key: cord-0335145-cyccdj3v authors: Chang, W. H.; Lai, A. G. title: Psychiatric disorders and self-harm across 26 adult cancers: cumulative burden, temporal variation, excess years of life lost and unnatural causes of deaths date: 2021-10-09 journal: nan DOI: 10.1101/2021.10.07.21264703 sha: b4da8d9ec70b8c0effad63281a0401d53288df3f doc_id: 335145 cord_uid: cyccdj3v Background Cancer is a life-altering event causing considerable psychological distress. However, population-representative variations in the total burden of psychiatric episodes across cancer types and treatment modalities have not been examined. We sought to estimate the risk of self-harm after incident psychiatric disorder diagnosis in patients with cancer, and the risk of unnatural deaths after self-harm. Design, Setting, Participants Population-based cohort study with multiphase study designs. Population-based linked patient records in England (1998-2020) from primary care practices, hospitals, cancer registry and death registry were employed. We identified 459,542 individuals age [≥] 18 years with an incident diagnosis of a site-specific cancer of interest. Main outcome measures Using outpatient and inpatient records, we identified patients with five psychiatric disorders of interest: depression, anxiety disorder, schizophrenia, bipolar disorder and personality disorder. Cumulative burden for all psychiatric events was estimated using the mean cumulative count method. We considered 10 cancer treatment regimens, 11 chemotherapy drug classes, deprivation status and 21 non-cancer comorbidities in stratified analyses. Propensity score matching was employed to identify controls who did not have any record of a psychiatric disorder of interest. For each psychiatric disorder category, we fitted a Cox regression model to estimate the risk of self-harm. We also estimated the risk of all-cause mortality and excess years of life lost comparing patients with and without psychiatric disorders. A separate matched cohort was generated to explore the risk of suicide and unnatural deaths following self-harm. Results Depression was the most common psychiatric disorder in patients with cancer, where some of the highest cumulative burdens were observed in patients with testicular cancer (98.05 per 100 individuals [CI: 83.08-127.25]), cervical cancer (78.74 [73.61-90.14]) and Hodgkin lymphoma (69.87 [61.05-69.48]) by age 60. Patients who received chemotherapy, radiotherapy and surgery had the highest cumulative burden of psychiatric disorders, while patients who received radiotherapy alone had the lowest burden. Patients treated with alkylating agent chemotherapeutics had the highest burden of psychiatric disorders while those treated with kinase inhibitors had the lowest burden. Among patients with cancer, 5,683 individuals were identified as having an incident self-harm episode. A previous diagnosis of psychiatric disorder before self-harm was at least twice as prevalent than a subsequent diagnosis of psychiatric disorder where the prevalence ratio was the highest in patients with brain tumours (5.36, CI: 4.57-6.14). Younger individuals were more likely to be diagnosed with mental illness before the first self-harm episode. However, individuals from more deprived regions (2.46, CI: 2.32-2.60) and individuals with [≥]4 pre-existing comorbidities (2.19, CI: 1.92-2.46) were less likely to be diagnosed with mental illness before self-harm. Patients with mental illness had a higher cumulative burden of self-harm events compared with matched controls. All mental illnesses were associated with an increased risk of subsequent self-harm, where the highest risk was observed within 12 months of the mental illness diagnosis. Risks of self-harm during the first year in matched cohorts were as follow: depression (adjusted HR 44.1, CI: 34.0-57.1), anxiety disorder (HR 21.1, CI: 16.4-27.0) and schizophrenia (HR 7.5, CI: 5.0-11.2). Patients with cancer and psychiatric disorder experienced excess years of life lost. Patients who harmed themselves were 6.8 times more likely to die of unnatural causes of death compared with controls within 12 months of self-harm (HR 6.8, CI: 4.3-10.7). The risk of unnatural death after 12 months was markedly lower (HR 2.0, CI: 1.5-2.7). Conclusions This study quantifies the total burden of psychiatric events and self-harm in patients with cancer. The cumulative burden of psychiatric events varies across cancer type, treatment regimen and chemotherapy type. Incident psychiatric disorder diagnoses were significantly associated with increased