Advertisement

Predictors of Missed Follow-up Visits in the National Traumatic Brain Injury Model Systems Cohort Study

Open AccessPublished:June 13, 2022DOI:https://doi.org/10.1016/j.apmr.2022.05.003

      Abstract

      Objective

      To identify key variables that could predict risk of loss to follow-up (LTFU) in a nationally funded longitudinal database of persons with traumatic brain injury.

      Design

      Secondary analysis of a prospective longitudinal cohort study.

      Setting

      Traumatic Brain Injury Model System (TBIMS) Centers in the US.

      Participants

      A total of 17,956 TBIMS participants (N=17,956) with interview status data available were included if eligible for 1-, 2-, 5-, 10-, 15-, or 20-year follow-ups between October 31, 1989, and September 30, 2020.

      Interventions

      Not applicable.

      Main Outcome Measures

      Follow-up data collection completion status at years 1, 2, 5, 10, 15, and 20.

      Results

      Information relevant to participants’ history, injury characteristics, rehabilitation stay, and patterns of follow-up across 20 years were considered using a series of logistic regression models. Overall, LTFU rates were low (consistently <20%). The most robust predictors of LTFU across models were missed earlier follow-ups and demographic factors including Hispanic ethnicity, lower education, and lack of private health insurance.

      Conclusions

      Efforts to retain participants in such social disadvantaged or minority groups are encouraged given their disproportionate rate of LTFU. Repeated attempts to reach participants after a previously missed assessment are beneficial because many participants that missed 1 or more follow-ups were later recovered.

      Keywords

      List of abbreviations:

      GCS (Glasgow Coma Scale), LTFU (loss to follow-up), PTA (posttraumatic amnesia), TBI (traumatic brain injury), TBIMS (Traumatic Brain Injury Model Systems)
      Since 1987, the National Institute on Disability, Independent Living, and Rehabilitation Research has continuously funded the Traumatic Brain Injury Model Systems (TBIMS) program. The TBIMS National Database is the largest longitudinal database of persons with traumatic brain injury (TBI), with greater than 18,000 participants enrolled and tracked up to 30 years post injury. By collecting data regarding preinjury function and psychosocial factors, injury characteristics, rehabilitation variables, and postinjury conditions and functioning, the TBIMS database is instrumental in providing information on TBI recovery outcomes and trajectories. However, the generalizability and replicability of findings from all longitudinal studies are affected by participant retention, potentially introducing systematic bias.
      • Gustavson K
      • von Soest T
      • Karevold E
      • Røysamb E.
      Attrition and generalizability in longitudinal studies: findings from a 15-year population-based study and a Monte Carlo simulation study.
      Loss to follow-up (LTFU) rates in prior TBI studies range from 47%-60% in the first 2 years after enrollment
      • Corrigan JD
      • Harrison-Felix C
      • Bogner J
      • Dijkers M
      • Terrill MS
      • Whiteneck G.
      Systematic bias in traumatic brain injury outcome studies because of loss to follow-up.
      ,
      • Jourdan C
      • Bayen E
      • Bahrami S
      • et al.
      Loss to follow-up and social background in an inception cohort of patients with severe traumatic brain injury: results from the PariS-TBI study.
      and are higher than LTFU observed in other populations with neurologic conditions (eg, <20% in stroke).
      • Veerbeek JM
      • Kwakkel G
      • van Wegen EE
      • Ket JC
      • Heymans MW.
      Early prediction of outcome of activities of daily living after stroke: a systematic review.
      Higher LTFU in samples with TBI could relate to characteristics of the population with trauma, including younger age and psychological, behavioral, and interpersonal ramifications of traumatic injuries.
      • Jourdan C
      • Bayen E
      • Bahrami S
      • et al.
      Loss to follow-up and social background in an inception cohort of patients with severe traumatic brain injury: results from the PariS-TBI study.
      ,
      • Corrigan JD
      • Selassie AW
      • Orman JAL.
      The epidemiology of traumatic brain injury.
      ,
      • Langlois JA
      • Rutland-Brown W
      • Wald MM.
      The epidemiology and impact of traumatic brain injury: a brief overview.
      However, LTFU in the first 5 years in the Spinal Cord Injury Model Systems is reported to be <33%,
      • Kim H
      • Cutter GR
      • George B
      • Chen Y.
      Understanding and preventing loss to follow-up: experiences from the Spinal Cord Injury Model Systems.
      indicating potential unique TBI-related factors affecting LTFU.
      Prior research suggests sociodemographic factors may predict LTFU in TBI studies. Several studies report racial and ethnic minority groups are more likely to be lost to follow-up,
      • Corrigan JD
      • Harrison-Felix C
      • Bogner J
      • Dijkers M
      • Terrill MS
      • Whiteneck G.
      Systematic bias in traumatic brain injury outcome studies because of loss to follow-up.
      ,

      Harrison-Felix C, Dijkers M. Predictors of loss to follow-up in a longitudinal study of persons with traumatic brain injury. Detroit: Rehabilitation Institute of Michigan Del Harder Research Day; 1998.

      • Krellman JW
      • Kolakowsky-Hayner SA
      • Spielman L
      • et al.
      Predictors of follow-up completeness in longitudinal research on traumatic brain injury: findings from the National Institute on Disability and Rehabilitation Research Traumatic Brain Injury Model Systems program.
      • Sander AM
      • Lequerica AH
      • Ketchum JM
      • et al.
      Race/ethnicity and retention in traumatic brain injury outcomes research: a Traumatic Brain Injury Model Systems National Database study.
      as are individuals with a history substance misuse.
      • Corrigan JD
      • Harrison-Felix C
      • Bogner J
      • Dijkers M
      • Terrill MS
      • Whiteneck G.
      Systematic bias in traumatic brain injury outcome studies because of loss to follow-up.
      ,
      • Jourdan C
      • Bayen E
      • Bahrami S
      • et al.
      Loss to follow-up and social background in an inception cohort of patients with severe traumatic brain injury: results from the PariS-TBI study.
      ,
      • Sander AM
      • Lequerica AH
      • Ketchum JM
      • et al.
      Race/ethnicity and retention in traumatic brain injury outcomes research: a Traumatic Brain Injury Model Systems National Database study.
      ,
      • Corrigan JD
      • Bogner JA
      • Mysiw WJ
      • Clinchot D
      • Fugate L.
      Systematic bias in outcome studies of persons with traumatic brain injury.
      Other demographics shown predictive of LTFU in TBI studies include younger age,
      • Sander AM
      • Lequerica AH
      • Ketchum JM
      • et al.
      Race/ethnicity and retention in traumatic brain injury outcomes research: a Traumatic Brain Injury Model Systems National Database study.
      unmarried status,

      Harrison-Felix C, Dijkers M. Predictors of loss to follow-up in a longitudinal study of persons with traumatic brain injury. Detroit: Rehabilitation Institute of Michigan Del Harder Research Day; 1998.

      unemployment,
      • Corrigan JD
      • Harrison-Felix C
      • Bogner J
      • Dijkers M
      • Terrill MS
      • Whiteneck G.
      Systematic bias in traumatic brain injury outcome studies because of loss to follow-up.
      ,
      • Jourdan C
      • Bayen E
      • Bahrami S
      • et al.
      Loss to follow-up and social background in an inception cohort of patients with severe traumatic brain injury: results from the PariS-TBI study.
      and less education.
      • Krellman JW
      • Kolakowsky-Hayner SA
      • Spielman L
      • et al.
      Predictors of follow-up completeness in longitudinal research on traumatic brain injury: findings from the National Institute on Disability and Rehabilitation Research Traumatic Brain Injury Model Systems program.
      ,
      • Sander AM
      • Lequerica AH
      • Ketchum JM
      • et al.
      Race/ethnicity and retention in traumatic brain injury outcomes research: a Traumatic Brain Injury Model Systems National Database study.
      Vos et al
      • Vos L
      • Williams MW
      • Spielman L
      • et al.
      Understanding loss to follow-up in a longitudinal study of people with traumatic brain injury.
      recently examined predictors of LTFU in the first 15 years after injury from 1 TBIMS site located in Texas, finding that socially disadvantaged group membership, low education, employment status, problematic substance use, payor source, cause of injury, residence at rehabilitation discharge, and results of earlier follow-up attempts were all associated with LTFU. However, payor source and Hispanic ethnicity (particularly Spanish speakers born outside the US) were the most meaningful predictors in adjusted models. Vos
      • Vos L
      • Williams MW
      • Spielman L
      • et al.
      Understanding loss to follow-up in a longitudinal study of people with traumatic brain injury.
      recommended a larger study using the full TBIMS database to include greater analytical power, longer follow-up periods, and additional participant characteristics to further elucidate risk for LTFU. The findings from their regional sample may not represent the broader TBIMS database because of the low sample size and potential location-specific factors.
      Therefore, our objective was to identify demographic, preinjury, injury-related, and rehabilitation predictors of LTFU in the TBIMS National Database at 1, 2, 5, 10, 15, and 20 years post injury. We hypothesized that participants from socially or economically disadvantaged groups, those with problematic substance use or violent cause of injury, and those who missed prior follow-up attempts are more likely to be lost to follow-up.

      Methods

      Participants and follow-up procedures

      TBIMS study participants came from 23 centers in the US. TBIMS research activities are approved by local institutional review boards. Participants or proxy responders gave informed consent for participation. Eligibility criteria for TBIMS were injury age 16 years or older, moderate-severe TBI (posttraumatic amnesia [PTA]>24 hours, trauma-related intracranial neuroimaging abnormalities, loss of consciousness >30 minutes, or emergency department Glasgow Coma Scale [GCS] score <13), and acute care hospitalization within 72 hours of injury followed by hospitalization in a designated TBIMS inpatient rehabilitation facility. For the current study, TBIMS participants were included if eligible for 1-, 2-, 5-, 10-, 15-, or 20-year follow-ups between October 31, 1989, and September 30, 2020, (data from longer post injury were too sparse to allow comparable analyses). TBIMS follow-up data are primarily collected over the telephone but may be collected in person. When participants with TBI are unreachable or unable to respond because of impairment, data are obtained from a proxy respondent. The TBIMS Centers are funded through competitive application every 5 years; thus, participating TBIMS sites vary across funding cycles. To reduce the effect of LTFU because of defunding sites, the National Institute on Disability, Independent Living, and Rehabilitation Research has funded selected sites with high volume enrollment to continue follow-up data collection at certain time points. Rigorous guidelines for maximizing retention are in place, including repeated required contact attempts, quarterly mailings, and strategies for successful contact.

