Coincident Patterns of Suicide Risk Among Adult Patients with a Primary Solid Tumor: A Large-Scale Population Study.

BACKGROUND: Suicide rate is much higher in cancer patients than in general population. This study examined the suicide risk in survivors of primary solid tumor across 19 cancer sites considering risk coincident patterns based on area-based SES indicators.
METHODS: A retrospective search of the SEER database was conducted. Independent risk factors for suicide were identified using the Cox proportional-hazards model. Exploratory factor analysis and cluster analysis were used to create coincident patterns of SES factors.
RESULTS: Suicide risk was higher for patients with a primary solid tumor who were older, male, white, unmarried, had no insurance, poorly differentiated, distant metastasis and did not undergo active treatment (especially surgery). The suicide risk was higher for patients living in areas with economic and education disadvantage, high levels of immigration and crowding, and high levels of residential instability. Concomitant presence of high economic and education disadvantage, high immigration and crowding levels and low residential instability, showed the highest risk of suicide.
CONCLUSION: In order to mitigate suicidal risk, clinicians should pay more attention to patients who are older, male, white, not married, high levels of cancer severity, not received active treatment (especially surgery), and having no insurance. Identifying coincident patterns of suicide help further screen high suicidal risk patients based on area-based socioeconomic status.

Introduction

Cancer and suicide are both major causes of death worldwide,1 and patients with cancer have a suicide rate that is double that of the general population.2,3 Cancer affects patients in many ways, including increasing psychological distress, worsening psychosocial conditions, and decreasing the quality of life. Many of these factors are related to suicide risk.4–6

Similar to how unmanaged distress is associated with suicidal behavior in the general population, depressive disorders (including major and minor depression, and demoralization) affect up to 60% of patients with cancer.7 The side effects of cancer treatment, such as fatigue and pain, also cause a depressive mood, which is difficult to distinguish from cancer-related depression.8 Identifying patients with a high risk of suicide is vital for maintaining ongoing increases in cancer survival rates.

Previous studies have reported that patients across several cancer sites have an increased risk of suicidal death.9–12 Zaorsky NG has identified specific suicide risk factors in US cancer patients, including male gender, advanced disease, white race, and unmarried status.13 In patients with hematological cancers, higher suicide risk was found in those without active treatment.14 However, few studies were found to investigate suicide risk factors in whole patient population with a primary solid tumor. Cancer is commonly divided into solid cancers and hematological cancers, and solid tumors make up most of all cancers. Solid tumors share a similar staging method and cancer-directed surgery is the main therapy for solid tumors. Therefore, it is necessary to further elucidate underlying suicide risk factors in solid tumor patients.

Socioeconomic conditions play an important role in predicting suicide risk in the general population.15 Several studies have found an inverse relationship between the socioeconomic status (SES) and the incidence of suicide. The following socioeconomic indicators have been observed to determine this relationship: basic levels of education,16 employment in blue-collar occupations,17 and low income.18 However, more information is needed to identify the socioeconomic predictors for suicide risk among cancer patients.

This study, based on the Surveillance, Epidemiology, and End Results (SEER) database from the United States, analyzed the suicide risk among patients with a primary solid tumor. The aim of this study is to determine the suicide risk among patients with a primary solid tumor while identifying coincident patterns with high suicide risk according to the area-based socioeconomic status.

Materials and Methods

Data Source

This was a retrospective study based on the 18-registry SEER database (with additional treatment fields). Patients' data were retrieved using the SEER*Stat software (version 8.3.6). SEER is a publicly available, nationally representative, population-based cancer database which contains more than 8 million cancer cases. SEER database spanned 4 decades and cover 28% of the United States population, which is considered a valid source of cancer incidence and survival data in the United States.19–21 SEER has developed and maintained high-quality, validated data on causes of death among cancer survivors, providing insight into relative and cause-specific deaths in this population. Publicly available de-identified data from the SEER database were exempt from requiring a medical ethics review, and informed consent was not required.

