Towards PErsonalised PRognosis for children with traumatic brain injury: the PEPR study protocol
================================================================================================

* Cece C Kooper
* Jaap Oosterlaan
* Hilgo Bruining
* Marc Engelen
* Petra J W Pouwels
* Arne Popma
* Job B M van Woensel
* Dennis R Buis
* Marjan E Steenweg
* Maayke Hunfeld
* Marsh Königs

## Abstract

**Introduction** Traumatic brain injury (TBI) in children can be associated with poor outcome in crucial functional domains, including motor, neurocognitive and behavioural functioning. However, outcome varies between patients and is mediated by complex interplay between demographic factors, premorbid functioning and (sub)acute clinical characteristics. At present, methods to understand let alone predict outcome on the basis of these variables are lacking, which contributes to unnecessary follow-up as well as undetected impairments in children. Therefore, this study aims to develop prognostic models for the individual outcome of children with TBI in a range of important developmental domains. In addition, the potential added value of advanced neuroimaging data and the use of machine learning algorithms in the development of prognostic models will be assessed.

**Methods and analysis** 210 children aged 4–18 years diagnosed with mild-to-severe TBI will be prospectively recruited from a research network of Dutch hospitals. They will be matched 2:1 to a control group of neurologically healthy children (n=105). Predictors in the model will include demographic, premorbid and clinical measures prospectively registered from the TBI hospital admission onwards as well as MRI metrics assessed at 1 month post-injury. Outcome measures of the prognostic models are (1) motor functioning, (2) intelligence, (3) behavioural functioning and (4) school performance, all assessed at 6 months post-injury.

**Ethics and dissemination** Ethics has been obtained from the Medical Ethical Board of the Amsterdam UMC (location AMC). Findings of our multicentre prospective study will enable clinicians to identify TBI children at risk and aim towards a personalised prognosis. Lastly, findings will be submitted for publication in open access, international and peer-reviewed journals.

**Trial registration number** NL71283.018.19 and NL9051.

*   Magnetic resonance imaging
*   Paediatric intensive & critical care
*   Paediatric neurology
*   Paediatric radiology
*   Paediatric neurosurgery

#### Strengths and limitations of this study

*   This study will use a unique multidimensional approach to develop a more innovative personalised prognostic model to account for the heterogeneous outcome post paediatric traumatic brain injury (TBI).

*   The study design is optimised for clinical implementation by (1) selecting predictors of which the great majority is available at time of discharge from acute care, (2) aligning the timing of MRI assessment with follow-up according to clinical guidelines and (3) using outcome measures that are sensitive to impairments in a range of crucial domains of daily life functioning.

*   This study will test the added value of selected advanced MRI metrics that have shown promising prognostic potential for outcome in children with TBI.

*   This study will test the added value of machine learning approaches for the development of complex outcome prediction.

*   The resulting prognostic models will not be readily available for clinical practice, since external validation is required in order to assess clinical implementability in the hospital setting.

## Introduction

Traumatic brain injury (TBI) has an estimated annual worldwide prevalence of 69 million cases and is the leading cause of disability in children and young adults.1 2 The impact of TBI can have enduring effects on different aspects of daily life functioning, including motor, neurocognitive, behaviour and school functioning.3–8 Importantly, children show large differences in the nature and extent of TBI consequences, which are likely the result of the complex interplay between injury characteristics (ie, neuroimaging findings, severity of acute symptoms, vital parameters) and environmental factors (ie, premorbid functioning, socioeconomic status (SES), interventions).8–10 Due to the distinct heterogeneity in TBI and lack of good prognostic tools, clinicians are insufficiently able to properly inform the patient and family on expected outcome and are withheld to tailor care to the individual risk profile of the child.2 11 Considering the considerable morbidity and wide range of potential developmental disadvantages in children with TBI, better insights on TBI prognosis in children are much needed to improve appropriate family support, provide monitoring and intervention in a timely course, as well as effectively prevent poor outcome at the level of the individual child.12

In current clinical practice, widely used tools for head injuries include (1) the Glasgow Coma Scale score, (2) symptoms present in the acute phase (loss of consciousness, amnesia) and (3) if present, CT-based information that are combined and used for diagnostic purposes (stratifying into mild, moderate and severe TBI).13 14 Unfortunately, such tools are highly insufficient to predict the multifactorial outcome differences present across the spectrum of TBI severity.15–17 Several more advanced multivariate prognostic models for TBI outcome exist, yet suffer from important limitations for use in children.18 19 First, the vast majority defines outcome as ‘death’ or ‘severe disability’ instead of more fine-grained outcomes of threatened daily life functions (eg, motor, neurocognitive, behavioural and school functioning). Second, existing models have almost exclusively been developed in adult patients, thereby not accounting for the developmental aspects of brain functioning that are crucial for outcome prediction in children.20 Third, advanced multimodal MRI (ie, targeting brain volume, white matter integrity, structural and functional connectivity and neurometabolites) has not been integrated in the existing prediction models, while each of these MRI techniques has shown promising prognostic potential when studied in isolation.17 21–23 More specifically, measures of fractional anisotropy and resting-state network connectivity are considered to be strongly implicated in the neurocognitive and behavioural impairments of children with TBI.21 23 24

The limitations of existing prognostic models highlight the importance of research into the development of innovative prognostic models for outcome of paediatric TBI. Such models should be developed to move towards a more personalised prognosis. Given the complexity of TBI and its outcome,11 the development of accurate prognostic models is likely to require a rich source of multidimensional data that is brought down into a concise and clinically manageable set of predictors.10 Existing models have traditionally been developed using conventional statistical methods (eg, logistic and linear regression) which may not harvest the full predictive potential of rich data sources in complex real-life outcome prediction.25 Machine learning offers alternative models with high flexibility (eg, decision trees, support vector machines), allowing more accurate data modelling.25 Indeed recent application of machine learning in the prediction of global outcome after paediatric TBI has shown to improve the accuracy of prediction as compared with conventional statistical models.26 27 Yet to date, the value of machine learning for the development of prognosis on more fine-grained yet crucial outcome domains in multifactorial disease conditions such as TBI, remains largely unexplored.28 29

The current study aims to move towards personalised prognostic models for outcomes of paediatric TBI in crucial domains of child development (ie, motor, neurocognitive, behavioural and school functioning), based on multidimensional data covering premorbid and (sub)acute clinical characteristics. Furthermore, we aim to determine the added value of advanced neuroimaging metrics as well as machine learning algorithms for the development of a prognostic model. The primary result should be a prognostic model that may function as a practical tool for clinicians in daily care. Thereby, this study may contribute to a better family support, better planning of early rehabilitation and follow-up, preventing unnecessary care for children in whom good recovery is expected, and facilitate adequate monitoring and treatment of children with a high-risk of adverse outcome.

