Source of data

The study will be a prospective, observational study of two temporally staggered cohorts of patients with trauma, who are emergency department admissions into a Major Trauma Centre in the United Kingdom, with an episode inception defined as the traumatic event (figure 1). The first cohort will facilitate development of the prediction model to inform the screening tool, and the second will enable validation of the prediction model through evaluation of the predictive performance of the model and tool.13 14 There will be an interval of at least 6 months between recruitment into the two respective cohorts. The prospective design enables control of unwarranted influences and enables a stronger case for inferring causal relationships. The nature of the study necessitates predictive statistical modelling.15 This protocol is written in line with the TRIPOD (transparent reporting of a multivariable prediction model for individual prognosis or diagnosis) statement,16 in which recommendations are given for the reporting of prediction model development and validation.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Study design. CNS, central nervous system.

Self-reported and physical assessment predictive data will be collected at baseline over a period of up to 14 days (or duration of inpatient stay if shorter), which will commence immediately following recruitment. Biomarker data collection will occur throughout the same baseline period, but can commence prior to recruitment providing assent is gained from a legal consultee. The outcome data will be collected at 6 months postinjury (working definition of chronic pain)17, the point of evaluation of an individual’s absolute risk of poor outcome (objectives 1, 2 and 3). In addition, selected data will be measured at 3 and 12 months postinjury to explore the clinical course of recovery following injury in the shorter and longer terms (objective 4).

Participants

Participants will be recruited from the register of a Major Trauma Centre in the United Kingdom for a period of up to 24 months (planned start date January 2018). All consecutive eligible patients will be approached for recruitment until the sample size is achieved.

Eligibility criteria

Inclusion criteria: adult patients aged ≥16 years who are admitted to emergency department of the Major Trauma Centre, with their main criteria for admission being acute musculoskeletal trauma within the preceding 14 days, and a capacity to use and understand written and verbal English language and a mental capacity to provide informed consent (eg, no confusion, delirium, severe cognitive impairment or severe mental illness, defined by a score of ≤6 on the Abbreviated Mental Test).18 The primary reason for including patients with trauma occurring up to 14 days, is to be inclusive of patients who are critically ill and/or with diminished mental capacity initially following their trauma, patients requiring surgery as a result of the trauma and representative of the broad trauma population.

Exclusion criteria: exclusions will be made where the patient has an acute intracranial lesion (eg, bleed) combined with a score of ≤14 on the Glasgow Coma Scale19 (a 15-item measure of consciousness impairment with adequate reliability20 that is routinely taken in patients with trauma at hospital admission), where there is evident brain or central nervous system injury or impairment, long-term neurocognitive disorders (such as brain tumour, multiple sclerosis, Alzheimer’s and Parkinson’s diseases and so on), comorbid cancer, the presence of an ongoing rheumatological condition, prolonged use of corticosteroids or terminal illness with short life expectancy.

Withdrawals: participants will be informed that they are free to withdraw from the study at any time, without needing to provide reason. In the event of death within 3 months of being recruited, participants will be automatically withdrawn from the study and the primary predictive analysis. Baseline data of all withdrawn participants will be kept and compared with those of retained participants to assess for any differences.

Recruitment

Based on feasibility data (site data from the Trauma Audit and Research Network), it is estimated that at least 1000 eligible patients with trauma will be approachable for recruitment over a 24-month period, and that 50% would be expected to consent to participation. A dedicated team of research nurses will be available to recruit patients 7 days per week (from 0700 to 1930).

Because of impairments resulting from their injuries, some otherwise eligible patients will lack the mental capacity to provide informed consent when first approached to enrol in the study. Recruitment into the study will therefore be undertaken under the guidance and provision of the (UK) Mental Capacity Act 2005 for research in emergency situations. If the patient lacks sufficient capacity to consent, written assent for study participation will be sought from a legal consultee to begin biomarker data collection (blood samples), with informed consent for full recruitment (and subsequent data collection) being sought from the patient only if, and when, they regain sufficient capacity to provide this. If the patient does not regain capacity to provide consent within 14 days of their trauma, they will not be recruited into the study, biomarker data collection will cease and any blood samples already collected will be destroyed before analysis.