      National Data and Statistical Center. Traumatic Brain Injury Model Systems National Data and Statistical Center. Available at: https://www.tbindsc.org/. Accessed July 5, 2021.

      The guidelines include using letters to participants, directory assistance, internet sites, hospital records, death search, inmate search, and location services.

      Variables included in analysis

      The primary study outcome was dichotomous: lost to follow-up vs followed. Participants coded as refused, unable to reach, or withdrawn at scheduled study follow-up were considered lost to follow-up, while participants for whom follow-up data were collected were coded as followed. In cases when a participant withdrew from the study, they were considered lost to follow-up for that year and excluded from analyses of later years. Similarly, once persons were coded as dead they were excluded from all subsequent analyses. Those incarcerated or due for follow-up during a period of lost funding were excluded from the analysis for that follow-up year but remained eligible for inclusion at subsequent years.
      Predictor variables consisted of demographic and patient history items defined and coded according to the TBIMS Syllabus.

      National Data and Statistical Center. Traumatic Brain Injury Model Systems National Data and Statistical Center. Available at: https://www.tbindsc.org/. Accessed July 5, 2021.

      A list of predictor variables and definitions is found in table 1. Few participants (n=39, <1%) with missing sex, those with missing age at injury, or those who resided at correctional institutions were excluded from all analyses. Missing values for some categorical variables were recoded as other/unknown. Other categorical variables were collapsed because of small proportions in various levels (eg, divorced and separated; full-time and part-time students). Prior follow-up attempt outcomes (followed, unable to reach, refused, incarcerated, no funding) were also included as predictors.
      Table 1Predictor variable definitions
      PredictorDefinition
      AgeAge variable defined as age at time of injury
      TFCDays from injury until able to follow commands, in days
      PTADays from injury until out of posttraumatic amnesia, in days
      FIM scoreFIM total score at discharge from acute rehabilitation
      Acute care length of stayDays from acute care admit to acute care discharge
      Rehabilitation length of stayDays spent in rehabilitation
      Computed total length of stayTotal of acute care length of stay and rehabilitation length of stay
      Year of injuryInjury year, grouped into 1985-1990, 1991-2000, 2001-2010, 2011-2020
      SexDefined as female, male, or unknown
      Race and ethnicityDefined as White, Black, Hispanic, Native American, Asian/Pacific Islander, and a combination of Other and Unknown
      Marital statusMarital status at time of injury, defined as single, married, widowed, divorced or separated (combined), and other or unknown (combined)
      EducationEducation at time of injury categorized as <12 y (combined); GED or high school diploma (combined); associate's degree, trade school, or some college (combined); bachelor's degree; master's or doctoral degree (combined); and other/unknown (combined)
      ResidenceResidence at time of injury defined as private residence, assisted living (nursing, subacute, hospital care, adult homes), homeless and hotel/motel (combined); correctional institutions; other/unknown (combined)
      EmploymentEmployment at time of injury defined as competitively employed, student (full-time and part-time combined), special education and special employment (combined), homemaker, retired (for any reason), unemployed (includes those looking and not looking for work), volunteer, and other/unknown (includes all remaining categories)
      GCSGrouping based on GCS into mild (13-15), moderate (9-12), severe (3-8), intubated/sedated, and missing
      Cause of injuryDefined as violent (gunshot, assault, other violence), vehicular (motor vehicle, motorcycle, bicycle, ATV/ATC, and other vehicular), sports-related (water, field/track, gymnastic, winter, air sports, other sports), pedestrian, fall, and other and unknown (combined)
      Problem substance useProblem use defined as having taken illicit drugs, binge drinking (5+ drinks in a setting in the last month, or heavy drinking (14+ drinks for men per week, 7+ drinks for women per week)
      Primary rehabilitation payorPayor source for acute rehabilitation defined as insured (private insurance, workers’ compensation, auto insurance), Medicare, and other and unknown (combined)
      Prior statusDefined as followed, incarcerated, unable to reach, refused to complete, or no funding available to attempt follow-up
      Abbreviations: ATC, all-terrain cycle; ATV, all-terrain vehicle; GED, General Equivalency Diploma.

      LTFU data analysis

      Analyses included several univariable and multivariable logistic models with 1 random effect (study site) to control for site-related differences in LTFU. A series of unadjusted (univariable) and adjusted (multivariable) logistic mixed models were fitted using participant history variables, demographic factors, rehabilitation variables, payor data, and prior study follow-up status as predictors. Reference groups were assigned to be the most common group for each variable, with the exception of education (reference group was set as <12 years) because these reference groups allow for direct comparison with results from prior TBI retention studies.
      • Corrigan JD
      • Harrison-Felix C
      • Bogner J
      • Dijkers M
      • Terrill MS
      • Whiteneck G.
      Systematic bias in traumatic brain injury outcome studies because of loss to follow-up.
      ,
      • Jourdan C
      • Bayen E
      • Bahrami S
      • et al.
      Loss to follow-up and social background in an inception cohort of patients with severe traumatic brain injury: results from the PariS-TBI study.
      Post hoc tests to assess for differences in expected frequencies were conducted using Fischer exact tests because of some low-frequency groups.
      Natural logarithmic transformation was used for length of stay, FIM scores, PTA, and time to follow commands to reduce skewness and avoid convergence problems of the maximum likelihood estimates of the model parameters. Because the associations between the log odds of outcome with age and with year of injury were not linear, we used their natural cubic spline representation as predictors in the models rather than losing precision from collapsing data into categories. To reduce the number of overlapping variables in the adjusted models, total length of stay and GCS were entered into models because these variables had fewer missing data points and performed better in univariate models than other indicators of length of care and brain injury severity.

      Results

      Description of study sample

      The study sample included 17,956 TBIMS participants with interview status data available. Follow-up status by year is displayed in table 2. Of those enrolled, 96.8% (17,384) were eligible (ie, due for assessment) for 1-year follow-up, 88.9% (15,963) were eligible for 2-year follow-up, 69.8% (12,524) were eligible for 5-year follow-up, 44.0% (7893) were eligible for 10-year follow-up, 23.2% (4165) were eligible for 15-year follow-up, and 8.0% (1434) were eligible for 20-year follow-up. The number of participants included in unadjusted and adjusted regression models was slightly lower because of missing values in covariates. Characteristics of the sample are presented in table 3. For frequencies of categorical predictor variables before some groups were combined, please refer to supplemental table S1. For a comparison of the predictor variables across different study cohorts, see supplemental table S2.
      Table 2Total participant status by year
      StatusYearYearYearYearYearYear
      1, n (%)2, n (%)5, n (%)10, n (%)15, n (%)20, n (%)
      Attempted
      Followed14,752 (82.16)12,856 (75.08)9486 (57.32)5817 (37.31)2933 (19.92)1077 (7.52)
      Unable to reach1603 (8.93)1716 (10.02)1540 (9.31)1172 (7.52)565 (3.84)219 (1.53)
      Refused133 (0.74)122 (0.71)66 (0.40)42 (0.27)17 (0.12)7 (0.05)
      Withdrew265 (1.48)169 (0.98)180 (1.09)132 (0.85)62 (0.42)20 (0.14)
      Not attempted
      Incarcerated188 (1.05)262 (1.53)227 (1.37)164 (1.05)89 (0.60)25 (0.17)
      No funding443 (2.47)838 (4.89)1025 (6.19)566 (3.63)499 (3.39)86 (0.60)
      Ineligible
      Dead568 (3.16)405 (2.37)776 (4.69)739 (4.73)334 (2.23)171 (1.19)
      NA4 (0.02)4 (0.02)2 (0.01)4 (0.03)2 (0.01)0 (0.00)
      Not due0 (0.00)751 (4.39)3247 (19.62)6957 (44.62)10221 (69.43)12721 (88.80)
      NOTE. Total enrolled in TBIMS was 17,958. Percentages were calculated based on total attempted, not attempted, and ineligible by year. Two cases were removed because they did not have any valid interview status recorded.
      Abbreviation: NA, missing data in the TBIMS database.
      Table 3Demographics, injury characteristics, and predictors of the potential samples at each time point
      VariableTotal Enrolled (N= 17,956), Mean ± SDYear 1 (n=16,719), Mean ± SDYear 2 (n=14,834), Mean ± SDYear 5 (n= 11,257), Mean ± SDYear 10 (n=7154), Mean ± SDYear 15 (n=3568), Mean ± SDYear 20 (n=1321), Mean ± SD
      Continuous characteristics
      Age at injury (y)42.33±19.7641.71±19.4140.80±18.9538.61±17.6735.54±15.4633.19±13.7330.11±11.12
      Brain injury severity, TFC (d)10.08±21.5310.02±21.3210.18±21.4010.23±20.2110.70±19.7511.12±21.6312.49±24.95
      Brain injury severity, PTA (d)31.14±34.1131.09±34.1531.55±34.4932.30±33.8733.18±32.6333.34±32.7138.22±35.62
      FIM score89.74±23.1590.54±22.4390.87±22.4292.06±22.0394.21±21.6096.81±21.4899.34±21.20
      Computed total LOS (d)47.30±36.3247.51±36.6547.93±36.9748.30±36.6848.40±35.0048.32±35.8654.44±39.03
      Acute care LOS (d)20.55±17.4420.65±17.4920.84±17.6120.92±17.1720.82±16.5619.92±15.6720.91±17.18
      Rehabilitation LOS (d)26.78±26.4826.88±26.8227.09±26.9627.38±26.8827.58±25.3428.40±26.9733.53±28.49
      VariableMedian,

