Nocturnal wakefulness and suicide risk across months and methods of suicide

Abstract

Objective:

Insomnia is a risk factor for suicide, and the risk of suicide after accounting for population wakefulness is disproportionately highest at night. This study investigated whether this risk varied across months and/or methods of suicide.

Methods:

Time, date, method (e.g. firearm, poisoning) and demographic information for 35,338 suicides were collected from the National Violent Death Reporting System for the years 2003–2010. Time of fatal injury was grouped into one-hour bins and compared to the estimated hourly proportion of the population awake from the American Time Use Survey for 2003–2010. Negative binomial modeling then generated hourly incidence risk ratios of suicide. Risks were then aggregated into four categories: Morning (06:00 to 11:59), Afternoon (12:00 to 17:59), Evening (18:00 to 23:59), and Night (00:00 to 05:59).

Results:

The risk of suicide was higher at night across all months (p<0.001) and methods (p<0.001). The average nocturnal risk across months was 3.18 (SD 0.314), with the highest risk in May (3.90) and the lowest in November (2.74). The average nocturnal risk across methods was 3.09 (SD 0.472), with the highest risk for fire (3.75) and the lowest for drowning (2.44). Additionally, nocturnal risk elevated within all demographics (all p<0.001). However, there were no month-by-time or method-by-time interactions across demographics (all p>0.05).

Conclusions:

Regardless of month or method, the incidence risk of suicide at night is higher than any other time of day. Additionally, demographic subgroups did not differentially experience higher risks across months or mechanisms at night.

INTRODUCTION

Sleep and circadian rhythms play a significant role in suicidality. Pigeon and colleagues reported that disturbed sleep increased the risk of suicide ideation, attempts, and completions by 2 to 3 times, and that this effect was specifically related to insomnia and nightmares¹. Insomnia increases the risk of suicide attempts by 3.5 times in adults, with a 5.5 times increase in the 25–44 year old demographic². Insomnia is strongly associated with current suicidality in undergraduates^3–5 and military personnel^6,7. Insomnia also predicts future suicidality across adults^6,8–12. Case-control studies show that insomnia and sleep disturbance are more prevalent among suicide completers than among matched controls^13,14. From these studies, it is clear that difficulty with sleep contributes to suicidal behavior.

Three recent studies highlight nocturnal wakefulness as a risk factor for suicide^15,16. Perlis and colleagues extracted the time of fatal injury for over 35,000 suicides, combined this with estimations of the waking population size, and determined that the number of suicides at night vastly exceeded the number of expected suicides given the proportion of the population awake at that time¹⁵. This equated to a 3.6 times risk of suicide at night that was consistent across demographics. Additionally, Ballard and colleagues experimentally investigated the role of nocturnal wakefulness in suicidality with a single-night polysomnography study. They found participants with suicidal ideation were more likely to be awake at 4AM, that wakefulness between 4 and 5AM was associated with suicidal thinking the next day, and that these effects were independent of depression severity and other covariates. Finally, there is a circadian pattern in suicides among individuals who are heavily intoxicated (blood alcohol levels of 80 mg/dl or more), with a peak in suicide counts at 9:00PM¹⁷. Thus nocturnal wakefulness is associated with a heightened risk of suicidal thinking and behavior and this risk may be modifiable by alcohol or other substances.

Given the concordance between epidemiological and experimental data, it is imperative to determine the mechanisms connecting suicidality to nocturnal wakefulness. The interpersonal-psychological theory of suicide suggests that suicidality is associated with isolation and thwarted belongingness, both of which can be exacerbated when individuals are awake and alone during the night^18–20. From a neurobiological perspective, nocturnal wakefulness in the context of sleep deprivation is associated with reduced frontal cortical connectivity and impaired cognition^21–23, both of which may increase suicidality through altered decision-making, poor impulse control, and disrupted risk/reward evaluations. Unfortunately, the paucity of data and inconsistent methods limits inferences on these mechanisms²⁴.

However, where experimental data are lacking, population level data can provide insight. Suppose that the nocturnal risk of suicide varied across calendar months, specifically by increasing during the summer. This would lend support to current theories about temperature and light exposure as environmental factors driving suicidal behavior^25–30. Alternatively, suppose the nocturnal risk of suicide was disproportionately elevated for a particular method of suicide. This would provide useful information for public health interventions regarding that specific method. Therefore, the present study explored whether the nocturnal risk of suicide varied across months or methods of suicide, and whether this risk was influenced by demographic or geographic factors.

