Rift Valley Fever (RVF) is a mosquito-borne zoonotic disease often linked to climatic factors. However, identifying the onset and duration of outbreaks can be hindered by high numbers of zero counts in sparse surveillance data. We analyzed monthly RVF data from Kenya (2015–2022), compared the performance of widely used outbreak-detection methods, and developed Negative Binomial models augmented with a hidden Markov process and zero-inflation to address data sparsity. Our proposed framework estimates monthly probabilities of transitioning between endemic and epidemic phases, improving outbreak detection under various sparsity levels. Simulations and real-world applications show that this model outperforms established algorithms by reducing both false positives and negatives. When applied to Kenyan RVF data, the model suggests a weak negative association with rainfall and a weak positive association with temperature, potentially reflecting underreporting’s influence on estimated climatic effects. Overall, these findings underscore the potential of the proposed method to improve outbreak detection and public health decision-making in sparsely reported settings.
Angelakis et al. (Fri,) studied this question.