Once informed consent is gained and the participant recruited, following a minimum 1 hour lead time after the informed consent process (to reduce patient burden), members of the research team will visit the patient at their bedside to collect baseline self-reported data via questionnaires (table 1). On the next available working day following completion of the questionnaires (again, to reduce patient burden), members of the study team will return to the patient to conduct the first physical (quantitative sensory testing) assessment. At each visit, if deemed necessary, the capacity of the participant will be checked using an Abbreviated Mental Test18 (a score of ≤6 is indicative of reduced capacity), and asked if they are happy to proceed with data collection.

Outcome

The outcome for the prediction model is an individual’s absolute risk of poor outcome (chronic pain and disability) at 6 months postinjury. Outcome will be measured using the Chronic Pain Grade Scale (CPGS),21 which combines pain intensity and pain-related disability over the preceding 6 months into a single measure of pain severity. The CPGS has previously been used to assess the severity of body-wide chronic pain in general population,22 primary care23 and post-trauma24 populations. Each item of the CPGS relates to at least one of the three categories of the International Classification of Functioning, Disability and Health (ICF)25: impairment, activity limitations and restricted participation. Furthermore, all ICF categories are encompassed by the CPGS.26 The CPGS is a unidimensional scale, with good internal consistency across different pain populations; Cronbach’s alpha of 0.84 to 0.91 in back pain, 0.79 for headache and 0.84 for temporomandibular pain.21 27 With regards to construct validity, cross-sectional and longitudinal studies of general practice patients have shown high scores on the CPGS, indicating greater chronic pain, to be associated with higher rates of unemployment, greater pain impact scale scores, greater use of opioid analgesics and physician visits, depressed mood and lower self-rated health status.21 27 28 For convergent validity, the CPGS has been found to have good correlation with equivalent dimensions of the SF-36.27 28 In terms of responsiveness, changes in score over time in patients with chronic musculoskeletal pain correlated significantly with changes in SF-36 scores.29 The CPGS has also been shown to have good test–retest reliability in primary care patients with back pain (weighted kappa 0.81, 95% CI 0.65 to 0.98).27

Although pain persistence is not used in assigning pain grade, a measure of pain days in the prior 6 months is included in the CPGS.30 The responses on the remaining seven items are used for computing scores for the three subscales of the CPGS21: characteristic pain intensity, disability score and disability points. The characteristic pain intensity score (range: 0–100) is obtained by calculating the mean of three pain intensity measurements: ‘at the present time’, the ‘worst pain’ in the preceding 6 months and the ‘average’ pain over the preceding 6 months. The disability score (range: 0–100) is obtained through the mean ratings of how much the pain has interfered in performing activities of daily living, recreational, social and family activities, and work (including housework) activities in the last 6 months. The disability points are scored 0–3 and are derived from a combination of ranked categories of the number of disability days (the number of days that the respondent was away from usual activities in the preceding 6 months due to pain) and disability score. Based on these scores, the participant’s chronic pain and disability status can then be classified into one of the five ordinal categories of chronic pain severity21: no pain (Grade 0), low disability and low intensity pain (Grade I), low disability and high intensity pain (Grade II), high disability and moderately limiting intensity pain (Grade III), and high disability and severely limiting intensity pain (Grade IV). As in previous studies, poor outcome will be defined as grade ≥II.23 31–34

Candidate predictors

Candidate predictors have been selected that are: (1) reliable and valid measures of their domain, and (2) have a theoretical association with the development of chronic pain. Both modifiable and non-modifiable candidate predictors are included. Candidate predictors are summarised in table 1, with further detail in the online supplementary file S1. Table 1 details important data that will be measured at 3, 6 and 12 months postinjury to explore the clinical course of recovery following injury in the shorter and longer terms. All data collection will be standardised through protocols and clinical report forms.

