Using country of origin to inform targeted tuberculosis screening in asylum seekers: a modelling study of screening data in a German federal state, 2002–2015

We used data from mandatory upon-entry screening for TB in asylum seekers between 2002 and 2015 from the main reception centre of the third largest federal state in Germany with about 10 million inhabitants (Baden Württemberg). The reception centre quasi-randomly received about 13% of newly arriving asylum seekers to Germany based on administrative quota. All newly arriving asylum seekers are registered with a unique identification number and are required to undergo a mandatory health examination according to national law [20] (§62 of the Asylum-Law in combination with §36 of the Infection Protection Act).

The health examination consists of symptom-based screening for all asylum seekers to identify communicable diseases and a compulsory chest X-ray (CXR) examination for all asylum seekers above 15 years (except pregnant women) regardless of the results of symptom-based screening. The protocol for children below 16 years consisted of a tuberculin skin test (2002–2010) and symptom-based screening (2011–2015). Pregnant women underwent screening for TB infection either by a tuberculin skin test or interferon-gamma release assay (IGRA) (2002–2015). For non-pregnant adults, the screening is not sequential and CXR can be regarded as the first screening test. Asylum seekers with presumptive TB undergo further diagnosis by means of a facultative chest computer scan, a sputum test (microscopy or PCR) and culture tests. Ultimately, an active TB case was defined by the fact that the physician decided to start treatment (including clinically diagnosed cases, smear-positive cases and/or culturally confirmed TB). Asylum seekers must reside in the reception centre until the final result of the mandatory TB screening is obtained and they are not transferred to counties and communes unless active TB is ruled out. Those diagnosed with active TB are put on treatment in a hospital setting and not transferred unless risk of transmission has been ruled out. Further details of the screening protocol (e.g. for children < 16 years and pregnant women) are provided elsewhere [15].

Estimates for TB incidence and prevalence (per 100,000 population) in asylum seekers’ countries of origin were taken from the WHO Global Burden of TB database [21]. Following classifications used by the European Centre for Disease Prevention and Control [22] and de Vries et al. [19, 23], we classified asylum seekers’ countries of origin as low-, intermediate-, or high-incidence country when its WHO estimated period-averaged (1990–2014) TB incidence (Additional file 1) ranged between 0 and 20, between 21 and 50, or > 50 per 100,000, respectively. This classification was used to calculate the proportion of asylum seekers in the screening data stemming from countries with a low or intermediate incidence of TB among the total population of asylum seekers screened in each year. Further categories with higher cut-offs coded as binary variables (≥ 100, ≥ 150, ≥ 200, ≥ 250 and ≥ 300 per 100,000, respectively) were additionally used for both descriptive and analytical purposes. We further used data on TB prevalence (all forms, per 100,000 population) from WHO as external (“prior”) information for a set of 11 countries for which it was possible to calculate the country-specific observed yield from the screening data with reasonable precision (see below for details).

We assessed the period-, country-specific and the annual yield of TB screening as well as the NNS to identify one case of TB in the German screening data as defined in Table 1.

Throughout, we adopted a Bayesian perspective, thus treating parameters as random variables and considering prior information (see also Table 1 for further details) [24]. To investigate the performance of the screening programme and the suggested link between TB burden in the country of origin and the yield of screening in asylum seekers, we applied four main strategies:

To compare the country-specific yield of TB in asylum seekers screened in Germany (2002–2015) with country-level WHO estimates of TB prevalence reported in the WHO global TB database (1990–2014), we chose countries based on following criteria: ≥ 5 cases identified during screening or ≥ 5000 asylum seekers screened, and availability of country-level estimates in the WHO global TB database. The determination was deemed to allow estimation of screening yield and regression model parameters with reasonable precision and resulted in a manageable set of countries for reporting of stratified analyses. Eleven countries in the screening data source fulfilled the two criteria: Afghanistan, Cameroon, Eritrea, Gambia, Georgia, Iraq, Macedonia, Pakistan, Russia, Somalia and Syria.

We then determined the prior distribution based on the WHO estimates for the TB prevalence of each country and year (Table 1) and obtained the posterior distribution for TB prevalence, interpreted as expected TB prevalence, based on the combined data. This expected TB prevalence was used to calculate the probability that the NNS of a certain country is above a given cut-off value t in order to inform the choice of thresholds for or against TB screening based on asylum seekers’ country of origin (Table 1).

