in the last 5 years, but direct transmission was unlikely. Cases >10 SNVs from all previous cases were considered genetically distinct and unlikely to share a common source with another case during the study.

Statistical Methods
Univariable associations between these factors and cases having acquired C. difficile from a previous, closely genetically-related case or having transmitted CDI onwards (≤2 SNV threshold) were investigated using Kruskal-Wallis/rank-sum (continuous factors) and chi-squared/exact (categorical factors) tests. For multivariable logistic models for acquisition, transmission, 120day recurrence, and 30-day mortality, severity score (missing in n = 124 [19%] where blood tests were not done) and source isolation at onset (missing in n = 269 [42%]; 150 pre-January 2011, when this was not collected, and 119 post-January 2011, due to poor documentation) were imputed multiply (50 times) using both chained estimating equations [16] (Stata version 14.1; including all factors above and all four outcome variables) and boxcox-transforming continuous variables for normality. Independent predictors were selected using backwards elimination (exit P > .1 to fit an exploratory model) using fractional polynomials to investigate non-linearity [17]. Independent effects of each ribotype with >25 cases (vs all other ribotypes) were considered for inclusion in main models. Sensitivity analyses included ribotype as an 8-level factor (7 common vs others combined), together with all factors identified as independently prognostic in any model or all other factors.
The study was approved by the Berkshire Research Ethics Committee (10/H0505/83) and the Health Research Authority (8-05[e]2010).
Median age at diagnosis was 76 years (inter-quartile range [IQR] 62-84), and 389 (59%) cases were female patients. The incidence of healthcare-associated CDI was 4.2 HO-HCFA and CO-HCFA infections per 10 000 overnight stays. The incidence of community-associated CDI was 143 CA and I infections per 100 000 Leeds community per annum. Incidence declined during the study period (Supplementary Figure S3) Figure S4).

C. difficile Transmission From Prior Cases
Out of the new cases, 567 (89%) had the same ribotype as at least one previous CDI in the study population, but only 227 (35%) were 0-2 SNVs from a previous isolate, suggesting recent acquisition from the symptomatic population, while 286 (45%) were >10 SNVs from all prior isolates (Table 1). These proportions were similar when restricting to cases from January 2011 onwards (175/449 [39%] 0-2 SNVs and 176/449 [39%] >10 SNVs, P = .20), demonstrating there was no significant difference in the proportion of cases with a recent donor when accounting for the potentially reduced pool of donors in the first months of the study.
Only 115 (18%) new CDI cases had any prior direct ward contact (same ward, same day) with a 0-2 SNV donor; this contact was for a median of 8 days (IQR 4-16; range 1-59), with a median of 20 days (IQR 6-45; range 1-191) between the donor's most recent positive and the recipient's first positive sample (counting the most recent donor if multiple donors were identified). Other potential transmission routes are summarized in Table 1.

Impact of Acquisition on Clinical Outcomes
All-cause 30-day mortality was 27% (62/227) for patients with a previous potential donor, but only 12% (34/286) for those with genetically distinct strains (P < .001). Recurrence (>14 days after first test) was also more frequent in cases with any potential donor (29%, 65/227) versus those without (16%, 47/286; P = .001). The association between acquisition and mortality did not persist after adjusting for other independent predictors ( Table 2; P = .06), but there was an independent association between acquisition and recurrence (P < .001). Importantly, other than ribotype-020 marginally increasing the risk of recurrence vs all other ribotypes combined (P = .06), there was no additional impact of ribotype on either mortality or recurrence after adjusting for the (stronger) impact of acquisition itself (P > .1); thus, any excess risk of adverse outcomes associated univariably with specific lineages (ribotype-027 in particular, with univariable P < .001 for associations with mortality and recurrence) was mediated through acquisition from a previous symptomatic donor.

