Research population
This study was conducted with data from the ‘Letsel informatie systeem’ (LIS) (Dutch injury patient surveillance system). LIS is an ongoing data collection in fourteen (out of approximately 90) hospitals in the Netherlands [
19]. The fourteen hospitals are a representative sample of hospitals in the Netherlands and consist of academic and non-academic and rural and urban hospitals. The LIS hospitals register information on age, sex, circumstances of trauma, cause of trauma, in-hospital health care consumption, such as length of hospital stay and nature of trauma of each patient that visits the emergency department due to a trauma or poisoning [
20]. After treatment of trauma, the patient was admitted to the hospital or discharged to the home environment or institution. The data used in this study were limited to the data collected in 2017. A stratified sample of patients received questionnaires 6 months and 12 months after trauma, with exclusion criterion for participation < 15 years old. The first questionnaire contained an informed consent form. Approval by the Medical Ethics Review Committee (METC) was not required according to the METC of the Academic Medical Center of the University of Amsterdam. The questionnaires contained questions on education level, comorbidity and HRQL. HRQL was measured with the EQ-5D-5L with cognitive domain and with the EQ-VAS.
HRQL data
The EQ-5D-5L is a measurement instrument that consists of five domains with each five response levels. The five domains are mobility, self-care, usual activities, pain/discomfort and anxiety/depression. Cognition was added as a sixth domain to the five existing domains. The response options of the 5-level version are no problems, slight problems, moderate problems, severe problems, and extreme problems/unable to. Responses to the five EQ-5D and six EQ-5D+C domains can be combined in so-called health profiles, which define the severity level, where 1 means no problems and 5 means extreme problems or unable to perform, e.g. ‘21421′ for EQ-5D and ‘132512′ for EQ-5D+C. An unweighted summary score (level sum score) can be calculated from the responses to the EQ-5D, ranging from 5 to 25 for the EQ-5D and 6 to 30 for the EQ-5D with cognition. To enable comparison between the two level sum scores, scores on the EQ-5D+C were recoded to the same scale as scores on the EQ-5D by multiplying the level sum score with 5/6. In addition to the six domains of the EQ-5D, the EQ-VAS is also part of the EQ-5D. The EQ-VAS requires respondents to rate their health on a scale from 0 to 100, where 0 represents the worst imaginable health state, and 100 represents best imaginable health. Furthermore, the impact of events scale-revised (IES-R), which informs on PTSD-related complaints, was included in the questionnaire. The IES-R is a self-report questionnaire that consists of 22 items, which measure intrusive re-experiences of the trauma, avoidance of trauma-related stimuli and hyper arousal symptoms [
21]. By combining the 22 items the total IES-score, ranging from 0 through 88, can be calculated. The cut-off score was set at 33, as advised by Creamer, Bell and Failla [
22], with scores below 33 representing no PTSD, and scores equal to or higher than 33 representing PTSD. Data were imputed for respondents with one or two missing items on the IES-R using simple imputation, based on responses to other items of the IES-R. In addition, the first questionnaire included 19 items regarding the presence of one or more chronic diseases prior to trauma to assess comorbidity. Comorbidity is defined as the self-reported presence of any co-existing medical diseases or disease processes additional to the trauma that the trauma patients sustained.
Data analysis
Data analyses were performed in SPSS version 25. Respondents were included in the analyses if an answer was provided for the EQ-5D+C and the EQ-VAS both 6 months and 12 months after trauma. The presence of traumatic brain injury (TBI) was determined based on the trauma registration in the LIS, and comorbidity was determined based on information from the follow-up questionnaire at 6 months. Comorbidity was divided in categories: no comorbidity; one comorbidity; two or more comorbidities. Furthermore, presence of PTSD was determined based on responses to the IES-R, which was included in the questionnaire at 6 months. Frequencies of socio-demographic characteristics were compared between responders and non-responders using Chi-square tests and Mann–Whitney U tests. A distributional effect in terms of ceiling was determined by defining the proportion of perfect health profiles (11111 for EQ-5D and 111111 for EQ-5D+C) among all observed profiles. A higher proportion of perfect health profiles indicates more ceiling.