We performed a sensitivity analysis using the most recent TB estimates for the year 2015 of the global burden of disease (GBD) study [26] as prior information instead of WHO estimates (see Additional file 1 for details).

The annual yield varied considerably in the observation period and ranged from 44 to 279 per 100,000 asylum seekers (Fig. 2a). The proportion of asylum seekers from low- and intermediate-incidence TB countries among the screened population ranged from 32.6% in the year 2004 to 65.4% in 2015 (Fig. 2b). Compared to the index year 2002, the number of asylum seekers screened in 2015 increased more than 10-fold, TB cases (2015) by 333%, and the proportion of asylum seekers from low- and intermediate-incidence TB countries by 38.2%, while the yield (per 100,000) decreased by 62.3% (Fig. 3).

The annual proportion of asylum seekers from low- and intermediate-incidence TB countries among the screened population was negatively correlated with annual TB yield (rho: − 0.08). The overall yield of TB in asylum seekers from countries with a WHO-estimated high incidence of TB was 4.17 times the yield in those from countries with a low and intermediate incidence of TB, adjusted for age, sex and period effects (Table 3).

The difference in observed yield among asylum seekers from different countries of origin decreased gradually when higher cut-offs of WHO-estimated incidence of TB were used to determine comparison groups (Fig. 4).

The estimated yield of screening in asylum seekers from the majority of countries was generally below 200 per 100,000, except for Cameroon, Georgia, Gambia, Sudan, Vietnam, Eritrea and Somalia. A considerable number of individuals, up to 10,000 and more asylum seekers from Iraq, Syria and the Kosovo respectively, were screened with a relatively low yield of TB (Fig. 5).

The WHO-estimated country-level prevalence of TB was close to the observed country-specific period yield of screening (2002–2015) in asylum seekers from Cameroon, Georgia, Russia, Somalia and Macedonia (Additional file 1: Figure S1). The observed period yield of screening was higher than the reported WHO estimates for asylum seekers from Eritrea and Gambia, while for asylum seekers from Afghanistan and Iraq the observed TB yield was lower (Additional file 1: Figure S1).

Introducing a threshold for screening to increase the pre-test probability would lead to “missed” TB cases (i.e. false negative individuals), and the absolute number would increase with rising screening threshold (Fig. 5a). For example, a threshold of 50 per 100,000 based on WHO-reported incidence of TB would lead to 21 (22%) undetected TB cases (Fig. 6a). However, a total of about 66,700 individuals would be exempt from screening at this threshold (Fig. 6b), corresponding to about 0.3 undetected TB cases per 1000 individuals not screened (Fig. 6c). Considering the time period (13 years), the rate of false negatives at this threshold corresponds to 2.4 undetected TB cases per 100,000 individuals per year. Increasing the threshold up-to 100 per 100,000 would lead to a marginal increase in the number of undetected TB cases per 1000 individuals, but higher thresholds come along with a higher number of undetected cases per 1000 and entail a much larger trade-off between gain in pre-test probability and loss in sensitivity (Fig. 6c).

Comparison of the different distributions of observed country-specific yield and WHO-estimated TB prevalence showed that the use of the prior information resulted in narrower density distributions of the posterior distribution and thus improved prediction for eight of the eleven countries (Fig. 7). For three countries (Afghanistan, Eritrea and Iraq) the agreement between WHO and the screening data appeared to be smaller than for the remainder of countries.

To use this information for predictions of screening efficiency, we plotted the probability to lie above a given NNS value (ranging from 0 to 50,000) for each country (Fig. 8). This showed that, for example in asylum seekers from Iraq, Macedonia and Syria, the probability is more than 80% that the NNS is higher than 2000 (a threshold that has been used in the Netherlands and Finland to decide whether or not screening should be initiated [8, 19]). The probability that the NNS is higher than 500 is less than 10% for asylum seekers from Cameroon, and close to zero for asylum seekers from Pakistan and Somalia. The expected screening efficiency in a given population can thus be estimated based on different NNS cut-offs.

Using GBD estimates of TB prevalence instead of WHO data as prior improved the prediction of TB only for three countries (Eritrea, Iraq, and Pakistan). In the sensitivity analysis, the German screening data had little influence on the posterior distribution (Additional file 1: Figure S3,) and the GBD priors had substantially higher influence on the posterior distribution due to their considerably narrower uncertainty bounds (Additional file 1: Figure S4).

