Assessment of pre-injury health-related quality of life: a systematic review

Insight into the change from pre- to post-injury health-related quality of life (HRQL) of trauma patients is important to derive estimates of the impact of injury on HRQL. Prospectively collected pre-injury HRQL data are, however, often not available due to the difficulty to collect these data before the injury. We performed a systematic review on the current methods used to assess pre-injury health status and to estimate the change from pre- to post-injury HRQL due to an injury. A systematic literature search was conducted in EMBASE, MEDLINE, and other databases. We identified studies that reported on the pre-injury HRQL of trauma patients. Articles were collated by type of injury and HRQL instrument used. Reported pre-injury HRQL scores were compared with general age- and gender-adjusted norms for the EQ-5D, SF-36, and SF-12. We retrieved results from 31 eligible studies, described in 41 publications. All but two studies used retrospective assessment and asked patients to recall their pre-injury HRQL, showing widely varying timings of assessments (soon after injury up to years after injury). These studies commonly applied the SF-36 (n = 13), EQ-5D (n = 9), or SF-12 (n = 3) using questionnaires (n = 14) or face-to-face interviews (n = 11). Two studies reported prospective pre-injury assessment, based on prospective longitudinal cohort studies from a sample of initially non-injured patients, and applied questionnaires using the SF-36 or SF-12. The recalled pre-injury HRQL scores of injury patients consistently exceeded age- and sex-adjusted population norms, except in a limited number of studies on injury types of higher severity (e.g., traumatic brain injury and hip fractures). All studies reported reduced post-injury HRQL compared to pre-injury HRQL. Both prospective studies reported that patients had recovered to their pre-injury levels of physical and mental health, while in all but one retrospective study patients did not regain the reported pre-injury levels of HRQL, even years after injury. So far, primarily retrospective research has been conducted to assess pre-injury HRQL. This research shows consistently higher pre-injury HRQL scores than population norms and a recovery that lags behind that of prospective assessments, implying a systematic overestimation of the change in HRQL from pre- to post-injury due to an injury. More prospective research is necessary to examine the effect of recall bias and response shift. Researchers should be aware of the bias that may arise when pre-injury HRQL is assessed retrospectively or when population norms are applied, and should use prospectively derived HRQL scores wherever possible to estimate the impact of injury on HRQL.


Background
Insight into the change from pre-to post-injury health status of trauma patients is important in order to derive population estimates of the impact of injuries on health-related quality of life (HRQL). The difficulty in measuring the impact of injuries is that the patient's HRQL after sustaining an injury may be influenced by factors other than the injury [1]. For instance, preexisting comorbidity may contaminate our estimates of the injury-related disability, since HRQL scores might incorporate the impact of one or more comorbid diseases instead of solely reflecting the impact of the injury. To overcome attribution bias (i.e., attributing post-injury HRQL scores solely to the injury when it may have been caused by other factors), information on pre-injury HRQL is vital to make valid estimates of the change from pre-to post-injury HRQL due to the injury under study. However, prospectively collected information on the pre-injury HRQL of injury patients is difficult to obtain.
This has led researchers to use alternative methods to assess the contrast between pre-injury and post-injury HRQL, such as use of patient recall or retrospective baseline scores (in other words, pre-injury HRQL that is assessed after sustaining the injury). However, retrospective baseline scores of pre-injury health status are potentially subject to bias [2,3]. Patients may remember their pre-injury HRQL as better or worse than it actually was (recall bias) [2]. Moreover, patients' perception on HRQL may change after the injury, due to a change in internal standards or values (response shift) [4]. This change in perception of HRQL after the injury may also affect the retrospectively assessed pre-injury HRQL.
Other methods are the application of general population norms (i.e., using normative values from the general population as a reference point for the health status before the injury), or the use of a matched non-injured comparison group as a baseline to assess the reduction in health due to the injury. The application of population norms or a matched non-injured comparison group may lead to an inaccurate estimate of the change in health status, as injured people may differ from the general non-injured population [5,6]. Research indicated that injured people have a higher prevalence of comorbidity, hospitalization, and health service utilization prior to their injury in comparison to non-injured people [5]. This suggests that pre-injury health status is worse compared to population norms and conflicts with the reported better pre-injury health status compared to the general population [6][7][8]. On the other hand, the injured population might be healthier and more likely to participate in activities, exposing them to a higher risk of injuries [6].
The current systematic review identifies the methods that are used to assess pre-injury health status of trauma patients and to estimate the change from pre-to postinjury HRQL due to an injury. Moreover, bias that may occur from these methods is examined, by comparing the reported pre-injury HRQL scores with population norms by calculating age-and gender-specific norm scores based on the demographics of the included study samples.
The objectives of this study are: 1) To identify the methods which are used to measure pre-injury HRQL; 2) To compare the reported pre-injury HRQL scores with calculated general age-and gender-adjusted norms; 3) To address the pre-injury HRQL scores per HRQL instrument and injury type; 4) To examine the change between pre-and post-injury HRQL in injury patients; and 5) To formulate recommendations for future studies on (pre-injury) HRQL.

