Extremes in water availability and suicide: evidence from a nationally representative sample of rural Indian adults

Abstract

Background:

Extremes in water availability, either exceptionally wet or dry conditions, can damage crops and may detrimentally affect the livelihood and well-being of people engaged in agriculture. We estimated the effect of water availability on suicide in rural India, a context where the majority of households are dependent upon agriculture.

Methods:

We used data from a nationally representative sample of 8.5 million people who were monitored for causes of death from 2001 to 2013. Water availability was measured with high-resolution precipitation and temperature data (i.e., the Standardized Precipitation Evapotranspiration Index). We used a fixed effects approach that modeled changes in water availability within districts (n=569) over time (n=13 years) to estimate the impact on suicide deaths. We restricted our analysis to rural areas and to deaths occurring during the growing season (June - March) among adults aged 15 or older, and controlled for sex, age, region, and year. We used Poisson regression with standard errors clustered at the district level and total deaths as the offset.

Results:

There were 9,456 suicides and 249,786 total deaths in our study population between 2001 and 2013. Compared to normal growing seasons, the percent of deaths due to suicide increased by 18.7% during extremely wet growing seasons (95% CI: 6.2, 31.2) and by 3.6% during extremely dry growing seasons (95% CI: −17.9, 25.0). We found that effects varied by age.

Conclusions:

We found extremes is water availability associated with an increase in suicide. Abnormally wet growing conditions may play an important, yet overlooked, role in suicide among rural Indian adults.

INTRODUCTION

In 2016, approximately 230,000 Indian adults died by suicide, which accounted for more than one-quarter of all suicides worldwide.¹ Suicide rates, exceptionally high throughout India, vary substantially by geographical area. Age-standardized rates are approximately three times higher in southern states compared with the rest of India, and rates are approximately two times higher among rural Indians compared with urban Indians.² This geographical patterning indicates that large-scale social, economic, and environmental determinants may play a key role in suicide risk.

Extremes in water availability, either very dry (drought) or very wet (flood) conditions, is a potentially important environmental driver of suicide in rural India. Agriculture is a source of livelihood for a majority of rural Indian households,³ and extremes in water availability are linked with reduced crop yields,^4,5 which pose major threats to livelihoods and well-being. National statistics collected from the National Crime Records Bureau (NCRB) find that bankruptcy and indebtedness are the top causes of suicide among farmers and cultivators.⁶ Poor crop yield may be especially detrimental to those engaged in subsistence agriculture: a study from the Indian state of Maharashtra found that crop loss and indebtedness were the main causes of suicide among small-scale farmers (88%) but not large-scale farmers (11%).⁷

While there are compelling reasons why extremes in water availability may affect suicide, strong population-based empirical research from India is lacking. A population-based study using suicide counts from the National Crime Records Bureau (NCRB) found that high temperatures – which the author inferred may damage crops – were associated with suicide.⁸ While this population-based study provides the most rigorous evidence to-date, the study had some noted limitations, including using temperature as a proxy for growing conditions⁹ and using suicide data from the NCRB.¹⁰ Using temperature to infer crop damage is challenging because there are well-established links between high temperature and suicide in general (non-agrarian) populations across the globe,^11-15 and thus observed effects may be due to high temperature and not crop damage. In addition, using NCRB data presents challenges because suicide is highly stigmatized in India, and validation work indicates the NCRB underestimates suicide by at least 25% among men and 36% among women – and even more so among young women –² which may lead to biased estimates.

In this study, we build upon prior research to provide a more comprehensive assessment of the link between water availability and suicide. We use a validated measure of water availability measured at the local (district) level linked with a nationally representative sample of 8.5 million people whose cause of death was ascertained through comprehensive procedures. Using this dataset, we use temporal and spatial variation in water availability to estimate the effect of water availability on suicide among rural Indian adults.

