Malaria trends in Ethiopian highlands track the 2000 'slowdown' in global warming

The study investigates whether the reversal from increasing to decreasing malaria incidence in the Ethiopian highlands around 2000 is linked to climate forcing—specifically a transient slowdown in global warming and associated variability (ENSO, PDO)—rather than solely to enhanced public health interventions. Highlands, with temperature-limited transmission and low immunity, are expected to be sensitive to warming. After increases from the 1970s–1990s, incidences declined in the 2000s, challenging the role of climate. The authors hypothesize that regional manifestations of the global warming 'slowdown' (circa 1998–2012) and shifts in climate variability reduced malaria transmission potential, preceding and facilitating later control measures.

Debate exists on climate change’s role in malaria in East African highlands, with prior work showing warming-associated increases through the 1990s, and subsequent declines in the 2000s raising questions about climate versus control. The 'global warming hiatus' sparked extensive climate literature attributing the apparent slowdown to oceanic heat redistribution, ENSO/PDO phases, volcanic forcing, and observational biases. ENSO and PDO modulate East African climate via ITCZ shifts, affecting rainfall and temperatures. Prior studies have emphasized rainfall changes during El Niño and long rains' downward trends; however, in highlands, temperature is a dominant constraint on transmission. Literature also discusses teleconnections among Pacific, Indian, and Atlantic basins, with Indian Ocean warming influencing Pacific variability and East African climate. Alternative malaria drivers (drug resistance, reporting changes) have been considered; vivax–falciparum co-variation argues against drug resistance as a sole cause of trends.

Data: Monthly confirmed Plasmodium falciparum (Pf) and P. vivax (Pv) cases from Debre Zeit (Oromia, Ethiopia) 1968–2007; surveillance aggregated consistently across 159 kebeles. Climate: Monthly Tmin, Tmax, and precipitation composites from 11 nearby stations (DZ-reg) and, for comparison, all 24 Oromia stations (ORO). Global/regional climate datasets: CRU TS4.01, NCEP/NCAR reanalysis; Niño3.4 (ENSO) and PDO indices. Analyses: (1) Singular Spectrum Analysis (SSA; embedding M=40) to decompose malaria and climate series into seasonal (HF ≤1 year), interannual (IA 1–10 years), and low-frequency (LF ≥10 years) orthogonal components; statistical significance tested against white/red-noise null models. Spectral characterization via MEM and MTM. (2) Scale-Dependent Correlation (SDC) spatial analysis to identify transient, lead–lag correlations between malaria and global SST anomalies during 1997/98 peak, mapping coherent regions (ENSO/PDO boxes, Indian Ocean). (3) Process-based stochastic transmission model (SEIQ with phenomenological vector delay): human compartments S, E, I, asymptomatic Q; a gamma-distributed delay (chain of m=2 stages) represents vector/parasite extrinsic effects in the force of infection. Transmission rate β(t) includes flexible seasonality (periodic B-splines), interannual temperature covariates (TEMP1: Feb–May; TEMP2: Jun–Sep) with a 17°C threshold, and environmental gamma noise. Reporting modeled via negative binomial with reporting rate and overdispersion. Parameter inference by maximum likelihood via iterated filtering (MIF) using R pomp, fitted to Pf cases 1980–1999; initial states for prediction obtained from the particle filter. Out-of-fit predictions generated for 2000 onward under counterfactual of no enhanced post-2004/2005 interventions, driven solely by observed temperature forcing. Robustness: alternative model with fixed 12-day human incubation; Tmin-based covariate models also examined. (4) Atmospheric modeling: AMIP-style CAM5 simulations with detrended SSTs retaining interannual/decadal variability; 1880s radiative forcing, repeating sea ice climatology, historical/CMIP5 GHG/ozone/aerosols. Correlated land/sea temperatures with Niño3.4 and PDO for 1979–2000 vs 2001–2016 to assess changes in teleconnections around 2000.