Risk Factors for Acquiring CDI From a Previous Case
The factors most strongly univariably associated with having a previous potential donor (0-2 SNVs) were older age, origin of infection (CA/I/CO-HCFA/HO-HCFA), more inpatient days in the 12 weeks pre-diagnosis, higher severity score, and ribotype (Table 3). More recent hospital exposures were associated with greater proportions of CDIs closely related to a prior case (Supplementary Figure S5, Supplementary Table S3), both at SNV thresholds consistent with direct and recent indirect transmission (suggesting these cases arose from a reduced hospital reservoir of strains). The range of ribotypes causing CDI was broad regardless of healthcare or community origin of infection, although ribotype-027 was particularly associated with close healthcare exposure (P = .02 vs other ribotypes combined; Supplementary Figure S6). The proportion of cases with potential donors at differing levels of genetic relatedness varied markedly according to CDI ribotype (Figure 1, Supplementary Table S4). The vast majority of ribotype-027 (102/105 [97%]) and ribotype-078 (54/57 [95%]) CDIs had a previous case within 0-10 SNVs; however, genetically close matches (0-2 SNV) occurred in 94% (99/105) and 46% (26/57), respectively. Furthermore, 64% (67/105) of ribotype-027 but only 11% (6/57) of ribotype-078 cases had shared time on a ward with a potential donor ≤2 SNV, supporting greater rates of hospital-based acquisition for ribotype-027 compared with ribotype-078. Other common ribotypes (e.g. 015 and 002) were much more genetically diverse; thus, lower proportions of these CDIs had prior strains within 0-10 SNVs (56% [35/62]  In a multivariable analysis, the odds of acquiring CDI from a previous symptomatic donor (0-2 SNVs) remained independently higher for specific ribotypes (particularly 027; to a lesser extent, 078/014/020/001/072), older individuals, and those with more inpatient days in the preceding 12 weeks; as expected, risk was lower in the first study month, where fewer previous cases had been sequenced (Table 2). There was no independent effect of origin of infection (CA/I/CO-HCFA/HO-HCFA) after adjusting for these factors (P = .71), which was plausibly better explained by total inpatient days in the last 12 weeks rather than time since last exposure, categorized as per surveillance criteria.

Risk Factors for Onward Transmission
A total of 228 (36%) CDI cases were potential donors (0-2 SNVs) to ≥1 subsequent cases. The factors most strongly univariably associated with being a potential donor were older age, male sex, earlier diagnosis in the study period, more inpatient days during the 12 weeks before diagnosis, higher severity score, and ribotype (Table 4). In 20/115 (17%) recipients, the most plausible transmission event (using the most recent donor if multiple donors were identified) followed a donor's second or third diarrheal sample, a median of 49 days (IQR 21-95; range 8-129) after the initial test. Overall, 45/115 (39%) of the most plausible donors had recurrent infections.
In a multivariable analysis, the odds of onwards transmission to a new symptomatic CDI case (0-2 SNVs) remained independently higher for specific ribotypes (particularly 027; to a lesser extent, 078/014/020/002/001/072), older individuals, men, and those with more inpatient days in the preceding 12 weeks; as expected, odds decreased slightly over the study period ( Table 2). After adjusting for these factors, there was no independent effect of donor origin of infection (P = .93) or inpatient days post-diagnosis (P = .47).

