Impact calculation method: PIF PAF applied directly to GBD data or via PMSLT

Use of Global Burden of Disease data In the development of ITHIM-R (https://github.com/ITHIM/ITHIM-R) we aim to address issues related to the calculation of years of life lost (YLL) and years lived with disability (YLD) from the Global Burden of Diseases (GBD) studies by the Institute for Health Metrics and Evaluation (IHME). In past versions of ITHIM estimates of the population impact fraction (PIF) were directly applied to YLLs and YLDs to indicate the overall change in disability-adjusted life years (DALYs). The main issues with this approach are:

1. In the GBD studies, YLLs are estimated as the product of the number of deaths and a reference life expectancy (the highest attained in the world) by age group and sex. If the intention is to compare YLLs at a global scale, then this can be defended on grounds of equity. At the local level, however, it may be best to produce estimates based on observed life expectancy for the setting of interest.
2. In the GBD studies, the quality of life for each YLL lost is not specified. In reality a proportion of the YLLs would be lived in ill-health. Thus by averting premature deaths additional years lived with ill-health will be created (i.e. a proportion of the YLLs will become YLDs).
3. In the GBD studies, YLDs are estimated as the number of prevalent cases by the disability weight for the cause of interest. Applying the PIF to prevalent YLDs is then problematic, since the relative risks in the calculation of the PIF relate to becoming a case (incidence) and not being a case (prevalence).
   The options we see are: i. Only calculate YLLs and deaths ii. Calculate YLDs, using the PIF, but do not add up to YLLs. This would give indicative information on which diseases are affected. It would be particularly useful for diseases such as depression with a high YLD but low YLL burden. iii. Move exclusively to the proportional multi-state life table. With this we can estimate health-adjusted life years from life years based on the population of interest observed mortality rates (hence life expectancy) and adjust them by the quality of life. So our issues can be solved, however, at a high data cost (and works on the data with Dismod II). Unless we can generate these data for users it is not reasonable to expect the typical ITHIM user to do this. iv. Adjust the GBD data so we can more accurately calculate the data we want

If we want to try option iv then the solutions we see are:

• Issue 1
• Adjust life expectancy using locally relevant data
• Issue 2
• Given that we know the prevalence YLDs at each age group and we know the age specific life expectancy (YLLs/deaths) it is possible to adjust the YLLs gained for the average level of disease at each group.
• Issue 3 with YLDs (and DALYs as the sum of YLLs and YLDs)
• Estimate incidence YLDs. This would imply requesting duration data to IHME since incidence YLDs are estimated as incidencedurationDW. Incidence available at the GBD results tool is at the cause level (e.g. ischemic heart disease), hence duration and disability weights should match. This may be challenging since the GBD provides disability weights linked to sequalae which are associated with health states. An option is to do our own derivation of disability weights from prevalent years lived with disability and prevalence (both available from the GBD) and request duration (if it comes at the “cause level).