Psychiatric Disorders Are Associated with Increased Risk of Sepsis Following a Cancer Diagnosis

Sepsis is a common complication of cancer and its treatment. Patients with cancer have a nearly ten times higher risk of sepsis, compared with individuals without cancer (1, 2). Sepsis is associated with markedly elevated healthcare use and medical costs as well as increased mortality (1–3). It is therefore important to identify risk factors for sepsis among patients with cancer, ideally during the entire disease course (from diagnosis until later survivorship), with the aim to improve cancer survival and reduce healthcare cost. Psychiatric disorders, especially depression, anxiety, and adjustment disorders are also common comorbidities among patients with cancer (4) and have a wide spectrum of effects on physical health, including modulating the risk of infections (5–7). One potential pathway that links together psychiatric disorders with infections is psychological stress, which is known to play an important role in the onset, maintenance, and aggravation of psychiatric disorders (8) as well as in modulating immune function (9, 10). Disparities in healthcare access in relation to psychiatric disorders and differential risk behaviors among patients with psychiatric disorders might be additional or alternative mechanisms linking together psychiatric disorders with infection, especially severe infections (6, 11, 12). Although recent studies have suggested that depression and other stress-related disorders are associated with severe infections in the general population (13, 14), whether a similar association exists among patients with cancer, a high-risk group for infections, remains unknown. To this end, we assessed the risk of subsequent sepsis, as a complication of infections, among cancer patients with precancer psychiatric disorders, in contrast to cancer patients without such disorders.

We identified 362,500 patients (age ≥30) with a newly diagnosed cancer from January 1, 2006 to December 31, 2014 in Sweden, based on the Swedish Cancer Register, which has been available since 1958 with a nearly 100% coverage of the entire country. From the Cancer Register, we obtained information on date of diagnosis and cancer type. Since 2004, the register has also collected relatively complete information on cancer stage at the time of diagnosis, except for hematologic malignancy and central nervous system (CNS) tumors. According to the 10th Swedish Revision of the International Classification of Diseases (ICD-10) codes, we further classified cancers into eight groups, including prostate cancer, breast cancer, colorectal cancer, lung cancer, non–melanoma skin cancer, CNS tumor, hematologic malignancies, and other cancers (Supplementary Table S1). We used the personal identification numbers uniquely assigned to all Swedish residents to follow these patients from date of diagnosis, until a diagnosis of sepsis, emigration out of Sweden, death, or December 31 2014, whichever occurred first, through cross-linkages to the Swedish Patient Register (for ascertainment of sepsis), Migration Register, and Causes of Death Register. The Swedish Patient Register includes discharge records from inpatient care (100% complete for the country since 1987) and outpatient based-specialist care (>80% complete for the entire country since 2001). Because we aimed to study the association of psychiatric disorders with sepsis subsequent to a cancer diagnosis, patients with cancer that had a diagnosis of sepsis before cancer diagnosis were excluded (N = 6,498), leaving 356,002 in the final analysis (98%; Supplementary Fig. S1).

Through the Swedish Patient Register, we identified clinical diagnoses of psychiatric disorders among the cancer patients between January 1, 2001 and the date of cancer diagnosis. Psychiatric disorders were defined through any first inpatient or outpatient hospital visit with a main diagnosis concerning the ICD-10 codes F10-F99. Patients with cancer that had a clinical diagnosis of psychiatric disorders before cancer diagnosis were defined as exposed to psychiatric disorders, whereas the other patients with cancer served as the unexposed group (i.e., reference). We further classified psychiatric disorders as substance abuse (ICD-10: F10–F16/F18–F19), depression (ICD-10: F32–F33), anxiety (ICD-10: F40–F41), stress reaction and adjustment disorder (ICD-10: F43), somatoform/conversion disorder (ICD-10: F44–F45), and other psychiatric disorders.

A clinical diagnosis of sepsis was defined as any first inpatient or outpatient hospital visit with a main diagnosis concerning sepsis, according to the Patient Register. The ICD-10 codes used to define sepsis are listed in Supplementary Table S2 (15, 16).

