The effects of heat and air pollution on mental-health related mortality

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Request accessible transcript

Transcript (if available)

Content

The Effects of Heat and Air Pollution on Mental-Health Related Mortality
By
Melissa K. Lorenzo

A Thesis Presented to the
FACULTY OF THE USC KECK SCHOOL OF MEDICINE
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
MASTER OF SCIENCE
(BIOSTATISTICS)

May 2022

Copyright 2022 Melissa K. Lorenzo

ii
Table of Contents
List of Tables ................................................................................................................................. iii
List of Figures ..................................................................................................................................v
Abstract .......................................................................................................................................... vi
Introduction ......................................................................................................................................1
Methods............................................................................................................................................3
Results ..............................................................................................................................................5
Single-exposure models .......................................................................................................7
Two-exposure models ..........................................................................................................8
Evaluation of Linearity Assumption for Single-exposure models .....................................11
Interaction Models .............................................................................................................13
Significant Mean-Centered Interaction Models .....................................................14
Maximum/Minimum Heat Percentile ................................................................................18
Discussion ......................................................................................................................................19
Conclusion .....................................................................................................................................20
References ......................................................................................................................................21

iii
List of Tables
Table 1: Demographical information of suicide victims. Environmental exposures
listed here pertain to the individual’s day of death. ........................................................................6
Table 2: Demographical information of homicide victims. Environmental exposures
listed here pertain to the individual’s day of death. ........................................................................6
Table 3: Associations between temperature, pollutants, and deaths due to suicide in
single-exposure models. ...................................................................................................................7
Table 4: Associations between temperature, pollutants, and deaths due to homicide in
single-exposure models. ..................................................................................................................7
Table 5: Correlation chart of exposures (all-cause mortality). .......................................................9
Table 6: Main effect estimates of two-exposure models for deaths due to suicide. .......................9
Table 7: Main effect estimates of two-exposure models for deaths due to homicide. ..................10
Table 8: P-values of linear, square, and cubic terms for exposure in polynomial models
to assess for non-linearity in exposure-response. .........................................................................11
Table 9: Statistical measures of interaction terms regarding deaths due to suicide. ......................13
Table 10: Statistical measures of interaction terms regarding deaths due to homicide. ...............13
Table 11: Statistical measures of mean-centered main effects and interaction between
minimum temperature and ozone for deaths due to suicide. ........................................................14
Table 12: Statistical measures of mean-centered main effects and interaction between
maximum temperature and PM10 for deaths due to homicide. .....................................................16
Table 13: Statistical measures of mean-centered main effects and interaction between
minimum temperature and PM10 for deaths due to homicide. ......................................................17
iv
Table 14: Statistical measures of deaths due to suicide and homicide over various
percentiles of maximum temperature. ...........................................................................................18
Table 15: Statistical measures of deaths due to suicide and homicide over various
percentiles of minimum temperature. ...........................................................................................18

v
List of Figures
Figure 1: Smooth spline plots of maximum/minimum temperatures on case days.
Top row: deaths due to suicide. Bottom row: deaths due to homicide. ........................................12
Figure 2: Interaction plot between minimum temperature and ozone for deaths due
to suicide. ......................................................................................................................................14
Figure 3: Interaction plot between maximum temperature and PM10 for deaths
due to homicide. ............................................................................................................................15
Figure 4: Interaction plot between minimum temperature and PM10 for deaths
due to homicide. ............................................................................................................................17

vi
Abstract
Background: Climate change and its associated adverse outcomes is an area with much study.
Yet not much is known of the connection between air pollution and high heat -- two exposures
thought to increase with climate change -- to mental health related outcomes: specifically, deaths
due to suicide and homicide. This is particularly relevant for California, which is prone to
elevated temperatures and droughts, the latter of which possibly affecting air quality through
increased frequencies of wildfires. This analysis aims to explore the relationship between heat
and air pollution with suicide and homicide deaths in California during the years 2014 to 2019,
which included some of the hottest years on record.
Methods: Death certificate data for California from 2014 to 2019 were used in a case-crossover
study design. For each individual, date of death was defined as the case day and the other days
within the same month that were the same day of week were control days. Residential data for
decedents were used to assess exposure to daily temperature (maximum, minimum) and daily
average air pollution concentrations (particulate matter <10 µm [PM10] and <2.5 µm, nitrogen
dioxide, ozone). Deaths due to suicide and homicide were treated as separate outcomes in
conditional logistic regression models, adjusting for relative humidity. Single-exposure, two-
exposure (temperature and pollution), and interaction models were explored.
Results: Maximum daily temperature was significantly associated with deaths due to suicide (p-
value = 0.0381; odds ratio = 1.0044; 95% CI: 1.0002, 1.009), meaning that for every 1-degree
Celsius increase in maximum daily temperature, the odds of an individual dying by suicide
increases by about 0.44%. We observed pollutants to interact with minimum daily temperature:
the effect temperature has on mental-health mortality increased with lower exposure levels of
ozone and PM10.

vii
Conclusion: This analysis demonstrates a statistically significant linear relationship between the
odds of dying due to suicide or homicide and continuous maximum/minimum daily temperature.
More research is needed to confirm and better understand the observed interactions between air
pollutants and temperature. The results of this paper suggest raising awareness of the possible
connection between the effects of temperature on suicide and homicide deaths, as well as
bolstering preventive public health policies during hotter days.

