Reply: Non-transparent and insufficient descriptions of non-validated microbiome methods and related reproductive outcome results should be interpreted with caution.

Sir,

We thank Haahr et al. for their critical notifications towards our recent publication. We would like to elaborate on our study and on the points raised:

Concerning their first comment that the IS-pro technique has not been externally validated, we would like to point out that a brief PubMed search reveals a series of papers validating all aspects of the IS-pro technique (Budding ; Budding ; Daniels ; Budding ; Eck ; Eck ) and highlighting its applicability as a highly reproducible assay that can be applied to clinical diagnostics. This is in contrast to current sequencing approaches, which do not have the inter-hospital reproducibility required for clinical diagnostics (Lozupone ).

Furthermore, as can be read in our protocol paper (Koedooder ), we performed the analyses with both 16s rRNA gene sequencing and the IS-pro technique. The manuscript that compares these analyses with each other is in preparation, but the scope will be that the results yielded very similar vaginal microbiome profiles, with a median Pearson’s R2 of 0.97. This indicates a high level of similarity between 16s rRNA gene sequencing and IS-pro results.

We could have described the external validation group in more detail. However, restrictions in manuscript length forced us to make choices in what information to include. The baseline characteristics of the original study group showed no differences between women with a favourable and unfavourable profile. Moreover, the treatment in Germany is comparable to the treatment in the Netherlands.

Regarding the comments raised concerning the community state types (CSTs), it is important to note that there are no formal definitions of the CSTs, and true beta-diversity in vaginal samples is larger than what can be captured by the artificial classification in five CSTs. This is reflected in Fig. 1 of our paper. We decided to include the CST classification merely to provide easier insight into the data set, and CST classification is no part of our predictive algorithm.

Regarding the remark that `microbiome methods do not sufficiently take into account the total abundance of bacteria’, we wonder what `microbiome methods’ are referred to here. Moreover, we wonder what would be `sufficient’ for what purposes and what `misclassification’ is referred to. Some clarity in these important points would be essential to address them properly.

Concerning IS-pro, we would like to point out that this technique is semi-quantitative and certainly reflects the total abundance of bacteria. However, as Haahr et al. point out themselves, as we are comparing `relative’ abundances, total load is not important for this goal.

Concerning Fig. 3, we were aware of the error in Fig. 3, provided the correct version of Fig. 3, and in the meantime the journal has decided to publish a correction in the form of a corrigendum.

While this study was indeed set up as a prospective cohort study, we did use the pregnancy outcomes to optimize the cut-off levels of the predictive algorithm.

Finally, we fully agree with Haahr et al. that it is of the utmost importance to use validated and repeatable diagnostics. Therefore, we used the only tool that has—to our knowledge—met these standards (Budding ; Budding ; Daniels ; Budding ; Eck ; Eck ). Outside the published validations, the IS-pro assay has been Conformité Européenne/in-vitrodiagnostics (CE/IVD) marked to meet the highest international demands for standardization in clinical diagnostics.

Conflict of interest

None.