Modelling the lifetime health economic impact of scalable behavioural weight loss programmes
Some text in this section has been reproduced with permission from Ahern et al.51 and Bates et al.56 These are Open Access articles distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) licence, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The text below includes minor additions and formatting changes to the original text.
This WP evaluated the long-term health impacts and cost-effectiveness of the scalable weight loss programmes evaluated in WPs 1.1 (WRAP) and 1.2 (GLoW). It also systematically compared two different modelling methods to reduce uncertainty in the cost-effectiveness estimates and increase confidence in these findings.
Rationale
The full impact of weight loss interventions on diabetes and other weight-related illnesses is not realised within the timescale of a RCT. Modelling the lifetime impact of interventions on a range of weight-related diseases and related healthcare costs gives a fuller picture of the public health value for money of these interventions.
To evaluate the lifetime health economic impact of the interventions in WP1.1 and WP1.2, we used the NIHR School for Public Health Research (SPHR) Diabetes Prevention Model (version 4).57,58 This diabetes and cardiovascular health economic model is an individual patient microsimulation model that includes a range of comorbidities related to obesity and T2D. It has been used to inform decision-making about diabetes and obesity interventions by Public Health England, NHS England, the Department of Health and local commissioners, and the outcomes of this WP can inform practitioners and policy-makers about the impact of scalable weight loss programmes on incidence of T2D, related complications and efficient allocation of NHS resources.
To encourage greater confidence in the cost-effectiveness findings, we also undertook comparative modelling to consider the impacts of structural assumptions, risk models and trajectory methods on the results.
Objectives
- To model the long-term cost-effectiveness of the weight loss programmes evaluated in WP1.1 (WW) and estimate their impact on incidence of T2D and related complications in adults at high risk.
- To model the long-term cost-effectiveness of DEW (WP1.2) and its impact on incidence of complications in adults with a recent diagnosis of T2D.
- To assess impact on health inequalities by examining variation in long-term health impact and cost-effectiveness across different demographic groups.
- To compare the performance of two microsimulation models to improve confidence in estimates.
Lifetime health economic impact of a scalable weight loss programme in people at high risk
The analysis of the long-term health economic impact of the 12- and 52-week WW programmes for people at risk of T2D because of overweight or obesity was published alongside the effectiveness and within-trial cost-effectiveness analyses of WRAP (WP1.1) in Ahern et al.51 Additional details of the modelling methods and results are detailed in the appendix to the published paper (pp. 26–101).
Method
We used the NIHR SPHR Diabetes Prevention Model to estimate health outcomes and costs over a lifetime from an NHS and PSS perspective. We identified a representative sample of eligible individuals from the Health Survey for England 2014, updated their health status in annual cycles and calculated major health events, lifetime costs and QALYs. We estimated trajectories of metabolic risk factors for each intervention group (brief intervention and 12- and 52-week WW programmes) and simulated a natural history trajectory for comparison. We used WRAP trial data (WP1.1) to estimate the first 5 years and then assumed a linear return to natural history.
Key findings
Over a lifetime, we estimated that all three interventions (brief intervention and 12- and 52-week programmes) had lower expected lifetime costs and generated QALY gains compared with the do-nothing scenario (Table 2). The 12- and 52-week programmes were cost saving and generated QALY gains compared with the brief intervention. Relative to natural history, the 52-week programme generated greater QALY gains [0.0298 (simulated 95% CI −0.002 to 0.0688)] and lower estimated lifetime costs [−£424 (simulated 95% CI −926 to −51)] than did the 12-week programme [0.0248 (–0.0024 to 0.0599) and −£336 (−792 to 141)]. There was a 65% probability that the 52-week intervention was cost-effective over a lifetime, relative to brief intervention, when judged against the UK NICE threshold of £20,000 per additional QALY. The equivalent value for the 12-week intervention was 30%.