1

Introduction
The consequential dangers of climate change have long been discussed and is generally
well understood within the general population. However, this common knowledge does not often
venture into the notion of climate change, especially the effects of heat and air pollution,
affecting mental health outcomes. Climate change will inevitably bring about weather events of
higher frequency, severity, and duration, and there is enough evidence to suggest that such
changes will affect mental health (Berry 2019). Recent research emphasizes the effects of
modifiable environmental exposures on suicide risk, and a large number of studies propose that
several environmental factors may independently be responsible for an increase in suicide risk
(Cornelius 2021). Current literature suggests that there are proposed biologic mechanisms
involved in the risk of suicide: for instance, air particulates act as irritants that prompt systemic
and local inflammatory responses:
“The pathophysiology of air pollution associated suicide risk is hypothesized to be
neuroinflammatory, with air particulates acting as irritants that generate systemic and
local inflammatory responses. . . the data so far also show that the risk of suicide is
significantly worse when the air is particularly polluted” (Dumont 2020)

Regarding the effects of heat on suicide risk, another study showed that there was a
significant association between suicide attempts and 5-degree Celsius increase in temperature
during the summer months (defined as May 31 - Sep. 22); (odds ratio [OR] 1.59, 95%
confidence interval [CI] 1.22-2.08) and a significantly higher incident rate ratio [IRR] for
attempts in the summer compared to other seasons (IRR 1.08; 95% CI 1.00, 1.16) (Yarza 2020).
In this study, suicide attempts were more likely when temperatures are unusually high compared

2

to typical values for that day, especially in the summer. Those who made a prior attempt were
more susceptible to the effects of temperature increases over successive days.
Homicide, another mental health-related cause of death, was shown to be positively
associated with ambient temperature in certain cities in the United States: this association tended
to be stronger for homicide cases that occurred during hot seasons, during the nighttime, or on
the street (Xu 2020). Another study found an increase in homicide-related emergency room visits
associated with a 30-day average of daily mean levels of ozone, and that these ozone-
homicides/homicide-related injuries had significantly higher associations in the colder months
compared to warmer months (Nguyen 2021).
These existing findings serve as inspiration to pursue the aims of this thesis project,
which is to explore the individual associations between heat and air pollution with deaths due to
suicide and homicide. Heat will be defined as continuous daily temperature measured in degrees
Celsius; likewise, air pollution is defined as daily mean concentration of particulate matter of
diameter < 2.5 µm and <10 µm (PM2.5 and PM10) measured in μg/m
3
, as well as ozone (O3) and
nitrogen dioxide (NO2) in parts per billion (ppb). Suicide will be defined as intentional self-harm
through various methods (ICD-10 codes X60-X83), while homicide will be characterized as
assault through several means, including sequelae of assault (ICD-10 codes X85-Y87.1). These
two outcomes will be assessed based on the immediate cause of death listed on the decedent’s
death certificate.
It is desired that this project promotes a better understanding of mental health and its
available treatments as well as to push for a higher presence of support groups during the hottest
months of the year: “mental health impacts should be incorporated into plans for the public

3

health response to high temperatures and . . . psychological morbidity and mortality temperature
thresholds should be incorporated into hot weather warning systems” (Thompson 2018).
Methods
A case-crossover study design was used to assess the relationship between heat and air
pollution with mental health-related mortality. This study design allows individuals to serve as
their own control, thus allowing for control of all time-independent factors such as age (within
the same month), race/ethnicity, education level, co-morbidity status, health status, and
socioeconomic status. Rather than recording case individuals, case days are observed. Case days
are the day of an individual’s death, and control days are days in the weeks prior to or following
the individual’s death within the same month and falling on the same day of the week. For
example, if an individual passed on the third Tuesday of March 2017, the first, second, and
fourth Tuesdays of March 2017 will be counted as control days.
The California Department of Public Health’s vital statistics provided death certificate
data for deaths due to all causes occurring in California from 1 January 2014 to 31 December
2019, supplying demographical data related to the decedent. These deaths were then subset to
specific deaths of interests as defined in the ICD-10 codes mentioned in our introduction. Data
on ambient concentrations of PM2.5, PM10, NO2, and O3 were recorded from hourly and daily
observations from a network of more than 150 stations in California from 2014 to 2019 from the
US Environmental Protection Agency’s Air Quality System. Calculations for pollution
concentration were based on inverse distance squared weighting, and the recorded residential
address of the decedents were used to link them with this pollution data. Temperature data was
provided by daily gridMET data, a gridded dataset measuring daily meteorological variables for

