What is the association between medical contact for traumatic brain injury (TBI) and risk of suicide?
In this registry-based retrospective cohort study from Denmark that included 34 529 deaths by suicide over 35 years, individuals with medical contact for traumatic brain injury, compared with the general population without traumatic brain injury, had an increased risk of suicide, incident rate ratio of 1.90.
Traumatic brain injury may be associated with increased risk of suicide.
Traumatic brain injuries (TBIs) can have serious long-term consequences, including psychiatric disorders. However, few studies have assessed the association between TBI and risk of suicide.
To examine the association between TBI and subsequent suicide.
Design, Setting, and Participants
Retrospective cohort study using nationwide registers covering 7 418 391 individuals (≥10 years) living in Denmark (1980-2014) with 164 265 624 person-years’ follow-up; 567 823 (7.6%) had a medical contact for TBI. Data were analyzed using Poisson regression adjusted for relevant covariates, including fractures not involving the skull, psychiatric diagnoses, and deliberate self-harm.
Medical contacts for TBI recorded in the National Patient Register (1977-2014) as mild TBI (concussion), skull fracture without documented TBI, and severe TBI (head injuries with evidence of structural brain injury).
Main Outcomes and Measures
Suicide recorded in the Danish Cause of Death register until December 31, 2014.
Of 34 529 individuals who died by suicide (mean age, 52 years [SD, 18 years]; 32.7% women; absolute rate 21 per 100 000 person-years [95% CI, 20.8-21.2]), 3536 (10.2%) had medical contact: 2701 with mild TBI, 174 with skull fracture without documented TBI, and 661 with severe TBI. The absolute suicide rate was 41 per 100 000 person-years (95% CI, 39.2-41.9) among those with TBI vs 20 per 100 000 person-years (95% CI, 19.7-20.1) among those with no diagnosis of TBI. The adjusted incidence rate ratio (IRR) was 1.90 (95% CI, 1.83-1.97). Compared with those without TBI, severe TBI (absolute rate, 50.8 per 100 000 person-years; 95% CI, 46.9-54.6) was associated with an IRR of 2.38 (95% CI, 2.20-2.58), whereas mild TBI (absolute rate, 38.6 per 100 000 person-years; 95% CI, 37.1-40.0), and skull fracture without documented TBI (absolute rate, 42.4 per 100 000 person-years; 95% CI, 36.1-48.7) had an IRR of 1.81 (95% CI, 1.74-1.88) and an IRR of 2.01 (95% CI, 1.73-2.34), respectively. Suicide risk was associated with number of medical contacts for TBI compared with those with no TBI contacts: 1 TBI contact, absolute rate, 34.3 per 100 000 person-years (95% CI, 33.0-35.7; IRR, 1.75; 95% CI, 1.68-1.83); 2 TBI contacts, absolute rate, 59.8 per 100 000 person-years (95% CI, 55.1-64.6; IRR, 2.31; 95% CI, 2.13-2.51); and 3 or more TBI contacts, absolute rate, 90.6 per 100 000 person-years (95% CI, 82.3-98.9; IRR, 2.59; 95% CI, 2.35-2.85; all P < .001 for the IRR’s). Compared with the general population, temporal proximity since the last medical contact for TBI was associated with risk of suicide (P<.001), with an IRR of 3.67 (95% CI, 3.33-4.04) within the first 6 months and an incidence IRR of 1.76 (95% CI, 1.67-1.86) after 7 years.
Conclusions and Relevance
In this nationwide registry-based retrospective cohort study individuals with medical contact for TBI, compared with the general population without TBI, had increased suicide risk.
