Disability weights based on patient-reported data from a multinational injury cohort

Abstract Objective To create patient-based disability weights for individual injury diagnosis codes and nature-of-injury classifications, for use, as an alternative to panel-based weights, in studies on the burden of disease. Methods Self-reported data based on the EQ-5D standardized measure of health status were collected from 29 770 participants in the Injury-VIBES injury cohort study, which covered Australia, the Netherlands, New Zealand, the United Kingdom of Great Britain and Northern Ireland and the United States of America. The data were combined to calculate new disability weights for each common injury classification and for each type of diagnosis covered by the 10th revision of the International statistical classification of diseases and related health problems. Weights were calculated separately for hospital admissions and presentations confined to emergency departments. Findings There were 29 770 injury cases with at least one EQ-5D score. The mean age of the participants providing data was 51 years. Most participants were male and almost a third had road traffic injuries. The new disability weights were higher for admitted cases than for cases confined to emergency departments and higher than the corresponding weights used by the Global Burden of Disease 2013 study. Long-term disability was common in most categories of injuries. Conclusion Injury is often a chronic disorder and burden of disease estimates should reflect this. Application of the new weights to burden studies would substantially increase estimates of disability-adjusted life-years and provide a more accurate reflection of the impact of injuries on peoples’ lives.


Introduction
If resource allocation and policy for the reduction of the burden of health problems are to be effective, the burden posed by injuries needs to be carefully evaluated. The disability-adjusted life-year (DALY), as used in the Global Burden of Disease (GBD) 1990, 2010 and 2013 studies, 1,2 is based on both premature mortality -i.e. years of life lost -and years lived with disability (YLD). 3,4 The assignment of disability weights, to represent the decrease in health associated with specific diseases or injuries, is a fundamental step in the estimation of YLD. 3,5 Different approaches to estimating disability weights 3 can lead to substantially different estimates of DALYs and YLD. 6,7 In panel-based studies of health burden, a lay description -a vignette -is used to represent the health impact of the condition of interest on a hypothetical affected individual. Health professionals or representatives of the general population then give the health status of that affected individual a score, or panel-based disability weight, that ranges between zero -representing no disability or perfect health -and one -representing disability equivalent to death. 3,5 The limitations of such a panel-based approach include the uncertain generalizability of the resultant weights to different geographical and socioeconomic contexts, the difficulty of developing vignettes to represent complex and varied health impacts and the limited focus on the time-course of any disability. 4,5 In an alternative to the panel-based approach, selfreported data collected directly from affected individuals, using multi-attribute utility instruments -such as the EQ-5D standardized measures of health status -can be used to derive case-based disability weights. 3 An individual's responses to a standardized set of questions can be used to determine that individual's generic health state and then the health states of all respondents having a particular health problem can be used to assign a disability weight to that problem. It has been suggested that such case-based disability weights should be used to quantify injury burdens. [8][9][10] Two studies based on injury cohorts led to case-based weights that were larger than corresponding panel-based estimates, but both studies were limited by small sample sizes. 6,7 The GBD 2013 study incorporated case-based weights for some injury groups but was hampered by the limited availability of case-reported data. 11 As an adjunct or alternative to the use of panel-based weights in burden of disease studies, we used pooled patient-reported data, from six longitudinal injury-outcome studies, to create case-based weights for individual injury diagnosis codes and established nature-of-injury classifications.

Setting
Our investigation was based on the Validating and Improving Injury Burden Estimates Study (Injury-VIBES) cohort, which consists of participants' data from six longitudinal Objective To create patient-based disability weights for individual injury diagnosis codes and nature-of-injury classifications, for use, as an alternative to panel-based weights, in studies on the burden of disease. Methods Self-reported data based on the EQ-5D standardized measure of health status were collected from 29 770 participants in the Injury-VIBES injury cohort study, which covered Australia, the Netherlands, New Zealand, the United Kingdom of Great Britain and Northern Ireland and the United States of America. The data were combined to calculate new disability weights for each common injury classification and for each type of diagnosis covered by the 10th revision of the International statistical classification of diseases and related health problems. Weights were calculated separately for hospital admissions and presentations confined to emergency departments. Findings There were 29 770 injury cases with at least one EQ-5D score. The mean age of the participants providing data was 51 years. Most participants were male and almost a third had road traffic injuries. The new disability weights were higher for admitted cases than for cases confined to emergency departments and higher than the corresponding weights used by the Global Burden of Disease 2013 study. Long-term disability was common in most categories of injuries. Conclusion Injury is often a chronic disorder and burden of disease estimates should reflect this. Application of the new weights to burden studies would substantially increase estimates of disability-adjusted life-years and provide a more accurate reflection of the impact of injuries on peoples' lives. studies in five countries (Table 1). 19 The main aim of the Injury-VIBES study is to improve the measurement of nonfatal injury burden through analysis of pooled, de-identified, patient-level data. Our investigation was approved by Monash University's Human Research Ethics Committee.