| Incremental analysis (SE) (95% CI) | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Incremental costs | QALYs gained | HbA1c decrease (mmol/mol) | Weight loss (kg) | Incremental cost per QALY gained | Incremental cost per 1 mmol/mol reduction in HbA1c | Incremental cost per 1 kg reduction in weight | iNMB | |||
| £13,000 | £20,000 | £30,000 | ||||||||
| Base casea | −232 (487)(−1177 to 547) | −0.001 (0.01)(−0.02 to 0.03] | 0.84 (1.10)(−2.99 to 1.31) | 1.38 (0.83)(0.19 to 2.56) | 236,314 | DEW dominates DE | DEW dominates DE | 219 (514) (−550 to 1310) | 212 (549) (−679 to 1421) | 202 (619) (−892 to 1622) |
| No adjustmentb | −21 (498)(−928 to 893) | −0.01 (0.02)(−0.05 to 0.03) | 1593 | −151 (592) (−1142 to 1119) | −244 (682) (−1361 to 1255) | −377 (839) (−1801 to 1508) | ||||
| Complete case analysisc | −304 (556)(−1204 to 981) | 0.001 (0.01)(−0.02 to 0.02) | DEW dominates DE | 320 (592) (−1044 to 1285) | 329 (627) (−1088 to 1354) | 342 (692) (−1211 to 1493) | ||||
| Complete case analysis, no adjustmentd | −99 (554)(−1089 to 1109) | −0.01 (0.03)(−0.06 to 0.07) | 12,476 | −4 (755) (−1445 to 1521) | −60 (928) (−1746 to 1885) | −140 (1208) (−2219 to 2487) | ||||
Probabilistic sensitivity analysis (PSA) and subgroup analyses found that the net benefits of the intervention were broadly similar across deprivation quintiles and showed comparable results. All trial arms produce greater cost savings in groups at higher risk of complications, such as a population with T2D or those with very high BMI. Including the 5-year data from WP1.1 resulted in more favourable cost-effectiveness than an analysis that only considered the 2-year data.
Lifetime health economic impact of a scalable weight loss programme in people with a recent diagnosis of type 2 diabetes
The analysis of the long-term health economic impact of DEW in adults with a new diagnosis of T2D was published alongside the main trial analysis (WP1.2) in Mueller et al.54
Methods
To evaluate the lifetime cost-effectiveness of interventions in the GLoW trial, we adapted the NIHR SPHR Diabetes Prevention Model for a population with T2D. We generated a synthetic baseline population using participant characteristics from the GLoW trial (WP1.2) and data from the Health Improvement Network. Metabolic risk factor trajectories and diabetes-related outcomes were generated using modified UKPDS risk equations (adjusted to address overprediction of some outcomes). Healthcare costs, health-related quality-of-life decrements and QALYs were assigned to complications of diabetes.
In our base-case analysis, the eligible population received their intervention (DE or DEW) face-to-face or online, with proportions based on current care locally. To address uncertainties in model parameters, we conducted PSA to examine the impact of: (1) DE being entirely face-to-face or entirely online (as this had a substantial impact on costs), (2) diabetes duration, (3) BMI categories and (4) IMD.
Key findings
Full findings are described in Mueller et al.54 Table 3 shows the lifetime discounted NHS and PSS costs, discounted QALYs, and cost-effectiveness estimates for different scenarios, which make different assumptions about how DE is delivered. In our baseline scenario, DE is delivered via a mix of face-to-face and online services, reflecting local practice. DEW was estimated to be cost-effective, with an incremental cost-effectiveness ratio (ICER) of £2273. PSA found that DEW had a 97% probability of being cost-effective at a threshold of £20,000 per QALY.
| Number of PSA samples for uncertainty analysis = 2000 | Total discounted NHS costs per person (£) | Total discounted QALYs per person | Incremental costs (£) | Incremental QALYs | Incremental expected NMB (£): £20,000 threshold | ICER |
|---|---|---|---|---|---|---|
| Mixed F2F and online delivery | ||||||
| DE | 40,886 | 8.7023 | ||||
| DEW | 40,966 | 8.7376 | 80 | 0.0353 | 625 | 2273 |
| F2F only service delivery | ||||||
| DE | 40,994 | 8.7023 | ||||
| DEW | 40,966 | 8.7376 | −28 | 0.0353 | 733 | Dominant |
| Online only service delivery | ||||||
| DE | 40,739 | 8.7023 | ||||
| DEW | 40,966 | 8.7376 | 227 | 0.0353 | 478 | 6438 |
| Cost of DESMOND assuming optimistic uptake and full capacity for F2F (per person DESMOND cost £95) | ||||||
| DE | 40,822 | 8.7023 | ||||
| DEW | 40,966 | 8.7376 | 145 | 0.0353 | 5601 | 4097 |
Diabetes education and behavioural weight management programme was still estimated to be cost-effective when we assume that DE is only delivered as a (cheaper) fully online service and is dominant (lower costs and higher QALY gains) if we assume DE is only delivered in person. The finding that DEW is cost-effective was robust to different assumptions explored in subgroup and sensitivity analyses, including diabetes duration, BMI category and IMD quintile.
Stratified analysis by IMD did not identify substantial differences in the long-term health and NHS cost saving by socioeconomic status. The analysis indicated that the GLoW intervention was slightly more cost-effective for the most deprived group, compared with the least deprived group.