4

the continental United States. Using nearest grid or inverse distance weighting, block group
locations were assigned daily temperature and relative humidity during the years 2014 to 2019.
As is classically used in the context of case-crossover study designs, conditional logistic
regression was used as the primary data analysis method. Specifically, two separate conditional
logistic regression models were used to evaluate the association between deaths due to suicides
and homicides with heat and air pollution. From there, single-exposure models were evaluated,
and then two-exposure models with temperature paired with a pollutant were assessed. Finally,
interactions between temperature and a pollutant were tested for significance. All of these
models were adjusted for relative humidity with smooth temperature to control for potential
confounding.
The general form of our single-exposure models will be written as
𝑙𝑜𝑔𝑖𝑡 (𝜋 𝑠𝑢𝑖𝑐𝑖𝑑𝑒 (𝑥 )) = 𝛽 0
+ 𝛽 𝑇 𝑋 𝑇 + 𝑛𝑠 (𝑋 𝑅𝐻
)
𝑙𝑜𝑔𝑖𝑡 (𝜋 𝑠𝑢𝑖𝑐𝑖𝑑𝑒 (𝑥 )) = 𝛽 0
+ 𝛽 𝑃 𝑋 𝑃 + 𝑛 𝑠 (𝑋 𝑅𝐻
)
𝑙𝑜𝑔𝑖𝑡 (𝜋 ℎ𝑜𝑚𝑖𝑐𝑖𝑑𝑒 (𝑥 )) = 𝛽 0
+ 𝛽 𝑇 𝑋 𝑇 + 𝑛𝑠 (𝑋 𝑅𝐻
)
𝑙𝑜𝑔𝑖𝑡 (𝜋 ℎ𝑜𝑚𝑖𝑐𝑖𝑑𝑒 (𝑥 )) = 𝛽 0
+ 𝛽 𝑃 𝑋 𝑃 + 𝑛𝑠 (𝑋 𝑅𝐻
)
where T is maximum or minimum temperature, P is one of the four air pollutants, and RH is
relative maximum/minimum humidity. 𝑛𝑠 (𝑋 𝑅𝐻
) denotes a natural cubic spline with 3 degrees of
freedom.
The two-exposure models are
𝑙𝑜𝑔𝑖𝑡 (𝜋 𝑠𝑢𝑖𝑐𝑖𝑑𝑒 (𝑥 )) = 𝛽 0
+ 𝛽 𝑇 𝑋 𝑇 + 𝛽 𝑃 𝑋 𝑃 + 𝑛𝑠 (𝑋 𝑅𝐻
)

5

𝑙𝑜𝑔𝑖𝑡 (𝜋 ℎ𝑜𝑚𝑖𝑐𝑖𝑑𝑒 (𝑥 )) = 𝛽 0
+ 𝛽 𝑇 𝑋 𝑇 + 𝛽 𝑃 𝑋 𝑃 + 𝑛𝑠 (𝑋 𝑅𝐻
)
And the interaction models will be
𝑙𝑜𝑔𝑖𝑡 (𝜋 𝑠𝑢𝑖𝑐𝑖𝑑𝑒 (𝑥 )) = 𝛽 0
+ 𝛽 𝑇 𝑋 𝑇 + 𝛽 𝑃 𝑋 𝑃 + 𝛽 𝑇𝑃
𝑋 𝑇 𝑋 𝑃 + 𝑛𝑠 (𝑋 𝑅𝐻
)
𝑙𝑜𝑔𝑖𝑡 (𝜋 ℎ𝑜𝑚𝑖𝑐𝑖𝑑𝑒 (𝑥 )) = 𝛽 0
+ 𝛽 𝑇 𝑋 𝑇 + 𝛽 𝑃 𝑋 𝑃 + 𝛽 𝑇𝑃
𝑋 𝑇 𝑋 𝑃 + 𝑛𝑠 (𝑋 𝑅𝐻
)
This will result in a total of 12 single-exposure models, 16 two-exposure models, and 16
interaction models. Correlations between each exposure will be cautiously assessed to avoid
multicollinearity.
Models containing relative maximum/minimum humidity and an indicator variable of
whether a day had temperature at or above a specific percentile were used to assess the potential
relationship between deaths due to suicide or homicide and the event that a day was hotter than a
particular percentile:
𝑙𝑜𝑔𝑖𝑡 (𝜋 𝑠𝑢𝑖𝑐𝑖𝑑𝑒 (𝑥 )) = 𝛽 0
+ 𝛽 𝑄 𝑋 𝑄 + 𝑛𝑠 (𝑋 𝑅𝐻
)
𝑙𝑜𝑔𝑖𝑡 (𝜋 ℎ𝑜𝑚𝑖𝑐𝑖𝑑𝑒 (𝑥 )) = 𝛽 0
+ 𝛽 𝑄 𝑋 𝑄 + 𝑛𝑠 (𝑋 𝑅𝐻
)
where Q denotes the indicator variable of a day being at or above a particular percentile of
maximum/minimum temperature.
Results
In the years 2014 to 2019 in the state of California, there were 1,514,292 deaths. 24,250
of these deaths were due to suicide and 10,148 deaths were homicide cases. The average suicide
victim was approximately 49 years old, and more victims were male than female.

6

Table 1: Demographical information of suicide victims. Environmental exposures listed here pertain to the individual’s day of
death.
Deaths due to Suicide (n = 24,250)
Variable Mean (SD) / Count (% Frequency)
Age, years 49.43 (19.25)
Maximum Temperature, degrees Celsius 24.27 (7.10)
Minimum Temperature, degrees Celsius 11.81 (5.27)
O3 (ppb) 43.45 (14.33)
PM10 (µg/m
3
) 25.90 (18.61)
PM2.5 (µg/m
3
) 10.08 (8.02)
NO2 (ppb) 10.40 (7.40)
Sex
Male 18,726 (77.2)
Female 5,524 (22.8)
Maximum Relative Humidity, percentage 80.04 (18.24)
Minimum Relative Humidity, percentage 37.00 (18.41)

