A Nationwide Cohort Study of the Association Between Hospitalization With Infection and Risk of Death by Suicide

Importance  Findings suggest that infections might be linked to the development of psychiatric disorders and suicidal behavior. Large-scale studies are needed to investigate the effect of infection on the risk of suicide.

Objective  To estimate the association between hospitalization with infection and the risk of death by suicide.

Design, Setting, and Participants  Nationwide, population-based, prospective cohort study with more than 149 million person-years of follow-up. Data were analyzed with survival analysis techniques and were adjusted for sex, age, calendar period, cohabitation status, socioeconomic status, and the Charlson Comorbidity Index. Individual data were drawn from Danish longitudinal registers. A total of 7.22 million individuals 15 years or older living in Denmark between January 1, 1980, and December 31, 2011, were observed during a 32-year follow-up period.

Main Outcomes and Measures  The risk of death by suicide was identified in the Danish Cause of Death Register. Incidence rate ratios (IRRs) and accompanying 95% CIs were used as measures of relative risk.

Results  In 7 221 578 individuals (3 601 653 men and 3 619 925 women) observed for a total of 149 061 786 person-years, 32 683 suicides were observed during the follow-up period. Among the suicides, 7892 (24.1%) individuals had previously been diagnosed as having an infection during a hospitalization. Hospitalization with infection was linked to an elevated risk of suicide, with an IRR of 1.42 (95% CI, 1.38-1.46) compared with those without prior infection. Dose-response relationships were observed with respect to the number of hospital contacts for different infections. For example, having 7 or more infections was linked to an IRR of 2.90 (95% CI, 2.14-3.93). The number of days of treatment for infections was associated with an elevated risk of suicide in a dose-response relationship. More than 3 months of hospital treatment was linked to an IRR of 2.38 (95% CI, 2.05-2.76). The population-attributable risk associated with hospitalization with infection accounted for 10.1% of suicides.

Conclusions and Relevance  An increased risk of death by suicide was found among individuals hospitalized with infection in prospective and dose-response relationships. These findings indicate that infections may have a relevant role in the pathophysiological mechanisms of suicidal behavior.

Infection is one of the most common causes of illness in the world,¹ and it is increasingly recognized that infection and inflammation can have a critical role in psychiatric disorders^2-6 and in suicidal behavior.^7-10 Suicide is a major health problem worldwide.¹¹ However, while psychological predictors of suicide, such as psychiatric disorders and history of suicide attempts,¹² have been examined extensively, there is a dearth of longitudinal studies investigating the effect of biological factors, such as infection.

Although certain infectious agents infect the brain directly, others reach the brain from the periphery, and yet others generate molecular mediators of inflammation that cross from the periphery into the brain and thereby increase the risk of suicide.¹³ Examples of agents infecting the brain are influenza B virus¹⁴ and the parasite Toxoplasma gondii, which have been linked to suicidal behavior.^15-18 In addition, an association between T gondii and phenotypes of suicidal behavior, such as impulsivity and aggression, has been reported in healthy adults¹⁹ and psychiatric patients.²⁰

Inflammation might also lead to dysregulation of tryptophan metabolites,^21,22 with state-dependent elevation of the excitotoxic metabolite quinolinic acid in the cerebrospinal fluid²³ and increased brain barrier free crossing of the metabolite kynurenine (a precursor of quinolinic acid), increased in the blood from patients with suicidal behavior compared with healthy and psychiatric controls.²⁴ Quinolinic acid is an N-methyl-d-aspartate receptor agonist with the capability of altering glutamate neurotransmission.²⁵ Furthermore, site-specific changes in brain concentrations of quinolinic acid were reported in patients with severe depression who died by suicide.^26,27 In addition, postmortem brain samples of patients who died by suicide have shown increased messenger RNA levels of inflammatory cytokines.^28,29 Elevated levels of peripheral inflammatory markers, such as cytokines, have also been found in the blood of people who attempted suicide and patients with high ratings of suicidal ideation.^30-32 However, most previous studies^{9,14,16,23,24,26,28-32} have been limited by small sample size and cross-sectional study designs.

