Health economics

Introduction

The aim of this health economic evaluation is to assess whether the treatment of patients with TRBD with pramipexole in addition to mood stabilisers (treatment) is cost-effective compared to placebo (control). For this purpose, the following section presents a detailed within-trial outcome, cost and incremental cost-effectiveness analysis of the PAX-BD randomised, double-blind, placebo-controlled trial conducted in the UK between 2018 and 2022. Most of the trial was conducted during the COVID-19 pandemic facing major challenges and had to be terminated early due to funding reasons. Therefore, care contacts and cost information are representative of this given time period and the sample size available for the economic evaluation is much more limited than originally foreseen.

Methods

The health economic analysis plan was published in the PAX-BD trial protocol paper,57 and is reported following the Consolidated Health Economic Evaluation Reporting Standards guidelines.88

Study population

All base-case economic analyses focused on the same ITT population with 12-week outcomes (n = 36) as used for the main efficacy analysis. Further sensitivity analyses were conducted on a per-protocol health economic (PP-HE) sample including those participants who had at least moderate-level depression at treatment initiation (baseline) and had some follow-up health economic data beyond baseline, and excluding one participant in the placebo arm whose 12-week QIDS-SR score was completed outside of the compliance window (note that this participant was NOT excluded from the per protocol population used in the sensitivity analysis in Chapter 3). For a detailed description of the samples analysed, see Appendix 4, Tables 68 and 69.

Economic evaluation methods

The economic evaluation compared the within-trial patient-level costs, outcomes and incremental cost-effectiveness between the pramipexole and the placebo arms using current gold standard methods.89 Relevant outcome and resource use data were collected electronically through the TrueColours clinical platform,90 at baseline and at weeks 12, 24, 36 and 48.

As per the NICE health technology evaluation manual (NICE 2022), the economic evaluation was conducted primarily from the NHS and PSS perspective. A secondary evaluation from a broader societal perspective included also lost productivity and informal care costs since these were shown as major cost components in the latest cost-of-illness trial for bipolar disorder in the UK.91 Due to the limited sample size, results are reported not only for the full 48 weeks trial follow-up, but also for the first 12 weeks trial period matching the primary clinical end point. Since the time horizon of the analyses did not exceed 1 year, no discounting of either the costs or the outcomes results was necessary.

Outcome measurement and valuation

The base-case health economic analysis is a cost–utility analysis with quality-adjusted life-years (QALYs) gained and calculated based on utility values developed from HRQoL information collected via the EQ-5D-5L questionnaire as recommended by most health technology institutes.89,92,93 UK-specific health state utilities were calculated using the ‘EQ-5D-5L Crosswalk Index Value Calculator’ tool provided by the EuroQol Group as currently required by NICE.92 Outcome results for the EQ visual analogue scale (EQ VAS) are also reported.

Due to known limitations of the EuroQol-5 Dimensions (EQ-5D’s) sensitivity for capturing broader quality-of-life benefits in mental health,82 secondary cost-effectiveness analyses using patient-reported capability well-being measurement tools previously validated within the mental health context were also carried out.94,95 For this purpose, we measured broader capability well-being based on the generic ICECAP-A questionnaire valued by its UK tariff set,96,97 and based on the more mental health-specific OxCAP-MH questionnaire using its standardised scores,81,82 and calculated capability-weighted life-years (CWLYs) gained.91

Costing methods

In terms of treatment costs, pramipexole was prescribed electronically and posted out to participants. The prescription of the trial drug hence did not require regular direct specialist contact. Instead, participants were regularly contacted and followed up by trial researchers with additional psychiatrist supervision and information review. Treatment costs in the current economic evaluation reflect the given care provision in the PAX-BD trial and may not be fully representative outside of the trial setting.