METHODS

Study population

Data came from the Million Deaths Study (MDS), which monitored the causes of death between the years 2001 and 2013 among approximately 8.5 million individuals living in randomly selected rural ‘small areas’.^16-18 These small areas were delineated by the decennial Indian census, which parsed all of India, encompassing India’s 28 states and 7 union territories, into approximately 1 million small areas. Using these small areas defined by the census, the Registrar General of India’s Sample Registration System (SRS) randomly selected urban and rural small areas to be monitored for all deaths twice a year. We restricted our sample to rural areas, which were defined by the Indian census as areas that met at least one of the following criteria: i) the area had less than 5,000 people, ii) more than 25% of the male working population engaged in agricultural pursuits, or iii) the area had a density less than 1000 per square mile.¹⁹

Rural areas were randomly selected by the SRS twice. In 2001, 4410 rural sample units, covering approximately 5.0 million people, were selected from the small areas defined by the 1991 census.¹⁶ In 2004, 4433 rural sample units, covering approximately 3.5 million people, were selected from the small areas defined by the 2001 census.¹⁷ Each small rural area was composed of approximately 150 households. We used data from both sampling frames (i.e., 2001-2003 & 2004-2013). The selected SRS areas are representative of the Indian population at the state level. Because suicide is rare among children, we restricted our analysis to people aged 15 or older. Our main unit of analysis was district, which on average contained approximately eight small rural areas selected by the SRS.

We included all districts that had small areas classified as rural by the SRS and were included in the MDS study. The MDS encompasses all of India, which was divided into 605 districts at the time of the study. Due to the small geographical area covered in the districts encompassing Delhi (n=9) and Mumbai (n=2), we combined these areas so that there was one district representing Delhi and one representing Mumbai, resulting in 594 districts. We also excluded three districts because they were island territories (i.e., Andaman and Nicobar; Lakshadweep) because effects may be different among people living on islands. Among the remaining 591 districts, 575 had rural small areas that were selected and followed by the SRS. District boundaries within India change over time, and we used district definitions preceding our study, from approximately 1998, so that we could model changes in consistent districts over time. We linked these districts with water availability data (i.e., SPEI) derived from another dataset, which is described in the variables section. Within India, districts have multiple names and multiple alterative spellings, and in a few cases (n = 6) we could not successfully link districts from these two data sources. Our final sample included 569 districts.

Variables

The underlying cause of death for all deaths occurring in the selected small areas was determined with an enhanced and validated form of verbal autopsy based on the World Health Organization’s 2012 instrument,²⁰ and which includes structured questions and a half-page local language narrative of the symptoms and treatments preceding death. In each home with a recorded death, a non-medical SRS field surveyor conducted a structured interview with the next of kin.^16,21 The completed forms were randomly assigned to two independent physicians who were specially trained to determine the underlying cause of death.^16,18 In cases where physicians did not agree on the cause of death, disagreements were adjudicated by a senior third physician. Cause of death was categorized using the International Classification of Disease, 10^th edition (ICD-10), and suicide deaths were defined as those eventually assigned ICD-10 codes X60 to X84 (intentional self-harm). The SRS field surveyor also collected basic demographic information for each decedent, including age, gender, and urban or rural location.

The exposure was water availability during the summer (Kharif) and winter (Rabi) growing seasons, which we combined into one 10-month growing period from June 1 to March 31. Water availability was measured with the Standardized Precipitation Evapotranspiration Index (SPEI), which is derived by taking the difference between precipitation amount and potential evapotranspiration during a specified time period.²² Thus, values at approximately zero indicate normal water availability, positive values denote a surplus in water, and negative values denote a dearth in water.²³ Validation work indicates that the SPEI can detect both extremely wet and dry conditions. A study in China found that positive (wet) and negative (dry) SPEI values corresponded to major drought and flood events recorded in the historical record,²⁴ and a study comparing SPEI values with ecosystem productivity found them highly correlated, especially in the Southern Hemisphere (r = 0.93).²⁵ Another study compared a number of drought indices (i.e., the Standardized Precipitation Index, plus four versions of the Palmer drought severity index) with streamflow, soil moisture, tree ring width, and wheat crop yields, and found that the SPEI most accurately captures drought.²³

We derived SPEI values using high-resolution (i.e., 0.05° degree) daily precipitation and temperature data, which we used to calculate average district-level SPEI values during the growing season (June 1 - March 31) for each year during our study period. Methods describing how these values are derived is available elsewhere, and the data are available publicly.²⁶ We used established cut-points to create indicator variables for moderate (−0.8 to −1.2), severe (−1.2 to −1.6), extreme (−1.6 to −2.0), and exceptional (−2.0 or less) drought,²⁷ and we applied corresponding cut-points to classify moderate (0.8 to 1.2), severe (1.2 to 1.6), extreme (1.6 to 2.0), and exceptional (2.0 or more) wet conditions. Normal conditions were defined as SPEI values between −0.8 and 0.8.