The research on the neurobiological basis of suicidality is in rapid development, but large-scale studies are needed to clarify the role of infection in suicide. In the present study, we investigated associations between infectious diseases and the risk of death by suicide using Danish nationwide registers. To address possible causality, we examined the risk of suicide associated with the number of hospitalizations with different infections, the number of days hospitalized with treatment for infection, and the time since the last hospitalization with treatment for infection.

Box Section Ref ID

All individuals 15 years or older who were living in Denmark between January 1, 1980, and December 31, 2011, were included in our analyses. Those who turned 15 years old or who immigrated into the country were included beginning on the respective dates. The cohort consisted of 7 221 578 individuals followed up until death, emigration from Denmark, disappearance, or December 31, 2011, whichever came first.

The Danish Civil Registration System contains complete and continuously updated administrative data on all residents living in Denmark.³³ A unique personal identification number assigned to each person enables accurate and complete linkage between various national registries. Data were obtained from the following registers: the National Hospital Register,³⁴ the Psychiatric Central Research Register,³⁵ and the Cause of Death Register.³⁶ Diagnoses were recorded according to the diagnostic system of the International Classification of Diseases, Eighth Revision (ICD-8) until January 1, 1994, after which date the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10)was used. All registers have full national coverage, and information from the registers are anonymized when used for research. This study was approved by the Danish Data Protection Agency. Because of the nature of the data, informed consent was not required.

A history of infection was defined as 1 or more infection diagnoses listed since 1977 in the National Hospital Register.³⁴ After 1995, outpatient visits and visits to emergency departments were included. All infections were grouped into subcategories, including pathogen (ie, bacterial, viral, and remaining pathogens), as well as type of infection (ie, sepsis, hepatitis, gastrointestinal tract, skin, respiratory tract, urological, genital, pregnancy-related [among women 49 years or younger], otitis media, central nervous system, and human immunodeficiency virus [HIV] or AIDS infections). All ICD-8 and ICD-10 codes for the included pathogens and types of infection are listed in eTable 1 in the Supplement. In addition, we assessed the effect by the number of hospitalizations with different infections (measured as different 2-digit ICD codes [eg, A40 and A41]), the number of days hospitalized with treatment for infection (measured as the number of bed-days in the hospital), and the time since the last hospitalization with treatment for infection (measured as the number of days since discharge from a hospitalization with infection).

Data on psychiatric diagnoses were obtained from the Psychiatric Central Research Register,³⁵ which lists hospitalizations since 1968. After 1995, visits to psychiatric outpatient clinics and psychiatric emergency units were included. For the examined psychiatric diagnoses, the date of onset was defined as the first date of a recorded diagnosis in the register. Included diagnoses are listed in eTable 2 in the Supplement.

The outcome of interest was death by suicide. Suicide was identified in the Cause of Death Register as ICD-8 codes E950 to E959 or ICD-10 codes X60 to X84 or Y87 or if the manner of death was listed as suicide.

We used logistic regression models, assuming a Poisson distribution to calculate incidence rate ratios (IRRs) and 95% CIs. We measured the risk of suicide among individuals diagnosed as having an infection relative to individuals without an infection. The risk estimates were obtained for different infection categories, including pathogen and type of infection, number of hospitalizations with different infections, number of days hospitalized with treatment for infection, and time since the last hospitalization with treatment for infection. No differences were noted between the sexes except that women had a higher incidence of death by suicide after infection, which is why the basic model was adjusted for sex, age (15-29, 30-49, 50-64, 65-79, and ≥80 years), and calendar period (1980-1989, 1990-1999, and 2000-2011). In the fully adjusted model, we further adjusted for cohabitation status (cohabitation vs no cohabitation), socioeconomic status (working, unemployed, disability pension, early retirement, student, or other), and the Charlson Comorbidity Index (none vs ≥1 comorbid disorders).³⁷ To investigate whether psychiatric disorders might have a mediating role, the risk of suicide relative to the timing of infection and psychiatric diagnosis was examined in a sensitivity analysis. We also calculated the population-attributable risk³⁸ and used the IRR of suicide from the fully adjusted model with the following formula:

Population-Attributable Risk = P_e(IRR_e − 1) / 1 + P_e(IRR_e − 1),where P_e is the prevalence of exposure and IRR_e is the IRR of suicide due to that exposure. All included covariates except sex were time-varying covariates and were updated either on the exact date of a change in status or on a yearly basis. For all the estimates, the same individual could, at different stages, contribute with risk time to different groups (eg, number of infections).

We conducted a sensitivity analysis by restricting the sample to a subcohort born after 1962, among whom more complete coverage of exposure to infections was secured (ie, only childhood exposure was not included). In a second sensitivity analysis, the sample was restricted to persons with no previous psychiatric diagnoses.

All data presented in this article were analyzed using statistical software. The software package used was SAS, version 9.2 (SAS Institute Inc).

The cohort consisted of 7 221 578 individuals (3 601 653 men and 3 619 925 women) observed for a total of 149 061 786 person-years. During the follow-up period, 809 384 (11.2%) individuals had a hospitalization with infection. A total of 32 683 suicides were observed during the follow-up period. Of those, 7892 (24.1%) individuals had previously been diagnosed as having an infection during a hospitalization.

In the fully adjusted model, the relative risk of suicide was increased among individuals with any hospitalization with infection, with an IRR of 1.42 (95% CI, 1.38-1.46) compared with individuals without infection (Table 1). Except for pregnancy-related infections, all examined types of infections were linked to an increased risk of suicide, which was most pronounced for HIV or AIDS, hepatitis, respiratory tract, and sepsis infections.

The risk of suicide increased in a dose-response relationship with the number of hospitalizations with different infections (Figure, A). The highest incidence of suicide was found for having 7 or more infections, with an IRR of 2.90 (95% CI, 2.14-3.93), while individuals with one hospitalization had an IRR of 1.34 (95% CI, 1.30-1.38) compared with those with no hospitalization. After receiving treatment for infection in the hospital for 1 to 4 days, the IRR was 1.46 (95% CI, 1.41-1.52), while the risk was 2.38 (95% CI, 2.05-2.76) after 94 days in treatment (Figure, B). The time since the last hospitalization with treatment for infection was linked to an elevated risk of suicide (P < .001), with the strongest effect after 1 and 2 years compared with those without infection (Figure, C).

Our analysis yielded a population-attributable risk associated with hospitalization with infection of 10.1% of suicides. The implication is that 1 in 10 suicides could be prevented if infections were to be eliminated entirely, provided that the association was causal.

In a sensitivity analysis, hospitalization with infection was still significantly associated with suicide when we restricted the study sample to individuals who had never been diagnosed as having a psychiatric disorder (ie, comparable to adjusting for psychiatric disorders). A 1.31-fold (95% CI, 1.27-1.36) higher risk of suicide was observed among persons who did not have a history of any psychiatric diagnosis. Similarly, restricting the sample to individuals with no diagnosis of schizophrenia or affective disorder was linked to an elevated risk of suicide (IRR, 1.40; 95% CI, 1.36-1.44). When only assessing the effect of infections among individuals who had never been diagnosed as having a substance use diagnosis, a significant effect was also observed (IRR, 1.35; 95% CI, 1.36-1.39).

To examine the time relationship, we stratified the analysis to individuals with a history of psychiatric diagnosis according to the time of infection. We found that individuals diagnosed as having an infection before a psychiatric diagnosis had an IRR of 1.21 (95% CI, 1.15-1.28) for suicide compared with individuals with a psychiatric diagnosis and no infection (Table 2). Individuals diagnosed as having an infection after a psychiatric diagnosis had an IRR of 0.93 (95% CI, 0.88-0.98) compared with individuals with a psychiatric diagnosis and no infection.

