CPRD-Katie-CVD-HF

Introduction

 

Target trial specification

Protocol component	Target trial protocol	Emulation
Eligibility criteria	Inclusion: Patients with T2D, aged >=18 years treated with antihyperglycaemic medication (those receiving non-MFN as first line therapy are weird). Exclusion: History of CVD, CKD (stage 3a-5) or HF (as separate guidance/contraindications exist for some groups). Current insulin treatment (as indicates advanced diabetes). Prior use of SGLT2i. Sensitivity analysis: only those on MFN and no prior use of second-line therapies (ever/in previous year as per Lancet paper?)	Inclusion: Patients with T2D, aged >=18 years treated with antihyperglycaemic medication. Exclusion: History of CVD, CKD (stage 3a-5) or HF (as separate guidance/contraindications exist for some groups) Current insulin treatment. Prior use of SGLT2i or SU (although we don't have complete records for everyone going back to diagnosis, but have to assume number who go off and then back on drug is small) Sensitivity analysis: only those on MFN and no prior use of second-line therapies (ever/in previous year as per Lancet paper?)
Treatment assignment	Participants randomly assigned to SGLT2i or placebo	Emulate random assignment to SGLT2i or SU (no evidence of CVD effect) using: * Overlap weighting * IPTW * Propensity score matching *Regression adjustment Sense check will be comparing DPP4 (third arm) to SU
Treatment initiation	Initiation of SGLT2i or placebo (and no concurrent treatment initiations)	Initiation of SGLT2i or SU (and no concurrent treatment initiations). Initiation no earlier than 01/01/2013 as this is when SGLT2is began being prescribed in UK. At least 90 days of pre-initiation data available so we know this is an initiation (can do sensitivity analysis with more data as per other papers). Patients can be in both arms at different times - unlike target trial.
Treatment strategy	Given medication as part of trial. Further additional treatment with DPP4, SU, MFN, Glinide (not TZD, SGLT2i, GLP1 [due to CVD effects]) after the treatment initiation period is allowed as needed.	Prescription recorded in primary care records - we cannot be sure patient took all medication. Assume stopped if gap of 6 months or more. Further additional treatment with DPP4, SU, MFN, Glinide (not TZD, SGLT2i, GLP1 [due to CVD effects] or insulin [due to exclusion]) after the treatment initiation period is allowed as needed.
Follow-up	Starts at treatment initiation. For ITT ends up to 5 years after initiation or if experience outcome. Censor if lost to follow up or death	Starts at treatment initiation. For ITT ends up to 5 years after initiation or if experience outcome. Censor if end of GP records (deregistration), or start TZD/SGLT2i/GLP1/insulin, or death.
Outcomes	HF: new onset HF CVD: New onset 3-point MACE	HF: any recording of HF in GP/HES/ONS (death) data CVD: any recording of MI (broader ICD10 [HES] codelist including history/complications) or stroke (broader ICD10 [HES] codelist including history/complications) in GP/HES data, or CV death in ONS data. Sensitivity analysis: HF: HF as primary reason for hospitalisation in HES or primary death cause in ONS data CVD: MI/stroke (narrower ICD10 [HES] codelists) as primary reason for hospitalisation in HES or CV death as primary cause of death in ONS data. Can see if broader codelists actually amke any difference
Causal contrasts of interest	ITT effect of SGLT2i over placebo on time to first HF / 3-point MACE	ITT effect of SGLT2i over SU on time to first HF / 3-point MACE
Analysis plan to estimate causal contrasts of interest

 

Risk score (QDiabetes-Heart Failure [QDHF] 2015 and QRISK2 2017) information

QDiabetes-Heart Failure paper: https://bmjopen.bmj.com/content/5/9/e008503 QRISK2 paper: https://www.bmj.com/content/336/7659/1475 QRISK2 2013 codelists: https://qrisk.org/QRISK2-2013-Code-Lists.pdf (link no longer works)

 

Variable coding

The following table provides a brief overview of how we have defined the variables for these risk scores in CPRD, and how this might differ to the variable definition used to validate these risk scores / the way in which QRISK2 has previously been automatically run on GP systems in the UK (where known; information from papers and QRISK 2013 codelists as linked above). All our codelists are available here: https://github.com/Exeter-Diabetes/CPRD-Codelists/.