Methods
Relevant studies were identified through systematic literature searches in the databases EMBASE, MEDLINE (via Ovid SP), Cochrane Central, PubMed, Web of Science, SCOPUS, PsycINFO, CINAHL, Lilacs, Scielo, ScienceDirect, and ProQuest. Grey literature was examined via Google Scholar. Search strategies were developed in consultation with a search expert, and included a combination of subheadings and text words (Appendix). Reference lists and citation indices of the included papers and relevant reviews were inspected to identify additional relevant citations.

Study selection
We included studies that assessed the pre-injury HRQL of injury patients, published in English in peer-reviewed journals until July 6, 2015. We included studies on general injury populations, as well as injury-specific studies (e.g., traumatic brain injury or hip fractures). There was no restriction in the methods of patient selection used in the studies (e.g., samples drawn from the Emergency Department (ED), hospital, or outpatient programs). HRQL was conceptualized as an individual's perception of how an illness and its treatment affect physical, mental, and social aspects of his/her life [9]. Studies that assessed only some domains of HRQL (e.g., functional status, activities of daily living, mobility, mental health) were excluded. We included studies that assessed the HRQL of patients before the injury, whether assessed before the injury or retrospectively. Studies that solely used population norms, as a substitute of pre-injury HRQL, were excluded. For studies using data from the same study sample, one study was chosen as the reference study by giving priority to the study that focused on reporting pre-injury HRQL summary scores or utility scores (e.g., instead of percentage of problems per HRQL domain).

Data extraction and methodological quality
The first review author (AS) screened all titles and abstracts and deleted obviously irrelevant papers. Two independent review authors (AS and SP) screened the remaining citations on title and abstract and those obtained in full text. Results from both reviewers were compared by a third review author (JH) and any disagreement was be resolved by discussion between the three authors. We extracted information on the participants (age and gender), injury (type, severity, and mechanism), the assessment of pre-injury HRQL (instrument, procedure, and timing), and recovery of injury patients (change between pre-and post-injury HRQL).
The methodological quality of the studies was evaluated with four elements of the STROBE checklist [10] which were most relevant to the quality of reported pre-injury HRQL by injury type: setting, participants, data sources/measurement, and study size. In addition, risk of bias was assessed using items from the Research Triangle Institute item bank for observational studies on attrition bias ("Impact missing data adequately assessed") and reporting bias ("No important primary outcomes missing") [11].

Statistical analysis
Pre-injury HRQL scores from the study samples were compared with norm scores derived from the general population. To provide population norms for all studies, we used norms by age and sex groups of the EQ-5D (UK population) [12], SF-36 [13], and SF-12 [14] (both US population) to calculate age-and gender-adjusted norms based on the demographics in the study samples.
Heterogeneity between pre-injury HRQL scores was assessed with the Q-statistic and I 2 -statistic, using a random-effects model in a Microsoft Excel spreadsheet [15]. The Q-statistic is a Chi 2 -test for heterogeneity, which assesses whether observed differences in results are compatible with chance alone. A significant Q (low p-value) indicates heterogeneity among the HRQL scores and a variation that is beyond chance [16]. The I 2 -statistic describes the percentage of variation across studies that is due to heterogeneity rather than chance, with an I 2 value of 25% or lower is associated with low heterogeneity, 50% indicating substantial heterogeneity, and 75% or higher indicating high heterogeneity [17]. In case of substantial or high heterogeneity, pooled results should not be calculated, or at the very least, should be interpreted with caution.