We also restricted the cohort to those born after 1962. These individuals with infection had a higher risk of suicide (IRR, 1.40; 95% CI, 1.33-1.49) compared with those without infection.

To our knowledge, this investigation is the largest study to date to examine infections as a predictor of death by suicide. In this nationwide population-based cohort study, we found a 42% significantly increased risk of death by suicide in those with a history of hospitalization with infection and subsequent death by suicide. The data are consistent with a dose-response relationship with respect to the number of hospitalizations with different infections and the number of days hospitalized with treatment for infection. Also, an increased risk of suicide was noted relative to the time since the last hospitalization with treatment for infection. These risk estimates remained significantly increased after adjusting for important confounding variables, including sex, age, calendar period, cohabitation status, socioeconomic status, and the Charlson Comorbidity Index.

In this large-scale study, we found that all pathogens are linked to an elevated risk of suicide, as well as most types of infections. The only exceptions were pregnancy-related infections and otitis media infections. While the protective effect of parenthood on an individual’s risk of suicide³⁹ could explain the lack of association with pregnancy-related infections, otitis media infections often occur early in life and generally induce a lower systemic response, which might explain a lack of influence. Types of infection, such as sepsis, respiratory tract, and genital infections, induce a more severe systemic response that could be more likely to affect the brain, which adds support to the hypothesis of a biological link between infections and death by suicide.¹³

The highest risk of suicide was found among individuals with hepatitis infection and HIV or AIDS infection, which is in line with previous findings that these patients report higher levels of suicidal ideation and behavior than the general population.^40-42 Furthermore, HIV or AIDS infection is strongly linked with psychiatric disorders (eg, depression and substance use).^41,42 However, after excluding individuals with psychiatric diagnoses, including substance use diagnoses, the risk estimates herein remained significant, implying that this finding may only explain part of the association.

Psychiatric disorders are potential mediators, and a psychiatric diagnosis before an infection can act as a potential effect modifier. In a sensitivity analysis, an increased risk of death by suicide was found among individuals diagnosed as having an infection before a psychiatric disorder compared with individuals with a psychiatric diagnosis and no infection. Notably, the risk of suicide was not elevated among those with an infection after a psychiatric diagnosis. However, it is well known that individuals with severe mental illness experience increased mortality after infection.⁴³ Hence, individuals diagnosed as having a psychiatric disorder and then admitted for an infection seem to die of causes other than suicide. An effect related to infections was also noted herein after excluding individuals diagnosed as having psychiatric disorders, including substance use. This finding suggests that the association between infection and death by suicide is not solely explained by psychiatric disorders. Still, substance use may be underdiagnosed among psychiatric inpatients.⁴⁴

As previously mentioned, there are several potential causal links between infection and suicide. Our findings support the literature in this field, in which small-scale studies^{9,14,16,23,24,26,28-32} have linked infections, proinflammatory cytokines, and inflammatory metabolites to increased risk of suicidal behavior. In addition, infection with T gondii has been associated with suicide attempts.^15-18

The association between infection and suicide could also be an epiphenomenon or mediated through other related predictors. The psychological effect of being hospitalized with a severe infection might affect the risk of suicide. However, after adjusting for the effect of long-term and physical diseases using the Charlson Comorbidity Index, the association between infection and suicide was still significant. It is also possible that immune alterations that lead to or are the consequence of past or long-term infections, as well as potentially reactivating infections, could lead to suicidal behavior.¹³ Furthermore, infection and treatment with antibiotics can have an effect on the human microbiome,⁴⁵ which could also influence the associations indicated by recent research on the microbiome-gut-brain axis and its effects on human behavior.⁴⁶ Bacterial infections were only slightly more associated with suicide than viral infections, indicating that the associations are less likely to merely be due to antibiotic treatment affecting the microbiome.

Although psychiatric disorders and previous suicide attempts might constitute stronger predictors of suicide,¹² we found that hospitalization with infection accounted for a population-attributable risk of 10.1%. Admittedly, infections might not be amenable to direct interventions. While our findings do not exclude the possibility that genetic vulnerability factors for suicide exist, they indicate that genetic factors and environmental factors, such as infections, are both linked to suicide.