Data sets
We investigated persons with injury aged at least 18 years who were included in two Australian registries -that is, the Victorian State Trauma Registry 16

Injury classifications
When possible, weights were initially calculated for each of the four-character principal diagnosis codes listed in the 10th revision of the International statistical classification of diseases and related health problems (ICD-10) 20 and then mapped to each of the 47 injury groups used in the GBD 2013 study, 11 each of the 39 EUROCOST classification groups 21 and each of the European Injury Data Base groupings. 22 The ICD-10 codes for the cases from the USA were derived from the ICD-9 codes used in the data set. The Dutch data set only categorized injuries into the European Injury Data Base groupings. Although we could recategorize the Dutch patients into the injury groups used in the GBD 2013 study, we could not use the data from these patients to estimate weights for individual ICD-10 diagnosis codes.

Disability weights
In general, the patients' responses to the questions in the three-level EQ-5D questionnaire were used to estimate disability weights. The questionnaire is designed to record a respondent's self-reported health status in terms of five topics: (i) anxiety/depression; (ii) mobility; (iii) pain/discomfort; (iv) self-care; and (v) usual activities. For each of these topics, a respondent is asked if they have no problems, some problems or extreme problems. 23 The three-level EQ-5D questionnaire was used for the Australian cases from 2009 onwards and for all the injury cases included in the participating British, Dutch and New Zealand data sets. For all the other cases we considered, the recorded responses to the questions in the 12-item Short Form Health Survey 24 had to be translated into EQ-5D responses. 24 EQ-5D responses are used to calculate a preference score for each respondent. Such scores can range from −0.59 to 1.00. Negative values and values of zero and one indicate, respectively, respondents who have health states that are worse than death or equivalent to death and respondents who are in perfect health. 1 Disability weights were calculated at three time points -that is at three, six and 12 months post-injury -by subtracting the EQ-5D preference scores for respondents with a particular health problem from the age-and sexspecific norms. 23 The average EQ-5D differences at each time point were multiplied by a factor corresponding to the length of the period over which the disability weight applied and then these weighted disability averages were summed to provide an annualized or time-averaged disability weight. Thus, the calculated averages at three, six and 12 months were multiplied by 3/12, 3/12 and 6/12, respectively, with the resulting three weighted disability averages then summed together to produce a single disability weight. The nine-month outcomes from the Dutch data set were included in the 12-month estimates. Weights calculated at 12 months post-injury -hereafter called 12-month weights -were assumed to represent both the degree of residual disability at 12 months and the expected lifelong disability. 12,25 We compared our new disability weights with the one-year Integration of European Injury Statistics weights 21 and the long-term weights -for treated cases when weights for treated and untreated cases were given separately -of the GBD 2013 study. 11 The former represent injured cases admitted to hospital while the latter represent cases who warrant "some form of health care in a system with full access to health care". 1, 21 We calculated new disability weights separately for cases admitted to hospital and for other cases who only

Results
Across the six data sets and three different time points we investigated, there were 29 770 injury cases with at least one EQ-5D score -9003, 20 929 and 24 894 responses were recorded at three, six and 12 months post-injury, respectively. The mean age of the respondents was 51 years, most of them were male and almost a third of them had had road traffic injuries. The proportion of the cases from each data set that had been admitted to hospital ranged from 25% to 100% (Table 2). To save space, we have not reported weights for European Injury Data Base groupings but these are available from the corresponding author.

GBD 2013 injury categories
There were insufficient case numbers to calculate new disability weights for admitted cases in 14 of the 40 nature-ofinjury categories used in the GBD 2013 study (  (Table 3). However, the new weights for hospitalized cases of severe traumatic brain injury and spinal cord lesion at neck level were lower than the corresponding GBD 2013 weights (Table 3). Long-term outcome data for injury cases not admitted to hospital were only available for 16 of the nature-of-injury categories used in the GBD 2013 study ( Table 4). The new disability weights for such cases were much lower than the corresponding weights for the admitted cases and several were near zero -indicating that long-term disability is unlikely to occur (Table 4).