Because prior research found that crop loss due to extremes in water availability was associated with suicide in certain regions,²⁸ we created an indicator variable for region, which was defined as Central (Chhattisgarh, Madhya Pradesh, Uttarakhand, Uttar Pradesh), Eastern (Bihar, Jharkhand, West Bengal, Odisha), Northeast (Arunachal Pradesh, Assam, Manipur, Meghalaya, Mizoram, Nagaland, Sikkim, Tripura), Southern (Andhra Pradesh, Karnataka, Kerala, Lakshadweep, Puducherry, Tamil Nadu), Western (Dadra and Nagar Haveli, Daman and Diu, Goa, Gujarat, Maharashtra), and Northern (Haryana, Himachal Pradesh, Punjab, Rajasthan, Delhi, Chandigarh, Jammu and Kashmir). A map of these regions is shown in Appendix 1.

Analysis

We used temporal and spatial variation in water availability to estimate the effect of changes in water availability occurring within the same districts over time on corresponding changes in suicide counts using fixed effects regression. We chose a fixed effects approach over a random effects approach because information on potentially important confounders of the relation between drought and suicide, such as a district’s level of economic development or dependence upon agriculture, were unavailable. A fixed effects approach, where districts serve as their own controls, accounts for unmeasured differences in time-invariant characteristics of districts by design. This includes differences in baseline suicide rates, which prior research indicates varies considerably by geographical area.²

We created a dataset with four separate strata for each possible sex and age combination (i.e., male or female; aged 15 to 40, or 40 and older) for each district (n=569) and each year (n=13), resulting in 29,588 observations. Within each of these strata, we calculated the total number of deaths, the number of deaths due to suicide, and the corresponding measure of water availability, which was measured by the SPEI. Water availability was initially defined with a 9-level indicator variable (i.e., normal conditions; moderate, severe, extreme, or exceptional drought; moderate, severe, extreme, or exceptional wetness). Wald tests indicated effect estimates were similar for many categories of wetness or dryness, and thus we combined categories into 3 levels, including normal (SPEI between −1.6 and 1.6), extremely wet (SPEI greater than 1.6), and extremely dry (SPEI less than −1.6). We estimated standard errors clustered at the district level and used total deaths as the offset.²⁹ We allowed trends in suicide rates to differ by region with an product term between year fixed effects and region.

We modeled suicide death counts with Poisson regression, which assumes that the conditional distribution of suicide counts in each district d during growing season g is Poisson:

$$y_{dg}=\sim Poisson(u_{dg})$$

The model is expressed by the equation:

$$u_{dg}=\text{exp}(\propto +\beta XSPE{I}_{dg}+{\sigma }_d+{\rho }_r+{\tau }_y+{\pi }_{yr}+{\theta }_f+\delta X_{dg}+\text{ln}(tdeath{s}_{dg}))$$

where

u_dg is the number of suicides in each district d in growing season g

SPEI is the Standardized Precipitation Evapotranspiration Index for each district d in each growing season g

σ_d are district fixed effects

ρ_r are region fixed effects

τ_y are year fixed effects (to control for secular trends in suicide rates)

π_yr is an interaction term between region and year (to allow trends in suicide to vary by region)

θ_f is an indicator variable for sampling frame (i.e., 2001-03, 2004-2013)

X_dg are basic demographic covariates (age, sex) at the district-growing season level

tdeaths_dg is the total number of deaths in each district d in growing season g.

The effect of extremes in water availability may vary by region or demographic factors. We tested if our effects differed by these factors by including interaction terms between 1) region and SPEI, 2) sex and SPEI, and 3) age and SPEI. We tested interactions with Wald tests. In instances where we found significant interactions, we included the interaction in our model, and also included an additional parameter to allow for differential secular trends for demographic factors (e.g., age X year).

