Recent and forecasted increases in coccidioidomycosis incidence in California linked to hydroclimatic swings

Abstract

Coccidioidomycosis, or Valley fever, is an infectious disease caused by inhalation of Coccidioides spp., fungi found primarily in soils of the southwestern United States. Prior work showed that coccidioidomycosis cases in California sharply increase by nearly 2-fold following wet winters that occur one- and two-years following drought. Statewide drought between 2020-2022 followed by heavy precipitation during the 2022-2023 winter raised concerns over potential increases in coccidioidomycosis cases in the fall of 2023, prompting California Department of Public Health (CDPH) to issue public health alerts. As anticipated, California saw a near record number of cases in 2023, with 9,054 provisional cases reported. During the 2023-2024 California wet season, precipitation was 115% the long-term average, furthering concerns about continued high coccidioidomycosis risk. We developed an ensemble model to forecast coccidioidomycosis cases in California in 2024-2025. Using this model, we predicted a total of 11,846 cases (90% PI: 10,056-14,094) in California between April 1, 2023, and March 31, 2024, encompassing the preliminary state report of 10,593. Our model forecasted 12,244 cases statewide between April 1, 2024, and March 31, 2025 - a 62% increase over the cases reported during the same period two years prior, and on par with the high incidence seen in 2023. The Southern San Joaquin Valley (5,398 cases, 90% PI: 4,556-6,442), Southern Coast (3,322, 90% PI: 2,694-3,961), and Central Coast (1,207 cases, 90% PI: 867-1,585) regions are expected to see the largest number of infections. Our model forecasts that disease incidence will exhibit pronounced seasonality, particularly in endemic regions, with cases rising in June and peaking in November at 1,411 (90% PI: 815-2,172) cases statewide - 98% higher than the peak two years prior (714) and nearly as high as the peak in 2023 (1,462). Near-term forecasts have the potential to inform public health messaging to enhance provider and patient awareness, encourage risk reduction practices, and improve recognition and management of coccidioidomycosis.

Fig. 1.

Schematic of the progressive time-series cross-validation approach used to generate weights for the ensemble model. Each progressive fold added an additional year to the training data (green) and predicted the following test year (beige). For example, we fit the models to data from 2000–2011 and predicted cases in 2012. We then fit the model to data from 2000–2012 and predicted cases in 2013, and so on. Individual models were then weighted by the inverse of their out-of-sample error across test years and used to forecast cases from Jan. 1^st, 2023 – Mar. 31^st, 2025 (purple).

Fig. 2.

Statewide monthly coccidioidomycosis cases, 2000–2024. Black dots indicate confirmed cases reported between 2000–2022, x’s indicate the provisional cases reported between Jan. 1, 2023, and Jul. 31, 2024, the green line represents the ensemble model fit to the observed case data (R² = 0.87), and the purple line indicates the ensemble model predicted (Apr. 1, 2023 – Mar. 31, 2024) and forecasted (Apr. 1, 2024 – Mar. 31, 2025) cases between Apr. 1, 2023, and Mar. 31, 2025. 90% prediction intervals are represented by the colored bands.

Fig. 3.

Regional monthly coccidioidomycosis cases, 2015–2024. Black dots indicate confirmed cases reported between 2015–2022, the green line represents the ensemble model fit to the observed case data, and the purple line indicates the ensemble model predicted (Apr. 1, 2023 – Mar. 31, 2024) and forecasted (Apr. 1, 2024 – Mar. 31, 2025) cases between Jan. 1, 2023, and Mar. 31, 2025. 90% prediction intervals are represented by the colored bands.

Fig. 4.

Forecasted regional monthly coccidioidomycosis cases between January 1, 2023, and March 31, 2025, under varying future climates in 2024–2025. The baseline climate scenario represents the 50^th percentile of precipitation between 1981–2023 and extrapolated monthly average temperatures assuming a 42-year linear trend. The 20^th and 80^th percentile precipitation scenarios assume the baseline temperature scenario, and the 3°F warmer or cooler temperature scenarios assume the baseline precipitation scenario (i.e., 50^th percentile).