We collected information about region of residence, marital status, and educational level from the Swedish Longitudinal Integration Database for Health Insurance and Labor Market. We calculated Charlson Comorbidity Index score (17) within the five years prior to cancer diagnosis for all patients according to the Patient Register. Cancer stage was ascertained by the TNM or FIGO information from the Cancer Register, including limited-stage localized (T-localized/N0/M0 or FIGO 0, I), advanced-stage localized (T-advanced/N0/M0 or FIGO II), regional spread (any T/N+/M0 or FIGO III), advanced (any T/any N/M+ or FIGO IV), and unknown stage (detailed information is available from https://www.encr.eu/sites/default/files/pdf/extentofdisease.pdf; refs. 18–20). We used information on cancer subtypes, including lymphoma, myeloma, leukemia, myelodysplastic syndrome, and myeloproliferative neoplasms for hematological malignancy, instead of stage. Unknown was used as the cancer stage for CNS tumors. As the risk of sepsis might differ between patients with and without a history of severe infection before cancer diagnosis, we also obtained information on history of inpatient care due to infection within the three months before cancer diagnosis through the Patient Register. ICD-10 codes used to identify infections are listed in Supplementary Table S2 (21, 22).

We first described the characteristics of patients with cancer, including sex, age at diagnosis, calendar period of diagnosis (2006–2010 or 2011–2014), region of residence (Southern Sweden, Central Sweden, or Northern Sweden), marital status (single, married or cohabiting, or divorced or widowed), educational level (>12 years, 9–12 years, <9 years, or unknown), type of cancer, cancer stage, infection history, and Charlson Comorbidity Index score (0, 1, or ≥2). We calculated the unadjusted incidence rates (IR) of sepsis among patients with or without precancer psychiatric disorders during the entire follow-up, using number of cases divided by accumulated person-years at risk. To demonstrate the temporal pattern of absolute rate of sepsis along with time since cancer diagnosis, we first fitted a flexible parametric survival model to estimate hazards of sepsis among these two groups of patients, allowing the hazards to change over time (i.e., nonproportional hazards). We then used flexible parametric survival models to estimate the HRs and their corresponding 95% confidence intervals (CI) for sepsis, after adjusting for other covariates. We first performed models that allow for nonproportional hazards for the effect of exposure, that is, the HRs were allowed to change over time since cohort entry. Because we observed constant HRs over time, we also performed proportional hazards models to estimate average HRs for the entire follow-up period. Time since cohort entry was used as the underlying time scale in all analyses and the analyses were performed in two models. In model 1, analyses were adjusted for sex, age at cancer diagnosis, calendar period, region of residence, and type of cancer, as potential confounders. In model 2, we further adjusted for covariables that might both confound and mediate the studied association, including marital status, educational level, cancer stage, infection history, and Charlson Comorbidity Index score. We hypothesized that while sex, age, calendar period, region of residence, and type of cancer might be potential confounders for the association of psychiatric disorders with sepsis, marital status, educational level, cancer stage, infection history and Charlson Comorbidity Index score might instead serve as potential mediators linking together precancer psychiatric disorders with sepsis after cancer diagnosis. A detailed description of the statistical models is provided in the Supplementary Methods.

The association of precancer psychiatric disorders with sepsis might vary between patients with different sociodemographic and clinical characteristics, that is, patient characteristics might interact with the effect of precancer psychiatric disorders on sepsis. For instance, the risk of sepsis following a cancer diagnosis might differ by sex, age, socioeconomic status, disease severity, susceptibility to infection, and general health status. We therefore performed additional analyses to assess the interactions between precancer psychiatric disorders and sex, age, calendar period, region of residence, marital status, educational level, cancer stage, infection history, or Charlson Comorbidity Index score. The differences between subgroup-specific HRs were assessed by introducing interaction terms to the flexible parametric models and the corresponding statistical significance was assessed by likelihood ratio tests. As the association might also vary for different cancer types and individual psychiatric disorders, we also separately analyzed patients with different cancers, as well as for different individual psychiatric disorders. The differences between group-specific HRs were assessed by introducing interaction terms to the flexible parametric models and the corresponding statistical significance was assessed by likelihood ratio tests. As the association might also vary for different cancer types and individual psychiatric disorders, we also separately analyzed patients with different cancers, as well as for different individual psychiatric disorders. We found in a previous study that the risk of psychiatric disorders started to increase quickly from approximately one year before cancer diagnosis (23), which might be attributable to the experience of psychological distress in relation to the emerging cancer symptoms or the clinical evaluation of a potential cancer (24), or simply the cancer-related inflammation (25). In another sensitivity analysis, we restricted the definition of precancer psychiatric disorders to the ones diagnosed more than one year before cancer diagnosis.

Analyses were performed using statistical software (SAS, version 9.4; SAS Institute and Stata, version 15.1; StataCorp LP). P < 0.05 was used to indicate statistical significance. The study was approved by the Regional Ethical Review Board in Stockholm, Sweden.

