Search

Methotrexate is widely used as a first-line treatment for rheumatoid arthritis (RA). While methotrexate can be very beneficial to some people with RA, some do not respond to the treatment, and others can develop serious adverse drug reactions (ADRs). This thesis investigates the potential economic impact of introducing a precision medicine intervention based on a hypothetical predictive model of response and ADRs to methotrexate to improve the cost-effectiveness of using methotrexate as a first-line treatment for RA. A combination of methods was used to meet the objectives of this thesis. Two systematic reviews were conducted to identify: (i) published economic evaluations of methotrexate to treat RA and (ii) existing predictive models that could be used to target methotrexate based on response to treatment and/or ADRs. A microcosting study was employed to design and attach a cost to the hypothetical precision interventions aimed at the use of methotrexate as a first-line treatment for RA. To conduct an economic evaluation of these interventions compared with the current practice, a discrete event simulation model was conceptualised, developed, and implemented. An early model-based Cost-Effectiveness Analysis (CEA) was then conducted to estimate the indicative cost-effectiveness of the precision interventions and key drivers of cost-effectiveness. The review of economic evaluations of methotrexate identified and critiqued 20 studies and found that despite the long history of methotrexate use in RA treatment there was limited evidence to support its cost-effectiveness in the management of early RA. The review of predictive models found 12 studies, most of which did not attempt to build a predictive model and assess its predictive value. Two of the proposed predictive models were selected as the examples to design the hypothetical precision interventions. The microcosting study with sensitivity analyses estimated the unit costs of using a genetic and clinical predictive model to be used as parameter values in the early-CEA. The early-CEA found that the interventions were not cost-effective when compared with the current practice using currently available data and outlined key areas of uncertainty that should inform the further research on this topic. This thesis has shown that using early-CEA in the context of precision medicine approaches that are not yet fully developed was feasible and useful to drive future research agendas. Using existing data and current assumptions about the alternative management strategies, there was considerable doubt that the precision medicine approaches for methotrexate based on predictive models would be cost-effective. Future studies should aim to build on this early-CEA, using the feedback from this thesis as part of an iterative process to generate economic evidence for targeting the use of methotrexate as a first-line treatment for RA.

Cerebral palsy (CP) refers to a cluster of developmental disorders of movement and posture. This lifelong disability causes activity restriction and limits the individuals’ participation in everyday activities. Further, due to the CP being a lifelong disability, interventions and treatment impose a significant burden on the people with CP, their families and the health care system. Resources are scarce and governments need to know the best way to allocate these resources. A common method to address this requirement is using economic evaluation to provide essential information for resource allocation within the health care sector. Cost-utility analysis (CUA), provides a universal outcome for the economic evaluation of health care interventions that are expressed as a quality-adjusted life-year (QALY). QALYs combines quality and quantity of life into a single index. Quality of life for QALYs is valued by the preference weights given for different health states from the preference based instruments. Previously published systematic reviews have demonstrated that a limited number of CUAs were conducted for interventions for the CP population. The studies indicated that the current available preference based measures do not perform well in CP population, and there are no condition-specific preference based measures for CP currently available for CUA use in this population. Therefore, the main objective of this research is to develop a preference based measure for CP to be used in CUA. Preference based measures have two components; a health state classification system and a value set for the health states from the classification system. The first step was to develop the CP-specific classification system. The classification system was derived from a validated and widely used CP quality of life instrument, the Cerebral palsy quality of life (CPQOL). Factor analysis and Rasch analysis were applied to develop the classification system. Rasch analysis and experts opinion were used to evaluate the construct and content validity of the classification system at the development stage. The classification system consists of six domains: "Social wellbeing and acceptance", "Physical health", "Communication", "Pain and discomfort", "Manual ability" and "Sleep". Each domain is presented by one item and each item has 5 response levels. The new instrument is named the Cerebral Palsy 6 Dimension (CP-6D). The next step was to generate a scoring (utility) algorithm for the CP-6D. Utility weights are accrued from the preference of people using preference elicitation methods. A Discrete Choice