risk of subsequent self-harm, where risks varied across psychiatric diagnostic categories and follow-up periods. Patients who harm themselves experienced the highest risk of dying from unnatural deaths within the first year of self-harm. We provide an extensive knowledge base to help inform collaborative cancer-psychiatric care initiatives by prioritising patients who are most at risk. Linked electronic health record data sources 137 138 Two primary care linked electronic health record (EHR) databases, GOLD and Aurum, from We identified incident primary site-specific cancer cases in individuals aged 18 years or older. Incident cancers were defined as the first diagnosis of cancer occurring at least 1 year after 156 the start of the up to standard (UTS) date. The UTS date is a practice-based quality metric 157 and an indication of research-quality patients and periods of quality data recording 9 . Patients 158 had at least one year of UTS follow-up prior to the start of the study period (01-01-1998). First, we analysed the cumulative burden of psychiatric disorders after cancer diagnosis. Since 161 patients may experience recurrent psychiatric events, we employed the mean cumulative 162 count method to not only capture the first occurrence of the event, but also subsequent 163 occurrences. The cumulative burden method reflects a summarisation of all events that occur 164 in a population by a given time 10 (see cumulative burden section for details). Second, we analysed temporal variations in the first psychiatric disorder diagnosis in relation 167 to the first event of self-harm. The proportion of patients being diagnosed with a psychiatric 168 disorder (for the first time) before and after the first self-harm event was calculated and plotted. Third, we analysed the cumulative burden of recurrent self-harm events after the diagnosis of 171 psychiatric disorders in patients with cancer. Fourth, we estimated the risk of self-harm after diagnosis of psychiatric disorder in patients 174 with cancer. We identified incident cases of psychiatric disorder (patients who had a history of 175 psychiatric disorder before cancer diagnosis were excluded) ( Figure S1 ). Patients with a 176 history of self-harm before or at the time of the first diagnosis of psychiatric disorder were also 177 excluded. For each psychiatric disorder, the first date of diagnosis (index date) was used as 6 Fifth, we estimated the risk of suicide and other causes of death following self-harm in patients 189 with cancer. We identified incident self-harm episodes (patients who had a history of self-harm 190 before cancer diagnosis were excluded). The first date of self-harm (index date) was used as 191 the date that follow-up started. Controls were identified by propensity score matching by age 192 at cancer diagnosis, cancer type, sex, Index of Multiple Deprivation and primary care practice 193 ID. For controls, the index date of its corresponding matched case was the date that follow-up 194 started. Patients were followed up until death, date of deregistration from the practice or date 195 of administrative censoring (31-10-2020), whichever occurred first ( Figure S2 ). Phenotypes for CPRD GOLD were generated using version 2 Read codes. Phenotypes for 203 CPRD Aurum data were generated using a combination of SNOMED CT, Read version 2 and 204 EMIS Web codes. Phenotypes for HES were generated in ICD-10. EHR phenotypes for self-205 harm were obtained from a previous study 14 . We employed primary care, secondary care and 206 NCRAS records to identify patients aged ≥ 18 years with an incident primary site-specific 207 cancer. We considered 26 cancer types: bladder, bone, brain, breast, cervix, colon and 208 rectum, gallbladder and biliary tract, Hodgkin lymphoma, kidney and renal pelvis, leukaemia, 209 liver and intrahepatic bile duct, lung and bronchus, melanoma, multiple myeloma, non- Hodgkin lymphoma, oesophagus, oropharynx, ovary, pancreas, prostate, small intestine, 211 spinal cord and nervous system, stomach, testis, thyroid and uterus. We considered five 212 psychiatric diagnostic categories: depression, anxiety disorders, schizophrenia, schizotypal 213 and delusional disorders, bipolar affective disorder and mania and personality disorders. Prevalent non-cancer physical comorbidities recorded before index date entry were identified 216 from primary and secondary care records. We considered the following 21 comorbidities: heart 217 failure, chronic obstructive pulmonary disease, HIV, hepatic disorders (i.e., alcoholic liver 218 