Effects and predictive performance of multilayer environmental exposures on coccidioidomycosis: a longitudinal surveillance study
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Coccidioidomycosis (Valley fever) is a soilborne mycosis endemic to the US Southwest whose incidence has increased markedly in recent decades. Environmental conditions are thought to influence the soil-dwelling lifecycle of Coccidioides ; however, most prior studies have relied on above-ground meteorological conditions–primarily precipitation and air temperature (AT)–with few examining subsurface soil moisture (SM) and soil temperature (ST), which may more directly influence the fungal lifecycle. No study of coccidioidomycosis, or any other environment-sensitive soilborne mycosis, has examined how deeper-layer soil conditions relate to disease incidence, despite the prevailing soil-sterilisation hypothesis implicating deeper soil as a potential fungal refugium. Furthermore, nonlinear exposure–lag–response relationships for key dust-dispersion exposures–including PM 10 , a potential proxy for airborne spore concentration, and wind speed–remain uncharacterised. We aimed to estimate and compare the associations between coccidioidomycosis incidence and environmental exposures across multiple above- and below-ground layers, and to evaluate their independent and combined predictive performance. This ecological time-series study analysed 185,486 reported cases of coccidioidomycosis in Arizona’s hyperendemic tri-county region (Maricopa, Pima, and Pinal) during 1997–2024. We developed a mechanism-informed multilayer environmental framework comprising one dust-dispersion layer (PM 10 , wind speed) and four soil–climate layers–meteorological (precipitation, AT), topsoil (0–10 cm), midsoil (10–40 cm), and deepsoil (40–100 cm) SM and ST. We fitted distributed lag non-linear models (DLNMs) with season-specific interaction terms to estimate exposure–lag–response associations between each environmental layer and coccidioidomycosis incidence. We then developed a two-stage stacked ensemble machine learning framework to assess each layer’s independent predictive performance (stage 1) and integrate them into a unified forecast (stage 2), which was evaluated using a strictly held-out test period. At concurrent lags (1–3 months prior to reporting), coccidioidomycosis incidence was primarily associated with dustier, windier, and drier conditions, cooler air temperatures, and warmer topsoil. For each IQR increase, PM 10 showed the most consistent concurrent associations, with significant positive incidence rate ratios (IRRs) across all four incidence seasons at lags 1–2 (ranging from 1.04 [95% CI 1.00–1.08] to 1.37 [1.26–1.48]). Across lags 1–36 months, all four soil–climate layers exhibited nonlinear, non-monotonic, and season-dependent associations with coccidioidomycosis incidence, characterised by alternating wet–dry and cool–warm oscillations. Topsoil displayed the most frequent significant associations, with moisture–temperature signals attenuating progressively from topsoil through midsoil to deepsoil. A depth-dependent lag structure was observed for both moisture and temperature, in which significant positive IRRs emerged at progressively shorter lags with increasing soil depth, accompanied by vertical divergence across depths at the same lag windows. For example, for fall incidence, positive moisture IRRs appeared at precipitation lag 15 (1.10 [1.00–1.21]), topsoil SM lag 9 (1.22 [1.15–1.30]), midsoil SM lags 8–9 (up to 1.23 [1.11–1.37]), and deepsoil SM lags 4–5 (up to 1.08 [1.01–1.16]); at these same lags, deepsoil SM was positively associated with incidence whereas topsoil SM and precipitation remained negatively associated. During the held-out test period (2021–2024), the multilayer ensemble generally captured seasonal and interannual variation well, including the timing and approximate magnitude of most peaks, outperforming all single-layer models. Although individual layers had slightly lower test RMSEs, their test gap ratios were substantially higher (0.27–0.67 vs 0.00), indicating that the ensemble generalised far more reliably. All five environmental layers contributed to the final ensemble forecast; the dust-dispersion and topsoil layers received the highest importance, with PM 10 ranked as the most important predictor group. The best-performing of four pipeline configurations relied solely on environmental inputs available within one week, enabling the model to function as a near-real-time nowcast. This study provides the first evidence linking multilayer environmental exposures to coccidioidomycosis incidence across both temporal and vertical dimensions, offering new quantitative support for the prevailing soil-sterilisation and grow-and-blow hypotheses and demonstrating that a multilayer framework could improve both mechanistic understanding and predictive performance. The framework could be generalised to other endemic settings and readily extended with new data and methods to inform surveillance and public health preparedness. These findings support incorporating multilayer lagged environmental exposures into both effect estimation and forecasting systems to better prepare endemic regions for anticipated warming, drying, and increasingly variable climatic conditions.