Experiment (DCE) is an ordinal preference elicitation method that asks participants to choose between two or more discrete alternatives. A systematic review was conducted to find the best approach to apply a DCE for valuing health for a multi attribute instrument. The review illustrated that there is not a golden approach to apply a DCE. However, there were some similarities between the studies. For instance, most of the studies used an online survey within the general population. To choose an approach, researchers need to know the features of the value sets that they want to produce before using the value sets in decision making. The systematic review also confirmed that there are no CP-specific instruments that used a DCE to value health. To develop the utility algorithm for CP-6D, DCE was applied. When using DCEs to value health, they produce utility values on a latent scale; hence, the weights generated from the DCE need to be anchored onto a full health-dead scale to calculate QALYs. To anchor the values, DCEtto method was used. In DCEtto duration is added as an attribute when designing the DCE. The valuation study was done in two stages; at first, a pilot study was done with a zero prior design, the priors for the final design were estimated from the pilot study data. The main valuation study was conducted as an online survey. The sample was a representative sample of Australian general population in age and sex. A total of 2002 adults aged 18 years and above completed the survey. The survey included a set of DCEtto tasks, basic social-demographic questions, the CP-6D, and a generic preference based measure (AQoL-4D). DCEtto data were analysed using conditional and mixed logit. All estimated coefficients from the data were in the expected direction and order for all the dimensions. In both methods, all coefficients were negative as expected, except level two of Physical health, which was not statistically significantly different from zero. The Australian utility values of CP-6D ranges from -0.582 to 1.000. The utility algorithm will enable the calculation of utility values for CP-6D in economic evaluations. The same sample as the DCE study was used to evaluate the construct and criterion validity of the CP-6D, as outlined in the COSMIN checklist. The validity was then evaluated using the AQoL-4D in a general population. There was a moderate correlation between the CP-6D and the AQoL-4D (~0.64), which suggested that the instruments are measuring a similar concept. However, the correlation between domains and items were small, this could be due to different factors. Socio-demographic variable changes had the same effect on both instruments. ANOVA and t-test were used to show the changes in socio-demographic variables; the effect size was medium, and all the results were significant (p-value < 0.001). This study filled a major gap in the literature in providing health state valuation for CP. A new health state classification system and utility algorithm completes a new preference based Health-Related Quality of Life (HRQoL) measure for CP, however further use and external validation is required. This study has some limitations. Due to the resource constraints and feasibility, the CP-6D was developed to evaluate CP interventions in economic evaluations among the Australian population also like other patient preference settings it can be used for informing patient care and clinical decisions. However, there is a high potential that the instrument to be used in other countries, future research is recommended to estimate the scores in different countries. Further, the CP-6D should be validated rigorously with people that have CP in a clinical setting. Future studies are recommended to assess whether there is any significant difference between health state utility values for CP-6D derived from the general population and the preferences of people with CP. The CP-6D will facilitate the assessment of CP related interventions using a CUA framework and will assist resource allocation through economic evaluations for CP interventions and treatments, the most common motor disability among children.

Oral diseases are considered a significant public health problem due to their high prevalence and impact on people’s day-to-day activities and quality of life (QoL). Further, treatment of oral diseases imposes a substantial burden on health care systems, families and individuals. Cost-utility analysis (CUA) is a type of full economic evaluation preferred by most health technology assessment agencies to prioritise health care interventions in the context of limited resources. CUA compares the effects of interventions, in the form of a summary of health outcomes that incorporates both quantity and quality of life years. CUA is an important approach to evaluate oral heath interventions, as oral diseases have considerable impact on people’s QoL. Hence, the application of CUA in oral health research is worthy of exploration. Two systematic literature reviews were conducted to evaluate the application of CUA in oral health interventions and the paediatric QoL instruments used in oral health research. The first systematic review was performed to evaluate the usage of CUA in oral health interventions, the methods used and the reporting quality of CUA in publications on oral health interventions. This review identified an increasing trend of CUA use in oral health research over time, especially from 2011 to 2016. The majority of CUA publications were of good