## Methods and analysis

### Study design

This controlled observational multicentre study will use a prospective longitudinal design. Children with a clinical diagnosis of TBI will be enrolled for 6 months after initial hospital admission. Data collection will take place at three time points: (1) during hospital stay, (2) 1 month post-injury (MRI assessment) and (3) 6 months post-injury (outcome assessment). The total duration of the study depends on the inclusion rate and is expected to be over 2 years (2021–2023). An overview of the study design and procedures for children with TBI is displayed in figure 1. Neurologically healthy children will be enrolled in the control group for outcome assessment only.

![Figure 1](http://bmjopen.bmj.com/https://bmjopen.bmj.com/content/bmjopen/12/6/e058975/F1.medium.gif)

[Figure 1](http://bmjopen.bmj.com/content/12/6/e058975/F1)

Figure 1 
The study design for children with TBI. TBI, traumatic brain injury.

### Study population

This study will prospectively recruit a multicentre cohort of children diagnosed with mild-to-severe TBI from a research network of hospitals. Trauma-level 1 Dutch University Medical Centres and general hospitals in the geographical area of the Amsterdam University Medical Centre (initiator site) qualify as a participating centre. Thereby, seeking to recruit a representative sample of children with TBI from primary school onwards. The inclusion and exclusion criteria for participation are displayed in box 1. We will use a clinical diagnosis of TBI instead of a research diagnosis of TBI for inclusion in the TBI group. Although this may lead to a more heterogeneous study sample of children with TBI due to practice variation in the adherence to national guidelines stipulating criteria for assessment and treatment of mild TBI,30 this will also result in a study sample that better represents the clinical population of children with TBI. Exclusion criteria relating to very poor motor or cognitive outcome (exclusion criteria 2 and 3) will be registered as an outcome, to investigate the potential bias that may be introduced by lack of outcome assessment in a specific subsample of children with very poor outcome. Inclusion will be complete when a sample of 210 children with TBI has been included (see Sample size estimation). Demographically matched neurologically healthy children will be recruited, mainly via schools, out-of-school care facilities, sports clubs and through existing collaborations with healthcare institutions in the geographical area of the participating centres. Children in the control group will be matched to children with TBI on age, sex and SES,31 with a 1:2 ratio requiring a control sample of 105 children.

Box 1 
### Inclusion and exclusion criteria for study participation

#### Inclusion criteria

*   Inhabitant of the Netherlands.

*   Fluent in the Dutch language.

*   4–18 years old.

*   No documented and/or parent-reported diagnosis of a neurological disorder (other than TBI*).

*   A clinical diagnosis of mild-to-severe TBI according to a paediatrician or paediatric neurologist.*

#### Exclusion criteria

*   Absence or withdrawal of written informed consent.

*   Severe motor disability that interferes with outcome assessment at the time of assessment.

*   Inability to comprehend testing instructions at the time of assessment.

*   Somatic disorders unrelated to TBI that affect outcome assessments at the time of assessment.*

### Patient and public involvement

No patient involved.

### Sample size estimation

Sample size calculation was performed according to EMGO+ guidelines,32 which universally apply to the candidate methods that are described for model development. Consequently, 10–15 observations are required per predictor in the model (15 was chosen for a liberal calculation of the required sample size). The minimum required sample size was calculated for an advanced, yet clinically relevant and implementable prognostic model. Hence, the model complexity was set to a maximum of 10 predictors, in turn defining the minimum required sample at (10×15=) 150 children. Considering Dutch medical ethical guidelines for clinical research, children as from 8 years of age are eligible for MRI scanning in research. Therefore, we set the minimum required sample size (150 children) as the target sample size for the MRI subsample of children (8–18 years old). Considering that we also aim to recruit an age-balanced TBI sample, we calculated the target sample of children with TBI per age year in the MRI subsample aged 8–18 years old (150/(18 years–8 years)=15 children per age year) and applied this to the age range of the whole study sample aged 4–18 years old (15×(18 years–4 years)=210 children) to arrive at the target sample size in the whole study sample of N=210. The resulting sample sizes allow detecting small-to-medium-sized group differences (*f*=0.18), assuming a statistical power of 80%, alpha set at 0.05 and two-sided testing using analysis of variance.33

### Protocol

#### Measurements during hospital admission (TBI children only)

All children admitted to a participating hospital with the clinical diagnosis of mild-to-severe TBI will be screened for eligibility by the on-call paediatric neurologist. Children and their parents (ie, legal guardian) will be informed on the study by the researcher, potential questions will be answered and appropriate informed consent procedures will be conducted depending on age and/or incapacitation of the participant (following article 3, 4, 6 and 9 of the Medical Research Involving Human Subjects Act). Then, participants and/or parent(s) will be asked to fill out questionnaires on demographics (10 min) and premorbid functioning (15 min). Questions on premorbid functioning (assessing family34 and behavioural35 functioning) will be collected at the time of hospital admission to limit the contamination of assessment of premorbid functioning with potential consequences of TBI. The chosen questionnaires will lend beneficial insights in possible mediating factors of outcome, assessing the presence of a social support system and allows adjusting for SES, which is known to be important for TBI recovery.36 SES will be defined as the average level of parental education ranging from 1 (no education) to 8 (postdoctoral education).37 A multidisciplinary set of clinical predictors will be collected during hospital stay using standardised forms for nurses and physicians, integrated in the electronic medical record. The forms will strictly adhere to relevant care guidelines,38–40 thereby facilitating systematic prospective data collection as well as contributing to clinical practice. Table 1 provides an overview of all study measures that will be recorded during hospital admission. See online supplemental appendix 1 for a full listing of clinical measures that will be collected.