The cost analysis used further resource use data collected with an adapted version of the HEQ.98,99 The HEQ’s development was based on previous versions of the Client Service Receipt Inventory instrument100 and the Trimbos/iMTA questionnaire for Costs associated with Psychiatric illness (TiC-P).101 Collected resource use data included all hospital and community health and social care services, medication, productivity losses and informal care. For medication costs, only mental health-relevant prescription medications were considered and categorised as antipsychotics, mood stabilisers, antidepressants, hypnotics or other mental health-related medications. Additional cost categories for the broader societal perspective included the costs of informal care provided by family members or friends as well as costs due to absenteeism from work for participants who were employed or self-employed.

Costing was conducted using UK national-level unit costs for a common costing year of 2020–1 as reported in Table 66 in Appendix 4. Lost productivity costs were estimated using the human capital approach where time off work was multiplied by the average daily national salary for participants who were employed or self-employed.102,103 All costs are expressed in Great British pounds (GBP) for the financial year 2020–1. Unit costs referring to earlier years were inflated using the NHS Cost Inflation Index (NHSCII) as reported in the Personal Social Services Research Unit (PSSRU) Cost of Health and Social Care 2021 report.104

Data processing

All available outcome and resource use data underwent thorough plausibility checks. Furthermore, three forms of missing data were encountered and underwent imputation. Firstly, participants did not always submit their self-reported outcome information at the planned follow-up time point within the compliance period of ± 3 weeks. When reported outcome data were available outside this window, we used linear extrapolation between the previous and the given data reporting time point and imputed missing outcome data points according. For the calculation of QALYs or CWLYs gained, all available and extrapolated outcome data points were taken into consideration. Secondly, participants may have submitted incomplete HEQ data where key information was based on more than one question. For instance, the frequency of visits was reported but not their length. In these cases, missing length information was derived from the mean of the available observations in the relevant trial arm. All other missing data were imputed using the gold standard method of multivariate imputation via chained equations based on the missing at random assumption.105,106 Missing values were filled in based on predictive mean matching adjusted for age, sex and trial arm using three donors and repeated for M = 24 sets of imputed values, where M as a rule-of-thumb corresponded to the percentage of missing values in the data set.

Cost-effectiveness analysis

For the economic analyses, outcomes were adjusted to their baseline values and reported as difference-to-baseline to avoid any potential bias due to the considerable observed difference in the mean baseline EQ-5D-5L, ICECAP-A and OxCAP-MH levels between the pramipexole and placebo arms. Difference in HRQoL and capability well-being gains between the groups over time was analysed using multiple regression analysis with age, sex and trial arm as explanatory variables. For the calculation of QALYs and CWLYs gained, HRQoL and capability well-being states were assumed to change linearly between the two data collection points.

Cost-effectiveness results were expressed as incremental cost-effectiveness ratios (ICERs) which correspond to the difference in costs divided by the difference in the given outcome between the pramipexole and the placebo arms, and are reported also as net monetary benefits (NMBs) and net health benefits (NHBs) for the base-case analysis at threshold values of GBP20,000/QALY gained and GBP30,000/QALY gained.

Non-parametric bootstrapping was used to generate a joint distribution of the mean incremental costs and effects through 1000 simulations based on random draws.107 The bootstrapping approach allows calculating the 95% uncertainty intervals of the ICERs to represent uncertainty around the point estimate and controls for data outliers. As per probabilistic medical decision-making techniques, uncertainty around the main cost-effectiveness estimates is also presented on the cost-effectiveness plane and by cost-effectiveness acceptability curves (CEACs).108,109 The cost-effectiveness plane has four quadrants, each representing a different decision-making scenario. Bootstrapped ICERs falling in the SE quadrant show pramipexole being more effective and less costly than placebo (dominant), falling in the NW quadrant show pramipexole being less effective and more costly than placebo (dominated), falling in the SW quadrant show pramipexole being less effective and less costly than placebo, and falling in the NE quadrant show pramipexole being more effective and more costly. CEACs are based on the net benefit principle and show the probability that pramipexole is cost-effective at a range of maximum threshold values (ceiling ratio) that a decision-maker might be willing to pay for an additional unit of improvement in outcomes.