Literature search
The extensive search strategy identified 2,286 unique titles of potentially relevant articles (Fig. 1). Screening of the titles and abstracts resulted in a selection of 383 articles that appeared to meet all selection criteria. After screening and selection of the full text papers, we retrieved 31 studies described in 41 publications. The main reasons for exclusion were not measuring preinjury health status, not reporting on injuries or only reporting part of the outcomes on HRQL.

Methodological quality
Over half (n = 19) of the 31 articles included in our review reported on attrition. Most studies faced several problems in the participation of eligible patients, as patients refused to participate (n = 15), could not be contacted (n = 6), did not complete the HRQL assessment (n = 6), had died (n = 5), or were not able to respond to the questionnaires (e.g., due to the consequences of the trauma, n = 3). Overall, response rates ranged from 60 to 98% in 17 of the 22 studies that reported on response rates.
Limited variation existed in the selection of samples between the studies. Most patients were recruited during or after a treatment in a (pediatric) hospital (n = 21), while others were selected from a specialized burn center (n = 2) [20,28], sports center (n = 1) [23], or nursing home facility (n = 1) [30].
In four out of the 31 studies, the measurement of preinjury HRQL was one of the primary aims [8,18,46], while in all other studies pre-injury HRQL scores were used to assess the change in HRQL after the injury or to validate HRQL instruments.
All but two studies in this review retrospectively assessed the pre-injury HRQL of patients, by asking them to recall their HRQL before the injury occurred. Only two studies provided prospectively collected preinjury health status of participants [18,46] (articles in bold and italics in Table 1): the Medical Expenditure Panel Survey (MEPS) [18] and the Seguimiento Universidad de Navarra (SUN) [46] cohort. These studies used data from longitudinal cohort studies in which participants who were initially non-injured were followed for several years, by means of questionnaires comprising the SF-36 [46] or SF-12 [18]. In addition, only one of the included studies measured the recalled pre-injury health status of trauma patients and not their post-injury HRQL [6], while all other studies measured both pre-and post-injury HRQL.
Within-study comparisons of pre-injury HRQL between injury patients or with controls showed that patients who were injured due to a motor vehicle injury or who sustained a TBI had significantly lower mental health at baseline [18,27,44,46] and lower scores across all HRQL domains [46] compared to those without a motor-vehicle injury or TBI (Table 2). Higher pre-injury HRQL was found in those who survived than those who eventually died during follow-up (significant differences found on the SF-36 PF, RP and GH [24], no significant differences found between EQ-5D scores [30]) and in those recovered than those not recovered at follow-up (not significant) [8].

Change between pre-and post-injury HRQL
Most studies used a longitudinal design (n = 23) with multiple follow-up measurements over time (n = 18), often measuring post-injury HRQL at three months, six months and/or 12 months. All studies showed a decrease in post-injury HRQL compared to their pre-injury levels of HRQL (Table 2). Looking at the EQ-5D, only one out of the 12 studies showed full recovery to preinjury HRQL at one year after the injury [38], while the other studies still reported reduced levels of HRQL postinjury. Looking at the SF-36 and SF-12, injuries showed to have the highest impact on the physical component of HRQL (reduction in PCS with 15 to 30 points from preinjury to first post-injury assessment) compared to the mental component of HRQL (reduction in MCS with 5 to 9 points) [23,27,29,32]. At the final follow-up measurement, both prospective studies showed almost full recovery to pre-injury HRQL levels on the PCS and full recovery on the MCS [18,46], while only one retrospective study showed such recovery on the PCS [48] or MCS [28].

