RISK OF SUICIDE MORTALITY AMONG CANCER PATIENTS: A META-ANALYSIS OF OBSERVATIONAL STUDIES Raffaella Calati (a), Elizabeth Mostofsky (b), Valentina Di Mattei (c), Philippe Courtet (a) raffaella.calati@gmail.com (a) INSERM U1061, La Colombière Hospital, University of Montpellier, Montpellier, France; University of Montpellier, Montpellier, France; Department of Psychiatric Emergency & Post-Acute Care, Lapeyronie Hospital, CHU Montpellier, Montpellier, France; FondaMental Foundation, Créteil, France (b) Department of Epidemiology, Harvard School of Public Health, Boston, MA, United States; Cardiovascular Epidemiology Research Unit, Department of Medicine, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, MA, United States (c) Vita-Salute San Raffaele University, Milan, Italy; Clinical and Health Psychology Unit, IRCCS San Raffaele Hospital, Milan, Italy ABSTRACT: Introduction: Suicide rates among patients with cancer are higher than ones in the general population. Objective: This meta-analysis aims to estimate the suicide risk in patients with cancer. Methods: We searched MEDLINE, PsycINFO, and the Cochrane Library to identify articles published before September 2016, examining the association between suicide [death (SD), attempt (SA), ideation (SI)] and any form of diagnosed cancer. Results: We initially identified 4,882 records and after unsuitable studies were removed, our search yielded 104 publications of which 15 were included in the meta-analyses. Patients with cancer had higher risk of SD (seven studies, 247,978 participants; odds ratio [OR]=1.55, 95% CI=1.23-1.96, p=0.0002) compared with those without cancer (among case-control studies focused on SD versus living controls). Among studies focused on SD versus other deaths, patients with cancer had higher risk of SD (two studies, 23,839 participants; OR=1.53, 95% CI=1.03-2.27, p=0.03). No difference has been detected for risk of SA (four studies, 8,147,762 participants) and for SI (three studies, 42,700 participants). The majority of the included studies have a high quality at the STROBE statement. Conclusion: The assessment of suicide risk in this population is crucial.
INTRODUCTION Individuals diagnosed with cancer are at elevated risk of suicide. A number of reviews have been published on the link between suicide, in particular suicidal ideation and suicide death, and cancer [1-3].
AIM
RESULTS 15 studies have been included in total (SD: 10; SA: 4; SI: 3). Cancer subjects Study or Subgroup
Events
METHODS
Events
Odds Ratio
Total Weight
Odds Ratio
M-H, Random, 95% CI
M-H, Random, 95% CI
1.1.1 SD versus living controls (all studies)
Strand et al. 2011
3
2762
98
25583
6.1%
0.28 [0.09, 0.89]
22
12669
23631
7847960
10.9%
0.58 [0.38, 0.88]
Fang et al. 2012
786
534154
13284
5539086
12.3%
0.61 [0.57, 0.66]
Webb et al. 2012
30
591
843
17742
11.2%
1.07 [0.74, 1.56]
151
107861
125
107736
11.9%
1.21 [0.95, 1.53]
97
299
2003
8101
11.8%
1.46 [1.14, 1.87]
104
452
498
3149
11.9%
1.59 [1.25, 2.02]
9
17
77
222
7.0%
2.12 [0.79, 5.71]
Jia et al. 2014
14
20
186
380
7.1%
2.43 [0.92, 6.47]
Miller et al. 2008
19
83
109
1325
10.0%
3.31 [1.91, 5.73]
13551284
100.0%
1.18 [0.80, 1.76]
Lu et al. 2013
Mohammadi et al. 2014 Bolton et al. 2015 Voaklander et al. 2008
To investigate the relationship between cancer and suicide, we performed the first meta-analysis on this topic. We assessed different aspects of suicidal thoughts and behaviors [suicidal ideation (SI), suicide attempt (SA) and suicide death (SD)]. Concerning cancer, we considered any type of cancer. Secondarily, we performed a number of sensitivity analyses to account for the hypothesized between-study heterogeneity.
