Q fever: Evidence of a massive yet undetected cross‐border outbreak, with ongoing risk of extra mortality, in a Dutch–German border region

Abstract Background Following outbreaks in other parts of the Netherlands, the Dutch border region of South Limburg experienced a large‐scale outbreak of human Q fever related to a single dairy goat farm in 2009, with surprisingly few cases reported from neighbouring German counties. Late chronic Q fever, with recent spikes of newly detected cases, is an ongoing public health concern in the Netherlands. We aimed to assess the scope and scale of any undetected cross‐border transmission to neighbouring German counties, where individuals unknowingly exposed may carry extra risk of overlooked diagnosis. Methods (A) Seroprevalence rates in the Dutch area were estimated fitting an exponential gradient to the geographical distribution of notified acute human Q fever cases, using seroprevalence in a sample of farm township inhabitants as baseline. (B) Seroprevalence rates in 122 neighbouring German postcode areas were estimated from a sample of blood donors living in these areas and attending the regional blood donation centre in January/February 2010 (n = 3,460). (C) Using multivariate linear regression, including goat and sheep densities, veterinary Q fever notifications and blood donor sampling densities as covariates, we assessed whether seroprevalence rates across the entire border region were associated with distance from the farm. Results (A) Seroprevalence in the outbreak farm's township was 16.1%. Overall seroprevalence in the Dutch area was 3.6%. (B) Overall seroprevalence in the German area was 0.9%. Estimated mean seroprevalence rates (per 100,000 population) declined with increasing distance from the outbreak farm (0–19 km = 2,302, 20–39 km = 1,122, 40–59 km = 432 and ≥60 km = 0). Decline was linear in multivariate regression using log‐transformed seroprevalence rates (0–19 km = 2.9 [95% confidence interval (CI) = 2.6 to 3.2], 20 to 39 km = 1.9 [95% CI = 1.0 to 2.8], 40–59 km = 0.6 [95% CI = −0.2 to 1.3] and ≥60 km = 0.0 [95% CI = −0.3 to 0.3]). Conclusions Our findings were suggestive of widespread cross‐border transmission, with thousands of undetected infections, arguing for intensified cross‐border collaboration and surveillance and screening of individuals susceptible to chronic Q fever in the affected area.


| INTRODUC TI ON
Following outbreaks in other parts of the Netherlands, the Dutch border region of South Limburg experienced a massive single-point source outbreak of Q fever related to a local dairy goat farm, counting 253 notified cases of acute human Q fever and an estimated 9,000 undetected infections across the entire region, in 2009 (Hackert et al., 2012;van Leuken et al., 2013). Q fever is a bacterial zoonosis caused by Coxiella burnetii, transmitted to humans from infected ruminants, primarily by the airborne route (Cutler, Bouzid, & Cutler, 2007;Maurin & Raoult, 1999;Parker, Barralet, & Bell, 2006). In acute disease, flulike illness is usually prominent. Most Q fever infections are mild or asymptomatic and self-limiting (Hackert et al., 2012;van der Hoek et al., 2012). However, a small percentage of infected individuals may develop chronic Q fever, which often goes unnoticed for years after infection but is usually fatal if left untreated. In addition, a substantial proportion of infected individuals may suffer symptoms referred to as Q fever fatigue syndrome, which may persist for years, with major health-related consequences (Morroy et al., 2016).
In South Limburg, the distribution of notified cases of acute human Q fever followed a west-east gradient of decreasing incidence from the source, following westerly winds predominant at the time of the outbreak, towards and up to the Dutch-German border (Hackert et al., 2012). In the same year, only six cases of acute Q fever (notifiable under German law) were reported from the entire German federal state of North Rhine-Westphalia, none of whom was a resident of the two districts bordering South Limburg (Heinsberg and Aachen). In the five years preceding the outbreak (2004)(2005)(2006)(2007)(2008), a total of 42 cases were reported from North Rhine-Westphalia, just one of whom lived in Aachen (2006). In 2010, the year following the outbreak, North Rhine-Westphalia counted a total of 14 cases, only 2 of whom were from Aachen (Robert Koch Institute, 2016).
Whereas these data suggest that cross-border transmission from South Limburg to neighbouring German counties was negligible, it is rather unlikely that airborne transmission stopped short of the Dutch-German border. A Belgian study suggests that a limited degree of transmission took place in westerly direction across the Dutch-Belgian border, but does not quantify the extent of transmission (Naesens et al., 2012). Recent spikes of newly detected cases of late chronic Q fever in the Netherlands, with a high burden of extra mortality, show that the Dutch epidemic is far from over and should be seen as an ongoing public health concern with reason for unabated alertness (Radboud University Medical Center, 2018). This may be even more the case in a population unknowingly exposed to Q fever, where the risk of delayed or overlooked diagnosis of chronic Q fever may be even higher. We therefore aimed to assess the scope and scale of any undetected cross-border transmission to neighbouring German counties associated with the regional outbreak in South Limburg.