METHODS

Datasets

The datasets used for this analysis have been described previously¹⁵. Briefly, suicide data were collected from the United States National Violent Death Reporting System (NVDRS; www.cdc.gov/violenceprevention/nvdrs/³¹) for the years 2003 to 2010. The NVDRS collects data from participating states through medical examiner and police reports. These include the estimated time of death/fatal injury, the month in which the suicide occurred, the age (including minors), race, ethnicity, and sex of the victim, and the primary method used to commit suicide. Population wakefulness data were drawn from the American Time Use Survey (ATUS; www.bls.gov/tus/data.htm³²) for the same years. ATUS data are collected annually by telephone from individuals 15 years and older across the United States. The ATUS provides information on the percentage of the population awake at each clock hour, including subdivisions by demographic (race, ethnicity, age, and sex), and month.

Variable Definitions

Clock hours were grouped into four time categories: night (0000 to 0559), morning (0600 to 1159), afternoon (1200 to 1759), and evening (1800 to 2359). Months and suicide methods (guns, asphyxia, poison, fall, vehicle, sharp weapons, drowning, and fire) were reported as part of the NVDRS dataset and assessed as categorical variables. Demographics included sex (Male or Female), age (15 to 24, 25 to 34, 35 to 44, 45 to 54, 55 to 64, 65 to 74, and 75 or older), race (White, Black, Asian, Other), and ethnicity (Hispanic, Non-Hispanic) based on the data available from the NVDRS and ATUS datasets. Geographic analyses were conducted in two ways. First, latitude effects were assessed by comparing states that were clearly at or above 40°N (“Above40”), or clearly at or below 35°N (“Below35”). The goal of this analysis was to address variations in seasonal light exposure by taking the most extreme data available. States in between these latitudes were not used. The second analysis divided states into four regions (West, Midwest, South, and Northeast) in accordance with the American Association of Suicidology³³ to determine regional effects on suicide risks.

Generating Incidence Risk Ratios

The statistical analysis used quasi-Poisson and negative binomial regression to generate incidence risk ratios. The choice between quasi-Poisson and negative binomial regression depended on the number of suicides observed in each category, as smaller counts are more accurately modeled using quasi-Poisson regression. Suicide counts were modeled as functions of time of day, month or method, and their interaction. Empirically, at least 5 suicides at each time were required for adequate model fit, so any categories which did not have at least 5 suicides per time of day were eliminated. Population wakefulness was estimated based on ATUS survey responses. These proportions were adjusted for the relevant month and demographic subgroup (sex, age, race, and ethnicity) and entered as an offset/exposure variable in the model. Modeling counts in this way produces an incidence risk ratio (IRR), which represents the increased risk of suicide at that level (e.g. month, time) as compared to a reference level. Ordinarily, this reference would be another level within the same categorical variable (e.g. comparing Blacks to Whites). Effect coding is an alternative approach which compares each level to the grand mean across levels (e.g. comparing Blacks to the average across all races). To avoid arbitrary assignment of a reference level, effect coding was used in both the by-months and by-methods analyses. For months, effect coding was used for time, month, and the interaction. For methods, effect coding was used only for time, as comparisons were made only within method. Effect coding was achieved using package ‘wec’ in R³⁴. Statistical tests of significance were conducted using one or two-way ANOVA depending on the particular model used. All analyses were conducted using R (version 3.5.1) and graphs were generated using ggplot2³⁵.

RESULTS

Summary of the Data

A total of 35,338 suicides were extracted from the NVDRS dataset between 2003 and 2010. Some cases included missing data and were excluded as appropriate in analyses (e.g. month of death was present, but not method). Table 1 shows the suicide counts by Time of Day, Month, and Method of Suicide across demographic variables. By raw counts, most suicides occurred in the Afternoon (N=11,381), in May (N=3,196), and by firearms (N=21,397). More males committed suicide than females (N=28,700 males vs. N=6,636 females), and suicide counts were highest among 45–54 year olds (N=7,252), whites (N=31,239), and non-Hispanics (N=33,394). The number of suicides was roughly equivalent between states categorized as Above40 (N=6,165) and Below35 (N=6,995). By region, most suicides were in the South (N=19,996), with fewer suicides in the West (N=7,620), Midwest (N=4,150), and Northeast (N=3,504).

