Table 1 IMPACT_NCD key assumptions and limitations

Sociodemographic module

Migration is not modelled explicitly in the model. However, the model outputs are calibrated to ONS population projections, which include migration (we used the ONS principal migration assumptions). Nevertheless, we assume that migrants have similar characteristics to the local population.

Social mobility is not considered.

Decile groups of the index of multiple deprivation (DIMD) are a relative marker of (area) deprivation with several versions since 2003. We have used the 2015 version and assume it is constant throughout the simulation⁵⁴.

Exposure module (see Supplementary Methods Exposure module section p7).

We assume that the surveys used are truly representative of the population. For example, the adjustments for selection bias in the Health Survey for England are adequate.

On average, simulants remain in the same exposure quintile group throughout their life.

The linear correlations in exposure quintile groups remain constant over time (i.e. the clustering of exposures in some subpopulations).

We assume that trends in risk factor exposures continue and follow log-linear trends.

Disease module

We assume multiplicative risk effects (see Supplementary Methods Disease Incidence section p10).

We assume a log-linear exposure-response for the continuous risk factors.

We assume that the effects of the risk factors on incidence and case fatality are equal (see Supplementary Methods Mortality section p16).

We assume a mean lag time between exposure and outcome of about 4–5 years for most exposure/outcome pairs, except for cancers, for which we assume a mean lag time of 9 years (see Supplementary Methods Disease Incidence section p10).

We assume 100% risk reversibility for all exposures except smoking. We allow smoking to have a cumulative effect on the risk for COPD and lung, breast, and colorectal cancers.

We assume that trends in disease incidence are attributable only to trends of the relevant modelled risk factors or other diseases modelled.

We assume that the linked primary care data used to model disease trends over time represents England’s adult population.

We assume that trends in disease incidence continue to follow log-linear trends (other than pain, see below).

For cancers, we assume that survival 10 years after diagnosis equals remission.

For all conditions other than cancer (see point above), pain, constipation, asthma, alcohol problems, and anxiety and depression, we assume conditions are chronic.

For pain, we modelled the incidence of pain based on the incidence in 2013 due to data quality issues over time with the prescription data.

For anxiety and depression and constipation, we did not calibrate to the observed trends in incidence rates because their projections led to implausible rates.