Real-world data analysis to inform modelling

Download (PDF)

Table 3 shows data analysis opportunities that can help with model inputs that meet this reference case extension or supporting information about model design.

Any real-world data analysis should follow the principles set out in the NICE real-world evidence framework. When estimating baseline rates, these might be for a "behavioural intervention" arm that focuses on healthy diet and physical activity, and for a population that is not having bariatric procedures or incretin agonists. Data should be relevant to an NHS population in England; data from NHS populations across the UK are considered generalisable and can be used.

**Table 3: Useful real-world data analyses**
Parameters or analysis	Details	Examples of suitable data sets
Baseline population characteristics	Extract age-sex, weight and comorbidity distributions by stratum in the prevalent population	GP medical records (for example, Clinical Practice Research Datalink or OpenSafely) linked to hospital episode statistics (HES)
Choosing health outcomes and comorbidities	Test association between weight loss or gain and health outcome changes	GP medical records linked to HES and general registry data
Baseline risk of health events and outcomes	Estimate baseline rates of all‑cause mortality, acute clinical events and incident comorbidities by stratum (defined in table 1), and preferably by age-sex group and obesity subgroup	GP medical records linked to HES and Office for National Statistics (ONS) Civil Registrations of Death
Risk equations for health events and outcomes	Conduct multivariate analysis that includes age, sex and risk factors (weight or body mass index [BMI], systolic blood pressure, HbA1C and cholesterol level), by stratum (defined in table 1), and preferably by age-sex group and obesity subgroup	GP medical records linked to HES
Baseline weight change over time	Estimate weight change over time, by stratum (defined in table 1) and preferably by age-sex group and obesity subgroup	GP medical records
Weight change after discontinuation	Estimate BMI change over time after discontinuation of incretin agonists or after bariatric procedures, by stratum (defined in table 1), and preferably by age-sex group, obesity subgroup, and reason for discontinuation	GP medical records
Procedure-related complications	Estimate the proportion of people with complications for each procedure, and the number of complications for each procedure, as people may experience more than one complication per procedure	National registries (for example, National Bariatric Surgery Registry) or GP medical records linked to HES
Health state utilities	Multivariate analyses to determine EQ-5D, with BMI, age, sex and various comorbidities or health events as covariates	Registries or surveys (for example, Health Survey for England)
Health state costs	Multivariate analyses to determine healthcare cost, with BMI, age, sex and various comorbidities or health events as covariates	GP medical records linked to HES, combined with unit costs from standard UK sources (for example, National Cost Collection).
Resource use of behavioural interventions	Type, frequency and duration of healthcare professional contacts in addition to any other resource use (for example subscriptions)	GP medical records or surveys

Further research on data inputs is required to populate models for genetic or syndromic obesity.