Article Text
Abstract
Background: Disability weights are necessary to estimate the disability component (years lived with disability, YLDs) of disability-adjusted life years. The original global burden of disease approach to deriving disability weights ignores temporary consequences of injury.
Objectives: To develop and apply novel empirical disability weights to improve estimates of the non-fatal burden of injury.
Methods: A set of 45 disability weights was derived for both permanent and temporary consequences of injury, using the annual profile approach. A population panel (n = 143) provided the values. The novel set of disability weights was then linked to epidemiological surveillance data on the incidence of injury in The Netherlands to calculate YLD resulting from permanent and temporary consequences of injury.
Results: The empirical disability weights for injury consequences varied from minor (corneal abrasion, 0.004) to very severe (quadriplegia, 0.719) health loss. Increasing disability weights by level of severity were found, as illustrated by concussion (0.02), versus moderate brain injury (0.193), versus severe brain injury (0.540). Application of these new disability weights showed a 36% increase in YLD as the result of unintentional injury.
Conclusions: YLD calculations based on global burden of disease disability weights underestimate the size of the injury problem by ignoring temporary health consequences. Application of novel empirical disability weights, derived using the annual profile approach, may improve calculations on the burden of non-fatal injury.
Statistics from Altmetric.com
Worldwide, injuries are recognized as a major concern in clinical and public health. To reduce the number of injuries, a range of preventive interventions have been implemented, and more are being considered depending on opportunities and resources. However, resources available for prevention are limited. For policy makers, quantification of the burden of disease or injury and the subsequent determination of the relative attributes of avoidable risk factors to this burden are vital tools for priority setting.
The burden of disease concept was introduced by the World Health Organization in 1996.1 It represents the total health loss at the population level, and is quantified in terms of disability-adjusted life years (DALYs). DALYs summarize (a) time lost through premature mortality (years of life lost), and (b) “healthy” time lost while living with a disability (years lived with disability (YLDs)) in a single quantity. An essential factor for establishing YLDs is the disability weight, a value assigned to living with disability.1 This value, anchored between 0 and 1, reflects the effect of a specific health condition and is usually based on the preferences of a panel of judges. For this, dedicated preference measurement methods are used.23
For the global burden of disease (GBD) approach—which aims to estimate the total burden of disease worldwide as the sum of the burden of all separate diseases—a large set of disability weights had to be derived. Over 100 specific disease categories with a disability component have been distinguished.1 The initial approach used by Murray et al45 was based on choices that have subsequently been challenged. Firstly, to arrive at the GBD disability weights, the disability-causing health states were presented to the panel of judges as a limited diagnostic description of the disease, derived from the terminology of the International Classification of Diseases (ICD). Apart from the disadvantage that only physicians could be involved in the evaluation, later studies showed that, even for expert panels, adding a generic description of functional health status to the diagnostic disease label is necessary to standardize the stimulus.6–8 Secondly, in computing the DALYs, the GBD approach assumes independence between duration and disability and requires that the health state remains constant over time,910 an assumption that is untenable for many diseases and injury consequences.
To overcome the operational limitations of the GBD approach, Stouthard et al7 developed an alternative approach, the annual profile approach (APA). The most important feature of the APA is that the course of the health state—the disability profile—is described over a period of 1 year, alleviating the time constraints of the GBD approach.10 This allows derivation of disability weights for health states with an acute onset, episodic diseases such as epilepsy, and health states characterized by complex and heterogeneous recovery patterns, such as injury consequences.
Injuries encompass a wide range of health states that may have both temporary and permanent health consequences. Although temporary consequences of injury have a relatively small effect on health-related quality of an individual life, they are highly prevalent, and, collectively, their impact on health burden is significant.11–13 By ignoring temporary consequences of injury, the GBD approach may typically underestimate the burden of injury and specific external causes, and therefore affect prioritization of resources for injury prevention. However, the GBD disability weights are currently used internationally to assess the burden of non-fatal injury, as alternative disability weights are not available.14
In this study, our objective was to develop and apply novel empirical disability weights to improve estimates of the burden of non-fatal injury.
METHODS
Derivation of disability weights
Injury selection and health-state description
For the selection of injury consequences, the previously derived European injury classification EUROCOST was used.15 This classification is compatible with the International Statistical Classification of Diseases, Injuries and Causes of Death-9th revision (ICD-9) classification system. The 39 EUROCOST injury groups were subdivided into 45 injury stages that are homogeneous in terms of healthcare use, disability, as well as treatment and prognosis.
