Published drought and immunity models of US West Nile neuroinvasive disease fitted to 1999-2013 did not forecast 24 state counts in 2014-2025 better than simple baselines

Question: do published climate and immunity drivers of US West Nile virus neuroinvasive disease (WNND) forecast years that the original studies did not use? Approach: we wrote a test plan, then used CDC ArboNET state counts for 1999-2025, NOAA nClimDiv state climate and Census population. We kept the published state coefficients of Paull et al. (2017) for cumulative incidence (immunity), May-August PDSI (drought) and precipitation in 24 states, fitted only the intercepts to 1999-2013, and forecast 2014-2025 with the observed climate. We also refitted pooled negative binomial models with state fixed effects and tested the national temperature correlation of Hahn et al. (2015). Results: our state refits reproduced the published coefficients for 1999-2013. The published models had a skill of +0.44 (95% interval +0.32 to +0.53) against last-year persistence in 1999-2013, but -0.24 (-0.55 to +0.06) in 2014-2025 and -0.48 (-0.87 to -0.14) against a five-year mean. They predicted that mean cases would fall to 29% of the 1999-2013 mean; the observed value was 80%, and all 18 states with an immunity term had more cases than predicted. Without the immunity term, the models were equal to the baselines. The pooled drought coefficient was small in both periods (-0.049 and -0.033 per PDSI unit; intervals include zero; published -0.14), and refitted state drought slopes for 2014-2025 did not agree with the published slopes (Spearman 0.02). Observed national counts (mean 1321 per year) were above the published 95th-percentile projection (1331) in 8 of 12 years, although 2014-2025 was less dry. The national correlation with annual temperature decreased from r = 0.92 (2004-2012) to 0.28 (2013-2025). Limitations: state-level data, no temperature-R0 or freeze terms, observed-climate hindcasts, reporting changes, and wide intervals from 12 forward years.