Sensitivity analyses

We performed additional analyses to assess the robustness of our results. Our main analysis estimated the effect of changes in water availability on the corresponding change in suicide within the same time period (June-March), and we performed additional analyses to test whether water availability preceded suicide. First, we excluded deaths (both suicides and total deaths) occurring during the first two months of the growing season (i.e., June and July), because these deaths may be attributable to the conditions of the preceding growing season. Thus, these analyses estimate the effect of water availability, between June and March, on deaths occurring between August and March. Second, we estimated effects with 1 and 2 month time lags. This analysis was conducted by summing total death and suicide counts during a time period that was one month (July-April) and two months (August-May) later than the growing season, and then estimating the lagged effect of water availability during the growing season (June-March) on suicides between July and April (1 month lag) and August to May (2 month lag). Third, we assessed the cumulative effect of water availability over the preceding two years by including both current year and past year SPEI values in the same model.

We used total deaths as the offset, and a key assumption is that, within districts, water availability does not substantially affect non-suicide death counts. While specific causes of death might be affected by water availability in India (e.g., snake bites, malaria), an increase or decrease in these other types of death will only lead to bias if total death counts are affected. We tested this assumption by estimating the relation between water availability and non-suicide deaths.

We also conducted two falsification tests to examine if our main analysis might be biased.³⁰ First, we identified a related type of death that should not be affected by water availability, deaths from unintentional injuries. These deaths were predominantly due to road traffic accidents (35%) and falls (29%), which we believe should not be affected by long-term water availability over an entire growing season. Second, we estimated the effect of water availability on suicide in urban areas, which encompassed 340 districts in our sample. In both analyses, finding an association between water availability and our outcome suggests that our main analysis may be biased, whereas finding no association provides evidence that our assumptions may be reasonable. We estimated these associations with the same models used in our main analysis, namely, a fixed effects model that adjusted for age, sex, year, region, district, and sampling frame, and allowed trends to differ by region (year X region).

RESULTS

There were 249,786 deaths and 9456 suicides that occurred during the growing season among adults aged 15 and older living in rural India in our study population. The most common modes of suicide were poisoning (40%), hanging or strangulation (37%), and burning (10%). Table 1 shows that the percent of deaths due to suicide was highest among those aged 15 to 40 and people living in rural areas, and that suicide rates were almost as high in women as men (male-to-female suicide ratio = 1.1). There was substantial variation in the percent of deaths due to suicide by district, which was highest in the South (Figure 1).

Table 1.

Characteristics of suicide deaths, Million Deaths Study, India, 2001-2013

	Suicide deaths	Total deaths	Percent of deaths due to suicide
Age
0 to 15	268	110017	0.2
15 to 40	9380	72873	12.9
40 or older	4518	399083	1.1
Sex
Male	8154	327151	2.5
Female	5743	255052	2.3
Location
Urban	2350	123555	1.9
Rural	11816	458711	2.6
Time of year
Growing season (June 1 - March 31)	11547	491472	2.3
Non-growing season (April 1- May 31)	2527	87490	2.9

Figure 1.

Percent of deaths due to suicide by district, Million Deaths Study, India, 2001-2013

Note: Districts excluded from analysis include: i) those that were composed solely of urban areas (n=16), ii) island territories (n=3), and iii) districts that had data sources that could not be successfully linked (n=6).

There were differences in water availability between districts, and water availability changed substantially during wet and dry years (Figure 2). Extremely wet conditions were more common than extremely dry conditions (Figure 3): on average, each year 1.7% of districts experienced extreme or exceptionally dry conditions, and 6.6% experienced extreme or exceptionally wet conditions. There were notable fluctuations in water availability by year, and there were also fluctuations in the percent of deaths due to suicide (Figure 4).

Figure 2.

Standardized Precipitation Evapotranspiration Index (SPEI) values during a dry (2002) and wet year (2010).

^a10-month average SPEI values from June 1, 2002 to March 31, 2003.

^b10-month average SPEI values from June 1, 2010 to March 31, 2011.

Notes:

1. Negative values (orange or red) indicate dry conditions, and positive values (blue) indicate wet conditions.

2. Districts excluded from analysis include: i) those that were composed solely of urban areas (n=16), ii) island territories (n=3), and iii) districts that had data sources that could not be successfully linked (n=6).

Figure 3.

Frequency of extremely dry (SPEI ≤−1.6) and extremely wet (SPEI ≥ 1.6) conditions from 2001 and 2013.

