Emerging pathogens associated with acute respiratory infections in children in Hanoi, Vietnam: an analysis of microbiology assay data from 2019 to 2023

We sincerely thank the reviewer for the thorough and constructive critique which has largely improved our manuscript. The reviewer's three comments collectively called for a more rigorous statistical framework to move beyond descriptive visual analysis. We have addressed all three concerns by implementing a comprehensive Interrupted Time Series (ITS) analysis using segmented Negative Binomial and Poisson regression models, with testing volume as an offset and seasonality controlled via Fourier terms. Below we respond to each comment individually and describe the specific analytical additions to the revised manuscript.

Reviewer Comment 1
"The analysis is currently limited to the visual inspection of plots. While descriptive, this approach lacks the statistical power to quantify changes in level, trend, or seasonality associated with specific intervention periods. Incorporating formal time-series modeling, including confidence intervals and hypothesis testing, is necessary to support causal interpretations."

Response: We thank the reviewer for this central methodological critique. We fully agree that formal statistical quantification is essential to support causal interpretation of the observed pathogen dynamics. In the revised manuscript, we have added a formal Interrupted Time Series (ITS) analysis for all pathogen groups.
Specifically, we implemented segmented Negative Binomial regression (glm.nb function of the “MASS” package) for Mycoplasma pneumoniae IgM) and bias-reduced Poisson regression (brglmFit method of the “brglm2” package) for bacterial PCR Panel 4, bacterial culture, and viral assays) to model monthly pathogen counts.
Model family was selected based on data characteristics. For Mycoplasma pneumoniae IgM serology, a Negative Binomial regression model (glm.nb function of the “MASS” package) was used, as monthly counts were consistently non-zero throughout the study period and confirmed overdispersion was present (estimated dispersion parameter θ ≈ 36; likelihood ratio test vs. Poisson: p < 0.001). For all remaining pathogens, bias-reduced Poisson regression (brglmFit method of the “brglm2” package) was used. These pathogens exhibited extended periods of near-zero counts during the COVID-19 suppression phase, causing the Negative Binomial likelihood to be undefined (glm.nb failed to converge due to an indeterminate dispersion parameter). Bias-reduced estimation prevents infinite coefficient estimates arising from complete or quasi-complete separation, which is a known consequence of sparse count data in segmented regression.
All models included:
(1) Phase indicator variables capturing immediate level changes at each epidemiological transition (pre-pandemic baseline → COVID-19 suppression → re-emergence/rebound);
(2) Slope terms capturing gradual trend changes during the re-emergence period;
(3) Fourier seasonality terms (sin and cos with a 12-month period) to analytically control for recurring annual cycles;
(4) Offset for testing volume: log(total tests performed) was included as an offset in all models, effectively standardizing all outcomes to positivity rates (see also our response to Comment 2).
Results are presented as Incidence Rate Ratios (IRRs) with 95% Wald confidence intervals and p-values in newly added Table 2. Observed vs. Expected plots generated from the models (newly added Figure 3, Figure 4, Figure 5, Figure 6) visually confirm the model fit and serve as counterfactual comparisons.
We have revised the manuscript section Methods, Results and Discussion accordingly to reflect these changes.

Reviewer Comment 2
"The study focuses on absolute counts of positive tests rather than standardized metrics, such as positivity rates adjusted for testing volume. Given that testing intensity fluctuated significantly during the study period, the current metrics risk misinterpretation. Supplementing the data with positivity rates or modeled expected counts would provide a more robust basis for inference."

Response: We agree that fluctuations in testing volume are a significant potential source of bias in surveillance data, particularly during the COVID-19 pandemic when testing patterns changed substantially. We have directly addressed this concern in two complementary ways:
(a) Regression offset for testing volume: In all ITS regression models, we included log(total tests) as an offset term. This mathematically transforms the outcome from absolute counts into a rate (positivity rate), ensuring that estimated IRRs reflect genuine changes in pathogen prevalence rather than artifacts of testing intensity fluctuations.
(b) Expected vs. Observed counterfactual plots: We generated model-based expected counts for each pathogen under a counterfactual scenario in which no pandemic-related changes occurred (i.e., all intervention indicator variables set to zero, preserving only baseline seasonality and testing volume). These plots (Figure 3, Figure 4, Figure 5, Figure 6) visually demonstrate the excess or deficit in observed cases relative to what would be expected based on 2019 baseline rates, adjusted for contemporaneous testing volume. The 95% confidence intervals around the expected counts further allow readers to assess whether observed deviations are statistically meaningful.
This dual approach provides robust, testing-volume-adjusted evidence that the observed pathogen re-emergence patterns represent genuine epidemiological shifts rather than consequences of increased testing activity.

