1. Lithium in drinking water for prevention of male suicide Nobuyoshi Ishii, M.D., Takeshi Terao, M.D., Ph.D., Yasuo Araki, M.D., Kentaro Kohno, M.D., Yoshinori Mizokami, M.Sc., Ippei Shiotsuki, M.D., Koji Hatano, M.D., Mayu Makino, M.D., Kensuke Kodama, M.D., Noboru Iwata, Ph.D. Department of Neuropsychiatry, Oita University Faculty of Medicine, Oita, JAPAN

2. Introduction There is growing evidence from epidemiological studies that lithium in drinking water may protect against suicide. An inverse association between lithium levels in drinking water and suicide rates was first reported in the USA for 27 Texas counties (Schrauzer et al, 1990). In Japan, we showed the inverse association between lithium levels in drinking water and suicide rates for 18 municipalities of Oita prefecture (Ohgami et al, 2009) whereas another study found no association for 47 subdivisions in the East of England (Kabacs et al, 2011). A nationwide Austrian study, however, found an inverse association between lithium levels in drinking water and suicide rates after adjustment of population density, per capita income, proportion of Roman Catholics, as well as the availability of mental health service providers (Kapsuta et al, 2011). Recently, the Austrian group reconfirmed an inverse association between lithium levels in drinking water and suicide rates after adjustment of county-based population density, age, gender, race/ethnicity, median income per household, poverty and unemployment rates in Texas (Blüml et al, 2013). In the present study, we focused on Kyushu island as the development of our previous study (Ohgami et al, 2009) which investigated Oita prefecture which was one of 8 prefectures in Kyushu island. In 2011, the total, male, female population of Kyushu island was 14 646 121, 6 952 255, and 7 693 866, respectively. Kyushu island had 118 cities, 119 towns and 37 villages, totally 274 municipalities.

3. Methods Suicide data Taking the large differences in gender and age distribution of individual municipality populations into account, the standardized mortality ratio (SMR) of suicide in 2011 was calculated for each individual municipality. SMR is an indirect method of adjusting a mortality rate, defined as the number of observed deaths in an individual municipality population divided by the number of expected deaths compared with the gender- and age- matched general population. We examined the data from Ministry of Health Labour and Welfare on suicide, and from Statistics Bureau, Ministry of Internal Affairs and Communications, and calculated total, male and female suicide SMRs for 2011 across all the 274 municipalities. Measurement of lithium levels in drinking water From 2010 to 2013, tap water samples (mainly, from the main rail station and/or the municipal office) of each municipality were taken and their lithium levels were measured by using mass spectroscopy analyzed by a third party. This method can measure very small amounts of lithium; the minimal amount of lithium that can be measured is 0.1 ppb (0.1 μg/l). If lithium levels of drinking water were measured at multiple points in the same municipality, the mean value was calculated. Although lithium levels were measured once, we confirmed a very small fluctuation in levels because the correlation coefficient between the lithium levels and those re-measured after 1 year in the same places was 0.998 (Ohgami et al, 2009).

4. Adjustment factors Although we initiated a crude model of the association of lithium levels in drinking water and suicide SMRs (total, male and female) without any adjustment of the confounding factors (crude model), as afore- mentioned, the associations were further investigated adjusting for proportion of one-person households, proportion of ≧ college education, and proportion of engaging in primary industry (adjusted model 1). Finally, further adjustment was performed with complete unemployment rate, proportion of elderly people, annual marriage rate, annual mean temperature and annual postal savings per person (adjusted model 2). Statistical analysis When a procedure that treats all of the data equally would give less precisely measured data points more influence than they are entitled to and vice versa would give highly precise data points too little influence, weighted least squares regression analysis was performed by attempting to give each data point its proper influence over the parameter estimates. Also in the present study, due to greater differences in population size across the 274 municipalities, weighted least squares regression analysis adjusted for the size of each population was used to investigate the association of lithium levels in drinking water and the suicide SMRs. Multiple regression analyses were used to predict suicide SMRs (total, male and female) by lithium levels in drinking water and the adjustment factors. This study was approved by the ethics committee of Oita University Faculty of Medicine.