Table 1:

Suicide counts by time of day, month, and method across demographic factors.

		Sex		Age							Race				Ethnicity		Latitude		Region
	N=35,338	F	M	15–24	25–34	35–44	45–54	55–64	65–74	>75	White	Black	Asian	Other	Non-Hispanic	Hispanic	>40	<35	Midwest	Northeast	South	West
	Total	6636	28700	5000	5659	6740	7252	4686	2688	2943	31239	2701	526	686	33384	1618	6165	6995	4150	3504	19996	7620
Time of Day	Morning	1706	7705	967	1222	1653	1956	1454	922	1183	8408	651	123	188	8896	425	1649	1784	1067	1058	5185	2082
	Afternoon	2299	9081	1456	1601	2169	2498	1643	916	956	10155	814	180	172	10839	444	2127	2272	1426	1117	6506	2314
	Evening	1703	7519	1452	1649	1891	1920	1086	585	498	8123	757	136	156	8711	423	1504	1916	940	873	5414	1977
	Night	928	4395	1125	1187	1027	878	503	265	306	4553	479	87	170	4938	326	885	1023	717	456	2891	1247
Month	Jan	525	2288	422	409	549	568	375	213	249	2492	212	38	56	2648	143	510	579	335	292	1583	598
	Feb	515	2137	377	454	494	537	332	214	214	2339	214	43	48	2514	105	451	563	298	241	1518	587
	Mar	607	2520	444	468	566	686	430	228	277	2778	236	45	53	2939	152	557	631	336	310	1764	713
	Apr	572	2347	410	509	555	586	415	175	233	2582	219	39	53	2754	136	490	590	340	303	1696	572
	May	596	2600	436	534	669	599	399	257	259	2819	238	53	68	3028	141	561	616	359	338	1797	697
	Jun	564	2449	455	476	555	615	403	224	260	2659	229	49	58	2846	137	520	591	378	300	1691	641
	Jul	563	2617	437	517	604	663	423	231	270	2806	251	40	71	3001	158	553	615	378	292	1793	712
	Aug	595	2452	435	501	580	623	412	224	231	2711	234	40	49	2888	135	512	603	347	320	1737	637
	Sep	575	2369	420	455	529	657	389	248	221	2579	244	35	76	2787	124	526	539	349	294	1650	648
	Oct	522	2483	433	489	584	587	387	233	267	2670	218	50	49	2824	149	536	613	359	281	1714	647
	Nov	508	2256	370	440	575	550	345	220	236	2424	219	47	59	2621	119	487	527	338	277	1531	611
	Dec	494	2182	361	407	480	581	376	221	226	2380	187	47	46	2534	119	462	528	333	256	1522	557
Method	Asphyxia	1291	5646	1512	1577	1519	1197	528	190	197	5796	616	234	217	6275	605	1231	1232	963	1119	3429	1418
	Drowning	134	286	40	71	77	99	62	42	27	313	79	21	3	404	14	120	68	49	102	209	58
	Fall	273	670	165	156	206	202	102	50	59	780	97	48	4	885	50	263	102	95	244	419	182
	Fire	75	165	29	39	50	70	38	9	5	177	42	15	5	226	11	61	34	29	59	114	37
	Gun	2794	18603	2760	3070	3636	4146	3113	2144	2405	19261	1587	132	352	20482	681	3390	4636	2265	1138	13212	4759
	Poison	1764	2113	265	488	929	1173	666	172	171	3593	175	36	62	3687	165	750	761	544	489	1941	884
	Sharp Weapon	96	457	37	71	110	147	91	44	50	488	38	12	11	523	30	106	82	70	127	264	90
	Vehicle	169	590	170	146	174	171	55	23	15	639	60	26	28	699	56	193	63	96	217	283	163

Suicide Months Analysis

The first question was whether, after accounting for population wakefulness, the hourly incidence risk ratio of suicide varied across months. The results are shown in Figure 1. May showed the highest risk (3.9 ± 0.48), while November had the lowest risk (2.7 ± 0.34). However, a two-way ANOVA for month, time of day, and an interaction showed that the incidence risk varied significantly across time of day (p<0.001), but not across months (p=0.33) or by the interaction (p=1.00). A post-hoc Wald test showed that incidence risk at Night was significantly higher than any other time of day (p<0.001). The average incidence risk at night across months was 3.18 (SD: 0.314).