Note: Districts excluded from analysis include: i) those that were composed solely of urban areas (n=16), ii) island territories (n=3), and iii) districts that had data sources that could not be successfully linked (n=6).

Figure 4.

Percent of deaths due to suicide and average Standardized Precipitation Evapotranspiration Index (SPEI) values by region, 2001 to 2013

Notes:

1. Years span from June 1 to March 31^st of the following year

2. Standardized Precipitation Evapotranspiration Index (SPEI) values above 0 denote wet conditions, and values below 0 denote dry conditions.

In analyses restricted to deaths in rural areas and occurring during the growing season, models that adjusted for district, year, sample frame, sex, age, region, and allowed for secular trends in suicide by region (year X region), we found that the percent of deaths due to suicide increased during growing seasons with extremes in water availability (Table 2). The predicted percent of deaths due to suicide increased from 6.1 (95% CI: 5.9, 6.3) during normal growing seasons to 7.2 (95% CI: 6.5, 8.0) during extremely wet growing seasons, which corresponded to a relative increase of 18.7 percent (95% CI: 6.2, 31.2) and an absolute increase of 1.1 percentage points (95% CI: 0.4, 1.9). The predicted percent of deaths due to suicide also increased slightly during extremely dry conditions (6.3; 95% CI: 5.0, 7.6), which corresponded to a relative increase of 3.6 percent (95% CI: −17.9, 25.0) and an absolute increase of 0.2 percentage points (95% CI: −1.1, 1.5) compared to normal conditions. However, effect estimates were imprecise and included the null.

Table 2.

Percent of deaths due to suicide (95% CI) as a result of extremes in water availability, as measured by the Standardized Precipitation Evapotranspiration Index (SPEI), rural India, 2001-2013

	Normal (−1.6> SPEI <1.6)	Extremely dry (SPEI ≤ −1.6)			Extremely wet (SPEI ≥ 1.6)
	Predicted % of deaths due to suicide	Predicted % of deaths due to suicide	Percent change from normal conditions	Absolute change from normal conditions	Predicted % of deaths due to suicide	Percent change from normal conditions	Absolute change from normal conditions
Main effects
Adjusted for districta	2.8 (2.7, 2.9)	2.8 (2.2, 3.4)	0.3 (−22.5, 23.2)	0.0 (−0.6, 0.6)	3.0 (2.7, 3.3)	8.0 (−3.3, 19.3)	0.2 (−0.1, 0.5)
Fully adjustedb	6.1 (5.9, 6.3)	6.3 (5.0, 7.6)	3.6 (−17.9, 25.0)	0.2 (−1.1, 1.5)	7.2 (6.5, 8.0)	18.7 (6.2, 31.2)	1.1 (0.4, 1.9)
By age
Aged 15 to 40c	12.1 (11.8, 12.5)	10.4 (7.7, 13.1)	−14.7 (−37.1, 7.7)	−1.8 (−4.5, 0.9)	14.0 (12.5, 15.6)	15.5 (1.9, 29.2)	1.9 (0.3, 3.5)
Aged 40 or olderc	1.0 (1.0, 1.1)	1.4 (1.0, 1.9)	42.8 (1.4, 84.1)	0.4 (0.0, 0.9)	1.3 (1.1, 1.5)	25.5 (4.4, 46.6)	0.3 (0.0, 0.5)

^aIncludes fixed effect for district

^bModel controlled for year, sampling frame, age, sex, and region; included an interaction term between region and year (region X year); and a fixed effect for district

^cModel controlled for year, sampling frame, age, sex, and region; included interaction terms between SPEI and age (age X SPEI) and region and year (region X year); and a fixed effect for district.

