1 INTRODUCTION

Cancer is associated with an increased risk for completed suicide.1-3 Zaorsky et al.‘s4 recent register study confirmed a 4.4-fold increased rate of death by suicide drawing on representative data from over 8 million cancer patients. While the 0.154% rate of deaths by suicide shows that completed suicide is nonetheless a rare event in patients with cancer, this is not the case for suicidal ideation. The systematic review of Kolva et al.5 indicates widely ranging suicidal ideation prevalences between 0.7% and up to 46.3% among individuals with cancer. Although the one-item assessments used in many included studies do not respect the heterogeneity of suicide-related thoughts and omit suicidal behaviors, the state of research enheartens further detailed investigation.

Manifestations of thoughts of death and suicidality1 in cancer may range on a continuum from death acceptance over death wishes, with or without a desire to hasten death, to actual suicide plans.6, 7 Especially death wishes in advanced cancer patients can have different meanings and degrees of suicidal intent. Qualitative interview studies have subtyped their meanings and functions.8, 9 The wish to die may for example express a sense of letting go, a hypothetical exit plan to assure control in case the situation is perceived as unmanageable, or current intolerable despair and suffering.10 The nature of the cancer stressor and its association with existential fears of uncertainty, functional impairments, loss of autonomy, uncontrollable suffering, burdening others, or being left alone,11 can make it difficult to distinguish suicidal ideation and behavior from other adjustment reactions in patients with cancer. Much of the distress underlying suicidal ideation and plans (e.g. requests for physician-assisted suicide) is directed towards a hypothetical future state where suffering may become intolerable.6 The complexity of suicide-related phenomena in cancer raises interest in their typology.

In non-cancer samples, latent class analyses have identified subtypes of self-injurious thoughts and behaviors.12 The heterogeneous suicide indicators used interfere with cross-study comparisons, yet results often showed a major large class without suicidal ideation and behavior, and smaller classes that were distinguished by suicidal ideation and plans alone vs. individuals who had attempted suicide.13, 14

The known association between mental disorders and an increased risk for suicidal ideation has been confirmed in cancer samples.15-17 Other risk factors for suicidal ideation in cancer patients include a higher physical symptom burden, living alone and advanced cancer (stage IV),5 while findings have been inconsistent for sex and age. This pattern deviates from the risk factors for completed suicide in cancer: younger age, male sex, white race, and early stage cancer.4

There has, to our knowledge, been no attempt so far to classify suicidality profiles among patients with cancer using quantitative methods. The main goal of the present study is to (1) identify subtypes characterized by specific features of suicidality in a mixed sample of patients with cancer that is representative for all tumor entities in Germany. Such subtypes can be determined by latent class analysis. We applied this method to explore whether the patterns of thoughts of death and suicidality endorsed by individuals with cancer form distinct subgroups.2 We further aimed to investigate (2a) how the resulting classes were characterized by demographic and disease-related patient variables, and (2b) their association with mental disorders.

2 METHODS 2.1 Participants and procedures

We analyzed data from an epidemiological cross-sectional study.19, 20 Patients were recruited while receiving oncological treatments from oncological inpatient clinics at acute care hospitals, specialized outpatient cancer care facilities, and cancer rehabilitation centers across Germany. Exclusion criteria were age younger than 18 or older than 75 years, severe cognitive or physical impairment, and language barrier. All patients provided written informed consent. Participants were screened with the Patient-Health-Questionnaire-9 (PHQ-9). Patients with sum scores ≥9 were further assessed by a standardized diagnostic interview for mental disorders. Patients with sum scores <9 were randomly assigned to the interview. All statistical analyses were weighted to control for the oversampling of patients with PHQ-9 scores ≥9. All patients completed a set of self-report questionnaires.20 The research ethics committees of the local medical association in each study center approved this study (reference numbers 2768, 61/09, 244/07, S-228/2007, 50155039, 107/07, 200–2007).

Of 5889 eligible patients, 4020 (68%) agreed to participate in the study. As reported elsewhere,19 non-participants were younger, but did not differ in sex. 1202 patients (30%) screened ≥9 on the PHQ-9 and were assigned to the standardized interview (CIDI-O), of which 903 (75%) completed the interview. 2818 patients (70%) screened <9 on the PHQ-9, of which 1508 (54%) were randomly assigned to the CIDI-O. Among these, 1238 (82%) completed the interview. In total, 2141 patients completed the CIDI-O and were analyzed in this study.

A detailed overview of demographic and disease-related sample characteristics as well as descriptive information about study variables is given in the supplement, table S1.

2.2 Measures

Demographic data were collected by a standardized questionnaire. Disease-related characteristics were obtained from medical charts. Incurable cancer was defined by stage IV diagnosis for solid tumors and evaluation of curative vs. palliative treatment intent by the attending physician in hematological cancers.