Selection of Patients

The suicide risk of patients diagnosed with a primary solid tumor between 1975 and 2016 was analyzed. The following 19 most-common sites for primary solid tumors in US patients were analyzed: brain, breast, cervix uteri, colorectum, esophagus, gallbladder, kidney, larynx, liver, lung and bronchus, skin (melanoma), oral cavity, ovary, pancreas, prostate, small intestine, stomach, thyroid, and urinary bladder. Patients who were younger than 18 years, diagnosed via autopsy or death certificate, or whose detailed information was not available were not included in this study.

Variable Selection

The following data were collected from the SEER database: site of cancer, age at diagnosis, sex, race, marital status, grade, stage, year of diagnosis, insurance status, therapy methods, cause of death, and survival time. After excluding patients younger than 18 years old, age groups were classified by every 20 years (18–39, 40–59, 60–79, and ≥80 years old). Sex was classified as male or female. Race was classified as black, white, or other. Grade was classified as Grade1 (well differentiated), Grade2 (moderately differentiated), Grade3 (poorly differentiated), Grade4 (undifferentiated). Stage was classified as localized, regional, distant. Marriage was classified as married, unmarried (including divorced, separated, or widowed), or never married (single or domestic partner). Insurance status was classified as insured (including Medicare or Medicaid) or uninsured. The included therapy methods were surgery, radiation, and chemotherapy. Year of diagnosis was grouped by every ten years from 1975 to 2016.

The crude and age adjusted suicide rate from 1975 to 2107 was extracted from SEER database Rate Session. The Rate was displayed as cases per 100 000 and the standard population was 2000 US std Population (19 age groups – Census P25-1130).

Outcome Definition

Mortality codes in SEER are assigned from death certificates, completed by the doctor caring for the patient at the time of demise. Patients were considered to have committed suicide if the cause of death was coded as: suicide and self-inflicted injury (50,220). Survival time in SEER is measured in months, and the minimum time is 1 month to any event. Living patients were censored at the time when the data were collected.

Socioeconomic Status Indicators

Socioeconomic status indicators are calculated using the Census American Community Survey (ACS) 5-year files. Various county-level SES indicators were examined. Education (3 variables): with an education of less than 9th grade, less than high school graduate, and at least a bachelor’s degree. Poverty (4 variables): the percentage of persons below poverty, families below poverty, persons below 150% poverty, and persons below 200% poverty in the county of residence. Crowding (1 variable): the percentage of households with more than one person per room. Income and employment (3 variables): unemployment (the percentage of persons over age 16 years who were unemployed) and median family and median household income measured in 2013 inflation-adjusted dollars. Relocation (8 variables): the percentage of individuals in the county age 5 years old older and the percentage of individuals in the county age 1 year and older who remained in the same house, moved within the county, moved to a different county within the same state, and moved to a different state. Immigration (4 variables): the percentage of persons who were foreign born, percentage of language isolated and percentages (ages ≥1 year) and (ages≥5 years) within the county that moved to the US from another nation.

Statistical Analysis

In this study, baseline clinicopathological features for patients who were or were not suicide decedents were evaluated and compared based on Chi-Squared test. A survival analysis was performed using the Cox proportional-hazards model to calculate hazard ratios, with the survival time being from diagnosis to suicide. Non-suicide deaths and living patients were censored. Twenty-three county-level socioeconomic variables available in the SEER database were subjected to an exploratory factor analysis using varimax rotation in order to reduce the number of variables. Each patient was assigned a factor score based on the result of exploratory factor analysis. A k-means clustering method was applied to these factor scores to form distinct groups within all patients based on heterogeneity of the SES factor structures. A univariate stratified Cox proportional-hazards model was used to determine how SES factors and SES cluster groups affect the suicide risk in cancer patients. All statistical tests were two sided, with statistical significance being determined by P<0.05. All statistical analyses were performed using R software ().