disease, non-alcoholic fatty liver disease, hepatic failure, liver fibrosis and cirrhosis and portal 219 hypertension), stroke (i.e., ischaemic stroke, stroke not otherwise specified, transient 220 ischaemic attack, intracerebral haemorrhage and subarachnoid haemorrhage), myocardial 221 infarction, vascular disease (i.e., peripheral arterial disease, Raynaud's syndrome, and venous 222 thromboembolic disease) and abnormal glucose metabolism (i.e., type 1 and type 2 diabetes, 223 diabetic neurological complications and diabetic ophthalmic complications). For stratified 224 analyses, patients were divided into the following categories: 0 additional comorbidity (i.e., 225 patients without diagnosis of any of the above non-cancer comorbidities), 1 additional 226 comorbidity, 2 comorbidities, 3 comorbidities and 4+ comorbidities. We analysed ten cancer treatment variables: (i) all chemotherapy (i.e., everyone who received We estimated the cumulative burden of five psychiatric disorders and self-harm episodes 247 using the previously described and validated mean cumulative count (MCC) method 10,15 . Given that psychiatric and self-harm episodes can recur, we employed the MCC method which 249 considers recurrent eventscumulative incidence only considers the first occurrence of the We performed Cox proportional hazards regression to determine: 1) the risk of self-harm after 265 psychiatric disorder diagnosis and 2) the risk of suicide and other causes of death following 266 self-harm. Cox regression was performed on matched cohorts and further adjusted for non-267 cancer comorbidities, cancer treatment and the presence of additional psychiatric disorders. Proportional hazards assumption was evaluated using the Schoenfeld residuals. We also 269 estimated the cumulative incidence for: 1) all-cause mortality after psychiatric disorder 270 diagnosis and 2) all-cause and cause-specific mortality after self-harm. Analyses on specific is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint Between 1998 and 2020, we identified 459,542 individuals age ≥ 18 years with an incident 294 diagnosis of a site-specific cancer of interest. Patient characteristics were presented in Table 295 S1. We analysed cumulative burden of five psychiatric disorders (depression, anxiety 296 disorders, schizophrenia, schizotypal and delusional disorders, bipolar affective disorder and 297 mania and personality disorders) following cancer diagnosis and compared results across 26 298 cancer diagnostic groups. Cumulative burden was the highest for depression, followed by 299 anxiety disorders ( Figure 1A , When comparing across treatment modalities, patients who received all three chemotherapy, 320 radiotherapy and surgery had the highest cumulative burden of psychiatric disorders ( Figure 321 1B, Among patients with cancer, 5,683 individuals were identified as having an incident self-harm 346 episode (self-harm event after cancer diagnosis) ( Figure S2 ). We estimated temporal variation 347 in the first diagnosis of a psychiatric disorder in relation to the time of the first episode of self-348 harm. We observed that across cancer types, a previous diagnosis of psychiatric disorder prior 349 to self-harm was at least twice as prevalent than a subsequent diagnosis of psychiatric 350 disorder ( Figure 2 ). Prevalence ratio was the highest in patients with brain tumours (5.36, CI: 351 4.57-6.14), followed by prostate cancer (4.30, CI: 4.08-4.52), Hodgkin lymphoma (4.17, CI: Propensity score matching was used to identify patients with and without a specific psychiatric 371 disorder ( Figure S1 ). The number of patients with incident psychiatric disorder were as follow: Table S5 ). All psychiatric disorders were significantly associated with an increased risk of subsequent 390 self-harm ( Figure 3B , Table S6 ). Self-harm risk was observed to change according to the is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 9, 2021. ; https://doi.org/10.1101/2021.10.07.21264703 doi: medRxiv preprint CI: 4.6-5.8), anxiety disorder (HR 4.0, CI: 3.5-4.5), schizophrenia (HR 2.9, CI: 2.4-3.5), bipolar 398 disorder (HR 12.1, CI: 7.5-19.5) and personality disorder (HR 9.0, CI: 2,9-28.5) ( Figure 3B , 399 Table S6 ). Across all psychiatric disorders, the cumulative incidences for all-cause mortality were 405 consistently higher in cases than controls ( Figure 4A ). We estimated excess years of life lost 406 (YLL) which is the average number of years that patients with psychiatric disorder lose in 407 excess of that found in patients without psychiatric