### Supplementary data

[[bmjopen-2021-058975supp001.pdf]](pending:yes)

View this table:
[Table 1](http://bmjopen.bmj.com/content/12/6/e058975/T1)

Table 1 
Demographic, premorbid and clinical measures

### MRI at 1 month post-injury (TBI children aged ≥8 years only)

According to Dutch medical ethical guidelines for scientific research, MRI will only be collected in children aged ≥8 years with negative screening for MRI contraindications. For eligible participants, one MRI session will be planned at 1 month post-injury with a 2-week time window at the Spinoza Centre for Neuroimaging, situated at the campus of the Amsterdam UMC. The chosen time point of the MRI assessment reflects a compromise between early measurement and the potentially confounding influence of brain oedema on advanced neuroimaging during the acute phase.41 Moreover, this time window aligns with routine follow-up of children after hospital admission for TBI according to the Dutch clinical guideline38 and enables MRI assessment of children with more severe injuries. At time of visit, actual MRI acquisition will take 40 min using a Philips 3T Achieva. The scanning protocol includes both conventional and advanced MRI scan types, all displayed in table 2 together with the accommodating predictors that will be extracted from the data. Moreover, we will compare the prognostic value of promising experimental MRI scans (ie, Diffusion Tensor Imaging, Resting-State Functional MRI, Magnetic Resonance Spectroscopy) to conventional CT and MRI (ie, T1 and Susceptible Weighted Imaging). Diffusion Tensor Imaging and Resting-State Functional MRI will be used to extract measures of structural and functional connectivity.24 42 43 Single Voxel Magnetic Resonance Spectroscopy using Point RESolved Spectroscopy (PRESS, positioned in the Corpus Callosum) will be used as a non-invasive measure to quantify neurometabolite levels. This measure complements the assessment of structural and functional connectivity and has previously shown to be relevant for neurocognitive outcome.44

View this table:
[Table 2](http://bmjopen.bmj.com/content/12/6/e058975/T2)

Table 2 
MRI (3T) scanning protocol

### Functional outcome assessment

At 6 months post-injury (TBI group) or after obtaining informed consent (control group) functional outcome will be assessed in a standardised manner by research assistants in the participating hospitals with an estimated duration of 1½ hours. An overview of the measures of functional outcome is displayed in table 3. To thank children for participation they will be given a small present after being debriefed and travel expenses will be reimbursed. Participants can choose out of a small selection of age-appropriate presents (worth around €5—for children aged 4–11 years, eg, colouring books, wooden games and worth around €10—for children aged >12 years, eg, sports attributes, card games).

View this table:
[Table 3](http://bmjopen.bmj.com/content/12/6/e058975/T3)

Table 3 
Measures of functional outcome

Motor functioning will be assessed using the Movement Assessment Battery for Children (second Dutch edition; M-ABC-2).45 The M-ABC-2 is a standardised and widely used test battery to assess motor skills in children aged 3–16 years but also allows measurement of motor skills in adolescents.46 Nevertheless, we will explore potential ceiling effects in participants aged 17 and 18. The M-ABC-2 contains eight items measuring: (1) manual dexterity, (2) throwing and catching and (3) balance. The total score is the sum of the three components and is transformed into age-adjusted standard scores, indicative as an overall measure of motor functioning. The test has adequate psychometric properties45 and has an estimated total duration of 20–40 min (depending on the child’s age).

Intelligence will be assessed using the revised Dutch Wechsler Intelligence Scales.47–49 Depending on the child’s age either the Wechsler Preschool Primary Scale of Intelligence, Wechsler Intelligence Scale for Children or Wechsler Adult Intelligence Scale will be assessed. For all versions, age-adjusted full-scale IQ will be estimated using a short form (assessing the subtests Vocabulary, Similarities, Matrix Reasoning and Block Design), with adequate validity and reliability in estimating intelligence.50 The short form has a duration of approximately 45 min.

Parents will fill out the Dutch version of the Child Behaviour Checklist (CBCL), which is a widely used measure for behavioural and emotional problems focusing on the past 6 months of children.51 Either the preschool or school-age version of the CBCL will be completed depending on the age of the child. The questionnaires contain 100-items (preschool) to 113-items (school-age), providing a total score, broadband scales and small band scales. The broadband scales discriminate between externalising and internalising problems. The small bands discriminate between numerous types of behavioural problems such as somatic disorders, anxious/depressed, social problems, attentional problems and aggressive behaviour. An adaption of the Dutch preschool and school-age version of the CBCL, the Teacher Report Forms (TRFs), will be completed by teachers and allow direct comparison with outcomes from the CBCL.51 For patients that have not returned to school and stay in an inpatient or residential programme, the TRF will be completed by the daily care medical staff, if possible. Both CBCL and TRF typically take 20 min to complete.

School functioning will be assessed in the subsample of children attending primary school. Dutch Pupil Monitoring System52 results will be requested through primary school teachers and include information prior to the injury as well as 6 months-post injury. The Dutch Pupil Monitoring System developed by the National Institute of Educational Measurement in the Netherlands is to obtain reliable data systematically on pupil learning progress during their entire primary school career.53 Test packages are developed for all six age groups between 6 and 12 years-old and allow seamless charting of academic development across these ages. We will assess packages developed for arithmetic’s, spelling and technical reading.54 Based on the expected age range of this group (6–12 years), this information will be available for a subsample of (3.5×30=) 105 children with TBI. The size of this subsample allows building a separate highly relevant prediction model for school outcome with a maximum of seven predictors.

### Analyses

#### Data preprocessing and score constructions

Outliers (z-score >3.29 or z-score <−3.29) will be identified in all variables and will be assessed for measurement errors carefully. If no evidence is found for a measurement error, outliers will be rescaled using Winsorizing.55 Variables with missing values >10% will be discarded from further analysis. Data missing at random or completely at random will be imputed using multiple imputations. Voluminous data (eg, continuous measurements of vital signs) will be reduced using principal components analysis and/or classification/regression trees to control the number of available predictors for the model.

Each age and sex standardised functional outcome score (motor development, intelligence, behavioural functioning and school performance) will first be transformed to z-scores, where the z-score describes the difference between each TBI participant’s score and the mean of the demographically-matched control participant. Second, the z-scores will be adjusted for the influence of premorbid functioning (family34 and behavioural35 functioning) by adding these variables as predictors to the linear regression analyses on each z-score pertaining to an outcome domain. The demographic and premorbid adjusted z-scores will then be retrieved by extracting the standardised residuals of these regression analyses. Since children with high premorbid functioning and significant decrement in functioning can still perform in the average range of the general population, the demographic and premorbid adjustment procedure will increase the sensitivity of outcome prediction.

All MRI data will be preprocessed using the Functional MRI of the Brain Software Library (FSL).56 T1 data will be assessed using volumetric analysis.57 With regard to Diffusion Tensor Imaging data, preprocessing will involve correction for motion and eddy-currents and automated imputation of volume data.58 White matter microstructure will be assessed by the primary diffusion measures (fractional anisotropy and mean diffusion). Spatial correlations between white matter microstructure and outcome measures will be assessed using tract-based spatial statistics as well as voxel-based white matter parameters, using Diffusion Tensor Imaging ToolKit registration.59 60 Structural connectivity will be assessed using probabilistic fibre tracking on the diffusion data.61 Functional connectivity will be assessed using temporal correlations in brain activity between brain areas.62 Global and local network parameters of structural as well as functional connectivity will be extracted from the resulting connectivity matrices using the application of graph theory.63 All spectroscopy data will be processed using the LCModel package.64 The spectroscopy data will be used to extract the concentrations of the following metabolites sensitive to TBI depending on quantification reliability within the population, possibly including *N*-acetyl aspartate, choline, myo-inositol, creatine, glutamine and glutamate.44 65 Data reduction and predictor selection techniques will be used to handle the large number of MRI-derived predictors per subject (ie, ensemble averaging, independent component analysis and selective regularisation of MRI data).

