Suicide is a serious public health problem accounting for 1.4% of all deaths in 2012 worldwide [1
]. In Denmark, around 600 people, corresponding to 1% of all deaths, commit suicide every year since 2005 [2
]. For comparison, this number is more than three times higher than the number of deaths due to traffic accidents [3
]. Apart from being a personal tragedy, suicide causes profound suffering for families, relatives, and communities [1
]. Suicide is a multifactorial event caused by a complex interaction between psychosocial, genetic, and environmental factors [4
]. While a wide range of psychosocial factors (such as abuse, loss, or stressful life events [5
]) have been investigated in terms of their influence on suicide, fewer studies have considered the potential impact of environmental exposures.
Lithium (Li) is a naturally occurring element in drinking water mainly originating from weathering of minerals in the subsurface. Levels in drinking water vary across the world for example with levels up to 12.9 µg/L in the Aomori prefecture in Japan [6
] and up to 219 µg/L in Texas [7
], yet everyone is exposed to some amount [8
]. In clinical practice it is well established that lithium has a mood-stabilizing and suicide-preventive effect in individuals suffering from affective disorders [9
]. Therapeutic doses vary from 600 to 2400 mg per day [11
] and are magnitudes higher than levels reported from both surface and groundwater sources of drinking water [12
]. However, the low lithium exposure through drinking water occurs continuously throughout the lifespan.
Recently, a series of ecological studies [6
] has addressed the potential effect of lithium in drinking water on suicide in the general population. Although evidence is pointing in the direction of an inverse association [19
], contradictory findings have been reported in different regions and a causal relation is uncertain. An essential limitation of previous studies [6
] is that they have been conducted at an aggregate regional level prone to the possibility of ecological fallacy, where associations may be falsely overestimated compared to those established by individual-level analysis [20
]. Therefore, for the first time, we investigate the association at an individual level using prospectively collected nationwide Danish register data. The unique personal identification number and the comprehensive Danish registers enabled follow-up of the entire Danish adult population (age ≥21 years, 3.7 million individuals) over the course of 22 years. All Danish drinking water is of groundwater origin and clear geographic patterns in lithium levels have been found [21
]. Individuals living in Eastern Denmark are exposed to lithium levels more than 10 times higher than individuals in Western Denmark [21
]. The individual-level data allow for investigation of the effect of a long-term exposure to lithium in drinking water, while accounting for people moving residences and thereby changing exposure. In addressing issues of spatial autocorrelation (see also Helbich et al. [14
]), where dependence among observations violates standard statistical techniques, spatial analyses were conducted using a Bayesian conditional autoregressive (CAR) model. Overall, the aim of the present study was to investigate the effect of naturally occurring lithium in drinking water on the incidence of suicide at an individual level, hypothesizing that lithium in drinking water has a potentially protective effect on the risk of suicide in the general population. The supposition is that people, who in periods have been exposed to a high level of lithium in their drinking water, have a stabilized mood without large fluctuations, which lowers their risk of later experiencing mood disturbances so severe that they end up taking their own life.
The stability of lithium levels in groundwater over time was evaluated (Table 1
). The results show that most of the variance in the lithium measurements was due to the anticipated variation between boreholes (74.7%) and between different sampling-depths within the same borehole (16.9%). The differences between measurements over time, i.e. the variation in lithium levels within the same borehole and sampling-depth on different days (residual variance 0.07), accounted for 8.4% of the total variation in the groundwater lithium measurements. As this is considered a relatively small proportion, the results of the analysis indicated that the lithium levels in Danish groundwater were reasonably stable over time.
The mean lithium level of the 151 drinking water measurements was 11.6 µg/L (standard deviation (SD): 6.8 µg/L) ranging from 0.6 µg/L in Western Denmark to 30.7 µg/L in Eastern Denmark (Figure 1
). The interpolation of the point data done by kriging rendered a prediction map and a corresponding prediction variance (i.e., error) map (Figure 2
). Similar prediction maps were found for all kriging models regardless of model specifications. Inverse distance weighting (IDW) was used as an alternative spatial interpolation method to validate the prediction obtained by kriging and rendered equivalent results. The mean drinking water lithium level for each of the 275 municipalities (based on the prediction map in Figure 2
A) is shown in Figure 3
Baseline characteristics of the study population on 1 January 1991 are shown in Table 2
by five-year TWA lithium exposure levels.
The study population consisted of a total of 3,740,113 individuals aged 21 years or older of which 51.5% were women. The cohort had a total of 66,813,931 person-years at risk during the study period. A total of 14,151 individuals, equivalent to 0.38% of the study population, committed suicide during the study period. The suicide rate decreased during the study period from 29.7 per 100,000 person-years in 1991 to 18.4 per 100,000 person-years in 2012. The suicide rates at the municipality level ranged from 6.1 to 36.0 suicides per 100,000 person-years at risk.
The spatial regression analysis showed no protective effect of five-year TWA lithium exposure level through drinking water on suicide rate adjusted for differences in gender, age, employment, civil status, and calendar year (Figure 4
). The results indicated an increasing suicide rate by increasing five-year TWA lithium exposure level. Estimates of the covariates are shown in the Supplementary Materials, Table S1
The supplementary analyses overall showed the same results as the main spatial regression analysis (Supplementary Materials, Table S2
). Repeating the main analysis with the 10-year TWA lithium exposure level showed similar results with no protective effect of an increasing five-year TWA lithium exposure level on suicide rate. The analysis excluding the capital area of Copenhagen, the semi-adjusted analysis and the non-spatial analysis showed the same overall trend as the spatial regression analysis. Changing the study design to a matched case-control study or to an ecological study showed similar results with no protective effect of an increasing five-year TWA lithium exposure level on the incidence of suicide.
