Infectious Disease · Medical Literature
[Correspondence] Improving prediction of future Mycoplasma pneumoniae epidemics using data-driven transmission models
Abstract
Following the global re-emergence of Mycoplasma pneumoniae infections in 2023–24 after the COVID-19 pandemic,1 a key question is when, and to what extent, these infections will surge in the near future. This question is particularly relevant because the most recent epidemic placed a considerable burden on health-care systems worldwide.2 By extending our transmission modelling,2 we aimed to enable such epidemic predictions and advance recent modelling approaches.3 This advancement was based on the premise that short-term to medium-term forecasts can be improved by integrating data-driven transmission models, realistic reporting processes, and validation with contemporary surveillance data.
Clinical relevance
Can data-driven transmission models that integrate realistic reporting processes and are validated against contemporary surveillance data improve short-term to medium-term predictions of future Mycoplasma pneumoniae epidemic timing and magnitude? __SNAPSHOT__ {"studyDesign":"Correspondence (modelling study); extension of a previously published epidemiological and transmission modelling study using a Bayesian data-driven susceptible-infected-recovered framework","setting":"International; primary model validation performed using national surveillance data from Denmark (Statens Serum Institut, Copenhagen); affiliated institutions include the UK Health Security Agency, University Children's Hospital Zurich, and Cardiff Metropolitan University, among others within the ESGMAC MAPS Study Group","sampleSize":"Not reported in the abstract; surveillance data spanning January 1, 2011 to March 31, 2026 were used for model fitting and prospective validation in Denmark","population":"Population-level M. pneumoniae surveillance data from Denmark and the broader ESGMAC MAPS global network; no individual patient-level data reported in this correspondence","intervention":"An extended data-driven transmission model integrating: (1) a Bayesian mechanistic susceptible-infected-recovered framework, (2) explicit modelling of NPI effects on transmission and reporting, and (3) scenario-based behavioral assumptions post-2025 (return to pre-COVID-19 behavior vs. continuation of modified post-pandemic behavior)","comparator":"Compared against an alternative modelling approach (Nielsen and colleagues, PNAS 2025) that used a stochastic mechanistic model fit to Danish decadal surveillance data","primaryOutcome":"Accuracy and concordance of model-generated short-term to medium-term epidemic forecasts with observed M. pneumoniae detection data through March 31, 2026; secondary outcome was estimation of NPI-related increases in testing and reporting effort","followUp":"Model training data: January 1, 2011 – March 31, 2020 (pre-NPI baseline); NPI-period inference: April 1, 2020 – March 31, 2025; prospective validation window: April 1, 2025 – March 31, 2026","fundingSource":"Not reported in the abstract or correspondence text"}
Key findings
The extended Bayesian transmission model estimated that the increase in testing and reporting effort from the start of NPIs was 282% (95% credible interval [CrI]: 262–307%). Under both behavioral assumption scenarios (return to pre-COVID-19 population behavior from April 1, 2025, or continuation of modified post-pandemic behavior), the model consistently predicted low M. pneumoniae incidence over the next 1–2 years, followed by a return to pre-COVID-19 pandemic seasonal dynamics. Observed Danish surveillance data through March 31, 2026 (the sixth year following NPI implementation) corroborated these predictions, showing persistently low detection levels after the conclusion of the 2023–24 global epidemic. Importantly, the model produced substantially different short-term forecasts compared to the competing model of Nielsen and colleagues. Specific numeric forecasts (e.g., absolute predicted case counts, predicted epidemic peak timing, formal forecast accuracy metrics such as RMSE or interval coverage) are not reported in the abstract. No adverse events are applicable to this modelling correspondence. No NNT/NNH is calculable.
Practice implications
This correspondence establishes a proof-of-concept that integrating data-driven transmission models with explicit, quantitative adjustment for NPI-related changes in both transmission and surveillance reporting can produce epidemic forecasts for M. pneumoniae that are concordant with prospectively observed population-level data. The study demonstrates that the post-2023–24 epidemic period is likely to be characterized by low M. pneumoniae activity for at least 1–2 years regardless of whether population contact behavior returns immediately to pre-pandemic norms or remains partially modified—a clinically reassuring signal that may allow health systems some lead time before the next major wave. However, several important caveats apply. First, the model was validated using data from a single country (Denmark), which has historically robust surveillance infrastructure and well-characterized multi-year epidemic cycles; generalizability to settings with heterogeneous testing practices, varying macrolide resistance rates (notably >60–70% in parts of Asia), or less complete surveillance is uncertain. Second, neither individual patient-level outcomes (hospitalizations, ICU admissions, mortality) nor antimicrobial resistance trajectories were modeled. Third, the competing modelling group (Nielsen et al., PNAS 2025) reached a broadly concordant but quantitatively different forecast (moderate-to-high incidence predicted in the 2025–2026 season using a stochastic framework), underscoring that forecast uncertainty remains substantial and model choice materially affects predictions. Fourth, as this is a Correspondence, the full methodological reporting necessary for independent critical appraisal—including model structure equations, prior specifications, complete sensitivity analyses, and formal external validation metrics—is only partially accessible without the full-length paper. The 2023–24 re-emergence placed a considerable burden on health-care systems worldwide, including increased hospitalizations in adults in some countries. Clinicians and public health planners should treat current model-based forecasts of near-term low activity as probabilistic, not deterministic, and maintain surveillance readiness. __LIMITATIONS__ Correspondence format: the full model specification, parameter uncertainty analyses, and validation metrics are not completely reportable in this abbreviated format; this summary is based on abstract text only, which significantly limits critical appraisal,Single-country prospective validation: the model was validated against Danish surveillance data only; applicability to other national contexts with different surveillance infrastructure, demographic structures, or immune landscapes is unproven,Behavioral scenario uncertainty: future population contact behavior post-2025 is unobservable and must be assumed; neither scenario may accurately reflect real-world mixing patterns,Reporting artifact: the estimated 282% increase in testing effort during the NPI period is a key latent variable that, if mis-specified, could substantially bias transmission parameter estimates and downstream forecasts,No patient-level clinical outcomes modeled: the framework cannot predict severity, hospitalization rates, ICU demand, or mortality—the outcomes most relevant to health-system planning,Macrolide resistance not incorporated: growing macrolide resistance (particularly in East Asia and the Americas) may alter the clinical impact of future epidemics in ways not captured by this transmission-only model,Stochasticity: M. pneumoniae dynamics are known to be noise-driven and exhibit mode-hopping between annual and multiannual cycles, which limits the reliability of medium-term point forecasts even from well-calibrated models,Discordance with competing model: the Nielsen et al. stochastic model predicted moderate-to-high incidence in 2025–26, directly contradicting the low near-term incidence forecast of this model, reflecting unresolved methodological uncertainty in the field,Funding source not reported __BOTTOM_LINE_CATEGORY__ promising_but_preliminary __BOTTOM_LINE__ This correspondence presents a data-driven Bayesian transmission model that, when explicitly adjusted for NPI-related changes in both transmission and surveillance reporting effort (estimated at +282%, 95% CrI 262–307%), accurately forecasted the low M. pneumoniae detection observed in Denmark through March 2026 following the 2023–24 global epidemic. Both behavioral scenarios modeled predict a continued 1–2 year period of low activity before return to pre-pandemic cyclical dynamics, though this conflicts with a competing stochastic model predicting moderate-to-high activity in 2025–26. Clinicians and infection control planners should monitor ongoing surveillance closely while interpreting these forecasts as probabilistic estimates subject to substantial stochastic and behavioral uncertainty.