Discussion
This systematic review summarized the methods that were used to assess pre-injury health status and to estimate the change from pre-to post-injury HRQL. All but two of the 31 studies in our review used retrospective assessment (recall) to assess pre-injury HRQL. The studies most often applied the SF-36, followed by the EQ-5D or SF-12, by means of questionnaires or face-to-face interviews. Recalled pre-injury HRQL scores consistently exceeded general population norms, except in a limited number of studies on injury types of higher severity (e.g., traumatic brain injury and hip fractures). All    (7) Diagnosis: PCS 41 (11) Hip fracture had lower preinjury HRQL than wrist facture (significant) or vertebral fracture. Scores showed recovery after 6 m. After 1y, scores were not significantly different from pre-fracture.    Hip fracture had lower pre-injury HRQL than wrist (significant) or vertebral fracture. Scores at 6 m were significantly lower than preinjury. After 1y, scores were not significantly different from pre-fracture values.

Pons
studies reported reduced post-injury HRQL compared to pre-injury HRQL. Both prospective studies reported that patients had recovered to their pre-injury levels of physical and mental health, while in all but one retrospective study patients had not returned to their reported preinjury levels of HRQL, even years after the injury. Prospective assessment is the preferred method to determine pre-injury HRQL as it is not subject to bias that may occur due to experiencing an injury. In our review, only two out of the 31 studies used prospective assessment of pre-injury HRQL. These studies used longitudinal data from the Medical Expenditure Panel Survey (MEPS) among the US general population [18] and the Seguimiento Universidad de Navarra (SUN) cohort comprising university graduates in Navarra, Spain [46]. Both prospective studies reported lowest pre-injury mental health on the SF-36 (MCS 47) [46] as well as SF-12 (MCS 49) [18] of all studies in our review, which otherwise all used retrospective assessment. These prospective studies indicate that the retrospective assessment and population norm approach are highly likely to be biased.
Our review shows that the retrospectively assessed pre-injury HRQL systematically differed from the ageand gender-adjusted norms we calculated based on population data on the EQ-5D, SF-36, and SF-12. Despite the use of different HRQL instruments, recalled pre-injury HRQL scores in our review consistently exceeded these adjusted population norms. An exception to this were samples including patients with a hip fracture [30,35,36,39], motor vehicle injury [18,46], vertebral fracture [33] or TBI [27], that reported poorer preinjury HRQL than our calculated adjusted norms. These injury patients are likely to be less healthy than their counterparts [18,27,44,46], in terms of socioeconomic status [18], comorbidity [18,49], or frailty and older age [12,49,50].
The difference between retrospectively assessed preinjury HRQL and population norm scores might be caused by several reasons.
Recall bias may have influenced the outcomes of the retrospective assessment, as patients may have remembered their pre-injury HRQL differently than it actually was [2,51,52]. Patients may, for example, have overestimated their health status before the injury, resulting in higher recalled pre-injury HRQL than seen in the general population.
Response shift might have occurred, as patients' perception of HRQL may have changed due to the injury and a change in health [4]. After having had experience with poor HRQL, patients may have inflated the rating of their health status before the injury [53].
Nevertheless, some researchers argue for the use of retrospective assessment of pre-injury HRQL, as this method applies one internal standard of HRQL values (reference point) in the assessment of both pre-injury HRQL and post-injury HRQL [4,53]. According to  them, such a reference point is essential for the interpretation of the change from pre-to post-injury HRQL, since patients may have changed their judgement of HRQL due to new insights since the injury (e.g., although a patient has a serious injury, he/she has seen others who are far worse off ), or patients have become used to their new health state. However, both recall bias and response shift might result in an overestimation of the pre-injury HRQL by patients. This is underpinned by our finding that, even years after the injury, in all but one retrospective study patients had not returned to their reported levels of pre-injury PCS and MCS, while recovery to pre-injury HRQL levels was seen in both prospective studies.
Moreover, selection bias may have threatened the validity of the findings from the studies included in our review, as the study populations were often not randomly selected from the injury population for which the findings are reported [54]. For example, studies had excluded patients with pre-existing morbidities (e.g., physical illness, cognitive impairment), as it was anticipated that these patients would be difficult to follow up. Exclusion of patients with impairments before the injury may have increased the overall pre-injury HRQL scores of these study samples, as healthier participants were recruited.
In contrast, attrition bias may have decreased the overall pre-injury HRQL scores measured in the studies, as a Fig. 3 Pre-injury SF-36 and SF-12 scores by injury type and in comparison to population norm scores. Studies in bold and italics prospectively measured pre-injury HRQL. 1 Adjusted by the age and sex distribution in the study population, based on the weighted health state index by age and sex [13,14]. 2 Final post-injury measurement: at 3 [27], 6 [43] or maximal 9 months post-injury [18], 1 year post-injury [26,28,29,32,33,48], 2 years post-injury [44], or 4-8 years post-injury [46]. Heterogeneity: PCS Chi 2 = 12.48, df = 11, (p = 0.33), I 2 = 12%; MCS Chi 2 = 11.88, df = 11, (p = 0.37), I 2 = 7%. MVC: injury due to motor vehicle crash; Ortho: orthopedic injury; TBI: traumatic brain injury Fig. 2 Pre-injury EQ-5D scores by injury type and in comparison to population norm scores. 1 Adjusted by the age and sex distribution in the study population, based on the weighted health state index by age and sex [12]. 2 Final post-injury measurement: at discharge [35], 1 year post-injury [8,30,[36][37][38][39], 2 years post-injury, or 2-7 years post-injury [42] higher proportion of the non-participants were less educated [26], cognitively impaired [38], victim of intentional injury [6], shorter hospitalized [21] and had lower injury severity [28,29,44], less pain [34], better mental health [34]. These factors are all expected to be associated with better HRQL and incorporation of these patients may have resulted in higher pre-injury HRQL scores. Additionally, pre-injury HRQL levels may have increased after loss of follow up, resulting in higher pre-injury HRQL in the final study sample with complete response compared to the eligible study sample [32].
Finally, retrospectively assessed pre-injury HRQL scores may differ from the population norms as injury populations may differ from the general population. The findings of the retrospective assessments (recall) in our review suggest that injured populations are generally healthier than the general population. Previous studies reported that, as injured populations might be healthier, they are more likely to participate in activities, exposing them to a higher risk of injuries [6]. However, the comparisons of injury patients with matched controls in our review showed injury patients to be less healthy than their counterparts, as they reported significantly lower pre-injury mental health than controls [18,27,44,46] and lower scores across all HRQL domains [46]. Previous research showed that injury patients had a higher occurrence of comorbidity, higher admission rates to the hospital, higher health service utilization, and a lower socioeconomic status prior to their injury in comparison to uninjured people [5,18]. It is argued that the general population has not been exposed to a similar injury experience as the injury population, which emphasizes the use of retrospective assessments over the application of general population norms to estimate the impact of injury on HRQL [7].