5. Results Table 1: Total suicide SMRs and lithium levels in drinking water ModelLithium and adjustment factorsβPAdjusted R 2 Crude modellithium−0.1750.0310.024 Adjusted model 1lithium propotion of elderly people proportion of one-person households proportion of ≧ college education proportion of engaging in primary industry −0.146 −0.297 0.301 −0.504 0.321 0.047 0.013 0.002 0.0003 0.004 0.199 Adjusted model 2lithium proportion of elderly people proportion of one-person households proportion of ≧ college education proportion of engaging in primary industry complete unemployment rate annual marriage rate annual mean temperature annual postal savings −0.118 −0.202 0.304 −0.229 0.507 0.319 −0.020 −0.087 −0.067 0.104 0.338 0.012 0.185 0.00004 0.001 0.908 0.414 0.435 0.243 Total suicide SMRs Lithium levels

6. Table 2: Male suicide SMRs and lithium levels in drinking water ModelLithium and adjustment factorsβPAdjusted R 2 Crude modellithium−0.2280.0050.046 Adjusted model 1lithium propotion of elderly people proportion of one-person households proportion of ≧ college education proportion of engaging in primary industry −0.195 −0.359 0.209 −0.517 0.276 0.008 0.002 0.027 0.00009 0.013 0.211 Adjusted model 2lithium proportion of elderly people proportion of one-person households proportion of ≧ college education proportion of engaging in primary industry complete unemployment rate annual marriage rate annual mean temperature annual postal savings −0.166 −0.203 0.189 −0.199 0.467 0.352 0.066 −0.064 −0.030 0.021 0.256 0.080 0.205 0.0001 0.0003 0.679 0.531 0.719 0.266 Table 3: Female suicide SMRs and lithium levels in drinking water ModelLithium and adjustment factorsβPAdjusted R 2 Crude modellithium0.0040.9570.000 Adjusted model 1lithium propotion of elderly people proportion of one-person households proportion of ≧ college education proportion of engaging in primary industry 0.023 0.067 0.263 −0.106 0.185 0.772 0.611 0.014 0.486 0.124 0.048 Adjusted model 2lithium proportion of elderly people proportion of one-person households proportion of ≧ college education proportion of engaging in primary industry complete unemployment rate annual marriage rate annual mean temperature annual postal savings 0.035 0.073 0.279 −0.026 0.240 0.079 −0.053 −0.060 −0.103 0.670 0.770 0.049 0.899 0.079 0.491 0.783 0.615 0.286 0.033

7. Discussion In the present study, lithium levels were significantly and inversely associated with total and male suicide SMRs across 274 municipalities in Kyushu island in the crude model and in the adjusted model 1 where the associations were further investigated adjusting for proportion of one-person households, proportion of ≧ college education, and proportion of engaging in primary industry. However, in the adjusted model 2 where further adjustment was performed with complete unemployment rate, proportion of elderly people, annual marriage rate, annual mean temperature and annual postal savings per person, male suicide SMRs, but not total or female suicide SMRs, remained significantly and inversely associated with lithium levels in drinking water. These findings suggest that male suicide SMRs may be inversely associated with lithium levels in drinking water and that the combination of the significant association of lithium with male suicide SMRs and the non-significant association of lithium with female SMRs might have produced significant associations of lithium with total suicide SMRs in the crude model and the adjusted model 1 but not in the adjusted model 2.

8. To the best of our knowledge, this is the first report demonstrating a gender difference of the inverse association between suicide rate and lithium levels in drinking water. Although there has been a report showing little gender difference in clinical response to lithium treatment of bipolar and related affective disorders (Viguera et al, 2000), with regard to suicide, among the lithium-unexposed population men had a 3.57-fold higher rate (95%CI, 3.12-4.07) of suicide than women whereas among patients purchasing lithium the rate of suicide did not differ significantly for men and women (Kessing et al, 2005). This report may indirectly indicate the gender difference in the anti-suicidal effect of lithium treatment (i.e., the anti-suicidal effect for men> that for women), which is in agreement of our study’s findings. At least partially, it appears likely that anti-aggressive effects of lithium could potentially prevent suicide in the general population in those who take lithium contained drinking water for a long time (Terao, 2008; Terao et al, 2009). If male aggression is greater and more responsive to lithium than female aggression, this could explain gender differences in the present findings, but more studies are required to further investigate this possibility. In conclusion, the present findings suggest that lithium in drinking water may be associated with the low risk of male suicide in the general population.