We identified 22,609 patients with cancer with a diagnosis of psychiatric disorders before cancer diagnosis, and 333,393 patients without such a diagnosis. The median age at diagnosis was 64 for patients with cancer with precancer psychiatric disorders and 68 for patients with cancer without such comorbidities. Among patients with precancer psychiatric disorders, 50% were male; among patients without pre-cancer psychiatric disorders, 54% were male.

Compared to patients without psychiatric disorders, patients with psychiatric disorders were in general more likely to be female, diagnosed with cancer at a younger age and at a later calendar period, non-cohabitating, and have a higher prevalence of infection history and a higher Charlson Comorbidity Index score (Table 1).

During a median follow-up of 2.6 years, a total of 13,005 cases of sepsis were identified. Among these cases, 11,104 (85.4%) were ascertained through inpatient care, whereas 1,901 (14.6%) through outpatient care. We identified 872 incident cases of sepsis among patients with cancer with precancer psychiatric disorders (IR, 14.8 per 1000 person-years) and 12,133 cases among patients with cancer without precancer psychiatric disorders (IR, 11.6 per 1,000 person-years; Table 2). Patients with precancer psychiatric disorders had constantly higher hazard of sepsis, compared with patients without precancer psychiatric disorders, during the entire follow-up apart from the very brief time period immediately after cancer diagnosis (Fig. 1A). Compared with the reference, we found an increased rate of sepsis among patients with cancer with precancer psychiatric disorders. The increased rate was noted immediately after cancer diagnosis and persisted until the end of follow-up (Fig. 1B). Because of the constant HRs noted during the entire follow-up, all HRs presented below are derived from models assuming proportional hazards for effect of precancer psychiatric disorders. The increased rate was similar in both the simpler (HR, 1.31; 95% CI, 1.22–1.40) and full (HR, 1.26; 95% CI, 1.18–1.35) models. The increased rate did not vary greatly by sex, calendar period of diagnosis, region of residence, educational level, infection history, or Charlson Comorbidity Index score, but was slightly greater among younger patients, non-cohabitating patients, and patients with lower stage cancer (Table 2).

Largely similar results were found for all major cancer types, except hematologic malignancy and non-melanoma skin cancer (Table 3). The strongest associations were observed for breast cancer, prostate cancer, and the group of other cancers. Similar results were also found for most individual psychiatric disorders studied, with the strongest association noted for substance abuse, followed by depression and other psychiatric disorders (Table 4). When restricting the definition of precancer psychiatric disorders to psychiatric disorders diagnosed more than one year before cancer diagnosis, a similar overall association and temporal pattern of the association were noted (Supplementary Fig. S2).

On the basis of a large national cohort, we found that preexisting psychiatric disorders before cancer diagnosis were associated with a higher subsequent rate of sepsis among patients with cancer. Such an association was noted immediately after cancer diagnosis and persisted up to nine years after cancer diagnosis. The association did not differ greatly by sex, calendar period, region of residence, educational level, infection history, or Charlson Comorbidity Index score, but was slightly stronger among patients diagnosed at younger age, non-cohabitating patients, and patients with a lower stage cancer.

Sepsis is a common complication among patients with cancer due to immunosuppression caused by intensive cancer treatment, the hospital setting, or factors associated with the underlying malignancy itself (26). Sepsis is known to be associated with longer hospital stay and higher medical cost (26). Despite intensive care, sepsis further leads to higher mortality among patients with cancer (26, 27). Psychiatric disorders are also common concomitants of cancer (4) and have been suggested to play a significant role in somatic illnesses, including severe infections and sepsis (5–7, 13, 14). The potential role of psychiatric disorders on sepsis among patients with cancer has however rarely been investigated. Our results provide therefore new knowledge to substantiate an elevated risk of sepsis in relation to psychiatric disorders among patients with cancer, a high-risk population for sepsis. Such knowledge could provide evidence for early prevention and intervention for vulnerable individuals, and subsequently improve prognosis of patients with cancer and optimize the use of medical resources. In this study, we focused on psychiatric disorders before cancer diagnosis as the exposure. However, increased risk of newly onset psychiatric disorders after cancer diagnosis has also been reported. Whether and how postcancer psychiatric disorders influence the risk of sepsis among patients with cancer remains therefore to be examined, taking into full consideration of cancer treatment and progression.

