Abstract
From August to November 2017, Madagascar has endured an outbreak of plague. A total of 2119 cases of plague has been confirmed, causing until now a death toll of 195. Public health interventions have been introduced, preventing new cases and deaths. However, it is likely that the outbreak could reappear as plague is endemic in the region and typically only last until April annually. We collected real-time data from various official reports. We described the outbreak’s characteristics and reported estimates of the key transmission parameters using statistical and mathematical modelling approaches. Plague’s epidemic curve depicts a propagated outbreak with multiple peaks, caused by continuing exposure. Optimal climate conditions for rat flea to flourish were observed during the epidemic. Estimate of the Plague’s reproduction number based on pneumonic data was 6.9. The main mode of transmission is human-to-human with a much higher transmission rate than that of flea-to-human mode. With a potential of continuing exposure to infected rat fleas until April 2018, current public health efforts should be maintained at the high level. While efforts in controlling vector to prevent the appearance of new index cases can be critical, maintaining interventions targeting reduce human-to-human transmission is key to prevent large-scale outbreaks.
1. Introduction
One of the deadliest natural disasters in human history was reported as the Black Death — attributed to the bacterium Yersinia pestis — killing about 50 to 200 million people in the 14th century [1]. Although plague was naturally widespread in ancient times, plague outbreaks occurred following the deliberate use and propagation of this disease, serving as a bioweapon [2]. This lethal bacterium can derive in several forms of plague maintaining its existence in a cycle involving rodents and their fleas [3]. While sanitation and public health surveillance have greatly reduced the likelihood of a plague pandemic, isolated plague outbreaks are lethal threats to humankind.
This gram negative can result in different clinical forms of plague: bubonic, pneumonic and septicemic [1]. Human infection is primary driven by bubonic plague, as a result of being bitten by infected fleas. Additionally, direct contamination with infective material can be an alternative transmission route [1]. Patients with bubonic plague can develop sudden onset of fever, headache, chills, tender and painful lymph nodes [4]. While plague can be successfully treated with antibiotics, if untreated, the bacteria can disseminate from the lymph nodes into the bloodstream causing secondary septicemic plague. In addition to the symptoms presented in the bubonic plague, patients with septicemic plague undergo abdominal pain and possibly bleeding into the skin and other organs, at the same time skin and other tissues may turn black and die, especially on fingers, toes, and the nose [3]. However, the most fulminant form of the disease is driven by pneumonic plague that turns out to be the only form of plague that can spread from person to person by infectious droplets. The incubation period of pneumonic plague is shorter than in the other forms of the disease, usually 1 to 2 days, leading to a high fatality rate despite immediate treatment with antibiotics [4].
Plague epidemics of infectious diseases continue to pose a threat to humans, reporting continuous annual occurrence of plague cases in five countries: Mada-gascar, Tanzania, Vietnam, China, and the USA [1, 5]. On 13 September 2017, the Madagascar Ministry of Public Health notified WHO of an outbreak of pneumonic plague [4]. Since then, a total of 2417 cases of plague has been confirmed, causing until now a death toll of 209 [6]. Public health interventions have been introduced in Madagascar, preventing new cases and deaths. However, it is likely that the outbreak could reappear as plague is endemic in the region and could last until April 2018 [4]. Descriptive and numerical analyses of the plague outbreak could facilitate studies in evaluating the spread of diseases as well as targets for disease control and prevention.
2. Materials and Methods
Outbreak Data - Cumulative Cases
Data were manually inputted from separate reports of WHO [6], including the cumulative total numbers of clinical cases (confirmed, probable, and suspected). The data can be found at the following link systemsmedicine/plague2017/Cumulative
Outbreak Data - by Disease Forms
Data were digitized from the figure reported from WHO [6], including the incidences classified by the three forms of the plague disease: pneumonic, bubonic, and septicemic. The data can be found at the following link systemsmedicine/plague2017/Classification and the digitized figure can also be found at the same repository.
Temperature and Precipitation Data
Data were requested from the National Centers for Environmental Information (Order #1133340 Custom GHCN-Daily CSV). The data can be found at systemsmedicine/plague2017/Climate.
