The North West London Diabetes Cohort (NWLDC) was established to provide a systematic characterisation of a large diabetes population drawn from electronic health records, with the explicit aim of creating a foundation for complications research and prognostic modelling. The cohort addresses an identified gap in many predictive modelling studies, which often prioritise model accuracy while under-reporting the descriptive characterisation of the underlying population. By documenting demographic composition, clinical measures and complication incidence patterns in detail, the NWLDC is intended to support development of reliable prognostic tools and evidence-based prevention strategies for diabetes complications. The notably diverse, multiethnic population enables investigation of ethnic disparities in complication risk and model performance.
At baseline the cohort comprised 337,271 patients with recorded diabetes. Case counts by diabetes type were: 279,067 patients with type 2 diabetes, 17,638 with type 1 diabetes, 33,590 with gestational diabetes and 6,916 with unspecified diabetes. The dataset includes historical diagnoses with the earliest recorded diabetes diagnosis dated to January 1932, and data were updated through 27 May 2025. The cohort therefore spans long-term longitudinal follow-up for many individuals and permits incidence and temporal sequence analyses of complications.
Comprehensive baseline data captured on cohort participants include demographic variables (age, sex, Deprivation Index and ethnicity) and routinely recorded clinical measures. Key clinical measures available across the cohort are glycated haemoglobin (HbA1c), body mass index (BMI), blood pressure, lipid profiles and estimated glomerular filtration rate (eGFR). These variables support characterisation of cardiometabolic risk status at baseline and enable adjustment or stratification in prognosis and complication analyses.
The NWLDC monitors 14 major diabetes complications longitudinally. In the descriptive analyses presented for patients with type 2 diabetes, the most common recorded complication was diabetic retinopathy, with an incidence of 74.6 per 1,000 person-years. Hypertension followed with an incidence of 51.0 per 1,000 person-years, and kidney disease with 31.4 per 1,000 person-years. These incidence rates were derived from longitudinal follow-up within the EHR dataset and provide a baseline landscape of complication burden in the cohort.
Cumulative incidence analyses accounted for mortality as a competing risk using the Aalen-Johansen estimator. These analyses demonstrated significant ethnic disparities in complication risk: Black, Asian, mixed and other ethnic groups showed elevated risks compared with White patients. The use of a competing-risk estimator acknowledges that mortality can preclude the occurrence of non-fatal complications and provides more appropriate estimates of absolute cumulative incidence in the presence of differential mortality.
Time-varying Cox proportional hazards models were used to examine temporal associations between complications. These models identified strong clustering between cardiovascular and renal complications, supporting the concept of a cardiometabolic-renal syndrome within the diabetes population. Such clustering highlights the interdependence of vascular and renal pathways in patients with diabetes and underscores the need to consider comorbid complication trajectories when building prognostic models and planning interventions.
The cohort profile notes that mental health conditions, specifically depression and anxiety, were prevalent across the disease timeline. These conditions occurred both before and after the recorded diabetes diagnosis, indicating that mental health comorbidity is common and temporally distributed in relation to diabetes onset. The presence of mental health conditions has implications for integrated care, risk stratification and potentially for complication risk modelling.
The NWLDC is planned to serve as a platform for developing and validating prognostic models for diabetes complications, enabling risk stratification and targeted intervention strategies. Future work described for the cohort includes: incorporation of medication data to refine diabetes type classification; examination of the effectiveness of antidiabetic medications in preventing different complications; and assessment of demographic differences in prognostic model performance and prediction accuracy. To better characterise lifestyle factors and recorded clinical advice, further interrogation of EHR data will examine the recording of dietary advice, referrals to weight management schemes and the presence of alcohol consumption codes. These planned extensions aim to improve both explanatory analyses and predictive modelling built from the NWLDC resource.