Each of these 45 injury stages was presented separately on a vignette (a preformatted A4 size sheet of paper that describes the functional consequences of an injury stage with four quadrants of disease information.) The left upper quadrant provided the disease label and clinical description. The upper right quadrant provided a generic description. For this generic description, we used the extended version of the EuroQol 5D (EQ 5D).71617 The lower right quadrant described the course of the condition over 1 year. Conditions with short duration were presented as a patient who, in an otherwise healthy year, experiences, for instance, the health effect of a concussion for 2 weeks, whereas conditions with long-term consequences were presented as a patient who experiences, for instance, the health effect of a moderate brain injury throughout the whole year. The lower left quadrant gave a visual representation of the injury.
For the generic descriptions of the 43 injury stages, we used data from a survey administered to a representative sample of patients who attended an emergency department in The Netherlands because of injury.18
Health-state evaluation
To elicit preferences for the 43 injury stages, two evaluation techniques were used: the visual analogue scale (VAS) and the time trade-off (TTO).19 The VAS evaluation technique requires participants to score the injury stage on a vertical scale graded from 0 (worst imaginable health state) to 100 (best imaginable health state). With TTO, the participants were asked how much time they would be willing to “trade” in order to be restored from the presented disease stage to full health. All vignettes were valued independently according to both methods.
Panel participants and data collection
The panel participants were lay people who applied to participate in the Mild Diseases and Ailments Study, conducted in 2003.20 For that study, 560 people were recruited from the general public through an advertisement in a newspaper that is freely available throughout The Netherlands. These people were used as a sampling framework for our study on injury disability weights. For this purpose, a random sample of 250 people was contacted by mail; 170 of these were willing to participate.
Initially, the panel participants attended a 3 h panel meeting, during which they evaluated 10 vignettes. Then they received a postal questionnaire. We developed three versions of the questionnaire, each containing 22 different vignettes in randomized order. The participants received €50 for participating in the study.
Data analysis
Firstly, we calculated VAS weights and TTO weights for each response using the following formulas:
VAS weight = 1– (VAS score/100)
TTO weight = TTO score/365
To establish whether the injury stages were ranked in a consistent order, the ranking of a vignette according to VAS and TTO weights was compared using the Spearman’s rank correlation coefficient and the Pearson’s rank correlation coefficient. We also performed regression analysis to determine whether age, sex and education level had independent significant effects on the VAS and TTO weights.
The disability weights were also calculated on the basis of both mean VAS and mean TTO weights using a log-transformation procedure.21 Previous studies have shown that the evaluation technique used to elicit preferences may affect the value of the preferences.21–23 The transformation procedure aims to adjust for this potential effect of the evaluation technique.
Calculation of burden of non-fatal injury (YLD)
The following surveillance systems provided national data on injury incidence and their consequences:
The Dutch Injury Surveillance system: a permanent registry of injuries treated at the emergency department of 17 hospitals.18
The Second Dutch National Survey of General Practice: a nationwide survey, including data on visits to a general practitioner for injury; performed by The Netherlands Institute for Health Services Research (NIVEL) in 2001.24
A follow-up survey administered to a representative sample of patients with injury up to 2 years after treatment at the emergency department of a hospital.18
To calculate YLD, incidence data from the surveillance systems were multiplied by mean duration and the log-transformed disability weight derived from the panel study. Injury consequences for which at least 50% of the participants were not willing to “trade off” any time were considered to be trivial and were excluded from the YLD calculation—that is, corneal abrasion and superficial injury, for which 72% and 75%, respectively, of the participants were not willing to trade off any time.
RESULTS
Disability weights
A total of 143 participants attended the panel meeting, 136 (95%) of whom responded to the questionnaire. The response rates of the three versions of the postal questionnaire were 94%, 98% and 96%. Table 1 gives some details of the participants.
