This is a very impressive New England Journal of Medicine study from an Oxford University based group, using whole genome sequencing to really dissect relatedness of C. difficile isolates over a 5 year period. The study evaluates how many cases of C. difficile infection (CDI) were caused by isolates that were genetically related to previous symptomatic cases. This is not quite the same thing as evaluating how much CDI is hospital-acquired, mainly because the test used to detect CDI in the study has been phased out due to poor sensitivity, patients and staff were not screened for asymptomatic C. difficile carriage, and the environment was not sampled, so there was a large, unrecognized, hospital-based C. difficile reservoir from which horizontal transmission almost certainly occurred.

A major problem was the use of an Enzyme Immuno Assay (EIA) test kit to detect CDI. Whist these tests were used pretty much universally in the UK at the time of the study, they have now been shown to be very unsatisfactory. The sensitivity of EIA for the detection of CDI has been as low as 50% in some studies. Put another way, for every case of CDI that is detected, one goes undetected. This is crucially important in the context of this study, where the undetected CDI cases would contribute to the burden of asymptomatic carriers, which together would contribute to transmission. It’s also worth noting that C. difficile could not be cultured from 25% of stool samples that were EIA-positive, suggesting that the test may have had poor specificity too. The authors did try to ‘control’ for this problem, by effectively assuming that all stool specimens tested for CDI were positive in a sensitivity analysis, but this did not really help in explaining genetically related cases with no discernable epidemiological links.

There is also a technical point about the definition of ‘genetically distinct’ in terms of whole genome sequencing. If two isolates differ by 11 base pairs across the whole genome, do they originate from the same strain? It’s difficult to tell. In this study, they used a fairly conservative measure of relatedness: >10 single nucleotide variants (SNVs) was considered ‘genetically distinct’, and ≤2 SNVs was considered ‘genetically related’. This may have over-estimated apparent genetic heterogeneity. To be fair, the authors did perform a careful ‘validation’ study to determine the clock speed of mutation in their isolates by sequencing the first and list isolates obtained from a sample of patients, finding that 0-2 SNVs were expected for isolates <124 days apart. Even using these conservative measures of relatedness, the majority (55%) of isolates were related (‘not genetically distinct’ to be precise) to others in the collection (≤10 SNVs) and around a third of isolates were ‘genetically related’ to others in the collection (≤2 SNVs).

The authors performed detailed work to explore epidemiological associations between genetically related isolates (Figure). No acute- or community-based epidemiological links could be identified for 36% of the 333 genetically related cases, which perhaps supports the presence of unrecognized symptomatic cases or asymptomatic carriers.

Figure: Epidemiology relationships between 333 genetically related cases. ‘Ward contact’ = shared time on the same ward; ‘Hospital contact’ = shared time in the same hospital, without direct ward contact; ‘Ward contamination’ = admitted to the same ward within 28 days of the discharge of a symptomatic patient; ‘Same GP’ = no hospital contact, but shared the same GP; ‘Same postcode’ = no hospital contact, but shared the same postal code).

The overall rate of CDI was low, at <1 per 1000 patient days and it is noteworthy that the prevalence of genetically related and genetically distinct cases declined during the study period. I suspect if the same study had been performed for the period of 2000-2005, when more hospital transmission was almost certainly occurring, then a far higher proportion of isolates would have been genetically related.

I fear that this study will be used by some to ‘prove’ that horizontal transmission of C. difficile in healthcare settings is now uncommon, and most hospital-onset cases can be explained away by “CA-CDI”. Due to the poor sensitivity of the diagnostic kit combined with the likelihood of asymptomatic human carriage and environmental contamination, this study does not answer the question of how much CDI is hospital-acquired. It does, however, suggest that horizontal transmission from known symptomatic cases may be less common that we thought.

Article citation: Eyre DW, Cule ML, Wilson DJ et al. Diverse sources of C. difficile infection identified on whole-genome sequencing. N Engl J Med 2013; 369: 1195-1205.

Photo credit: Annie Cavanagh. Wellcome Images.