We examined the association between magnesium in drinking water and incident AF in Denmark. Though we found a wide contrast between low and high exposure to magnesium in drinking (ranging from 0.5 to 62 mg/L, mean = 13.9 mg/L) in this nationwide register-based study, we found no strong association between magnesium in drinking water and incident AF. AF incidence may be lower when exposed to a magnesium concentration of 5–10 mg/L in drinking water, compared to an exposure of 0.1–5 mg/L, though an overall small positive association was found. Higher incidence rates were observed among those aged 80 years and more, and among those in the lowest education group. An inverse association was found among those in the highest education group for all exposure groups compared to the lowest exposure group.
The estimated magnesium exposure interval (0.1–62.0 mg/L) and mean (13.9 mg/L) in the present study was wider compared to previous studies from the Nordic countries and the Netherlands on magnesium in drinking water and CVD, e.g. 1.17–5.87 mg/L (mean = 3.0 mg/L) in Finland , and 1.7–26.2 mg/L (mean 6.8 mg/L) in The Netherlands . However, in England a wider exposure interval was estimated: 2–111 mg/L (mean 19 mg/L) .
Our findings of no clear inverse association support the earlier findings in studies of incident CVD, where no clear association between magnesium in drinking water and incident acute myocardial infarct [22, 46] and incident coronary heart disease  was found. A linear inverse dose-response effect up to about 8 mg/L  has been suggested for the association between magnesium in drinking water and cardiovascular mortality. However, our results suggest an unexpected positive association, though we do observe a lower risk of AF when exposed to 5–10 mg/L compared to less than 5 mg/L. Although when sub-dividing magnesium exposure into 7 groups, the lower risk of AF was observed only at 5–7.5 mg/L compared to less than 5 mg/L. A positive association has also been found by Morris, Walker , between coronary heart disease incidence and a two-fold increase in magnesium intake from tap water (Hazard ratio (HR) = 1.10; 95% CI: 1.00–1.20), adjusted for age and other cardiovascular risk factors. No dose-response relationship was found when dividing magnesium exposure into tertiles (magnesium concentration: < 5.3 mg/day, 5.3–14.1 mg/day, > 14.1 mg/day).
The recommended daily intake (RDI) of magnesium ranges from 280 to 420 mg . Assuming an estimated daily tap water consumption of 1.85 L (estimated for the Swedish adult population, including boiled water ), the median magnesium intake in the highest exposure group accounts for around 10% of the RDI (9.9% for highest RDI - 14.9% for lowest RDI). In contrast, the coverage of daily magnesium intake through tap water was only 2% of the daily intake among adults aged 55–69 in the Netherlands , where no overall association was found between magnesium in tap water and mortality due to ischemic heart disease or stroke. Though their magnesium estimate was based on personal questionnaires and therefore expected to be more precise as a snapshot estimate, our magnesium estimate may be better as a long-term exposure estimate.
Both the identified interaction between magnesium and age and education, and the relatively different exposure-response pattern in the age stratified analyses, indicate that the association between increasing magnesium in drinking water and increasing AF incidence may be stronger for the oldest age group and lowest education group. However, since a too high body magnesium level is rare , it is probably something else that may explain the findings of a positive association. A possible explanation of the positive association might be due to different unaccounted causes of AF between the different age groups, e.g. hypertension. Furthermore, if the geographical variation in unaccounted risk resembles the geographical variation in magnesium in drinking water it could potentially influence the association. In addition, we found a lower risk of incident AF among individuals in the highest education group when exposed to > 5 mg/L as compared to 0.1–5 mg/L. In general, individuals with a high educational level have less comorbidities and better health behavior compared to individuals with a low educational level. Therefore, it suggests a small positive effect of magnesium in drinking water in the absence of strong individual risk factors and comorbidities (i.e. among individuals with a high educational level).
To our knowledge this is the first study on the association between magnesium in drinking water and incidence of AF and one of the first studies to use detailed and long-term registry data on both outcome, covariates, and magnesium exposure. The main strengths of the present study are the large study population, covering the entire Danish nation with administratively collected data selected independently of the aim of the study which limits selection bias  and a wide variability in magnesium exposure levels. In addition, AF is a well-defined diagnosis with a positive predictive value of AF at 92.6% in the National Patient register . The geographical and temporal variations in magnesium in drinking water are well-studied , and the long time series of exposure and AF make it possible to follow changes in exposure category for each individual, by including changes in residential address. Furthermore, a decentralized water supply structure in Denmark results in relatively small water supply areas, which to some extent decreases the risk of misclassification.