A link between psychiatric disorders and sepsis is plausible. One potential pathway is psychological distress, which is a known correlate of psychiatric disorders and modulates immune responses through the hypothalamic–pituitary–adrenal axis and sympathetic nervous system (9, 10). Mounting in vivo and in vitro evidence has supported that psychologic distress downregulates immune responses by suppressing antibody production and inhibiting T-cell responses (28, 29). Recent studies also indicated that psychological distress stimulates systemic low-grade inflammation by elevating plasma levels of IL6 and C-reactive protein (30, 31). Another potential pathway is the inequalities to health care among patients with psychiatric disorders in general, compared to individuals without such disorders. For instance, patients with psychiatric disorders tend to have less healthcare access in general, despite greater access to psychiatric care (6, 11). Delayed diagnosis and treatment for infections might therefore have contributed to the increased risk of sepsis among cancer patients with psychiatric comorbidities. Similarly, a lower degree of compliance to medical treatment among patients with psychiatric disorders (6, 11) might be an additional explanation. Finally, the noted association might also be attributable to differential behavioral factors between patients with and without psychiatric disorders. For instance, patients with substance abuse are more likely to experience infections through unsafe injection and sexual risk behaviors (12).

There was an increased risk of sepsis in relation to precancer psychiatric disorders among patients with most cancer types but not hematologic malignancy or non-melanoma skin cancer. The different results might be attributable to the different underlying diseases and their treatments. For instance, prophylactic use of antibiotics among patients with hematologic malignancy, who are highly susceptible to infections, might have contributed to the null association. The lack of association among patients with non-melanoma skin cancer might on the other hand be due to the general low likelihood of systemic treatment in this patient group.

A major strength of our study is the nationwide cohort design with large sample size and long and complete follow-up. Together with the prospectively and independently collected information on psychiatric disorders and sepsis, these strengths greatly reduced concerns about selection and information biases. The possibility to adjust for a handful of potential confounders was another strength of the study.

Our study also has limitations. One limitation of this study is the lack of information on cancer treatment. Cancer treatment, including surgery, chemotherapy, and radiotherapy, is generally invasive or intensive, and could increase the risk of sepsis through potential pathogen contamination or immunosuppression. Lack of treatment information therefore prevented us from assessing the role of specific cancer treatment on the association between precancer psychiatric disorders and sepsis. Because patients with cancer with psychiatric disorders might be more likely to receive less invasive and intensive treatment compared with cancer patients without psychiatric disorders (32), the noted association could have been an underestimate of the real association because of the possible lower rate of treatment-related sepsis among the exposed group. Another limitation of this study is the lack of data on lifestyle factors (e.g., smoking, nutrition, physical activity) that may also impact risk of sepsis, because individuals with psychiatric disorders have generally more unhealthy lifestyles compared to individuals without such disorders (33). Similar results observed among patients with different Charlson Comorbidity Index scores, which could be regarded as a proxy of overall health status, might partly alleviate such concern.

In conclusion, preexisting psychiatric disorders before cancer diagnosis were associated with an increased risk of sepsis among cancer patients. Our findings call for extended prevention and surveillance of sepsis among cancer patients with psychiatric comorbidities.

Q. Liu reports grants from Chinese Scholarship Council during the conduct of the study. T.M.-L. Andersson is involved in an ongoing public-private real-world evidence collaboration between Karolinska Institutet and Janssen Pharmaceuticals, however, the current project is not related to this collaboration. F. Fang reports grants from Swedish Cancer Society, grants from Swedish Research Council for Health, Working Life and Welfare, and grants from Karolinska Institutet during the conduct of the study. No potential conflicts of interest were disclosed by the other authors.

Q. Liu: Conceptualization, formal analysis, methodology, writing-original draft, writing-review and editing. H. Song: Conceptualization, methodology, writing-review and editing. T.M.-L. Andersson: Conceptualization, formal analysis, supervision, methodology, writing-review and editing. P.K.E. Magnusson: Supervision, writing-review and editing. J. Zhu: Methodology, writing-review and editing. K.E. Smedby: Conceptualization, resources, supervision, methodology, writing-review and editing. F. Fang: Conceptualization, supervision, funding acquisition, methodology, writing-review and editing.

This study was supported by the Swedish Cancer Society (no. CAN 2017/322), the Swedish Research Council for Health, Working Life and Welfare (no. 2017-00531), the China Scholarship Council (no. 201700260291), and the Karolinska Institutet (Senior Researcher Award and Strategic Research Area in Epidemiology).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.