Results

Characteristics of Patients

This study reviewed 6,990,572 patients in the SEER database who had a primary solid tumor. The cause of death was suicide for 11,389 (0.16%) of these patients. Patients who committed suicide were mostly male (83.2%), white (92.4%), and 60–79 years old at the time of their cancer diagnosis (57.0%). Cancer patients were mostly diagnosed during year 2005~2016 (52.0%), and the proportion of suicide patients became lower during the years of 2005~2016 (33.0%) comparing to non-suicide patients counterparty (52.1%). The cancer sites associated with the most suicide deaths were the prostate (30.1%), lung/bronchus (13.0%), urinary bladder (6.7%), and breast (8.0%), which together constituted more than half of all cancer patients. Table 1 describes the baseline clinicopathological characteristics of the included patients.

Table 1

Demographic Characteristics Among Patients

Variables	All Patients	Non_Suicide	Suicide Patients	P value	Variables	All Patients	Non_Suicide	Suicide Patients	P value
		Patients (%)	(%)				Patients (%)	(%)
Total number	6,990,572	6,979,183	11,389		Surgery				0.053
Age				<0.001	Yes	4,294,982 (61.43)	4,288,023 (61.44)	6959 (61.10)
19~39	381,174 (5.45)	380,534 (5.45)	640 (5.62)		No	2,513,890 (35.96)	2,509,797 (36.0)	4093 (35.94)
40~59	1,933,151 (27.65)	1,930,203 (27.66)	2948 (25.88)		Unknown	181,700 (2.60)	181,363 (2.59)	337 (2.96)
60~79	3,608,905 (51.63)	3,602,408 (51.62)	6497 (57.05)		Radiation				<0.001
80+	1,067,342 (15.27)	1,066,038 (15.27)	1304 (11.45)		Yes	2,060,952 (29.48)	2,057,893 (29.49)	3059 (26.86)
Sex				<0.001	No/Unknown	4,918,231 (70.52)	4,921,290 (70.51)	8330 (73.14)
Male	3,661,619 (52.38)	3,652,143 (52.33)	9476 (83.20)		Chemotherapy				<0.001
Female	3,328,953 (47.62)	3,327,040 (47.67)	1913 (16.80)		Yes	1,696,086 (24.26)	1,694,285 (24.3)	1801 (15.81)
Race				<0.001	No/Unknown	5,283,097 (75.74)	5,284,898 (75.72)	9588 (84.2)
White	5,759,699 (82.39)	5,749,172 (82.38)	10,527 (92.43)		Year of Diagnosis				<0.001
Black	711,346 (10.18)	710,972 (10.19)	374 (3.28)		1975~1985	564,447 (8.07)	562,835 (8.06)	1612 (14.15)
Other	453,723 (6.49)	453,260 (6.49)	463 (4.07)		1985~1995	841,336 (12.04)	839,022 (12.02)	2314 (20.32)
Unknown	65,804 (0.94)	65,779 (0.94)	25 (0.22)		1995~2005	1,947,453 (27.86)	1,943,744 (27.85)	3709 (32.57)
Marital Status				<0.001	2005~2016	3,637,336 (52.03)	3,633,582 (52.06)	3754 (32.96)
Unmarried	1,716,627 (24.56)	1,714,066 (24.56)	2561 (22.49)		Tumor Site
Married	3,897,993 (55.76)	3,891,651 (55.76)	6342 (55.69)		Brain	125,731 (1.80)	125,620 (1.80)	111 (0.97)	<0.001
Never married	899,265 (12.86)	897,693 (12.86)	1572 (13.80)		Breast	1,319,573 (18.87)	1,318,665 (18.89)	908 (7.93)	<0.001
Unknown	476,687 (6.82)	475,773 (6.82)	914 (8.03)		Cervix Uteri	92,003 (1.32)	91,916 (1.32)	87 (0.76)	<0.001
Stage				<0.001	Colorectum	441,642 (6.32)	440,851 (6.32)	791 (6.95)	<0.001
Localized	2,267,891 (32.44)	2,264,277 (32.44)	3614 (31.73)		Esophagus	83,055 (1.19)	82,863 (1.19)	192 (1.69)	<0.001
Regional	1,387,945 (19.85)	1,385,753 (19.86)	2192 (19.25)		Kidney	221,887 (3.17)	221,555 (3.17)	332 (2.92)	<0.001
Distant	1,240,144 (17.74)	1,238,807 (17.75)	1337 (11.74)		Larynx	65,173 (0.93)	64,897 (0.93)	276 (2.42)	<0.001
Unstaged	550,902 (7.88)	550,089 (7.88)	813 (7.14)		Liver	119,282 (1.71)	119,208 (1.71)	74 (0.65)	<0.001
Unknown	715,943 (10.24)	714,357 (10.24)	1586 (13.93)		Lung/Bronchus	1,108,196 (15.85)	1,106,711 (15.85)	1485 (13.04)	<0.001
Grade (%)				<0.001	Melanoma/Skin	325,320 (4.65)	324,691 (4.65)	629 (5.52)	<0.001
1	664,382 (9.50)	663,236 (9.50)	1146 (10.06)		Oral Cavity	132,806 (1.90)	132,324 (1.90)	482 (4.23)	<0.001
2	2,100,975 (30.05)	2,097,222 (30.05)	3753 (32.95)		Ovary	140,418 (2.01)	140,314 (2.01)	104 (0.91)	<0.001
3	1,603,765 (22.94)	1,601,121 (22.94)	2644 (23.22)		Pancreas	224,751 (3.22)	224,566 (3.22)	185 (1.62)	<0.001
4	284,009 (4.06)	283,599 (4.06)	410 (3.60)		Prostate	1,302,873 (18.64)	1,299,444 (18.62)	3429 (30.11)	<0.001
5	2,337,441 (33.44)	2,334,005 (33.44)	3436 (30.17)		Small Intestine	34,052 (0.49)	33,995 (0.49)	57 (0.50)	<0.001
Any Insurance				<0.001	Stomach	154,854 (2.22)	154,627 (2.22)	227 (1.99)	<0.001
Yes	2,749,185 (39.33)	2,746,680 (39.36)	2505 (21.99)		Gallbladder	55,403 (0.79)	55,219 (0.79)	184 (1.62)	<0.001
No	74,146 (1.06)	74,058 (1.06)	88 (0.77)		Thyroid	186,610 (2.67)	186,429 (2.67)	181 (1.59)	<0.001
Unknown	4,167,241 (59.61)	4,158,445 (59.58)	8796 (77.23)		Urinary Bladder	302,013 (4.32)	301,248 (4.32)	765 (6.72)	<0.001