We used the standardized computer-assisted Composite International Diagnostic Interview for Oncology (CIDI-O) to assess the 4-weeks prevalence of thoughts of death, death wishes, suicidal ideation and suicidal behavior, as well as mental disorders. Assessment included the following questions as part of the depression section: In the past 4 weeks, did you have a period of two weeks or more, when you 1) thought a lot about death, 2) had the wish to die, 3) felt so low you thought about committing suicide, 4) made a suicide plan? Question 5) asked individuals whether they had ever attempted suicide. The interview further assessed the 4-weeks prevalence of the following mental disorders: disorders resulting from general medical condition (organic mood disorders and organic anxiety disorder), substance use disorders (dependence and abuse of alcohol), mood disorders (unipolar and bipolar), anxiety disorders (including acute and post-traumatic stress disorder; without obsessive compulsive disorder), somatoform disorders, and eating disorders.21 The CIDI-Oncology further enables diagnosis of adjustment disorders in response to specific cancer-related stressors.22 In accordance with DSM-IV criteria, an adjustment disorder was diagnosed where distress was problematic, out of proportion to the clinical setting, causing impairment, no other axis I disorder was present, and symptoms did not persist for longer than 6 months.

We used the physical problem list of the Distress Thermometer (DT) to assess the presence of 21 common physical symptoms in cancer patients.23 The total physical symptom count (possible range from 0 to 21) was used as a measure of physical symptom burden.

2.3 Statistical methods

We calculated descriptive statistics including means, standard deviations, and frequencies for demographic and disease-related characteristics.

We used latent class analyses to define subtypes of suicidality as implemented in R package depmixS4.24 The latent class model was calculated from the five suicidality items assessed by the structured CIDI-interview as indicator variables. We determined the solution with an optimum number of classes by examining the following indices: loglikelihood value, Akaike's information criterion, Bayesian information criterion, likelihood ratio value, and entropy value.18

We then determined the extent to which mental disorders as well as demographic and disease-related characteristics differed in frequency across classes using multinomial regression analyses. Odds ratios (ORs) quantify the odds for having a mental disorder among members of the respective class compared to members of the reference class. We first conducted a bivariate multinomial regression analyses, where the association of each predictor with class membership was tested separately. We then performed a multivariate regression analysis with simultaneous entry of all demographic, disease, and disorder-related patient characteristics as predictors of class membership. This analysis determines the contribution of each predictor under control of the remaining variables in the model.

Analyses were weighted to control for the oversampling of patients with PHQ-9 depression scores ≥9. Missing values occurred in 0.1% of responses for the self-report measure of physical symptoms and were person mean-imputed. We used R version 3.6.125 for all analyses.

3 RESULTS 3.1 Subtypes of thoughts of death and suicidality: latent class analysis

A three-class solution yielded the best fit (see supplement Table S2) with lowest AIC and BIC values. The loglikelihood and likelihood ratio values were similar for three-, four-, and five-class solutions, but improved markedly from the two-class to the three-class solution. The entropy value of 0.82 suggests adequate latent class separation.

Table 1 and Figure 1 show the pattern of item endorsement probabilities for thoughts of death, death wishes, suicidal ideation and behaviors for all three classes. Of the total sample, 1905 (89%) individuals were assigned to class 1, 148 (6.9%) were assigned to class 2, and 88 (4.1%) were assigned to class 3. Class 1 was characterized by zero probability for all items and hence labeled “No Suicidality”. In class 2, labeled “Thoughts of Death”, the probability for frequent thoughts of death in general or one's own death was 0.79, and 0.18 for a wish to die over the past 4 weeks. Class 3, “Suicidal Ideation”, was characterized by a high probability of thoughts of suicide (0.97). Most members in class 3 also endorsed a wish to die (0.77) and thoughts of death (0.80). There was a moderate probability for suicide plans (0.36) and suicide attempts (0.14) in class 3.

TABLE 1. Probabilities for endorsement of suicide items across latent classes Class 1 No Suicidality (n = 1905) 89.0% Class 2 Thoughts of Death (n = 148) 6.9% Class 3 Suicidal Ideation (n = 88) 4.1% 1) Frequent thoughts of death 0.00 0.79 0.80 2) Wish to die 0.00 0.18 0.77 3) Suicidal ideation 0.00 0.00 0.97 4) Suicide plans 0.00 0.00 0.36 5) Suicide attempt 0.00 0.00 0.14 Note: Structured clinical interview questions: In the past 4 weeks, did you have a period of two weeks or more, when you 1) thought a lot about death, 2) had the wish to die, 3) felt so low you thought about committing suicide, 4) made a suicide plan? 5) Have you ever attempted suicide?