Data handling

Blood samples will be collected through the clinical and research nurse teams, while the participant is in the hospital, and either analysed immediately (C-reactive protein) or securely stored for subsequent analysis (cell-free DNA). Baseline self-reported questionnaires, pain and injury drawings, and physical assessments will be collected by one of three trained assessors from the study team. Inter-rater reliability studies (across two assessors) will first be conducted in both healthy and trauma populations to inform definitive testing protocols. The order of physical assessment data collection will be randomly assigned (using computerised randomisation software) according to the modality of testing and by site, to prevent order effects. Follow-up self-reported questionnaires will be posted to participants at their home addresses; with up to two postal reminders and a telephone call for non-response. All questionnaires will be formatted so that data can be scanned or entered directly into an electronic database. Following data entry, data will be checked by a second researcher for completeness and accuracy. In addition, regular audits of data collection and storage will be performed by an independent study management committee. Participant identifiable information will be securely stored within the hospital, in line with current United Kingdom data protection legislation, and only accessible by the site Principal Investigator and Research Nurse team who will not be involved in data analysis. All outcome measure data will be securely transferred within an anonymised database file to physically secure servers at the University of Birmingham, and stored for a period of 10 years in line with Research Governance procedures. Participants will receive usual care and interventions received will be recorded for descriptive analysis. Anonymised data will be analysed using IBM SPSS Statistics.

Sample size

In predictive modelling, a larger sample size enables lower bias and variance, and permits the prospective prediction of new observations.15 The number of predictors will be reduced using exploratory factor analysis. This process will ensure that the sample size provides at least 10 cases per candidate predictor, to adequately power the final regression analysis.35 36 Data will be collected for an estimated 300 participants per cohort (n=600 total) to allow for withdrawals (primarily expected deaths within the first 3 months) and losses to follow-up, so that final data are available for 250 participants per cohort (n=500 total).

Statistical analysis methods and management of missing data

For each cohort, potentially eligible patients, numbers examined for eligibility, confirmed eligible, recruited into the study, completing follow-up and analysed will be reported in a flow diagram. Reasons for non-participation, exclusion, drop-outs and withdrawal (eg, death) will be reported at each stage. Participant characteristics will be descriptively presented. For each variable of interest, the number of participants with missing data will be reported.

For the first cohort to develop the predictive model, an initial exploratory data analysis stage will summarise the data.15 Correlations between candidate predictive factors will be calculated at baseline. Outcome (CPGS) scores will be dichotomised into good and poor categories as described previously. Data reduction will use exploratory factor analysis to assess factor loading of candidate predictors (summary scores) on poor outcome at 6 months. This will enable the number of candidate predictors entered into the final model to be reduced to 25, which can be supported by the cohort sample of 250. This process reduces the risk of over-fitting the model and the risk of selecting the wrong variables due to correlation between predictor variables (multicollinearity).37

Statistical modelling for prediction has been planned a priori. To explore the influence of each predictive factor on poor outcome at 6 months, a logistic multivariable regression model will be fitted to the dichotomised outcome scores to calculate low and high risk of poor outcome. Odds ratios for each candidate predictive factor will be reported, adjusted for other factors and account for clustering (eg, level of injury severity). If necessary, multiple imputation38 will be used to deal with missing outcome data. The characteristics of those patients with and without 6-month data will also be compared, to inform whether patients with no 6-month data were missing at random. Reduced multivariable analyses will be considered if necessary (eg, removing one of two candidate predictive factors that are highly correlated at baseline), to examine the robustness of the conclusions.

Risk groups and development of the predictive screening tool

The predictive model will be used to develop a risk stratification tool to inform an individual’s absolute risk of poor outcome. The stratification tool will inform clinical decision-making for precision rehabilitation. Items will be selected for the model if they are statistically significantly (P<0.05) associated with poor outcome in the logistic regression analysis, and those deemed clinically important to retain using expert opinion (regardless of statistical significance, study steering group) to improve face validity for clinicians and avoid overfitting of the model.37 The regression model with included predictive factors will be fitted to the data from the first of the two cohorts to obtain a final set of parameter estimates (i.e. alpha and beta terms), which will be used to form the model. An important requirement of the stratification tool is that it should be sufficiently brief to facilitate use in clinical practice. Thus, we will look to simplify the model where possible to facilitate its use, but without important reduction in its predictive ability in terms of calibration and discrimination. For example, if multi-item questionnaire scores are included in the model, then we will evaluate whether just one of the questionnaire items is sufficient. Ideally, this process will result in a full and simplified model.

Development versus validation

For validation of the model, data from the second of the two cohorts will be compared with that of the first to enable analysis of the distribution of important variables, inclusive of demographic, predictor and outcome variables. The predictive performance of the screening tool (discrimination, calibration and goodness of fit) will be assessed using data from the second cohort. Data in both cohorts will be consistent in terms of setting, eligibility criteria, outcome and predictors.