In fully adjusted models, we found substantial variation in effects by age (p=0.01) but not region (p=0.32) or sex (p=0.78). Therefore, in additional models we allowed effects to vary by age by including a product term between age and SPEI and allowing these effects to vary by year (age X year). Among adults aged 15 to 40, extremely wet conditions corresponded to a relative increase of 15.5 percent (95% CI: 1.9, 29.2) and an absolute increase of 1.9 percentage points (95% CI: 0.3, 3.5) compared to normal growing conditions. However, extremely dry conditions did not correspond to an increase in suicide. Among adults aged 40 and older, extremely wet conditions corresponded to a relative increase of 25.5 percent (95% CI: 4.4, 46.6) and an absolute increase of 0.3 percentage points (95% CI: 0.0, 0.5) compared to normal conditions, and extremely dry conditions corresponded to a relative increase of 42.8 percent (95% CI: 1.4, 84.1) and an absolute increase of 0.4 percentage points (95% CI: 0.0, 0.9). The much larger absolute differences observed among those aged 15 to 40 compared to those 40 and older reflect the much higher percent of suicide deaths among this younger age group (12.9%) compared to the older age group (1.1%), as shown in Table 1.

In sensitivity analyses assessing temporality, we found very similar effect estimates to our main results when we excluded deaths occurring at the beginning of the growing season (Appendix 2), and when we added a one or two month time lag between water availability and suicide (Appendix 3 & 4). However, we did not find an association between extremes in water availability and suicide when we modelled the cumulative effect of past and current year (Appendix 5), indicating that water availability in the current year, but not preceding year, may be most relevant for suicide. Our analysis estimating the relation between water availability and non-suicide deaths found slight changes in non-suicide deaths (Appendix 6). The largest change was among adults aged 15 to 40 during extremely wet conditions, where suicide deaths decreased by 4.5 percent (95% CI: −6.3, −2.7) relative to normal conditions. These results indicate that non-suicide deaths are only slightly affected by water availability, and thus provides evidence that total deaths is an appropriate offset.

Our falsification tests were supportive of the conclusions from our primary analysis (Appendix 7 & 8). First, we did not find strong evidence that our negative control, unintentional injury deaths, was associated with a relative increase in either extremely dry conditions (relative change = 2.3; 95% CI: −9.5, 14.1) or extremely wet conditions (relative change = 4.9; 95% CI: −2.3, 12.0) compared to normal conditions. Second, we did not find evidence of an effect of either extremely dry conditions (relative change = −16.3; 95% CI: −50.5, 18.0) or extremely wet conditions (relative change = 29.6; 95% CI: −2.3, 61.6) compared to normal conditions in urban areas, although our analysis is likely underpowered to detect a difference; only 340 district had urban areas, and the sample size in urban areas was about a quarter of the size of rural areas (Table 1).

DISCUSSION

We found extremes in water availability – especially extremely wet conditions – associated with an increase in the percent of deaths due to suicide among adults living in rural areas. Our study adds to a broad and robust literature on the health effects of weather extremes,³¹ which are projected to increase due to climate change.³²

Research investigating water availability and mental health has focused almost exclusively on drought,^33-36 and drought as a cause of suicide within India has dominated public discourse.^37-41 To our knowledge, no studies before ours have investigated the link between suicide and water availability at both extremes. We found, perhaps surprisingly, extremely wet conditions more strongly associated with suicide than extremely dry conditions. Our results were robust to a number of sensitivity analyses.

We found substantial variation in the effect of water availability by age, with adults aged 15 to 40 experiencing the largest percentage point increase in suicide deaths as a result of extremely wet conditions. We are not aware of research that has investigated the mental health effects of water availability among different age groups, and thus this is an underexplored area of research. Younger people in rural areas may be particularly susceptible to water extremes due to the migration of young adults with better mental health to urban areas (i.e., the healthy migrant effect), which has been postulated as a mechanism in other settings such as China,⁴² or cohort effects, such that people born during different time periods have a different risk profile for suicide.⁴³ Future research could investigate potential reasons for this heterogeneity.

While our study could not investigate potential mechanisms, very wet conditions could be especially damaging to crops because of limited means to mitigate conditions: while extreme drought might be offset by irrigation, extremely wet conditions – such as flooding or torrential rain – may result in conditions that are less easily mitigated and more devasting to crops. Excess soil moisture is linked with crop loss in the US,⁴⁴ and research from India finds extreme wetness associated with diminished crop yields.^4,5 However, these Indian studies find extreme dryness associated with larger reductions in crop yield than extremely wet conditions,^4,5 although we found a stronger association between wet conditions and suicide. This difference could be due to additional mechanisms linking extremely wet conditions with suicide (e.g., cloud cover,^45,46 temperature^11-15,45), or extremely wet conditions could truly be more damaging to crops, but the extant literature from India is incomplete; this literature used rainfall as a proxy for drought and flood, which is a less accurate measure of water availability than measures (such as the SPEI) that integrate temperature.⁴⁷