Descriptive analyses
With the aim of facilitating modelling works, we described dynamics and patterns of variables that have previously shown to be relevant to plague outbreaks, including temperature and precipitation [7].
Statistical estimate of the reproduction number
We estimated the reproduction number (R0) of Yersina pestitis using data of pneumonic cases during the second (large) wave of the epidemic, i.e., visually defined from 22/09/17 onwards. The serial interval of plague was assumed gamma distributed with shape and scale parameters are 5.4 and 0.9, respectively [8]. We reported the R0 estimates using the following methods: exponential growth (EG) [9], maximum likelihood (ML) [10], and the sequential bayesian estimation (SB) [11].
Plague transmission model (PTM)
We conducted an epidemic simulation using a modified SEIR (Susceptible-Exposed-Infectious-Removed) model with two incorporated components reflecting the infection from infected rat fleas and the effects of public health interventions. A schematic illustration of the PTM is shown in Figure 1. Assuming only a small proportion (p) of the population will be exposed to the risk of being bitten by infected rat fleas. The flea density is approximated by a sinusoidal function fitted to Madagascar temperature. Dashed lines indicate where public health intervention would have the strongest effect, i.e., animals control and reducing contact with infected cases. The model equations are as follows:
where S, E, I describe Susceptible, Exposed, and Infectious and the subscripts b and p denote bubonic and pneumonic form, respectively. The model assumes that in a population of size S0, only a small part Sb = (1 − p)S0 is exposed to infected rat fleas. The flea-to-human and human-to-human transmission rates are denoted as α and β, respectively. The infected bubonic cases become infectious with a proportion ε progressing to pneumonic stage. We followed the approach of Aron and May [12], approximating the density of infected rat fleas as a sinusoidal function firf = A +Bsin(2π/12t)+Ccos(2π/12t) which was fitted to the average temperature of Madagascar in the period 1960–2008 [7]. Public health interventions are assumed reducing both human-to-human and flea-to-human transmission that has a logistic form fitv = 1 − 1/[1+exp(−(t − τ))] where τ denotes the time at which the intervention effect reaches half its potential. The infected cases are assumed to recover and die with the total rate of removal from the infected pool being δb and δb for pneumonic and bubonic cases, respectively.
We fitted the PTM to the daily data of pneumonic and bubonic cases during the large wave of the epidemic curve which was visually defined from 22/09/17 onwards. Model parameters were estimated using the global optimisation algorithm Differential Evolution [13]. Simulations and estimations were done in R using packages base [14], deSolve [15], and R0 [16] and in Python. Stochastic simulations were performed using a tau-leaping algorithm with a fixed time-step of ca. 15 minutes; code and data are publicly available at systemsmedicine/plague2017/Stochastic.
3. Results
Descriptive analyses
During August, bubonic cases appeared sporadically with almost no records of pneumonic form (fig. 2). An increase in number of pneumonic cases was not necessarily led by an increase in the number of bubonic cases (fig. 2). It seemed to be the epidemic curves include several waves of incidence overlapping each others.
Figure 3 shows that plague incidences emerged everyday in the weeks although in some weeks a smaller number of cases were reported in the weekends. No distinctive time lag was observed between the appearances of bubonic and pneumonic cases (fig. 3). The incidences were negligible during the period when Famadihana tradition was presumably practised. Precipitation measure exhibited no pattern before or during the outbreak but generally showed a dry climatic condition. Average temperature appeared increasing and reached a higher level (above 23 degree Celsius) around the same time as the outbreak.
Figure 4 shows that the temperature would typically remain at the level favoring the rat fleas (20–25°C [17]) in the upcoming months and until May in the next year.
Estimates of the reproduction number using pneumonic cases
Figure 5 shows that three estimation methods gave the same estimate for plague’s basic reproduction number during the epidemic growing phase, approximating 7. The effective reproduction number, however, could be seen quickly plunge toward unity.
Mathematical model of plague epidemics
Figure 6 shows that the model and the estimated parameters (table 1) capture well the dynamics of both pneumonic and bubonic data. Stochastic transitions could lead to larger or smaller waves of the dynamics. The parameters suggest that only a small fraction (1 per 10000) exposed to infected rat fleas is enough to generate the observed epidemic. The human-to-human transmission rate was much higher than the flea-to-human transmission rate. The average time spent in the exposed state before infectious was 5.614 and 1.537 days for bubonic and pneumonic cases, respectively.