      n (%)
      Median,

      n (%)
      Median,

      n (%)
      Median,

      n (%)
      Median,

      n (%)
      Median,

      n (%)
      Median,

      n (%)
      Year of injury2009200920082007200420011997
      1985-1990144 (0.80)139 (0.83)136 (0.92)132 (1.17)121 (1.69)80 (2.24)101 (7.65)
      1991-20002273 (12.66)2162 (12.93)1962 (13.23)1684 (14.96)1665 (23.27)1483 (41.56)1214 (91.90)
      2001-20107865 (43.8)7389 (44.2)6984 (47.08)6398 (56.84)5347 (74.74)2005 (56.19)6 (0.45)
      2011-20207674 (42.74)7029 (42.04)5752 (38.78)3043 (27.03)21 (0.29)0 (0)0 (0)
      Categorical characteristicsn (%)n (%)n (%)n (%)n (%)n (%)n (%)
      Sex
       Male13,211 (73.57)12,288 (73.50)10,889 (73.41)8273 (73.49)5317 (74.32)2639 (73.96)976 (73.88)
       Female4737 (26.38)4431 (26.50)3945 (26.59)2984 (26.51)1837 (25.68)929 (26.04)345 (26.12)
       Missing8 (0.04)0 (0)0 (0)0 (0)0 (0)0 (0)0 (0)
      Race and ethnicity
       White11,882 (66.17)11,056 (66.13)9817 (66.18)7430 (66.00)4671 (65.29)2269 (63.59)745 (56.40)
       Black3287 (18.31)3035 (18.15)2683 (18.09)2151 (19.11)1479 (20.67)821 (23.01)388 (29.37)
       Hispanic1973 (10.99)1866 (11.16)1657 (11.17)1163 (10.33)678 (9.48)313 (8.77)124 (9.39)
       Native American93 (0.52)90 (0.54)83 (0.56)68 (0.60)38 (0.53)18 (0.50)4 (0.30)
       Asian/Pacific Islander488 (2.72)466 (2.79)410 (2.76)310 (2.75)213 (2.98)110 (3.08)44 (3.33)
       Other/Unknown233 (1.30)206 (1.23)184 (1.24)135 (1.20)75 (1.05)37 (1.04)16 (1.21)
      Marital status
       Married5946 (33.11)5533 (33.09)4845 (32.66)3557 (31.6)2152 (30.08)1015 (28.45)312 (23.62)
       Single8195 (45.64)7725 (46.20)6979 (47.05)5503 (48.89)3768 (52.67)1981 (55.52)814 (61.62)
       Divorced/separated2808 (15.64)2598 (15.54)2303 (15.53)1769 (15.71)1067 (14.91)517 (14.49)183 (13.85)
       Widowed950 (5.29)827 (4.95)681 (4.59)415 (3.69)162 (2.26)54 (1.51)12 (0.91)
       Other/unknown57 (0.32)36 (0.22)26 (0.18)13 (0.12)5 (0.07)1 (0.03)0 (0)
      Education
       <12 y4398 (24.49)4111 (24.59)3680 (24.81)2959 (26.29)2062 (28.82)1144 (32.06)476 (36.03)
       GED/HS6171 (34.37)5700 (34.09)5035 (33.94)3809 (33.84)2412 (33.72)1138 (31.89)368 (27.86)
       Associates/trade/some college4110 (22.89)3894 (23.29)3523 (23.75)2691 (23.91)1672 (23.37)810 (22.7)324 (24.53)
       Bachelor's1950 (10.86)1817 (10.87)1586 (10.69)1125 (9.99)612 (8.55)284 (7.96)91 (6.89)
       Master's/doctorate1047 (5.83)962 (5.75)807 (5.44)520 (4.62)281 (3.93)107 (3.00)26 (1.97)
       Other/unknown280 (1.56)235 (1.41)203 (1.37)153 (1.36)115 (1.61)85 (2.38)36 (2.73)
      Residence
       Private17,494 (97.43)16,345 (97.76)14,514 (97.84)11,040 (98.07)7028 (98.24)3505 (98.23)1293 (97.88)
       Assisted living143 (0.8)118 (0.71)98 (0.66)60 (0.53)34 (0.48)11 (0.31)5 (0.38)
       Homeless/hotel204 (1.14)180 (1.08)153 (1.03)107 (0.95)67 (0.94)40 (1.12)17 (1.29)
       Other/unknown100 (0.56)76 (0.45)69 (0.47)50 (0.44)25 (0.35)12 (0.34)6 (0.45)
       Correctional institution15 (0.08)0 (0)0 (0)0 (0)0 (0)0 (0)0 (0)
      Employment
       Competitively employed10,861 (60.49)10,358 (61.95)9349 (63.02)7302 (64.87)4876 (68.16)2412 (67.60)866 (65.56)
       FT/PT student1110 (6.18)1070 (6.4)985 (6.64)823 (7.31)558 (7.80)327 (9.16)139 (10.52)
       Special education/employment44 (0.25)40 (0.24)33 (0.22)28 (0.25)18 (0.25)11 (0.31)7 (0.53)
       Homemaker372 (2.07)337 (2.02)289 (1.95)220 (1.95)138 (1.93)57 (1.60)14 (1.06)
       Retired3086 (17.19)2663 (15.93)2151 (14.50)1266 (11.25)483 (6.75)145 (4.06)12 (0.91)
       Unemployed2177 (12.12)1989 (11.90)1800 (12.13)1454 (12.92)981 (13.71)547 (15.33)248 (18.77)
       Volunteer109 (0.61)100 (0.60)86 (0.58)56 (0.50)25 (0.35)8 (0.22)3 (0.23)
       Other/unknown197 (1.10)162 (0.97)141 (0.95)108 (0.96)75 (1.05)61 (1.71)32 (2.42)
      Injury severity (GCS)
       Mild5793 (32.26)5285 (31.61)4489 (30.26)3127 (27.78)1746 (24.41)792 (22.20)237 (17.94)
       Moderate2092 (11.65)1950 (11.66)1723 (11.62)1356 (12.05)881 (12.31)469 (13.14)222 (16.81)
       Severe5890 (32.80)5570 (33.32)5074 (34.21)4066 (36.12)2674 (37.38)1510 (42.32)643 (48.68)
       Intubated/sedated3958 (22.04)3726 (22.29)3384 (22.81)2604 (23.13)1783 (24.92)767 (21.50)213 (16.12)
       Missing223 (1.24)188 (1.12)164 (1.11)104 (0.92)70 (0.98)30 (0.84)6 (0.45)
      Cause of injury
       Violent2022 (11.26)1886 (11.28)1697 (11.44)1351 (12.00)964 (13.47)536 (15.02)264 (19.98)
       Vehicle8870 (49.4)8407 (50.28)7658 (51.62)6196 (55.04)4168 (58.26)2117 (59.33)776 (58.74)
       Sports-related316 (1.76)306 (1.83)286 (1.93)232 (2.06)143 (2.00)66 (1.85)17 (1.29)
       Pedestrian1371 (7.64)1285 (7.69)1144 (7.71)846 (7.52)533 (7.45)286 (8.02)109 (8.25)
       Fall5267 (29.33)4737 (28.33)3959 (26.69)2572 (22.85)1306 (18.26)538 (15.08)143 (10.83)
       Other/unknown110 (0.61)98 (0.59)90 (0.61)60 (0.53)40 (0.56)25 (0.70)12 (0.91)
      Problem substance use
       No9223 (51.36)8572 (51.27)7545 (50.86)5500 (48.86)3105 (43.4)1235 (34.61)152 (11.51)
       Yes6716 (37.40)6271 (37.51)5626 (37.93)4395 (39.04)2891 (40.41)1381 (38.71)416 (31.49)
       Refused/unknown to informant2017 (11.23)1876 (11.22)1663 (11.21)1362 (12.10)1158 (16.19)952 (26.68)753 (57.00)
      Primary rehab payor
       Insurance9359 (52.12)8866 (53.03)7979 (53.79)6135 (54.50)4095 (57.24)2116 (59.3)743 (56.25)
       Medicare2741 (15.27)2388 (14.28)1925 (12.98)1150 (10.22)420 (5.87)122 (3.42)30 (2.27)
       Other/unknown5856 (32.61)5465 (32.69)4930 (33.23)3972 (35.28)2639 (36.89)1330 (37.28)548 (41.48)
      NOTE. For continuous variables, all items have <12% missing. For TFC, participants who never followed commands were included as number of days between rehabilitation discharge and date of injury plus 1. For those still in PTA at time of rehabilitation discharge, PTA date was included as number of days between rehabilitation discharge and date of injury plus 1. Year of injury was treated as continuous predictor in models but is displayed by decade above for reference purposes.
      Abbreviations: FIM, FIM score at rehabilitation discharge; FT, full-time; GED, General Equivalency Diploma; HS, high school; LOS, length of stay; PT, part-time; PTA, posttraumatic amnesia; TFC, time to follow commands.
      In general, the people recruited in the early years were younger, had higher severity of injury, were less often of Hispanic ethnicity, were less educated, more often had violent/vehicular injuries, and more often had refused/unknown data concerning substance use.
      At each follow-up year, there was a group of participants eligible for follow-up but not attempted because they were incarcerated or their study site temporarily lost funding (see table 2). Of those attempted, LTFU was 11.9% at year 1, 13.5% at year 2, 15.8% at year 5, 18.8% at year 10, 18.0% at year 15, and 18.6% at year 20.