### Prognostic model development

#### Primary analyses: multivariate models for functional outcome

Statistical analyses will be performed using R and SPSS with alpha set at 0.05 (two-sided). Model development will be performed according to Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) guidelines.66 Prognostic models will be developed for each of the four functional outcomes separately, all using predictors that were available at the moment of discharge. In addition, one prognostic model will be developed to predict the overall functional outcome (sum of z-scores) to identify children with general functional impairment.

Prediction models are anticipated to be developed using different types of candidate supervised machine learning techniques, among which decision trees and support vector machines. A reference model will be constructed using linear regression. Given the rapid developments in the field of data science, we will adapt our specific selection of candidate models to the state of the literature at the time of analysis. Model complexity will be determined using cross-validation (with a maximum of 10 predictors based on the minimum sample size calculation). Model performance will be assessed according to ABCD (ie, A calibration-in-the-large, or the model intercept; B calibration slope; C discrimination, with a concordance statistic; D and clinical usefulness, with decision-curve analysis) guidelines using measures of calibration and discrimination.67 The entire data set will be used for model training as recommended for smaller clinical samples,68 therefore internal validation will be performed using the bootstrap method and model performance will be corrected for optimism accordingly.69 Ultimately, the complete methodological process will be reported in the dissemination of the data and the best performing model will be presented.

#### Secondary analyses: additive value of MRI metrics and machine learning

Additional analyses are aimed at assessing the additive prognostic value of innovative MRI metrics (as compared with conventional CT and MRI metrics) and the additive value of machine learning based methods (as compared with linear regression). The value of advanced MRI metrics for outcome prediction will be tested in the MRI subsample (n=150) by the change in prediction model performance after adding advanced MRI metrics to the available predictors. Differences in model performance will be analysed using bootstrap CIs created for three widely used performance measures for regression-based prediction: root mean squared error, explained variance and mean absolute error.70 Then, the value of machine learning techniques for the development of personalised prognostic models will be compared with the reference model created with linear regression, also using the bootstrap CIs on the same performance measures as previously mentioned.71

### Ethics and dissemination

This study poses a negligible risk to the participating children and their parents. Study participation will not restrict any received clinical care as determined by physicians (additional CT or MRI, assessments or follow-up). All study procedures will be conducted according to the principles of the Declaration of Helsinki (2013) and will follow the Medical Research Involving Human Subjects Act (WMO). Participation in the study is voluntary and participants can leave the study at any time for any reason. Leaving the study will be without any consequences for clinical care. On completion of all study measures for all participants, we will provide families interested in the results of the study with a concise report. In addition, on request, families can retrieve a report with individual outcomes for measures with readily available normative data (eg, subtest of the Wechsler Intelligence Scale). The research data including a manuscript will be published in international peer-reviewed journals, preferably open-access. Publication topics will include (1) the development of prognostic models for functional outcome 6 months post TBI in children aged 4–18 years, (2) the relevance of neuroimaging metrics for functional outcome post TBI as well as (3) the potentially added value of machine learning as compared with conventional analyses for clinical prediction models.

## Ethics statements

### Patient consent for publication

Not applicable.

## Footnotes

*   Contributors CCK, JO, HB, ME and MK led the study concept and design, selected outcome measures and were involved in writing of the manuscript. PJWP helped with the development and selection of neuroimaging scans. CCK, ME, AP, JvW, DRB, MES and MH were (and continue to be) involved in data acquisition of the study. All authors read, critically revised and approved the final version of the manuscript.

*   Funding This work was supported by the Amsterdam University Medical Centre (location AMC) through the Department of General Pediatrics. Subsidising parties were Cornelia Stichting (no award/grant number), Janivo Stichting (2017576), Dr CJ Vaillantfonds (no award/grant number) and Amsterdam Research & Development (V.000296).

*   Competing interests None declared.

*   Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.

*   Provenance and peer review Not commissioned; externally peer reviewed.

*   Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.

[http://creativecommons.org/licenses/by-nc/4.0/](http://creativecommons.org/licenses/by-nc/4.0/)
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: [http://creativecommons.org/licenses/by-nc/4.0/](http://creativecommons.org/licenses/by-nc/4.0/).

## References

1.  Dewan MC, Rattani A, Gupta S. Estimating the global incidence of traumatic brain injury. J Neurosurg 2019;130:1080–97.
    
    [CrossRef](http://bmjopen.bmj.com/lookup/external-ref?access_num=10.3171/2017.10.jns17352&link_type=DOI) 

2.  Suskauer SJ, Houtrow AJ. Invited Commentary on "The Report to Congress on the Management of Traumatic Brain Injury in Children". Arch Phys Med Rehabil 2018;99:1–90.[doi:10.1016/j.apmr.2018.05.012](http://dx.doi.org/10.1016/j.apmr.2018.05.012)pmid:http://www.ncbi.nlm.nih.gov/pubmed/29906420
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 

3.  Nelson LD, Ranson J, Ferguson AR. Validating multi-dimensional outcome assessment using the traumatic brain injury common data elements: an analysis of the TRACK-TBI pilot study sample. J Neurotrauma 2017;34:3158–72.
    
