RESULTS
Consecutive samples sent for C. difficile testing at 6 hospitals were studied for 12 months (Table 1). In total, 1052/1098 (96%) of GDH/toxin-PCR screen-positive samples were available: 95/98 (97%) at hospital 1, 144/178 (81%) at hospital 2, 118/127 (93%) at hospital 5 and otherwise 100%. 974/1052 (93%) available samples were confirmed as C. difficile on culture. For the 5 hospitals with available testing data, 887/21539 (4.1%) of samples submitted for testing were culture-positive (Table 1); 971/974 (99.7%) culture-positive samples were successfully sequenced. Of sequenced culture-positive samples, 451/971 (46.4%) were EIA fecal-toxin-positive, 35–71% by hospital. By contrast, 851/971 (87.6%) were potentially toxigenic, that is, had toxin genes detected via sequence data. Hence, 400/851 (47.0%) samples containing potentially toxigenic C. difficile did not have fecal-toxin detected. In the 971 sequenced isolates, the most common ribotypes identified were 014, 015, 005, 002, 020, and 078 (Table 2). Ribotype-027(NAP1/ST-1) only accounted for 16 (2%) cases.
To view graphs and tables, please click the following link:
Relatedness to Prior Cases
The proportion of cases plausibly linked to a prior case by recent transmission varied by hospital. Of 851 sequenced potentially toxigenic cases, all were considered as potential sources of infection, but only the 652 obtained after the first 90 days of sampling at each hospital were assessed for linkage to a previous case. Across the 6 hospitals, 128/652 (20%, 95% confidence interval [CI] 17–23%) potentially toxigenic cases were genetically linked to a prior case from the previous 90 days. Hospital 2 had the fewest cases linked to a prior case, 7/105 (7%, 3–13%), hospital 1 had an intermediate number, 9/70 (13%, 6–23%), and hospitals 3–6 had similar numbers of linked cases, 37/153 (24%, 18–32%), 32/134 (24%, 17–32%), 18/76 (24%, 15–35%), and 25/113 (22%, 15–31%), respectively. Hospital 2 had significantly fewer linked cases than hospitals 3–6 (P ≤ .002), with weaker evidence for lower rates in hospital 1 than hospitals 3, 4, and 5 (P = .05, .07, .09, respectively). Overall, 48/128 (38%) of potential transmission recipients were fecal-toxin-negative (11–68% across hospitals, Figure 1A). Fecal-toxin detection in a recipient was associated with increased odds of having a potential transmission donor, odds ratio 1.67 (95% CI 1.12–2.48, P = .01).
In total, 59/128 (46%) putative transmission recipients were only linked to ≥1 fecal-toxin-positive potential donors, 50 (39%) to only fecal-toxin-negative donors, and 19 (15%) to both toxin-positive and toxin-negative donors. Considering the 667 cases occurring in the first 270 days at each hospital, that is, the cases with an opportunity to transmit to a sampled case within the next 90 days, 120 (18%) were potential donors. Fecal-toxin-positive and -negative cases were similarly infectious: the odds ratio for a fecal-toxin- positive case, compared to a fecal-toxin-negative case, being a potential transmission donor was 1.01 (95% CI 0.68–1.49, P = .97).
When only considering transmission to and from fecal- toxin-positive cases, fewer cases were genetically linked to a previous case within 90 days, 51/335 (15%, 95% CI 12–20%). We observed a different “ranking” of hospitals compared with the above analysis of linkage rates based on potentially toxigenic isolate-positive patients: hospital 3 had the greatest proportion of fecal-toxin-positive cases genetically related to a prior fecal-toxin-positive case, 31% (22–41%), and hospital 6 the lowest, 0% (0–9%) (Figure 1B).
Results were similar to those for all potentially toxigenic C. difficile (Figure 1A) if all C. difficile sequences, nontoxigenic as well as potentially toxigenic, were considered (Figure 1C). Considering only nontoxigenic isolates, very similarly to potentially toxigenic isolates, 19/96 (20%, 95% CI 12–29%) were genetically linked to a prior patient isolate from the previous 90 days.
There was no evidence that the number of linked cases varied during the study at any hospital (Figure 1D). Because different numbers of sequences were obtained from the different hospitals, we investigated how this affected the estimated proportions of cases linked to a prior case. Estimated proportions of linked cases were relatively stable once approximately 50 cases had been sequenced (Figure 2).