      Patterns of follow-up in participants with available 20-year data

      Data from participants included in the 20-year multivariable regression analyses (n=1249) were further examined to identify response patterns across the 6 follow-up assessments. The most common pattern (31.2% of participants) was “Followed” at all assessment points, yet it was not uncommon for participants to be followed after missing 1 or more assessment because of either LTFU or no funding (21.5% of participants). Repeated and persistent LTFU was rare. Common response patterns are displayed in table 4.
      Table 4Follow-up status patterns for those eligible for 20-y follow-up (n=1249)
      Year 1Year 2Year 5Year 10Year 15Year 20Frequency, n (%)
      Followed 6 times
       FollowedFollowedFollowedFollowedFollowedFollowed390 (31.2)
      Followed 5 times268 (21.5)
       FollowedLTFUFollowedFollowedFollowedFollowed42 (3.4)
       LTFUFollowedFollowedFollowedFollowedFollowed41 (3.3)
       FollowedFollowedNo fundingFollowedFollowedFollowed35 (2.8)
       FollowedFollowedLTFUFollowedFollowedFollowed32 (2.6)
       FollowedFollowedFollowedFollowedFollowedLTFU32 (2.6)
       FollowedFollowedFollowedLTFUFollowedFollowed29 (2.3)
       FollowedFollowedFollowedNo fundingFollowedFollowed23 (1.8)
      Followed 4 times240 (19.2)
       LTFULTFUFollowedFollowedFollowedFollowed22 (1.8)
       FollowedLTFULTFUFollowedFollowedFollowed18 (1.4)
       FollowedFollowedLTFULTFUFollowedFollowed17 (1.4)
      Followed 3 times155 (12.4)
       LTFULTFULTFUFollowedFollowedFollowed19 (1.5)
      Followed 2 times107 (8.6)
       LTFULTFULTFULTFUFollowedFollowed17 (1.4)
      Followed once56 (4.5)
      Never followed33 (2.6)
       LTFULTFULTFULTFULTFULTFU16 (1.3)
      NOTE. Pattern combinations were not shown that had <1% of participants. LTFU has the same definition as in the outcome, which includes those unable to reach, refused, and withdrew.

      Unadjusted and adjusted logistic regressions

      Unadjusted model results for each follow-up are displayed in table 5. To provide greater visual comparison of the effects, the multivariable (adjusted) results are displayed graphically in figs 1-Fig 2, Fig 3, Fig 4, Fig 5, Fig 6, Fig 7, Fig 8. The pattern of results is briefly described below, with emphasis on the adjusted results.
      Table 5Univariate predictors of follow-up status at years 1, 2, 5, 10, 15, and 20
      Continuous VariablesYear 1 n=16,038-16,719Year 2 n=14,252-14,834Year 5 n=10,840-11,257Year 10 n=6789-7154Year 15 n=3318-3568Year 20 n=1168-1321
      P ValueOR (95% CI)P ValueOR (95% CI)P ValueOR (95% CI)P ValueOR (95% CI)P ValueOR (95% CI)P ValueOR (95% CI)
      Age
      In Age and Years since injury where cubic splines were used, the reported OR corresponds to the interquartile odds ratio, while the likelihood ratio test P value corresponds to the overall significance of spline coefficients.
      <.001
      All results in which P<.05.


      0.71 (0.62-0.81)
      All results in which P<.05.
      ,
      P<.001.
      <.001
      All results in which P<.05.
      0.76 (0.67-0.86)
      All results in which P<.05.
      ,
      P<.001.
      <.001
      All results in which P<.05.
      0.74 (0.65-0.85)
      All results in which P<.05.
      ,
      P<.001.
      <.001
      All results in which P<.05.
      0.76-(0.65-0.89)
      All results in which P<.05.
      ,
      P<.001.
      .031
      All results in which P<.05.
      0.71 (0.56-0.91)
      All results in which P<.05.
      ,
      P<.01.
      .613
      Computed total length of stay.090.018
      All results in which P<.05.
      0.91 (0.84-0.98)
      All results in which P<.05.
      ,
      P<.05.
      .139.346.865.562
      Acute care length of stay.022
      All results in which P<.05.
      0.93 (0.87-0.99)
      All results in which P<.05.
      ,
      P<.05.
      .029
      All results in which P<.05.
      0.93 (0.87-0.99)
      All results in which P<.05.
      ,
      P<.05.


      .064.408.442.868
      Rehabilitation length of stay.171.047
      All results in which P<.05.
      0.93 (0.87-1.00)
      All results in which P<.05.
      ,
      P<.05.


      .365.193.976.647
      FIM total<.001
      All results in which P<.05.
      1.56 (1.32-1.85)
      All results in which P<.05.
      ,
      P<.001.
      < .001
      All results in which P<.05.
      1.82 (1.53-2.17)
      All results in which P<.05.
      ,
      P<.001.


      <.001
      All results in which P<.05.
      1.78 (1.47-2.16)
      All results in which P<.05.
      ,
      P<.001.
      <.001
      All results in which P<.05.
      1.53 (1.22-1.93)
      All results in which P<.05.
      ,
      P<.001.
      .631.571
      PTA.520.507.998.380.220.293
      Year of injury
      In Age and Years since injury where cubic splines were used, the reported OR corresponds to the interquartile odds ratio, while the likelihood ratio test P value corresponds to the overall significance of spline coefficients.
      <.001
      All results in which P<.05.
      0.42 (0.34-0.53)
      All results in which P<.05.
      ,
      P<.001.
      <.001
      All results in which P<.05.
      0.34 (0.27-0.43)
      All results in which P<.05.
      ,
      P<.001.
      <.001
      All results in which P<.05.
      0.62 (0.49-0.80)
      All results in which P<.05.
      ,
      P<.001.
      <.001
      All results in which P<.05.
      1.02 (0.84-1.22)<.001
      All results in which P<.05.
      1.50 (1.08-2.09)
      All results in which P<.05.
      ,
      P<.05.
      .004
      All results in which P<.05.
      1.20 (0.71-2.03)
      Severity (TFC).454.548.567.354.624.540
      Categorical variables
      Y1 status

      (ref: followed)
      --<.001
      All results in which P<.05.
      <.001
      All results in which P<.05.
      <.001
      All results in which P<.05.


      <.001
      All results in which P<.05.
      .023
      All results in which P<.05.
      Incarcerated--3.60 (2.18-5.92)
      All results in which P<.05.
      ,
      P<.001.
      3.41 (2.15-5.39)
      All results in which P<.05.
      ,
      P<.001.
      2.48 (1.41-4.36)
      All results in which P<.05.
      ,
      P<.01.
      0.73 (0.22-2.41)0.94 (0.21-4.33)
      Unable to reach--10.27 (9.06-11.65)
      All results in which P<.05.
      ,
      P<.001.
      5.88 (5.11-6.78)
      All results in which P<.05.
      ,
      P<.001.
      3.17 (2.68-3.75)
      All results in which P<.05.
      ,
      P<.001.
      2.90 (2.31-3.64)
      All results in which P<.05.
      ,
      P<.001.
      1.66 (1.19-2.31)
      All results in which P<.05.
      ,
      P<.01.
      Refused--10.04 (6.78-14.85)
      All results in which P<.05.
      ,
      P<.001.
      3.63 (2.07-6.38)
      All results in which P<.05.
      ,
      P<.001.
      2.97 (1.48-5.96)
      All results in which P<.05.
      ,
      P<.01.
      2.46 (0.89-6.82)0.18 (0.01-3.15)
      No funding--10.11 (5.30-19.28)
      All results in which P<.05.
      ,
      P<.001.
      2.27 (1.58-3.27)
      All results in which P<.05.
      ,
      P<.001.
      1.96 (1.35-2.86)
      All results in which P<.05.
      ,
      P<.001.
      0.86 (0.35-2.12)0.77 (0.22-2.66)
      Y2 status (ref: followed)----<.001
      All results in which P<.05.
      <.001
      All results in which P<.05.
      <.001
      All results in which P<.05.
      <.001
      All results in which P<.05.
      Incarcerated----2.86 (1.87-4.37)
      All results in which P<.05.
      ,
      P<.001.
      2.25 (1.40-3.63)
      All results in which P<.05.
      ,
      P<.001.
      2.24 (1.19-4.22)
      All results in which P<.05.
      ,
      P<.05.
      1.88 (0.67-5.27)
      Unable to reach----8.23 (7.20-9.41)
      All results in which P<.05.
      ,
      P<.001.
      4.44 (3.79-5.20)
      All results in which P<.05.
      ,
      P<.001.
      3.75 (3.01-4.66)
      All results in which P<.05.
      ,
      P<.001.
      2.06 (1.49-2.86)
      All results in which P<.05.
      ,
      P<.001.
      Refused----5.36 (3.16-9.07)
      All results in which P<.05.
      ,
      P<.001.
      3.14 (1.64-6.01)
      All results in which P<.05.
      ,
      P<.001.
      1.34 (0.39-4.56)1.43 (0.40-5.13)
      No funding----2.22 (1.65-3.00)
      All results in which P<.05.
      ,
      P<.001.
      3.15 (2.33-4.25)
      All results in which P<.05.
      ,
      P<.001.
      1.78 (0.98-3.24)1.55 (0.73-3.29)
      Y5 status

      (ref: followed)
      ------<.001
      All results in which P<.05.
      <.001
      All results in which P<.05.
      <.001
      All results in which P<.05.
      Incarcerated------2.95 (1.84-4.72)
      All results in which P<.05.
      ,
      P<.001.
      1.81 (0.89-3.67)3.12 (1.14-8.55)
      All results in which P<.05.
      ,
      P<.05.
      Unable to reach------6.61 (5.64-7.73)
      All results in which P<.05.
      ,
      P<.001.
      4.54 (3.62-5.69)
      All results in which P<.05.
      ,
      P<.001.
      2.48 (1.77-3.47)
      All results in which P<.05.
      ,
      P<.001.
      Refused------1.62 (0.61-4.30)4.99 (1.69-14.72)
      All results in which P<.05.
      ,
      P<.01.
      3.14 (0.59-16.77)
      No funding------4.63 (3.31-6.47)
      All results in which P<.05.
      ,
      P<.001.
      2.54 (1.42-4.54)
      All results in which P<.05.
      ,
      P<.01.
      2.12 (1.08-4.16)
      All results in which P<.05.
      ,
      P<.05.
      Y10 status

      (ref: followed)
      --------<.001
      All results in which P<.05.
      <.001
      All results in which P<.05.
      Incarcerated--------1.52 (0.71-3.24)2.21 (1.02-4.82)
      All results in which P<.05.
      ,
      P<.05.
      Unable to reach--------7.44 (5.90-9.37)
      All results in which P<.05.
      ,
      P<.001.
      4.31 (3.03-6.14)
      All results in which P<.05.
      ,
      P<.001.
      Refused--------3.63 (1.25-10.49)
      All results in which P<.05.
      ,
      P<.05.
      3.94 (0.73-21.30)
      No funding--------1.30 (0.55-3.06)2.12 (0.93-4.82)
      Y15 status

      (ref: followed)
      ----------<.001
      All results in which P<.05.
      Incarcerated----------2.45 (0.77-7.80)
      Unable to reach----------9.80 (6.61-14.52)
      All results in which P<.05.
      ,
      P<.001.
      Refused----------10.16 (1.97-52.51)
      All results in which P<.05.
      ,
      P<.01.
      No funding----------2.13 (0.92-4.95)
      Sex

      (ref: male)
      .002
      All results in which P<.05.