Suicide Risk According to Clinicopathological Characteristics

Figure 1 shows the suicide risk for patients with a primary solid tumor as analyzed using the Cox proportional-hazards model. For most cancer sites, Patients who were older, male, white, and presenting more serious illness (Grade 3 and 4, Stage of Regional and Distant) were more likely to die by suicide. Patients who were married, had insurance, and undergone active treatment therapies (surgery radiation and chemotherapy) had a lower suicide risk.

Figure 1

Impact of different clinical features on suicide risk by cancer sites.

There was a higher suicide risk for older patients with these cancer sites: kidney, lung/bronchus, skin (melanoma), thyroid, and urinary bladder, while younger patients with breast cancer indicated higher suicide risk. Radiation therapy was a protective factor for patients with breast, prostate, larynx, brain, or lung/bronchus cancer, while the suicide risk was increased in patients with colorectum cancer. Chemotherapy decreased the suicide risk in patients with breast, esophagus, prostate cancer.

Figure 1 also showed that the suicide risk significantly decreased during more recent years. Kaplan-Meier curve indicated a higher survival probability during the time interval of 1995~2016 and the trend of the curve did not indicate an aggregation of suicide occurrence with diagnosis time ().

Further subgroup analysis, which divided patients into male group and female group, showed a similar suicide risk for different gender except for age and chemotherapy factors. Female patients aged 18–59 years (compared to more than 60 years) were exposed with higher suicide risk, while older male patients were more likely to die by suicide. The result of subgroup analysis is shown in .