| Study area, study population and study period
The Meuse-Rhine Euroregion provided the administrative background for our study (Wikipedia contributors, 2017). Geographically, it covers an area of about 11,000 km 2 around the city corridor of Aachen (North Rhine-Westphalia, Germany), Maastricht (South Limburg, Netherlands), Hasselt (Limburg, Belgium) and Liège (Liège, Belgium).
The Dutch study area was defined by the approximate catchment area of a large regional general hospital (346 km 2 , 12 municipalities and 308,410 inhabitants).
The adjacent German study area was defined by the 122 postcodes of individual residents from North Rhine-Westphalia who donated blood at the RWTH Aachen University Hospital Blood Donation Centre in the first two months of 2010. For a summary of statistics regarding study area and study population, see Table 1. Of the 3,460 included German blood donors, the majority (n = 3,083, 89.1%) lived in postcode areas whose centroid was located within 40 km from the outbreak farm. The study was conducted from February 2009 (when first abortions were registered on the outbreak farm) to February 2010, when (a) no more incident human cases of acute Q fever were reported in the Dutch study area; (b) culling of pregnant dairy goats to prevent further transmission during the upcoming 2010 lambing season had been finalized; and (c) inclusion of German blood donors (January and February 2010) ended.

| Epidemiological investigation
To assess seroprevalence of Q fever in the Dutch, German and combined cross-border study area in relation to distance from the Dutch Conclusions: Our findings were suggestive of widespread cross-border transmission, with thousands of undetected infections, arguing for intensified cross-border collaboration and surveillance and screening of individuals susceptible to chronic Q fever in the affected area.

K E Y W O R D S
communicable disease control, Coxiella burnetii infection, international health regulations, one health, outbreaks, Q fever outbreak farm, we used various population samples and methods of analysis. An outline is given in Figure 1.

| Dutch study area
The outbreak affected a population largely naive to Q fever, according to a serological survey from the year predating the outbreak (2008). Human Q fever cases notified to the PHS South Limburg in 2009/10 (n = 253) were scattered downwind from the outbreak farm, following a gradient of declining attack rates in easterly direction from the source up to the Dutch-German border (Hackert et al., 2012). Using SPSS's curve fitting tool, we derived a curve of best exponential fit from aforementioned gradient, corresponding to the following model: 469.074733 * EXP (−0.321415 * [distance to outbreak farm in km]). Q fever seroprevalence in a convenience sample of adult residents from the outbreak farm's township (n = 120, aged ≥ 18 years, mean household distance from outbreak farm = 2.7 km) served as a baseline estimate for the calculation of seroprevalence rates across the entire distance from the outbreak farm up to the Dutch-German border according to our exponential model (Hackert et al., 2015). Underlying was the assumption that infections followed the same geographical gradient as notified cases. Seroprevalence in this township sample was age-and sexmatched within 10-year age strata according to the demographic distribution in our sample of German blood donors. To assess validity of our exponential model, we compared the calculated cumulative seroprevalence for the entire Dutch area with seroprevalence observed in a post-outbreak convenience sample of adults from the same area (Hackert et al., 2012).

