Background: Little is known about the transmission dynamics of Acinetobacter baumannii in hospitals. In the absence of comprehensive epidemiological data, mathematical modelling is an attractive approach to understanding transmission process. The statistical challenge in estimating transmission parameters from infection data arises from the fact that most patients are colonised asymptomatically and therefore the transmission process is not fully observed. Hidden Markov models (HMMs) can overcome this problem. We used such approach to characterise the transmission dynamics of A. baumannii in intensive care units (ICUs) in three hospitals in Melbourne, Australia.
Methods: A continuous-time structured HMM was developed. The HMM consisted of the hidden states – the number of patients colonised with A. baumannii (both detected and undetected) on the ICU ward, and the number of colonised patients detected at each time point (observations). The model input was monthly incidence data (60 months) of observed number of colonised patients. We estimated two parameters: the rate of cross-transmission between patients (β) and the rate of colonisation from sources independent of cross-transmission (sporadic acquisition, ν). The basic reproduction ratio (R0) and the proportion of colonisation due to cross-transmission (p) were also estimated. The parameters were estimated using the Bayesian framework with Markov chain Monte Carlo algorithm. The deviance information criterion was used to assess model fit.
Results: The model with cross-transmission and sporadic acquisition was superior to the models with only either of the acquisition sources. We estimated 96 – 98% of acquisition in Hospital 1 and 3 was due to cross-transmission; whereas most colonisation in Hospital 2 was acquired sporadically. On average, 4 (95% credible interval [95% CI] 3 – 5) and 2 (95% CI 1.8 – 3) new cases per month resulted from a colonised patient via cross-transmission in Hospital 1 and 3, respectively; whereas 2 (95% CI 1 – 3) new cases arose from sporadic acquisition in Hospital 2 every month. R0 for Hospital 1, 2 and 3 was 1.5 (95% CI 1.2 – 2), 0.02 (95% CI 0 – 0.2) and 1.6 (95% CI 1.3 – 2.2), respectively.
Conclusions: This study is the first to characterise the transmission dynamics of A. baumannii using mathematical modelling. We showed that HMMs can be applied to sparse data to estimate transmission parameters despite unobserved events. A. baumannii colonisation can be acquired both from cross-transmission and sporadically. Infection control in ICUs where A. baumannii is endemic should be optimised.