Victims of homicide were about 36 years old on average, and more individuals were male
than female.
For both suicide and homicide cases, measures of environmental factors are
approximately the same.
Table 2: Demographical information of homicide victims. Environmental exposures listed here pertain to the individual’s day of
death.
Deaths due to Homicide (n = 10,148)
Variable Mean (SD) / Count (% Frequency)
Age, years 35.59 (15.97)
Maximum Temperature, degrees Celsius 24.56 (7.11)
Minimum Temperature, degrees Celsius 12.14 (5.31)
O3 (ppb) 43.49 (15.05)
PM10 (µg/m
3
) 28.29 (18.31)
PM2.5 (µg/m
3
) 10.82 (8.16)
NO2 (ppb) 11.49 (7.70)
Sex
Male 8,498 (83.7)
Female 1,650 (16.3)
Maximum Relative Humidity, percentage 79.40 (17.90)
Minimum Relative Humidity, percentage 36.5118.24)

7

I. Single-Exposure Models
Table 3: Associations between temperature, pollutants, and deaths due to suicide in single-exposure models.
Deaths due to Suicide
Exposure Coefficient
Estimate
Odds 95% CI P-value
Maximum Temperature 0.0044 1.0044 (1.0002, 1.009) 0.0381
Minimum Temperature 0.0075 1.0076 (1.0017, 1.0135) 0.0120
O3 0.0005 1.0005 (0.9986, 1.0023) 0.6210
PM10 0.0009 1.0009 (0.9997, 1.0022) 0.1547
PM2.5 0.0001 1.0001 (0.9980, 1.0023) 0.8890
NO2 0.0006 1.0006 (0.9971, 1.0041) 0.7444

Table 4: Associations between temperature, pollutants, and deaths due to homicide in single-exposure models.
Deaths due to Homicide
Exposure Coefficient
Estimate
Odds 95% CI P-value
Maximum Temperature 0.0086 1.0086 (1.0023, 1.015) 0.0072
Minimum Temperature 0.0104 1.0105 (1.0015, 1.020) 0.0220
O3 -0.0009 0.0001 (0.9965, 1.0018) 0.5186
PM10 0.0015 1.0015 (0.9996, 1.0034) 0.1143
PM2.5 0.0014 1.0014 (0.9982, 1.0045) 0.3906
NO2 0.0025 1.0025 (0.9975, 1.007) 0.3256

After controlling for maximum relative humidity, there is enough evidence to suggest
that maximum daily temperature is significantly associated with deaths due to suicide (p-value =
0.0381; OR = 1.0044; 95% CI: 1.0002, 1.009). For every 1-degree Celsius increase in maximum
daily temperature, the odds of an individual dying by suicide in California in the years 2014 to
2019 increases by about 0.44% (or 2.22% for every 5-degree Celsius increase in maximum daily
temperature). Likewise, minimum temperature is significantly associated with death due to
suicide after controlling for minimum relative humidity (p-value = 0.0120; OR = 1.0076; 95%
CI: 1.0017, 1.0135). The odds of an individual dying by suicide increases by about 0.76% for

8

every 1-degree Celsius increase in minimum daily temperature (or 3.82% for every 5-degree
Celsius increase in minimum daily temperature).
As with deaths due to suicide, there is enough evidence to suggest that maximum daily
temperature is associated with deaths due to homicide after adjusting for maximum relative
humidity (p=0.0072; OR=1.0086; 95% CI: 1.0023, 1.015). For every 1-degree Celsius increase
in maximum daily temperature, the odds of an individual in California during the years 2014 to
2019 dying by homicide increases by approximately 0.86% (or 4.39% for every 5-degree
increase in maximum daily temperature).
These results are also reflected in minimum daily temperature after adjusting for
minimum relative humidity (p=0.0220; OR=1.0105; 95% CI: 1.0015, 1.020). For every 1-degree
Celsius increase in minimum daily temperature, the odds of an individual dying by homicide in
California in the years 2014 to 2019 increases by approximately 1.05% (or 5.34% for every 5-
degree Celsius increase in minimum daily temperature).
When considered individually, there is not enough evidence to suggest that any of the
four pollutants were significantly associated with deaths due to suicide or homicide.
II. Two-Exposure Models
Multicollinearity was carefully considered before running models with two exposures.
Table 5 displays the correlations between each exposure using all mortality data, since the results
of separate correlation charts of suicide and homicide cases are quite similar. This correlation
chart suggests that maximum and minimum temperature, PM10 and PM2.5, and O3 and maximum
temperature should not exist in the same model. The strong association between O3 and

9

maximum temperature intuitively makes sense since sunlight contributes to the production of
ozone.
Hence, we proceeded by including maximum or minimum temperature accompanied by a
pollutant in each model (and proceeded with caution when evaluating maximum temperature and
O3 in the same model).
Table 5: Correlation chart of exposures (all-cause mortality).
O3 PM10 PM2.5 NO2 Maximum
Temperature

Minimum
Temperature

O3 1 0.3125 0.0281 -0.1571 0.6616 0.4886
PM10 0.3125 1 0.6317 0.3412 0.3695 0.2005
PM2.5 0.0281 0.6317 1 0.3734 0.0502 -0.0308
NO2 -0.1571 0.3412 0.3734 1 0.0225 -0.1142
Maximum
Temperature
0.6616 0.3695 0.0502 0.0225 1 0.7675
Minimum
Temperature
0.4886 0.2005 -0.0308 -0.1142 0.7675 1