      0.84 (0.75-0.94)
      All results in which P<.05.
      ,
      P<.01.
      .010
      All results in which P<.05.
      0.86 (0.77-0.97)
      All results in which P<.05.
      ,
      P<.01.
      <.001
      All results in which P<.05.
      0.77 (0.68-0.88)
      All results in which P<.05.
      ,
      P<.01.
      .005
      All results in which P<.05.
      0.81 (0.70-0.94)
      All results in which P<.05.
      ,
      P<.01.
      .007
      All results in which P<.05.
      0.73 (0.58-0.92)
      All results in which P<.05.
      ,
      P<.01.
      .911
      Race

      (ref: White)
      < .001
      All results in which P<.05.
      <.001
      All results in which P<.05.
      <.001
      All results in which P<.05.
      < .001
      All results in which P<.05.
      < .001
      All results in which P<.05.
      <.001
      All results in which P<.05.
      Black1.23 (1.07-1.40)
      All results in which P<.05.
      ,
      P<.01.
      1.27 (1.11-1.46)
      All results in which P<.05.
      ,
      P<.001.
      1.21 (1.05-1.40)
      All results in which P<.05.
      ,
      P<.01.
      1.32 (1.11-1.56)
      All results in which P<.05.
      ,
      P<.01.
      1.15 (0.88-1.51)1.16 (0.79-1.72)
      Asian/Pacific Islander1.95 (1.51-2.53)
      All results in which P<.05.
      ,
      P<.001.
      2.00 (1.53-2.61)
      All results in which P<.05.
      ,
      P<.001.
      2.35 (1.76-3.14)
      All results in which P<.05.
      ,
      P<.001.
      2.14 (1.51-3.02)
      All results in which P<.05.
      ,
      P<.001.
      2.50 (1.45-4.32)
      All results in which P<.05.
      ,
      P<.01.
      2.79 (1.32-5.91)
      All results in which P<.05.
      ,
      P<.01.
      Native American1.72 (0.92-3.23)1.23 (0.61-2.484)0.90 (0.38-2.10)1.43 (0.59-3.48)5.82 (2.13-15.88)
      All results in which P<.05.
      ,
      P<.001.
      6.69 (0.86-52.27)
      Hispanic2.06 (1.77-2.38)
      All results in which P<.05.
      ,
      P<.001.
      2.42 (2.09-2.81)
      All results in which P<.05.
      ,
      P<.001.
      3.03 (2.57-3.56)
      All results in which P<.05.
      ,
      P<.001.
      3.16 (2.57-3.88)
      All results in which P<.05.
      ,
      P<.001.
      4.58 (3.31-6.34)
      All results in which P<.05.
      ,
      P<.001.
      3.17 (1.91-5.27)
      All results in which P<.05.
      ,
      P<.001.
      Other/unknown1.21 (0.79-1.88)1.41 (0.92-2.15)2.08 (1.35-3.20)
      All results in which P<.05.
      ,
      P<.001.
      2.74 (1.62-4.63)
      All results in which P<.05.
      ,
      P<.001.
      7.30 (3.47-15.36)
      All results in which P<.05.
      ,
      P<.001.
      1.57 (0.42-5.84)
      Education (ref: 11y or below)<.001
      All results in which P<.05.
      <.001
      All results in which P<.05.
      <.001
      All results in which P<.05.
      <.001
      All results in which P<.05.
      <.001
      All results in which P<.05.
      .044
      All results in which P<.05.
      GED/HS0.64 (0.57-0.72)
      All results in which P<.05.
      ,
      P<.001.
      0.57 (0.50-0.64)
      All results in which P<.05.
      ,
      P<.001.
      0.58 (0.51-0.66)
      All results in which P<.05.
      ,
      P<.001.
      0.59 (0.50-0.69)
      All results in which P<.05.
      ,
      P<.001.
      0.64 (0.50-0.80)
      All results in which P<.05.
      ,
      P<.001.
      0.64 (0.45-0.92)
      All results in which P<.05.
      ,
      P<.05.
      Trade/some college/associate0.52 (0.46-0.60)
      All results in which P<.05.
      ,
      P<.001.
      0.52 (0.45-0.60)
      All results in which P<.05.
      ,
      P<.001.
      0.46 (0.39-0.53)
      All results in which P<.05.
      ,
      P<.001.
      0.58 (0.49-0.69)
      All results in which P<.05.
      ,
      P<.001.
      0.65 (0.50-0.85)
      All results in which P<.05.
      ,
      P<.01.
      0.57 (0.39-0.85)
      All results in which P<.05.
      ,
      P<.01.
      Bachelor's0.53 (0.44-0.64)
      All results in which P<.05.
      ,
      P<.001.
      0.41 (0.33-0.50)
      All results in which P<.05.
      ,
      P<.001.
      0.37 (0.29-0.46)
      All results in which P<.05.
      ,
      P<.001.
      0.41 (0.31-0.54)
      All results in which P<.05.
      ,
      P<.001.
      0.50 (0.32-0.78)
      All results in which P<.05.
      ,
      P<.01.
      0.56 (0.29-1.10)
      Master's/doctorate0.39 (0.30-0.51)
      All results in which P<.05.
      ,
      P<.001.
      0.30 (0.22-0.40)
      All results in which P<.05.
      ,
      P<.001.
      0.35 (0.26-0.49)
      All results in which P<.05.
      ,
      P<.001.
      0.46 (0.32-0.68)
      All results in which P<.05.
      ,
      P<.001.
      0.36 (0.16-0.78)
      All results in which P<.05.
      ,
      P<.01.
      0.56 (0.18-1.74)
      Other/unknown1.72 (1.27-2.34)
      All results in which P<.05.
      ,
      P<.001.
      1.63 (1.18-2.24)
      All results in which P<.05.
      ,
      P<.01.
      1.38 (0.95-2.00)1.94 (1.28-2.95)
      All results in which P<.05.
      ,
      P<.01.
      1.24 (0.69-2.21)0.92 (0.37-2.28)
      Marital status (ref: single)<.001
      All results in which P<.05.
      <.001
      All results in which P<.05.
      .006
      All results in which P<.05.
      .397.603.125
      Married0.74 (0.66-0.83)
      All results in which P<.05.
      ,
      P<.001.
      0.77 (0.68-0.86)
      All results in which P<.05.
      ,
      P<.001.
      0.80 (0.71-0.90)
      All results in which P<.05.
      ,
      P<.001.
      Divorced/separated0.92 (0.80-1.06)1.01 (0.88-1.16)0.99 (0.85-1.15)
      Widowed1.01 (0.81-1.26)0.94 (0.74-1.19)0.92 (0.69-1.21)
      Other/unknown1.07 (0.41-2.79)1.20 (0.44-3.24)0.37 (0.05-2.83)
      Employment

      (ref: full-time employed)
      <.001
      All results in which P<.05.


