Psychiatric Illness Following Traumatic Brain Injury in an Adult HealthMaintenance Organization Population

Background  Psychiatric illness after traumatic brain injury (TBI) has been shown to be prevalent in hospitalized and tertiary care patient populations.

Objective  To determine the risk of psychiatric illness after TBI in an adult health maintenance organization population.

Design  Prospective cohort study.

Setting  Large staff-model health maintenance organization.

Participants  Nine hundred thirty-nine health plan members diagnosed as having TBI in 1993 and enrolled in the prior year, during which no TBI was ascertained. Three health plan members per TBI-exposed subject were randomly selected as unexposed comparisons, matched for age, sex, and reference date.

Main Outcome Measure  Psychiatric illness in the 3 years after the TBI reference date, determined using computerized records of psychiatric diagnoses according to the International Classification of Diseases, Ninth Revision, Clinical Modification, prescriptions, and service utilization.

Results  Prevalence of any psychiatric illness in the first year was 49% following moderate to severe TBI, 34% following mild TBI, and 18% in the comparison group. Among subjects without psychiatric illness in the prior year, the adjusted relative risk for any psychiatric illness in the 6 months following moderate to severe TBI was 4.0 (95% confidence interval [CI], 2.4-6.8) and following mild TBI was 2.8 (95% CI, 2.1-3.7; P<.001) compared with those without TBI. Among subjects with prior psychiatric illness, the adjusted relative risk for any psychiatric illness in the 6 months following moderate to severe TBI was 2.1 (95% CI, 1.3-3.3) and following mild TBI was 1.6 (95% CI, 1.2-2.0; P = .005). Prior psychiatric illness significantly modified the relationship between TBI and subsequent psychiatric illness (P = .04) and was a significant predictor (P<.001). Persons with mild TBI and prior psychiatric illness had evidence of persisting psychiatric illness.

Conclusions  Both moderate to severe and mild TBI are associated with an increased risk of subsequent psychiatric illness. Whereas moderate to severe TBI is associated with a higher initial risk, mild TBI may be associated with persistent psychiatric illness.

High rates of mood, psychotic, and substance abuse disorders followingtraumatic brain injury (TBI) have been found in hospitalized trauma and tertiarycare referral populations.^1-5 Apopulation-based secondary analysis of the New Haven portion of the NationalInstitute of Mental Health Epidemiologic Catchment Area Study⁶ bySilver and colleagues examined patients at varying time points after self-reported"severe head injury that was associated with a loss of consciousness or confusion,"^7(p937) making the TBI exposure susceptibleto recall bias. The study found that adjusted odds ratios (ORs) for all psychiatricdiagnoses except bipolar disorder were increased in the group reporting ahistory of head injury compared with the group reporting no history of headinjury. Data on injury severity and time since injury were not available.Few data exist on the longitudinal risk of psychiatric disorders in large,population-based ambulatory primary care settings following the entire spectrumof TBI severity, particularly mild TBI.

Studies have confirmed that disability caused by psychiatric disordersmay contribute to the disability associated with TBI.^2,8 Identifyingthe extent of psychiatric problems following TBI, particularly mild TBI, mayassist in targeting secondary and tertiary prevention efforts for TBI-relateddisability.

Psychiatric history has been shown to be a risk factor for post-TBIpsychiatric illness in some but not all studies.^1,9-11 Similarly,the relationship between TBI severity and prevalence of subsequent psychiatricdisorders has been inconsistent.^2,10-14 Theeffects of psychiatric history and TBI severity on risk of psychiatric problemsfollowing TBI have not been systematically and longitudinally studied in largeTBI and comparison samples.

The goals of this study were to determine the prevalence of psychiatricillness following mild as well as moderate to severe TBI in an adult healthmaintenance organization (HMO) population, the ORs for prevalent psychiatricillness and relative risks (RRs) for incident psychiatric illness followingTBI compared with patients without TBI, and the effect of prior psychiatricillness on the relationship between TBI and subsequent psychiatric illness.