Table 2 presents the mean and median VAS and TTO weights for the 45 injury stages. The results show that the participants evaluated the injury stages logically; lowest VAS and TTO weights were attributed to mild injuries, such as superficial injury (mean VAS 0.09, mean TTO 0.01), and highest to severe injuries, such as quadriplegia (mean VAS 0.89, mean TTO 0.64). This is illustrated by increasing disability weights by level of severity within specific types of injury. For example, concussion (mean VAS 0.20, mean TTO 0.01) is rated lower than moderate brain injury (mean VAS 0.55, mean TTO 0.27) and severe brain injury (mean VAS 0.80, mean TTO 0.33). Furthermore, the results from table 2 show that, if injuries were subdivided into an acute phase and a chronic phase, the acute phase had higher VAS and TTO values (less preferable). For example, the acute phase of multiple injury without brain injury (mean VAS 0.65, mean TTO 0.27) is rated higher than the stable phase of multiple injury without brain injury (mean VAS 0.49, mean TTO 0.23). In contrast with the VAS, the TTO ranking of some of the severe injury stages did not correspond to the severity of the injury stage.
The Spearman correlation coefficient between mean VAS and TTO values was 0.97, indicating a similar rank ordering of the injury stages. Pearson’s correlation coefficient was 0.93, indicating similar values of the injury stages.
The standard deviation of the TTO values (data not shown) was higher in the middle range, which was expected given the fixed end points of the scale. No significant effects of sex, age, or educational level on the VAS and TTO values of the injury stages were shown.
Table 2 also shows the log-transformed disability weights.
Calculation of burden of injury
The data sources showed that, in 2003 in The Netherlands, medical advice for the consequences of unintentional injuries was sought 1 714 000 times—that is, 11% of the population (population at risk 16 188 390). Almost half (48%, 830 000) of the people who sought medical advice were treated at the emergency department of a hospital. Of these, 5% (37 000) had long-term consequences (2 years after the injury).
Table 3 presents the number of YLDs due to unintentional injuries.
There were 116 000 YLDs due to unintentional injuries, 52% of which were due to home and leisure injuries, 27% to traffic injuries, 16% to sport injuries, and 5% to occupational injuries.
Short-term consequences of injury contributed 31 000 YLDs to the total estimate of YLDs due to unintentional injuries, an increase of 36% on top of the 86 000 YLDs due to long-term consequences. For home and leisure injuries, traffic injuries, occupational injuries, and sport injuries, short-term consequences contributed 17 000 YLDs (39%), 5500 YLDs (21%), 1900 YLDs (49%) and 6200 YLDs (50%), respectively.
DISCUSSION
This study derived a set of disability weights for injury consequences, varying from minor to very severe health loss, using the APA. The empirically derived disability weights were used to calculate the burden of injury due to unintentional injuries in The Netherlands. We found that inclusion of injuries with temporary consequences increased the non-fatal burden of injury in The Netherlands by 36%. Short-term consequences were particularly important in sport and occupational injuries: 50% increase in the number of YLDs.
In this study, we used the relatively new approach, APA, rather than the standard QALY/DALY approach (SQA) applied in the original GBD study.1 The SQA consists of adding up separately evaluated periods lived in a certain state of health. During these separate periods, state of health is assumed to remain constant. The implication of this approach is that health outcome characterized by dynamic and complex patterns such as injury consequences have be separated into various periods of constant health.
In contrast with the SQA, the APA avoids the break down of health outcome into time periods because it describes health outcome over the course of 1 year.10 This 1-year perspective allows the derivation of disability weights for health outcome with short-term consequences, which have been ignored in previous burden estimates. To obtain a societal perspective, in this study, preferences for injury stages were obtained by a panel derived from the general population, who evaluated each of the injury stages with both a rating scale and a trade-off technique. In contrast, the majority of the GBD disability weights were obtained by an expert panel with a rating scale only.25
The APA has been criticized by Vos, 26 who stated that the disability weights derived with the APA overvalue diseases and injuries with a mild and rapid course. According to this criticism, the burden of short-term consequences calculated in this study may be an overestimation, and the observed increase in burden of non-fatal injury may be an effect of the APA. However, Essink-Bot and Bonsel10 pointed out that the relatively high values of the disability weights derived with the APA are due to sensitivity problems at the end points of the scale. Hence, the relatively high disability weights are a result of the lack of discrimination between conditions of low severity rather than the time presentation.
Key points
The original global burden of disease approach ignores temporary health consequences because of methodological constraints.
The annual profile approach used to derive novel empirical disability weights alleviates the constraints of the global burden of disease disability weights.
