Participation in this AF screening program varied evidently between parishes, in particular among high risk individuals, i.e. individuals that would benefit the most from anticoagulation treatment in case of an AF diagnosis. In addition, low participation was linked to higher stroke incidence and higher proportion of immigrants in the target population.
The reasons for this geographical variation could be manifold. High risk individuals might have had disabilities which affected their participation. Individuals with dementia and individuals staying at nursing homes were most probably non-responders to the screening invitation, but also in most cases not eligible to OAC treatment in case of a new AF diagnosis. A more thorough examination or interview of non-responders could have identified patients with different disabilities making participation in the screening program difficult. These data was not always available in the medical records. However, the prevalence of a previous stroke, heart failure and diabetes mellitus was significantly higher among non-respondents which would imply higher risk of disability, but data on the level of disability after stroke was not part of our data collection.
The two parishes with the lowest participation in the screening program had the highest levels of elderly immigrants among the population. Decreased ability to understand the written invitation might have impaired participation. The invitation was written in Swedish and had no condensed information or links in other languages. The participants had to schedule their index visit via a telephone call to the study center. One might speculate that this procedure affected participation and that providing the participants with a pre-scheduled appointment could increase the response rate.
In our screening study, all examinations were performed at one hospital clinic (located in parish K, see Figure 1). We speculate that problem with getting to the clinic may have affected participation in our study. In a multi-practice, cluster randomized AF screening study in the UK comparing systematic screening with opportunistic screening and no screening in routine care among individuals with age > 65 years, participation in systematic screening was 53% but varied between 22 and 68% at practice level . Participation was higher among invited individuals < 75 years of age, but the yield of newly diagnosed AF was higher among those > 75 years of age. Among non-participants who stated their reason for not attending, problem with getting to the clinic was the most frequently stated reason . In an AF screening study among 75-year old inhabitants in a Norwegian community reported by Tveit and coworkers, participation was 82% . This study offered home visits to disabled inhabitants. It was reported that the study population had higher average income and higher education levels than national average. Both home visits and socioeconomic status might have influenced participation in this study .
Furthermore, our screening program was not accompanied by a media campaign.
In cancer screening settings, several factors that may influence participation have been addressed; and interventions as well as targeted actions for improving participation have been proposed [22–24].
Our results yielded useful information needed to intervene for improved screening uptake. The result points towards specific neighborhood areas for possible target actions. Yet, it might be feasible to use a modified invitation procedure for the total study group. Nevertheless, we anticipate that such an intervention will have greatest potential effect in the parishes with relatively high share of immigrants.
Although the share of immigrants and the share of non-participants not speaking Swedish fluently was unknown to us, the screening uptake might be increased by modifying the invitation letter, giving brief information and links in several languages.
Previous studies on screening for AF among elderly people have reported the significance of easy access to the screening center . Performing the ECG screening closer to the participant, for instance in collaboration with primary care might increase screening uptake. As an extension of this modification, performing ECG screening via home visit in selected cases might increase screening uptake as well, although selection of these cases might prove difficult. Screening partly via home visits would also increase screening costs and would perhaps negatively influence cost effectiveness.
Participation in established screening programs in Sweden is high. For instance, 83% of invited 65-year-old men accepted to participate in aortic abdominal aneurysm screening in the Uppland region in Sweden . Since our screening study was neither accompanied by a media campaign nor being part of an established screening procedure, screening uptake in a future routinely performed program might be higher than in this pilot study.
These measures will be subject to forthcoming studies. Neighborhood areas with a high proportion of immigrants among the elderly and elevated stroke incidence, which enforce screening aiming at stroke prevention, are the primary targets for action.
We have demonstrated that geo-maps on participation in our initial screening program for silent AF, along with target-population-based geo-maps on proportion of immigrants and ischemic stroke incidence, can provide valuable information in order to tailor efforts to improve participation in future screening programs.
Clinical and geographical data were not available from the 75 individuals who actively denied participation. If these 75 individuals were unevenly distributed among the 12 parishes, this might have altered our results slightly.
We were able to geo-code the study population into 12 different parishes. The empirical Bayes model applied means that we performed a “global” smoothing across the 12 parishes. A fully Bayseian approach provides an alternative option. By applying such an approach, a “local” smoothing is added, i.e. allowing also for dependency between rates in adjacent areas. A fully Bayesian approach is reasonable provided feasible geo-coding of the study population into several small areas. With the 12 different areas (parishes) considered, however, the fully Bayesian approach in Rapid Inquiry Facility along with free software for Bayesian data analysis, WinBugs  yielded an overly marked smoothing: For the total study group, those smoothed non-participation ratios varied between 0.95-1.06 across the 12 parishes and, for the high risk individuals, between 0.94-1.11.
Regarding individual-level data for the non-participants, there was limited access to socio-demographic characteristics due to ethical restrictions. In fact, the individuals of the study population could only be stratified on stroke risk factors (we used the CHADS2-score) and sex, in order the perform Bayesian smoothing of non-participation rates across the parishes. Provided richer data, geographically-weighted regression offers an interesting alternative method to identify spatial discrepancies in observed-to-expected ratios .
Data on thromboembolic risk factors among non-participants were collected from medical records which has certain limitations.
Our data were too spare for analyzing participation among high risk individuals defined by a more strict definition than CHADS2 ≥ 2.