- Long-term covariation: LF components of Tmin and malaria (Pf and Pv) show highly concordant trends with a pronounced increase through the 1990s and a decline starting around 2000; correlations between long-term temperature variability and malaria exceed r>0.95 (p<0.01). - Seasonal dynamics: Seasonal components of Tmin, rainfall, and malaria covary, with a distinct dip in seasonal amplitude around 2000; the envelope (amplitude modulation) of malaria seasonality tracks Tmin LF trend (all SSA components significant at p<0.005 or p<0.001). - Interannual variability: IA components of Pf, Pv, Tmin, and PDO co-amplify in the late 1980s–1990s, then contract post-2000. Strong El Niño events (1997/98, 2002, 2006) align with peaks in regional climate anomalies and malaria. - Transmission modeling: A temperature-driven stochastic SEIQ model fitted to 1980–1999 Pf data predicts a reversal and decline in epidemic size post-2000 under counterfactual constant (pre-2004) control conditions; predicted mean monthly cases closely match observations up to 2004, diverging thereafter consistent with enhanced interventions. Results are robust to fixing a longer human incubation period (12 days) and to Tmin-based covariates. - Teleconnections: SDC maps during 1997/98 show significant positive correlations between malaria and SST anomalies in canonical ENSO (Niño3.4) and PDO regions, indicating synergistic warm phases impacting Ethiopia. Similar patterns appear with 2 m air temperature anomalies. - Changing climate controls: CAM5 AMIP correlations indicate a weakening ENSO influence and a strengthening PDO influence on East African climate around the 2000s transition, consistent with observational time-series analyses. - Alternative explanations: Parallel Pf and Pv responses to temperature trends argue against drug resistance (relevant only to Pf historically) as the driver of the turnaround; accounting for population and surveillance minimizes reporting biases.

Findings support a strong, multi-scale coupling between climate variability and malaria dynamics in Ethiopian highlands. The transient global warming slowdown and associated shifts in ENSO/PDO altered regional temperatures, reducing transmission suitability around 2000. Temperature-driven model predictions match the observed post-2000 decline before intensification of control, implying climate changes preceded and facilitated intervention impacts, preventing 1990s-scale epidemics. Seasonal, interannual, and decadal components of malaria coevolve with Tmin and large-scale ocean–atmosphere modes, establishing a coherent chain from global drivers to regional climate to local disease. Alternative hypotheses (drug resistance, reporting expansion) are inconsistent with the parallel Pf–Pv patterns and modeling framework incorporating population changes. The observed weakening of ENSO and strengthening of PDO teleconnections post-2000 in atmospheric simulations aligns with the epidemiological shift. These results highlight the need to integrate climate information into public health planning in unstable transmission settings to avoid complacency during climate-favorable periods and to anticipate reversals as climate conditions change.

The study demonstrates that malaria incidence trends in the Ethiopian highlands closely tracked a transient slowdown in regional warming around 2000, itself linked to global climate variability (ENSO, PDO). A temperature-driven transmission model reproduces the observed decline under constant pre-2004 control, indicating climate facilitated and acted synergistically with later enhanced interventions. The work underscores climate’s persistent, multi-timescale influence on epidemic-prone highlands and the importance of considering environmental context in malaria control and elimination strategies. Future research should: (i) extend analyses to other East African highlands (e.g., Kenya) with long-term epidemiological records; (ii) develop explicit models for P. vivax with relapse dynamics; (iii) incorporate spatial heterogeneity (elevation gradients) into transmission models; (iv) improve climate–disease attribution with longer, curated climate and entomological datasets in endemic regions; and (v) further elucidate evolving teleconnections among Pacific, Indian, and Atlantic basins affecting East African climate.

- Data constraints: Sparse, inconsistent African climate station coverage, particularly in earlier decades; reliance on composites and reanalysis/model simulations to address gaps. - Modeling scope: Transmission model parameterized for P. falciparum only; P. vivax relapse dynamics not explicitly modeled. Spatial heterogeneity (elevation) not included; cases aggregated regionally. - Attribution windows: Limited time span reduces confidence in firm conclusions about decadal PDO impacts post-2000. - Surveillance variation: Some uncertainty in surveillance coverage before 2004, though population adjustments and inclusion of neighboring facilities mitigate biases. - Climate modes: Indian Ocean Dipole roles at higher frequency were not analyzed in depth for this question; focus placed on ENSO/PDO. - Counterfactual assumption: Predictions assume unchanged pre-2004/2005 control levels; unobserved changes in local practices could contribute residual effects.