This study reports findings from the largest documented cohort of primary amoebic meningoencephalitis, providing the first systematic analysis of host factors, inflammatory markers, and clinical outcomes under standardised contemporary management. Four principal findings emerge. First, diabetes mellitus is the dominant host risk factor for mortality—an association that held across a range of sensitivity analyses. Second, mechanical ventilation and early treatment appear to improve survival. Third, inflammatory biomarkers, including NLR, do not predict outcome. Fourth, asthma shows a protective association that, whilst intriguing, requires cautious interpretation.

The case fatality rate of 45·5% represents a marked departure from the near-universal lethality that has characterised PAM since its recognition^1,2. Historical CFRs have consistently exceeded 97%, with survival limited to isolated case reports^5,6. Several factors likely contribute to the improved outcomes we observed. The outbreak context heightened clinical awareness, enabling earlier diagnostic consideration¹². The standardised treatment protocol ensured consistent delivery of amphotericin B ± miltefosine to all patients¹⁸. A robust critical care infrastructure with over half of patients receiving ICU admission and mechanical ventilation may have prevented deaths from respiratory failure and other complications. That mechanical ventilation showed a protective trend (OR 0·53; p = 0·094) is consistent with the value of aggressive supportive care.

Diabetes mellitus emerged as the only statistically significant predictor of mortality in the adjusted model (aOR 2·59; p = 0·048), though we note that the confidence interval is relatively wide and approaches unity at its lower bound. Diabetic patients demonstrated CFR of 66·7% compared with 39·4% in non-diabetic individuals (Table 4). The bootstrap-resampled confidence interval excluded unity (95% CI 1·06–8·74), and the E-value of 4·62 suggests moderate robustness to unmeasured confounding²¹. The association strengthened rather than attenuated in simpler models (OR 3·05–3·23), suggesting the primary model estimate may be conservative. There was no significant interaction with treatment timing, indicating diabetes confers risk regardless of how quickly treatment is initiated.

From a mechanistic standpoint, diabetes impairs multiple components of anti-amoebic immunity: neutrophil chemotaxis, phagocytic capacity, and complement activation are each compromised in hyperglycaemic states^23,24. The blood–brain barrier shows increased permeability in diabetes due to chronic endothelial dysfunction²⁵. This finding fits with the broader pattern of diabetes as a risk factor for severe outcomes across infectious diseases, including bacterial meningitis and COVID-19^24,26. Clinically, diabetic patients presenting with possible PAM may warrant immediate aggressive intervention and close monitoring.

The failure of inflammatory markers to predict mortality represents a departure from patterns observed in other central nervous system infections. In bacterial meningitis, elevated NLR (typically >10–15) predicts poor outcome, reflecting the prognostic value of systemic inflammatory responses^16,27. Similarly, NLR predicts mortality in sepsis and COVID-19 pneumonia^28,29. In this PAM cohort, however, NLR showed no association with outcome (p = 0·360; Table 5). NLR was, in fact, paradoxically lower in non-survivors (median 12·2) than survivors (median 14·0). IL-6, TNF-α, and IL-1β showed similar non-discrimination (Fig. 5A). These data suggest that in PAM, systemic inflammatory responses are dissociated from clinical outcome; the magnitude of the inflammatory response does not appear to determine survival.

The relationship between pathogen burden and clinical parameters warrants further comment. Although CSF pathogen burden did not differ between survivors and non-survivors (p = 0·827), it correlated significantly with admission neurological status (Spearman r = −0·40, p 5B): patients presenting with lower GCS scores harboured higher amoebic loads. This dissociation pathogen burden tracks with neurological severity at presentation but not with ultimate survival, suggesting that factors beyond initial parasite load, such as the host inflammatory response, timing of treatment, and comorbidities, determine whether a given degree of neurological insult proves fatal. The finding also implies that the relationship between luminal CSF burden and parenchymal tissue invasion may be non-linear, with organisms sequestered within brain tissue rather than circulating freely in CSF. N. fowleri causes rapid, fulminant destruction of olfactory bulb and frontal lobe tissue within 24–72 h of CNS invasion^3,13. The inflammatory responses measured at presentation likely represent downstream phenomena following irreversible neurological injury, rather than causal drivers of ongoing damage. This interpretation has clinical implications: unlike bacterial meningitis, where modulating inflammation can improve outcome³⁰, anti-inflammatory strategies may have limited benefit in PAM once clinical presentation has occurred.