This prospective cohort study was conducted using computerized recordsfrom the Group Health Cooperative of Puget Sound (GHC), a consumer-governedregional HMO that serves approximately 450000 members in the Puget Sound areaof western Washington State. As a staff-model HMO, it builds and owns itsfacilities, and the physicians and other staff are employees of the healthplan. The GHC population is broadly representative of the greater Seattlepopulation in terms of age, sex, race, and marital status; GHC enrollees haveslightly higher educational attainment and less representation at the highend of income distribution.¹⁵ This study useddata obtained in conjunction with a previously reported case-control studyof psychiatric risk factors for TBI.¹⁶ Subjectsreceived medical care from one of GHC's facilities, located in 6 counties.Data on health plan members were derived from GHC's computerized databases,which included information on all inpatient and outpatient visits and diagnoses,all prescriptions dispensed from GHC pharmacies, age, sex, and insurance type(Medicare, Medicaid, GHC individual or family plan, GHC group plan, or otherplan), which was used as a proxy for socioeconomic status. Each GHC memberhas a unique and permanent number that makes linkage of all utilization possible;GHC tracks enrollment closely and has a low rate of member disenrollment (13.1%for 1992 through 1993), which made it possible for us to maintain a stablestudy base. Diagnoses in 1992 were recorded on 95% of all visit records.¹⁵ Only about 7% of GHC members had dual insurance coverage,so ascertainment of utilization was nearly complete. The study was approvedby the institutional review boards of GHC and the University of Washington,Seattle.

The TBI-exposed group included patients 15 years or older diagnosedat an emergency department, hospital, or outpatient clinic as having a TBIin 1993. The following diagnostic categories and codes from the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM)¹⁷ wereused to identify TBI: fracture of the vault or base of the skull (800.0-801.9);other, unqualified, and multiple fractures of the skull (803.0-804.9); andintracranial injury, including concussion, contusion, laceration, and hemorrhage(850.0-854.1). These diagnostic categories for TBI were used by the Centersfor Disease Control and Prevention in their TBI surveillance studies.¹⁸ If a person received a TBI diagnosis at more than1 visit in 1993, the first diagnosis was considered the incident TBI. Thereference date for cases was the date of this incident TBI diagnosis. Patientswith TBI were required to have been continuously enrolled at GHC for the yearprior to their TBI diagnosis to ensure that records were available to assessindicators of prior psychiatric illness. We included patients from 6 countiesin the Puget Sound area that provided complete health service utilizationdata. To maximize the likelihood of ascertaining incident TBI cases, subjectswho had an ICD-9-CM diagnosis of TBI in the yearprior to their reference date were excluded from the TBI-exposed group.

Severity of TBI was dichotomized into mild TBI and moderate to severeTBI using the categorization criteria of the Centers for Disease Control andPrevention.¹⁹ The TBI exposure was consideredto be mild if ICD-9-CM codes indicated brief (<1hour) or no loss of consciousness and no documented traumatic intracraniallesions. The TBI exposure was considered to be moderate to severe if ICD-9-CM codes indicated prolonged loss of consciousnessor a documented traumatic intracranial or brain lesion. Persons whose TBIseverity was undetermined were excluded from the study.

To compare the effects of recent incident TBI exposure with no recentTBI exposure, 3 subjects per TBI-exposed patient were selected at random fromGHC enrollment files and were frequency matched with TBI-exposed patientsby sex, age in 5-year groups (15-19 years up to ≥95 years), and enrollmentat the time of the TBI-exposed patient's reference date. Similar to the patientswith TBI, TBI-unexposed patients had to be GHC members on their assigned referencedate as well as continuously during the year prior to this date and couldnot have received a TBI diagnosis during that year.

Psychiatric illnesses in the year prior to and 3 years following thereference date were ascertained using 3 separate indicators: presence of apsychiatric diagnosis, filling of a prescription for psychiatric medication,or utilization of psychiatric services. Presence of a psychiatric illnesswas recorded in 6-month blocks for the 3 years following the reference date.The denominator for each period was the number of subjects who were enrolledat any time during that period. Therefore, except for those who were disenrolledas a result of death on the reference date all patients had psychiatric illnessoutcome data in the first 6 months after the reference date.

