Chronic kidney disease (CKD) affects over 10% of the population worldwide. Complications include cardiovascular morbidity, death, and progression to kidney failure requiring dialysis. The number of CKD cases is growing, with implications for patients and healthcare services. Action is needed to support earlier diagnosis and better targeted care with the potential to delay disease progression and reduce associated complications. Increasing availability of electronic healthcare records (EHRs) can support study of CKD and its progression in the general population. Regular kidney function testing which is recommended in clinical practice has the potential to capture patients with CKD and subsequent progression of disease. Large sample sizes and long duration of follow-up can support study of rare outcomes such as kidney failure and decline in kidney function which may progress slowly over many years in some patients. However, there is variation in recognition of CKD in clinical practice and there are challenges in detecting CKD due to its asymptomatic nature in the early stages, relying on blood tests to detect. Availability of kidney function test results is likely to depend on patient risk factors, healthcare seeking behaviours and healthcare provider factors (“informative testing”). This may lead to selection bias and impact reliability of research findings. This thesis aimed to explore and highlight the challenges resulting from issues of data quality and completeness inherent to EHRs when used to study the epidemiology of progression of CKD, and to present approaches to overcome these challenges. Firstly, a systematic review showed substantial risks of selection bias in previous research, due to selection procedures for study inclusion and completeness of data captured during follow-up for outcomes. Generally, large proportions of patients were excluded from analysis due to missing data, with little reflection on the implications of bias in study results and unrepresentative samples. Statistical methodology varied widely, with varying capability of handling missing data. Secondly, a feasibility study investigated data quality and completeness for kidney function tests conducted in UK primary care. Testing was uncommon in adults overall, but there was high frequency of repeat testing in patients with risk factors for CKD, with the potential to capture most patients with CKD. However, reasons for missing data weren’t clear, and data may be disproportionately missing due to certain risk factors or management in secondary care. Data quality issues due to historical laboratory reporting problems led to underestimation of decline in kidney function but impact was small in most patients. Thirdly, we studied the association between GP practice completeness of diagnostic coding for CKD and patient-level hospitalisation outcomes in patients with CKD in England. The use of a practice-level exposure aimed to reduce risks of unmeasured confounding. Being registered at a higher coding practice was associated with lower rates of hospitalisations for CV events, after adjustment for other practice factors. Finally, we developed new risk prediction equations for kidney failure requiring dialysis, using EHRs capturing the entire healthcare system in Stockholm, Sweden. Previously validated equations require data that is not routinely collected in most patients, but our analysis included 98% of patients identified with CKD, by including predictor variables that are routinely available. New models were precisely estimated and achieved high discrimination. EHRs hold huge value to study progression of CKD due to large sample size and long duration of regularly collected kidney function tests. However, issues of informative missingness and sampling bias have not been appropriately acknowledged and addressed in previous research. Future research should ensure that research questions can be answered with available data using appropriate statistical techniques, and with improved transparency of potential for selection bias. Research in this thesis has strengthened evidence for the importance of diagnostic coding for CKD in clinical practice in reducing risk of complications of CKD, by enabling improved patient care. New risk models have the potential to improve equality of healthcare, enabling risk prediction in all patients with CKD, but require validation in the UK.