The protective association of asthma (univariable OR 0·37; p = 0·021) should be interpreted with caution, as it may partly reflect confounding by corticosteroid use, age distribution, or differences in healthcare-seeking behaviour among asthmatic patients. In a post hoc sensitivity analysis adjusting for age, sex, and district of residence, the association was attenuated (aOR 0·42; 95% CI 0·17–1·05; p = 0·063). We therefore regard this observation as hypothesis-generating rather than confirmatory. Several biological mechanisms could plausibly underpin a genuine protective effect: the type 2 immune polarisation characteristic of asthma, with heightened eosinophil activity, might confer some degree of anti-parasitic defence; chronic airway inflammation could prime mucosal immune responses at the olfactory epithelium; and inhaled corticosteroids commonly used in asthma management might modulate excessive neuroinflammation^3,14. These hypotheses require formal testing in experimental models and in cohorts where medication data are available.

The impact of treatment timing showed a threshold effect, with very early initiation (≤2 days from symptom onset) associated with substantially reduced mortality (CFR 29·6 versus 49·5%; OR 0·43; p = 0·084; Table 3 and Fig. 4). This trend attenuated at longer intervals, suggesting a critical early window during which therapeutic intervention is most effective^5,6. The median time to treatment of 4 days in this cohort exceeds this apparent therapeutic window for most patients. This observation underscores the imperative for heightened clinical suspicion and rapid diagnostic capacity in endemic regions^19,20.

The epidemiological profile of this outbreak carries public health implications^8,9. The biphasic temporal pattern sporadic cases from January through July followed by a surge from August onwards (Fig. 2) suggests environmental amplification during monsoon and post-monsoon conditions when warm stagnant water favours N. fowleri proliferation. The involvement of piped municipal water (19·5% of exposures) and borewells (15·5%) alongside natural water bodies challenges the traditional conception of PAM as exclusively associated with recreational freshwater contact (Table 1). Domestic water practices may represent important transmission routes^2,11. Geographic clustering across six districts (Table 2) and sustained transmission over 11 months indicate established environmental reservoirs requiring surveillance and remediation¹². As climate change elevates water temperatures globally, similar outbreaks may emerge in previously unaffected regions^8,10,31.

This study has several limitations that merit candid acknowledgement. The observational design cannot establish causality; the associations we report, including that between diabetes and mortality, may be confounded by unmeasured factors. Thirty-three per cent of patients remained under treatment at database closure, introducing uncertainty in CFR estimates, although sensitivity analyses spanning best-case to worst-case scenarios yielded a relatively narrow range (30·5–63·5%) and the GCS-stratified imputation closely approximated our primary estimate (Fig. 3B). The modest explanatory power of the multivariable model (pseudo R² = 0·084) indicates that measured clinical variables at presentation account for only a fraction of outcome variance; genetic susceptibility, pathogen virulence heterogeneity, and the precise timing of irreversible neurological injury likely contribute substantially but were not captured. We lacked individual-level data on medication use—particularly inhaled corticosteroids in asthmatic patients, which limits interpretation of the asthma association. The uniform treatment protocol precluded evaluation of therapeutic regimen comparisons. The single-outbreak setting in Kerala may limit generalisability to other populations and healthcare contexts. However, the low R² is itself informative, suggesting that measured clinical parameters at presentation have limited prognostic value—consistent with the hypothesis that outcome is determined early in infection.

Several lines of investigation should follow from these findings. Whole-genome sequencing of N. fowleri isolates from this outbreak would clarify whether virulence heterogeneity contributes to the observed outcome variation. Host genetic studies particularly of innate immune pathways including complement, toll-like receptors, and eosinophil granule proteins—may identify susceptibility loci that explain why some individuals succumb rapidly whilst others survive. The apparent therapeutic window suggested by the ≤2-day treatment analysis calls for evaluation of ultra-early empirical protocols in endemic regions, potentially including pre-hospital amphotericin B administration. Finally, systematic environmental surveillance of domestic water sources, including piped municipal supplies and borewells, is needed to guide public health interventions as climate change expands the geographic range of N. fowleri.