reporting quality and provided conclusions concerning the most cost-effective intervention among the different options compared; hence, these will assist in healthcare decision-making. However, among 23 CUAs published in dentistry, only four were conducted among paediatric populations and all four were related to dental caries interventions. As children are the main target group of the public health care system, the second systematic review was focused to identify the generic or disease-specific paediatric QoL instruments used in oral health research. This review provided an overview of 11 oral health-specific QoL instruments and five generic instruments used in oral health research among children and adolescents. Of these 11 oral health-specific QoL instruments, none were preference-based QoL measures (PBMs), whereas, two (CHU 9D and EQ-5D-Y) of the five generic instruments were PBMs. The background analysis identified only limited CUAs conducted among paediatric populations on oral health interventions and found no condition-specific PBMs in oral health to be used in CUA in this population. The limited number of CUAs identified in the first review was likely due to the fact that there is no oral health-specific paediatric PBM for use in economic evaluations. PBMs are important to assist estimation of more-accurate utility values for economic evaluations. Existing oral health-related QoL instruments are non-PBM and hence, cannot be used to generate utility values. The background analysis identified the necessity of paediatric utility measures to quantify outcomes in terms of the quality-adjusted life years (QALYs) and to promote the economic evaluation of oral health care interventions using CUA among children and adolescents. Therefore, the main aim of this study was to develop an oral health-specific preference-based QoL measure to facilitate the identification of high-value oral health interventions in adolescent populations. All four paediatric CUAs identified in the first systematic review were all related to interventions to prevent dental caries, the most prevalent childhood oral disease. This indicates that the economic evaluations of paediatric oral health studies were mainly concentrated on dental caries; thus, the availability of a paediatric CSPBM for dental caries would facilitate better evaluation of oral healthcare interventions among paediatric populations. Therefore, dental caries was the focus of the PBM. Preference-based measures consist of two components: a health state classification system and a set of utility weights that enable the generation of utility values for the health states defined by the classification system. The classification system for the PBM was developed as a de novo measure using literature reviews, qualitative interviews with adolescents and expert opinion using the Delphi technique. A systematic literature search of paediatric oral health-related QoL instruments was used to identify possible domains and items to be included in the classification system. Studies eliciting utility values for oral health outcomes and clinical dentistry references were also reviewed. Based on the findings, a draft classification system was developed and refined using semi-structured interviews with a convenience sample of 15 12–17-year-old adolescents who had active caries or previous experience with dental caries. The draft classification system was further refined and validated by a group of dental experts, using a modified Delphi technique. The classification system consists of five items (pain/discomfort, difficulty in eating food/drinking, worried, ability to participate in activities, and appearance) and each item has a four-level severity-based response scale. The resulting preference-based measure is named Dental Caries Utility Index (DCUI). The target group of this instrument was set as children above 12 years of age, considering that mixed dentition ends at 12 years and given the cognitive burden required for younger children to understand the concept and wordings of PBMs. Further, the Flesch Kincaid reading score of the finalised instrument was 64.6, indicating that adolescents aged 12–13 years can easily understand the classification system. The next stage was to generate a utility algorithm for the DCUI classification system. Discrete choice experiments (DCEs) were selected as the preference elicitation technique to value health states generated by the DCUI classification system. DCEs produce utility values on a latent scale; therefore, it is important to anchor them onto a full health-dead scale to calculate QALYs. As there were no previous studies to identify the most suitable method for dental caries health state valuation, two pilot studies were conducted prior to the main survey with two anchoring approaches: DCE with duration (DCETTO) and DCE with visual analogue scale (DCEVAS). The pilot surveys were conducted online among a sample from the Australian general population. Two separate DCE designs were created using Ngene software. The DCETTO design included duration as an additional attribute, whereas DCEVAS included DCE choice tasks with five attributes from the classification system and a separate visual analogue scale (VAS) task for the purpose of anchoring. Conditional logit was used to model the DCE data. Modelled DCETTO coefficients were anchored using the coefficient for the duration. Modelled DCEVAS data were anchored based on two methods: using worst heath state of VAS and mapping DCE