Individuals with a history of traumatic brain injury (TBI) have been shown to have higher rates of nonfatal deliberate self-harm, suicide, and all-cause mortality than members of the general population.1-3 Individuals with TBI may experience significant physical, cognitive, and emotional symptoms that place them at higher risk of suicide.4 A recent systematic review3 supported this premise by reporting an association of increased risk of suicide among TBI survivors compared with individuals with no TBI. However, most previous studies examining the relation of TBI and completed suicide have been limited by methodological shortcomings, such as small sample sizes, in particular very low numbers of suicide cases with TBI (maximum of 105 cases in studies that also included a control population).5-12 These factors have compromised more detailed analyses, for instance with respect to severity, timely relation, and control of confounding variables. Several studies2,13 have used standardized mortality rate (SMR) calculations based on governmental released age- and sex-mortality rates to estimate the risk of suicide associated with TBI. These estimates have varied widely, with reported SMRs ranging from 0.82 to more than 4, and they have lacked appropriate confounding control.2,7,8,10,14 Recently, a Swedish register-based study reported a 3-fold higher risk of suicide in patients with TBI than in the age- and sex-matched general population; however, this estimate was not adjusted for important risk factors related to both TBI and suicide, such as preexisting psychiatric illness and nonfatal deliberate self-harm.15
The primary objective of this retrospective cohort study was to examine the association between TBI and subsequent suicide.
An anonymized data set was used for research purposes, and the project was approved by the Danish Data Protection Agency (journal number 2012-58-0004). Hence according to Danish legislation, informed consent from participants was not required.
All individuals who were alive and living in Denmark during the study period were included in our analyses. In total, the cohort comprised 7 418 390 individuals aged 10 years or older from January 1, 1980, who were followed up until their dates of death or emigration from Denmark or December 31, 2014, whichever came first. By using the unique personal identification number assigned to each person in Denmark, linkage of data between various national registries was possible. We retrieved data from the Danish Civil Registration System,16 the Database for Integrated Labour Market Research,17 the National Hospital Register,18 the Psychiatric Central Research Register,19 and the Cause of Death Register.20 All registers have full national coverage and contain continuously updated administrative data on all residents living in Denmark. Diagnoses in the National Hospital Register18 and the Psychiatric Central Research Register19 were recorded according to the diagnostic system of the International Classification of Diseases, Eighth Revision (ICD-8) until January 1, 1994; after that, ICD-10 codes were used. Private psychiatric hospital treatment does not exist in Denmark; however, about 1% of somatic hospital beds are located in private hospitals.21
Traumatic brain injury was recorded in the National Patient Register, which covers 3 different forms of medical contacts. Since 1977, TBI was recorded for inpatient treatment and from 1995, contacts on outpatient visits and visits to emergency units were included in the registers. If a patient was registered twice within a month for a TBI, this was considered to be a recording of the same event. Traumatic brain injury was categorized into the following types: (1) mild TBI (defined as concussion), (2) skull fracture without documented TBI, and (3) severe TBI (head injuries with evidence of structural brain injury) (see the ICD-codes in eTable 1 in the Supplement). This categorization is based on the definition given by the American Congress of Rehabilitation Medicine and has been used in prior population-based studies of TBI.22,23 In the analyses of TBI severity, these categories were mutually exclusive and individuals were categorized according to severity, ranging from mild TBI through skull fracture without documented TBI to severe TBI.
To assess the association between TBI and suicide in greater depth, we included the following covariates: number of medical contacts for likely distinct TBI events (0, 1, 2 or ≥3), accumulated number of days in hospital treatment for TBI (<1, same day discharge; 1; 2; 3; 4; 5; 6-14; and ≥15 days), age at first TBI (0-10, >11-15, >16-20, >21-40, >41-60, and ≥61 years), and time since last medical contact for TBI (0-6 months, >6-12 months, >1-2 years, >2-3 years, >3-4 years, >4-5 years, >5-6 years, >6-7 years, ≥7 years since discharge).
From the Cause of Death Register,20 we retrieved the outcome measure on death by suicide (ICD-8 codes E950-E959 or ICD-10 codes X60-X84, Y87).
We also obtained diagnoses of fractures not involving the skull or spine (non–central nervous system [CNS]–related fractures), to explore if fractures that occur to the head vs other fractures were associated with a higher risk of suicide and to adjust for this indicator of other injuries. From the National Patient Register, diagnoses of epilepsy were also included because people with epilepsy might have a higher frequency of TBI and psychiatric disorders.24-26 In addition, estimates were adjusted for long-term physical diseases using the Charlson comorbidity index27 (eTable 1 in the Supplement).