Our study is the first to investigate the effect of naturally occurring lithium in drinking water on the incidence of suicide at an individual level with more than 20 years of follow-up. The comprehensiveness of our data and analyses makes a pronounced contribution to previous findings [19
] and demonstrates that there does not seem to be a protective effect of exposure to low levels of lithium on the incidence of suicide with lithium levels below 30 μg/L in drinking water.
The main advance of our study compared to previous ecological studies, is the use of prospectively collected individual-level data following the entire Danish adult population over the course of 22 years. The use of these data avoided selection bias [42
] and enabled computation of a five-year TWA lithium exposure level accounting for people moving residence and thereby changing exposure during the study period. Calculation of a five-year TWA lithium exposure level based on a one-time lithium measurement was possible as lithium levels were found to be stable over time when analyzing groundwater data. Further, it was possible to incorporate updated information on all covariates each year, making the confounder adjustment more exact. This was particularly important for the adjustment for employment and civil status, since especially changes in these factors might affect the incidence of suicide.
The lithium exposure assessment in the present study was based on drinking water samples from public waterworks. This is appropriate for Denmark, since the public water supply is the main source of drinking water, and bottled water consumption is amongst the lowest in the EU (20 L per person per year) [43
]. Kriging was used to estimate lithium exposure levels at locations that were not sampled. Kriging is an interpolation method developed for continuous variables. Stationarity is an assumption of kriging, which means that the spatial correlation structure (i.e., the semivariogram parameters) should be constant across locations within the study area. Gotway and Wolfinger showed that despite deviations from stationarity and misspecified semivariograms, the kriging estimates are relatively unbiased [46
]. Maps of kriging estimates of the lithium level in drinking water in the present study were very similar for different combinations of semivariogram parameters. Spatial interpolation using IDW as a non-parametric method also resulted in a similar map. It is therefore concluded that ordinary kriging used in the present study is a suitable method for deriving a map of estimated lithium levels. The interpolation of the lithium point measurements assumes that individuals are exposed to the level of lithium found at the waterworks close to their residential location. This assumption is reasonable due to the highly decentralized water supply in most areas of Denmark [47
]. This is, however, not true for the capital area of Copenhagen, where drinking water is supplied by several large waterworks outside the city and mixed before reaching the tap at the consumer. Yet, sensitivity analyses excluding individuals residing in the capital area of Copenhagen did not significantly alter the results. Since Copenhagen is the largest urban area in Denmark, the sensitivity analysis also indicated that confounding due to urbanicity is not likely to have occurred. Using actual water supply areas instead of interpolation of point measurements might increase the certainty of the lithium exposure estimation (e.g., [46
]). However, this approach was not used in the present study as lithium measurements were not available from all water supply areas.
The study also has some limitations. As mentioned previously, lithium in drinking water is not monitored routinely in Denmark. The lithium measurements used in the present study were obtained in a sampling campaign of 151 public waterworks supplying approximately half the Danish population. Although temporal stability of groundwater lithium measurements was seen, measurements of lithium in drinking water would be needed from more waterworks for a number of years to examine the stability of lithium in drinking water over time. Further, drinking water is not the only source of lithium exposure as the element is also present in some amounts in food, e.g., by uptake from vegetables through the soil [8
]. Addressing the issue of other potential sources of lithium intake would be relevant in future research. Additionally, it would be relevant to evaluate the potential effect of lithium prescriptions on the association, although a study from 2015 found that suicide and lithium levels in drinking water were not a function of lithium prescription rates across Austria [48
In the present study, Danish drinking water lithium levels were found to range from 0.6 to 30.7 µg/L with a mean level of 11.6 µg/L (median 10.5 µg/L) and a standard deviation (SD) of 6.8 µg/L. Where the mean level was comparable to levels found in previous studies [6
], the range in the present study was generally more narrow. Previous studies that found a significant association with suicide consistently reported the highest lithium exposure levels with up to 59 µg/L in the Oita prefecture in Japan [16
], 121 µg/L in Greece [13
], and 219 µg/L in Texas [7
]. Conversely, studies with the lowest levels up to a maximum of 12.9 µg/L in the Aomori prefecture in Japan [6
] and 21 µg/L in the east of England [15
] did not find an association, like in our present study in Denmark. The lack of variation in lithium levels in the present study may have challenged our analyses. Nevertheless, the results from our present study, together with the studies from Japan and the east of England, show that exposure to very low levels up to 31 µg/L does not seem to have an effect on the incidence of suicide.
Lithium’s biochemical mechanisms of action are complex and not fully understood. In addition to its mood-stabilizing, antidepressive, and antimanic effects in individuals with bipolar disorder, human studies suggest that lithium in therapeutic doses has an anti-suicidal effect [10
]. This may be mediated through its mood-stabilizing properties or directly through a reduction of aggressiveness and impulsivity, characteristics that are associated with an increased incidence of suicide [51
]. In our present comprehensive study, we did not find that low natural levels of lithium reduced the incidence of suicide. Thus, further studies are needed to investigate whether there is an association between suicide and natural lithium exposures higher than those observed in the present study.