For all statistical analyses, a two-tailed p-value of < 0.05 was used as the threshold for statistical significance. To compare trial arm means, paired two-tailed t-tests were applied. All analyses were carried out in Excel (Microsoft Office 2016, Microsoft Corporation, Redmond, WA, USA) and STATA version 15 (2017. StataCorp LLC, TX, USA).

Sensitivity analyses

We investigated the potential impact of linear extrapolation in HRQoL and capability well-being between health states by assuming the alternative that these changes happen at the beginning of each given time period in a sensitivity analysis. Further sensitivity analyses in terms of clinical patient characteristics were carried out for the PP-HE sample that excluded three participants whose depression severity was below 11 on the QIDS scale at treatment initiation (baseline), and one patient who did not provide any health economic data beyond the baseline. All methods and scenarios otherwise mirrored the main health economic evaluation. Due to the low number of trial participants who provided complete data (no single missing data point) over the entire 48-week follow-up (a total of 15 participants), a complete case analysis was not carried out.

Results

Study population

The demographic and main clinical characteristics of patients included in the ITT sample at baseline (total: n = 36, pramipexole: n = 16, placebo: n = 20) are shown in Table 28 and Table 29 in Appendix 3. Socioeconomic characteristics are described in Table 67 in Appendix 4. The same baseline patient characteristics of the PP-HE sample (total: n = 32, pramipexole: n = 15, placebo: n = 17) are provided in Table 68 in Appendix 4. As described in Chapter 3, there were no significant differences between the arms except in the case of depression severity being lower in the pramipexole arm.

Missing data

The level of missingness for all used outcome and resource use data is reported in Appendix 4, Table 69. Overall, missingness was below 5% at 12 weeks and around 25% of all planned health economics-related data points over 48 weeks including missingness as a result of withdrawal due to early closure of the trial.

Outcome results

The two arms differed considerably (but not significantly) in all health economic outcome measures at baseline with participants in the pramipexole arm having on average higher HRQoL and capability well-being outcomes at treatment initiation (see Appendix 4, Table 70). Observed HRQoL and capability well-being (EQ-5D-5L, EQ VAS, ICECAP-A, OxCAP-MH) all improved for both arms at 12 weeks and over the entire 48 weeks follow-up when taking the mean over all available data collection points at 12, 24, 36 and 48 weeks for the available cases into consideration (Figure 15). None of the observed changes at 12 weeks relative to baseline were statistically significant at the 95% confidence level due to the comparatively large confidence bounds corresponding to the small sample size. However, average improvement was more dominant in the pramipexole arm for the later follow-up stages, resulting in statistically significantly higher incremental HRQoL (EQ-5D-5L index) gain (+ 0.07, 95% CI 0.01 to 0.13) in comparison to placebo over 48 weeks as shown in the multiple regression analysis adjusted for age, sex and baseline values (Table 14). The secondary health economic outcomes showed a similar significant added benefit of pramipexole over 48 weeks for the ICECAP-A index (+ 0.08, 95% CI 0.02 to 0.15) (Table 15) and for the OxCAP-MH score (+ 4.03, 95% CI 0.24 to 7.81) (Table 16).

FIGURE 15 Health economic outcome results. (a) EQ-5D-5L (NW); (b) ICECAP-A (NE); (c) OxCAP-MH (SW); and (d) EQ VAS (SE) for the comparative ITT sample. (Figure not reproduced here; see the original article.)

