Re-evaluating the D in DALY's for Mental Illness

Re-evaluating the D in DALYS for Mental Illness

Principal Investigators: Carole Siegel, Ph.D. & Eugene Laska, Ph.D. with Bedirhan Ustun, M.D. (World Health Organization), and Gary Haugland, M.A..

PROJECT GOALS

This project, conducted in conjunction with the World Health Organization, will develop a global measure of client outcome that encompasses multiple domains.

In most studies in which interventions for persons with mental disorders are compared, outcomes are assessed that cover multiple domains, and results are usually reported specific to each domain. However, a global appraisal of outcome is often desirable as a basis for ranking or selecting interventions. This is especially true in cost-effective analysis (CEA). To date, there is no satisfactory method for combining outcome measures across domains, particularly when they are based on different units. We propose to develop a global measure of outcome based on domain outcome measures where the unit of measurement is time.

Generically, the approach that we propose will construct a time-adjusted measure of multiple outcomes assessed from structured instruments. The method will be operationalized specifically for DALYs, the measure used in the World Bank report on the Global Burden of Illness to assess the burden of illnesses (including mental illnesses around the world). In the current version of the DALY methodology, the level of disability of an illness -- a quantity used to adjust time -- is based on a qualitative ranking of disability states. Specifically, we propose to:

Develop a methodology to form an adjusted time measure from client level data from multiple assessments of illness-related health states or disablements;
Operationalize the methodology for the WHO Disability Assessment Instrument;
Develop a protocol, in conjunction with WHO, to:

i) calculate the time adjustment weights,
ii) test their reliability, and
iii) compare the values obtained with those used in the World Bank report and those obtained by WHO for the same samples using preference-based assessments.

RESEARCH ACTIVITIES AND RESULTS

Aim 1:

We will develop a method for adjusting the time spent in a defined period of time, T, to reflect the quality of life or the burden of disability experienced within that period. As examples, T can represent a three-month follow-up period of a clinical trial, or the time from onset of illness to death (the time frame of DALYs). Suppose that measures are available on K mental health states or conditions of living experienced by a person during T that are derived from one or more structured measurement instruments (e.g., scores on symptom severity, work activity, and social activity). In the general case, the units of measurements are not time but are ordinal values representing, in some sense, the level of severity or the disadvantage resulting from the condition being measured. Let mk denote the measurement of state k and tk represent the corresponding slot of time spent in state k during period T for k = 1,2,..., K. The correspondence of state and time may not be known and more than one state can be operant in the same time slot.

The following simplifying assumptions are made so that we can illustrate the line of thinking that will be used in this project. First, assume the ti's are known, are non-overlapping and St i = T. This is the framework used in the Quality-adjusted Time Without Symptoms and Toxicity (Q-TWiST) method: a quality-adjusted survival analysis that integrates quality of life considerations into the comparison of (more than benign) treatments in a clinical trial (Goldhirsch et al., 1989). T is the time from trial entry to death and clear-cut health state transitions are delineated. Time is decomposed into mutually exclusive intervals, t1 = time with treatment toxicity, t2 = time without symptoms and toxicity and t3 = time in relapse. We present two approaches to measurement development corresponding to this simple situation.

Approach 1. We make the assumption that the value or utility of time to a person (or society) is directly proportional to the measurement of the health state of the person in the time slot, i.e., the value of the time in tj is proportional to mj. Let m*j represent the average value of the jth health state in a "healthy" person. Then Si(mi/m*i) ti is a health state adjusted amount of time of a person with health state measurements mi (the adjusted score will vary between 0 and 1). To obtain a population level measure two options are possible: the client level measures can be averaged to the population level; or the mi can first be averaged over the population to obtain, say, E(mi) and then ti is adjusted by a population based weight such as E(mi)/m*i.

Approach 2. Since the measures mj are not likely to be interval-scaled, panels of clients and clinicians are asked to estimate the value of time diminution implied by a measurement outcome. Thus if mi has response categories mij, j = 1,2,..J. the panel would set aij, the time diminution weight for response j. The time adjusted measure for an individual whose responses are a1j1, a2j2...,aKjK is Si(aiji) ti.

Methodological work required: In general, the mapping of the health state measurements into the operant time domain will not be known (i.e., we will not know how mi -->ti). Furthermore, the ti's are probably not mutually exclusive (i.e., many health states may be operant in the same time slot). The first task will be to develop a map of the measurement domain into the time domain. In this case ti will represent the distinct time slot in which a distinct set of one or more health states is operant (i.e.,{m}i -->ti) where{m}i is the set of measurements operant in time slot ti. A panel of approximately 15 persons comprised of experts on the health states, consumers and family members will be asked to develop the mappings, and for approach 2 to set aij. Several techniques can be used to facilitate the development. These include case analysis, idealized patient scenarios and personal experiences.

The second task is to develop weights to adjust the value of the time ti corresponding to the set {m}i. Concepts from multi-attribute theory will be applied in which preference measures are developed for multi-attribute states in terms of the preferences for individual attributes. Here, health state i corresponds to attribute i. The preference for health state i will correspond to the weight mi/m*i (or aij). Additive and multiplicative functions of the preferences for the individual attributes will be considered as candidate measures based on the properties inherent in the function. These involve assumptions made on marginal conditions.

Aim 2:

The method will be operationalized for two illnesses: schizophrenia and unipolar major depression using the Disability Assessment Instrument (DAI) being developed by WHO. DAI measures illness-related levels of disability that a person is currently experiencing. Items are organized within domains. The domains for the short version of the instrument include:

Understanding and interacting with the world,
Getting around,
Self Care,
Life activities,
Getting on with people, and
Involvement in citizenship and economic life.

Items and domains will be reviewed to determine whether reaggregation of items into other domains or direct item mapping into time is required to facilitate the forming of the measurement to time mappings The time frame T for this application is age of illness onset to death. The time period will be decomposed into three sub-periods: around onset, mid-years of illness, and late illness phase. For each illness for each of these three sub-periods, the time mapping will be performed as described above.

Aim 3:

In order to develop the weights, data will be required on either ill (approach 2) or healthy and ill populations (approach 1) of similar characteristics (except for illness) for the sub-periods. The protocol to conduct this phase of the study will be written in year 2 of the project, but the project will only commence after the DAI instrument has been pilot tested. WHO has received funding to apply the DAI on samples that represent the illnesses and time periods discussed above, and to develop the weights and compare various approaches.

This approach should provide, at a minimum, the same level of discrimination among disorders as preference based approaches. If it does not, this would indicate that the method does not have the same level of "discriminant" validity as currently used methods (e.g., the World Bank method based on person trade-off ratings). Further, we will examine the ordering of the values obtained in comparison to the orderings obtained by other methods. We are limited in validation approaches since there are no gold standards to refer to, and predictive validity testing would entail budgets of large magnitude. We remark that the preference ratings used in the DALY calculation used by the World Bank were not validated. As discussed above in the literature review, weights obtained by the same panels differ for the different preference methods. Discrepancies that are found among methods will be discussed with consumers, providers and family members. This will provide a level of "face" validation of the proposed methodology.

SIGNIFICANCE OF FINDINGS/POLICY IMPLICATIONS

The significance of this project will be useful in many mental health services situations -- to monitor outcomes in managed care, to conduct cost-effectiveness analysis, as well as its application in this project to measure the global burden of mental disorders.

(Months): 0-6: Development of methodology; 6- 21: Operationalize method using DAI; 21-24: Data collection protocol, 21-36: Applying methods to data collected by WHO collaborators.

Updated: 3/23/99

[Top]