Strengths of the study are the use of nationwide registers with individual-level data, which allows investigation of outcomes with low frequency, such as suicide. By adjusting for a range of time-varying covariates, our risk estimates gained precision. All data were collected prospectively and studied during a long follow-up period of 32 years. There is free access to and treatment in Danish hospitals, full national coverage in all registers, and no loss to follow-up.

Our study had some limitations. First, data on infections were limited to inpatient hospitalizations before 1995. Consequently, milder infections treated in primary care settings or untreated infections that did not require hospital admission were not included during this period. The inclusion of all outpatient visits from 1995 onward might render the risk estimates as conservative because more visits for less severe infections are included. Future research should investigate infections among inpatients only, and hospital admissions with infections in early childhood and their effect on outcomes should also be investigated, which was beyond the scope of the present study. Second, we cannot determine whether the hospital treatment itself or disability due to severe infection might explain some of the observed risks, although the possibility is minimized by the fact that elevated risks were also noted years after the last hospitalization. There is also the possibility of mediation, in which risk factors for suicide, such as depression, may be associated with self-care issues and thereby linked to the incidence and severity of infections. Third, we did not have information on hospitalizations before January 1, 1977. However, a subcohort born after 1962 with full adult lifetime follow-up in all registers supported the association between infection and suicide. Fourth, there is a dearth of knowledge about suicide risk among minority groups in the Danish society, although persons with a foreign background were still included in the analyses. Fifth, residual confounding by variables not captured in the registers, such as undiagnosed infections or psychiatric disorders, cannot be excluded, and it is possible that the risk estimates might be influenced by circumstances for which we did not or were unable to adjust for. It would also have been preferable to adjust for previous self-harm in individuals and their relatives. However, this event is underrecorded in Danish hospital data and could potentially lead to bias in the results.⁴⁷ Likewise, underrecording of suicides is possible, although the registration of suicides has been found to be reliable in Denmark.⁴⁸

Risk of suicide was observed in both prospective and dose-response relationships. Our findings indicate that infections may have a relevant role in the pathophysiological mechanisms of suicidal behavior. Provided that the association between infection and the risk of death by suicide was causal, identification and early treatment of infections could be explored as a public health measure for prevention of suicide. Still, further efforts are needed to clarify the exact mechanisms by which infection influences human behavior and risk of suicide.

We found an increased risk of death by suicide among individuals hospitalized with infection. The association remained after adjusting for a range of possible confounders.

Accepted for Publication: May 23, 2016.

Corresponding Author: Helene Lund-Sørensen, BM, Mental Health Centre Copenhagen, Kildegårdsvej 28, DK-2900 Hellerup, Capital Region of Denmark, Denmark (helene.lund@sund.ku.dk).

Published Online: August 10, 2016. doi:10.1001/jamapsychiatry.2016.1594.

Author Contributions: Ms Lund-Sørensen and Dr Erlangsen had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Benros, Madsen, Sørensen, Nordentoft, Erlangsen.

Acquisition, analysis, or interpretation of data: Lund-Sørensen, Benros, Madsen, Sørensen, Eaton, Postolache, Erlangsen.

Drafting of the manuscript: Lund-Sørensen, Benros, Sørensen.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Lund-Sørensen, Madsen, Nordentoft, Erlangsen.

Obtained funding: Lund-Sørensen, Benros, Nordentoft, Erlangsen.

Administrative, technical, or material support: Lund-Sørensen, Benros.

Study supervision: Benros, Madsen, Postolache, Nordentoft, Erlangsen.

Conflict of Interest Disclosures: None reported.

Funding/Support: This project was funded with a unrestricted scholarship grant from the Lundbeck Foundation. Dr Postolache’s contributions were also supported by a Distinguished Investigator Award from the American Foundation for Suicide Prevention.

Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.