onto VAS. A total of 200 participants completed the DCETTO survey and 191 participants completed the DCEVAS survey; there was no statistically significant difference between the two samples in relation to their sociodemographic characteristics. Further, there was no statistically significant difference between the participants’ self-reported difficulty in understanding and completing the valuation tasks between two approaches. The coefficient estimates from the unadjusted DCETTO model showed that the duration coefficient was in the expected direction and significant. Of the 15 coefficients estimated for each level in five dimensions, five were non-significant and one was not in the expected direction (level 2 of ‘difficulty in eating food/drinking’). The coefficients of all dimensions except ‘difficulty in eating food/drinking’ and ‘ability to participate activities’ were ordered as expected (i.e., higher utility decrements were associated with increasing severity levels). Therefore, an adjusted model was estimated by combining both levels 1 and 2 of ‘difficulty in eating food/drinking’ and levels 3 and 4 of ‘ability to participate activities’. All coefficients estimated from this adjusted model were in the expected direction and were logically consistent. Of the 14 coefficients estimated, 10 including the coefficient for duration were significant. Therefore, anchoring of the coefficients onto the full health-dead scale was performed based on the adjusted DCETTO model. In the DCEVAS approach, all coefficients estimated from the unadjusted model were in the expected direction, except that of level 2 of ‘appearance’, and the magnitude of the coefficients increased with the severity level of each dimension. Of the 15 coefficients estimated, four were non-significant. Therefore, an adjusted model was estimated by combining levels 1 and 2 of the dimension ‘appearance’. The anchoring of DCE data onto the full health-dead scale was completed based on adjusted DCEVAS, as all coefficients were in the expected direction, logically consistent and, of the 14 coefficients estimated, 11 were significant. The resultant utility values from these two valuation approaches were compared. Rescaled coefficients from the DCEVAS were in the expected order and significant compared to the rescaled coefficients from the DCETTO approach. Further, DCETTO produced more disperse utility decrements; as a result, the severe health states were valued as worse than death. This is unreliable for a condition like dental caries, in which participants may be reluctant to trade life years to avoid being in a severe dental caries health state. As the pilot data revealed that the DCEVAS model performed optimally and produced more-reliable utility values for dental caries health states compared to the DCETTO approach, the DCEVAS approach was utilised for the main survey. The main valuation survey was conducted as an online survey of an age and sex representative sample of the adult Australian general population. A total of 995 adults completed the survey. The survey included a set of DCE tasks and VAS tasks, basic social-demographic questions, the DCUI, a generic preference-based measure (EQ-5D-5L) and an oral health QoL instrument (OHIP-14). DCE data were modelled using conditional logit. All estimated coefficients from the DCE data were in the expected direction and order for the five dimensions. All coefficients were statistically significant except for the second levels of the ‘worried’ and ‘appearance’ domains. The estimated coefficients were rescaled onto full health-dead scale using two methods: VAS worst heath state and mapping DCE onto the VAS. Both of these methods produce largely similar utility values for the DCUI health states. The mean absolute error value for the DCE estimates based on the mapping approach was lower compared to anchoring based on VAS worst health state. Therefore, the final utility algorithm was generated based on the rescaled coefficients from mapping DCE onto VAS. The Australian-specific tariff of DCUI ranges from 0.1681 to 1.0000. The utility algorithm will enable the calculation of utility values from the participants’ responses for DCUI in economic evaluations. This study has certain limitations. Due to resource constraints and feasibility concerns, the DCUI was developed for dental caries economic evaluations among Australian adolescents. However, there is great potential for the DCUI to be used in adolescents in other countries, as an adult measure or to evaluate oral health conditions other than dental caries. Therefore, future research is recommended in these areas. Further, the DCUI should be validated rigorously with target group adolescents at dental clinic settings. Due to the methodological constraints, health state valuation was conducted using an adult sample, a common method to elicit preferences for paediatric PBMs. Therefore, future studies are recommended to assess whether there is any significant difference between the health state utility values for DCUI derived from an adult sample and the preferences of an adolescent sample. A new health state classification system and utility algorithm completed the new Preference-based QoL measure for dental caries: Dental Caries Utility Index (DCUI). The DCUI will facilitate the assessment of oral health interventions using a CUA framework and will aid resource allocation through economic evaluations for dental caries, the most prevalent childhood disease among Australian adolescents.