Psychiatric diagnoses were determined using ICD-9-CM codes and were categorized as follows: acute reaction to stress oradjustment reaction (308, 309); alcohol or drug intoxication, withdrawal,or dependence (291.0-292.9, 303.0, 303.9, 304, 305); anxiety (300.0, 300.2,300.3, 799.2); depression (296.2, 296.3, 296.82, 296.9, 300.4, 311); hyperkineticsyndrome of childhood (314); malaise or fatigue (300.5, 780.7); organic psychoticmental disorders (290.0-290.9, 293.0-294.9); organic nonpsychotic mental disorders(310, 780.09); schizophrenia, hallucinations, or paranoia (295, 297.0-299.9,780.1); somatoform disorders (300.1, 300.6-300.9, 306, 307.8, 307.89); orother psychiatric disorders (307, 316, V40.2-V40.9, V62.81, V62.89, V65.9).

Subjects were considered to have filled a psychiatric medication prescriptionif automated GHC pharmacy data indicated that a prescription for a psychiatricmedication in any of the following classes was filled in the 3 years afterthe reference date: antidepressants, lithium, anxiolytics, antipsychotics,or psychostimulants. Surveys in 1985 and 1986 showed that more than 90% ofall medications prescribed at GHC were filled in GHC pharmacies, and a studyfound that 97.6% of patients treated with antidepressant medications between1991 and 1992 filled their prescriptions at GHC pharmacies.¹⁵ Becauseantidepressants are commonly used for other conditions, they were consideredto be for a psychiatric indication only if the prescription was filled within60 days of a depression diagnosis. Anxiolytics were considered to be for apsychiatric indication if the prescription was filled within 60 days of ananxiety diagnosis. Psychostimulants were excluded if there was a diagnosisof narcolepsy in the year prior to the reference date.

Utilization of psychiatric services was ascertained based on computerizedrecords of inpatient psychiatric hospitalizations, outpatient mental healthclinic visits, and inpatient stays or outpatient visits for alcohol or drugtreatment in the year before and 3 years after the reference date. Out-of-planmental health care use is rare owing to comprehensive mental health serviceswith small copayments in GHC.

To maximize clinical validity, the psychiatric illness indicators werefurther divided into the following clinical categories: (1) affective disorders(depressive or anxiety disorder diagnosis; antidepressant, lithium, or anxiolyticprescription), (2) psychotic disorders (schizophrenia, hallucinations, orparanoia or organic psychotic disorder diagnosis; antipsychotic prescription),(3) substance abuse disorders (alcohol or drug intoxication, withdrawal, ordependence diagnosis), (4) adjustment reaction (acute reaction to stress,adjustment reaction, malaise, or fatigue), (5) somatoform disorder, (6) organicnonpsychotic mental disorder, (7) hyperactivity (hyperkinetic syndrome ofchildhood; psychostimulant prescription), and (8) other psychiatric disorder.An overall summary determination of psychiatric illness was made based onthe presence of any psychiatric diagnosis, psychiatric medication prescription,or psychiatric utilization during the year before and 3 years after the referencedate.

Comorbid injuries associated with incident TBI were determined using ICD-9-CM codes for fractures (805-829), internal injuries(860-869), open wounds (870-897), crushing injuries (925-929), injury to thenerves or spinal cord (950-957), and other injuries (958-959) during the 6months before and 6 months following the reference date.

To adjust for medical comorbidity, we initially considered the JohnsHopkins Ambulatory Care Group (ACG) Case Mix Adjustment System,²⁰ ameasure of relative health status designed to be indicative of expected resourceconsumption. This system uses age, sex, and ICD-9-CM diagnosesto assign a person to 1 of 106 mutually exclusive categories.²¹ Althoughadjustment with the ACG system may be considered for continuous or large-countoutcomes, its many categories make it less useful for dichotomous outcomes.Consistent with the cost-prediction design goal of the ACG, we used the logarithmof total costs in the year prior to TBI as a proxy for medical comorbidityand compared it with ACG adjustment. With the most prevalent outcomes, inwhich ACG adjustment was possible, logarithm of total costs and ACG adjustmentwere nearly interchangeable as adjustments to the effect of TBI, with logarithmof total costs tending to be a slightly stronger confounder.