Impact of Testing Frequency
The proportion of originally tested samples that were stored and then culture-positive was similar across the 5 hospitals with testing data, 3.8%–4.3% (P = .89, Table 1). In contrast, testing rates ranged from 98 to 239 samples per 10000 bed-days. There was no association between the estimated proportion of cases linked to a previous case within 90 days and testing rates (P = .19 for all potentially toxigenic cases, Figure 3A, and P = .60 for fecal-toxin-positive cases only, Figure 3B). For comparison, Figure 3B also displays rates of linked cases for previously published data from Oxford and Leeds.
Figure 2: Proportion of potentially toxigenic cases linked to a previous potentially toxigenic case by hospital and number of sequences obtained. Abbreviation: SNP, single-nucleotide polymorphism.
Adjustment for Ribotype
After adjustment for locally circulating ribotypes, estimates of the proportion of potentially toxigenic cases related to a previous potentially toxigenic case within ≤2 SNPs and ≤90 days remained largely unchanged (Figure 4A). Using the same model, per-ribotype estimates for the proportion of related cases, adjusted for differences across hospitals, showed more variation (Figure 4B, Table 2 for unadjusted proportions). Ribotype-027 had significantly more related cases (adjusted proportion, 57%, 95% CI 29–81%, n = 12) than the comparison group of all other ribotypes (11%, 7–18%, P = .002, n = 124), as did ribotype-002 (25%, 15–38%, P = .04, n = 53), 012 (50%, 29–71%, P = .001, n = 22), and 087 (44%, 23–67%, P = .005, n = 18).
Adjustment for Completeness of Testing
As only 144/178 (81%) of GDH-positive samples at hospital 2 were retrievable for culture we assessed the likely impact of these missing samples on the estimated proportion of linked cases by simulating transmission and sampling at a theoretical hospital (Figure S1). As sampling becomes increasingly less complete, the estimated proportion of linked cases declines proportional to the probability of a case being sampled. Applying our simulation to hospital 2 provides a revised estimate of 8% of cases being linked to a prior case (see Supplement for details).
……………………….
DISCUSSION
Here, we demonstrate the value of WGS as a tool to estimate different rates of C. difficile transmission across institutions. Sequencing consecutive C. difficile isolates from routine testing over one year, we found transmission rates varied between 6 hospitals. Considering all patients with potentially toxigenic C. difficile, irrespective of fecal-toxin status, in the best performing hospital only 7% of patients’ isolates were sufficiently genetically related to a previous isolate from another patient to support transmission (8% adjusting for incomplete sampling). By contrast, approximately 3–4-fold more isolates (22–26%) were related in 4 of the other hospitals. These results remained similar after adjusting for the locally circulating strains.
Restricting to only patients with fecal-toxin-positive CDI, we confirmed previous findings that only a minority of CDI cases arise from contact with another symptomatic case: 35% in Oxford [10], 35% in Leeds [11], and 37% of ribotype-027 cases in Liverpool [12], were genetically linked to a previous case, with only a subset of these cases sharing time and space on the same hospital ward.
Applying the criteria for linking cases used in the present study to the Oxford and Leeds data sets, 38% of cases in Oxford were linked to a previous case in 2008 falling to 19% in 2010, and 30% of cases were similarly linked in Leeds. Across the 6 study hospitals, serving a range of populations, toxin-positive CDI linkage rates were all <15% with the exception of hospital 3, where 31% of cases were linked. It is likely the lower linkage rates in the current study in part reflect the falling incidence of ribotype-027 [11], associated with more onward transmission in this study, likely as a result of national fluoroquinolone restriction [27] but may also represent changes in infection prevention and control practice.
Our findings also support the recently reported role in transmission of GDH-positive patients with toxigenic C. difficile, but no detected fecal-toxin [28]. By sequencing all GDH-positive cases, we were able to compare the probability of fecal-toxin-positive and toxin-negative patients being potential sources of transmission, that is, having C. difficile genetically linked to a subsequent C. difficile isolate in another patient. Fecal-toxin-negative patients were similarly infectious to fecal-toxin-positive patients: fecal-toxin status did not affect the odds of being a potential transmission source. Strategies to identify and institute infection control measures around patients with potentially toxigenic C. difficile without detected fecal-toxin are therefore likely to reduce overall CDI incidence, although may be more costly, for example if toxin gene PCR is used as an initial screen rather than GDH EIA. Toxin-positive patients, that is, CDI cases, were more likely to have an identified potential transmission donor, than toxin-negative patients. This is in keeping with previous observations that recent C. difficile acquisition is associated with increased risk of disease, whereas long-term carriage is relatively protective [29].