Trend in Suicide Rates, 1975–2017

Figure 2 shows the trend of suicide rate (adjusted by age) in US across 19 sites of solid tumors. In general, suicide rate was much higher in male patients, and the suicide rate was decreasing both in men and women during the year from 1975 to 2017. For male patients, suicide rate was higher in cancer sites of prostate, lung/bronchus, Colorectum and Urinary Bladder (Figure 2A). For female patients, suicide rate kept at lower levels, and breast cancer accounted for a higher proportion of suicide risk (Figure 2B). The crude suicide rate during the year 1975 to 2017 is shown in .

Figure 2

Trend in suicide rate for different gender, 1975–2017. (A) Suicide rate for male patients, (B) suicide rate for female patients.

Formation and Description of Socioeconomic Status Indicators: SES Factors and SES Clusters

Factor analysis of the 23 SES variables resulted in a solution involving 4 factors that explained 100% of the variance in the original variables. shows the correlation coefficients between the variables and factors (factor loading) and the proportion of information extracted from each variable. Factor 1 (the economic and education disadvantage factor) captured variables related to economic and education disadvantage, such as higher rates of community-level poverty, lower income, higher unemployment, a lack of higher education, and lower rates of immigration within the past year. Factor 2 (the residential instability factor) captured variables related to housing instability, such as lower rates of residents staying in the same house during the past year and higher rates of them moving within counties. Factor 3 (the immigration and crowding factor) captured variables related to immigration, such as higher rates of community-level language isolation, not being born in the US, crowding, education at or below 9th grade, and immigration to the US within the past year.

Cluster analysis of the SES factor scores identified four distinct SES groups. Figure 3A illustrates the patterns of factor scores across the clusters, while Table 2 presents the county-level socioeconomic characteristics of patients with a primary solid tumor falling into each of the four cluster groups. Cluster 1 is distinguished by medium economic and education disadvantage factors (Factor 1), high residential instability (Factor 2) and medium levels of immigration and crowding (Factor 3). Cluster 2 is characterized by high levels of economic and education disadvantage (Factor 1), low levels of residential instability (Factor 2), and high levels of immigration and crowding (Factor 3). Cluster 3 is distinguished by the high proportion of immigration and crowding (Factor 3), moderate levels of residential instability factors (Factor 2) and lowest levels of economic and education disadvantage (Factor 1). Cluster 4 is distinguished by all SES factors being low.

Figure 3

Patterns of SES factor scores across SES clusters. (A) All patient pattern, (B) male patient pattern, (C) female patient pattern.

Table 2

Characterization of Socioeconomic Variables by Clusters

	Cluster1 Mean (SD)	Cluster2 Mean (SD)	Cluster3 Mean (SD)	Cluster4 Mean (SD)
Economic and educational disadvantage factor
% < High School	36.21 (9.39)	24.11 (6.89)	43.97 (7.06)	30.36 (9.79)
% At least Bachelors	10.03 (3.07)	20.13 (4.35)	11.66 (2.50)	9.89 (3.49)
% Families below poverty	9.84 (3.41)	15.68 (3.78)	6.70 (1.64)	8.27 (3.00)
% Persons below poverty	14.32 (4.23)	19.95 (4.34)	10.00 (2.08)	11.52 (3.66)
% Persons <150% of poverty	22.74 (5.89)	31.37 (5.14)	16.45 (3.05)	19.20 (5.74)
% Persons <200% of poverty	31.28 (7.12)	41.64 (5.52)	23.20 (3.89)	27.29 (7.44)
% Unemployed	4.06 (1.81)	10.10 (3.84)	6.42 (1.70)	3.68 (1.60)
Median household income	6311.64 (1259.23)	5252.59 (998.44)	8888.98 (1079.95)	6441.88 (1380.38)
Median family income	7774.62 (1504.68)	6134.01 (932.89)	10,519.18 (1416.86)	7996.28 (1677.14)
Immigration + Crowding factor
% <9th Grade	6.47 (1.58)	9.18 (2.01)	5.96 (0.95)	6.17 (1.67)
% Crowding	3.21 (1.69)	7.20 (4.00)	6.24 (2.37)	1.86 (1.00)
% Foreign born	12.93 (6.77)	20.82 (12.10)	28.71 (6.73)	8.86 (5.50)
% Language isolation	3.70 (1.85)	8.11 (4.95)	8.63 (2.26)	2.81 (1.98)
% Moved to US (5+ yrs)	0.87 (0.46)	0.56 (0.22)	1.15 (0.36)	0.45 (0.26)
% Moved to US (1+ yrs)	0.88 (0.46)	0.56 (0.22)	1.16 (0.36)	0.45 (0.27)
Residential instability factor
Age 5+ years
% Moved within county	10.41 (1.51)	8.45 (1.62)	6.96 (1.27)	6.44 (1.54)
% Moved across state	3.16 (1.17)	1.17 (0.45)	1.88 (0.83)	1.83 (0.78)
% No moving	82.12 (2.56)	87.38 (2.08)	87.33 (1.65)	88.09 (2.09)
Age 1+ years
% Moved within county	10.59 (1.49)	8.70 (1.67)	7.09 (1.30)	6.62 (1.58)
% Moved across state	3.19 (1.19)	1.18 (0.47)	1.90 (0.86)	1.87 (0.79)
% No moving	81.89 (2.52)	87.11 (2.17)	87.17 (1.62)	87.82 (2.12)