Table 6: Main effect estimates of two-exposure models for deaths due to suicide.
Deaths due to Suicide
Pair of Exposures Coefficient
Estimate
Odds 95% CI P-value Wald P-
Value
Maximum Temperature 0.0045 1.0045 (1.0000, 1.009) 0.0500 0.01
O 3 -0.00002 0.9998 (0.9979, 1.002) 0.8319
Maximum Temperature 0.0055 1.0055 (1.0006, 1.010) 0.0292 0.009
PM 10 0.0007 1.0007 (0.9994, 1.002) 0.3130
Maximum Temperature 0.0004 1.004 (0.9998, 1.008) 0.0633 0.009
PM 2.5 -0.0001 0.9999 (0.9978, 1.002) 0.9381
Maximum Temperature 0.0050 1.0050 (1.0002, 1.010) 0.0417 0.02
NO 2 -0.0011 0.9989 (0.9951, 1.003) 0.5756
Minimum Temperature 0.0074 1.0074 (1.0014, 1.013) 0.0148 0.009
O 3 0.0003 1.0003 (0.9983, 1.002) 0.7860
Minimum Temperature 0.0082 1.0083 (1.0015, 1.015) 0.0164 0.009
PM 10 0.0007 1.0007 (0.9994, 1.002) 0.2726
Minimum Temperature 0.0078 1.0079 (1.0019, 1.014) 0.0101 0.009
PM 2.5 0.00003 1.0000 (0.9979, 1.002) 0.9776
Minimum Temperature 0.0086 1.0087 (1.0025, 1.015) 0.0058 0.020
NO 2 0.0007 1.0007 (0.9967, 1.005) 0.7359

10

Table 7: Main effect estimates of two-exposure models for deaths due to homicide.
Deaths due to Homicide
Pair of Exposures Coefficien
t Estimate
Odds 95% CI P-value Wald P-
Value
Maximum Temperature 0.0112 1.0113 (1.0045, 1.018) 0.0011 0.003
O 3 -0.0026 0.9974 (0.9946, 1.000) 0.07845
Maximum Temperature 0.0083 1.0083 (1.0011, 1.016) 0.0246 0.01
PM 10 0.0011 1.0011 (0.9991, 1.003) 0.2764
Maximum Temperature 0.0091 1.0091 (1.0027, 1.016) 0.0055 0.009
PM 2.5 0.0007 1.0007 (0.9975, 1.004) 0.6571
Maximum Temperature 0.0095 1.0095 (1.0023, 1.017) 0.009 0.02
NO 2 -0.0008 0.9992 (0.9937, 1.005) 0.7869
Minimum Temperature 0.0111 1.0112 (1.0021, 1.020) 0.0157 0.01
O 3 -0.0019 0.9981 (0.9953, 1.001) 0.1944
Minimum Temperature 0.0081 1.0081 (0.9980, 1.018) 0.1160 0.05
PM 10 0.0013 1.0013 (0.9993, 1.003) 0.2060
Minimum Temperature 0.0108 1.0109 (1.0017, 1.020) 0.0196 0.02
PM 2.5 0.0009 1.0009 (0.9977, 1.004) 0.5817
Minimum Temperature 0.0102 1.0102 (1.0010, 1.020) 0.0293 0.04
NO 2 0.0009 1.0009 (0.9952, 1.007) 0.7496

There is still enough evidence to suggest that maximum and minimum temperature are
associated with deaths due to suicide after adjusting for various pollutants, as well as relative
humidity. However, there is not enough evidence to suggest an association between the four
pollutants and deaths due to suicide, even after adjusting for maximum and minimum daily
temperature and relative humidity.

11

III. Evaluation of Linearity Assumption for Single-Exposure Models
Table 8: P-values of linear, square, and cubic terms for exposure in polynomial
models to assess for non-linearity in exposure-response

Exposure Variable Term P-value
Deaths due to Suicide
Maximum Temperature Linear 0.1891
Squared 0.5971
Cubic 0.6410
Maximum Temperature Linear 0.0063

Squared 0.8950

Cubic 0.2568
Deaths due to Homicide

Maximum Temperature Linear 0.0761

Squared 0.7239

Cubic 0.6871
Maximum Temperature Linear 0.0032

Squared 0.3015

Cubic 0.1303

A polynomial approach was used to test for non-linearity in our single-exposure models.
Maximum and minimum temperature were first centered upon their mean to reduce correlation
between the polynomial terms. Then, these mean-centered variables were raised to the squared
and cubic powers. The model had either death due to suicide or homicide as its outcome and
included maximum/minimum temperature as a linear term, then raised to the second and third
degree. Maximum/minimum relative humidity was included as a covariate as a potential
confounding factor similar to the main models.
At the 0.05 significance level, the p-values associated with each polynomial term do not
provide enough evidence for a non-linear relationship. There is not enough evidence to support
that a square or cubic term fits the data well; in other words, it cannot be concluded that there is
sufficient evidence for non-linearity. These findings support a linear relationship between
maximum/minimum temperature and deaths due to suicide and homicide.

12

Figure 1: Smooth spline plots of maximum/minimum temperatures on
case days. Top row: deaths due to suicide. Bottom row: deaths due to
homicide

We also inspected evidence for non-linearity visually. Figure 1 provides a visual
interpretation of the relationship between maximum/minimum temperature and deaths due to
suicide and homicide based on models with a natural cubic spline term on exposure with degrees
of freedom (df) = X. All smooth spline plots imply a linear relationship between the exposure
and outcome, except for the plot between minimum temperature and homicide (bottom right). In
this plot, the risk score for homicide decreases until approximately 5°C, the degree at which it
begins to increase. The 95% confidence intervals are also quite wide in this range of minimum
temperature since there are fewer observations observed at the range’s endpoints, thus indicating
the uncertainty around the estimated slope of the smooth spline.