Cost results

Observed level of resource use

Resource utilisation is shown in Appendix 4, Table 71, while Appendix 4, Table 72 shows the main cost results. The two arms did not differ significantly in their baseline costs measured over 4 weeks prior to treatment initiation. Other than in the case of direct treatment costs (£20.00, 95% CI £17.47 to £22.53), the two arms did not differ in any of the measured cost categories significantly during the trial period corresponding to the small sample size either. When looking at total health and social services costs, the pramipexole arm in comparison to the placebo arm was on average cost saving both over 12 weeks (−£52.48, 95% CI −£391.64 to £286.67) and over 48 weeks (−£798.65, 95% CI −£1882.17 to £284.87), although the differences were not significant. From the societal perspective, the pramipexole arm was on average more costly both over 12 weeks (£794.73, 95% CI −£2698.82 to £4288.29) and over 48 weeks (£1170.73, 95% CI −£8941.89 to £11,283.35), although the differences were not significant.

Cost-effectiveness results: base-case analysis (QALY gained)

Mean incremental costs, QALYs and cost–utility results are reported in Table 17 both over 12 weeks and over 48 weeks. From the NHS + PSS perspective, we found pramipexole on average more effective and cost saving than placebo for both time periods. From the societal perspective, pramipexole was shown being on average more effective and more expensive. Uncertainty around the mean ICERs was large with statistically non-conclusive 95% CIs in all cases.

Uncertainty in the cost-effectiveness results is also presented probabilistically via the cost-effectiveness plane and CEAC based on bootstrapped results from the NHS + PSS perspective in Figure 16 and from the societal perspective in Figure 17 for both time periods.

FIGURE 16 Uncertainty in the cost-effectiveness results (NHS + PSS perspective). Note: cost-effectiveness plane with bootstrapped ICERs for pramipexole vs. placebo treatment (left: 12 weeks, right: 48 weeks); CEAC showing the probability of pramipexole being cost-effective in comparison to placebo treatment at different willingness-to-pay thresholds for QALY gained (left: 12 weeks, right: 48 weeks). (Figure not reproduced here; see the original article.)

FIGURE 17 Uncertainty in the cost-effectiveness results (societal perspective) over 12 weeks and 48 weeks follow-up for the comparative ITT sample. Note: cost-effectiveness plane with bootstrapped ICERs for pramipexole vs. placebo treatment (left: 12 weeks, right: 48 weeks); CEAC showing the probability of pramipexole being cost-effective in comparison to placebo treatment at different willingness-to-pay thresholds for QALY gained (left: 12 weeks, right: 48 weeks). (Figure not reproduced here; see the original article.)

In Figure 16, we can see that from the NHS + PSS perspective bootstrapped ICERs mainly fall within the SE quadrant of the cost-effectiveness plane for both time periods resulting in an around 90% probability of pramipexole being cost-effective at £30,000/QALY gained.

In Figure 17, we can see that from the societal perspective most bootstrapped ICERs fall within the NW quadrant of the cost-effectiveness plane and the probability of pramipexole remains under 50% at £30,000/QALY gained.

The NMB at 12 weeks (NHS + PSS perspective) was £165.91 (95% CI −£177.08 to £536.21) assuming a willingness-to-pay (WTP) threshold of £20,000, and £222.63 (95% CI −£205.54 to £672.53) assuming a WTP threshold of £30,000. The corresponding NHB was 0.01 QALY (95% CI −0.01 QALY to 0.03 QALY) and 0.01 QALY (95% CI −0.01 QALY to 0.02 QALY), respectively. In line with the findings presented above, the NMB and NHB for pramipexole improve when considering the longer follow-up period of 48 weeks, for example, with an NMB of £1448.15 (95% CI −£295.39 to £3057.81) at £20,000 and £1772.90 (95% CI −£542.81 to £3964.20) at £30,000, respectively. Appendix 4, Tables 73 and 74 provide summaries of the NMB and NHB results for the main analyses of the ITT and PP-HE patient samples.

Cost-effectiveness results: secondary outcome analyses (CWLYs gained)

The secondary cost-effectiveness analyses used CWLYs gained based on the ICECAP-A index and the OxCAP-MH score as alternative outcome measures.

ICEpop CAPability measure for Adults

The secondary cost-effectiveness analysis results based on CWLY gained using the ICECAP-A are shown in Table 18, Figures 18 and 19. In line with the health outcome analysis results, the probability of pramipexole being cost-effective is somewhat lower than in the base-case analysis, but still around 70% over 48 weeks from the NHS + PSS perspective.

