Estimating Chikungunya prevalence in La Réunion Island outbreak by serosurveys: Two methods for two critical times of the epidemic

Chikungunya fever is an arbovirosis caused by Chikungunya virus (CHIKV), a mosquito-transmitted alphavirus belonging to the Togaviridae family [1, 2]. CHIKV was first isolated in 1952, during a Tanzanian outbreak [3]. It circulated in Africa and Asia, where periodic outbreaks were described in the past 50 years. In some areas, attack rates had reached 80 to 90% [1, 2]. Between February 2005 and August 2006, a large Chikungunya fever outbreak swept the Indian Ocean islands [4, 5], including La Réunion Island since April 2005, an overseas French department of 787,836 inhabitants (Figure 1). The mosquito specie involved in La Réunion outbreak was Aedes (A.) albopictus [6]. Most CHIKV infections were symptomatic [7] and characterized by a dengue-like illness of sudden onset combining high fever, poly-arthralgia, myalgia, headache, asthenia and rash [8, 9].

In La Réunion, the epidemic pattern was monitored through a regional surveillance-system managed by the Cellule Interrégionale d'Epidémiologie (CIRE) based on "suspected cases", defined as subjects with a sudden fever (T > 38.5°C) and incapacitating arthralgia [10, 11]. This surveillance-system relied on self-reports, emergency stays, physician declarations, biology laboratories activity, and active case-finding around the cases reported by a sentinel physician network [11]. At the beginning of the outbreak it consisted in an active and retrospective case detection around the cases declared, and then, when the incidence sharply increased (by December 2005), in an estimation of the cases obtained from reports of a sentinel network [12].

Before the explosion of the epidemic in mid-January 2006, a herald wave occurred during the previous rainy season; between March and July 2005, and led the CIRE to record about 3,000 suspected cases of Chikungunya [13]. Later on and until December 2005, low case rates were recorded without interruption. An exponential increase of the cases reported was observed in late December 2005, and January 2006 with a peak in February 2006 [11] (Figure 2). On February 15^th 2006, the CIRE estimated 157,000 suspected cases of Chikungunya, i.e. a prevalence rate of 20.3%. On July 5^th 2006, the CIRE estimated the burden more than 266,000, i.e. a prevalence rate of 34.3% [14].

The purpose of the study was to refine the estimates of attack rates provided by the surveillance-system for the population of La Réunion Island at two critical times of the 2005–2006 outbreak. That is why we conducted two serosurveys, the first using stored sera of pregnant women during the epidemic upsurge aimed at assessing the extent of the outbreak, the second using a random sample of the population aimed at giving a precise idea of the herd immunity in the post-epidemic era.

In the current paper, we demonstrate the usefulness of two different epidemiological methods to assess the burden of a Chikungunya outbreak at two critical times of its evolution.

The rapid seroprevalence survey conducted at the peak of La Réunion Island epidemic on sera from pregnant women provided an 18.2% seroprevalence rate (or a rough estimate of 143,000 people infected) by February 15^th, 2006. This result was obtained at a very low cost, at a time when the attack rate of the infection in the population and the herd immunity were unknown. Since 99% of the 162 tested sera harbored IgM anti-CHIKV antibodies and only one IgG anti-CHIKV antibodies only, it excluded a previous (recent) significant circulation of the virus in the island and thus suggested that the outbreak emerged into a naïve population. This result is in agreement with the 20% prevalence rate calculated for all parturient women delivered at the GHSR maternity in mid-February 2006 [16], and with the magnitude of 26% reported in pregnant women by April 2006 in Mayotte [17].

The rough estimate by the rapid serosurvey in pregnant women is slightly higher but of the same magnitude than the 130,000 cumulative number of suspected cases deducted from the CIRE at the same time (attack rate of 16.5%) [18, 19]. It is therefore noteworthy that the rate observed in a targeted population of pregnant woman gives a valuable insight upon the magnitude of the attack rate in the community, and beyond, of the herd immunity. This congruent result with the CIRE data suggests that in February 2006 pregnant women yet behaved like everyone and that their level of exposure to CHIKV was not different to that observed for anybody else. In other words, the message aimed at crystallising the pregnant woman as vulnerable person to the threat of Chikungunya [20] and the measures aimed at reducing her exposure [21], e.g., wear of long clothing, free distribution of insecticide-treated nets and repellents, soon implemented in the maternities of La Réunion, were not effective against A. albopictus bites, a vector which was proven to exercise a diurnal activity [6].

The slight difference between pregnant women and community may account for a selection phenomenon, the pregnant women being not representative of the community. However, the prevalence was higher in pregnancy which argues against a significant selection bias, since more than 50% of serum collections came from south and west labs at a time when transmission was still predominant in theses micro-regions, leading to an unexpected geographical adjustment on the transmission level.