13

IV. Interaction Models
Table 9: Statistical measures of interaction terms regarding deaths due to suicide.
Deaths due to Suicide
Pair of Exposures Coefficient
Estimate
P-value
Maximum Temperature* O3 -0.00006 0.5150
Maximum Temperature* PM10 0.00001 0.890
Maximum Temperature* PM2.5 -0.0001 0.4268
Maximum Temperature* NO2 0.00004 0.848
Minimum Temperature* O3 -0.0003 0.055
Minimum Temperature* PM10 -0.00001 0.9667
Minimum Temperature* PM2.5 -0.0003 0.1298
Minimum Temperature* NO2 0.0001 0.6352

Table 10: Statistical measures of interaction terms regarding deaths due to
homicide.
Deaths due to Homicide
Pair of Exposures Coefficient
Estimate
P-value
Maximum Temperature* O3 -0.00006 0.6570
Maximum Temperature* PM10 -0.00002 0.0845
Maximum Temperature* PM2.5 -0.00036 0.1378
Maximum Temperature* NO2 -0.00045 0.1442
Minimum Temperature* O3 -0.00005 0.8203
Minimum Temperature* PM10 -0.00025 0.0705
Minimum Temperature* PM2.5 -0.00022 0.3890
Minimum Temperature* NO2 -0.00030 0.4740

There is not enough evidence to suggest an interaction between maximum temperature
and any pollutant when the outcome was deaths due to suicide. However, there was evidence of a
borderline statistically significant negative interaction between minimum temperature with ozone
exposure (p-value = 0.055). No other pollutant interaction term was statistically significant for
minimum temperature.

14

A marginally significant negative interaction between maximum temperature and PM10
(p-value = 0.0845) as well as minimum temperature and PM10 (p-value = 0.0705) exists for the
outcome of homicide. No other temperature-pollutant interaction term was statistically
significant for homicide.
i. Significant Mean-Centered Interaction Models

Figure 2: Interaction plot between minimum temperature and ozone
for deaths due to suicide.

Table 11: Statistical measures of mean-centered main effects and interaction
between minimum temperature and ozone for deaths due to suicide.
Deaths due to Suicide
Coefficient P-Value
Minimum
temperature
0.0075 0.0130
Ozone 0.0009 0.3672
Min.Temp*Ozone -0.0003 0.0546

15

We know that there is statistically significant interaction between minimum temperature
and ozone for death due to suicide. The interaction coefficient estimate is less than 0, meaning
that the suicide mortality effect of minimum temperature is lowered as levels of ozone increases,
and vice versa (e.g., temperature mortality effect increased with lower ozone exposure). Figure 2
illustrates that when ozone levels are one standard deviation lower (-14.33 ppb; red line) than the
mean (blue line), minimum temperature has a stronger effect on death due to suicide, as
indicated by the steeper slope. Similarly, when ozone levels are one standard deviation higher
(+14.33 ppb; green line) than the mean (blue line), the effect of minimum temperature on death
due to suicide is lower.

Figure 3: Interaction plot between maximum temperature and
PM10 for deaths due to homicide.

16

Table 12: Statistical measures of mean-centered main effects and interaction
between maximum temperature and PM10 for deaths due to homicide.
Deaths due to Homicide
Coefficient P-Value
Maximum
temperature
0.0083 0.025
PM10 0.0009 0.3551
Max.Temp* PM10 -0.0002 0.0845

The coefficient estimate of the interaction term between maximum temperature and PM10
is once again less than 0, which means that the homicide mortality effect of maximum
temperature is lowered as levels of PM10 increases, and the maximum temperature mortality
effect increases with lower exposure levels of PM10. Figure 3 shows that when PM10 levels are
one standard deviation (-17.71 µg/m
3
red line) are lower than the mean (blue line), maximum
temperature has a stronger effect on death due to homicide, as indicated by the steeper slope.
Likewise, when PM10 levels are one standard deviation higher (+17.71 µg/m
3
; green line) than
the mean (blue line), the effect of maximum temperature on death due to homicide is lower.

17

Figure 4: Interaction plot between minimum temperature and PM10
for deaths due to homicide.

Table 13: Statistical measures of mean-centered main effects and interaction
between minimum temperature and PM10 for deaths due to homicide.
Deaths due to Homicide
Coefficient P-Value
Minimum
temperature
0.0087 0.0866
PM10 0.0007 0.5076
Min.Temp* PM10 -0.0003 0.0705

An antagonistic relationship is observed once more, as the coefficient estimate of the
interaction between minimum temperature and PM10 is less than 0: the homicide mortality effect
of minimum temperature is lowered as levels of PM10 increases; conversely, the temperature
mortality effect increases with lower PM10 exposure levels. Figure 4 demonstrates that when PM10
levels are one standard deviation lower (-17.71 µg/m
3
; red line) than the mean (blue line),
maximum temperature has a stronger effect on death due to homicide. When PM10 levels are one