FIGURE 18 Uncertainty in the cost-effectiveness results (NHS + PSS perspective) over 12 and 48 weeks follow-up for the comparative ITT sample based on the secondary ICECAP-A outcome measure. Note: cost-effectiveness plane with bootstrapped ICERs for pramipexole vs. placebo treatment (left: 12 weeks, right: 48 weeks); CEAC showing the probability of pramipexole being cost-effective in comparison to placebo treatment at different WTP thresholds for CWLY gained (left: 12 weeks, right: 48 weeks). (Figure not reproduced here; see the original article.)

FIGURE 19 Uncertainty in the cost-effectiveness results (societal perspective) over 12 and 48 weeks follow-up for the comparative ITT sample based on the secondary ICECAP-A outcome measure. Note: cost-effectiveness plane with bootstrapped ICERs for pramipexole vs. placebo treatment (left: 12 weeks, right: 48 weeks); CEAC showing the probability of pramipexole being cost-effective in comparison to placebo treatment at different WTP thresholds for CWLY gained (left: 12 weeks, right: 48 weeks). (Figure not reproduced here; see the original article.)

Oxford CAPabilities questionnaire-Mental Health

The secondary cost-effectiveness analysis results based on CWLY gained using the OxCAP-MH are shown in Table 19, Figures 20 and 21. In line with the primary health outcome analysis results, we found the probability of pramipexole being cost-effective about the same as in the base-case analysis, around 90%, over 48 weeks from the NHS + PSS perspective.

FIGURE 20 Uncertainty in the cost-effectiveness results (NHS + PSS perspective) over 12 and 48 weeks follow-up for the comparative ITT sample based on the secondary OxCAP-MH outcome measure. Note: cost-effectiveness plane with bootstrapped ICERs for pramipexole vs. placebo treatment (left: 12 weeks, right: 48 weeks); CEAC showing the probability of pramipexole being cost-effective in comparison to placebo treatment at different WTP thresholds for CWLY gained (left: 12 weeks, right: 48 weeks). (Figure not reproduced here; see the original article.)

FIGURE 21 Uncertainty in the cost-effectiveness results (societal perspective) over 12 and 48 weeks follow-up for the comparative ITT sample based on the secondary OxCAP-MH outcome measure. Note: cost-effectiveness plane with bootstrapped ICERs for pramipexole against placebo treatment (NW: 12 weeks, NE: 48 weeks); CEAC showing the probability of pramipexole being cost-effective in comparison to placebo treatment at different WTP thresholds for CWLY gained (SW: 12 weeks, SE: 48 weeks). (Figure not reproduced here; see the original article.)

Sensitivity analyses

Per-protocol health economics sample

Including in the economic evaluation only participants who fulfilled the ‘per protocol’ criteria had considerable impact on the cost-effectiveness results. In the PP-HE analysis, pramipexole was shown being significantly more effective than placebo in terms of QALYs gained over 48 weeks with the probability of being cost-effective from the NHS + PSS perspective increasing close to 100%. Relevant results are reported in Table 20, Figures 22 and 23. Separate trial population, outcome and cost results for the PP-HE analysis are shown in Appendix 4, Figure 32, Tables 68, 75 and 76.

FIGURE 22 Uncertainty in the cost-effectiveness results (NHS + PSS perspective) over 12 and 48 weeks follow-up for the PP-HE sample. Note: cost-effectiveness plane with bootstrapped ICERs for pramipexole vs. placebo treatment (left: 12 weeks, right: 48 weeks); CEAC showing the probability of pramipexole being cost-effective in comparison to placebo treatment at different WTP thresholds for QALY gained (left: 12 weeks, right: 48 weeks). (Figure not reproduced here; see the original article.)