EUROCOST injury groups
Annualized new disability weights were calculated for admitted cases sustaining injuries in 31 EUROCOST groups (Table 5). These new weights were lower than the corresponding Integration of European Injury Statistics weights for all but three groups -facial fractures, open facial wounds and spinal cord injuries (Table 5) -and higher than the corresponding new weights for cases not admitted to hospital, several of which were close to -or less than -zero (Table 6).

ICD-10 diagnosis codes
Within the data sets we investigated, there were at least 30 cases admitted to hospital for each of 80 ICD-10 codes (

Discussion
We found differences between our new weights, which were based entirely on case-reported outcomes, and the corresponding GBD 2013 weights, which were based on a combination of panelbased and case-outcome studies. It could be argued that our new weights are not directly comparable with the GBD 2013 weights, due to distinctly different approaches to weight generation, although either set of weights could be used to derive population-based measures of injury burden. The GBD studies primarily relied on the responses of a public panel or panel of experts when faced with a standardized set of brief descriptors. Our new weights are entirely based on case-reported outcomes from cohort studies in high-income countries. The GBD studies, our study and other epidemiological studies designed to generate disability weights have generally not explicitly considered the extent to which factors such as socioeconomic status, access to high-quality care, environmental barriers or resilience, adaptation and the coping strategies of injured individuals can influence the lived experience of injury-related disability.
One argument for the preferential use of panel-based weights is the potential for individuals with chronic conditions to adapt and underestimate disease burden. 26 In general, however, our new weights -like the case-based Integration of European Injury Statistics weights -were substantially higher Table 2. Demographics of the patients from six injury cohorts who had an eligible EQ-5D summary score at three, six and/or 12 months post-injury

Research
New disability weights for injury burden studies Belinda J Gabbe et al.
than the largely panel-based GBD 2013 weights. This difference was especially marked for the more common categories of injury such as fractures and dislocations. In a previous study, estimates of injury burden based on data collected from the general public were generally found to be lower than those estimated from the experiences of the injured, particularly for categories of injury that are generally perceived to be less severe, such as sprains and fractures. 6 However, those living with spinal cord injury reported less disability than that predicted by the general public. 6 The general public's overestimation of the burden of disability resulting from some severe injuries may reflect the limitations of the New disability weights for injury burden studies Belinda J Gabbe et al.
vignette to convey the variability in disability within injuries adequately. This could explain why our new weights for severe traumatic brain injury and spinal cord lesion at neck level are substantially lower than the corresponding GBD 2013 weights. A perceived benefit of the case-based approach is the capacity to evaluate variation in disability within an injury group.
An argument for favouring estimates of disease burdens based on the perceptions of the general public over those based on the responses of the diseased has been that people living

Research
New disability weights for injury burden studies Belinda J Gabbe et al.
with a disease may have difficulty in placing their experiences in the context of other diseases. [26][27][28] Our new weights were based on the measurement of case-reported outcomes using validated multi-attribute utility instruments. Such instruments use population preferences to create norms for health states rather than for specific conditions. Their use helps to place the experience of people living with injury into a wide context.
Our new weights reflect the deviation of actual patient function from populationbased norms.
The panel-based approach requires a brief lay description of what living with a particular condition is like for a typical case. The description of a typical injury case is difficult because of the potential variation in the severity of the injury and in the injury's impact on the injured person's life. In the GBD 2013 study, the lay description of a spinal cord lesion below neck level, as used in the GBD 2010 study, was revised to include "and no urine and bowel control". This revision led to a sixfold increase in the corresponding disability weight -from 0.047 to 0.296. 11 In the case-based approach, the problems associated with the variable scope and specificity of lay descriptions are avoided. New disability weights for injury burden studies Belinda J Gabbe et al.
The results of our analysis indicated that all categories of injury treated via hospital admission -and most categories of injury treated only in emergency departments -were associated with persistent measurable disability. They also provided evidence of long-term disability for several injury groups where specific long-term weights were not provided by the GBD 2013 study. Similarly, where the GBD 2013 study provided long-term weights only for so-called untreated cases -for example for cases of fracture of the femur, radius or ulna -the corresponding new weights were relatively high, even though the new weights were based on cases recruited directly from health-care services in high-income countries that presumably, had access to relatively well resourced treatment.
Many EUROCOST and GBD injury groups combine several types of injury. The combination of several conditions into a single group -for which a single weight is estimated -is not problematic if the outcomes of the combined conditions are similar. Injuries of a single nature from a single body region, such as fractures within the shoulder, are often bundled together in this manner. However, our new disability weights for individual ICD-10 diagnosis codes (Table 7 and Table 8; available at: http://www.who.int/bulle.tin/volumes/94/10/16-172155) indicate considerable heterogeneity in disability experienced by patients with fractures in the same body region or even the same bone. For example, the new weights indicate that clavicle fractures have a much lower disability weight than Table 6. New disability weights for the nature-of injury-groups used by EUROCOST, as derived from the responses of patients, from six injury cohorts, who presented at emergency department but were not admitted to hospital Nature-of-injury group a n Mean new weights (95%CI)