Only few countries perform targeted screening based on TB incidence in the country of origin (e.g. the United Kingdom and Switzerland), and incidence thresholds at which screening is initiated vary, ranging from > 15, > 40, > 50 to > 100 per 100,000 population [4, 8]. In Germany, no targeted approach exists and all migrants in collective accommodation centers aged 16 years and above undergo compulsory screening (except for pregnant women and children who may be considered for screening based on sub-national regulations [29]). The results of our study question such indiscriminate screening policies. Other countries use NNS thresholds to decide whether or not to continue or discontinue screening in a given population. In the Netherlands, for example, screening is ceased at an estimated NNS > 2000 [19]. These thresholds are, however, to a certain degree arbitrary and based on practical experience and the question is how to best choose a threshold. Making decisions about targeted screening programmes requires that questions around sensitivity, timing of screening, and efficiency and cost-effectiveness are addressed.

Our study provides an empirically derived alternative to previous decision-making for or against screening in this context. Limiting screening to a smaller fraction of asylum seekers will, by necessity, miss some rare cases originating from low-incidence countries. The important normative question at the societal or health system level is therefore what is valued as a good or acceptable balance between sensitivity of screening and efficiency and cost-effectiveness of screening. Introducing a threshold means to abandon the traditional idea that entry screening should detect all TB cases among migrants, which is likely to be an inefficient and perhaps also an unrealistic target. Based on the analysis of change in sensitivity when increasing the pre-test probability, our data suggests that a threshold, based on WHO-reported TB incidence between 50 and 100 per 100,000, would entail the lowest trade-off between loss in sensitivity and efficiency of screening. When introducing a screening threshold at 50 per 100,000, the number of false negative individuals, i.e. asylum seekers with TB going unnoticed, per 1000 individuals is relatively low (although it cannot be ruled out that these would have been also identified due to clinical symptoms or contact investigations). While screening all individuals upon-entry is an approach that might be feasible in times of low immigration, it is technically, practically, ethically, and economically questionable, especially in times when numbers of immigrants peak. Passive case finding approaches and improved access to primary care [30] could complement a targeted approach at such a threshold to ensure that the few individuals with TB among those exempt from screening are noticed as soon as an infection turns into disease and becomes symptomatic. If the purpose of a programme is to increase efficiency, the threshold could probably be even increased to 100 per 100,000 without substantial loss in sensitivity.

Furthermore, the estimated country-specific proportion of detected TB cases conditional on NNS-based cut-offs allows assessing the impact on screening efficiency when using different NNS thresholds for screening of asylum seekers from a given country. The data provided by our study can be used to calculate such targets and estimate marginal costs of finding additional TB cases in asylum seekers from low-incidence countries to guide such decisions. Even with universal screening and more if screening is selective, the country has to expect that some TB cases will be undetected and will appear after entry. Research in a large cohort of immigrants to the United Kingdom has shown that the risk of TB incidence among immigrants screened negative before entry is highest during the first two to 4 years after immigration [17]. Furthermore, the proportion of cases detected before entry are relatively low compared to those developing TB after immigration [17]. Data from Germany is consistent with these findings, showing that cases detected upon entry are only a small fraction of cases becoming incident in the following years [31]. Providing universal access to health care and social protection [32] and complementing targeted upon-entry screening with adequate (i.e. culturally, ethically and economically acceptable) programmes for post-migration follow-up for TB [16, 33] is thus of crucial importance, also for TB control in Germany.

Our study adds further complexity to the question of “pre-entry, post-entry or no tuberculosis screening” [11] among immigrants by considering asylum seekers as a socially constructed, heterogeneous group [16]. Asking whom to screen among asylum seekers instead of whether or not to screen [11] all would be of special importance for countries at the external borders of the EU such as Italy, Greece and Spain with a high number of forced migrants arriving every year. These countries currently apply an ‘all or nothing principle’ with respect to TB screening (Greece and Spain actively screening all, and Italy none of the incoming refugees [8, 9]). A more nuanced approach would consider the expected TB prevalence based on asylum seekers’ countries of origin. Such an approach should be closely linked with the question of the design and optimal timing for post-migration follow-up [17, 33] screening for asylum seekers from countries at highest risk for TB [16].