Figure 1:

The risk of suicide by time of day across months. The incidence risk ratios were much higher at night as compared to the average across all times of day. Risk peaks in May and October, although these increases were not statistically significant. Night: 12AM to 6AM; Morning: 6AM to 12PM; Afternoon: 12PM to 6PM; Evening: 6PM to 12AM. Risk ratios are plotted as mean ± standard error.

In subgroup analyses, each demographic was evaluated separately for an effect of time of day by month. These results are presented in Figure 2; women tended to have slightly higher nighttime risks than men, nighttime risk appeared to decrease with age, and Hispanics tended to have higher nocturnal risks than non-Hispanics. However, a two-way ANOVA found that, while IRRs varied significantly across time for men and women, all ages, all races, Hispanics and non-Hispanics, all latitudes, and all regions (all p<0.001), they were not significantly different across months, nor was there a significant month by time of day interaction (all p>0.05).

Figure 2:

The nocturnal risk of suicide across months by demographic characteristics. Incidence risk ratios were generated within demographic categories (e.g. for males separately from females), so comparisons across demographics were not made. No statistically significant effects for time of day by demographic were noted. Risk ratios are plotted as mean ± standard error.

Suicide Method Analysis

The second question was whether the incidence risk ratio of suicide varied by time across suicide methods. An initial two-way ANOVA evaluating time by method showed that risk varied significantly by time of day and by method (both p<0.001), but that the interaction was not significant (p=0.3026). Thus, no method had a significantly higher risk at a specific time than any other method at that same time. Post-hoc Wald tests found that the risk was highest at night and for guns (both p<0.001). The average incidence risk at night across methods was 3.09 (SD 0.472). However, suicide counts were heavily skewed across methods, with more than half of the suicides involving firearms. Therefore, the risk of each method at each time of day was evaluated independently, as opposed to comparing risks across methods. Effect coding was used to compare the risk at each time to the grand mean across the day within each method. Figure 3 shows the IRRs for each method at each time.

Figure 3:

The risk of suicide by time of day across methods of suicide. The incidence risk ratios were higher at night as compared to the average risk across all times of day. While vehicles and fires showed an elevated risk, this increase was not statistically significant. Night: 12AM to 6AM; Morning: 6AM to 12PM; Afternoon: 12PM to 6PM; Evening: 6PM to 12AM. Risks are plotted as mean +/− standard error. Risk ratios are plotted as mean ± standard error.

As in the by-months analysis, IRRs were evaluated across sex, age, race, ethnicity, latitude, and region. These results are presented in Figure 4. Nocturnal risk for each method did not vary by sex, age, race, ethnicity, or geography. A two-way ANOVA showed that IRRs varied significantly by time of day and method for all demographic groups (all p<0.001), reiterating that the risk of suicide is not equivalent across all times or every method. However, the interaction between time of day and method was not significant for any subgroup (all p>0.05). Thus, there was no method of suicide that was systematically more common in any particular demographic or geographic group.

Figure 4:

The nocturnal risk of suicide across suicide methods by demographic characteristics. Incidence risk ratios were generated within demographic categories, so no comparisons across demographics were made. The only statistically significant effect present was the increased nocturnal risk of death by fire in northern states; no other statistical differences were noted. Risk ratios are plotted as mean ± standard error. Categories for which no ratio is present did not have at least 5 suicides at each time of day, and were therefore excluded from analysis.

DISCUSSION

This study found that the nocturnal risk of suicide is approximately three times that of the across day average after accounting for population wakefulness. Additionally, the increased risk at night remained constant regardless of month or method of suicide. While differences may exist across demographics and methods, these differences are not dependent on time of day. These findings reinforce previous results in suggesting that nocturnal wakefulness is a significant risk factor for suicide.

