Table 1. Predictors considered, and how they are represented in CPRD and in our models.

Category	Predictor	Representation in CPRD	Representation in models	In QA	In QA+	In T
Demographics	Age	Year of birth	Computed based on mid-year of year of birth	*	*	*
	Sex	Binary variable (male/female)	Binary variable	*	*	*
	Ethnicity	Ethnicity (categorical value)	Categorical variable	*	*	*
Lifestyle and family history	Socioeconomic status	Index of Multiple Deprivation	Numeric variable on a scale of 1 to 5	*	*	*
	BMI	Weight measurement recorded repeatedly in various clinic visits	BMI based on height and most recent recorded weight	*	*	*
	Smoking status	Current tobacco use, in terms of number of cigars/cigarettes per day (recorded repeatedly)	Categorical variable for latest status: non-smoker, ex-smoker, light smoker (less than 10 cigarettes/day), moderate smoker (10–20 cigarettes/day), heavy smoker (more than 20 cigarettes/day), smoker (amount not recorded)	*	*	*
	Alcohol intake	Current alcohol consumption, in terms of units of alcohol per day (recorded repeatedly)	Categorical variable for latest status: non-drinker, ex-drinker, trivial (less than 1 unit/week), light (1–2 units/week), moderate (3–6 units/week), heavy (7–9 units/week), very heavy (more than 9 units/week), drinker (amount not recorded)	*	*	*
	Family history of chronic disease	Binary variable (yes/no)	Binary variable (yes/no)	*	*	*
	Strategic health authority (region)	Categorical variable	Categorical variable	*	*	*
	Marital status	Categorical variable	Categorical variable		*	*
Use of care	Previous emergency admissions	Read Code and date of event	Number of occurrences during last year	*	*	*
			Time since last occurrence (in days)			*
	Prior GP visits (consultations)	Read Code and date of event	Number of occurrences during last year		*	*
			Time since last occurrence (in days)			*
			Total duration spent in GP visits (minutes)			*
Clinical diagnoses (comorbidities)	Diabetes, atrial fibrillation, cardiovascular disease, congestive cardiac failure, venous thromboembolism, cancer, asthma or COPD, epilepsy, falls, manic depression or schizophrenia, chronic renal disease, chronic liver disease or pancreatitis, valvular heart disease, treated hypertension, rheumatoid arthritis or SLE, depression (QOF definition)	Read Code and date of entry	One separate binary variable for each disease, 16 variables in total	*	*
			Time since first diagnosis (in days)—1 separate variable for each disease, 16 variables in total			*
	Arthritis, connective tissue disease, hemiplegia, HIV/AIDS, hyperlipidaemia, learning disability, obesity, osteoporosis, peripheral arterial disease, peptic ulcer disease, substance abuse	Read Code and date of entry	One separate binary variable for each disease, 11 variables in total		*
			Time since first diagnosis (in days)—1 separate variable for each disease, 11 variables in total			*
Clinical measures and laboratory tests	Systolic blood pressure, haemoglobin, cholesterol/HDL, liver function test (γ-GT, aspartate aminotransferase, or bilirubin), platelets, ESR	Numeric value for result and date of measurement	Binary (yes/no) variable for if recorded—1 variable per test	*	*
			Numeric variable for most recent result—1 variable per test	*	*	*
			Binary variable for abnormal result—1 variable per test	*	*	*
			Time since the latest result (in days)—1 variable per test			*
Prescriptions	Statin, NSAID, anticoagulant, corticosteroid, antidepressant, antipsychotic	Date of prescription if applicable	Binary (yes/no) variable for if prescription exists	*	*	*

γ-GT, γ-glutamyl transferase; COPD, chronic obstructive pulmonary disease; CPRD, Clinical Practice Research Datalink; ESR, erythrocyte sedimentation rate; GP, general practice; HDL, high-density lipoprotein; NSAID, non-steroidal anti-inflammatory drug; QOF, Quality and Outcomes Framework; SLE, systemic lupus erythematosus.