Notes: All SES variables with percentage of persons living within a county with that characteristics calculated by Census American Community Survey (ACS) 2013–2017; median family income and median household income are in dollars; ANOVA p value <0.05 for all variables.

Factor analysis was used to concentrate SES factors in male and female subgroups and the four identical factors were extracted. Figure 3B and C illustrate the patterns of factor scores across the clusters in male and female patients. For male, Cluster 1 is distinguished by all SES factors being low. Cluster 2 has highest immigration and crowding (Factor 3), medium rates of other two factors. Cluster 3 is distinguished by high levels of residential instability (Factor 2), factors related to low levels of other two factors. Cluster 4 is distinguished by high levels of economic and education disadvantage (Factor 1) and low levels of other two factors. For female patients, Cluster 1 is distinguished by highest economic and education disadvantage factors (Factor 1), factors related to medium levels of residential instability (Factor 2), and lowest levels of immigration and crowding (Factor 3). Cluster 2 is distinguished by high economic and education disadvantage factors (Factor 1) and immigration and crowding (Factor 3), low rate of residential instability (Factor 2). Cluster 3 is distinguished by highest levels of residential instability (Factor 2), medium immigration and crowding levels (Factor 3) and low economic and education disadvantage (Factor 2). Cluster 4 has low levels of all factors. present the statistics of area-based SES factors under the different clusters among cases in different gender.

Identifying the Suicide Risk Factors and Groups

The suicide risk associated with each SES factor and cluster was analyzed using the Cox proportional-hazards model. Factor 1 and Factor 3 were adverse factors for the suicide risk, while Factor 2 was a protective factor for suicide (after adjusting for age, race, sex and insurance status) (Figure 4A). Figure 4B shows the highest risk cluster: patients in areas with high economic and education disadvantage, high immigration and crowding levels, low level of residential instability were identified as having higher suicide risk (Cluster 3) (OR=1.13, 95% CI=1.11–1.45) (after adjusting for age, race, sex and insurance status).

Figure 4

Adjusted hazard risk of primary solid tumor associated with socioeconomic factors (A) and clusters (B).

In male and female patient subgroup, Factor 1 and Factor 3 were adverse factors for the suicide risk, while Factor 2 was a protective factor for suicide (). This result was similar to whole patient analysis. Male patients in areas with medium economic and education disadvantage, medium residential instability, and high immigration and crowding levels were identified as having a higher suicide risk (OR=1.14, 95% CI=1.12–1.16) (Figure 5A). For female patients, living in areas with medium levels of economic and education disadvantage, high immigration and crowding levels and low residential instability were shown a higher suicide risk (OR=1.21, 95% CI=1.18–1.24) (Figure 5B).