The population-based survey provided a 38.2% prevalence rate in the post-epidemic era, i.e., about 300,000 people infected, by October 20, 2006 [7]. At the same time, the CIRE data, published on the Institut national de Veille Sanitaire website, estimated at 266,000 (34.3%) the number of people infected [14]. Thus, the seroprevalence in the community was slightly higher than the 34.3% estimated for suspected cases (P < 0.001), but of the same magnitude. This difference might correspond to undeclared cases to the CIRE, i.e., (1) the inapparent cases (5.0%), although these would be compensated by an approximately equal proportion of false positives (4.5%); (2) patients who did not consult and performed auto-medication; (3) patients who did not match the clinical criteria for "suspected cases", i.e., sudden fever > 38.5 C° and incapacitating arthralgia. Based on laboratory confirmations of atypical presentations, Renault et al. calculated that approximately 3% of patients did not fulfil this definition [11]. Importantly, the SEROCHIK survey showed that 25.8% of CHIKV-infected subjects did not declare fever combined to arthralgia. This significant discrepancy may result both from an information bias due to the structure of the questionnaire (subjects who were not aware of their infection did not answer to questions about clinical signs), or a memory bias due to the time interval between the SEROCHIK survey and the onset of symptoms (2 to 15 months) that could preclude mild cases to remind their symptoms. However, the declaration-based surveillance system based on suspected cases may have also underestimated the attack rate at the epidemic phase. The difference was particularly notable from April to December 2005 (Figure 3), when the incidence was less than 500 new cases per week (Figure 2) and the surveillance relied upon active and retrospective case detection around the cases declared. Ditto, it was verified from June 2006 (Figure 3), when the incidence dropped dramatically shortly before the epidemic stopped in August (Figure 2). It could be explained by a lower PPV for each symptoms (< 85 to 92%) and for the clinical definition (< 87%), as the transmission was low (Table 3). Indeed, it is well known that when the incidence of a communicable disease is low, its contribution to the clinical forms that can evoke it decreases, with a consequent decline of the PPV for clinical signs to identify the disease [22]. Moreover, the possibility of concomitant circulation of other infections, such as Influenza or Dengue [23], may have challenged the diagnosis of Chikungunya [24], especially between April and December 2005, or from June 2006.

Another important aspect of Chikungunya disease disclosed in the course of La Réunion outbreak is the low proportion of inapparent forms (16%), in comparison with those usually observed for other arbovirosis, such as Dengue Fever (> 50%) [25, 26] or West-Nile virus infection (> 70%) [27, 28].

Finally, it is noteworthy that the 38.2% seroprevalence rate observed in the post-epidemic phase [7] gives a valuable insight upon the herd immunity and a clearer picture of susceptible people who could be infected in the future (62.8%). Importantly, the seroprevalence observed in La Réunion Island was far inferior to those reported recently from the Kenyan island of Lamu (75%) [29] and the Grande Comoro Island (63%) [30], two areas where CHIKV emerged before to reach La Réunion [4, 5].

Several hypotheses may explain this discrepancy in prevalence rates: 1°) Kenyan and Comorian climates are less prone to seasonal variations and therefore more conducive to a sustainable transmission; 2°) A. aegypti, the classical vector of Chikungunya, involved in Kenyan and Comorian outbreaks, keeps a better capability to spread the disease in domestic environment than the less anthropophilic A. albopictus; 3°) for the same reason, the density of susceptible hosts would be less important in peridomestic environment in La Réunion, than in Kenyan and Comorian homes invaded by highly anthropophilic A. aegypti; 4°) in La Réunion, floods brought by the cyclone Diwa drove away larvae of A. albopictus from gullies and hastened the decrease in transmission from early March 2006; 5°) effective vector control measures combining eradication of breeding sites, adulticide and larvicide treatments contributed to limit the density of vectors throughout the Reunion outbreak; 6°) In La Réunion, the herd immunity was gained more rapidly in the littoral plains (where most of the population lives in contact with vectors) which reduced transmission as the entry in the dry austral winter. Indeed, some micro-geographical differences in prevalence rates would have been not detected by the SEROCHIK survey (whose sampling plan was aimed at discriminating between micro-regions but not within), and thereby prevalence rates in littoral plains would have far exceeded the overall 38.2% rate, which would have led to the premature decline of the epidemic due to effective herd immunity.

Prevalence rates of 60 to 70% were necessary to delay resurgences to 20 to 30 years in areas where CHIKV had circulated before [31]. According to a recent mathematical model [32], it was concluded that in the best-fitting case (reproductive number of 3.7), the attack rate would have been of 73% which suggests that, with only 38.2% of people infected, a re-emergence in La Réunion Island cannot be excluded for the next years, as long as viral strains circulate in the region. However, the scenario observed in La Réunion seems thwart this mathematical prediction, because so far, no case of Chikungunya has been scientifically confirmed since August 2006. This could be explained at municipality level by heterogeneity of reproductive numbers that would have been very sensitive to local interventions in vector control [33].

Congruent estimates of Chikungunya attack rate were observed at the upsurge of La Réunion Island outbreak, either using clinical declaration of suspected cases (16.5%), or a rapid serosurvey in pregnant women (18.2%). In contrast, a discrepancy was observed in the post-epidemic era, when clinical diagnosis underestimated the attack rate (34.3%) in comparison to seroprevalence estimate (38.2%; 95% CI, 35.9 to 40.6%). Thus, a rapid serosurvey in a targeted population can be helpful to assess the extent of epidemics at time of emergency when large seroprevalence studies cannot be done. Beyond this indication, our findings suggest that prospective real time surveillance of attack rates in pregnant women would serve as a good model for population monitoring in the event of Chikungunya outbreaks. However, although it may fail to detect micro-geographical differences in prevalence rates at municipality level, a population-based serosurvey can still be useful to refine the clinical estimates and to assess more precisely the herd immunity. Moreover, only a representative survey can bring an overview on risk factors and other conditions facilitating the transmission. Finally, this work speaks to the usefulness of serosurveys for the quantification of epidemics, as to cost-containment in public health. It also provides valuables clues for monitoring epidemics in high and low-incomes countries.