BGPT: Paper Review: Long-term impact of the SARS-CoV-2 pandemic on respiratory viruses in Germany

Explore by Goal

Quick Explanation Copied

Core finding (paper claims): Using long-term German multiplex surveillance (CVN; 2010–2024), the authors report large, virus-specific disruptions in seasonality after the SARS‑CoV‑2 pandemic onset—largest for RSV and influenza—followed by partial “regaining” of pre-pandemic patterns as NPIs waned, while coinfection patterns were comparatively stable.

Long Explanation

Visual paper review: “Long-term impact of the SARS‑CoV‑2 pandemic on respiratory viruses in Germany”

BMC Public Health — DOI: 10.1186/s12889-025-23983-8

What the authors did (pipeline)

Data source & scope: CVN multiplex clinical virology testing in Germany, covering 2010–2024 and 18 respiratory viruses (after filtering to single-virus tests).
Cleaning / filtering: remove biologically implausible records and problematic subtype labeling; remove combitests; exclude non-German sites; exclude specific HBoV positivity cluster and mislabelled influenza coinfection entries.
Seasonality disruption metric: compute monthly smoothed counts, compare against baseline from past 10 years using a log2-based disruption statistic (DS), then compute SDI as the RMS deviation of DS from zero disruption within chosen time windows.
Cross-virus comparison: cluster viruses by disruption profiles (hierarchical clustering with average linkage using squared Euclidean distance).
NPIs relationship: correlate infection counts (Spearman and Pearson checks; absolute vs relative counts) with Oxford Stringency Index over Jan 2020–Dec 2022; analysis is monthly aggregation.
Coinfection analysis: compare coinfection counts pre-pandemic vs post-pandemic onset; due to smaller post-2020 sample sizes, they do not use the more inferential pairwise “increase/decrease vs chance” method from their prior work.

Data integrity & filtering (key statistical landmines)

Because the SDI and correlations are computed from positivity-based surveillance counts, preprocessing decisions (and testing-policy variability) can materially change conclusions. Below is exactly what the manuscript reports they excluded.

Core results (as stated)

1) Seasonality disruption is virus-specific.

The authors report the largest disruptions for viruses with strong pre-pandemic seasonality—especially RSV and influenza, with RSV described as having the highest SDI among studied viruses.

2) Many viruses “regain” seasonality after early disruption.

Their conclusion frames pandemic disruption as significant initially, with a tendency toward return as measures loosen.

3) Coinfection patterns: “surprisingly similar” before vs after.

They report overall coinfection patterns remain broadly similar, with notable post-onset changes related to HBoV coinfections.

4) NPIs correlate (often negatively) with infection counts, nonlinearly.

They report many viruses show moderate negative Spearman correlations with stringency; Pearson correlations suggest nonlinearity and that the relationship may be nonlinear.

A skeptical interpretation (what must be true for these claims to hold)

Testing-policy confounding risk (measurement bias):

They explicitly acknowledge that testing policies vary over time and location, so neither absolute nor relative counts are bias-free. This matters because SDI is computed from counts and log-ratios vs baseline medians; biased sampling could create apparent “missing seasons” or shifts.

SDI hyperparameter sensitivity (algorithmic stability):

The SDI depends on smoothing window w, pseudocount alpha, and baseline seasons d. The manuscript states that changing these values “drastically” alters results and calls for parameter sensitivity across datasets as future work—so cross-study generalization is uncertain.

Correlation ≠ causal effect of NPIs:

The paper correlates stringency with infection counts and disruption statistics. But stringency is a composite proxy for many behavioral and policy changes, and it is likely correlated with other drivers (e.g., healthcare seeking, testing volumes, mobility), making causal attribution hard. The authors mention the composite nature of NPIs and the difficulty disentangling individual interventions.

Coinfection inferential limitation (smaller post-2020 samples):

They avoid their earlier “increase/decrease vs chance” statistical approach due to smaller post-2020 sample sizes, using before/after comparisons instead. That reduces the strength of evidence for changes in coinfection propensity.

Visual summaries built from the manuscript’s stated extracted numbers

These visuals do not replicate time-series figures (the raw month-by-virus values are not provided in your prompt), but they do summarize the study’s scale and the filtration logic that underpins SDI and correlation results.

Paper critique (evidence strength by claim type)

Claim category	What the paper reports	Evidence strength (skeptical)	Primary limitation(s) highlighted by authors
Seasonality disruption ranking	RSV shows highest SDI; influenza also highly disrupted; perennial viruses less so	Moderate-to-strong (quant metric defined, but depends on hyperparameters & counts bias)	Testing policy variability; SDI parameter dependence acknowledged
NPIs association	Many viruses show moderate negative Spearman correlations; Pearson suggests nonlinearity	Moderate (associations; causal direction not established)	Composite NPIs correlate with multiple factors; disentangling is difficult
Coinfection changes	Coinfection patterns pre vs post are “surprisingly similar”; HBoV-related coinfections increase	Weak-to-moderate (less formal statistics post-2020)	Reduced post-2020 sample sizes prevented their earlier inferential coinfection test; they use before/after count comparisons

What would most strongly DISPROVE or overturn the paper’s main conclusions?

Show that observed “missing peaks” or shifts are replicated by testing-volume/positivity detection artifacts rather than true transmission changes—i.e., after correcting for time-varying testing intensity and site mix, SDI collapses toward baseline. (The authors themselves highlight testing-policy bias as a threat.)
Demonstrate SDI ordering is not robust to SDI hyperparameters (w, alpha, d). The authors acknowledge parameter dependence can drastically alter results and propose systematic parameter sensitivity for future work.
Recompute coinfection statistics with a framework that accounts for post-2020 sampling differences (rather than only before/after comparisons) and show the HBoV-related increases disappear. The paper notes limited sample sizes constrained their inferential method.

Author reviews (bespoke BGPT links)

Open reviews by each listed full-name author.

Feedback:

Updated: April 24, 2026