Table 1. Overview of variables.

Variable	Definition
Sociodemographics
Age	Years
Sex	Male | Female
Education: Highest achieved	Primary, Secondary, Tertiary
Risk factors
BMI	Weight in kilograms/(Height in metres^2)
In the last two years, have you stopped smoking?	Yes | No
Chronic health conditions
Has a doctor told you that you have the following conditions [ Diabetes/ Cancer/ Lung disease/ Hypertension/ Heart disease/ Stroke]?	Yes | No
Functional limitations
ADLs: Because of a health or memory problem, do you have difficulty doing any of the following activities: dressing bathing, eating?	Total difficulties (/3): 0 | 1 | 2 | 3
IADLs: Because of a health or memory problem, do you have difficulty doing any of the following activities: preparing a hot meal, shopping for groceries, making telephone calls, taking medications, managing money?	Total difficulties (/5): 0 | 1 | 2+
Mortality
Mortality	Confirmed died via GRO or family member
Health care utilisation in prior 12 months
ED visits	Count
Inpatient admissions	Count
Outpatient visits	Count

Legend: (I)ADLs: (Instrumental) Activities of Daily Living. BMI: Body mass index. ED: emergency department.

Sources: All variables were self-reported in the TILDA CAPI except for BMI, which was measured by a nurse in a health assessment centre at Wave 1 and self-reported in all subsequent waves. This created an inconsistency problem, where Wave 1 BMI was higher than at later Waves for the majority of participants. For data consistency we used BMI as an outcome in Wave 2 onwards only, and we used BMI as a predictor in all waves after adjusting the Wave 1 data to match the distribution of later Waves (e.g. if a participant had BMI in the 60 ^th percentile in Wave 1 then we adjusted their Wave 1 BMI to the 60 ^th percentile value for later waves).

Notes: Each diagnosis question was asked individually except for heart disease, which combines diagnosis of any one of heart attack, congestive heart failure, angina and cardiac arrhythmia. TILDA asked about six ADLs and six IADLs, but the current version of the model runs on the publicly available harmonised dataset, which contains only three and five of these respectively. Future iterations of the model will use the full dataset and model all six difficulties in each index. ADLs were used from Wave 2 onwards because of a measurement inconsistency: Wave 1 CAPI asked, “Because of a health or memory problem, do you have difficulty [dressing, including putting on shoes and socks]?” Wave 2 onwards asked, “Because of a health or memory problem, do you have difficulty [dressing]?”