Beyond efficiency, our study has also implications with respect to equity and ethical aspects. Screening strategies that take into account country-specific risk avoid overprovision of services [34], allow to allocate resources where needs are highest (vertical equity), and are therefore an essential step towards tailored and appropriate high-value care [35, 36] for asylum seekers. Targeted screening is, however, a double-edged sword. Ethical implications from strategies targeting asylum seekers based on country of origin are to avoid stigmatisation. This requires that clinicians, policy makers, politicians and public health services effectively communicate the fact that these groups area “at higher risk” of having TB, not “a higher risk” for importing TB [37].

Prediction algorithms for targeted screening could be further improved by combining clinical, diagnostic [2, 13] and country-specific parameters. A combination with age, sex and socioeconomic status could further improve estimates of expected yield. This is especially relevant to further decrease the potential loss in sensitivity, and reduce the trade-off between sensitivity and high pre-test probability. Clinical information on co-morbidities (e.g. HIV, or diabetes) and socioeconomic factors were, however, not available in our data. The empirical derivation of such algorithms and their validation is further complicated because TB is a rare disease (thus, large samples are required) and available data often contain little detail on personal characteristics. We were able to estimate parameters for only 11 of 81 countries with reasonable precision. Pooling of large datasets across countries with comparable screening protocols is needed to enable further data analysis, especially as the overall yield in our study was lower than those reported in other studies on asylum seekers [3, 10]. This would allow replicating and applying our approach to other countries of origin. Initiatives such as the E-DETECT TB project co-funded by the European Commission [8] may be instrumental to this end. Furthermore, cost-effectiveness studies based on the data on thresholds and NNS provided in this study, or in studies with a larger sample size with more TB cases could be performed to generate further guidance for the decision-making towards targeted screening programmes.

We used a mono-centric data source spanning 14 years of upon-entry TB screening in one of the largest German federal states, which is comparable in size and population to Belgium or the Netherlands. The sample can be seen as representative with respect to age and sex distribution for asylum-seekers in Germany. The composition of the target population with respect to countries of origin may vary between federal states due to administrative procedures in the asylum process. However, this variation affects only few countries of origin, and the majority of asylum seekers are distributed to federal states based on a quasi-random administrative process [29]. It is hence likely that the composition of the population over 13 years resembles those of Germany as a whole, although direct comparison with administrative data provided by the Federal Agency for Migration and Refugees is difficult.

We only focused on active TB cases as the primary aim of upon-entry screening is to avoid transmissions in asylum seekers’ shelters. Including asylum seekers with latent TB in the analysis would have increased the yield estimates, but would have been less relevant for informing screening programmes.

Despite the large sample used the generalizability to other EU countries may be limited due to the heterogeneity in screening protocols [8]. It may also be the case that countries of first arrival of asylum seekers in Southern Europe have different (and higher) screening yields. However, we are not aware of any systematic analysis of differences in screening yield depending on the migration trajectory.

Given the low yield of TB screening and the uncertainty around estimates of observed screening yield, external evidence on TB burden should generally be considered in decisions for or against screening of asylum seekers from a given country since the uncertainty can be reduced. Depending on the general agreement between prevalence in country of origin and empirical screening data this may be sensible for some countries (e.g. Syria), but not for all countries (e.g. Afghanistan). It generally appears sensible where no or very sparse evidence exists from screening studies, where estimated TB prevalence and observed screening yield are not in conflict, and where no evidence exists that the screened population is substantially different from the general population with respect to TB risk. It might be sensible to add some “extra-uncertainty” to the country-specific prior (i.e., not applying the prevalence as the prior distribution 1:1 but to make it less informative) to consider the possibility that asylum and general populations differ with respect to TB risk. We have shown that WHO data is more reasonable to use as prior information instead of GBD estimates. The uncertainty in WHO data is a study limitation, and using period-averaged estimates as proxy of country-specific TB risk may not reflect the “true” risk at time of emigration. But this approach seemed as both practicable and reasonable for the vast population in our sample. Although GBD estimates appear more robust, and thus intuitively better, their narrow uncertainty bounds imply such a high certainty that no other source of information (in this case observed TB prevalence taken from screening data) was able to influence the distribution of data. This was not unproblematic as the deviance between observed screening data and prior information was higher for GBD data than for WHO data. The reason for this may be the use of period prevalences for both WHO (1994–2014) and German screening data (2002–2015), while using point prevalences (2015) for GBD data.