Psychiatric illness and nonfatal deliberate self-harm are associated with suicide,28 so data on contacts for these issues were retrieved from the Psychiatric Central Research Register,19 including all diagnoses given during hospital contact to inpatients since 1969 while outpatient and emergency department contacts were added in 1995 (eTable 1 in the Supplement). Contacts due to deliberate self-harm were identified using ICD-8 codes 950-959 or ICD-10 codes X60-X84 or when the reason for contact was indicated as deliberate self-harm using both somatic and psychiatric hospital registries. In addition, patient contacts at a psychiatric hospital where a diagnosis of injury or poisoning (ICD-10 codes S51, S55, S59, S61, S65, S69, T36-T50, or T52-T60) had been recorded and no record of a substance misuse disorder existed (ICD-10 codes F11-F19) were considered to be deliberate self-harm.
The Danish Civil Registration System16 and the Database for Integrated Labour Market Research17 include data on sex, age, and marital status (never married, married or registered partnership, divorced, widowed, or unknown), cohabitation status (cohabitation, no cohabitation), educational level (elementary school, vocational training, high school, university degree, ongoing or missing), and socioeconomic status (working, unemployed, disability pension, early retirement, student, and other or missing).
Incidence rate ratios (IRRs) were estimated using adjusted Poisson regression models and time-varying variables (all variables were time-varying except for sex) using SAS (SAS Institute Inc; version 9.4). This method approximates Cox regression.29 All incidence IRRs provided in the Tables represents results of between group tests of difference. First, our primary outcome was to estimate the risk of suicide among individuals diagnosed and discharged alive with TBI relative to individuals without head injuries. The basic model was adjusted for sex, age, and calendar period. In the fully adjusted models, we further adjusted for marital status, cohabitation status, socioeconomic status, other injuries, epilepsy, the Charlson comorbidity index (0, 1, 2, 3, 4, 5, or ≥6 chronic disorders),18,27 “pre-TBI psychiatric disorders,” ie, psychiatric disorders diagnosed before any medical contact for TBI, and equivalently “pre-TBI deliberate self-harm.” Second, in a range of fully adjusted models, we examined how suicide was associated with different measures of TBI, such as (1) TBI severity, (2) number of TBI contacts, (3) days in treatment for TBI, (4) age at first TBI, (5) injury type, and (6) time since last TBI. We also tested the following covariates for trend: number of medical contacts for likely distinct TBI events, accumulated number of days in hospital treatment for TBI, and time since last medical contact for TBI. Test of trend was 2-sided and performed with the Cochran-Armitage test.
No adjustment was carried out for post-TBI psychiatric disorders and post-TBI deliberate self-harm because these potentially may act as mediators between TBI and suicide. To explore this further, we estimated incidence IRRs of the risk of suicide associated with before and after TBI psychiatric diagnosis and deliberate self-harm, respectively, in analyses confined to patients with TBI. In addition, multiplicative interaction analyses were carried out between TBI and pre-TBI psychiatric disorders and pre-TBI deliberate self-harm, respectively. In eTables 2 and 3 in the Supplement, the results of test of interaction terms are provided in the footnotes.
We also tested the overall association between TBI and suicide in a sub-cohort of individuals 18 years or more old born after 1962 to validate the association among individuals with full register follow-up as well as in a subcohort excluding those who received a TBI diagnosis in a deliberate self-harm episode.
Overall, the level of statistical significance was P <.05, and tests were 2-sided. To minimize type I errors due to multiple testing, all presented P values are Bonferroni corrected with a factor 53 equal to the total number of carried out tests, and statistically significant estimates (with a P <.00095 after Bonferroni correction) were noted in the Tables.
Of the 7 418 391 living residents of Denmark during the 1980-2014 follow-up period (observed for a total of 164 265 624 person-years; Table 1), 567 823 had received a diagnosis of TBI (mean age at first TBI, 34.3 years [SD, 23.6 years], 41% women). Of the total population, 423 502 individuals (5.7%) were diagnosed with a mild TBI, 24 221 (0.3%) with skull fracture, and 120 100 (1.6%) with severe TBI. In all, 34 529 individuals died by suicide, mean age of 52 years (23 238 men; 11 291 women) and an overall absolute rate of 21.0 per 100 000 person-years (95% CI, 20.8-21.2). Among the 34 529 suicides, 3536 (10.2%) had previously been diagnosed with TBI (2578 men; 958 women), including 2701 with mild TBI, 174 with skull fracture without documented TBI, and 661 with severe TBI (Table 2).
