Abstract
<jats:p>Objectives: To examine the conformance to type 2 diabetes NICE guidelines across cardiovascular risk strata; and to quantify geographical variation in treatment pathways following the COVID-19 pandemic, encompassing guideline changes. Design: We carried out a retrospective observational study using linked electronic health records across England. Process mining, a data driven method that can reconstruct clinical treatment pathways, was applied to map 12-month treatment trajectories after treatment initiation. Conformance with NICE NG28 (2022) was quantified using a structural similarity index. Further, behavioural and entropy-based similarity (capturing treatment variability and complexity) measures were used to assess sequencing and heterogeneity of treatment. Setting: Primary and secondary care in England datasets within the National Health Service England Secure Data Environment (NHSE SDE), analysed first at national level and then across 42 Integrated Care Boards (ICBs) which are the devolved health care geographical delivery regions in England. Participants: 822,650 individuals with newly diagnosed T2DM between 1-February-2022 and 1-November-2025, stratified into low cardiovascular risk (LR-C; QRISK3<10), high risk (HR-C; QRISK3>=10 or receiving statins/blood pressure lowering treatment), and established cardiovascular disease (eCVD-C). Participants were followed for 12 months after first dispensed glucose lowering therapy. Main outcome measure: First line therapy, treatment intensification and switching within 12 months; change in glycated haemoglobin (HbA1c); quantified conformance to NICE recommended pathways; and regional variation in broader similarity measures. Results: Metformin monotherapy was the dominant initiation strategy in LR-C and HR-C cohorts (92.4% and 90.2%, respectively), whereas eCVD-C showed lower uptake of metformin (68.9%) and higher uptake of SGLT2 inhibitors (26.3%). Intensification from metformin to combination therapy was infrequent across all cohorts (<1%), although HR-C demonstrated the highest treatment transitions and switching behaviour. Dispensed SGLT2 inhibitor use was nearly threefold higher in eCVD-C (26.7%) than in LR-C (9.0%) or HR-C (10.7%). Overall, conformance to NICE-recommended pathways remained modest nationally, particularly in LR-C and HR-C. Across 42 ICBs, substantial regional heterogeneity in treatment pathways and guideline conformance was observed, with conformance ranging from 0.29 to 1.00 in LR-C pathways, 0.40 to 1.00 in HR-C pathways, and 0.54 to 0.92 in eCVD pathways. Conclusion: National T2DM treatment pathways for post-pandemic showed higher alignment to NICE guidelines in eCVD-C compared to the other risk groups, with ongoing gaps and large regional variations in other risk groups. Process mining offers a scalable approach to monitor implementation of guideline recommended care that could support learning health systems. Using T2DM during and post COVID-19 pandemic as a case study, this work demonstrates how these methods can assess the use of existing and innovative therapies, identify gaps and guide future adoption to ensure recommended treatments reach the right patient groups.</jats:p>