Non cancer subjects
Total
Waern et al. 2002
Subtotal (95% CI) Total events
658908 1235
40854
Heterogeneity: Tau² = 0.33; Chi² = 159.09, df = 9 (P < 0.00001); I² = 94% Test for overall effect: Z = 0.83 (P = 0.41) 1.1.2 SD versus living controls (case-control studies)
Webb et al. 2012
30
591
843
17742
16.1%
1.07 [0.74, 1.56]
151
107861
125
107736
21.4%
1.21 [0.95, 1.53]
97
299
2003
8101
21.0%
1.46 [1.14, 1.87]
104
452
498
3149
21.3%
1.59 [1.25, 2.02]
9
17
77
222
4.6%
2.12 [0.79, 5.71]
Jia et al. 2014
14
20
186
380
4.7%
2.43 [0.92, 6.47]
Miller et al. 2008
19
83
109
1325
10.9%
3.31 [1.91, 5.73]
138655
100.0%
1.55 [1.23, 1.96]
Mohammadi et al. 2014 Bolton et al. 2015
Voaklander et al. 2008 Waern et al. 2002
Subtotal (95% CI)
Total events
109323
424
3841
Heterogeneity: Tau² = 0.05; Chi² = 15.80, df = 6 (P = 0.01); I² = 62%
IN MEMORY OF ELISA BOGNETTI
Test for overall effect: Z = 3.67 (P = 0.0002) 1.1.3 SD versus living controls (cohort studies)
Strand et al. 2011 Lu et al. 2013 Fang et al. 2012
3
2762
98
25583
0.4%
0.28 [0.09, 0.89]
22
12669
23631
7847960
2.9%
0.58 [0.38, 0.88]
786
534154
13284
5539086
96.8%
0.61 [0.57, 0.66]
13412629
100.0%
0.61 [0.57, 0.65]
1.28 [1.06, 1.55]
Subtotal (95% CI) Total events
549585 811
37013
Heterogeneity: Tau² = 0.00; Chi² = 1.81, df = 2 (P = 0.40); I² = 0% Test for overall effect: Z = 13.67 (P < 0.00001)
1.1.4 SD versus other deaths Marshall et al. 1983 Quan et al. 2002
155
569
4854
21504
56.3%
95
155
727
1611
43.7%
1.93 [1.37, 2.70]
23115 100.0%
1.53 [1.03, 2.27]
Subtotal (95% CI) Total events
724 250
5581
Heterogeneity: Tau² = 0.06; Chi² = 4.23, df = 1 (P = 0.04); I² = 76% Test for overall effect: Z = 2.13 (P = 0.03)
0.1
0.2
0.5
SD lower in cancer
1
2
5
10
SD higher in cancer
Figure 1. Forest plot of studies comparing SD rates in individuals with and without cancer.
Secondary results: No difference has been found concerning both SA and SI. The majority of the included studies have a high quality at the STROBE statement.
CONCLUSION A cancer diagnosis constitutes a major stressor that affects the risk of fatal outcomes as suicide death. The period immediately after the cancer diagnosis seems to be the one linked to the higher risk [4]. Shared risk factors and sources of confounding should be accounted in future analyses. Limitations: the number of included studies was restricted; the absence of data with respect to the time since diagnosis and tumor characteristics (e.g., tumor type, stage, etc.) limited the possibility of exploring their effect on the association between cancer diagnosis and suicide.
REFERENCES: 1. Anguiano, L., et al., A literature review of suicide in cancer patients. Cancer Nurs, 2012. 35(4): p. E14-26. 2. Spoletini, I., et al., Suicide and cancer: where do we go from here? Crit Rev Oncol Hematol, 2011. 78(3): p. 206-19. 3. Robson, A., et al., The risk of suicide in cancer patients: a review of the literature. Psychooncology, 2010. 19(12): p. 1250-8. 4. Fang, F., et al., Suicide and cardiovascular death after a cancer diagnosis. N Engl J Med, 2012. 366(14): p. 1310-8.
No conflict of interest.
This meta-analysis was performed according the reporting checklist proposed by the Meta-analysis of Observational Studies in Epidemiology (MOOSE) group. Search strategy and selection criteria We conducted a systematic search of PubMed, PsycINFO and Cochrane Library electronic databases until September 2016. Studies were included if: they were published in a peer-reviewed journal; they were written in English; they reported rates of SI, SA or SD for both participants having received a diagnosis of cancer and participants not affected. Any reported form of cancer was included. Studies were excluded if: they focused on suicidal or cancer patients only; they had no control group. Outcomes The primary outcome was SD rate in participants with cancer versus participants without it. Secondary outcomes were: SA and SI in participants with cancer versus participants without it. Quality assessment and data analysis The methodological quality of studies was assessed according to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement. Data were analyzed with Cochrane Collaboration Review Manager Software (version 5.3) and Comprehensive Meta-analysis (version 2.2). Individual and pooled odds ratios and associated 95% confidence intervals were calculated. The test of heterogeneity was assessed with the Chi2 goodness of fit and I2 statistics. Data were analyzed using random-effects models. A funnel plot was created and the Egger’s test was also used. All p values were two-tailed and statistical significance was set at the 0.05 level.