Suicide Months Analysis

Seasonal effects on suicide have been described in both the northern and southern hemispheres. Holopainen and colleagues found that suicide peaks occurred in May and October, which they hypothesized was due to dramatic changes in temperature³⁰. This reinforced an earlier finding that Finnish suicides were highest in spring, when solar radiation was highest³⁶. A study of young people in the more equatorial country of Turkey found a peak in suicide attempts in the summer³⁷. In Brazil, Bando and colleagues found that male suicide peaked in late spring and reached a minimum in late fall, while females peaked in summer and reached their minimum in winter³⁸. The increased risk for suicide in spring was also present in Danish suicide completers with a history of mood disorders³⁹. The springtime increase in suicides has been correlated with increased light exposure, possibly through modulation of serotonergic activity^25–29. Whatever the mechanism, season appears to play a role in suicide risk.

Contrary to this literature, the present study found that the risk of suicide was highest at night across all months, and that no month showed a significantly different risk than any other. While Figure 1 appears to show an increase in nocturnal suicide risk in May and October, this finding was not statistically significant and cannot be interpreted further. If a seasonal effect does exist for suicide risk, it is likely insignificant compared to the time of day, and future research on seasons and suicide should account for time of day.

Suicide Methods Analysis

It is interesting that the nocturnal risk of suicide did not vary across methods of suicide. From a practical standpoint, some methods might be less accessible at night than others (e.g. there are fewer cars/trains active at night to be hit by). Nevertheless, in accordance with the hypofrontality theory advanced by Perlis and colleagues, this suggests that nocturnal wakefulness may be more associated with the initial decision to commit suicide than the choice of method²³. This also accords with other evidence of disinhibited behaviors at night, such as those around food choices. Individuals who go to bed later are more likely to consume hedonistic and energy-rich foods, such as fast foods and sodas^40,41. Similarly, a survey of Australian children and adolescents found that those who were awake later at night had a higher intake of “extra foods” and a lower diet quality⁴². In the case of suicide and food choice, nocturnal wakefulness may be driving impulsive decision-making through impaired cortical activity. However, experimental studies are needed to explore this hypothesis.

Strengths and Limitations

This study has several strengths. Despite using a different statistical approach, the average nocturnal risk of suicide was in line with the risk reported by Perlis and colleagues¹⁵. Additionally, the analysis was based on a large sample size (over 35,000 suicides), with estimated population wakefulness adjusted for time of day and relevant demographics. However, this study has several limitations. First, it is unclear whether individuals who committed suicide at night were continuously awake until their suicide, or whether they had recently woken up. This distinction is important, as it might separate sleep deprivation from circadian effects. Second, the adjustment for wakefulness was not based on individuals who specifically have suicidal ideation, plans, or attempts. Thus, if suicide attempters systematically differ from the general population in the timing of wakefulness this approach would not account for this difference. Additionally, we were unable to separate minors out from the age 15 to 24 demographic as we did not have separate ATUS wakefulness estimates for minors and adults in this category. Finally, the small number of minority suicides reported prevented the calculation of some incidence risk ratios.

Future Directions

It is critical to understand what biological and psychological mechanisms are at work in individuals who commit suicide at night. If middle insomnia leads to increased catastrophic thinking, hopelessness, and thwarted belonging, then treatment of insomnia may help vulnerable individuals to avoid times of dangerous thinking. Alternatively, if nocturnal hypofrontality contributes to suicidal decision-making, then separating the role sleep and circadian rhythms on this hypofrontality may inform future interventions. Finally, these results need to be replicated in other suicidal behaviors, including attempts, planning, and ideation. This would help to determine whether nocturnal wakefulness serves as an acute trigger for suicide attempts, or whether accumulation of nocturnal wakefulness promotes all forms of suicidal behavior.

Conclusions

Suicide is a growing crisis, and a greater understanding of precipitating factors is needed. In accordance with previous work, this report shows that suicide risk is higher at night than any other time of day, and that this risk does not vary significantly by month, method, or demographic characteristics. The implication that nocturnal wakefulness is a universal factor for suicide risk suggests that biological or psychological mechanisms may play a role outside of social and environmental factors. Future work is needed to fully assess sleep and nocturnal wakefulness as acute risk factors for suicidal behavior.

Clinical Points.

• Increasing evidence supports nocturnal wakefulness as a risk factor for suicide.

• The suicide risk associated with nocturnal wakefulness does not appear dependent on season or method in which the suicide occurs.

• Sleep disturbances in patients with a history or risk of suicidality should be treated to reduce suicide risk.