Figure 5

Adjusted hazard risk of primary solid tumor associated with socioeconomic clusters for different gender. (A) Adjusted hazard risk for male patients, (B) adjusted hazard risk for female patients.

Discussion

Suicide is a complex phenomenon that has significant sociological and psychiatric implications. This study focused on the risk of suicide among 6.99 million patients with a primary solid tumor, and analyzed the impacts of clinical and socioeconomic features. Although the occurrence of suicide is difficult to predict, our study indicated that patients who are older, male, white, not married, high levels of cancer severity, have not received active treatment (especially surgery), and have no insurance have a higher risk of suicide.

Our findings for patients with a primary solid tumor appear to be consistent with reports related to other types of cancer. Previous studies have shown that older cancer patients and members of the general population are more likely to commit suicide,10,22,23 which is consistent with the present findings. Older patients usually have a greater disease burden and social psychological pressures. Interestingly, younger and middle-aged patients with female genital system cancer (breast cancer, ovary cancer and cervix uteri) were found to have higher suicide risk, and the age difference of suicide risk among breast cancer patients was statistically significant. These findings were consistent with other studies.24,25 Studies reported that young women who developed invasive breast cancer experienced greater absolute and relative functional losses in physical role function, bodily pain, social function, and mental health, compared with middle-aged or elderly women with incident breast cancer.26 However, few studies further explained the pathological and psychological effect of the age difference for suicide risk in patients with ovary and cervix uteri cancer. This is possibly due to younger women suffered more from psychological burden losing sexual function and hormonal disorder from cancer and hormonotherapy. However, more studies are needed to explain these findings.

Being male was a risk factor for suicide in our study, which is consistent with the findings of other studies involving cancer patients.27 This is partially attributable to males being more likely to be aggressive and alcoholics, which is involved with mental disorder. The present study also corroborated the previous finding that race has a significant impact on suicide. The Centers for Disease Control and Prevention reported that the suicide rate in the US is lower in black people than in other races.28 The risk of dying from suicide was reportedly more than twice as high for white than for black patients with non-small-cell lung carcinoma.29 Our study similarly found a lower suicide risk in non-white patients for most cancer sites, which might be explained by their religious beliefs, family support, and culture of rejecting suicide.29

Providing social and medical support to patients with cancer plays a significant role in suicide prevention. We found that divorced or widow patients had a higher suicide risk, which might be due to them receiving less social support.30 Moreover, patients who had never married had a lower suicide risk than divorced or widow patients, which suggests that the latter patients tend to suffer more from a depressive mood.

The suicide risk was closely associated with the illness severity. In our study, patients with poorly differentiated and distant metastasis had higher suicide risk, which is due to their psychological pressure, fear of death and the pain of illness. Determining the optimal treatment options in cancer patients is complicated by treatments being affected by the illness severity. Few studies have systematically investigated the suicide risk in patients who have undergone chemotherapy and radiation therapy. Our study found that radiation and chemotherapy decreased suicide risk for many cancer sites. Nevertheless, cancer-related surgery was found to be a protective factor for almost all the cancer sites, which is consistent with previous findings.13,31–33 However, cognitive impairments of chemotherapy regimens were not distinguished in SEER database which may also induce suicidal ideation.34 For example, taxane-based chemotherapies confer risk for significant psychological symptoms,35 while some other drugs may only display mild or no symptoms of depression. Furthermore, patients with any type of insurance showed a lower suicide risk in our study, probably because the availability of insurance meant greater treatment support. Also, patients who can afford insurance tend to have a better economic status. Future developments in medical technology will result in treatment methods being more effective and efficient in alleviating the patients’ suffering and improving prognoses.