18

standard deviation higher (+17.71 µg/m
3
; green line) than the mean (blue line), the effect of
maximum temperature on death due to homicide is lower.
V. Maximum/Minimum Heat Percentile
Table 14: Statistical measures of deaths due to suicide and homicide over various percentiles of maximum temperature.
Maximum Temperature Over Various Percentiles
Coefficient Odds 95% CI P-value
Death due to Suicide
90
th
percentile (n =2,383) 0.0226 1.0228 (0.9688, 1.0799) 0.4147
95
th
percentile (n = 1,270) 0.0322 1.0355 (0.9634, 1.1071) 0.3637
97
th
percentile (n = 776) 0.0289 1.0293 (0.0444, 1.1218) 0.5111
99
th
percentile (n = 247) -0.0069 0.9932 (0.8590, 1.1483) 0.9261
Death due to Homicide
90
th
percentile (n = 1,083) 0.0614 1.0633 (0.9805, 1.1531) 0.1377
95
th
percentile (n = 554) 0.0299 1.0304 (0.9270, 1.1453) 0.5793
97
th
percentile (n = 322) -0.0576 0.9440 (0.8264, 1.0784) 0.3962
99
th
percentile (n = 114)

-0.0091 1.0092 (0.8145, 1.2504) 0.9335

Table 15: Statistical measures of deaths due to suicide and homicide over various percentiles of minimum temperature.
Minimum Temperature Over Various Percentiles
Coefficient Odds 95% CI P-value
Death due to Suicide
90
th
percentile (n = 2,557) 0.0249 1.0253 (0.9700, 1.0837) 0.3779
95
th
percentile (n = 1,302) 0.0441 1.0450 (0.9746, 1.1206) 0.2159
97
th
percentile (n = 790) 0.0563 1.0579 (0.9707, 1.1530) 0.1998
99
th
percentile (n = 256) 0.0025 1.0025 (0.8692, 1.156) 0.9723
Death due to Homicide
90
th
percentile (n = 1,154) 0.0658 1.0680 (0.9818, 1.162) 0.1254
95
th
percentile (n = 592) 0.0840 1.0876 (0.9784, 1.209) 0.1196
97
th
percentile (n = 361) 0.0642 1.0663 (0.9380, 1.212) 0.3263
99
th
percentile (n = 125)

0.1304 1.1393 (0.9261, 1.402) 0.2172

It was eventually found that there is not enough evidence to establish a statistically
significant relationship between deaths due to suicide/homicide and days that were hotter than

19

the 90
th
, 95
th
, 97
th
, or 99
th
percentile of maximum/minimum temperature after controlling for
relative humidity.
Discussion
After adjusting for relative humidity, same-day maximum/minimum temperature was
significantly associated with suicide mortality risk, even after individually adjusting for levels of
ozone, nitrogen dioxide, PM10, and PM2.5. There was some evidence of a negative interaction
between daily minimum temperature and ozone, indicating that the effect of minimum
temperature lowers as ozone increases. Further research is needed to understand this potentially
complex relationship in a practical context.
Both maximum and minimum temperature were also related to deaths due to homicide
after adjusting for relative humidity, and after adjusting individually for levels of ozone, nitrogen
dioxide, and PM2.5 as well. The effect of minimum temperature on homicide loses significance
after controlling for levels of PM10 (from p-value = 0.0220 to 0.1160), but effect for minimum
temperature did not markedly change after adjusting for PM10, meaning that PM10 was not a
confounder of the minimum temperature-homicide mortality relationship.
The percentile-based exposure effects were all nonsignificant, had wide 95% confidence
intervals, and had nonsignificant p-values. This is possible due in part to low numbers when
using this percentile-based exposure assessment strategy. Coefficient estimates of the effect of
minimum temperature over various percentiles were all greater than 0, but have nonsignificant p-
values, and the fact that each percentile stratum has a small sample size must be taken into great
consideration. The risk of dying by homicide increases by 6.3% on days that had maximum
temperature at or above the 90
th
percentile. Also, this risk of dying by homicide increases by

20

13.9% on days that had minimum temperature at or above the 99
th
percentile. Of course, since
these two results have nonsignificant p-values, there is not enough evidence at the standard 0.05
significance level to support these statements, but perhaps with a larger sample size and further
research, similar effect sizes could be observed.
The specific definition of a p-value is loosely defined as the probability of observing an
event actually being due to chance. So, the probability that the event “the risk of dying by
homicide increases by 6.3% on days that had maximum temperature at or above the 90
th

percentile” occurring solely by chance is approximately 13.8%, which is not too far from the
standard 5% threshold. Closely considering these p-value estimates and keeping a literal
interpretation in mind, these results should be taken into at least some consideration as an
interesting outcome of the percentile models. 18.8% is still nevertheless over the 5% standard
threshold: future research is needed, and a particularly large sample should be collected and
analyzed to verify the validity of these results.
Conclusion
The effects of heat are indeed significantly associated with the risk of dying by suicide
and homicide. At the same time, there is little evidence to suggest that ozone, nitrogen dioxide,
PM10, and PM2.5 are significantly associated with the outcomes of interest. Considering the fact
that the state of California is reaching record-high temperatures, it is vital that attention towards
suicide prevention is raised during hotter days, which could be seen as a duration of “high risk
times”. Resources for mental therapy and promotion towards healthier cognitive behavioral
practices could be bolstered and fortified during hotter days could prevent more cases of both
homicide and suicide from occurring.