Reviewer Comment 3
"Though some factors such as shifting admission policies, vaccination coverage, and viral variants are mentioned narratively, they are not controlled for analytically. Without adjusting for these time-varying confounders, it is difficult to definitively attribute pathogen 're-emergence' to COVID-19 interventions rather than external variables."

Response: We thank the reviewer for raising this important issue of confounding. We agree that controlling for time-varying confounders strengthens the causal interpretation of our findings. We have addressed this at two levels:
(1) Analytical control for seasonality: All ITS models include Fourier terms (sin and cos with 12-month period) that analytically capture recurring annual fluctuations in testing behavior, healthcare-seeking behavior, and pathogen biology. These seasonal terms serve as proxies for a range of annually recurring confounders including climate, school calendar cycles, and seasonal variation in hospital admission patterns.
(2) ITS design as a structural control for policy-related confounding: The segmented ITS design explicitly models distinct epidemiological phases corresponding to identifiable COVID-19 policy events in Vietnam: (i) pre-pandemic baseline (January 2019 to February 2020); (ii) COVID-19 suppression with social distancing and lockdown (March 2020 to March 2022); (iii) post-restriction re-emergence (from April 2022 onward). By modeling these policy-defined transitions as separate phases with distinct intercepts (level changes) and slopes (trend changes), the ITS approach inherently controls for the major time-varying contextual factors such as changes in hospital admission volume, dominant viral variant periods, and rolling out of COVID-19 vaccination that coincided with these policy transitions.
We acknowledge in the revised Discussion and Limitations that individual-level data on vaccination status and variant genotypes were not available, and that the single-center design may introduce unmeasured confounding related to hospital-specific admission policies. These are stated as explicit limitations. Nonetheless, the consistency of our findings across multiple independent pathogen groups (bacterial and viral), and their alignment with outbreak reports from Vietnamese health authorities and international literature, indicates that the observed patterns reflect real epidemiological phenomena rather than confounded effects.

The authors would like to thank the reviewer for his contributive comments which have helped improve our paper.
We would like to response to the comments below.

1. Thank you for pointing out redundancy error in the caption of Figure 1. There is repeated information that makes the caption of Figure 1 much longer than it is supposed to be. We will fix this. The description/ protocol for microbiological assays is already included in the Methods section (started with “The microbiology assays included in the analysis were:” …. .
These microbiological assays are done in Vinmec hospital for patients with corresponding indications (e.g. assays for Mycoplasma pneumoniae (MP) are indicated and done if the patient has signs/ symptoms suggesting MP infection and assays for MP are not routinely done for all patients with any acute respiratory infections).

2. Regarding the most appropriate public health measure(s) for governments to implement to contain future respiratory outbreaks: We believe that the measures to contain future respiratory outbreaks include maintaining availability and coverage of vaccines for outbreak-prone respiratory pathogens (e.g. maintaining Flu, HI, SP,… vaccination for community) alongside with maintaining public health interventions to reduce/ prevent exposure and transmission (e.g. mask, handwash,…) as well as improving host health (e.g. proper nutrition for children) before, during and after the pandemic.

3. The outbreaks shown by our data corresponds with actual outbreaks observed by authorities in Vietnam during the same time period. For example, the HAdV and Mycoplasma pneumoniae (MP) outbreaks correspond to the outbreak of HAdV and MP reported by authorities in Vietnam at the same time period (https://moh.gov.vn/hoat-dong-cua-dia-phuong/-/asset_publisher/gHbla8vOQDuS/content/cap-nhat-moi-nhat-ve-benh-virus-adeno-so-tre-mac-a-tang-len-1-406-co-7-ca-tu-vong ; https://baochinhphu.vn/bo-y-te-yeu-cau-phan-tich-dich-te-hoc-cac-ca-mac-va-tu-vong-do-virus-adeno-102220921182250996.htm ; https://benhviennhitrunguong.gov.vn/tre-mac-viem-phoi-do-mycoplasma-gia-tang-huong-dan-phong-benh.html ). We also previously published molecular data of HAdV and MP outbreaks (doi: 10.1371/journal.pntd.0011311 ;  doi: 10.3201/eid3005.231632). The Flu A and Flu B outbreaks observed in our data also correspond with the Flu outbreak reported by Vietnam authorities at the same time period (https://baochinhphu.vn/su-tro-lai-cua-dich-cum-sau-dai-dich-covid-19-102220831115300348.htm ; https://thainguyen.gov.vn/y-te-giao-duc/-/asset_publisher/L0n17VJXU23O/content/dich-cum-mua-co-dien-bien-bat-thuong-sau-covid-19?inheritRedirect=true ). 

4. Many thanks for sharing the two articles and thanks for your question. We keep archiving, observing our microbiological assay data over time as surveillance for potential future outbreaks of infections. We also store some samples for further molecular analysis in case of observed emerging outbreaks.

Putting all these together, we believe that our data helps anticipate the impact of emerging respiratory outbreaks and guide preparedness efforts.