The absolute rate of suicide in individuals with hospital contact for TBI was 40.6 per 100 000 person-years (95% CI, 39.2-41.9) compared with 19.9 per 100 000 person-years (95% CI, 19.7-20.1) in those with no hospital contact for TBI, for a difference of 20.7 per 100 000 person-years (95% CI, 19.3-22.1). The IRR was 2.64 (95% CI, 2.55-2.74) in the model adjusted for sex, age, and calendar period and was 1.90 (95% CI, 1.83-1.97) in the fully adjusted model. Furthermore, the fully adjusted analyses showed an increased risk of suicide by TBI severity: the absolute rate for mild TBI was 38.6 per 100 000 person-years (95% CI, 37.1-40.0) with an IRR of 1.81 (95% CI, 1.74-1.88); 42.4 per 100 000 person-years, skull fracture (95% CI, 36.1-48.7) with an IRR of 2.01 (95% CI, 1.73-2.34, P < .001), and 50.8 per 100 000 person-years, severe TBI (95% CI, 46.9-54.6) with an IRR of 2.38 (95% CI, 2.20-2.58, P < .001) compared with individuals with no medical contact for TBI (Table 2). Individuals with a severe TBI also had a higher risk of suicide than individuals with a mild TBI (between-group difference IRR, 1.32; 95% CI, 1.21-1.44; P < .001) but not significantly different compared with those who had a skull fracture (IRR, 1.18; 95% CI, 1.00-1.40; P = .048). There was no significant difference in suicide risk between those with a mild TBI and a skull fracture (IRR, 1.11; 95% CI, 0.96-1.30; P = .17).
A higher suicide rate was noted with increasing number of medical contacts for likely distinct TBI events (test for trend, P < .001); compared with those with no TBI contact, 1 contact was associated with an IRR of 1.75 (95% CI, 1.68-1.83, P < .001) whereas 2 contacts had an IRR of 2.31 (95% CI, 2.13-2.51), and 3 or more contacts had an IRR of 2.59 (95% CI, 2.35-2.85), ie, not significantly different from 2 contacts (P ≥.99) (Figure, A). Moreover, a higher suicide frequency was noted relative to increasing number of days in treatment for TBI (test for trend, P < .001). As seen in Figure, B, the IRR for individuals hospitalized at least 1 day was 1.78 (95% CI, 1.69-1.88), whereas IRRs of more than 2 were noted for those whose treatment had lasted at least 3 days compared with individuals with no medical contact for TBI (see eTable 4 in the Supplement). Temporal proximity since last medical contact for TBI was associated with risk of suicide (test for trend, P < .001), with an IRR of 3.67 (95% CI, 3.33-4.04) the first 6 months and 1.76 (95% CI, 1.67-1.86) after 7 years compared with the background population. The risk of suicide with in the first 6 months after the TBI incident was also significantly higher (test of between group difference IRR, 2.10 (95% CI, 1.89-2.34; P < .001) compared with more than 7 years after a TBI.
In Figure, D, the results presented were not adjusted for non–CNS-related fractures, and it shows that those with TBI had an IRR of 2.00 (95% CI, 1.93-2.08; P < .001) for suicide after a medical contact for TBI but also that those with fractures not involving the skull or spine had a higher rate of suicide with an IRR of 1.15 (95% CI, 1.12-1.19) than did the background population. Compared with individuals with a non–CNS-related fracture, those with a TBI had a significantly higher risk of suicide (IRR, 1.73; 95% CI, 1.66-1.81; P < .001).
The fully adjusted analyses showed that suicide rates were significantly elevated for all age groups of first TBI compared with individuals without TBI (Table 3). Those who had a first medical contact and were between the ages of 16 and 20 years had the highest suicide risk (IRR, 3.01; 95% CI, 2.74-3.30) compared with individuals with no TBI and also were at significantly higher risk of suicide than those experiencing TBI in all other age-groups (test of between group difference IRR, 1.65; 95% CI, 1.50-1.82; P < .001).