      <.001
      All results in which P<.05.
      <.001
      All results in which P<.05.
      <.001
      All results in which P<.05.
      .030
      All results in which P<.05.
      .010
      All results in which P<.05.
      Full-time/part-time student1.01 (0.82-1.24)1.03 (0.85-1.26)0.84 (0.68-1.05)0.96 (0.75-1.23)1.01 (0.71-1.46)0.79 (0.45-1.38)
      Special education/employment2.10 (1.01-4.36)
      All results in which P<.05.
      ,
      P<.05.
      2.37 (1.11-5.08)
      All results in which P<.05.
      ,
      P<.05.
      1.11 (0.44-2.81)1.37 (0.46-4.05)2.28 (0.57-9.00)1.72 (0.32-9.13)
      Homemaker1.07 (0.77-1.50)0.92 (0.65-1.32)0.95 (0.65-1.39)1.13 (0.72-1.78)1.26 (0.61-2.63)6.33 (2.09-19.21)
      All results in which P<.05.
      ,
      P<.01.
      Retired0.96 (0.83-1.11)0.73 (0.63-0.86)
      All results in which P<.05.
      ,
      P<.001.
      0.83 (0.69-1.00)
      All results in which P<.05.
      ,
      P<.05.
      0.85 (0.65-1.10)1.23 (0.77-1.99)1.54 (0.40-5.92)
      Unemployed1.36 (1.19-1.57)
      All results in which P<.05.
      ,
      P<.001.
      1.54 (1.35-1.77)
      All results in which P<.05.
      ,
      P<.001.
      1.52 (1.32-1.76)
      All results in which P<.05.
      ,
      P<.001.
      1.49 (1.26-1.77)
      All results in which P<.05.
      ,
      P<.001.
      1.57 (1.23-2.02)
      All results in which P<.05.
      ,
      P<.001.
      1.23 (0.85-1.79)
      Volunteer0.87 (0.45-1.69)0.72 (0.34-1.49)0.36 (0.13-1.02)0.90 (0.30-2.69)
      Model failed to converge or there were too few cases.
      Model failed to converge or there were too few cases.
      Other/unknown2.50 (1.72-3.62)
      All results in which P<.05.
      ,
      P<.001.
      1.81 (1.20-2.73)
      All results in which P<.05.
      ,
      P<.01.
      2.06 (1.32-3.22)
      All results in which P<.05.
      ,
      P<.01.
      1.92 (1.13-3.27)
      All results in which P<.05.
      ,
      P<.05.
      0.94 (0.43-2.04)2.83 (1.22-6.59)
      All results in which P<.05.
      ,
      P<.05.
      Problematic substance use (ref: no)<.001
      All results in which P<.05.
      <.001
      All results in which P<.05.
      <.001
      All results in which P<.05.
      <.001
      All results in which P<.05.
      .334.040
      All results in which P<.05.
      Yes1.38 (1.23-1.53)
      All results in which P<.05.
      ,
      P<.001.
      1.33 (1.19-1.48)
      All results in which P<.05.
      ,
      P<.001.
      1.45 (1.28-1.63)
      All results in which P<.05.
      ,
      P<.001.
      1.24 (1.07-1.43)
      All results in which P<.05.
      ,
      P<.01.
      1.49 (0.83-2.66)
      Unknown2.90 (2.53-3.32)
      All results in which P<.05.
      ,
      P<.001.
      2.85 (2.48-3.27)
      All results in which P<.05.
      ,
      P<.001.
      3.07 (2.64-3.57)
      All results in which P<.05.
      ,
      P<.001.
      2.05 (1.73-2.45)
      All results in which P<.05.
      ,
      P<.001.
      1.96 (1.11-3.46)
      All results in which P<.05.
      ,
      P<.05.
      Place of residence (ref: private)<.001
      All results in which P<.05.
      <.001
      All results in which P<.05.
      .015
      All results in which P<.05.
      .007
      All results in which P<.05.
      .020
      All results in which P<.05.
      .169
      Assisted living facility0.78 (0.43-1.44)0.80 (0.42-1.51)0.90 (0.44-1.85)1.02 (0.44-2.38)3.08 (0.84-11.28)
      Homeless/hotel2.14 (1.49-3.07)
      All results in which P<.05.
      ,
      P<.001.
      2.35 (1.62-3.41)
      All results in which P<.05.
      ,
      P<.001.
      2.08 (1.33-3.27)
      All results in which P<.05.
      ,
      P<.01.
      2.44 (1.42-4.17)
      All results in which P<.05.
      ,
      P<.01.
      1.65 (0.77-3.57)
      Other/unknown2.05 (1.19-3.53)
      All results in which P<.05.
      ,
      P<.01.
      1.61 (0.88-2.92)1.14 (0.55-2.38)1.80 (0.75-4.33)4.27 (1.27-14.40)
      All results in which P<.05.
      ,
      P<.05.
      Cause of injury (ref: vehicular)<.001
      All results in which P<.05.
      <.001
      All results in which P<.05.
      <.001
      All results in which P<.05.
      .010
      All results in which P<.05.
      .382.832
      Violence1.41 (1.22-1.63)
      All results in which P<.05.
      ,
      P<.001.
      1.51 (1.31-1.75)
      All results in which P<.05.
      ,
      P<.001.
      1.31 (1.12-1.54)
      All results in which P<.05.
      ,
      P<.001.
      1.25 (1.04-1.51)
      All results in which P<.05.
      ,
      P<.05.
      Sports-related0.96 (0.63-1.48)0.77 (0.50-1.20)0.47 (0.26-0.84)
      All results in which P<.05.
      ,
      P<.05.
      0.65 (0.36-1.17)
      Fall0.93 (0.83-1.06)0.83 (0.73-0.94)
      All results in which P<.05.
      ,
      P<.01.
      0.98 (0.85-1.12)0.98 (0.82-1.16)
      Pedestrian1.26 (1.06-1.50)
      All results in which P<.05.
      ,
      P<.01.
      1.16 (0.97-1.39)1.17 (0.96-1.42)1.36 (1.09-1.71)
      All results in which P<.05.
      ,
      P<.01.
      Other/unknown1.80 (1.08-3.01)
      All results in which P<.05.
      ,
      P<.05.
      1.37 (0.79-2.37)1.04 (0.51-2.15)0.82 (0.33-2.02)
      Payer source (ref: insurance)<.001
      All results in which P<.05.
      <.001
      All results in which P<.05.
      <.001
      All results in which P<.05.
      <.001
      All results in which P<.05.
      <.001
      All results in which P<.05.
      .303
      Medicare0.92 (0.79-1.09)0.84 (0.71-0.99)
      All results in which P<.05.
      ,
      P<.05.
      0.97 (0.79-1.18)0.97 (0.73-1.29)1.39 (0.84-2.29)
      Other1.68 (1.51-1.86)
      All results in which P<.05.
      ,
      P<.001.
      1.82 (1.64-2.02)
      All results in which P<.05.
      ,
      P<.001.
      2.05 (1.83-2.30)
      All results in which P<.05.
      ,
      P<.001.
      1.91 (1.67-2.19)
      All results in which P<.05.
      ,
      P<.001.
      1.78 (1.46-2.18)
      All results in which P<.05.
      ,
      P<.001.
      GCS (ref: severe)<.001
      All results in which P<.05.
      .774.057.017
      All results in which P<.05.
      .177.443
      Mild1.08 (0.95-1.21)1.05 (0.89-1.24)
      Moderate1.19 (1.02-1.39)
      All results in which P<.05.
      ,
      P<.05.
      1.30 (1.07-1.58)
      All results in which P<.05.
      ,
      P<.01.
      Sedated/intubated0.90 (0.78-1.03)0.93 (0.77-1.11)
      NOTE. Follow-up status was dichotomized to Followed=0, LTFU (unable to reach, withdrew, refused)=1. Incarcerated cases at outcome were excluded. Problematic substance use is illicit drug use, binge drinking, or heavy alcohol use in the 12 mo prior to enrollment;
      P values from Likelihood ratio test are reported for overall significance, whereas Wald P values are reported for individual levels in categorical variables. Due to rounding, some significant predictors have confidence intervals including 1.0. Length of stay variables, FIM, PTA, and TFC were log transformed.
      Abbreviations: CI, confidence interval FIM, FIM score at rehabilitation discharge; ref, reference group for paired comparison; GED, General Equivalency Diploma; HS, high school; OR, odds ratio; TFC, time to follow commands, in days.
      low asterisk In Age and Years since injury where cubic splines were used, the reported OR corresponds to the interquartile odds ratio, while the likelihood ratio test P value corresponds to the overall significance of spline coefficients.
      All results in which P<.05.
      P<.001.
      § P<.01.
      P<.05.
      Model failed to converge or there were too few cases.
      Fig 1
      Fig 1Plots for odds of LTFU by age. Natural splines with degree of freedom 3 were used for modeling age in multivariable models. FU1 refers to follow-up at year 1, and this notation is similar to other follow-up years. Reference categories were assigned for categorical variables, along with median injury year, mean total length of stay (log transformed), and mean FIM (log transformed). Random center effect was set to 0. Significant effects were evident at follow-up year 1, P<.05. Interquartile odds ratios and 95% confidence intervals in each follow-up year: FU1: 0.89 (0.74-1.06); FU2: 1.00 (0.83-1.21); FU5: 0.86 (0.70-1.05); FU10: 0.80 (0.64-1.00); FU15: 0.74 (0.52-1.05); FU20: 1.10 (0.68-1.80).

      Demographic factors

      Numerous significant relationships were evident between demographic variables and LTFU across unadjusted and adjusted models (see table 5 and figs 1-Fig 2, Fig 3, Fig 4, Fig 5, Fig 6, Fig 7, Fig 8). Age, race and ethnicity, sex, marital status, and education associations with LTFU varied across follow-up years, with race and ethnicity and education being significant most often. In adjusted models, those who were married were significantly less likely to be LTFU than those who were single at the time of injury only at year 1. At year 5 follow-up, female participants had slightly lower odds of LTFU than male participants. Age was a significant predictor in adjusted models only for follow-up year 1, with the highest odds of LTFU among participants in their early 30s. While the significance of pairwise comparisons varied across the assessment time points, there was a clear trend for less education at time of injury (particularly ≤11 years) to be associated with greater odds of LTFU. Findings related to race were relatively robust in adjusted models, with some racial minority groups (Hispanic, Asian, Native American, Other/unknown) having higher odds of LTFU at most assessment time points compared with White participants.
      Additional Fisher exact tests were performed to better understand these race and ethnicity effects. Those born outside (vs within) the US and those who did not speak English as their primary language (vs those who did) had higher rates of LTFU through 15 years (P<.05). Those proportions remained higher for language (P=.01), but not country of birth (P=.06), at year 20. Those born outside the US disproportionately spoke Spanish or languages other than English (P<.001) and most often identified as Hispanic (P<.001) or Asian/Pacific Islander (P<.001). Within the Hispanic subgroup, those born outside the US disproportionately spoke Spanish or languages other than English as a primary language (P<.001), and those who spoke Spanish as a primary language were disproportionately lost to follow-up at year 1 (P<.001), year 2 (P<.001), year 5 (P<.01), year 10 (P<.001), and year 20 (P=.046). Those born outside the US were disproportionately lost to follow-up at years 1 (P<.001), 2 (P<.001), 5 (P<.001), and 10 (P=.006). Within the Asian/Pacific Islander group, those born outside the US disproportionately spoke languages other than English as primary language (P<.001). For them, neither language nor country of origin had a significant effect on follow up status (at any follow-up year).

      Patient history

      There were significant findings for employment, problematic substance use, and place of residence in several unadjusted models but fewer effects evident in adjusted models. Residence was a significant predictor only at years 1 and 15. Compared with those living in private homes at the time of injury, higher odds of LTFU were associated with living in unknown or temporary housing (year 1) and in assisted living at time of injury (year 15). While problematic substance use carried higher odds for LTFU at year 1, only unknown substance use was a significant predictor of LTFU in years 1, 2, and 5. No patient history variables emerged as significant in adjusted models for follow-up years 10 or 20.