CONCLUSIONS
Our study identified risk factors connected with healthcare-associated transmission of CDI, and measured the clinical consequences of recent C. difficile acquisition from a symptomatic case, having robustly defined genetic relatedness between cases using WGS. We examined a large cohort of cases using a sensitive approach to sampling and diagnosis based on toxin testing, which has been shown to have a high predictive value for true CDI [19,20]. By combining WGS and conventional ribotyping, we found that genetic relatedness of strains to previous cases varies significantly according to ribotype, suggesting C. difficile Interestingly, however, we did not demonstrate an independent relationship between ribotype and adverse outcome, as some previous studies have shown; recurrence and mortality were more strongly associated with strain acquisition from a recent case than with strain type.
We found that predictors of recent strain acquisition and onward transmission included more inpatient days during the 12 weeks before diagnosis, older age, and certain ribotypes. These data could help to identify a high-risk population for healthcare-associated strain acquisition that could be used when planning future preemptive C. difficile interventions, such as vaccination [21] or prophylaxis [22]. Interestingly, the total number of days spent in hospital in the 12 weeks prior to infection was more closely associated with strain acquisition than conventional community vs. hospital definitions [3].
Our results provide new evidence that certain C. difficile ribotypes have higher levels of in-hospital transmissibility. Specifically, ribotype-027 strains were predominantly clonal, with approximately two-thirds of ribotype-027 CDIs having Abbreviations: CA, community associated: CDI, Clostridium difficile infection; CO-HCFA, community-onset healthcare facility-associated; HO-HCFA, hospital-onset healthcare facility-associated; I, indeterminate; SNV, single nucleotide variants. a including any case with a subsequent case within the 0-2 SNV threshold as a potential donor as the specific transmission route is unknown. recent ward contact with a genetically matched case. This epidemic ribotype has been associated with more frequent poor outcomes and is known to be a recently evolved clone [7,23]. Conversely, ribotype-078, a strain also associated with poor outcomes [24,25], was largely clonal, but only 11% of ribotype-078 CDIs occurred in patients with recent ward contact with a genetically matched case. Other authors have also shown low patient-to-patient transmission for this strain [26]. C. difficile ribotype-078 is commonly found in livestock [27,28], hence food and farm exposure are potential alternative environmental sources [29][30][31]. Notably, other common disease-causing lineages (eg, 015, 002) demonstrated few genetic matches, so were unlikely to have been recently acquired from a common source. We show that 30-day mortality and 120-day disease recurrence were each more likely in cases where CDI was plausibly acquired from a recent donor. The association between recent acquisition and recurrence persisted after adjusting for other important factors, including age and ribotype; hence, this association was not due to confounding by ribotype-027 in particular. The association between recent acquisition and outcome may reflect a host contribution (eg, multiple comorbidities) predisposing a patient to both the acquisition and the poor outcome. Nevertheless, these findings strongly support the importance of infection control interventions in targeting symptomatic CDI cases to reduce the risk of transmission to vulnerable patients, and so prevent particularly poor outcomes in these individuals. Furthermore, in an era where nosocomial CDI incidence is increasingly viewed as a healthcare quality indicator [32], our data suggest that evidence demonstrating control of in-hospital C. difficile transmission (ie, lack of strain-relatedness between cases) could be used as a specific measure of prevention and control effectiveness.
Our study has several limitations. In particular, it is impossible to confirm patient-to-patient C. difficile transmission retrospectively, even with detailed epidemiological data and WGS. The few sequence failures may have led to donor/ recipient genetic relationships being missed. Transmission events outside the geographical boundaries of Leeds could also have been missed. We did not examine asymptomatic carriers and, therefore, our results apply only to transmission leading from and to disease; there is growing evidence that asymptomatic carriers may play a key role in wider C. difficile transmission [12,33,34]. However, as WGS-matched samples from the entire study period were considered as possible donors, the influence of clinically significant (but unknown) intermediate carriers has been accounted for, as two cases related by a common source were still linked to each other in the study. Furthermore, a low threshold for patient sampling and automatic C. difficile testing of diarrheal specimens will have enhanced our detection of possible C. difficile donors; such UK-recommended practice has been associated with a marked decrease in CDI incidents [5,35]. The incidence of ribotype-027 has decreased in the UK since the study period; the proportion of cases with a donor (or recipient) may be lower if the study is repeated. Data for some potentially influential factors were incomplete or not available (including proton-pump inhibitor use, antibiotic use, and co-morbidities), and so we have not considered these in our main analyses. Understanding the possible role of antibiotic use in transmission/acquisition of C. difficile is complex, not least given heterogeneity of exposure (including polypharmacy). Interestingly, antibiotic use in the 7 or 90 days before diagnosis was not strongly associated with either transmission or acquisition from a previous symptomatic donor (both P > .2; Supplementary Table S5) in patients for whom these data were available. Lastly, data were not available on diarrhea severity or duration; however, CDI severity criteria and repeat tests were used as proxies for these.
In summary, WGS and detailed epidemiological data demonstrate that acquisition of a C. difficile strain that caused a previous symptomatic case is associated with poorer clinical outcome. Targeting preemptive interventions to patients at high risk of being donors or recipients will likely further reduce C. difficile transmission and infection rates within healthcare institutions. This will help to improve patient experiences and outcomes. Notably, we also demonstrated variable healthcare-associated transmission of C. difficile strain types, suggesting CDI is not a homogeneous entity, but is likely to have different reservoirs and modes of transmission.

Supplementary Data
Supplementary materials are available at Clinical Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.