Population-based assessment of health, healthcare utilisation, and specific needs of Syrian migrants in Germany: what is the best sampling method?

Internal and external displacement due to the violent conflicts in the Middle East belongs to the most dramatic ongoing humanitarian emergencies in the last decades [1, 2]. Among the worst cases is the war in Syria, which not only lowered the population’s life expectancy by about 20 years but also forced a large number of people to migrate and seek refuge in other countries [3, 4]. Also after arrival in neighbouring or European host countries, migrants constitute an especially vulnerable group that suffers from a significant burden of disease [5]. They also may have special needs and specific characteristics with respect to legal status, health status, health-related risk factors, healthcare utilisation, barriers to access healthcare, or disadvantages because of low income [6‐8]. Therefore, host countries urgently need to create strategies and healthcare policies to cope with this novel situation and to tailor specific interventions that meet the needs of this vulnerable population ensuring quality healthcare [1, 9, 10]. As a basis for such strategies and policies, reliable scientific evidence is vitally required [11]. In Germany, however, although being the European country hosting the largest number of Syrian migrants, the special needs of migrants have been investigated very little with studies carried out so far assessing health mainly in small samples of asylum seekers living in accommodation centres limiting the generalisability of these results [12, 13].

For the purpose of obtaining representative results in population-based cross-sectional surveys the method of choice in Germany has traditionally been random sampling via the population registries as these registries capture the population almost gapless. Also migrants are recorded in these registries shortly after their arrival. Nevertheless, it is questionable if random sampling via the population registry is the optimal method for recruiting migrants. Previous studies using traditional sampling methods like random sampling via the population registry frequently reported the problem that survey response in migrants was even lower than the currently also decreasing willingness to respond among the autochthonous population [14, 15]. Migrants are hence considered as one of the populations most hard to reach for research [16]. The low response in this sub-population could be due to language barriers, fear of the individual answers being reported to immigration authorities, lack of time or accessibility, communication difficulties, or lack of translated study instruments [17‐20]. In addition, migrants frequently live hidden and under precarious conditions, partly without legal status [1]. For this reason they may not be fully covered in population registries [21], but may be captured by alternative sampling strategies.

In order to increase survey response in so-called hard-to-reach populations, new methods were developed [16]. One of them is the respondent-driven sampling strategy, a network sampling technique that starts with a small convenience sample of the target population (“seeds”) which is asked to complete the survey (potentially online) [22‐24]. They are then asked to invite a limited number of their contacts (“peer-recruited participants”) who are also members of the target population by using recruitment coupons [25]. This way, the sample is expanded in recruitment waves and the dependence on the initial convenience sample is reduced [26]. Participants receive incentives for completing the survey and for successfully recruiting other respondents [25]. This double-incentive strategy increases the motivation for the seed and the peer-recruited participant to complete the survey as well as recruit further participants.

In medical research, respondent-driven sampling has been mostly used to recruit drug users [25, 27], sex workers [28, 29], and men who have sex with men [30, 31]. Beyond that, respondent-driven sampling has been shown to be an effective tool for recruiting migrants in very diverse settings and populations such as migrants from the former Soviet Union living in Poland [32], mobile migrant workers in Thailand [33], or sub-Saharan migrants in Morocco [34]. In Germany, however, so far only one study in health research employed respondent-driven sampling as sampling strategy while it has not yet been employed for sampling migrant populations [35, 36].

Therefore, as preparation for a large-scale cross-sectional study on health-related aspects of Syrian migrants living in Germany, we aimed to identify the sampling strategy most suitable for obtaining representative results. More specifically, our objective was to assess whether respondent-driven sampling is a better sampling approach to gain access to the Syrian population in Germany than traditional random sampling via the population registry. For answering this study question, we aimed at evaluating the following criteria:

The present study intended to assess the most suited sampling method for a large-scale epidemiological study investigating health-related aspects in Syrian migrants living in Germany and to evaluate if respondent-driven sampling is a better means for this undertaking than traditional random sampling. To achieve this aim, we compared random sampling and respondent-driven sampling with respect to the absolute number of recruited participants, invitees’ willingness to participate and characteristics of the study populations. Our results provide evidence that selection bias plays a major role with different sampling methods yielding study populations with different characteristics indicating that respondent-driven sampling is very helpful to get access to specific subgroups of the target population that can hardly be reached by traditional random sampling.

Via random sampling through the population registry, only a small number of participants were recruited with a very low response of 14%. This number is yet a little bit lower in comparison with the response in another German cross-sectional feasibility study including asylum-seekers [13]. Given the already mentioned general difficulty in recruiting representative samples of migrant populations, those numbers are not entirely surprising. Nevertheless, such a low response is clearly not sufficient to be able to be reasonably sure to obtain representative results as such results are most likely affected by selection bias [13]. In order to obtain more representative study samples of migrants, previous studies recommended sub-group-specific activities or participatory methods including peers in data collection [13, 15].

Using respondent-driven sampling, we applied one of the strategies that actively involve peers in the recruitment process. With the relatively small numbers of 16 seeds only, a much larger number of participants could be included via respondent-driven sampling as compared to traditional random sampling. This suggests that respondent-driven sampling is a suitable method for recruiting a reasonable number of participants with comparatively moderate effort. Having said that, the comparison between the two sampling strategies with respect to the number of recruited participants is, of course, somewhat arbitrary. Whereas in respondent-driven sampling this number depends, among other aspects, on the number of seeds, the number of participants recruited via the population registry is subject to the size of the random sample drawn. Nevertheless, we think that the aspect of absolute numbers of participants is still worthwhile to be discussed as the findings of this pilot study at least provide some information, on how many seeds we need to choose or how large the random sample needs to be to get a certain number of participants. For example, applying the above-mentioned official number of about 4000 individuals with Syrian citizenship registered in Munich to our observed response of 13% would mean that in total we could expect to recruit about 500 participants via the population registry.