The proportion of subjects with psychiatric illness indicators withina particular period (p = prevalence) reveals thecumulative effects of exposure regardless of time of onset. Adjusted exposureeffects are estimated with the OR (odds = p/[1-p]; OR = odds [exposed]/odds [unexposed]). Because of thepotential for correlation among the psychiatric illness indicators for eachsubject, OR estimates were computed using generalized estimating equations.Separate regression analyses were reported for subgroups with and withoutpsychiatric illness in the prior year. We used SAS version 8 statistical software(SAS Institute Inc, Cary, NC) for this analysis.

The proportion of subjects whose first psychiatric illness indicatoroccurs during a particular period (incidence) shows the effect of exposureon new cases. The RR expresses incidence among exposed subjects as a proportionof the incidence among unexposed subjects. The RR estimates were computedusing complementary log-log generalized linear models for fixed-interval,interval-censored data.²² Based on the a priorihypothesis of strong early effects of TBI with decreasing risk thereafter,we modeled the first 6-month RR separately along with a linear trend in RRfor intervals 2 through 6. Because of the small number of substance abuseoutcomes, the ORs and RRs for substance abuse are reported for a combinationof mild TBI exposure and moderate to severe TBI exposure. Inferences are basedon omnibus tests of any effect of TBI and nested tests for trends in RR acrosstime. In addition, descriptive tests and confidence intervals (CIs) for individualperiods are reported for comparison of the relative strength of TBI effectsacross time. Stata version 7 (Stata Corp, College Station, Tex) and S-PLUSversion 6.1 (Insightful Corporation, Seattle) statistical software were usedfor this analysis.

In 1993, 939 GHC members who were enrolled in the 6-county study regionwere diagnosed as having a TBI. These members had been enrolled in GHC forat least 1 year prior to their index TBI and had no evidence of a TBI in thatyear. Most TBIs were classified as mild (n = 803; 85.5%) and were diagnosedin an outpatient setting (n = 410; 43.7%) or in the emergency department (n= 388; 41.3%), whereas 141 (15.0%) were diagnosed in the hospital. The overallannual incidence rate of TBI at GHC among all ages was 475.2 per 100000 person-years.¹⁶

Although 531 (57%) of all TBI-exposed subjects and 494 (62%) of mildTBI–exposed individuals were aged 15 to 44 years, 77 (57%) of thosewith moderate to severe TBI were 65 years or older. There were slightly morewomen (51%) than men. The TBI and comparison groups were similar with regardto insurance type; 74% were group plan subscribers or dependents, 4% receivedMedicare, and less than 1% received Medicaid. The TBI events were distributedevenly across months, although there was a slightly lower proportion in thewinter months.

The prevalence of psychiatric illness categories for the year priorto and the 3 years following the reference date are shown in Table 1. Participants with TBI generally had higher rates of psychiatricillness indicators compared with those without TBI. Among subjects with moderateto severe TBI, 49% had evidence of a psychiatric illness in the year followingTBI compared with 34% in those with mild TBI and 18% in non-TBI comparisons.The prevalence for subjects with TBI gradually dropped during subsequent yearsbut usually remained greater than for the comparison group. Most psychiatricdiagnoses were made by non–mental health professionals, with 48% madeby family practice physicians (46% of TBI-exposed subjects and 50% of comparisons).During the 3 years, 60 subjects (44%) who had moderate to severe TBI, 183(23%) who had mild TBI, and 621 (22%) without TBI were dropouts as a resultof death or GHC disenrollment. Death on the reference date accounted for 47%(n = 28) of all dropouts in the moderate to severe TBI group compared with8% (n = 15) in the mild TBI group and 1% (n = 9) in the comparison group.

Table 2 summarizes the adjustedOR estimates during the 3 years following TBI, stratified by psychiatric illnessin the prior year, for prevalent affective disorders, psychotic disorders,and substance abuse disorders as well as diagnostic, medication prescription,utilization, and any psychiatric illness indicator categories. Time pointsfor which P<.05 are marked with a dagger for comparisonof the strength of effect across time. Adjusted ORs for subjects with priorpsychiatric illness were computed separately so that the TBI effect estimatesand all adjustments accounted for the effect modification due to prior psychiatricillness. For most outcomes, the estimated effects caused by TBI were greateramong those without prior psychiatric illness and were often stronger formild TBI. With a few exceptions, primarily in subjects with prior psychiatricillness who sustained mild TBI, the OR estimates decreased across time.