Application of the set of newly derived disability weights showed a 36% increase in years lived with disability due to unintentional injury.
Highly prevalent injuries with temporary consequences should not be ignored in burden of injury estimates, provided that there is a threshold to distinguish trivial from minimal relevant injuries.
To avoid the problem of disability weights of low-severity health states, it is important to distinguish trivial conditions from relevant conditions. Trivial conditions have a minor effect on the health-related quality of individual lives, but may be highly prevalent. Inclusion of these conditions in the calculation of burden of disease or injury may result in a large number of YLDs, which stand out compared with severe, but much less common, conditions. As a result, trivial injuries may become the focus of attention rather than severe injuries. This problem was also pointed out by Sim and Mackie, 27 who emphasized the importance of reliable indicators for measuring non-fatal injuries. Therefore, in this study we used the preference-based relevance threshold developed by Bonsel et al20 to distinguish trivial injury consequences from minimally relevant injury consequences. The preference-based threshold was met if the median TTO value was greater than 0—that is, if at least 50% of the participants were willing to trade-off any time. If not, the health state was regarded as trivial according to the majority of the population and therefore excluded from the calculation of burden of injury. This threshold is based on societal preferences, which concurs with the societal perspective of the burden of disease concept, and it corresponds to the majority rules principle of most democratic voting systems. The results show that two injury types did not meet the threshold, namely corneal abrasion and superficial injury. These two injury types were therefore excluded from the YLD calculations. Inclusion of these two injury types would result in a 3% increase in the number of YLDs (from 116 000 to 120 000).
To meet the societal perspective of the burden of disease concept, it is important that the disability weights reflect the preferences of the society under study.28 The disability weights derived in this study are based on preferences of a sample of the Dutch population. This, however, does not mean that the use of these disability weights is limited to The Netherlands. Two previous studies that compared health-state evaluations among residents of several countries found that ranking of health states is similar across countries.2930 Secondly, the disability weights from the 1997 Dutch Disability Weights Study have been applied in burden of disease studies of many other countries, such as Australia, New Zealand, South Africa, Serbia, and Spain.731–34
Considering the duration and complexity of the evaluation task, the response rates of the panel study were high, with 57% of invited people actually participating in the panel session and 95% of the participants returning the follow-up questionnaire. These high response rates may be explained by the relatively high incentive of €50, the careful design of the study materials to enhance information processing, and the use of reminders for the follow-up questionnaire. Two panel studies with similar study design and incentives yielded equally high response rates.2035
A limitation of this study is that, although the panel overall ranked the injury stages consistently and logically, the TTO values showed some unexpected inconsistencies for severe health states. With the TTO technique, participants seemed less able to discriminate between severe injury stages, suggesting that minor inconsistencies might be caused by the conceptually difficult trade-off evaluation technique. The VAS on the other hand is conceptually less difficult, and its values did not show inconsistencies. However, the VAS lacks the trade-off feature; it does not ask for sacrifice of something valuable in order to assess the undesirability of the health state. Consequently the preferences elicited with the VAS give less information about the relative desirability of a state of health compared with other health states and are therefore regarded as less appropriate for eliciting preferences and, for this reason, less appropriate for basing disability weights on.21 To adjust for the minor TTO inconsistencies, the disability weights are based on both VAS and TTO values.
We conclude that the disability weights that we have derived in this study may be used to overcome some of the limitations of the GBD disability weights. Furthermore, we conclude that highly prevalent injuries with temporary consequences should not be ignored in estimates of burden of injury, provided that there is a threshold of triviality.
IMPLICATIONS FOR PREVENTION
The burden of disease concept is used increasingly for priority setting in healthcare and prevention. However, calculations of burden of disease based on GBD disability weights ignore temporary health consequences, resulting in an underestimation of the size of the injury problem. This novel derived set of disability weights for injury consequences includes highly prevalent short-term consequences, improving burden of injury estimates. This allows policy makers to base their decisions regarding resource allocation on a more complete burden of injury.
Acknowledgments
We are grateful to the Ministry of Health, Welfare and Sport (VWS), The Netherlands and the Consumer Safety Institute, The Netherlands for sponsoring this study. We thank M F Janssen (Academic Medical Center, Amsterdam, The Netherlands) for help and advice with the organization of the panel study.
REFERENCES
Footnotes
Competing interests: None.