As another criterion for evaluating the feasibility of our sampling methods, we compared the two study populations with total sample drawn from the population registry used as a proxy for the characteristics of the target population with respect to gender and citizenship. In the total sample obtained from the registry and in the population registry based study population a substantially higher proportion of subjects held German citizenship than in the respondent-driven sampling study population. Additionally, comparing the two study populations, participants recruited via the population registry reported more often to hold a university degree, to have a larger social network and to be better acculturated than the study population obtained via respondent-driven sampling. Moreover, population registry participants were older, smoked less and indicated a higher prevalence of chronic diseases and more frequent healthcare utilisation than respondent-driven sampling participants. That population registry participants were older could be an explanation for the health and lifestyle-related differences between the two groups. In total, these results suggest that sampling via the population registry resulted in a study population that lived longer in the host country and had a higher level of adaption to the host society while respondent-driven sampling was more suited to identify younger migrants that are less well integrated. This may be an indication of undercoverage of the population registry especially missing more recently immigrated persons.

Our findings thus imply that population-based cross-sectional surveys investigating sociodemographic and health-related aspects of migrants may yield different results depending on the sampling method. This observation is not only helpful for the interpretation of existing studies but also for the planning of future research. Following this, to not only recruit a very specific subgroup of migrants depending on the recruitment method but to be able to obtain more representative results, a combination of probabilistic and non-probabilistic sampling techniques might be a worthwhile option. Such an approach could also include additional sampling strategies as quota sampling [32], cluster sampling [44], convenience sampling methods like block-walking [45], or other non-probability sampling techniques [46]. This might also help to overcome limitations of respondent-driven sampling such as inability to access individuals who are socially isolated, non-random selection of seeds, inaccurate reporting of the network size leading to biased results, access to the online questionnaire by ineligible participants, or difficulties to recruit samples with mixed ethnic structure [36, 47, 48]. Combining probabilistic and non-probabilistic sampling methods, however, has several potential pitfalls that would need careful consideration. For instance, potential participants may be invited to the study via two or more of the selected recruitment methods. This issue could be addressed by requiring participants to create a unique personal identifier when completing the questionnaire or by screening incoming questionnaires for identical information. Furthermore, such an approach would lead to complex datasets containing data from both randomly and non-randomly selected participants. Hence, researchers would need to apply appropriate techniques such as the application of ‘pseudo-weights’ [49] and statistical software packages especially designed for such complex data [50].

As far as the authors are aware, this is the first German study exploring the feasibility of respondent-driven sampling to recruit hard-to-reach-populations for medical research. It is also one of the very first projects to implement research on health status, healthcare utilisation, and specific needs of Syrian migrants living in Germany. Scrutinising these questions is of high public health relevance as it is very likely that migration will continue to be a major issue for many societies [51]. It imposes specific challenges with respect to humanitarian aspects and healthcare [52]. The latter is exacerbated by specific difficulties with respect to communication or cultural, legal, and bureaucratic hurdles that need to be taken into account [10, 53]. Despite those difficulties, elucidating specific needs of migrants is of uppermost importance not only for developing appropriate health policy strategies but also for unveiling health inequalities and for ensuring dignified treatment of the migrants [8, 54, 55].

When interpreting our findings, it should be taken into account that we conducted our study in one of the German cities with the biggest proportion of migrants and a relatively good infrastructure of migrant organisations. Hence, we cannot be entirely sure to what extent our findings can be generalised to other locations where migrants are lower in numbers, less networked, potentially living more hidden, and thus even harder to access, especially as respondent-driven sampling suffers from difficulties in recruiting isolated and sparsely networked persons [48]. Moreover, there is the potential pitfall of individuals being inclined to participate multiple times in order to get the financial incentives. We tried to control this by the use of individual access codes and by additionally checking if incoming questionnaires contained exactly the same answers as previous questionnaires or conspicuously implausible answer patterns, e.g. always selection of the first answer option. Another limitation of respondent-driven sampling is that it does not allow calculating a response rate. Hence, we could not directly compare the response rates between the two methods, what on the other hand makes the comparison of the characteristics of the two study populations even more important. As, the main objective of our study was to evaluate sampling methods, the distribution of the variables that we measured in our sample should, however, not be taken as a valid and reliable description of the characteristics of Syrian migrants living in Germany. If at all, our numbers can be seen as very preliminary first findings. In that regard, it also needs to be mentioned that the city of Munich belongs to the wealthiest locations in Germany. Thus, the characteristics of our study population, e.g. educational level, might be different to migrants living in less prosperous regions. To get reliable information on health status and healthcare needs of Syrians and other ethnic minority groups in Germany, carefully designed representative studies are urgently needed. For the planning of such investigations our study provides important information concerning the selection of the sampling and recruitment strategy.

As expected, response was very low among the random sample drawn from the population registry indicating a considerable level of selection bias. Using a relatively low number of seeds, a larger number of participants were recruited via respondent-driven sampling. The two recruitment strategies led to study populations with substantially different characteristics suggesting that population-based surveys investigating migrants’ health may provide different results depending on the sampling method. Therefore, our results indicate that respondent-driven sampling is a useful way to gain access to specific subgroups of the target population that are hard to reach via traditional random sampling. To avoid recruitment of only one very specific subgroup, future studies may consider mixed sampling approaches combining various probabilistic and non-probabilistic recruitment strategies.