    

4.  Rosema S, Crowe L, Anderson V. Social function in children and adolescents after traumatic brain injury: a systematic review 1989-2011. J Neurotrauma 2012;29:1277–91.[doi:10.1089/neu.2011.2144](http://dx.doi.org/10.1089/neu.2011.2144)pmid:http://www.ncbi.nlm.nih.gov/pubmed/22260408
    
    [CrossRef](http://bmjopen.bmj.com/lookup/external-ref?access_num=10.1089/neu.2011.2144&link_type=DOI) 
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=22260408&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 
    
    [Web of Science](http://bmjopen.bmj.com/lookup/external-ref?access_num=000303384600001&link_type=ISI) 

5.  Moen KT, Jørgensen L, Olsen A, et al. High-level mobility in chronic traumatic brain injury and its relationship with clinical variables and magnetic resonance imaging findings in the acute phase. Arch Phys Med Rehabil 2014;95:1838–45.[doi:10.1016/j.apmr.2014.04.014](http://dx.doi.org/10.1016/j.apmr.2014.04.014)pmid:http://www.ncbi.nlm.nih.gov/pubmed/24814461
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 

6.  Babikian T, Asarnow R. Neurocognitive outcomes and recovery after pediatric TBI: meta-analytic review of the literature. Neuropsychology 2009;23:283–96.[doi:10.1037/a0015268](http://dx.doi.org/10.1037/a0015268)pmid:http://www.ncbi.nlm.nih.gov/pubmed/19413443
    
    [CrossRef](http://bmjopen.bmj.com/lookup/external-ref?access_num=10.1037/a0015268&link_type=DOI) 
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=19413443&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 
    
    [Web of Science](http://bmjopen.bmj.com/lookup/external-ref?access_num=000265662000002&link_type=ISI) 

7.  Li L, Liu J. The effect of pediatric traumatic brain injury on behavioral outcomes: a systematic review. Dev Med Child Neurol 2013;55:37–45.[doi:10.1111/j.1469-8749.2012.04414.x](http://dx.doi.org/10.1111/j.1469-8749.2012.04414.x)pmid:http://www.ncbi.nlm.nih.gov/pubmed/22998525
    
    [CrossRef](http://bmjopen.bmj.com/lookup/external-ref?access_num=10.1111/j.1469-8749.2012.04414.x&link_type=DOI) 
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=22998525&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 

8.  JA V, Babikian T, Asarnow RF. Academic and language outcomes in children after traumatic brain injury a meta analysis. Except Child 2011;77:263–81.
    
    [Web of Science](http://bmjopen.bmj.com/lookup/external-ref?access_num=000287979400002&link_type=ISI) 

9.  Ryan NP, van Bijnen L, Catroppa C, et al. Longitudinal outcome and recovery of social problems after pediatric traumatic brain injury (TBI): contribution of brain insult and family environment. Int J Dev Neurosci 2016;49:23–30.[doi:10.1016/j.ijdevneu.2015.12.004](http://dx.doi.org/10.1016/j.ijdevneu.2015.12.004)pmid:http://www.ncbi.nlm.nih.gov/pubmed/26739435
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 

10. Au AK, Clark RSB. Paediatric traumatic brain injury: prognostic insights and outlooks. Curr Opin Neurol 2017;30:565–72.[doi:10.1097/WCO.0000000000000504](http://dx.doi.org/10.1097/WCO.0000000000000504)pmid:http://www.ncbi.nlm.nih.gov/pubmed/28938340
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 

11. Covington NV, Duff MC. Heterogeneity is a hallmark of traumatic brain injury, not a limitation: a new perspective on study design in rehabilitation research. Am J Speech Lang Pathol 2021;30:974-985.[doi:10.1044/2020\_AJSLP-20-00081](http://dx.doi.org/10.1044/2020_AJSLP-20-00081)pmid:http://www.ncbi.nlm.nih.gov/pubmed/33556261
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 

12. McCrea MA, Manley GT. State ofthe science on pediatric mild traumatic brain injury progress toward clinical translation. JAMA Pediatr 2018;172:141–56.
    
    

13. Marmarou A, Lu J, Butcher I, et al. Prognostic value of the Glasgow coma scale and pupil reactivity in traumatic brain injury assessed pre-hospital and on enrollment: an impact analysis. J Neurotrauma 2007;24:270–80.[doi:10.1089/neu.2006.0029](http://dx.doi.org/10.1089/neu.2006.0029)pmid:http://www.ncbi.nlm.nih.gov/pubmed/17375991
    
    [CrossRef](http://bmjopen.bmj.com/lookup/external-ref?access_num=10.1089/neu.2006.0029&link_type=DOI) 
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=17375991&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 
    
    [Web of Science](http://bmjopen.bmj.com/lookup/external-ref?access_num=000245187900006&link_type=ISI) 

14. Kuppermann N, Holmes JF, Dayan PS, et al. Identification of children at very low risk of clinically-important brain injuries after head trauma: a prospective cohort study. Lancet 2009;374:1160–70.[doi:10.1016/S0140-6736(09)61558-0](http://dx.doi.org/10.1016/S0140-6736(09)61558-0)pmid:http://www.ncbi.nlm.nih.gov/pubmed/19758692
    
    [CrossRef](http://bmjopen.bmj.com/lookup/external-ref?access_num=10.1016/S0140-6736(09)61558-0&link_type=DOI) 
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=19758692&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 
    
    [Web of Science](http://bmjopen.bmj.com/lookup/external-ref?access_num=000270688200032&link_type=ISI) 

15. Mittl RL, Grossman RI, Hiehle JF, et al. Prevalence of MR evidence of diffuse axonal injury in patients with mild head injury and normal head CT findings. AJNR Am J Neuroradiol 1994;15:1583–9.pmid:http://www.ncbi.nlm.nih.gov/pubmed/7985582
    
    [Abstract/FREE Full Text](http://bmjopen.bmj.com/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6NDoiYWpuciI7czo1OiJyZXNpZCI7czo5OiIxNS84LzE1ODMiO3M6NDoiYXRvbSI7czoyNjoiL2Jtam9wZW4vMTIvNi9lMDU4OTc1LmF0b20iO31zOjg6ImZyYWdtZW50IjtzOjA6IiI7fQ==) 

16. Sigmund GA, Tong KA, Nickerson JP, et al. Multimodality comparison of neuroimaging in pediatric traumatic brain injury. Pediatr Neurol 2007;36:217–26.[doi:10.1016/j.pediatrneurol.2007.01.003](http://dx.doi.org/10.1016/j.pediatrneurol.2007.01.003)pmid:http://www.ncbi.nlm.nih.gov/pubmed/17437903
    
    [CrossRef](http://bmjopen.bmj.com/lookup/external-ref?access_num=10.1016/j.pediatrneurol.2007.01.003&link_type=DOI) 
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=17437903&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 

17. Konigs M, Pouwels P, van Heurn L. Relevance of neuroimaging for neurocognitive and behavioral outcome after pediatric traumatic brain injury. Brain Imaging Behav 2017:1–12.
    
    

18. Perel P, Edwards P, Wentz R, et al. Systematic review of prognostic models in traumatic brain injury. BMC Med Inform Decis Mak 2006;6:1–10.
    
