Study design and data sources
This historical cohort study used data from clinical audit and routinely collected health records. Our cohort was defined using primary care data from the National Chronic Kidney Disease Audit (NCKDA) research database, and secondary care data from the Myocardial Ischaemia National Audit Project (MINAP) and Hospital Episode Statistics (HES) Admitted Patient Care (APC).
The NCKDA [24, 25] aimed to audit and improve primary care health services in England and Wales for people with CKD or CKD risk factors (Additional Table 1). The audit included 10% of English General Practices (approximately 1.7 million people with CKD or risk factors for CKD) who invested in audit software and volunteered to participate in the audit, and is now used as a research database to study long-term outcomes of this population [22, 24, 25]. The NCKDA collected complete historical patient-level data for eligible participants from general practices in two main cross-sectional extracts between 2014 and 2016. People who died between extracts, opted-out of data-sharing (person or practice-level), or people who changed GPs were excluded in the second extract. People included in the NCKDA were generally representative of the English population in terms of age and sex .
HES Admitted Patient Care data are collected to compensate hospitals for services provided by the NHS. In England, all hospitalisations funded by the NHS (approximately 99%) are captured by HES data . Diagnoses during hospitalisation are recorded using International Classification of Diseases 10th Edition (ICD-10) codes.
MINAP, part of the National Institute of Cardiovascular Outcomes Research (NICOR) audit and research programme, aims to audit all type 1 AMIs admitted to hospitals in England and Wales. Data are collected on patient characteristics, laboratory tests, comorbidities, processes of care, and treatment received during AMI hospitalisation [27, 28].
Office of National Statistics (ONS) data were linked to these primary and secondary care data to determine death dates  (Additional Table 2). The Index of Multiple Deprivation (IMD) patient-level data were linked as a proxy for socioeconomic status (SES) .
HES and MINAP-linked data were available up to 31 March 2017. Anyone with an AMI hospitalisation between the final NCKDA extract and the end of HES/MINAP linked data were included in the cohort. The ONS linked follow-up death data were available up to 15 September 2019.
We included people captured by the NCKDA research database with an AMI hospitalisation recorded in MINAP, HES, or both between 2015–2017, after the final NCKDA cross-sectional extract in which the person appeared . We identified incident AMI hospitalisations and AMI subtypes (STEMI, NSTEMI) in HES using ICD-10 codes (Additional Table 3) recorded in the first diagnostic position of the first episode of the spell, and in MINAP using an algorithm which uses discharge diagnosis, cardiac marker levels, and electrocardiogram results (Additional Table 4). People with CKD risk factors, but no eGFR in the primary care record (n = 118), were excluded.
We calculated the baseline eGFR from the most recent serum creatinine value recorded in primary care prior to the index AMI hospitalisation using the MDRD equation . We defined eGFR categories using the same cut-points KDIGO recommends for the definition of CKD stages: Category 1–2 (eGFR 60-120 mL/min/1.73m2), 3a (eGFR 45–59), 3b (eGFR 30–44), and 4–5 (eGFR 0–29) .
Our primary outcomes were all-cause death during the first AMI hospitalisation recorded during the study period (the index AMI hospitalisation), and all-cause death during follow-up, for those who survived the index AMI hospitalisation. Variables used to define death date are described in Additional Table 5. Secondary outcomes were treatments received during hospitalisation: (1) Angiography and/or percutaneous coronary intervention (PCI), and (2) coronary artery bypass graft (CABG) (Additional Table 6). Other secondary outcomes, among survivors of the index AMI hospitalisation, were AMI re-admission and cardiovascular-specific death post-index AMI discharge.
Potential confounding variables available in our dataset were age at AMI hospitalisation (continuous), sex, ethnicity (white, other), IMD quintile, smoking status (non-smoker, ever smoker), receipt of dialysis or kidney transplant, prior AMI, and comorbidities including chronic obstructive pulmonary disease (COPD), type 2 diabetes mellitus (T2DM), heart failure, unstable angina, cerebrovascular disease, hypertension, and peripheral vascular disease. We defined these covariates using a combination of primary and secondary care data (Additional Table 7) . We categorised each hospital centre which contributed patient-level data to this study into two main categories: (1) PCI always available and (2) PCI services not always available (Additional Table 8).
We described baseline characteristics of the study population stratified by eGFR category. We used multivariable logistic regression to estimate the adjusted odds ratios comparing the odds of death during the index AMI hospitalisation (primary outcome) and the odds of invasive management (angiography and/or PCI, coronary artery bypass graft (CABG)) across eGFR categories. We also calculated predicted percentages from the adjusted logistic regression models using recycled predictions, since odds ratios can be misleading when the outcome is common . We looked at these associations in the overall study population and stratified by AMI subtype (STEMI and NSTEMI). We tested for a linear trend in the association between eGFR category and the odds of receiving angiography and/or PCI using a likelihood ratio test.
We used Cox regression to investigate the association between eGFR category and outcomes post-index AMI hospitalisation among survivors, including all-cause mortality (primary outcome), cardiovascular-specific mortality, and AMI re-admission, after confirming the proportional hazard assumption using a global test on the Schoenfeld residuals over time. We first calculated crude rates for each outcome stratified by eGFR stage by dividing the number of outcome events by the total person-time study participants contributed following discharge from the index AMI hospitalisation. We reported these crude rates per 100 person-years. In our multivariable models, we specified a priori to adjust for age (continuous), sex, ethnicity, IMD quintile, COPD, T2DM, heart failure, and prior AMI as we anticipated these to be the most important confounders for this study population.
We repeated the main analyses, stratifying by (1) centre type, to understand the impact of PCI availability on the association between eGFR category and the odds of receiving angiography and/or PCI; (2) and relevant comorbidities (prevalent T2DM and heart failure), since it is possible people with these comorbidities experience different management and outcomes compared with those without. We also repeated all main analyses after excluding people with prior AMI (n = 1,883) as previous coronary intervention may impact subsequent care.
We conducted a complete case analysis, excluding people with missing ethnicity and/or IMD data (n = 107). Discharge dates were missing in 19% of MINAP and 1% of HES records. We imputed missing discharge dates using the median number of days in-hospital from non-missing records (5 and 4 days in MINAP and HES, respectively) .
Patient and public involvement
This study benefited from similar patient and public involvement as described in a related study . The creation and maintenance of the NCKDA research database, including its record linkages and necessary section 251 permissions benefited from the support of the Kidney Care UK patient organisation (https://www.kidneycareuk.org/). Feedback from patient members of the UK Renal Registry Patient Council (https://renal.org/patients/patient-council) supported a further planned record linkage of renal and cardiac data.