Suicide among cancer patients.

Our purpose is to identify cancer patients at highest risk of suicide compared to the general population and other cancer patients. This is a retrospective, population-based study using nationally representative data from the Surveillance, Epidemiology, and End Results program, 1973-2014. Among 8,651,569 cancer patients, 13,311 committed suicide; the rate of suicide was 28.58/ 100,000-person years, and the standardized mortality ratio (SMR) of suicide was 4.44 (95% CI, 4.33, 4.55). The predominant patients who committed suicide were male (83%) and white (92%). Cancers of the lung, head and neck, testes, bladder, and Hodgkin lymphoma had the highest SMRs ( > 5-10) through the follow up period. Elderly, white, unmarried males with localized disease are at highest risk vs other cancer patients. Among those diagnosed at < 50 years of age, the plurality of suicides is from hematologic and testicular tumors; if > 50, from prostate, lung, and colorectal cancer patients.


Dear Editors,
Enclosed, please find our revised manuscript titled, "Suicide among cancer patients," for consideration in Nature Communications. In the manuscript, we have addressed all of the points brought up by our reviewers. In this letter, we explain how we address the reviewers' concerns.
Additionally, we have reviewed and completed the journal manuscript checklist, the reporting summary, and the editorial policy checklist. We provide these files with the upload.

Referees' Comments:
Reviewer #1 (Remarks to the Author): This well-written and succinct manuscript described rates of suicide among cancer patients. The manuscript is characterized by several strengths, including an important topic, large sample size, and appropriate statistical analyses. The manuscript adds new knowledge to existing literature on this topic.