Probability profile of thoughts of death, death wishes, suicidal ideation and behavior across the latent classes Structured clinical interview questions: In the past 4 weeks, did you have a period of two weeks or more, when you 1) thought a lot about death, 2) had the wish to die, 3) felt so low you thought about committing suicide, 4) made a suicide plan? 5) Have you ever attempted suicide?

Association of thoughts of death and suicidality subtypes with sociodemographic and disease-related characteristics.

Table 2 presents differences in sociodemographic and disease-related characteristics across the three latent classes. Odds ratios indicate whether members of class 2 and class 3 were more or less likely to carry the respective characteristic compared to the reference group class 1.

TABLE 2. Frequency of demographic and disease-related characteristics across latent classes, bivariate associations Class 1 No Suicidality (n = 1905) 89.0% Class 2 Thoughts of Death (n = 148) 6.9% Class 3 Suicidal Ideation (n = 88) 4.1% N (% of class 1) N (% of class 2) ORa 95% CI N (% of class 3) ORa 95% CI Age < 40 100 (5.3) -a 14 (9.4) 1.87** 1.17 to 2.98 6 (6.5) 1.25 0.62 to 2.50 Age ≥ 70 327 (17.2) - 16 (10.7) 0.58* 0.38 to 0.89 9 (10.1) 0.54* 0.31 to 0.95 Female sex 906 (47.5) - 86 (57.9) 1.52** 1.16 to 1.99 67 (76.8) 3.66*** 2.45 to 5.46 Married/cohabiting 1562 (82.0) - 109 (73.7) 0.62** 0.44 to 0.86 53 (60.7) 0.34*** 0.23 to 0.50 High school/university education 598 (31.4) - 64 (43.3) 1.67*** 1.28 to 2.19 33 (37.7) 1.32 0.93 to 1.88 Incurable cancer 430 (22.5) - 42 (28.8) 1.39* 1.10 to 1.87 11 (12.3) 0.48** 0.29 to 0.81 Recurrent cancer 376 (19.7) - 38 (25.3) 1.38* 1.01 to 1.88 18 (21.0) 1.08 0.71 to 1.64 Inpatient treatmentb 849 (44.5) - 58 (39.5) 0.81 0.62 to 1.07 29 (32.6) 0.60** 0.42 to 0.87 <12 Months since first diagnosis 1224 (64.2) - 73 (49.2) 0.54*** 0.40 to 0.73 46 (52.6) 0.62* 0.43 to 0.90 Physical symptom count, mean (SD)c 5.21 (3.63) - 7.63 (3.87) 1.17***d 1.13 to 1.22 7.80 (3.94) 1.19***d 1.13 to 1.24 Breast cancer 369 (19.3) - 38 (26.2) 1.48* 1.09 to 2.01 30 (34.1) 2.15*** 1.50 to 3.10 Prostate cancer 340 (17.9) - 14 (9.0) 0.46*** 0.29 to 0.72 5 (5.8) 0.28*** 0.14 to 0.58 Colorectal cancer 253 (13.3) - 22 (14.6) 1.11 0.76 to 1.63 5 (10.1) 0.74 0.42 to 1.29 Lung cancer 169 (8.9) - 12 (8.2) 0.91 0.56 to 1.48 7 (8.0) 0.89 0.47 to 1.67 Gynecologic cancer 152 (8.0) - 13 (8.6) 1.08 0.67 to 1.74 13 (14.5) 1.95** 1.20 to 3.20 Hematologic cancer 139 (7.3) - 14 (9.4) 1.33 0.84 to 2.10 9 (10.1) 1.44 0.81 to 2.54 Stomach/esophageal cancere 77 (4.0) - 5 (3.4) - - 2 (2.2) - - Kidney/urinary tract cancere 73 (3.8) - 3 (2.1) - - 1 (0.7) - - Head and neck cancere 57 (3.0) - 6 (3.9) - - 3 (2.9) - - Bladder cancere 44 (2.3) - 4 (2.6) - - 2 (2.9) - - Pancreatic cancere 46 (2.4) - 3 (1.7) - - 1 (1.4) - - Malignant melanomae 34 (1.8) - 3 (2.1) - - 0 (0.0) - - Abbreviations: CI, confidence interval; OR, odds ratio. a Class 1 is the reference group for odds ratios, interpretation: the OR of 1.87 for example indicates that the odds of being in class 2, compared to class 1, is 1.87 times higher for patients younger than 40 years than for those 40 years or older. b Acute oncologic inpatient treatment (excluding rehabilitation inpatients). c Possible range: 0-21. d Interpretation for OR: with every additional physical symptom, the odds of being in class 2, compared to class 1, increase by 1.17. e No ORs calculated due to low n in classes 2 and 3. *p < 0.05, **p < 0.01, ***p < 0.001.