Individuals who were diagnosed with a psychological illness after their TBI had a higher risk of suicide (IRR, 4.90; 95% CI, 4.55-5.29; P < .001) than did those with TBI only, as were those who had engaged in deliberate self-harm after experiencing their TBI (IRR, 7.54; 95% CI, 6.91-8.22; P < .001; Table 4). Likewise, individuals who had been diagnosed with a psychological illness before their TBI had a higher risk of suicide (IRR, 2.32; 95% CI, 2.10-2.55; P < .001) than did those with TBI only, as were those who had engaged in deliberate self-harm before experiencing their TBI (IRR, 2.85; 95% CI, 2.53-3.19; P < .001). Analyses of interaction showed a negative association between TBI and prior psychiatric diagnosis or prior deliberate self-harm, thus among individuals who had both preexisting psychological illnesses or had engaged in self-harm prior to their experiencing a TBI were at lower risk of suicide than those who had psychological illness or had engaged in deliberate self-harm but who did not experience a TBI (P < .001 for interaction terms; eTable 2 and eTable 3 in the Supplement).
The sensitivity analyses including only individuals with full register follow-up born after 1962 supported the overall association between TBI and suicide in the younger population, with a slightly higher association between TBI and suicide (IRR, 2.42; 95% CI, 2.25-2.59; eTable 5 in the Supplement). The association between TBI and suicide in analyses excluding individuals who received a TBI diagnosis as a result of a deliberate self-harm episode yielded an IRR of 1.88 (95% CI, 1.81-1.95).
In this registry-based, retrospective, cohort study involving all Denmark residents, those with medical contact for TBI compared with the general population without TBI had an increased risk of suicide. Additional analyses revealed that the risk of suicide was higher for individuals with severe TBI, numerous medical contacts, and longer hospital stays. Analysis further showed that these individuals were at highest risk in the first 6 months after discharge. The association between TBI and suicide is likely to be partly mediated by post-TBI psychiatric symptoms because the risk of suicide among those who developed a psychiatric diagnosis or engaged in deliberate self-harm after a TBI diagnosis was higher than among individuals with only a TBI diagnosis. Traumatic brain injury constitutes a major public health problem with many serious consequences; furthermore, medical contact due to TBI had occurred prior to 10.2% of suicides. The absolute suicide rate in Denmark was 21 per 100 000 person-years in the 1980-2014 period, but it was almost twice as high among individuals with TBI, 41 per 100 000 person-years.
This study reports a lower difference between those with TBI and the general population than what has been reported in other studies, which in most cases have described this relative difference (although expressed as odds ratios, hazard rates, or in SMRs) to be somewhat higher than 2-fold.2,10,11,13-15 Some of the previously reported estimates were only adjusted for sex and age, although a few were also adjusted for race, income, or marital status, and these resemble the basic adjusted findings in this study (Table 2), for which an incidence RR of 2.64 (95% CI, 2.55-2.74) was obtained. Only 2 previous studies11,12 have reported estimates that were adjusted for pre-TBI psychiatric diagnoses; however, both studies were based on selected subgroups, ie, children11 and military veterans (90% male population)12 and were limited by small numbers of TBI-related suicide cases. As such no other studies, to our knowledge, have previously provided adjusted estimates for important confounders, such as pre-TBI psychiatric diagnosis, epilepsy, other fractures, a range of somatic comorbidity, and pre-TBI deliberate self-harm, which decreases the association between TBI and suicide. Still, a significantly higher suicide rate after medical contact for TBI was found. Furthermore, some findings in our study affirmed those reported in previous studies, such as suicide risk being associated with the number of medical contacts for TBI,6,9 an increased rate with increased TBI severity,2 and a higher suicide rate among individuals who experience a first TBI in young adulthood.2,11 Moreover, the risk of suicide was substantially higher after TBI than after non–CNS-related fractures, indicating that the association between TBI and suicide was not merely due to injury proneness. Furthermore, the interaction analyses indicating that in individuals with a pre-TBI history of either a psychiatric diagnosis or an engagement of deliberate self-harm, a TBI was associated with a lower risk of suicide than among those who only had a psychiatric diagnosis or engaged in deliberate self-harm. This seems paradoxical and might be due to increased medical attention after the TBI or possible TBI induced initiative apathy among those who in addition to a psychiatric history or deliberate self-harm experience a TBI, reducing suicide events that otherwise would have occurred.