Research in context
Evidence before this study
We searched PubMed, Scopus, and Google Scholar for literature published in English from database inception to July 17, 2026. The first search combined environmental exposure terms (“environment*” OR “climat*” OR “meteorolog*” OR “weather” OR “soil moisture” OR “temperature” OR “precipitation” OR “rain” OR “humidity” OR “moisture” OR “drought” OR “wind” OR “dust” OR “particulate matter” OR “PM10” OR “PM2.5”) with coccidioidomycosis terms (“coccidioidomycosis” OR “Valley fever” OR “ Coccidioides ”). The second search used the same environmental terms combined with broader soilborne mycosis terms (“soilborne mycosis” OR “soilborne mycoses” OR “endemic mycosis” OR “endemic mycoses” OR “histoplasmosis” OR “blastomycosis” OR “paracoccidioidomycosis” OR “coccidioidomycosis” OR “Valley fever” OR “ Coccidioides ”). Prior studies have generally supported the theory that antecedent alternating wet–dry and cool–warm periods increase coccidioidomycosis incidence, yet most relied on above-ground meteorological variables—primarily precipitation and air temperature—that might not adequately reflect subsurface conditions where the fungus grows. A few studies incorporated soil moisture data, but relied on correlation or univariate analyses without adjusting for potential confounders. Our recent study was the first to assess topsoil (0–10 cm) moisture and temperature effects on coccidioidomycosis incidence within a multivariable framework, but was restricted to topsoil layer and linear modelling, leaving nonlinear exposure–lag–response relationships and the potential role of deeper soil layers unexplored. Critically, no study in coccidioidomycosis—or any other environment-sensitive soilborne mycosis—has examined how deeper-layer (>10 cm) soil conditions relate to disease incidence, even though the soil-sterilisation hypothesis suggests that deeper soils might serve as fungal refugia. Nor has any study characterised the nonlinear exposure–lag–response relationship between PM 10 , a potential proxy for airborne spore concentration, and coccidioidomycosis incidence. Finally, no study has evaluated or compared the effects and predictive performance of multilayer environmental exposures for coccidioidomycosis or other environmentally sensitive soilborne mycoses.
Added value of this study
To our knowledge, this is the first study to use a comprehensive multilayer environmental framework—spanning one dust-dispersion and four soil–climate layers (meteorological, topsoil, midsoil, and deepsoil)—to any soilborne mycosis. Within this framework, we used DLNMs and a novel two-stage stacked ensemble approach to assess, for the first time, both the associations and predictive performance of environmental exposures across all five layers. This enabled the first characterisation of nonlinear exposure–lag–response relationships for PM 10 and subsurface SM and ST in coccidioidomycosis research. We found that increased incidence was generally associated with concurrent dustier, windier, and drier conditions, cooler air temperatures, and warmer topsoil, preceded by alternating wet–dry and cool–warm oscillations across layers. PM 10 exhibited consistent positive associations with incidence at concurrent lags across all four seasons and emerged as the most important predictor in the ensemble forecast, jointly providing the first support for the recently proposed dust-borne atmospheric transport hypothesis. The multilayer design extended evidence of alternating wet–dry and cool–warm cycles to midsoil and deepsoil for the first time, although this cyclical signal was most pronounced in the topsoil and attenuated progressively with depth. Across moisture and temperature variables, we identified depth-dependent lag structures in which significant positive associations appeared at progressively shorter lags with increasing soil depth, accompanied by vertical divergence across layers at the same lag windows—providing the first quantitative evidence in the vertical dimension for both the dominant soil-sterilisation and grow-and-blow hypotheses. These patterns suggest that deeper soils might function as a buffered subsurface refugium for Coccidioides , preserving favourable moisture and thermal conditions longer than shallower layers. Our two-stage ensemble framework demonstrated that combining multilayer environmental information achieved superior prediction over any single-layer approach. The selected pipeline—relying solely on environmental data available within one week of the target month—could serve as a near-real-time nowcast, generating incidence estimates well before finalised surveillance data become available. Our ensemble framework offers a flexible, modular architecture that can be readily extended with new predictors and candidate models to further refine forecasting performance. Collectively, these results offer the first evidence connecting multilayer environmental exposures to coccidioidomycosis incidence across both temporal and vertical dimensions, provide new quantitative support for the prevailing mechanistic hypotheses, and show that a multilayer approach could enhance both mechanistic understanding and predictive performance.
Implications of all the available evidence
Our results suggest that coccidioidomycosis dynamics might be associated with complex, depth-stratified hydroclimatic cycles that above-ground environmental data alone cannot fully capture, highlighting the potential value of incorporating subsurface soil data into environmental health studies of soilborne mycoses more broadly. Although the best-performing candidate models, the predictive performance of individual layers, and their relative contributions within the ensemble might vary across endemic settings, the approach used by most existing forecasting studies—relying on above-ground meteorological or dust-related variables alone—is unlikely to be sufficient. The primary contribution of this work might lie less in any particular set of candidate models or predictors than in the multilayer framework itself, which provides a modular and extensible architecture for integrating heterogeneous environmental information across vertical and temporal dimensions. The selected prediction pipeline—relying solely on environmental inputs available within one week—could function as a near-real-time nowcast, generating incidence estimates well before many contemporaneously exposed patients were diagnosed given the prolonged diagnostic pathway for coccidioidomycosis. Nowcast-identified high-incidence periods could support public health preparedness by prompting earlier clinical consideration and targeted patient counselling—particularly as patients with prior awareness of coccidioidomycosis have been shown to be diagnosed substantially earlier and to seek testing more proactively. Importantly, the modular architecture is readily extensible with forecast-derived predictors, enabling a shift from nowcasting to prospective early-warning forecasting. The evidence, including findings from this study, suggests that anticipated climatic changes in the southwestern USA—including intensifying drought, continued warming, and potentially increasing dust emissions—might escalate coccidioidomycosis burden and expand its endemic range, underscoring the need for improved surveillance and forecasting tools. Future efforts in coccidioidomycosis surveillance, effect estimation, and prediction might benefit from adopting and refining this multilayer framework and evaluating its applicability in other endemic settings and, potentially, in other environment-sensitive soilborne mycoses.
