When available for all relevant comparators, use high-quality directly measured outcomes (for example, atherosclerotic cardiovascular disease [ASCVD] events or incidence of type 2 diabetes mellitus [T2DM]) to model relative treatment effects, including trends in treatment effects over time. (required)
Absolute effects should be modelled by multiplying the relative treatment effect by a baseline event rate. Baseline event rates should be estimated using real-world evidence when available. (recommended)
Baseline event rates should be specific to strata and health state and relevant to an NHS population in England, ideally also by age and sex subgroup. (recommended)
When directly measured health outcomes are unavailable or cannot be applied consistently across all relevant comparators, model the effectiveness of interventions through changes in weight and other risk factors, used as inputs to validated risk prediction tools or risk equations. (required).
When risk-factor-based modelling is used to predict changes in the incidence of ASCVD and T2DM, treatment-related changes in the following risk factors should be considered:
weight or body mass index (BMI) (recommended)
systolic blood pressure (recommended)
HbA1c (recommended)
cholesterol (high-density lipoprotein and low-density lipoprotein). (recommended)
Baseline estimates of risk factors (including weight) should be based on real world evidence and be specific to stratum and health state and relevant to an NHS population in England. Treatment impact on risk factors should be based on evidence from trials. (recommended)
Justify the selection of validated risk prediction tools or risk equations to estimate modelled outcomes and events. (required)
Risk prediction tools applicable to the stratum or subgroup and to the NHS context in England should be used. (recommended)
When using multi‑year risk estimates (such as 10‑year risks) convert them into annual probabilities precisely to prevent overestimation. (required)
When data for high-quality directly measured health outcomes is available for some (but not all) comparators, use it to validate, and when necessary, calibrate, outcomes predicted using risk equations or risk prediction tools. (required)
Even when direct evidence is used to measure health outcomes, weight (baseline and treatment effect) should be modelled over time. This is for the adjustment of utilities based on weight. (recommended)
For the rationale for these statements, see the rationale and supporting information section on treatment effects.
Predictions of treatment effects beyond trial follow-up need to be biologically and clinically plausible, for example, by taking into account data on the natural history of weight change, and long-term adherence to treatment. (required)
Long-term weight trajectory assumptions across behavioural intervention, pharmacological intervention (allowing variation by mechanism of action), minimally invasive and surgical bariatric procedures should be differentiated to reflect their distinct clinical benefits. This may be informed by real-world longitudinal data. (recommended)
When extrapolation beyond the data observation period is needed, incorporate the following scenario analyses to explore uncertainty in weight trajectory over time while on treatment, including:
applying the same natural history weight gain to both interventions and comparators (required)
assuming no natural history weight gain in interventions and comparators (required)
exploring partial waning of treatment effect relative to the initial magnitude of treatment benefit on weight loss over time. (required)
When weight is not used to predict health outcomes, incorporate scenarios around intervention treatment effect waning in a sensitivity analysis for these outcomes. (required)
For the rationale for these statements, see the rationale and supporting information section on treatment effects in the longer term.
Include treatment duration or discontinuation and its effect on resource use in the model for interventions and comparators, including behavioural intervention only comparators. (required)
Beyond trial follow-up, reduce relative treatment effect to reflect discontinuation or dose reduction, as treatment effects should apply only to people continuing with the (full-dose) treatment. (required)
The heterogeneity across people in post-treatment weight regain should be incorporated to capture the long-term health outcomes adequately. (recommended)
When medicines have been discontinued, do not include NHS resource use associated with concomitant behavioural interventions unless there is evidence that behavioural interventions without medicine have continued in clinical practice in England. (required)
Treatment discontinuation should be modelled over time as follows:
rate of discontinuation over time should be specific to the population stratum and treatment and relevant to an NHS population in England (recommended)
baseline discontinuation rates should be sourced from real-world evidence and relative differences in discontinuation rates between interventions from trial data. (recommended)
When evaluating interventions that have a waning of the treatment effect, threshold analyses should be included to identify optimal treatment durations. (recommended)
Account for weight regain and subsequent impact on health states and events after treatment discontinuation using evidence from long-term studies when available:
when long-term studies for the intervention are not available, use evidence on weight regain for other treatments with a similar mode of action and efficacy or clinical expert opinion (required)
use rate of weight gain over time that is specific to the population stratum and treatment and relevant to an NHS population in England (required)
include scenario analyses informed by clinical opinion to test the impact of different weight regain rates or duration of regain, for example, return to initial baseline weight after a specific number of years of discontinuation (required)
include a scenario where weight returns to a higher level than the underlying natural trajectory of weight gain on no treatment (required)
include a scenario where weight returns to pre-treatment weight in a shorter timeframe than in the base case analysis. (required)
For the rationale for these statements, see the rationale and supporting information section on treatment discontinuation.
Use a consistent mortality modelling approach within the model structure; ideally mortality rates should be specific to both the health state and the body mass index (BMI) category. If specific data is not available, then the model needs to either apply BMI-adjusted all-cause mortality ratios or condition-specific mortality ratios (not both) to avoid the risk of double counting. (required)
When possible, mortality rates (cardiovascular mortality and non-cardiovascular mortality) should be specific to the population stratum and health state and relevant to an NHS population in England. (recommended)
BMI-adjusted mortality should be applied to non-CVD mortality only, with CVD mortality modelled separately through ASCVD health states. Alternatively, if using BMI-adjusted all-cause mortality, CVD death should not be modelled as a separate outcome. When BMI-adjusted mortality is used, granularity in BMI categories is preferred: age-sex-specific hazard ratios reflect the non-linear and age-dependent relationship between BMI category and mortality risk. (recommended)
When all-cause mortality evidence is available from trials use it to validate and, if necessary, calibrate the overall mortality effect predicted in the model. (required)
Sensitivity analyses should be used to explore the impact of length of time of living with overweight or obesity on mortality. (recommended)
For the rationale for these statements, see the rationale and supporting information section on mortality.