Abstract: Cost-utility analysis (CUA) assesses the cost-effectiveness of health technologies by comparing their costs and health outcomes. The utility is incorporated in the health outcome measures of CUA, and the EQ-5D-5L is one of the most common instruments to estimate utility values. When utility values are not available, mapping from non-preference-based instruments to a preference-based instrument is a popular technique. When CUAs use different preference-based measures, mapping between these measures can transfer the utility values and allow for better comparison across CUAs. However, there are many concerns regarding studies reporting mapped utility values, such as extrapolation issues and the uncertainty of this methodology. The quality of mapping studies has become an important criterion when using them in economic evaluations. The first study of my thesis assessed the reporting quality of mapping studies onto the EQ-5D-5L, especially their completeness of information for CUA applications. The second study developed a novel mapping algorithm from the Edmonton Symptom Assessment System Revised: Renal (ESAS-r: Renal) to the EQ-5D-5L among patients with end-stage renal disease (ESRD). The first objective of my systematic review was to identify new mapping studies onto the EQ-5D-5L by updating a previous systematic search made by the Health Economics Research Centre (HERC). The second objective was to assess all the EQ-5D-5L mapping studies on their reporting quality, especially the completeness of information for CUA, with the use of two reporting quality checklists. The third objective was to explore whether using reporting quality checklists was associated with improved reporting quality. The review identified 14 new studies since 2018 which were not included in the HERC database. In the assessment of all 39 published studies (including 25 from the HERC database), the overall reporting quality was mostly good. In several areas I identified problems that would require improvements including 1) estimation of predicted utilities, 2) reporting variances, covariances, and error terms, 3) final model calculation examples, 4) parameter uncertainty, and 5) individual uncertainty. A preliminary comparison showed that the checklists could help to improve the reporting quality of the studies. The second study of this thesis mapped the ESAS-r: Renal to the EQ-5D-5L in patients with ESRD using data from the Evaluation of Routinely Measured Patient-reported Outcomes in Hemodialysis Care (EMPATHY) trial, a multi-centre clustered randomized-controlled trial of routine measurement of patient-reported outcomes in hemodialysis units in Northern Alberta. Several models were explored in the mapping analysis using data from one study arm, including linear models, censored dependent variable models, mixture models and response mapping. Internal validation was conducted to evaluate the predictive power of the models, and the validation sample was from another arm of the EMPATHY trial. Statistical fit and predictive power were measured by mean absolute error (MAE) and mean squared error (MSE), which, along with theoretical backgrounds, were the selection criteria for the best model. The final sample size for model estimation was 506, after excluding missing records (missing rate: 57.6%). All models produced relatively similar statistical fit and predictive power (Estimation: MAE: 0.056 - 0.120, MSE: 0.007 - 0.028; Validation: MAE: 0.136 - 0.155, MSE: 0.032 - 0.046). All models showed great prediction properties for relatively healthy health states, but poor prediction properties for worse health states. With the consideration of all selection criteria, the generalized estimating equations (Estimation: MAE: 0.120, MSE: 0.027; Validation: MAE:0.140, MSE:0.034 ) and generalized linear models (Estimation: MAE: 0.116, MSE: 0.028; Validation: MAE: 0.136, MSE: 0.034) on selected ESAS-r: Renal items were considered the best models. Since the models have not been externally validated, they should be applied in populations with similar patient characteristics as our study sample. Overall, mapping is a feasible and useful technique to estimate the utility values for conducting CUA. The issues identified in current mapping studies could inform further mapping studies on how to improve reporting quality, especially ensuring the completeness of information for employing mapping algorithms in CUA. The empirical mapping study on ESAS-r: Renal provided mapping algorithms which could be used to predict utility values for patients with ESRD when only ESAS-r: Renal is available.

Purpose: Germline mutations in BRCA1 and/or BRCA2 genes (gBRCA1/2m) are associated with an increased risk of breast cancer (BC) and ovarian cancer (OC). The aim of this study was to estimate the efficiency of providing germline BRCA1/2 testing to high-grade epithelial ovarian cancer (HGEOC) patients without family history of OC or BC and the subsequent testing and management of their relatives with gBRCA1/2m in Spain. Methods/patients: Incident HGEOC patients without family history of OC or BC who were gBRCA1/2m carriers and their relatives were simulated in a 50-year time horizon. The study compared two scenarios: BRCA1/2 testing vs no testing, using the perspective of the Spanish National Health Service. Cancer risk among gBRCA1/2m carriers was estimated based on their age and whether they had undergone risk-reducing surgeries. Direct healthcare costs and utilities of patients who developed EOC and BC were also included. A probabilistic sensitivity analysis (PSA) with 5 thousand simulations was developed considering ± 25% of the base-case value. Results: The BRCA1/2-testing scenario amounted to €13,437,897.43 while the no-testing scenario amounted to €12,053,291.17. It was estimated that the screening test improved the quality of life among the patients' relatives by 43.8 quality-adjusted life years (QALYs). The incremental cost-utility ratio (ICUR) was €31,621.33/QALY in the base case. The PSA showed that 89.12% of the simulations were below the €50,000/QALY threshold. Conclusion: Providing this screening test to HGEOC patients and their relatives is cost-effective and it allows one to identify a target population with high risk of cancer to provide effective prevention strategies.