Among subjects without prior psychiatric illness, the OR estimates forprevalent affective disorder after mild TBI were greatest in the first 2 years.In those with prior psychiatric illness, the estimated ORs for affective disorderwere not individually significant until the second year after mild TBI. Inthe first year after TBI, the estimated OR for psychotic disorder was greatestfor moderate to severe TBI exposure. However, the OR estimate increased insubjects with moderate to severe TBI without a psychiatric history and inthose who had mild TBI with a psychiatric history. Although substance abusewas rare among GHC enrollees, precluding stratification by TBI severity, theOR estimate was highest in the first year among subjects without a psychiatrichistory and in the second year in those with a psychiatric history.

In examining the major psychiatric illness indicators (diagnosis, prescription,utilization, and any indicator), the OR estimates for prevalent psychiatricillness conferred by mild TBI were individually statistically significantat 18 of 24 time points compared with 4 of 24 time points in the moderateto severe cohort. Among subjects with mild TBI without prior psychiatric illness,the estimates tended to decrease; however, in those with prior psychiatricillness, they tended to increase.

Figure 1 shows the predictedcumulative incidence of any psychiatric illness within the TBI exposure groupsfor subjects with and without prior psychiatric illness. As expected, thosewith prior psychiatric illness had a higher incidence of subsequent psychiatricillness. This may explain the smaller OR and RR estimates (Table 3) for the additional effect due to TBI in this subgroup.Subjects who sustained a moderate to severe TBI also developed subsequentpsychiatric illness at a higher rate, particularly in the first 6 months.However, among those without prior psychiatric illness, there seems to bea "catch-up" phenomenon whereby those with no TBI and those with mild TBIclose the gap in proportions at later time points. This may explain OR andRR estimates that are lower than 1 at the later time points.

Adjusted estimates of the RR for incidence of affective, psychotic,and substance abuse disorders and any psychiatric illness indicator are presentedin Table 3. Statistically significanteffects of TBI on incident affective disorders were seen in patients withoutprior psychiatric illness, particularly in the mild TBI cohort. However, RRestimates for incident psychotic disorders were greater for the moderate tosevere TBI cohort compared with subjects who had mild TBI; CIs were largeowing to the rarity of the outcomes. The estimated RR for substance abusewas significant in the group with no prior psychiatric illness, with individualsignificance continuing through month 30. In the category of any psychiatricillness, the estimated increase in risk of subsequent psychiatric illnessconferred by TBI was initially greater in subjects without prior psychiatricillness and in those with moderate to severe TBI compared with mild TBI. However,nominally statistically significant risks were sustained longer in the mildTBI group.

For subjects without prior psychiatric illness, being female (adjustedRR = 1.2; 95% CI, 1.1-1.4) and having higher costs in the prior year (adjustedRR = 1.4; 95% CI, 1.3-1.6, for groups differing by a factor of 10) were associatedwith a higher risk of any psychiatric illness indicator in the 3 years followingTBI. For those with prior psychiatric illness, sex was not a significant riskfactor, but costs were (adjusted RR = 1.3; 95% CI, 1.1-1.6). Comorbid injurieswere not associated with higher risk of psychiatric illness in either psychiatrichistory stratum. Prior psychiatric illness significantly modified the effectof TBI on subsequent psychiatric illness (P = .04)and was also a strong predictor of subsequent psychiatric illness (P<.001). The RRs for any psychiatric illness in the 3-year follow-upperiod for prior vs no prior psychiatric illness were 2.9 (95% CI, 2.4-3.3)for the TBI-unexposed group, 2.5 (95% CI, 2.0-3.1) for the mild TBI group,and 3.0 (95% CI, 1.7-5.0) for the moderate to severe TBI group.