21

References
Berry, H. L., Bowen, K., & Kjellstrom, T. (2010). Climate change and mental health: a causal
pathways framework. International journal of public health, 55(2), 123–132.
https://doi.org/10.1007/s00038-009-0112-0
Cornelius, S. L., Berry, T., Goodrich, A. J., Shiner, B., & Riblet, N. B. (2021). The Effect of
Meteorological, Pollution, and Geographic Exposures on Death by Suicide: A Scoping
Review. International journal of environmental research and public health, 18(15), 7809.
https://doi.org/10.3390/ijerph18157809
Dumont, C., Haase, E., Dolber, T., Lewis, J., & Coverdale, J. (2020). Climate Change and Risk
of Completed Suicide. The Journal of nervous and mental disease, 208(7), 559–565.
https://doi.org/10.1097/NMD.0000000000001162
Gao, J., Cheng, Q., Duan, J., Xu, Z., Bai, L., Zhang, Y., Zhang, H., Wang, S., Zhang, Z., & Su,
H. (2019). Ambient temperature, sunlight duration, and suicide: A systematic review and meta-
analysis. The Science of the total environment, 646, 1021–1029.
https://doi.org/10.1016/j.scitotenv.2018.07.098
Nguyen, A. M., Malig, B. J., & Basu, R. (2021). The association between ozone and fine
particles and mental health-related emergency department visits in California, 2005-2013. PloS
one, 16(4), e0249675. https://doi.org/10.1371/journal.pone.0249675
Thompson R, Hornigold R, Page L, Waite T. Associations between high ambient temperatures
and heat waves with mental health outcomes: a systematic review. Public Health. 2018
Aug;161:171-191. doi: 10.1016/j.puhe.2018.06.008. Epub 2018 Jul 12. PMID: 30007545.

22

Palinkas LA, Wong M. Global climate change and mental health. Curr Opin Psychol. 2020
Apr;32:12-16. doi: 10.1016/j.copsyc.2019.06.023. Epub 2019 Jul 2. PMID: 31349129.
Xu, R., Xiong, X., Abramson, M. J., Li, S., & Guo, Y. (2020). Ambient temperature and
intentional homicide: A multi-city case-crossover study in the US. Environment
international, 143, 105992. https://doi.org/10.1016/j.envint.2020.105992
Yarza, S., Vodonos, A., Hassan, L., Shalev, H., Novack, V., & Novack, L. (2020). Suicide
behavior and meteorological characteristics in hot and arid climate. Environmental
research, 184, 109314. https://doi.org/10.1016/j.envres.2020.109314

Asset Metadata

Creator Lorenzo, Melissa (author)

Core Title The effects of heat and air pollution on mental-health related mortality

Contributor Electronically uploaded by the author (provenance)

School Keck School of Medicine

Degree Master of Science

Degree Program Biostatistics

Degree Conferral Date 2022-05

Publication Date 04/15/2022

Defense Date 04/14/2022

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag Air pollution,Heat,Homicide,Mental Health,mortality,nitrogen dioxide,oai:digitallibrary.usc.edu:usctheses,OAI-PMH Harvest,ozone,particulate matter,suicide,temperature

Format application/pdf (imt)

Language English

Advisor Garcia, Erika (committee chair), Conti, David (committee member), Gauderman, James (committee member)

Creator Email melissalorenzo05@gmail.com,mklorenz@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC110964818

Unique identifier UC110964818

Document Type Thesis

Format application/pdf (imt)

Rights Lorenzo, Melissa

Type texts

Source 20220415-usctheses-batch-925 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email uscdl@usc.edu

Abstract (if available)

Abstract Background: Climate change and its associated adverse outcomes is an area with much study. Yet not much is known of the connection between air pollution and high heat -- two exposures thought to increase with climate change -- to mental health related outcomes: specifically, deaths due to suicide and homicide. This is particularly relevant for California, which is prone to elevated temperatures and droughts, the latter of which possibly affecting air quality through increased frequencies of wildfires. This analysis aims to explore the relationship between heat and air pollution with suicide and homicide deaths in California during the years 2014 to 2019, which included some of the hottest years on record.
Methods: Death certificate data for California from 2014 to 2019 were used in a case-crossover study design. For each individual, date of death was defined as the case day and the other days within the same month that were the same day of week were control days. Residential data for decedents were used to assess exposure to daily temperature (maximum, minimum) and daily average air pollution concentrations (particulate matter <10 µm [PM10] and <2.5 µm, nitrogen dioxide, ozone). Deaths due to suicide and homicide were treated as separate outcomes in conditional logistic regression models, adjusting for relative humidity. Single-exposure, two-exposure (temperature and pollution), and interaction models were explored.
Results: Maximum daily temperature was significantly associated with deaths due to suicide (p-value = 0.0381; odds ratio = 1.0044; 95% CI: 1.0002, 1.009), meaning that for every 1-degree Celsius increase in maximum daily temperature, the odds of an individual dying by suicide increases by about 0.44%. We observed pollutants to interact with minimum daily temperature: the effect temperature has on mental-health mortality increased with lower exposure levels of ozone and PM 10.
Conclusion: This analysis demonstrates a statistically significant linear relationship between the odds of dying due to suicide or homicide and continuous maximum/minimum daily temperature. More research is needed to confirm and better understand the observed interactions between air pollutants and temperature. The results of this paper suggest raising awareness of the possible connection between the effects of temperature on suicide and homicide deaths, as well as bolstering preventive public health policies during hotter days.