Informativity, expressed as classification power of EQ-5D and EQ-5D+C was determined with the Shannon Index (
H′) and Shannon Evenness Index (
J′) [
23]. Information on the ability to measure diversity in a population can be derived from these two indices [
24]. To calculate the Shannon Index, the formula:
H′ = − ∑
ci=1 pi 2log
pi was used, where p
i represents the proportion of people with one health profile, and C represents the total number of possible health profiles. A higher value of
H′ indicates that more information is captured by the measurement instrument. For EQ-5D, the total number of possible health profiles was 3125 (5 × 5 × 5 × 5 × 5), whereas for EQ-5D+C there were 15,625 possible health profiles. Next, the Shannon Evenness Index was calculated, based on the Shannon Index:
J′ =
H′/
H′
max, with
H′
max representing
2log
C (total number of possible health profiles). A higher value on the Shannon Evenness Index represents the capture of more information by the extra domain, and therefore increases the distinction between patients [
8]. As the assessment of
H′ using a sample of the total population will lead to an underestimation according to Pielou [
25], adjusted values were calculated for
H′ and
J′ to control for this. Adjustment magnitude was set at (
C − 1)/2
N. Classification efficiency was determined for subgroups of respondents with TBI, PTSD and neither TBI nor PTSD, both for 6-month and 12-month measurement.
Furthermore, convergent validity of the EQ-5D-5L and the EQ-5D-5L+C was determined by the strength of association between the EQ-5D-5L with and without cognition with the EQ-VAS. First, the level sum score was calculated for both EQ-5D and EQ-5D+C as the sum of all domains (e.g. health profile 33333 had a level sum score of 15). The level sum score ranges from 5 to 25 for EQ-5D and 6 to 30 for EQ-5D+C, with a higher score representing poorer health. Subsequently, after confirmation that the assumptions were met, Spearman’s rank correlation coefficients between EQ-5D (all versions) and EQ-VAS were determined for the 6-month and 12-month assessments.
Additionally, explanatory power of EQ-5D and EQ-5D+C were determined using multivariable linear regression analyses, as the assumptions of linear regression were met, with EQ-VAS as dependent variable. Independent variables consisted of dummy variables for the levels ‘slight problems, moderate problems, severe problems, and extreme problems/unable to’ for all EQ-5D domains including cognition for the EQ-5D+C, with ‘no problems’ as the reference category. All combinations of 5 out of the 6 EQ-5D+C domains were analysed separately for TBI/PTSD respondents and for respondents with neither TBI nor PTSD.
Discriminatory power
Mean level sum score was determined for both EQ-5D and EQ-5D+C at 6-month measurement. Scores were compared between subgroups of TBI/PTSD versus no TBI/PTSD, and no comorbidity versus one comorbidity versus two or more comorbidities, respectively. The level sum scores of EQ-5D and EQ-5D+C were compared within each subgroup using paired t-tests. Furthermore, level sum scores were compared between subgroups using one-way ANOVA.
Responsiveness
Mean difference between reported EQ-5D and EQ-5D+C responses collected at 6 months and 12 months post-trauma were analysed per domain to gain insight into the average change per domain over time. The mean difference was calculated per domain by subtracting the 6-month score from the 12-month score: \(\frac{D_{12}-D_{6}}{N},\) where D12 represents the score on the domain at 12 months, D6 the score on the same domain at the 6-month assessment, and N the total number of respondents. This resulted in a score between − 4 and 4.
Furthermore, EQ-5D and EQ-5D+C health profiles were compared at 6 months and 12 months using the Paretian classification of Health Change [
26]. Based on the difference between the two measurement moments, respondents were classified as follows: no change; no problems; improved health; worsened health; non-categorisable (mixed change). Difference in classification between EQ-5D and EQ-5D+C was compared for subgroups of the population, based on the presence of TBI and PTSD.
To quantify change over time per domain, the probability of superiority was calculated by dividing the number of respondents with positive change (in terms of health improvement over time) by the total sample size. Half of the respondents that were categorised as ‘no change’ were added to the number of respondents with positive change to account for ties. Both the Paretian classification of Health Change and the probability of superiority can be interpreted as measures of responsiveness to change.