Escrever um comentário

Major Issues

1. The word “transmission” should be used cautiously throughout. The clusters are inferred from genetic relatedness and collection dates, not observed transmission chains. This is reasonable, but the paper sometimes moves from “clustered infections” to “nosocomial transmission” quite strongly. The authors should consistently state that these are inferred nosocomial transmission clusters and that alternative explanations, such as repeated exposure to a persistent environmental reservoir or repeated introductions of highly prevalent clones, cannot always be separated.

2. Single-linkage clustering can create large chained clusters. The authors clearly explain that not all pairs within a cluster need to be within the genetic or temporal threshold. This matters especially for very large clusters, such as the ST307 cluster involving 188 neonates. The manuscript would be strengthened by reporting within-cluster maximum pairwise genetic distance, maximum temporal span, and perhaps a sensitivity analysis using complete-linkage or stricter cluster definitions for the largest clusters.

3. Sampling and denominator differences could bias site comparisons. The included studies differ in duration, culture practices, isolate storage/revival, sequencing success, clinical inclusion criteria, and whether CSF isolates were included. These factors can affect the probability of detecting clusters and the denominator of sequenced K. pneumoniae infections. The authors discuss this, but should more explicitly caution against ranking sites or regions by transmission burden without standardized sampling intensity.

4. Transmission proportion may be both conservative and biased in complex ways. Excluding one index case per cluster is conservative, but missed cases, colonization-only transmission, repeat isolates, and variation in blood-culture sensitivity could alter estimates in either direction. The paper should avoid presenting 57.7% as a precise attributable fraction and instead frame it as a lower-bound estimate under explicit assumptions.

5. Facility-level associations are underpowered and should be treated as exploratory. The observations about piped water availability and neonatal surgical facilities are important, but they are based on limited, post hoc facility-level data and small numbers of sites. These findings should be described as hypothesis-generating rather than evidence of specific determinants.

6. AMR-transmission associations are difficult to interpret causally. ESBL and carbapenemase genes are associated with clustered infections, but this could reflect lineage structure, treatment selection, blood-culture detection probability, hospital persistence, or true transmissibility. The authors discuss this well, but the abstract and conclusions should avoid implying that resistance itself directly increases transmission without qualification.

7. Clinical metadata limitations constrain interpretation. The absence of standardized admission dates, birth dates, disease-onset timing, ward movement, bed location, colonization screening, maternal isolates, environmental isolates, and IPC exposure data prevents resolution of transmission routes. The discussion should more clearly separate what this study can estimate, clustered invasive disease burden, from what it cannot estimate, exact route or source.

Minor Issues

1. Page 1 appears crowded, with the affiliation block overlapping the medRxiv footer in the rendered PDF. The title page should be reformatted.

2. Figures 1-3 are very informative but dense. Larger fonts or simplified main panels with detailed tables moved to supplements would improve readability.

3. Clarify early in the Results that the 68.0% and 57.7% are overall isolate-level estimates, while 53.3% and 33.5% are median site-level estimates.

4. Define “introduction” clearly when used for clusters and singleton cases, since it is an analytical unit rather than a directly observed importation event.

5. Consider presenting a short table of the largest clusters, including ST, site, duration, number of cases, and maximum pairwise SNV distance.

6. The manuscript should clarify how missing patient identifiers in the MLW Biobank may affect repeat-isolate exclusion and cluster size estimates.

7. The discussion of WHO empirical therapy implications is important, but it should note that genotype-predicted resistance does not fully replace phenotypic susceptibility testing.

8. Please ensure all supplementary tables and figures are easy to locate from the main text, since many key assumptions are in supplements.

Competing interests

The author declares that they have no competing interests.

Use of Artificial Intelligence (AI)

The author declares that they did not use generative AI to come up with new ideas for their review.