Variable	Which risk scores	How this was defined during validation / for QRISK2 in automatic GP systems	How we have defined it
Age	Both		Age from DOB; DOB definition
Sex	Both		Provided by CPRD
Ethnicity	Both	UK census Read codes, no details on how conflicting codes dealt with	Medcode list (includes all of QRISK2 Read codes + extras); hospital (HES) ethnicity used if GP missing; and our algorithm if conflicting codes are present as per our algorithm. Some missingness
Deprivation	Both	Uses patient postcode + lookup of Townsend Deprivation Scores	We only have IMD deciles, not Townsend Scores, so are using the median Townsend Score for LSOAs with same IMD decile as patient
Smoking status	Both	Read codes, but quantity (value) associated with code used to determine whether non/current light/current moderate/current heavy smoker (ignore values for ex-smoker codes). If value missing, use code type to determine category (light if unspecified)	Medcode list (includes all of QRISK2 Read codes and extras) and have tried to replicate their algorithm (see details). We only include values with valid/missing numunitid. Some missingness
Diabetes status (none/T1/T2; or just T1/T2 for QDHF)	Both	Type 1 and Type 2 (includes unspecified type) Read codes. Uses most recent code; if has both types on that day then Type 2	Our algorithms
Diabetes duration	QDHF		Our algorithm for diagnosis date
Angina or heart attack in 1st degree relative <60	QRISK2	Read codes	Medcode list (includes all of QRISK2 Read codes – including ones for angina/MI in relative <age 55 which differs from <60 stated on tool; and extras including CVD in 1st degree relative <60 [not specifically MI or stroke])
Heart attack, angina or stroke	QDHF		Medcode and ICD10 (HES) lists
CKD stage 4 or 5	Both	Read codes	Our algorithm which uses creatinine measurements and CKD stage 5 codes. If no baseline CKD stage, set to 0 not missing
Atrial fibrillation	Both	Read codes	Medcode list (doesn't include all of their AF Read codes, but theirs add <0.1% to occurrences) and ICD10 (HES) codes
On BP meds	QRISK2	QRISK (original) paper says they used beta-blockers, ACE-inhibitors, calcium channel blockers and thiazide-like diuretics. QRISK2 paper says this variable is ‘treated hypertension’ = diagnosis of hypertension (Read code list available) and ≥1 current prescription of at ≥1 antihypertensive agent. QRISK3 paper says ‘Use of drugs at baseline was defined as at least two prescriptions, with the most recent one no more than 28 days before the date of entry to the cohort’	Prodcode lists for the 4 types in QRISK paper, didn’t use hypertension codes as per QRISK2, used rule that had to have ≥2 prescriptions with last 1 within 28 days of index date as per QRISK3 paper
Rheumatoid arthritis	QRISK2		Medcode and ICD10 (HES) lists
These only identify ~half of the people as QRISK2 codelist. 2% of drug starts have RA at drug start currently
HbA1c	QDHF		Medcode list, only including if valid/missing numunitid. Some missingness.
Cholesterol:HDL ratio	Both	Separate cholesterol and HDL Read code lists	Medcode lists for HDL and total cholesterol, only including if valid/missing numunitid. HDL codelist doesn’t include one of their seven codes – adds <0.1% to total occurrences. Not using specific cholesterol:HDL ratio codes as these add <1% additional patients
SBP	Both	Read code list (one code only)	Medcode list (includes their one SBP Read code), only including if valid/missing numunitid
BMI	Both	Read code list (one code only)	Medcode list (includes their one BMI Read code), only including if valid/missing numunitid

 

Additional: QRISK3 paper states that for SBP, BMI, and smoking status, the most recent values before baseline were used. For cholesterol:HDL ratio, the most recent before or after baseline was used as long as it was before CVD diagnosis/censoring/statins. QDHF paper states that for continuous variables, the closest values prior to baseline or within 6 months after were used. We have used the most recent value prior to baseline (maximum of two years earlier, or 6 months for HbA1c) or up to 7 days after if closer.

 

Algorithm coding

Code for implementing the QDiabetes-Heart Failure 2015 and QRISK2 2017 algorithms was adapted from the open source software available on their respective websites (QDHF: https://qdiabetes.org/index.php; QRISK2 no longer available) and are included in the EHRBiomarkr R package.

 

Implementation

Those with biomarker values outside the range used by the online versions of the QDiabetes-Heart Failure 2015 and QRISK2 2017 calculators were excluded. The ranges were:

• QDHF:
  • Cholesterol:HDL ratio: missing or 1-11
  • HbA1c: 40-150 mmol/mol
  • SBP: missing or 70-210 mmHg
  • Age: 25-84 years
  • Also excluded if BMI<20 kg/m2 as very different from development cohort
• QRISK2:
  • Cholesterol:HDL ratio: missing or 1-12
  • SBP: missing or 70-210 mmHg
  • Age: 25-84 years
  • Also exclude if BMI<20 kg/m2 as very different from development cohort

 

Outcome variable definitions