Preventative interventions often demand that resources be consumed in the present in exchange for future benefits. Cost-effectiveness analysis is a tool to understand these trade-offs, and inform decision-making under resource constraints. Targeted testing and treatment (TTT) for latent tuberculosis infection (LTBI) consists in identifying people at high risk for LTBI for preventive treatment to decrease the risk of active tuberculosis disease (ATBD). The state of Tennessee began conducting TTT statewide in 2001. This study uses a decision tree to evaluate the cost and outcomes of TTT for LTBI in Tennessee, compared to passive ATBD case finding (PACF). Key probabilities were obtained from the Tennessee TTT program and the literature. Outcomes are measured in terms of Quality Adjusted Life Years (QALY). The cost-effectiveness threshold was $100,000/QALY saved. One-way sensitivity analyses around factors related to study design, the program’s environment, and program performance were conducted, as was probabilistic sensitivity analysis (PSA) which takes into account the uncertainty in multiple parameters simultaneously. The base case, with a 25-year analytic horizon and 3% discount rate, shows that TTT prevents 47 ATBD cases, and saves 31 QALYs per 100,000 patients screened at a societal cost of $12,579 per QALY saved. Sensitivity analyses identified value thresholds that would trigger a change in preferred policy. PSA shows that the likelihood that TTT would be cost-effective is low. Decision makers should carefully assess the characteristics of the local TB epidemic and expected program performance to determine whether TTT is preferable over PACF from a cost-effectiveness viewpoint.

Background: Only lamivudine has been included for patients with chronic hepatitis B (CHB) in the National List of Essential Drugs (NLED), a pharmaceutical reimbursement list in Thailand. There have also been no economic evaluation studies of CHB drug treatments conducted in Thailand yet. In order to fill this gap in policy research, the objective of this study was to compare the cost-utility of each drug therapy (Figure 1) with palliative care in patients with HBeAg-positive CHB. Methods: A cost-utility analysis using an economic evaluation model was performed to compare each drug treatment for HBeAg-positive CHB patients. A Markov model was used to estimate the relevant costs and health outcomes during a lifetime horizon based on a societal perspective. Direct medical costs, direct non-medical costs, and indirect costs were included, and health outcomes were denoted in life years (LYs) and quality-adjusted life years (QALYs). The results were presented as an incremental cost effectiveness ratio (ICER) in Thai baht (THB) per LY or QALY gained. One-way sensitivity and probabilistic sensitivity analyses were applied to investigate the effects of model parameter uncertainties. Results: The ICER values of providing generic lamivudine with the addition of tenofovir when drug resistance occurred, generic lamivudine with the addition of tenofovir based on the road map guideline, and tenofovir monotherapy were -14,000 (USD -467), -8,000 (USD -267) , and -5,000 (USD -167) THB per QALY gained, respectively. However, when taking into account all parameter uncertainties in the model, providing generic lamivudine with the addition of tenofovir when drug resistance occurred (78% and 75%) and tenofovir monotherapy (18% and 24%) would yield higher probabilities of being cost-effective at the societal willingness to pay thresholds of 100,000 (USD 3,333) and 300,000 (USD 10,000) THB per QALY gained in Thailand, respectively. Conclusions: Based on the policy recommendations from this study, the Thai government decided to include tenofovir into the NLED in addition to generic lamivudine which is already on the list. Moreover, the results have shown that the preferred treatment regimen involves using generic lamivudine as the first-line drug with tenofovir added if drug resistance occurs in HBeAg-positive CHB patients.