Strengths and limitations
Our review included studies on the pre-injury HRQL from children, adolescents, and adult patients, with various injury types, using a range of HRQL instruments. Moreover, this review compared the reported pre-injury HRQL scores with general population norms, calculated for each study based on the reported mean age and gender distribution of the study sample, to identify bias that may occur from the different methods to assess preinjury HRQL.
There are limitations to this review that need to be addressed. First, there was no restriction in the methods of patient selection used in the studies. Therefore, the studies in this review included samples retrieved from a variety of injury settings (e.g., hospital or outpatient programs). Their conclusion may not be applicable to injury patients from other injury settings. However, most studies selected their patients during or after treatment in a (pediatric) hospital or specialized treatment center, which may enhance the generalizability of their results to patient populations with similar case mix.
Second, the review included studies with patient samples from a broad range of injury types and injury severity levels, which may have complicated the comparability of the results between studies. Nonetheless, this way we were able to provide a full oversight of the pre-injury health status of injury patients and the differences in pre-injury HRQL between injury types.
In addition, there are limitations to the studies included in our review. First, more than half of the included studies had difficulties in recruiting research participants, as patients often could not be contacted, had died, refused to participate, or did/could not complete questionnaires. The studies often reported limited generalizability of their results due to differences between the eligible patients and study participants, loss to follow-up, their limited number of subjects, and recruitment of participants from a single center.
In some studies pre-injury HRQL was assessed after a long period of time since the injury, for example several months up to years after the injury [8,42,44]. This longer time frame may have increased the recalled preinjury HRQL scores [31], as these studies also reported the highest pre-injury HRQL scores on the EQ-5D (0.94-0.99) compared to the studies that used shorter time frames. However, these three studies assessed the HRQL of a relatively young injury population. Moreover, no differences were found between the time frame and pre-injury HRQL in studies that used the SF-36 or SF-12.
Finally, unfortunately not all studies reported the HRQL scores in the text or tables (e.g., only in graphs). After contacting the authors, in three publications HRQL scores had to be manually obtained from the graphs presented in the article [36,40,45]. This may have resulted in some small differences in the levels of pre-and/or post-injury HRQL.