FIGURE 23 Uncertainty in the cost-effectiveness results (societal perspective) over 12 and 48 weeks follow-up for the PP-HE sample. Note: cost-effectiveness plane with bootstrapped ICERs for pramipexole vs. placebo treatment (left: 12 weeks, right: 48 weeks); CEAC showing the probability of pramipexole being cost-effective in comparison to placebo treatment at different WTP thresholds for QALY gained (left: 12 weeks, right: 48 weeks). (Figure not reproduced here; see the original article.)

Alternative quality-adjusted life-year calculation (change-at-beginning)

Assuming changes in health status at the beginning of each time period rather than linearly between time points had no impact on the results or conclusions, the probability of pramipexole being cost-effective remained around 90% over 48 weeks from the NHS + PSS perspective. Relevant results are reported in Appendix 4, Table 77, Figures 33 and 34.

Discussion

This health economic evaluation assessed whether pramipexole treatment of patients with treatment-resistant BD in addition to mood stabilisers (treatment) is cost-effective compared to placebo (control) as administered within the PAX-BD trial.

Since the PAX-BD trial was terminated early, the health economic evaluation suffered from the same limitation in terms of sample size as the main clinical analysis. Furthermore, the two arms of the trial had some imbalance in their health economic outcomes at baseline which we had to adjust for in our analyses. Addressing these limitations statistically, we found pramipexole being significantly more effective in terms of HRQoL and capability well-being over 48 weeks, but not over 12 weeks, which corresponds to the observed pattern of increasing pramipexole benefits over a longer duration. The additional health outcome gain over 48 weeks ranged between 4% and 8% depending on the outcome measure used. Cost-effectiveness analyses based on QALYs and CWLYs showed the probability of pramipexole being cost-effective in comparison to placebo over 48 weeks from the health and social care perspective being around 70–90%. In terms of the base-case analysis (QALY, NHS + PSS perspective), we found that almost 80% of the bootstrapped ICER estimates indicated pramipexole being more effective and less costly than placebo over 48 weeks (Table 21).

Since most of the PAX-BD trial was conducted during the COVID-19 pandemic, resource use and (treatment) cost results are indicative of the type of (mental) health care provided at that time and cannot be generalised to bipolar treatment costs outside the given context. On the other hand, these conditions equally impacted both arms of the trial. Since the economic evaluation looked at incremental effects and costs between the two trial arms, it is unlikely that any COVID-19-related resource use patterns impacted the incremental cost-effectiveness results from the NHS + PSS perspective significantly. This may not be true for broader societal costs such as lost productivity and informal care which could have been more significantly influenced by patients’ individual circumstances during the COVID-19 pandemic. We also observed large differences between the two arms in terms of the number of patients reporting informal care support or employment (see Appendix 4, Table 71). Therefore, given the small sample size and the added uncertainty, cost and cost-effectiveness estimates from the societal perspective should be treated with extreme caution. We nevertheless report these estimates as outlined in our original analysis plan.57

Besides the above limitations, the economic evaluation has multiple strengths. Health economic analyses were carried out fully blinded to group allocation and independently of the main clinical statistical analysis. Furthermore, a respectively low total of 25% missing data points across all variables needed imputation. In addition, we conducted cost-effectiveness analyses not only with QALYs based on the EQ-5D-5L but also using CWLYs based on two validated capability well-being measures (the generic ICECAP-A and the more mental-health-specific OxCAP-MH) and found the same outcome results and cost-effectiveness conclusions with all three outcome measures.

Furthermore, the conclusions were insensitive to the outcome extrapolation methods used. Including only patients with at least moderate depression at baseline (PP-HE sample) increased the probability of cost-effectiveness even further.

In summary, our results based on the PAX-BD trial suggest added benefits of pramipexole in comparison to placebo in terms of HRQoL and capability well-being and suggest a tendency towards reduced health and social care costs with a high probability of overall cost-effectiveness over 48 weeks. Ideally, these findings would have to be confirmed in a larger sample size.