4. Discussion
Mathematical models of infectious diseases have played a central role in understanding epidemics, providing an effective way of assessing disease transmission as well as evaluating disease control and prevention strategies [20]. Mathematical modeling has proposed new vaccination strategies against influenza infection [21]; supported public health strategies for containing emerging influenza pandemics [22, 23] and for the use of antiretroviral treatment for HIV-infected patients as a preventive measure [24]; reported real-time estimates of Ebola’s R0 to inform the outbreak situation [25], among others.
However, while it has been noted for the last Ebola outbreak [25], data sharing are still poor and WHO practices of reporting (separate PDF files) are putting constraints on modelling works. In the 2014 Ebola outbreak, most of the modelling studies were also done rather late in the process [26]. The solution can be as simple as putting a unique Excel file and update it, or better establishing a central website for all WHO outbreak reports, or alternatively, a data hub to encourage user contributed reports.
From modelling aspects, plague outbreak can be more challenging because: (1) there is a continuous input of flea-to-human transmission (fig. 2); this implies the observed epidemic curve can be a mixed of multiple waves of infected cases generated from different index cases. Thus, epidemic evaluations could risk to over- or under-estimate the consequences, e.g. the reproduction number or the end time of epidemics; (2) there are a known seasonal pattern of the plague epidemic [7] in which a direct measure of the rat flea population does not exist; (3) The flea-to-human transmission as well as the transition from bubonic cases to pneumonic cases appeared stochastically driven (fig. 2) and could be highly affected by interventions.
Here, the epidemic curves showed plague incidences appear sporadically during August (fig. 2). The pneumonic epidemic curve suggests that superspearders were likely to exist in order to generate the larger number of cases observed in the later periods. This can be observed from fig. 2 where the large increase in pneumonic cases was preceded by only a few bubonic cases. Estimates of the effective reproduction number also showed that a high estimate is needed to capture the early growing phase of the epidemic (fig. 5); which is almost double the previous estimates [8]. However, considering the potential mixed of epidemic waves and outbreak locations in the used data, the reproduction number could be overestimated. Approaches adjusting for the propagated outbreak data are needed to gain further insights in epidemic processes.
In addition, the bubonic cases appeared regularly making a continuous input of index cases to the human-to-human transmission network (fig. 2); as a result, the pneumonic epidemic curve seemed exhibiting multiple peaks. This suggests that vector control interventions are key to avoid potential next waves of the epidemic. Experiments have shown that an optimal climate for rat fleas to flourish is a dry climate with temperatures of 20–25°C [17]. These conditions were observed during the current outbreak: a generally dry weather with the optimal temperature coincided with the period of high epidemic activity. As further shown in fig. 4, these conditions would typically remain the same until next May, see further in Kreppel et al. [7]. This again stresses the role of vector control in preventing the next waves.
Nevertheless, the estimate of the human-to-human transmission rate is much higher than that of the flea-to-human transmission. This prompts that interventions aim at human-to-human transmission route need to remain at high level. Figure 7 shows extreme scenarios of the model trajectories assuming these speculations. The result shows that efforts, in cases of exhaustive of resources, should prioritise stopping human-to-human transmission route. In this case, while the bubonic cases would continue to appear they would not able to generate large size outbreaks. This is practical as with the usual public health intervention of early detection of the incidences would not only stop the human-to-human transmission but also provide a better chance for new bubonic cases to be treated early.
In this paper, we collected and described relevant data of the 2017 plague epidemic in Madagascar. We proposed a working mathematical model for evaluating and predicting epidemic consequences and what-if scenarios. We discussed potential drawbacks in modelling propagated epidemic data. We hope that the results would contribute informative insights for public health officers and provide a framework for further understanding the dynamics of plague outbreaks.
Authors contributions
VKN, EAHV: conceptualization and data curation. VKN: methodology, formal analysis, investigation, and visualization. CPR: stochastic simulations implementation and visualization. EAHV: funding acquisition, supervision. All authors: review & editing.
Acknowledgment
This work was supported by the Alfons und Gertrud Kassel-Stiftung.