Extremes in water availability may affect suicide by impacting crop production, and thus investigating the link between crop yield and suicide would provide support for this mechanism. Regrettably, we were not able to investigate crop yield, or factors related to crop yield (e.g., irrigation, crop types) directly; although crop yield data by district is publicly available, district boundaries changed over time, which precludes a longitudinal analysis between crop yield and suicide. Our study overcame this challenge by collecting suicide counts in districts that were consistently defined over time, and then linking these consistently defined districts to granular, geographical data on water availability. Thus, while our approach allows a longitudinal analysis of changes in water availability over time, it does not allow linkages with other data sources collected at the district level that could complement our main analysis.

Climate projections indicate extreme weather will increase in South Asia. Projections show the frequency of years with extremely high and extremely low rainfall will increase, and seasonal variation in rainfall will be amplified, with precipitation decreasing by 30% in the dry season and increasing by 30% in the wet season.⁴⁸ These predictions portend even more challenging growing conditions in the coming years and potentially even more suicides attributable to extremes in water availability. Indian governmental responses to extremes in water availability (specifically drought) have been criticized as reactive and ineffective,⁴⁹ and these projections highlight that a comprehensive governmental strategy to ameliorate the effects of water availability extremes is urgently needed. Our findings indicate public health and governmental monitoring and response to adverse growing conditions within India should encompass exceptionally wet conditions.

Our study has a number of strengths. First, we leveraged a unique population-based data source with stringent methods for ascertaining cause of death. Prior studies rely on suicide counts collected by the NCRB, which substantially under-counts suicide.² Second, we used an approach that controls for district-level differences in suicide rates with fixed effects regression. Suicide rates vary substantially by district, and by controlling for these differences, our study can provide accurate estimates of the effect of water availability on suicide. Third, we measured water availability with a validated scale, measured at the district level, which provides an accurate picture of growing conditions at local levels. Fourth, research investigating water availability and mental health has focused almost exclusively on drought,^33-36 and we investigated effects at both extremes.

Our study has limitations. First, because suicide was rare and our study aggregated deaths to small geographical areas, our assessment of effect heterogeneity by sex, region, and age may have been underpowered to detect differences. Relatedly, we could not investigate heterogeneity by more than two age categories due to the sparseness of suicide data. Second, because of limited demographic indicators collected by the MDS, we were not able to investigate effects among specific demographic groups that may be most affected by water availability, such as the poor or farmers. Third, there may be residual confounding of our effect estimates due to unmeasured time-varying characteristics of districts (e.g., economic downturns), as well as bias due to spillover effects from adjacent districts. Fourth, while we found compelling evidence that extremely wet conditions may affect suicide, our conclusions for extremely dry conditions were more ambiguous due to small effect estimates and wide confidence intervals. Finally, because we used the percent of total deaths due to suicide as the outcome, we cannot completely rule out the potential for confounding due to extreme growing conditions affecting other types of death, such as deaths due to malaria or snakebites. While our sensitivity analysis did not find non-suicide deaths strongly related to changes in water availability, we cannot rule this out as a source of residual confounding. Future research that replicates our findings and investigates effects with population denominators and for specific demographic groups, especially among different age groups, would be a welcome addition to the literature.

In summary, we found extremes in water availability, especially extremely wet conditions, associated with an increase in deaths due to suicide. Within India, exceptionally wet growing conditions should be considered an important population-level risk factor for suicide in public health and governmental interventions that aim to address these preventable deaths.

Highlights.

• Most research on extremes in water availability and suicide focuses on drought, and the effects of extreme wetness are not well known.

• Using a nationally representative sample of deaths occurring in India from 2001 to 2013 linked with a validated water availability scale, we investigated the effect of adverse growing conditions – both extreme drought and extreme wetness – on suicide.

• We found that extremely wet growing seasons were associated with an increase in the percent of deaths due to suicide. The percent of deaths due to suicide also increased slightly during extremely dry (drought) growing seasons, although effect estimates were attenuated and included the null.

• Extremely wet growing seasons may be an overlooked population-level risk factor for suicide in India.