      Injury-related predictors

      GCS was a significant predictor in some unadjusted models, but other injury severity markers (PTA, time to follow commands) showed no significant relationship and were not included in the adjusted analyses. In the adjusted models, those with mild injuries based on GCS vs those with severe injuries had higher odds of LTFU at year 1 only. Cause of injury was a significant predictor in several models, with adjusted effects showing violent cause of injury carried lower odds of LTFU (vs vehicular) at years 5 and 15. Year of injury using splines was significant (P<.05) in all models, and the estimated odds of LTFU in adjusted analyses are displayed in fig 2. The overall trend is that the odds of LTFU decrease as injury year increases.
      Fig 2
      Fig 2Plots for odds of LTFU by injury year. Natural splines with various degree of freedom were used for modeling injury year in multivariable models. FU1 refers to follow-up year 1, and the notation is similar to other follow-up years. Reference categories were assigned for categorical variables, along with mean age, mean total length of stay (log transformed), and mean FIM (log transformed). Random TBIMS Center effect was set to 0. Significant effects were noted at each of the follow-up years, P<.001. Interquartile odds ratios and 95% confidence intervals in each follow-up year: FU1: 0.41 (0.33-0.52); FU2: 0.56 (0.43-0.72); FU5: 0.84 (0.64-1.10); FU10: 1.67 (1.34-2.09); FU15: 1.87 (1.26-2.79); FU20: 2.52 (1.25-5.08).
      Fig 3
      Fig 3Forest plots for significant adjusted effects on follow-up status at year 1. n=16,129. Except for effects noted for age and year of injury depicted in the splines, only significant predictors were displayed in the forest plots. The reference groups for the above variables were the same as in the univariate analyses and (in order) are White, 11 y or below, Single, No problem use, Private residence, Private insurance, and Severe injury.
      Fig 4
      Fig 4Forest plots for significant adjusted effects on follow-up status at year 2. n=14,358. Except for effects noted for age and year of injury depicted in the splines, only significant predictors were displayed in the forest plots. The reference groups for the above variables were the same as in the univariate analyses and (in order) are Followed, White, 11 y or below, No problem use, and Private insurance.
      Fig 5
      Fig 5Forest plots for significant adjusted effects on follow-up status at year 5. n=10,917. Except for effects noted for age and year of injury depicted in the splines, only significant predictors were displayed in the forest plots. The reference groups for the above variables were the same as in the univariate analyses and (in order) are Followed, Followed, Male, White, 11 y or below, No problem use, Vehicular, and Private insurance.
      Fig 6
      Fig 6Forest plots for significant adjusted effects on follow-up status at year 10. n=6915. Except for effects noted for age and year of injury depicted in the splines, only significant predictors were displayed in the forest plots. The reference groups for the above variables were the same as in the univariate analyses and (in order) are Followed, Followed, Followed, White, 11 y or below, and Private insurance.
      Fig 7
      Fig 7Forest plots for significant adjusted effects on follow-up status at year 15. n=3407. Except for effects noted for age and year of injury depicted in the splines, only significant predictors were displayed in the forest plots. The reference groups for the above variables were the same as in the univariate analyses and (in order) are Followed, Followed, Followed, White, Private residence, and Vehicular.
      Fig 8
      Fig 8Forest plots for significant adjusted effects on follow-up status at year 20. n=1249. Except for effects noted for age and year of injury depicted in the splines, only significant predictors were displayed in the forest plots. The reference groups for the above variables were the same as in the univariate analyses and (in order) are Followed and Followed.

      Rehabilitation variables

      Length of stay and FIM at discharge showed significant relationships in follow-up years 1 and 2, with those with shorter lengths of stay and higher FIM score at higher odds for LTFU. Payor source was the most robust rehabilitation predictor in adjusted models. At the first 4 time points, those with Medicaid, unknown, or no insurance (eg, self-pay, hospital-paid) had greater odds of LTFU than those with commercial insurance (private, auto insurance, or workers’ compensation). For payor source, there were consistent effects showing those with Medicaid, unknown, or no insurance were more likely to be lost to follow-up than those insured. To further examine these effects, fig 9 shows further breakdown of the other group. For years 1-10, the lowest LTFU is observed in the Insurance and Medicare groups. For years 1-15, the highest LTFU is observed on self-paid. Medicaid, and others/unknown, and hospital LTFU levels remain in the middle (between self-paid and Insurance/Medicaid). For years 15 and 20, the counts in self-paid, Medicaid, and others/unknown are too low to conclude differences among them. The payer source variable was not fully independent from race and ethnicity in all follow-up years. For example, at year 1 follow-up, 75.1% and 68.8% of the White and Asian participants, respectively, were covered under either private insurance or Medicare, while for Hispanic, Black, and Native American participants, these percentages were 49.7%, 50.1%, and 45.9%, respectively.
      Fig 9
      Fig 9Loss to follow-up at each year by payor source.

      Prior follow-up

      Prior follow-up was the most robust predictor in adjusted models across years 2-20, with those who did not complete a prior follow-up (for various reasons) having greater odds of LTFU at later follow-up times. At year 20, only follow-up status at years 10 and 15 were significant predictors in the adjusted model, accounting for other effects.

      Discussion

      We found LTFU in the TBIMS database to be considerably better (consistently <20%) than the prior reports of high attrition in longitudinal TBI studies, potentially credited to the rigorous strategies to minimize attrition and use of both telephone and in-person assessment methods.
      • Corrigan JD
      • Harrison-Felix C
      • Bogner J
      • Dijkers M
      • Terrill MS
      • Whiteneck G.
      Systematic bias in traumatic brain injury outcome studies because of loss to follow-up.
      ,
      • Jourdan C
      • Bayen E
      • Bahrami S
      • et al.
      Loss to follow-up and social background in an inception cohort of patients with severe traumatic brain injury: results from the PariS-TBI study.
      Follow-up rates were also higher in the TBIMS data than the Spinal Cord Injury Model Systems data.
      • Kim H
      • Cutter GR
      • George B
      • Chen Y.
      Understanding and preventing loss to follow-up: experiences from the Spinal Cord Injury Model Systems.
      Many participants who missed a follow-up were later recovered. Nevertheless, the most robust predictor of late-stage follow-up was assessment completion status at earlier follow-up years. This finding replicates the results of an earlier single-center TBIMS study.
      • Vos L
      • Williams MW
      • Spielman L
      • et al.
      Understanding loss to follow-up in a longitudinal study of people with traumatic brain injury.
      These findings highlight the importance of maximizing retention early in longitudinal studies because multiple missed assessments may carry a cumulative risk for LTFU at later attempts.
      Consistent with trends seen by Vos et al,
      • Vos L
      • Williams MW
      • Spielman L
      • et al.
      Understanding loss to follow-up in a longitudinal study of people with traumatic brain injury.
      year of injury was a significant predictor in all adjusted models, and this should be considered in the context of the historical development of the TBIMS program. There have been numerous policy changes over time aimed at increasing participant retention, often capitalizing on technology and providing clear follow-up protocols. Our results generally support the success of these changes, with a significant reduction in odds of LTFU for more recently injured participants. Further, efforts to retain a participant after a missed follow-up are fruitful because examination of response patterns indicated many who missed an assessment were recovered at later follow-up years.
      Like prior studies,
      • Corrigan JD
      • Harrison-Felix C
      • Bogner J
      • Dijkers M
      • Terrill MS
      • Whiteneck G.
      Systematic bias in traumatic brain injury outcome studies because of loss to follow-up.
      ,
      • Krellman JW
      • Kolakowsky-Hayner SA
      • Spielman L
      • et al.
      Predictors of follow-up completeness in longitudinal research on traumatic brain injury: findings from the National Institute on Disability and Rehabilitation Research Traumatic Brain Injury Model Systems program.
      ,
      • Vos L
      • Williams MW
      • Spielman L
      • et al.
      Understanding loss to follow-up in a longitudinal study of people with traumatic brain injury.
      we found higher attrition rates for non-White compared with White participants. However, this investigation was one of the first
      • Sander AM
      • Lequerica AH
      • Ketchum JM
      • et al.
      Race/ethnicity and retention in traumatic brain injury outcomes research: a Traumatic Brain Injury Model Systems National Database study.
      ,
      • Vos L
      • Williams MW
      • Spielman L
      • et al.
      Understanding loss to follow-up in a longitudinal study of people with traumatic brain injury.
      ,

      National Data and Statistical Center. Traumatic Brain Injury Model Systems National Data and Statistical Center. Available at: https://www.tbindsc.org/. Accessed July 5, 2021.