Recommendations for future research
Our review clearly showed that recalled pre-injury HRQL systematically exceeded population norms. These differences in pre-injury HRQL may generate different estimates of the change in HRQL from pre-to postinjury due to an injury.
Researchers should use prospectively derived preinjury HRQL scores wherever possible to estimate the impact of injury on HRQL. If it is not feasible to prospectively assess the pre-injury health status of trauma patients, researchers should be aware of the bias that may arise when pre-injury HRQL is assessed retrospectively or when population norms are applied. Overall, more research is needed to examine the effect of recall bias and response shift on the reported levels of pre-injury HRQL among trauma patients, in which different methods to assess pre-injury HRQL are compared and within-study comparisons between reported pre-injury HRQL and population norms are made.
The results of our review imply that there are a number of methodological advances regarding preinjury HRQL interpretation left. Researchers should be aware of the different purposes the information on pre-injury HRQL of patients may have. For instance, pre-injury HRQL may be seen as a baseline health status to which patients are expected to return after the injury. On the other hand, pre-injury HRQL may be used to measure total loss in health, or may be used to offer insight into inter-patient differences in recovery after an injury.
In general, when assessing pre-injury HRQL, researchers should carefully consider and specify the timing of the assessment of pre-injury HRQL and the period of the pre-injury assessment. The time period shows to be one of the essential factors influencing patient recall, as recall bias is generally worse when asking for a recall over longer periods [55]. A short time frame within the injury and retrospective assessment of pre-injury HRQL may increase recall and may increase the correlation between pre-and post-injury measures [31]. This implies that pre-injury HRQL should be assessed as soon as possible after the injury, preferably within the first week after the injury [56]. Whether or not the measurement of pre-injury HRQL is the primary purpose of studies, publications on the measurement of HRQL should include information on the applied methods to measure HRQL.
Levels of pre-injury HRQL also may have been influenced by the use of telephone interviews. In our review, the highest or one of the highest pre-injury HRQL on the EQ-5D [42], SF-36 (PCS and MCS) [26], or SF-12 [43] were reported by studies that had conducted telephone interviews to assess the pre-injury levels of HRQL. Previous research indicated that telephone-administered questionnaires provide higher HRQL scores than self-administered questionnaires [57][58][59]. Preferably, the same method should be used for the assessment of both pre-injury and post-injury HRQL throughout the study, at all post-injury HRQL measurements and among all individuals.
Researchers should choose a validated HRQL instrument that has shown good performance in the type of injury under study, and that is sensitive to changes in HRQL and differentiate well between health states. In order to assess the change from pre-to post-injury HRQL, the same HRQL instrument should be applied throughout the study. Preferably, a HRQL instrument should be chosen for which national age-and genderadjusted population norms are available. In order to enable comparison of the impact of injuries on HRQL between studies, injury types and other diseases, it is recommended to report the pre-and post-injury HRQL scores for specific age and sex groups, which correspond to the age and sex distribution of the norm groups for the applied instrument.
Finally, to examine the change in HRQL due to the injury, a longitudinal design is recommended with multiple follow-up measurements over time (e.g., at 1-3 months, 3-6 months, and 6-24 months post-injury) [56].

Conclusions
So far, primarily retrospective research has been conducted to assess pre-injury HRQL. This research shows consistently higher pre-injury HRQL scores than population norms and a recovery that lags behind that of prospective assessments, implying a systematic overestimation of the change in HRQL from pre-to post-injury due to an injury. More prospective research is necessary to examine the effect of recall bias and response shift. Researchers should be aware of the bias that may arise when pre-injury HRQL is assessed retrospectively or when population norms are applied, and should use prospectively derived HRQL scores wherever possible to estimate the impact of injury on HRQL.