Individuals with missing information on residential address prior to the study period were excluded (1987–2001), based on the possibility that incident AF may have occurred when living abroad. Since AF can be a less severe disease, it is reasonable that not all individuals will register AF at the hospitals upon returning to Denmark.
In this study magnesium in drinking water at water supply area level was used as a proxy for the actual individual magnesium intake. In view of this, a common limitation in studies on drinking water and disease is the ecological nature of exposure which introduces a possibility of misclassification. In our register-based study, we were not able to include individual intake of magnesium from drinking water. Furthermore, the frequency of magnesium measurements at each waterworks ranges  and the water supply areas are assumed constant in time. No statistically significant difference was found between magnesium concentration measured at household and at waterworks, respectively . The regional concentration trend in magnesium in drinking water increases the likelihood for consuming the same magnesium concentration in drinking water at e.g. work compared to at the residential address. The two sensitivity analyses, one with only addresses within water supply areas with at least one water sample per every second year, and one with only addresses within water supply areas with constant concentration in time, did not change the results. The data file with the geographical distribution of water supply areas was created in 2013–2014 , with minor updates in 2017 , and are assumed stationary (2002–2015). Water supply areas covering more than one waterworks, might have been divided into smaller water supply areas prior to 2013, and magnesium concentration in drinking water might therefore have differed within the water supply area which has not been accounted for in this study. This might have led to misclassification of magnesium exposure, especially for years before and after the period where the water supply area data file was created. However, the patterns of magnesium concentration in drinking water make it likely that neighboring waterworks have similar magnesium levels, reducing the risk of misclassification.
Altogether, exposure misclassification in our study is expected to be independent of the incidence of AF and the effect on the result would be towards no association between magnesium in drinking water and incidence of AF.
AF is expected to be underestimated in the population , since AF may exist subclinical asymptomatic. Thereby, only the more severe or persistent AF is registered in the health registers. If the geographical variation in subclinical asymptomatic AF correlates with the magnesium exposure groups, this could potentially introduce a bias. Furthermore, little is known about the association between magnesium in drinking water and the latency time to AF. It has been suggested that it is the present magnesium exposure that is relevant for an effect . If a more instant response is expected, an error may have been introduced by calculating a 5-year mean magnesium concentration. However, by repeating the analyses using both a 1-year and 2-year weighted mean, we have accounted for the possibility of a short-term effect. On the other hand, if the development of AF is expected to be caused by long-term exposure to low level of magnesium in drinking water, the analyses where only the first exposure group was included may be more relevant.
When using administratively collected data, it is a common limitation that not all desired confounders are available in the registers . It seems unlikely that individual risk factors for incident AF effect the magnesium concentration in drinking water. To support this, no strong effect on the results was observed, when adjusting for socioeconomic position. However, the larger regional difference in magnesium concentration in drinking water may lead to the possibility of correlation between magnesium in drinking water at water supply area level and possible AF risk factors with similar geographical patterns at a regional scale. This could be an explanation for the findings of an overall positive association between magnesium in drinking water and increasing incidence of AF. While individual risk factors are important to consider, the place where each individual lives also has some significance . Geographical variations in contextual risk factors, e.g. environmental exposures or regional health service structures, could correlate with the geographical variation in magnesium exposure. Furthermore, Dummer  has highlighted the importance of understanding the geographical variation in health services and environmental exposures and their interrelations, when working with health-related risk exposures. Our results suggest that there are risk factors operating at a regional level that effect our results, since an inverse rather than a positive association between magnesium in drinking water and incident AF was found, when restricting the analysis to one administrative region.
Magnesium in drinking water may correlate with other inorganic chemical compounds in drinking water, and it is therefore plausible that these other compounds partly explain the results. In Denmark, magnesium in drinking water correlates with e.g. calcium . However, it has been concluded in several studies that the association with CVD is more likely to be related to magnesium than calcium in drinking water [15, 46]. Other possible chemical compounds that may be associated with AF risk factors and correlate with magnesium in drinking water include e.g. sodium. However, compared to the dietary sodium intake, the fraction from drinking water is small .
It has been argued, that the effect of magnesium in drinking water may only be found in populations where the dietary magnesium intake is insufficient , though by including a large population, we would expect to detect even a small effect, if any. In the age-stratified analyses, we saw a different association in the age group ≥80 years, which is considered as a more vulnerable group in relation to magnesium deficiency .