    [CrossRef](http://bmjopen.bmj.com/lookup/external-ref?access_num=10.1186/1472-6947-6-1&link_type=DOI) 
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=16396679&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 

19. Huth SF, Slater A, Waak M, et al. Predicting neurological recovery after traumatic brain injury in children: a systematic review of prognostic models. J Neurotrauma 2020;37:2141–9.[doi:10.1089/neu.2020.7158](http://dx.doi.org/10.1089/neu.2020.7158)pmid:http://www.ncbi.nlm.nih.gov/pubmed/32460675
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 

20. Anderson V, Spencer-Smith M, Wood A. Do children really recover better? Neurobehavioural plasticity after early brain insult. Brain 2011;134:2197–221.[doi:10.1093/brain/awr103](http://dx.doi.org/10.1093/brain/awr103)pmid:http://www.ncbi.nlm.nih.gov/pubmed/21784775
    
    [CrossRef](http://bmjopen.bmj.com/lookup/external-ref?access_num=10.1093/brain/awr103&link_type=DOI) 
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=21784775&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 
    
    [Web of Science](http://bmjopen.bmj.com/lookup/external-ref?access_num=000294067700009&link_type=ISI) 

21. Bonnelle V, Leech R, Kinnunen KM, et al. Default mode network connectivity predicts sustained attention deficits after traumatic brain injury. J Neurosci 2011;31:13442–51.[doi:10.1523/JNEUROSCI.1163-11.2011](http://dx.doi.org/10.1523/JNEUROSCI.1163-11.2011)pmid:http://www.ncbi.nlm.nih.gov/pubmed/21940437
    
    [Abstract/FREE Full Text](http://bmjopen.bmj.com/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6Njoiam5ldXJvIjtzOjU6InJlc2lkIjtzOjExOiIzMS8zOC8xMzQ0MiI7czo0OiJhdG9tIjtzOjI2OiIvYm1qb3Blbi8xMi82L2UwNTg5NzUuYXRvbSI7fXM6ODoiZnJhZ21lbnQiO3M6MDoiIjt9) 

22. Palacios EM, Sala-Llonch R, Junque C, et al. Resting-state functional magnetic resonance imaging activity and connectivity and cognitive outcome in traumatic brain injury. JAMA Neurol 2013;70:845–51.[doi:10.1001/jamaneurol.2013.38](http://dx.doi.org/10.1001/jamaneurol.2013.38)pmid:http://www.ncbi.nlm.nih.gov/pubmed/23689958
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 

23. Dennis EL, Caeyenberghs K, Hoskinson KR, et al. White matter disruption in pediatric traumatic brain injury. Neurology 2021;97:e298–309.[doi:10.1212/WNL.0000000000012222](http://dx.doi.org/10.1212/WNL.0000000000012222)
    
    

24. Königs M, Pouwels PJ, Ernest van Heurn LW, et al. Relevance of neuroimaging for neurocognitive and behavioral outcome after pediatric traumatic brain injury. Brain Imaging Behav 2018;12:29–43.[doi:10.1007/s11682-017-9673-3](http://dx.doi.org/10.1007/s11682-017-9673-3)pmid:http://www.ncbi.nlm.nih.gov/pubmed/28092022
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 

25. James G, Witten D, Hastie T. An introduction to statistical learning. Springer Texts 2006;102.
    
    

26. Hale AT, Stonko DP, Brown A, et al. Machine-learning analysis outperforms conventional statistical models and CT classification systems in predicting 6-month outcomes in pediatric patients sustaining traumatic brain injury. Neurosurg Focus 2018;45:1–7.[doi:10.3171/2018.8.FOCUS17773](http://dx.doi.org/10.3171/2018.8.FOCUS17773)pmid:http://www.ncbi.nlm.nih.gov/pubmed/30453455
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 

27. Tunthanathip T, Oearsakul T. Application of machine learning to predict the outcome of pediatric traumatic brain injury. Chin J Traumatol 2021;24:350-355.[doi:10.1016/j.cjtee.2021.06.003](http://dx.doi.org/10.1016/j.cjtee.2021.06.003)pmid:http://www.ncbi.nlm.nih.gov/pubmed/34284922
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 

28. Iniesta R, Stahl D, McGuffin P. Machine learning, statistical learning and the future of biological research in psychiatry. Psychol Med 2016;46:2455–65.[doi:10.1017/S0033291716001367](http://dx.doi.org/10.1017/S0033291716001367)pmid:http://www.ncbi.nlm.nih.gov/pubmed/27406289
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 

29. Hoodbhoy Z, Masroor Jeelani S, Aziz A, et al. Machine learning for child and adolescent health: a systematic review. Pediatrics 2021;147. doi:[doi:10.1542/peds.2020-011833](http://dx.doi.org/10.1542/peds.2020-011833). [Epub ahead of print: 15 12 2020].pmid:http://www.ncbi.nlm.nih.gov/pubmed/33323492
    
    

30. NVK. Richtlijn Opvang Van Patiënten Met Licht Traumatisch Hoofd / Hersenletsel 2010;130.
    
    

31. Strenze T. Intelligence and socioeconomic success: a meta-analytic review of longitudinal research. Intelligence 2007;35:401–26.[doi:10.1016/j.intell.2006.09.004](http://dx.doi.org/10.1016/j.intell.2006.09.004)
    
    [CrossRef](http://bmjopen.bmj.com/lookup/external-ref?access_num=10.1016/j.intell.2006.09.004&link_type=DOI) 
    
    [Web of Science](http://bmjopen.bmj.com/lookup/external-ref?access_num=000249214800002&link_type=ISI) 

32. EMGO+. Progostic & Diagnostic Tests. Quality Handbook v 2.0.
    
    

33. American Statistical Association. G power 3.1 manual 2017.
    
    

34. Veerman JW, Janssen J, Kroes G, et al. ‘Vragenlijst Gezinsfunctioneren volgens Ouders (VGFO). Handleiding.’ 2012.
    
    

35. van Widenfelt BM, Goedhart AW, Treffers PDA, et al. Dutch version of the strengths and difficulties questionnaire (SDQ). Eur Child Adolesc Psychiatry 2003;12:281–9.[doi:10.1007/s00787-003-0341-3](http://dx.doi.org/10.1007/s00787-003-0341-3)pmid:http://www.ncbi.nlm.nih.gov/pubmed/14689260
    
    [CrossRef](http://bmjopen.bmj.com/lookup/external-ref?access_num=10.1007/s00787-003-0341-3&link_type=DOI) 
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=14689260&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 
    
    [Web of Science](http://bmjopen.bmj.com/lookup/external-ref?access_num=000187508000004&link_type=ISI) 

36. Catroppa C, Anderson VA, Beauchamp M, et al. New frontiers in pediatric traumatic brain injury: an evidence base for clinical practice. Taylor & Francis, 2016.
    