We found that the risk of psychiatric illness, ascertained using severaldifferent indicators, was significantly increased following both mild andmoderate to severe TBI in an adult HMO population. The magnitude and patternof risk for subsequent psychiatric illness was modified by whether patientshad a history of psychiatric illness in the year prior to their TBI. The alreadyhigh risk of subsequent psychiatric illness in those with prior psychiatricillness may have diluted the increase in risk due to TBI, accounting for thesmaller RR estimates in this subgroup. The risk of incident psychiatric illnesswas greatest in the first 6 to 12 months after TBI. In patients without priorpsychiatric illness, decreasing trends were generally evident in RR estimatesacross time. A few studies found that psychiatric history was a significantrisk factor for psychiatric problems after TBI.^1,11

The literature on the relationship between TBI severity and subsequentpsychiatric disorders is inconsistent, with study findings ranging from positiveor inverse to no relationship.^{2,10-14,23,24} Wefound that although survivors of moderate to severe TBI with no prior illnesshad a greater initial elevation in risk of any psychiatric illness, the mildTBI cohort exhibited a more prolonged pattern of elevated risk that eventuallysurpassed that of the moderate to severe cohort. Our findings also suggestthat the TBI-associated RR for incident psychiatric illness is highest shortlyfollowing injury in persons with no psychiatric history, whereas it may remainhigh or even increase in subsequent years in persons with prior illness.

Some psychiatric illnesses appear to persist after TBI. Table 2 shows sustained elevation and a delayed increase in ORsfor certain prevalent psychiatric illnesses, particularly in patients whohave mild TBI with prior psychiatric illness. This population often perplexeshealth care professionals with their persistent symptoms and disability.^25-27 The TBI may precipitatethe expression of underlying psychiatric distress, which may initially appearas nonspecific somatic or postconcussive symptoms such as headache, decreasedmemory, or irritability.^2,28 Ourdata suggest the need to control for medical comorbidity when studying therelationship between TBI and psychiatric illness, although comorbid injuriesdid not confound this relationship.

Affective disorders were common in this HMO population with TBI. Wefound that patients with mild TBI had higher ORs and RRs for affective disordersthan those with moderate to severe TBI. This finding is consistent with variousreports that more severe TBI is not necessarily associated with a higher riskof depression.^{2,10,12,13,24} Moreover,patients with a psychiatric history who sustained mild TBI had higher ORsfor prevalent psychiatric illness in the second and third years after TBIcompared with the first year. This subgroup seems to be at particularly highrisk for persistent psychiatric morbidity.

There was also a pattern of delay for the relatively higher risk ofpsychotic disorder in TBI-exposed subjects with prior psychiatric illness.A psychotic disorder in the first year after moderate to severe TBI may representdelirium due to the TBI or to intensive care unit–related psychosis.However, RR increases in the second and third years in those with prior psychiatricillness, particularly with moderate to severe TBI, are consistent with findingsfrom other reports of delayed psychosis after TBI.^3,29

Substance abuse disorders have been found to initially decrease aftermoderate to severe TBI, but there may be a delayed increase after the firstyear.^4,30 Our OR data replicatedthis finding in patients with prior psychiatric illness, although those withno prior psychiatric illness had an initial increase that decreased with time.Relative risk estimates of incident substance abuse consistently decreasedacross time for both groups. The different OR patterns of substance abusebetween those with and without prior psychiatric illness may be related toan initial heightened emphasis on alcohol abstinence among health care professionalstoward prior substance abusers. Nearly all substance abusers with TBI hadmild TBI.

From our findings, we hypothesize that psychiatric symptoms arisingimmediately after TBI may be etiologically related to the neurophysiologicaleffects of the injury, as supported by an early relationship between TBI severityand psychiatric risk. However, other factors such as psychological vulnerability(eg, personality and attribution styles), self-awareness of deficits, socialinfluences, and secondary gain may play subsequent roles, particularly insubjects with prior psychiatric illness and mild injury. These possible etiologicalcorrelates, which have been proposed by other investigators,^8,23,31,32 likelyinteract in complex ways and require further research.