disorder of the same age. Excess YLLs for 408 each psychiatric disorder across different age of onset were displayed as radar plots ( Figure 409 4B, Table S7 ). Younger age of psychiatric disorder onset was consistently associated with Table S7 ). Patients who self-harm had the highest risk of unnatural deaths within 12 months of 419 self-harm 420 421 Propensity score matching was used to identify cases (with self-harm) and controls (without 422 self-harm) for analyses on cause-specific mortality risk following self-harm. We identified 423 5,683 individuals with incident self-harm episodes and 18,407 matched controls ( Figure S2 ). Adults who harmed themselves were 6.8 times more likely to die of unnatural causes of death 425 compared with controls within 12 months of self-harm (HR 6.8, CI: 4.3-10.7). The risk of 426 unnatural death after 12 months was markedly lower (HR 2.0, CI: 1.5-2.7) ( Figure 5A , Table 427 S8). Cumulative incidence curves also demonstrated that the self-harm group had an 428 increased risk of dying from unnatural deaths, after adjusting for competing risk of natural 429 deaths ( Figure 5B ). . CC-BY-NC-ND 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 9, 2021. Principal findings We present the first study examining mental illness and self-harm events across 26 cancer 436 types. Utilising data from primary care practices and hospitals, we quantified the total burden 437 (not just the first event) of psychiatric disorder and self-harm. We also examined the 438 prevalence of mental health diagnoses before and after self-harm and demonstrate that 439 previous diagnoses of psychiatric disorders are important predictors of self-harm. We 440 observed considerable differences in the risk of self-harm across psychiatric disorders. Patients with depression had the highest risk of self-harm, especially within 12 months of 442 diagnosis, suggesting that patients require higher vigilance during this initial critical period. Interestingly, we found that schizophrenia was associated with a lower risk of self-harma 444 finding that is corroborated by another study conducted in Hong Kong 20 . Patients with mental 445 illness were significantly more likely to experience premature mortality. Furthermore, the risk 446 of suicide and other causes of death were significantly higher in patients who harm 447 themselves, particularly within 12 months of the first self-harm episode. Two models have been proposed as contributing to the underlying cause of psychiatric is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 9, 2021. ; https://doi.org/10.1101/2021.10.07.21264703 doi: medRxiv preprint report found that the chemotherapy 5-Fluorouracil induced manic episodes in a patient without 487 a history of psychiatric illness 32 . Since 5-Fluorouracil could penetrate the blood-brain barrier, 488 it is linked to neurotoxicity 33 and mania may be caused by injury to neurotransmitter pathways. There have been limited studies on mania caused by extracerebral cancer and is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 9, 2021. Prescribing of antidepressants will need to take into account potential contraindications or is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 9, 2021. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 9, 2021. Case (self-harm) and control (no self-harm) groups were obtained via propensity score 812 matching (see Figure S2 ). Controls were matched by age at cancer diagnosis, cancer type, is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint 30 40 50 60 70 80 30 40 50 60 70 80 30 40 50 60 70 80 30 40 50 60 70 80 30 40 50 60 70 80 30 40 50 60 70 80 30 40 50 60 70 80 30 40 50 60 70 80 30 40 50 60 70 80 30 40 50 60 70 80 30 40 50 60 70 80 30 40 50 60 70 80 30 40 50 60 70 80 30 40 50 60 70 is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 9, 2021. 5.36 ( 4.57 − 6.14 ) 5.36 ( 4.57 − 6.14 ) 5.36 ( 4.57 − 6.14 ) 5.36 ( 4.57 − 6.14 ) 5.36 ( 4.57 − 6.14 ) 5.36 ( 4.57 − 6.14 ) 5.36 ( 4.57 − 6.14 ) 5.36 ( 4.57 − 6.14 ) 5.36 ( 4.57 − 6.14 ) 5.36 ( 4.57 − 6.14 ) 5.36 ( 4.57 − 6.14 ) 5.36 ( 4.57 − 6.14 ) 5.36 ( 4.57 − 6.14 ) Bars on the left side of the graphs indicate the proportion of individuals being diagnosed with a psychiatric disorder before the first self-harm event. Bars on the right side of the graphs indicate the proportion of individuals being diagnosed with a psychiatric disorder after the first self-harm event. Prevalence ratios (prevalence of psychiatric disorder diagnosis before self-harm divided by prevalence of psychiatric disorder after self-harm) were shown, alongside 95% confidence intervals. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 9, 2021. Self-harm < 12 months Self-harm ≥12 months Self-harm at any time during follow-up Figure 3 . Total burden and risk of self-harm after diagnosis of psychiatric disorders in patients with cancer. Case (with psychiatric disorder) and control (no psychiatric disorder) groups were obtained via propensity score matching (see Figure S1 ). Controls were matched by age at cancer diagnosis, cancer type, sex, Index of Multiple Deprivation and primary care practice ID. (A) Cumulative burden of self-harm events in cases and controls. (B) Hazard ratios for risk of self-harm for each psychiatric disorder were further adjusted for non-cancer comorbidities, cancer treatment and presence of other psychiatric disorders. Self-harm risks during the first 12 months and subsequent years of follow-up were shown. Strata with low event numbers (n < 10) were not analysed. All data and 95% confidence intervals are provided in the supplementary tables. . Cumulative incidence of all-cause mortality and years of life lost after diagnosis of psychiatric disorders in patients with cancer. Case (with psychiatric disorder) and control (no psychiatric disorder) groups were obtained via propensity score matching (see Figure S1 ). Controls were matched by age at cancer diagnosis, cancer type, sex, Index of Multiple Deprivation and primary care practice ID. (A) Cumulative incidence curves for all-cause mortality after psychiatric disorder diagnosis in matched case and control groups. (B) Excess years of life lost (YLL) attributable to psychiatric disorders in patients with cancer. Radar plots depict the difference in YLL between matched cases and controls. Excess YLL was estimated based on the age of onset of a psychiatric disorder. All data and 95% confidence intervals are provided in the supplementary tables. . CC-BY-NC-ND 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted October 9, 2021. ; https://doi.org/10.1101/2021.10.07.21264703 doi: medRxiv preprint p < 0.001 p < 0.001 p < 0.001 Figure 5 . Risk of suicide and other causes of death following self-harm in patients with cancer. Case (self-harm) and control (no self-harm) groups were obtained via propensity score matching (see Figure S2 ). Controls were matched by age at cancer diagnosis, cancer type, sex, Index of Multiple Deprivation and primary care practice ID. (A) Hazard ratios for risk of suicide and other causes of death were further adjusted for non-cancer comorbidities, cancer treatment and presence of psychiatric disorders. Mortality risks during the first 12 months and subsequent years of follow-up were shown. (B) Cumulative incidence curves of death due to all causes, natural causes and unnatural causes after self-harm in matched case and control groups. A B Risk of self-harm after the diagnosis of psychiatric 699 disorders in Hong Kong, 2000-10: a nested case-control study. The Lancet 700 Impact of chemotherapy-induced alopecia distress 702 on body image, psychosocial well-being, and depression in breast cancer patients Fear of Cancer Recurrence A Scoping Review About Their Conceptualization and 706 Personality Disorders in Patients with Cancer Prevalence of depression in cancer patients: 711 a meta-analysis of diagnostic interviews and self-report instruments Depression and Anxiety in Patients With 714 Cancer: A Cross-Sectional Study Depression, cytokines, and pancreatic 716 cancer Psychiatric manifestations of paraneoplastic 718 disorders Paraneoplastic syndromes: when to suspect, how to confirm, and how to 720 manage Bipolar disorder in older adults: a critical review Acute-onset mania in a patient with non-small 724 cell lung cancer Psychiatric complications of 726 treatment with corticosteroids: review with case report Onset of manic episode during 729 chemotherapy with 5-fluorouracil 5-fluorouracil-induced 731 leukoencephalopathy in patients with breast cancer Paraneoplastic syndromes of the CNS Mania associated with small cell carcinoma of the lung Late-onset mania associated with corticosteroids and small 738 cell lung carcinoma End-of-life care among patients with schizophrenia 42 Open verdict v. suicide--importance to research Psychiatric disorder as a first 756 manifestation of cancer: A 10-year population-based study Intracranial mass lesions associated with late-759 onset psychosis and depression Management of depression in patients with cancer: 761 a clinical practice guideline Preventing the co-prescription of tamoxifen and fluoxetine in