Background
Numerous instruments measure outcomes in palliative care. However, these instruments are not preference-based and so are unable to yield quality-adjusted life years (QALYs). Conversely, the generic preference-based measures (PBM) are neither appropriate nor sensitive enough to measure changes in palliative care outcomes.
Aim
The objective of this thesis was to propose a new palliative-care health state classification system, called POS-E, and to estimate its preference weights so that it can be used in QALY calculations.
Methods and results
Design: Secondary data analysis followed by a cross sectional survey informed by the Health Technology Assessment guide for developing and testing methods for deriving preference-based measures of health from condition-specific measures (and other patient-based measures of outcome).
This thesis consists of two main stages. The first stage explored the feasibility of deriving QALYs “indirectly” by mapping a commonly used palliative care instrument – the Palliative Care Outcome Scale (POS) – onto a generic PBM (EQ-5D; 3-level) using regression techniques. Data from three studies which included both POS and EQ-5D (N = 783) were combined and then randomly split into estimation and validation data sets. The overlap between both measures was assessed using principal component analysis (PCA) and correlation matrix analysis. Subsequently, we attempted to map the POS to the EQ-5D using three regression models. The mean absolute error (MAE), adjusted R2, and predicted/observed plots were used to assess the quality of mappings. The results showed low correlations between both instruments and that the POS is associated with only two EQ-5D dimensions (pain; and anxiety/depression). No POS items loaded onto the mobility; self-care; and usual-activities dimensions of the EQ-5D. The mapping models performed poorly at predicting utilities from POS data (MAE > 0.3 and R2 <0.10). This suggests that the EQ-5D did not capture some important aspects of the POS, and therefore mapping is unlikely to provide an appropriate basis for indirectly estimating utilities for conducting economic evaluations in palliative care studies.
The second stage explored the feasibility of directly deriving palliative-care-specific QALYs using the POS. The methods used here were based on the Health Technology Assessment (HTA) guidance document on “developing and testing methods for deriving preference based measures of health from condition-specific measures (and other patient-based measures of outcome)”, Brazier et al, 2012. This stage was further divided into three phases consisting of the following:
1) Revising the palliative outcome scale (POS) into a simplified health state classification amenable to valuation using items selected using Rasch and factor analyses. This involved combining data from six studies of patients receiving palliative care (N = 1011) and splitting this into two random halves – development and validation data. Analysis was undertaken on the development data and results were validated by repeating the analysis on the validation dataset. Following this, a classification system made of seven items with 1–2 levels each was derived. The POS-E describes a total 1,458 discrete health state. From this, Rasch analysis identified 14 plausible health states that were appropriate for valuation;
2) A valuation survey of palliative-care patients and healthy volunteers using a modified time-trade-off technique (TTO) to derive preference weights for the 14 health states derived from phase (1) above, and a comparison of the difference between patient values and values from healthy volunteers using simple t-tests. 102 participants (52 palliative care patients and 50 healthy volunteers) were surveyed across five sites in England. Each respondent valued eight health states and the analysis was based on a total of 408 valuations. Patient values and healthy volunteer values were very similar, with some areas of divergence. Mean TTO values ranged from 0.21 to 1 for patients, and 0.22 to 0.99 for healthy volunteers, for the worst and best health states respectively.
All TTO values corresponded logically with the order of severity of the health state classification, thereby supporting the internal validity of the health state classification system.
3) Estimating the utility weights for the full set of health states using regression techniques. This involved using regression analysis to estimate utility values for all health states using the Rasch logit score for dimensions that are correlated (i.e. have unidimensional properties).
This involved using regression models to estimate the relationship between the utility values obtained from the valuation survey, and their corresponding Rasch logit score. Subsequently, this mathematical relationship was used to estimate the utility values of all other POS-E health states (which were excluded from the valuation survey) based on their respective Rasch logit scores. Having tested several models, the model chosen for estimating the mean preference values of all other POS-E health state was the one which included linear, quadratic, cubic, and quartic terms, as it had the least predictive error and best explained the variation in the preference values obtained from the valuation survey (as indicated by low RMSE and high R-squared values), and its coefficients were all statistically significant.
Conclusion
In conclusion, the output of this thesis is a palliative care specific preference-based measure (POS-E) that can be used to calculate QALYs for cost-utility analysis of palliative care interventions. This thesis has addressed most of the theoretical and methodological concerns of cost-utility analysis in palliative care. It demonstrated that it is feasible to obtain meaningful preference values from patients with advanced chronic illness, and also that patient values are similar to similar to those of healthy people. It also showed that the QALY is still a useful vehicle for quantifying joint mortality and morbidity impacts of palliative care at individual and population level. Given the widespread use of the POS in assessing palliative care outcomes in the UK and internationally, the POS-E is expected to facilitate broader economic evaluations of palliative care interventions using current and forthcoming POS data sets.
Original language | English |
---|
Awarding Institution | |
---|
Supervisors/Advisors | |
---|
Award date | 2018 |
---|