Psychiatric diagnosis, prescriptions filled, and health care utilizationwere all recorded at the time of medical service. Therefore, possible biasdue to differential recall between TBI-exposed and TBI-unexposed subjectswas eliminated. Because of the completeness and uniformity of the database,the determination of psychiatric illness was not subject to the limitationsof incomplete recall or information bias.³³ Also,because providers at GHC are not influenced by the risk that a psychiatricdiagnosis may not receive full reimbursement, as may be the case in fee-for-servicesettings, reporting bias of psychiatric diagnoses owing to reimbursement rateswas unlikely. Use of a comparison group allowed us to control for baselinefluctuations in psychiatric illness detection; there was a 1% increase peryear in overall psychiatric illness in the comparison group during the 4-yearstudy period, perhaps representing improved detection and coding efforts atGHC.

A limitation of this study is the possible lack of precision in TBIexposure measurement. Our study population had a higher proportion of womenthan other TBI populations; this may have been an ambulatory population withmore women reporting mild TBI. This higher representation of women may haveafforded the statistical power to find the association between female sexand risk of subsequent psychiatric illness. Insurance type was used as a proxymeasure for socioeconomic status. Although it was not shown to be a confounder,there could be residual confounding by socioeconomic status that was not controlledfor in the analyses.

Incident RR estimates that decline lower than 1, particularly for moderateto severe TBI exposure, suggest that TBI may advance the identification ofpsychiatric illness in groups that might have been diagnosed at a later date.The plot for no prior psychiatric illness best depicts this catch-up effect.Because individuals with TBI may visit their physician more frequently, suchas for postconcussive symptoms, it is more likely that a psychiatric illnessin these persons would be recognized and recorded; a person without TBI maynot go to a GHC clinic, and thus a psychiatric illness might not be recognized.Those with TBI would be more likely to have a subsequent TBI,³⁴ furtherincreasing the risk of psychiatric illness; therefore, our RR estimates mayinclude risk associated with subsequent TBIs.

The method used to determine psychiatric illness may have been subjectto misclassification. Psychiatric diagnoses assigned by non–mental healthprofessionals may be less sensitive and specific than those made by mentalhealth professionals. Subjects could have been incorrectly assigned to diagnosticcategories or may not have received a diagnosis when a psychiatric illnesswas present. However, this potential nondifferential misclassification dueto type of specialist would likely bias TBI effect estimates toward the null.

Moderate to severe TBI–exposed subjects were at higher risk fordropout, particularly resulting from higher injury-related fatality, thanthose with mild TBI and comparison subjects. It is possible that losses tofollow-up may have affected the estimates of ORs and RRs, especially for laterperiods. When comparing these results with those in other populations, considerationshould be given to differences in the distributions of disenrollment, death,and exposure severity within categories. The incident RR estimates may havebeen more resistant to disenrollment effects because they relied on incidentvs prevalent psychiatric illness.

Despite these limitations, we found both mild and moderate to severeTBI to be significant risk factors for subsequent psychiatric illness. Evidenceof psychiatric illness in the year prior to TBI significantly modified thisrelationship. Subjects with a psychiatric history had a substantially greaterrisk of post-TBI psychiatric illness. Their RR due to the effect of TBI wassmaller compared with those without prior illness; the already high incidenceamong controls with prior illness left little room for other effects. Theclinical implications of this study include evidence that patients shouldbe screened for psychiatric morbidity within the first 6 to 12 months followingTBI. However, continued vigilance for up to 3 years, particularly in patientswith mild TBI and prior psychiatric illness, is also supported. Health careprofessionals working in primary care settings may have the best opportunityto identify and treat psychiatric morbidity in patients with mild TBI. Itis also important to inquire about psychiatric history before attributingpsychiatric symptoms to TBI.

Corresponding author and reprints: Jesse R. Fann, MD, MPH, Departmentof Psychiatry and Behavioral Sciences, Box 356560, University of Washington,Seattle, WA 98195-6560 (e-mail: fann@u.washington.edu).

Submitted for publication February 20, 2003; final revision receivedJune 25, 2003; accepted July 3, 2003.

This study was supported by a grant from the National Center for InjuryPrevention and Control, Centers for Disease Control and Prevention, Atlanta,Ga.

We thank David Rubanowice, BS, for his computer-programming assistance.