Model comparison to reduce uncertainties
To increase confidence in our cost-effectiveness estimates, we compared two microsimulation models to improve confidence. This analysis is reported in detail in Bates et al.56
Method
To investigate the impact of structural assumptions, risk models and trajectory methods on health economic outcomes, we compared outcomes for the same decision problem produced by two models: the SPHR Diabetes Prevention Model and the Health Check (HC) model. The HC model was developed to evaluate the NHS Health Check programme59 and was chosen because the model code is open source, enabling detailed understanding of the structural similarities and differences, and because developers of the model agreed to collaborate on the protocol for cross validation.
We qualitatively compared the structural differences between the two models using publications, technical descriptions and the model code. The model set-up was standardised across the two models, with adaptations made to the HC model (independently then combined) to align (1) the baseline population, (2) discount rate and (3) baseline utility to that of the SPHR model.
The decision problem was the health economic outcomes of the WRAP trial described above, comparing the three weight management interventions (WW for 12 weeks, WW for 52 weeks and a brief intervention) and a simulated no-intervention scenario.
To understand how differences in the model structure drive differences in the model outcomes, we ran repeated simulations making iterative changes to make the models more similar and compared the outcomes. We explored four scenarios:
- Remove impact of treatment on HbA1c from SPHR (the HC model did not include HbA1C trajectories).
- Run the models with only the health conditions that are included in both models (diabetes, dementia and cardiovascular disease)
- Use The English Longitudinal Study of Aging data for trajectories in the SPHR model.
- Apply structural changes A, B and C together.
Key findings
Changes made to the HC model to standardise: (1) the baseline population, (2) discount rate and (3) baseline utility were found to significantly impact model outcomes, with discounting having the largest effect on the total QALYs. Figure 3 shows the incremental QALYs between each treatment (brief intervention, 12-week intervention and 52-week intervention) and simulated no-intervention scenario for each model (HC Model and SPHR: School for Public Health diabetes prevention model) for the main analysis and for the four scenarios.
FIGURE 3 Lifetime incremental QALYs in the WRAP trial. Reproduced with permission from Bates et al.56 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) licence, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The figure above includes minor additions and formatting changes to the original text. (Figure not reproduced here; see the original article.)
Estimates of long-term effectiveness of the weight management programme using the SPHR and HC models were more similar after adjusting for explainable differences, namely, removing effects of weight management on HbA1c, limiting the models to a common set of health outcomes and using a common data set to derive metabolic trajectories. Since the SPHR model includes stronger evidence on the effect of weight management on HbA1c, the results of this cross-model validation support the existing cost-effectiveness analysis and provide additional confidence in those estimates.
Strengths and limitations
Cross-model comparisons are not routinely performed but provide an important aspect of model validation that is important to ensure models are robust. Open-source access enabled in-depth model understanding of the aspects of the model structure that drive differences in outcomes. Structural differences between models posed challenges for standardisation. Determining relevant outcomes for comparison was difficult. Defining an acceptable difference magnitude was hindered by the absence of cost estimates. Resource constraints limited the extent of iterative adjustments. However, the validation process offered insights beyond standard methods of validation and has informed future model development by highlighting key areas for improvement.
Conclusions
Health economic modelling suggests that scalable BWMPs are likely to be cost-effective from an NHS and PSS perspective in both people at high risk of T2D and in people who have had a diagnosis of T2D in the past 3 years. This is because of the impacts of weight loss on weight-related diseases, even when weight lost is regained over time.
Incorporating observed 5-year weight trajectories in WP1.1 improved estimates of cost-effectiveness because observed weight trajectories demonstrated better weight loss maintenance than previously assumed. This demonstrates the importance of trajectories of weight regain in determining cost-effectiveness and this challenge is addressed in WP2. Cross-model comparison analysis and the robustness of these findings to different sensitivity analyses support the validity of these cost-effectiveness estimates. In all evaluations the cost-effectiveness estimates were similar across IMD quintiles, indicating that the long-term benefits are equally distributed across socioeconomic groups.
Interrelation with other work packages
This WP used data from the WRAP trial (WP1.1) and the GLoW trial (WP1.2) to estimate intervention effects used in the model. In addition, statistical analyses of the WRAP and GLoW trial data sets were used to estimate key inputs to the SPHR Diabetes Prevention Model.60,61
The SPHR Diabetes Prevention Model used in this WP was also used in WP2.3 to estimate the long-term impact of a novel weight loss maintenance programme and to inform decision-making about the value and optimal design of a definitive RCT.