Members of class 2 (thoughts of death) as well as class 3 (suicidal ideation) were significantly younger, more often female and less often married compared to class 1 (no suicidality). Classes 2 and 3 were also characterized by longer time since diagnosis, and a higher number of physical symptoms. Education was better in class 2.

Illness prognosis however showed a notably divergent pattern across the three classes. While incurable disease was significantly more frequent in class 2 (OR = 1.39, 95% CI 1.10 to 1.87), it was significantly less frequent in class 3 (OR = 0.48, 95% CI 0.29 to 0.81). Tumor entity distribution showed relatively low differences across classes, except for those attributable to sex (breast cancer, prostate cancer, and gynecological cancer).

Table 3 shows case numbers and prevalence rates of mental disorders across the three classes. Members of both class 2 (thoughts of death) and class 3 (suicidal ideation) were significantly more likely to suffer from a mental disorder compared to members of class 1 (no suicidality), with odds being 4.2 (adjustment disorders, class 3) to 10.2 (mood disorders, class 2) times higher (except for somatoform disorders). Interestingly, comparison of class 2 and three indicates similar frequencies of mental disorders, and no significant differences between odds ratios for any mental disorder as well as specific mental disorders as indicated by overlapping confidence intervals.

TABLE 3. Frequency of mental disorders across latent classes, bivariate associations Class 1 No Suicidality (n = 1905) 89.0% Class 2 Thoughts of Death (n = 148) 6.9% Class 3 Suicidal Ideation (n = 88) 4.1% N (% of class 1) N (% of class 2) ORa 95% CI N (% of class 3) ORa 95% CI Any mental disorderb 439 (23.0)

-a

118 (79.8) 13.30*** 9.49 to 18.40 68 (78.3) 12.00*** 7.95 to 18.20 Two or more mental disorders 255 (13.4) - 71 (48.1) 5.98*** 4.53 to 7.90 43 (48.6) 6.10*** 4.30 to 8.66 Any anxiety disorderc 168 (8.8) - 46 (31.3) 4.72*** 3.48 to 6.40 31 (35.5) 5.70*** 3.93 to 8.26 Anxiety disorder without other comorbid disorder 108 (5.7) - 18 (12.4) 2.37*** 1.56 to 3.61 17 (18.8) 3.88*** 2.26 to 6.10 Adjustment disorder 163 (8.5) - 49 (33.5) 5.39*** 3.99 to 7.28 25 (28.3) 4.22*** 2.85 to 6.24 Any mood disorderd 74 (3.9) - 43 (29.2) 10.20*** 7.27 to 14.30 22 (24.6) 8.08*** 5.25 to 12.40 Major depression 32 (1.7) - 14 (9.4) 6.17*** 3.67 to 10.40 9 (10.9) 7.21*** 3.94 to 13.20 Any somatoform/conversion disorder/syndrome 91 (4.8) - 13 (8.6) 1.87* 1.15 to 3.04 9 (10.1) 2.25** 1.26 to 4.00 Any disorder due to general medical condition 30 (1.6) - 12 (7.7) 5.28*** 3.01 to 9.24 7 (8.7) 6.00*** 3.11 to 11.60 Abbreviations: CI, confidence interval; OR, odds ratio. a Class 1 is the reference group for odds ratios, interpretation: the OR of 13.30 for example indicates that the odds of being in class 2, compared to class 1, is 13.30 times higher for patients with any mental disorder than for those without a disorder. b Alcohol abuse included but not analyzed separately due to rare prevalence of 0.3% in total sample. The prevalence of eating disorders was 0.0%19. c Assessment of agoraphobia, panic disorder, social phobia, specific phobias (animal, natural environment, blood-injection-injury, or situational type), generalized anxiety disorder, posttraumatic stress disorder, single panic attack, or anxiety disorder not otherwise specified. d Assessment of major depressive disorder, dysthymic disorder, bipolar I disorder, bipolar II disorder, single manic episode, or hypomania. *p < 0.05, **p < 0.01, ***p < 0.001.

The supplement table S3 shows the results of a multivariate regression model with simultaneous entry of all demographic, disease, and disorder-related patient characteristics as predictors of class membership. In the multivariate model, the ORs for mental disorders were similarly high in class 2 and somewhat lower but still significant in class 3 compared to the bivariate model. Demographic and medical characteristics including female sex, living alone, and non-incurable cancer remained significant predictors of class 3, but not of class 2 in the multivariate model. A higher physical symptom burden predicted membership of both classes 2 and three in the multivariate model.

4 DISCUSSION

T