Annualized At 12 months post-injury
Fracture/dislocation/strain/sprain of vertebrae/ spine  A major strength of our analysis was the large sample size -from multiple studies and health jurisdictions -which allowed weights to be estimated, for most commonly used injury classifications, for both hospital admissions and cases who were only treated in emergency departments. However, our analysis did have several limitations. The accuracy of the coding of injury diagnoses cannot be guaranteed, especially for cases attending emergency departments -whose injuries may not have been be recorded by a trained coder. Disability weights for some categories of injury were based on relatively small numbers of cases. We therefore provided 95% confidence intervals to indicate the precision of each weight estimate. Inconsistencies and errors in documentation from the GBD 2013 study 11 sometimes made it difficult to map ICD-10 codes to the relevant GBD 2013 injury group. The six data sets we employed differed in terms of follow-up rates and availability of EQ-5D data for each time point postinjury. Responder bias may have affected the British and Dutch data sets, which showed higher losses to follow-up than the other data sets. For some data sets, there was no collection of EQ-5D scores and we needed to estimate such scores from the responses to questions in the 12-item Short Form Health Survey.
For consistency and comparability, we mapped the principal diagnosis of each case to the EUROCOST and GBD 2013 injury groups. We did not take into account additional injury diagnoses even though disability at 12 months post-injury is known to increase with the number of injuries affecting the patient. 29 Future evaluation of injury weights should consider multiple injuries. Our method ignored recovery within three months and the data sets we used predominantly included cases of falls and road trauma. Penetrating injuries were underrepresented.
Our weights were also calculated using data from adult cases only. While the GBD studies do cover all age groups, the vignettes used in these studies have not accounted for differences between children and adults and the GBD weights have simply been assumed to be applicable to all ages. It is plausible that there are differences in the recovery trajectories of children and adults, although the magnitude of these differences is not yet known. Like the GBD 2013 weights, our new weights do not explicitly consider the presence of comorbidity. However, the new weights are calculated from responses to a multi-attribute utility instrument that included age-specific population preferences -and age is a partial proxy for comorbidity.
Our new weights were based entirely on data collected in high-income countries and it remains unclear if they could and should be applied to cases in low-and middle-income countries. Finally, we considered any disability reported 12 months post-injury as persistent. While some studies on injuries have shown little or no improvement after more than 12 months, 12,25 others have shown such late improvement as well as nonlinear recovery trajectories. 30,31 In conclusion, new case-based disability weights have been estimated for individual injury-related ICD-10 diagnosis codes and commonly used injury groups. In general, these weights were higher than the corresponding largely panel-based weights that have been estimated previously. Long-term disability was evident in all categories of injuries admitted to hospital. The findings indicate that injury is often a chronic disorder and burden of disease estimates should reflect this. The impact of applying the new disability weights to DALY calculations will depend on the injury incidence profile of the population studied. A similar case-based approach could be used to determine disability weights for other conditions. ■
Resultados Hubo 29 770 casos de lesiones con al menos una puntuación EQ-5D. La edad media de los participantes que ofrecieron información fue de 51 años. La mayoría de los participantes eran de sexo masculino y casi un tercio de sufrió lesiones por accidentes de tráfico. Los nuevos pesos de discapacidad fueron superiores en los casos de ingresos que en los limitados a los servicios de urgencias y mayores que los pesos correspondientes utilizados por el estudio de Carga Mundial de Morbilidad de 2013. La discapacidad a largo plazo fue común en la mayoría de las categorías de lesiones. Conclusión Una lesión suele ser un trastorno crónico y las estimaciones de la carga de morbilidad reflejan este hecho. La aplicación de los nuevos pesos a los estudios de carga podría aumentar sustancialmente la estimación de los años de vida ajustados por discapacidad, así como ofrecer un reflejo más preciso sobre el efecto de las lesiones en la vida de las personas.