Our study indicated that the age-adjusted rate for suicide was significantly decreased across almost all the solid tumor patients during the year from 1975 to 2017, which is consistent with other study.36 The improvement of living standard and medical technology may help decrease suicide risk for cancer patients. The enlargement of base number for cancer patients may also explain the decreasing suicide rate, since the development of cancer screening technology increased the diagnosis rate for cancer, especially in early stage. In our study, more than 50% of patients were diagnosed during the year from 2006 to 2016. However, studies also reported that the risk of suicide vs the general population is increasing: 1.9 in patients diagnosed 1973–2002, vs 4.4 for patients in the current work who were diagnosed 1973–2015.13,37 We did not discover the relationship between the suicide happening and the time of diagnosis from the aspect of total solid tumor patients. However, Nicholas G reported that certain cancer patients have higher standardized mortality rate (SMR) from suicide after the first year of diagnosis.13

When analyzing area-based SES factors, we used exploratory factor analysis to reduce the number of variables and empirically disentangle the complex underlying relationships between them, as well as to avoid the collinearity problem during modeling. The main principle underlying reducing the number of variables is based on the presence of strong correlations among similar factors.

The economic and education disadvantage factor (Factor 1), which corresponded to a socioeconomically vulnerable environment (lower education status, poverty, and unemployment), showed a higher suicide risk in cancer patients. Previous studies that investigated cancer patients and the general population support our result.38,39 Low-SES areas are often characterized by relatively few social control mechanisms, which may increase the likelihood of exposure to violence and feeling unsafe.

Previous studies have found lower suicide risks in immigration groups, which has been attributed to specific cultural/social aspects.40 This was not consistent with the effect of immigration/crowding (Factor 3) found in our study, possibly because the beneficial effects of pure immigration may be offset by other immigration-related aspects, since the immigration/crowding factor also included language isolation, crowding, and the lowest education levels. Patients living in areas with high level of residential instability (Factor 2) was a protective factor for suicide risk, which is consistent with the findings of studies of other population subgroups.41,42

Furthermore, cluster analysis was conducted with factor scores on these latent constructs with the aim of identifying distinct patterns of community-level SES indicators. This approach acknowledges that while these SES constructs often overlap, they may not all co-occur within specific communities, and their specific patterns of co-occurrence may indicate a higher suicide risk for certain cancer patients. We found that high-risk patients were clustered with these co-occurring features: high economic and education disadvantage, high immigration and crowding levels, and low residential instability (cluster 2). Interestingly, low residential instability was a protective factor analyzed alone, this was opposite in the result from cluster analysis. In real world, the coincidence of several risk factors presents accurate and true situation relative to single predictive factor analysis. One of the most important conclusions from our results is that interaction effects should be considered when analyzing the suicide risk in any population.

Subgroup analysis shows the similar characteristics of co-occurrence patterns with whole patient clusters. The high-risk pattern was clustered by factors with moderate/high economic and education disadvantage, high immigration/crowding and low residential instability. However, studies reported that different social roles determine the ways that men and women are expected to behave differently43. Women are more likely to suffer from stressors experienced by others,44 so they suffer more living in areas with high immigration/crowding levels. Men are more likely to report financial stress,45 they are highly sensitive to economic and education disadvantage.

This study was subject to several limitations: (1) the inherent limitations of its retrospective design, (2) our inability to obtain SEER data on various potentially important factors, such as anxiety, depression, and pain, which are closely related to suicidal ideation, (3) treatment or medication effects, functional status, pain, disfigurement, and other important quality-of-life variables, are not available in SEER database, (4) depressive-effect of chemotherapy and radiation therapy were not distinguished in SEER database, (5) possible bias in the cause of death recorded in the database due to misclassification, (6) the results not including individuals who attempted suicide, and (7) the socioeconomic factors being based on an area-based evaluation (ie, county-based attributes) rather than the personalized socioeconomic status.

Conclusion

This study found that the risk of suicide death was significantly higher among adult patients with a primary solid tumor who were older, male, white, not received active treatment, had no insurance, high levels of cancer severity and were unmarried. The age-adjusted suicide rate has decreased during more recent years. Factor analysis indicated that more attention should be paid to patients living in areas of economic and education disadvantage, high levels of immigration/crowding and high levels of residential instability. Cluster analysis further helped to identify that a high suicide risk was associated with the concomitant presence of high economic and education disadvantage, high immigration and crowding levels and low residential instability.