AUTHOR RESPONSE:
Thank you for your comments.
The manuscript could be strengthened by addressing two points. First, SMR is significantly higher among recently diagnosed patients (i.e., 2011-2014) compared to patients diagnosed before 2011 (Table 1). Nevertheless, the odds ratio for suicide among more recently diagnosed patients is significantly lower than patients diagnosed before 2011 (Table 2). It would be helpful if the discussion could more directly explain these seemingly discrepant results. AUTHOR RESPONSE: Thank you for your valuable suggestions. In the discussion section, we have made the following changes: In the discussion regarding Objective 1, which uses SMRs, we state: The SMRs are significantly higher among recently diagnosed patients (i.e., 2011-2014) compared to patients diagnosed before 2011 (Table 1). Patients diagnosed in more recent years have a shorter follow up time (i.e. until 2017) compared to those diagnosed in the 1970s-2000s. Since the SMRs are generally highest in the first few years after diagnosis vs > 5-10 years after diagnosis (per Figure 1), the SMRs for the most recent patients are skewed and are higher than patients diagnosed in prior years. Notably, since SMRs are a measure to the standardized population (the general US population in this case), SMRs should not be compared to each other, and the SMRs from Objective 1 should not be compared to the ORs in Objective 2, described below.
In the discussion regarding Objective 2, which uses ORs, we state: The OR of suicide of patients diagnosed in more recent years is lower than that diagnosed in previous years (e.g. 2011-2014 vs 1973-1980 in Table 2), suggesting that patients diagnosed in more recent years are less likely to commit suicide than patients diagnosed in previous years. This finding is likely secondary to the evolving characteristics of cancer patients in the USA; with a decrease in smoking rates (highest among elderly white males), there is a decrease in rates of lung cancer and human papilloma virus (HPV)-negative head and neck cancers, which have also historically been cancers of elderly white men. In contrast, with the advent of screening mammography and prostate specific antigen (PSA) testing in the 1990s, there has been a surge in the diagnoses of low-risk breast and prostate cancers. ORs compare the odds of suicide of the group of one group of cancer patients vs a reference group of cancer patients, unlike the SMRs in Objective 1, which compare relative risk of death vs. the entire US population, as a function of time after diagnosis. Thus, if there is a change in the rate of suicide in subpopulations of patients between two eras, this change will be reflected in the ORs, but not necessarily in the SMRs.
Second, it would be helpful to place results in the context of national requirements (or lack thereof) of screening for suicidality in cancer patients. AUTHOR RESPONSE: Thank you for your valuable comments. We discuss the available national resources for screening for suicidality in cancer patients. We have modified our recommendations for screening per the comments of Reviewer #2. We now state: The results of the current work suggest that suicide-prevention strategies may be aimed at those >50 years of age patients with cancer of the prostate, lung, colorectum, and bladder; as well as patients with leukemias, lymphomas, and germ cell tumors. We recommend that providers follow the evolving guidelines for monitoring distress and suicide prevention from the American College of Surgeons Committee on Cancer, the American Society of Clinical Oncology, the National Comprehensive Cancer Network, and Action Alliance for Suicide Prevention Reviewer #2 (Remarks to the Author): The manuscript examined the suicide rate among patients with cancer from SEER database, and found that suicide rate was higher in patients with cancer compared with that in the general population. Some factors associated with suicide among patients with cancer were also studied. The topic in this manuscript were investigated previously in both US and other countries, and the findings were similar to previous reports. See the most recent pulication using the same SEER database as the authors used in this manuscript (Kumar V, et al. 2017 American J Psychiatry). The manuscript, however, is a bit too compact (only ~2000 words) and lacks details of the rational, methods, results, discussion of the implications, etc.
For instance, in the Abstract, the authors stated that "The exposure was suicide, ..." , which is not right. Suicide is the main outcome, not the exposure. AUTHOR RESPONSE: Thank you for your valuable comment. We have corrected this typo.
The rational of using logistic regression model to examine the risk factors of suicide should also be clarified. The time from diagnosis to suicide could be more interesting. AUTHOR RESPONSE: Thank you for your valuable comment. We applied the logistic regression analyses to report odds ratios, and we did not perform a time-to-event analysis (such as a Cox regression model). In the manuscript, we now state: The choice of modeling depends on the study objectives and interests. In objective 2, we explored the relation between risk factors for patients committing suicide vs not committing suicide. We chose logistic regression because we were interested in whether the patients committed suicide or not, eventually, without a time factor until suicide. We did not perform a time-to-event analysis, which aims to explore the relation between risk factors and time to suicide, which was not our focus in the current work. Moreover, to fit a logistic regression model for a time-to-event analysis, we would need more events. In the current data set, we have relatively few events, 13311/8651569 (~0.15%), and the remaining patients are censored (we were not able to observe their events), which would decrease the reliability of a logistic regression model. We value your statistical insight, and we have clarified the methods section accordingly.
The conclusions state "We recommend clinics screen older patients with cancer of .... Patients with .... should be screened indefinitely" lack empirical support or specific practical strategies from this study. AUTHOR RESPONSE: Thank you for your valuable comment. We have revised this statement. We instead now summarize our findings and reference the evolving recommendations from national organizations. We now state: The results of the current work suggest that suicide-prevention strategies may be aimed at those >50 years of age patients with cancer of the prostate, lung, colorectum, and bladder; as well as patients with leukemias, lymphomas, and germ cell tumors. We recommend that providers follow the evolving guidelines for monitoring distress and suicide prevention from the American College of Surgeons Committee on Cancer, the American Society of Clinical Oncology, the National Comprehensive Cancer Network, and Action Alliance for Suicide Prevention.
In this study, suicide was defined according to the death certificate. Misclassification of suicide is thus possible and should be discussed. Suicidal ideation should definitely be analzyed as well if it is available in the SEER database. The observed association between cancer and suicide could be confounded by psychiatric disorders (or the like), the authors did not comment on confouding, confounding control, etc. in the study design or analysis. AUTHOR RESPONSE: Thank you for your valuable comments. We have revised the methods and discussion sections as follows: In the discussion we state: Further, there is a risk of bias and misclassification of suicide in the SEER. For example, in a review of cases of non-Hodgkin lymphomas, agreement in the subclassification of histologies between the expert review and the SEER registry record varied from 5% to 100%. An investigation of the California Cancer Registry, which contributes to the national SEER data revealed registry sensitivity of receipt of radiation therapy of only 72%. As of 2018, there has been limited research published regarding the misclassification of cause of death in the SEER database. Thus, we are unable to characterize misclassification of suicide in the current work.
Nonetheless, for suicide, there is likely little discrepancy in the cause of death, as compared to a cause of death like heart disease, which may be cause by the cancer treatment, underlying heart disease, or a combination of both. We agree with Sun and Trinh, in their assessment of the SEER database, that although there may be some errors in large registries, the errors are likely less frequent than those in hospital based databases and big data will continue to remain an integral part of hypothesis-generating exploratory analyses in medical research.
In the methods we state: Notably, SEER does not code comorbidities or diagnoses associated with suicide, including suicidal ideation, previous suicide attempts, or use of anti-depressive medications. The observed associations between cancer and suicide may be confounded by psychiatric disorders and the use of medications, but we are unable to control for these factors in the current work. These are limitations of the analysis and limit the interpretability of the results.