    

37. Statistics Netherlands. Standaard Onderwijsindeling 2006. education categorization standard.
    
    

38. Licht Traumatisch Hoofd/Hersenletsel. Nederlandse Vereniging voor Neurologie 2010.
    
    

39. Acute Neurologie Bij Een Licht Traumatisch Hoofd/Hersenletsel. Nederlandse Vereniging voor Neurologie.
    
    

40. ATLS SubcommitteeAmerican College of Surgeons’ Committee on TraumaInternational ATLS working group. Advanced trauma life support (ATLS®): the ninth edition. J Trauma Acute Care Surg 2013;74:1363–6.[doi:10.1097/TA.0b013e31828b82f5](http://dx.doi.org/10.1097/TA.0b013e31828b82f5)pmid:http://www.ncbi.nlm.nih.gov/pubmed/23609291
    
    [CrossRef](http://bmjopen.bmj.com/lookup/external-ref?access_num=10.1097/TA.0b013e31828b82f5&link_type=DOI) 
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=23609291&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 
    
    [Web of Science](http://bmjopen.bmj.com/lookup/external-ref?access_num=000319316600034&link_type=ISI) 

41. Roberts R, Mathias J, Rose S. Dti) findings following pediatric non-penetrating TBI: a meta-analysis. Dev Neuropsychol 2014;39:600–37.
    
    

42. Moreira da Silva N, Cowie CJA, Blamire AM, et al. Investigating brain network changes and their association with cognitive recovery after traumatic brain injury: a longitudinal analysis. Front Neurol 2020;11:1–11.
    
    [CrossRef](http://bmjopen.bmj.com/lookup/external-ref?access_num=10.3389/fneur.2020.00001&link_type=DOI) 
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=32116995&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 

43. Königs M, van Heurn LWE, Bakx R, et al. The structural connectome of children with traumatic brain injury. Hum Brain Mapp 2017;38:3603–14.[doi:10.1002/hbm.23614](http://dx.doi.org/10.1002/hbm.23614)pmid:http://www.ncbi.nlm.nih.gov/pubmed/28429381
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 

44. Babikian T, Alger JR, Ellis-blied MU. Whole brain magnetic resonance spectroscopic determinants of functional outcomes 2018;1645:1637–45.
    
    

45. Smits-Engelsman B. Movement ABC; Nederlandse Handleiding [Dutch Manual Movement ABC]. Lisse, The Netherlands: Swets, Zeitlinger, 1998.
    
    

46. Husby IM, Skranes J, Olsen A, et al. Motor skills at 23 years of age in young adults born preterm with very low birth weight. Early Hum Dev 2013;89:747–54.[doi:10.1016/j.earlhumdev.2013.05.009](http://dx.doi.org/10.1016/j.earlhumdev.2013.05.009)pmid:http://www.ncbi.nlm.nih.gov/pubmed/23810435
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 

47. Wechsler D. Wechsler adult intelligence Scale-Fourth edition (WAIS-IV). San Antonio: TX NCS Pearson, 2008.
    
    

48. Wechsler D. Wechsler preschool and primary scale of intelligence. Encycl Autism Spectr Disord 2021:5172–81.
    
    

49. Wechsler D. Wechsler intelligence scale for children–Fifth edition (WISC-V). Bloom MN Pearson, 2014.
    
    

50. Sattler JM. Assessment of children: cognitive foundations. JM Sattler San Diego, CA, 2008.
    
    

51. Verhulst F, van der Ende J. Handeling ASEBA Vragenlijsten Voor Leeftijden 6 t/m 18 Jaar: CBCL/6-18, YSR & TRF. ASEBA, 2013.
    
    

52. Gilijns P, Verhoeven L. Het CITO leerlingvolgsysteem: Met het oog op de praktijk [The CITO pupil monitoring system: Focus on practice]. Pedagog Stud 1992.
    
    

53. Vlug KFM. Because every pupil counts: the success of the pupil monitoring system in the Netherlands. Educ Inf Technol 1997;2:287–306.[doi:10.1023/A:1018629701040](http://dx.doi.org/10.1023/A:1018629701040)
    
    

54. Glas CAW, Geerlings H. Psychometric aspects of pupil monitoring systems. Stud Educ Eval 2009;35:83–8.[doi:10.1016/j.stueduc.2009.10.006](http://dx.doi.org/10.1016/j.stueduc.2009.10.006)
    
    

55. Tabachnick BG, Fidell LS. Using multivariate statistics. Essentials Polit Res 2020;7:173–208.
    
    

56. Jenkinson M, Beckmann CF, Behrens TEJ, et al. FSL. Neuroimage 2012;62:782–90.[doi:10.1016/j.neuroimage.2011.09.015](http://dx.doi.org/10.1016/j.neuroimage.2011.09.015)pmid:http://www.ncbi.nlm.nih.gov/pubmed/21979382
    
    [CrossRef](http://bmjopen.bmj.com/lookup/external-ref?access_num=10.1016/j.neuroimage.2011.09.015&link_type=DOI) 
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=21979382&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 
    
    [Web of Science](http://bmjopen.bmj.com/lookup/external-ref?access_num=000306390600032&link_type=ISI) 

57. Patenaude B, Smith SM, Kennedy DN, et al. A Bayesian model of shape and appearance for subcortical brain segmentation. Neuroimage 2011;56:907–22.[doi:10.1016/j.neuroimage.2011.02.046](http://dx.doi.org/10.1016/j.neuroimage.2011.02.046)pmid:http://www.ncbi.nlm.nih.gov/pubmed/21352927
    
    [CrossRef](http://bmjopen.bmj.com/lookup/external-ref?access_num=10.1016/j.neuroimage.2011.02.046&link_type=DOI) 
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=21352927&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 
    
    [Web of Science](http://bmjopen.bmj.com/lookup/external-ref?access_num=000290649300007&link_type=ISI) 

58. Behrens TEJ, Johansen-Berg H, Woolrich MW, et al. Non-invasive mapping of connections between human thalamus and cortex using diffusion imaging. Nat Neurosci 2003;6:750–7.[doi:10.1038/nn1075](http://dx.doi.org/10.1038/nn1075)pmid:http://www.ncbi.nlm.nih.gov/pubmed/12808459
    
    [CrossRef](http://bmjopen.bmj.com/lookup/external-ref?access_num=10.1038/nn1075&link_type=DOI) 
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=12808459&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 
    
    [Web of Science](http://bmjopen.bmj.com/lookup/external-ref?access_num=000183780400024&link_type=ISI) 