It is likely that differing clinical CDI testing thresholds applied across the study hospitals, despite each being guided by national recommendations; notably, testing rates varied more than 2-fold between hospitals (98–239 tests/10000 bed-days). However, despite this variation, the overall proportion of samples tested that were C. difficile culture-positive was very similar across hospitals (~4%). These 2 findings combined resulted in varying rates of potentially toxigenic C. difficile isolation, 4.2–8.2/10000 bed-days, and varying (fecal-toxin-positive) CDI rates, 1.8–5.7/10000 bed-days. As the proportion of samples that were C. difficile culture-positive was close to reported community asymptomatic C. difficile colonization rates (~4%), and lower than reported colonization rates in asymptomatic hospital inpatients, (~10%) [30], it is possible that the higher reported CDI rates in some study hospitals may reflect overascertainment; independent assessment of which symptomatic patients are tested for CDI would be required to resolve this with certainty [7]. As designed, the study did not measure the extent of transmission involving asymptomatic patients, and therefore it is likely that not all hospital-associated transmission is captured. However, as this was the case for all hospitals, comparisons can still be made between hospitals and with previous studies investigating symptomatic patients.
Interestingly, we did not find any evidence of a relationship between rates of C. difficile testing and proportions of cases that could be linked to a previous case. Differing sampling/testing will likely mean the study populations at each hospital varied, for example with some institutions potentially more likely to include milder CDI cases than others. It should also be noted that differences in the population sampled by a particular testing strategy may affect the proportion of cases linked differently to incomplete sampling of a given population. We quantified the impact of the latter through simulation. Unfortunately, incomplete sampling could appear very similar to the impact of good infection control, as both results in low proportions of linked cases. One study limitation is that we only sequenced 81% GDH-positive samples at hospital 2. However, we demonstrate it may be possible to adjust for incomplete sampling, providing missed cases as assumed missing at random, and the number of onward transmissions from each case was random.
Both a limitation and a strength of our approach is that it relies only on sequencing laboratory samples and sampling dates. We demonstrate this allows comparative hospital surveillance with very limited, and no personal, sensitive or confidential, data. However, without ward admission and patient contact data, it is possible some genetically linked cases do not represent direct transmission from other cases. Genetic links might also arise through indirect healthcare-associated transmission via unsampled hosts or the hospital environment. Additionally, a minority of cases, without healthcare exposure in the last 90 days, may still have been genetically linked. However, there is no obvious reason why genetically related community C. difficile exposures, and therefore the proportion of such cases linked, should vary across England at a population level, even if other CDI risk factors do vary geographically, for example, antimicrobial use. Therefore, although we analyze transmission within the populations served by each hospital, as most CDI cases have recent healthcare exposure, the overall proportion of linked cases is still likely to be a reasonable combined indicator of infection control performance around cases and more generally. Without patient-level identifiers some repeat tests from the same patient may have been wrongly assigned as transmission events; however, we anticipate this was uncommon; repeat testing within 28 days is discouraged in national guidelines [17], and such samples are frequently not routinely processed.
Our method of comparing infection control performance depends on culturing C. difficile, which is not routinely undertaken, and on sequencing at least 6 months of samples, at around US$100 per sample. However, if samples are stored, as recommended in England, C. difficile could be cultured and sequenced retrospectively if increased incidence was noted and then continued prospectively to monitor the impact of any interventions. The cost-effectiveness of such an approach needs further evaluation.
In summary, here we present a novel method that enables assessment of the extent of hospital-acquired infection transmission within healthcare institutions. This approach revealed differences in CDI transmission rates across 6 English hospitals. It demonstrates the potential of whole-genome sequencing as a nationwide tool to identify institutions with excellent and also suboptimal infection control and therefore has the potential to allow targeted efforts to reduce CDI incidence.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.