Sampling composition and sequencing effort in the detection of within-hospital Gram-negative genomic clustering: a prospective observational study at two tertiary-care hospitals

Background. Genomic surveillance of Gram-negative pathogens is generally selective, focusing on suspected outbreaks or resistant isolates. The extent of what such programmes miss remains unknown. It has not been quantified in a large, unselected dataset. Methods. We prospectively whole-genome sequenced isolates of Escherichia coli, Klebsiella pneumoniae, and Pseudomonas aeruginosa from two tertiary-care hospitals in Copenhagen and Hannover, regardless of resistance phenotype, specimen type, or ward. We analysed 25,888 genomes from 13,712 patients. Clonal clusters were defined by single-linkage clustering at [≤]30 SNPs, based on within-patient distances and validated against inpatient movement records. Analyses were patient-based, with hospital comparisons using matched 18-month periods. Results. Clonal clusters included 14.1% of patients in Copenhagen and 16.5% in Hannover. Most clustered patients had no multidrug-resistant isolate (71.3% and 50.0%) and had no intensive care unit (ICU)-associated isolate (86.1% and 68.7%). Sequencing blood-culture isolates alone, the usual trigger for genomic investigation, would have identified only 14 of 900 clustered patients in Hannover (1.6%). Excluding colonisation screening would have missed 240 (26.7%). No targeted strategy detected more clustered patients than random sampling of the same number of isolates. Detection increased with sequencing effort without reaching a plateau. After matching periods and adjusting for ICU association, resistance, and specimen mix, clustering remained more common in Hannover (adjusted odds ratio 1.49, 95% CI 1.28-1.73), suggesting sampling explained much, but not all, of the difference. Conclusions. Most Gram-negative clusters involved non-MDR organisms on general wards and would have been missed by surveillance focused on resistance or intensive care. What is selected determines which patients can be detected; the number of isolates sequenced determines how many are found. The observed 14-16% is a lower bound. Surveillance should sample broadly and representatively, and report what was sequenced and how much.