Traumatic brain injury is a major public health problem that has many serious consequences, including suicide. The high prevalence of TBI globally emphasizes the importance for preventing TBI in order to ameliorate its sequelae, such as increased suicide risk, which can be prevented resulting in saved lives. Falls or road traffic accidents30 account for the largest share of TBIs. Helmet use has a protective effect, especially falls related to bicycling31,32 and falls that occur at work.33
This study has several strengths. First, this is a large-scale cohort study that included 7 418 391 individuals, 34 529 suicides, and 35 years of follow-up. It compared individual-level data in the analyses that were adjusted for time-varying important and well-known risk factors of suicide. Second, it included only suicide death as an outcome, not including uncertain deaths like in other previous studies,15 and classification of suicide in the Danish Cause of Death Register has recently been found to be very reliable.34
This study also has several limitations. First, before 1995 medical outpatient contacts were not registered; thus, mild TBI incidents were treated in medical outpatient settings before 1995 and were not counted as individuals with TBIs, which may bias the estimates in a conservative direction. Second, no information on what treatment patients with TBIs received was available, which would have been useful to estimate whether different treatment regimens or subsequent follow-up would have reduced the suicide risk. Third, this study analyzed the number of medical contacts without further distinction among the 3 subtypes of medical encounter (hospitalization, emergency department, or outpatient); however, we evaluated the days in treatment as a measure of severity. Fourth, some individuals may not seek medical treatment after experiencing a mild TBI or for mild psychiatric disorders or for deliberate self-harm, consequently this would be unregistered and result in misclassification that bias estimates. Nevertheless, the risk of suicide was more increased after severe TBI than after mild TBI and also increased with the severity of the TBI when measured by the length of hospitalization for TBI, which might be a more accurate measure of TBI severity. Fifth, the National Patient Register has registered inpatient contacts since 1977; therefore, some individuals may have entered the study cohort with a pre-1977 incident of TBI, which may result in an underestimation of the risk estimate particularly in the elderly. Nevertheless, sensitivity analyses including only individuals with lifetime full registry data follow-up support the overall results.
In this nationwide registry-based retrospective cohort study, individuals with medical contact for TBI, compared with the general population without TBI, had an increased risk of suicide.
Corresponding Author: Trine Madsen, PhD, Danish Research Institute of Suicide Prevention, Mental Health Centre Copenhagen, Capital Region of Denmark, Kildegårdsvej 28; DK-2900 Hellerup, Denmark (firstname.lastname@example.org).
Accepted for Publication: June 25, 2018.
Author Contributions: Drs Madsen and 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.
Concept and design: All authors
Acquisition, analysis, or interpretation of data: Madsen, Erlangsen, Nordentoft, Benros.
Drafting of the manuscript: Madsen, Orlovska, Modaffy, Benros.
Critical revision of the manuscript for important intellectual content: Madsen, Erlangsen, Nordentoft, Benros.
Statistical analysis: Madsen, Erlangsen, Nordentoft, Benros.
Obtained funding: Nordentoft.
Administrative, technical, or material support: Orlovska, Modaffy, Nordentoft, Benros.
Supervision: Erlangsen, Nordentoft, Benros.
Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Dr Erlangsen reported receiving research grants from the Lundbeck Foundation. No other disclosures were reported.
Funding/Support: The study was partly funded by the Mental Health Services Capital Region Denmark and an unrestricted grant from the Lundbeck Foundation.
Role of the Funder/Sponsor: The funders of the study had no role in design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, or approval of the manuscript or the decision to submit for publication.
AL. Suicidal ideation and behaviours after traumatic brain injury: a systematic review. Brain Impair
. 2013;14(1):92-112. doi:10.1017/BrImp.2013.11Google ScholarCrossref
RL. Preventing suicide after traumatic brain injury: implications for general practice. Med J Aust
. 2007;187(4):229-232.PubMedGoogle Scholar
P. Traumatic brain injury, psychiatric disorders and suicide: a population-based study of suicide victims during the years 1988-2004 in Northern Finland. Brain Inj
. 2007;21(8):851-855. doi:10.1080/02699050701504265PubMedGoogle ScholarCrossref
An Integrated Database for Labour Market Research [in Danish]. Copenhagen, Denmark: Danmarks Statistik Printing; 1991.
M. Statistical Models in Epidemiology. Oxford, United Kingdom: Oxford University Press USA-OSO; 2013.