| German study area
A total of 3,460 out of 3,493 blood donors (99.1%), who visited the Blood Donation Centre in January/February 2010, were residents of North Rhine-Westphalia, while n = 39 were residents of remote German federal states and consequently excluded from our study. Seroprevalence was estimated using retention sample sera from the 3,460 included blood donors. The detection of IgG or IgM phase II antibodies by ELISA or of C. burnetii DNA by qPCR identified positive sera. Minimum age for blood donations in Germany is 18 years. Apart from standard exclusion criteria relating to donor blood safety, additional criteria were applied during the study

| Combined Dutch-German cross-border study area
A regression model, including distance zones of 20 km as dummy variables, adjusting for goat and sheep densities, veterinary Q fever notifications and sampling rates (i.e. number of individuals tested per 100,000 population in each postcode area), was used to assess the geographical relationship between seroprevalence (assumed to represent incidence rate of infection) and exposure dose (approximated by residential distance from the outbreak farm). In addition, we applied our fitted model, derived from the geographical distribution of attack rates in the Dutch study area, to predict seroprevalence rates in the German study area by distance from the outbreak farm, and tested the correlation between predicted and observed rates. We used Spatial

| Laboratory investigation
Laboratory procedures performed during the Dutch outbreak were described previously (Hackert et al., 2012). All retention samples of the German blood donors and the Dutch serum samples were screened for anti-Coxiella phase II IgG according to manufacturers' protocols (Serion ELISA classic, Institut Virion/Serion GmbH, Würzburg, Germany). A selection of negative sera (n = 128) and all positive or indeterminate sera was additionally tested for anti-Coxiella phase II IgM (Serion ELISA classic, Institut Virion/Serion GmbH, Würzburg, Germany) and for the presence of Coxiella DNA using qPCR. We essentially applied the qPCR assay described elsewhere with a slightly modified TaqMan probe (Klee et al., 2006).

| Veterinary data
Human Q fever cases from 2009 were linked to a single dairy goat farm in South Limburg, whose nearest distance to the Dutch-

| Statistical analysis
Statistical analyses were performed using IBM SPSS Statistics, version 21 (IBM Inc.). We performed bootstrapping on our estimates using non-linear regression to obtain more robust confidence interval estimates. We report B values along with their 95% confidence intervals derived from bootstrapping.

| Overall seroprevalence in the Dutch-German cross-border region
Our smoothed map shows a large area of adjacent postcodes affected by human Q fever surrounding the outbreak farm, spreading over distances of more than 40 km into the German border region.

| Seroprevalence over 20-km distance classes from the outbreak farm
Seroprevalence rates across radial 20-km distance classes declined with increasing distance from the outbreak farm (Table 2 and

