Incorporating incubation period distributions to precisely estimate the association between rainfall and Legionella infection

Background: Multiple studies have shown a positive relationship between weather events and, 1 to 2 weeks later, Legionnaires' disease (LD) cases. Narrowing this time window of association can help determine whether the mechanism linking rainfall and relative humidity to sporadic LD is direct or indirect. Due to the large number of daily water interactions and low incidence of LD, we propose a new Bayesian modeling approach to disentangle the potential for a direct vs. indirect exposure to precipitation.

Methods: Incubation period distributions were used to redistribute LD cases to their estimated day of exposure. Then Bayesian distributed lag models were fit to estimate cases per day of exposure with predictor variables for rainfall and absolute humidity. Sensitivity analyses explored the impact of relatively humidity, rainfall post-estimated date of exposure and randomized rainfall to validate our results.

Results: One standard deviation increase in rainfall 2 and 3 days prior to the date of estimated exposure was associated with an approximately 15% increase in LD risk (per day). When heavy rainfall occurred 0 to 3 days prior to estimated exposure, risk increased by more than 40%, peaking at a 51% increased risk of LD 2 days after heavy rainfall.

Discussion: Our findings of a 2- and 3-day lag between rainfall and the date of estimated exposure is consistent with an indirect link with rainfall, rather than a same-day exposure. Potential pathways that can indirectly link rainfall to LD cases include rainfall-mediated declines in public water supplies but greater environmental sampling research is needed.