59. Smith SM, Jenkinson M, Johansen-Berg H, et al. Tract-based spatial statistics: voxelwise analysis of multi-subject diffusion data. Neuroimage 2006;31:1487–505.[doi:10.1016/j.neuroimage.2006.02.024](http://dx.doi.org/10.1016/j.neuroimage.2006.02.024)pmid:http://www.ncbi.nlm.nih.gov/pubmed/16624579
    
    [CrossRef](http://bmjopen.bmj.com/lookup/external-ref?access_num=10.1016/j.neuroimage.2006.02.024&link_type=DOI) 
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=16624579&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 
    
    [Web of Science](http://bmjopen.bmj.com/lookup/external-ref?access_num=000238704700009&link_type=ISI) 

60. Zhang H, Yushkevich P, Alexander D, et al. Deformable registration of diffusion tensor MR images with explicit orientation optimization. Med Image Anal 2006;10:764–85.[doi:10.1016/j.media.2006.06.004](http://dx.doi.org/10.1016/j.media.2006.06.004)
    
    [CrossRef](http://bmjopen.bmj.com/lookup/external-ref?access_num=10.1016/j.media.2006.06.004&link_type=DOI) 
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=16899392&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 
    
    [Web of Science](http://bmjopen.bmj.com/lookup/external-ref?access_num=000240663600008&link_type=ISI) 

61. Behrens TEJ, Berg HJ, Jbabdi S, et al. Probabilistic diffusion tractography with multiple fibre orientations: what can we gain? Neuroimage 2007;34:144-55.[doi:10.1016/j.neuroimage.2006.09.018](http://dx.doi.org/10.1016/j.neuroimage.2006.09.018)pmid:http://www.ncbi.nlm.nih.gov/pubmed/17070705
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 

62. Beckmann CF, Smith SM. Probabilistic independent component analysis for functional magnetic resonance imaging. IEEE Trans Med Imaging 2004;23:137–52.[doi:10.1109/TMI.2003.822821](http://dx.doi.org/10.1109/TMI.2003.822821)pmid:http://www.ncbi.nlm.nih.gov/pubmed/14964560
    
    [CrossRef](http://bmjopen.bmj.com/lookup/external-ref?access_num=10.1109/TMI.2003.822821&link_type=DOI) 
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=14964560&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 
    
    [Web of Science](http://bmjopen.bmj.com/lookup/external-ref?access_num=000188858300001&link_type=ISI) 

63. van den Heuvel MP, Hulshoff Pol HE. Exploring the brain network: a review on resting-state fMRI functional connectivity. Eur Neuropsychopharmacol 2010;20:519–34.[doi:10.1016/j.euroneuro.2010.03.008](http://dx.doi.org/10.1016/j.euroneuro.2010.03.008)pmid:http://www.ncbi.nlm.nih.gov/pubmed/20471808
    
    [CrossRef](http://bmjopen.bmj.com/lookup/external-ref?access_num=10.1016/j.euroneuro.2010.03.008&link_type=DOI) 
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=20471808&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 

64. Provencher SW. Estimation of metabolite concentrations from localized in vivo proton NMR spectra. Magn Reson Med 1993;30:672-9.[doi:10.1002/mrm.1910300604](http://dx.doi.org/10.1002/mrm.1910300604)pmid:http://www.ncbi.nlm.nih.gov/pubmed/8139448
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 

65. Bartnik-Olson B, Alger J, Babikian T. The clinical utility of magnetic resonance spectroscopy in traumatic brain injury: recommendations from the ENIGMA MRS Working group 2019.
    
    

66. Collins GS, Reitsma JB, Altman DG, et al. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. Eur Urol 2015;67:1142–51.[doi:10.1016/j.eururo.2014.11.025](http://dx.doi.org/10.1016/j.eururo.2014.11.025)pmid:http://www.ncbi.nlm.nih.gov/pubmed/25572824
    
    [CrossRef](http://bmjopen.bmj.com/lookup/external-ref?access_num=10.1016/j.eururo.2014.11.025&link_type=DOI) 
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=25572824&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 

67. Steyerberg EW, Vergouwe Y. Towards better clinical prediction models: seven steps for development and an ABCD for validation. Eur Heart J 2014;35:1925–31.[doi:10.1093/eurheartj/ehu207](http://dx.doi.org/10.1093/eurheartj/ehu207)pmid:http://www.ncbi.nlm.nih.gov/pubmed/24898551
    
    [CrossRef](http://bmjopen.bmj.com/lookup/external-ref?access_num=10.1093/eurheartj/ehu207&link_type=DOI) 
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=24898551&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 
    
    [Web of Science](http://bmjopen.bmj.com/lookup/external-ref?access_num=000342232100007&link_type=ISI) 

68. Riley RD, Ensor J, Snell KIE, et al. Calculating the sample size required for developing a clinical prediction model. BMJ 2020;368:1–12.[doi:10.1136/bmj.m441](http://dx.doi.org/10.1136/bmj.m441)pmid:http://www.ncbi.nlm.nih.gov/pubmed/32188600
    
    [CrossRef](http://bmjopen.bmj.com/lookup/external-ref?access_num=10.1136/bmj.m1198&link_type=DOI) 
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=http://www.n&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 

69. Steyerberg EW, Harrell FE, Borsboom GJ, et al. Internal validation of predictive models: efficiency of some procedures for logistic regression analysis. J Clin Epidemiol 2001;54:774–81.[doi:10.1016/s0895-4356(01)00341-9](http://dx.doi.org/10.1016/s0895-4356(01)00341-9)pmid:http://www.ncbi.nlm.nih.gov/pubmed/11470385
    
    [CrossRef](http://bmjopen.bmj.com/lookup/external-ref?access_num=10.1016/S0895-4356(01)00341-9&link_type=DOI) 
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=11470385&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 
    
    [Web of Science](http://bmjopen.bmj.com/lookup/external-ref?access_num=000170019900003&link_type=ISI) 

70. Kuhn M. Building Predictive Models in R Using the **caret** Package. J Stat Softw 2008;28:1–26.[doi:10.18637/jss.v028.i05](http://dx.doi.org/10.18637/jss.v028.i05)
    
    [CrossRef](http://bmjopen.bmj.com/lookup/external-ref?access_num=10.18637/jss.v028.i05&link_type=DOI) 
    
    [PubMed](http://bmjopen.bmj.com/lookup/external-ref?access_num=27774042&link_type=MED&atom=%2Fbmjopen%2F12%2F6%2Fe058975.atom) 

71. Mooney CZ, Duval RD. Bootstrapping : A Nonparametric Approach to Statistical Inference 1993;95.