      to report that Hispanic participants, but not Black participants, were more likely to miss follow-up visits. Further, this study showed significant effects for Asian/Pacific Islander participants who were not reported in prior studies that may have been limited by smaller sample size. Factors relevant to TBI outcome,
      • Lequerica AH
      • Botticello A
      • O'Neill J
      • et al.
      Relationship between Hispanic nativity, residential environment, and productive activity among individuals with traumatic brain injury: a TBI Model Systems study.
      including native country and language of origin, were also related to retention in this study. As suggested by preliminary, underpowered analyses in Vos's regional study,
      • Vos L
      • Williams MW
      • Spielman L
      • et al.
      Understanding loss to follow-up in a longitudinal study of people with traumatic brain injury.
      post hoc analyses highlighted that for those of Hispanic origin, primary language spoken and country of origin were predictive of LTFU, potentially driving the effect for greater LTFU in Hispanic participants. These findings support the previous calls for culturally competent bilingual study personnel to increase retention and limit bias.
      • Sander AM
      • Lequerica AH
      • Ketchum JM
      • et al.
      Race/ethnicity and retention in traumatic brain injury outcomes research: a Traumatic Brain Injury Model Systems National Database study.
      ,
      • Sander AM
      • Ketchum JM
      • Lequerica AH
      • et al.
      Primary language and participation outcomes in Hispanics with traumatic brain injury: a Traumatic Brain Injury Model Systems study.
      In contrast to the Hispanic subgroup, for the Asian/Pacific Islander group, nativity and language did not explain the greater odds of LTFU. Thus, there may be other unknown cultural factors that explain this finding. Implementing policies to retain participants of racial and ethnic groups at risk for LTFU is crucial, especially because they may be already underrepresented in the TBIMIS database given the known disparities in access to rehabilitation services.
      • Sander AM
      • Ketchum JM
      • Lequerica AH
      • et al.
      Primary language and participation outcomes in Hispanics with traumatic brain injury: a Traumatic Brain Injury Model Systems study.
      • Pappadis MR
      • Sander AM
      • Struchen MA
      • Leung P
      • Smith DW.
      Common misconceptions about traumatic brain injury among ethnic minorities with TBI.
      • Asemota AO
      • George BP
      • Cumpsty-Fowler CJ
      • Haider AH
      • Schneider EB.
      Race and insurance disparities in discharge to rehabilitation for patients with traumatic brain injury.
      For example, validated translations of test and interview forms, community partnerships, and access to national resources may be helpful to diversify recruitment and improve retention of minority groups. Yet, it is also important to recognize the highly heterogeneous nature of these groups, and there are innumerable other moderating effects (eg, acculturation, bilingual status, immigration/citizenship status, economic/housing factors) that might warrant exploration.
      While many demographic factors, rehabilitation variables, and aspects of participants’ history predicted LTFU at various time points, no factor was consistent across all follow-up years. Consistent with an earlier study, 12 the most robust of these findings was payor source, which was a significant predictor in each of the first 4 follow-up years. A potential explanation for this finding is that payor source is a proxy for economic resources and access and/or familiarity with the health care system. The Vos et al study
      • Vos L
      • Williams MW
      • Spielman L
      • et al.
      Understanding loss to follow-up in a longitudinal study of people with traumatic brain injury.
      grouped Medicare with other insurance because of a lower sample size and found a slightly high risk for LTFU in the other group (odds ratio, 2.55), which could be a regional effect. Contrary to prior research, violent cause of injury was associated with decreased odds of LTFU at some time points, which may have been the result of considering adjusted effects and using vehicular trauma as a comparison. While the effects we observed for other demographic factors are consistent with prior literature,
      • Jourdan C
      • Bayen E
      • Bahrami S
      • et al.
      Loss to follow-up and social background in an inception cohort of patients with severe traumatic brain injury: results from the PariS-TBI study.
      ,
      • Krellman JW
      • Kolakowsky-Hayner SA
      • Spielman L
      • et al.
      Predictors of follow-up completeness in longitudinal research on traumatic brain injury: findings from the National Institute on Disability and Rehabilitation Research Traumatic Brain Injury Model Systems program.
      ,
      • Vos L
      • Williams MW
      • Spielman L
      • et al.
      Understanding loss to follow-up in a longitudinal study of people with traumatic brain injury.
      our finding on the effect of problematic substance use was somewhat surprising, that is, we found a higher LTFU in those with an unknown substance use history than either those without or with known problematic substance use. We speculate this could be because of use of a proxy respondent who may be less aware of the substance use history of the survivor of TBI. This finding potentially corresponds with prior reports of missing data values posing a risk for later LTFU.
      • Krellman JW
      • Kolakowsky-Hayner SA
      • Spielman L
      • et al.
      Predictors of follow-up completeness in longitudinal research on traumatic brain injury: findings from the National Institute on Disability and Rehabilitation Research Traumatic Brain Injury Model Systems program.
      ,
      • Vos L
      • Williams MW
      • Spielman L
      • et al.
      Understanding loss to follow-up in a longitudinal study of people with traumatic brain injury.
      As with prior studies, these findings shed light on important considerations to be made when interpreting findings on TBI outcomes from longitudinal studies. For instance, given that individuals from socially disadvantaged backgrounds may be underrepresented in longitudinal TBI studies, conclusions regarding clinical outcomes may be biased, limiting their generalizability.
      • Jourdan C
      • Bayen E
      • Bahrami S
      • et al.
      Loss to follow-up and social background in an inception cohort of patients with severe traumatic brain injury: results from the PariS-TBI study.
      These findings also highlight specific subgroups in need of particular focus both in recruiting and retention in research studies, which may indicate a need for increased attention, institutional support, and communication for adherence in clinical care.
      • Jourdan C
      • Bayen E
      • Bahrami S
      • et al.
      Loss to follow-up and social background in an inception cohort of patients with severe traumatic brain injury: results from the PariS-TBI study.
      Specifically, Hispanic individuals, especially those without English as a first language, persons with less education, and those lacking private health insurance are at greatest risk for LTFU and may benefit from additional targeted retention strategies to prevent systemic bias.
      A recent meta-analysis examining the effectiveness of different retention strategies in longitudinal studies reported that strategies that reduced barriers to participation, such as offering alternative methods of data collection, adapting materials for mixed abilities, and extended data collection windows, were the strongest predictors of higher retention rates, as opposed to creating a project community, using reminders, or using tracing strategies.
      • Teague S
      • Youssef GJ
      • Macdonald JA
      • et al.
      Retention strategies in longitudinal cohort studies: a systematic review and meta-analysis.
      However, reducing barriers for in-person visits may require strategies different from reducing barriers for telephone visits.

      Study limitations

      The findings of this paper regarding attrition rates and predictors may be unique to the TBIMS database; because of its inclusion criteria and specific policies regarding follow-up, the findings may not generalize to other populations or even other longitudinal studies of individuals with TBI. There are unmeasured variables we did not examine that may affect attrition, such as geographic factors, technology availability, social determinants of health, mobility of participants, participant expectations at time of consent, changes in research personnel, and level of trust and rapport with research staff. Further investigation of these variables may help formulate specific, tailored interventions for those at highest risk for attrition.

      Conclusions

      This investigation of study retention in persons with TBI found that persons who were successfully followed at earlier time points and those with private insurance were most likely to be followed at subsequent time points. Persons from ethnic and racial minority groups, those born outside the US, those who did not primarily speak English, and those with less formal education were more likely to be lost to follow-up. It is essential to maximize success of early follow-up visits among persons with TBI in longitudinal studies, especially those from disadvantaged socioeconomic and ethnic and racial backgrounds, because attrition poses a threat to the validity and generalizability of findings. With a better understanding of risk factors for LTFU, programming can be developed to enhance retention of specific groups.

      Appendix. Supplementary materials

      References

        • Gustavson K
        • von Soest T
        • Karevold E
        • Røysamb E.
        Attrition and generalizability in longitudinal studies: findings from a 15-year population-based study and a Monte Carlo simulation study.
        BMC Public Health. 2012; 12: 1-11
        • Corrigan JD
        • Harrison-Felix C
        • Bogner J
        • Dijkers M
        • Terrill MS
        • Whiteneck G.
        Systematic bias in traumatic brain injury outcome studies because of loss to follow-up.
        Arch Phys Med Rehabil. 2003; 84: 153-160
        • Jourdan C
        • Bayen E
        • Bahrami S
        • et al.
        Loss to follow-up and social background in an inception cohort of patients with severe traumatic brain injury: results from the PariS-TBI study.
        J Head Trauma Rehabil. 2016; 31: E42-E48
        • Veerbeek JM
        • Kwakkel G
        • van Wegen EE
        • Ket JC
        • Heymans MW.
        Early prediction of outcome of activities of daily living after stroke: a systematic review.
        Stroke. 2011; 42: 1482-1488
        • Corrigan JD
        • Selassie AW
        • Orman JAL.
        The epidemiology of traumatic brain injury.
        J Head Trauma Rehabil. 2010; 25: 72-80
        • Langlois JA
        • Rutland-Brown W
        • Wald MM.
        The epidemiology and impact of traumatic brain injury: a brief overview.
        J Head Trauma Rehabil. 2006; 21: 375-378
        • Kim H
        • Cutter GR
        • George B
        • Chen Y.
        Understanding and preventing loss to follow-up: experiences from the Spinal Cord Injury Model Systems.
        Top Spinal Cord Inj Rehabil. 2018; 24: 97-109
      1. Harrison-Felix C, Dijkers M. Predictors of loss to follow-up in a longitudinal study of persons with traumatic brain injury. Detroit: Rehabilitation Institute of Michigan Del Harder Research Day; 1998.

        • Krellman JW
        • Kolakowsky-Hayner SA
        • Spielman L
        • et al.
        Predictors of follow-up completeness in longitudinal research on traumatic brain injury: findings from the National Institute on Disability and Rehabilitation Research Traumatic Brain Injury Model Systems program.
        Arch Phys Med Rehabil. 2014; 95: 633-641
        • Sander AM
        • Lequerica AH
        • Ketchum JM
        • et al.
        Race/ethnicity and retention in traumatic brain injury outcomes research: a Traumatic Brain Injury Model Systems National Database study.
        J Head Trauma Rehabil. 2018; 33: 219-227
        • Corrigan JD
        • Bogner JA
        • Mysiw WJ
        • Clinchot D
        • Fugate L.
        Systematic bias in outcome studies of persons with traumatic brain injury.
        Arch Phys Med Rehabil. 1997; 78: 132-137
        • Vos L
        • Williams MW
        • Spielman L
        • et al.
        Understanding loss to follow-up in a longitudinal study of people with traumatic brain injury.
        Brain Inj. 2021; 35: 1-9
      2. National Data and Statistical Center. Traumatic Brain Injury Model Systems National Data and Statistical Center. Available at: https://www.tbindsc.org/. Accessed July 5, 2021.

        • Lequerica AH
        • Botticello A
        • O'Neill J
        • et al.
        Relationship between Hispanic nativity, residential environment, and productive activity among individuals with traumatic brain injury: a TBI Model Systems study.
        J Head Trauma Rehabil. 2019; 34: E46-E54
        • Sander AM
        • Ketchum JM
        • Lequerica AH
        • et al.
        Primary language and participation outcomes in Hispanics with traumatic brain injury: a Traumatic Brain Injury Model Systems study.
        J Head Trauma Rehabil. 2021; 36: E218-E225
        • Pappadis MR
        • Sander AM
        • Struchen MA
        • Leung P
        • Smith DW.
        Common misconceptions about traumatic brain injury among ethnic minorities with TBI.
        J Head Trauma Rehabil. 2011; 26: 301-311
        • Asemota AO
        • George BP
        • Cumpsty-Fowler CJ
        • Haider AH
        • Schneider EB.
        Race and insurance disparities in discharge to rehabilitation for patients with traumatic brain injury.
        J Neurotrauma. 2013; 30: 2057-2065
        • Teague S
        • Youssef GJ
        • Macdonald JA
        • et al.
        Retention strategies in longitudinal cohort studies: a systematic review and meta-analysis.
        BMC Med Res Methodol. 2018; 18: 1-22