| D ISCUSS I ON
Transmission of Q fever over distances of at least 30 km has been described before (Eldin et  Our study suggests that cross-border transmission, in spite of evidence for massive numbers of infections dispersed over a wide geographical area, went largely undetected, posing susceptible patients at risk of long-term sequelae, most notably chronic Q fever, due to delayed diagnosis, missed diagnosis or misdiagnosis. Extrapolating from our sample of blood donors, the estimated number of unreported infections in the affected German border area may have been as high as 11,000. Our data suggest that the risk of Q fever infections going undetected was higher on the German than on the Dutch side of the border, with ascertainment rates (i.e. numbers of notified adult cases vs. estimated numbers of infections) being at least 10 times lower on the German side (Hackert et al., 2015(Hackert et al., , 2012 (Morroy et al., 2016).
Seroprevalence in our study declined exponentially with increasing distance from the outbreak farm across four 20-km zones. The observed west-east gradient is compatible with dispersal of C. burnetii from the farm, given that transmission of Q fever is usually airborne, and westerly winds were predominant in the study area at the time of Q fever-related abortions in pregnant goats on the outbreak farm.
Observed seroprevalence rates in the German study area were higher than rates predicted by our exponential model, with moderate statistical correlation between the two. Alternative sources, such as contaminated manure, wildlife reservoirs and sheep flocks that migrate over longer distances and have been shown to carry C. burnetii in clinically inconspicuous animals, may have contributed to human infections in the German border region, but these phenomena would not explain the geographical distribution observed (Hermans, Jeurissen, Hackert, & Hoebe, 2014;Hilbert et al., 2012;Webster, Lloyd, & Macdonald, 1995). While reports of Q fever in livestock in the German border region during the study period may implicate cattle as potential sources, a recent study found that human contact with sheep and goats, rather than cattle, was a consistent risk factor in human outbreaks (Georgiev et al., 2013;Verso et al., 2016). Goat and sheep densities in the German cross-border area, however, were rather low (see Table 1). While goats and sheep may have contributed to our seropositive findings, they thus seem unlikely sources for large numbers of human infections spread over a wide area. In addition, our data did not reveal any plausible local source other than the outbreak farm.
Nevertheless, while the observed west-east gradient of seroprevalence rates in the population is consistent with airborne disper- Under-ascertainment and under-reporting of Q fever are usually attributed to its mild and non-specific clinical presentation.
Primary infections are often mild, sometimes resembling a common cold, or asymptomatic, diagnosed only retrospectively through systematic testing (Eldin et al., 2017). This phenomenon is mirrored by a growing number of studies where seroprevalence rates in the population exceed reported cases of symptomatic Q fever. One such study from Denmark showed a rate of 64% of asymptomatic primary infections, while a study from Italy reported 30 seropositive individuals with no related episodes of respiratory or febrile disease (Bacci et al., 2012;Verso et al., 2016). A recent study from Spain found fever-usually considered the hallmark of symptomatic infection-to be absent in almost a third of 39 Q fever positive cases, even though all cases without fever had pneumonia.
Interestingly, a systematic review by the same authors reported the absence of traditional risk factors, most notably animal exposure, in a majority of almost 1,500 included Q fever cases, with as much as 60% of cases living in urban areas, raising the possibility of airborne and other routes of transmission in these cases (Alende-Castro et al., 2018).
Seroprevalence studies from the Netherlands following the Dutch 2007-2010 Q fever epidemic revealed incident Q fever infections to exceed notified infections by factors of ten and higher (Hackert et al., 2015(Hackert et al., ,2012Hogema et al., 2012;van der Hoek et al., 2012,). A syndromic surveillance study that retrospectively identified three clusters of lower respiratory infections dating from 2005 and 2006 plausibly linked to the Dutch epidemic appears to confirm that even clusters of more severe disease may easily be missed by physicians (van den Wijngaard et al., 2011).
In the case of the cross-border outbreak we describe here, limited attention paid to the outbreak in South Limburg by regional Limburg, suggesting that the higher seroprevalence rate observed in 2010 blood donors may reflect a real increase in Q fever infections.
Overall precision of our data for the German study area was limited. Sample sizes of blood donors per postcode were small, particularly in postcodes located at larger distances from the outbreak farm.
For lack of individual blood donor data such as residential address and travel patterns, we had to use postcode centroids as a proxy, resulting in low resolution and possibly misclassification regarding exposure location.
When interpreting seroprevalence and incidence rates in blood donors, one always needs to realize that the study population consists of adult, healthy blood donors, not of randomly selected citizens. However, while donors in many cases poorly represent the general population, infections incurred through airborne transmission are generally independent of donor status, reducing biases caused by the comparison of donors and the general population. A Dutch study among blood donors showed that the age and sex distribution of the study population was very similar to the age and sex distribution of the notified Q fever cases in the Netherlands . A recent Australian study found the seroprevalence in blood donors to be lower than in the general population, but indicates that different laboratory methods and population sampling may account for some of the differences (Gidding et al., 2019). Seroprevalence in our group of blood donors also may have underestimated the true seroprevalence in the general population, due to the selection process with excluded donor candidates with signs of acute or recent infection, and those with known risk exposures.
Ideally, our findings should be replicated by serological studies of preserved pre-and post-outbreak human samples from other Dutch-German border regions that likely were affected by the Dutch epidemic in 2007-2010. Meanwhile, in the absence of such studies, our findings argue for intensified and harmonized cross-border communicable disease control, including public health communication to professionals, public and media, as well as exchange of data suitable for surveillance, detection and early warning. In addition, we urgently recommend that patients, who live in affected areas and have predisposing conditions, serological evidence or clinical symptoms consistent with persistent focalized (chronic) Q fever infection, should be considered for low-threshold screening, keeping in mind that chronic Q fever may have atypical presentations (Melenotte, Million, & Raoult, 2020).