Obesity is common and has been associated with type 2 diabetes (chronic progressive disease secondary to insulin insufficiency). Weight loss surgery can help people live longer, healthier lives and can sometimes reverse type 2 diabetes.
Lots of people could benefit from weight loss surgery but very few people are offered this surgery. But it is not widely accessible and there exist disparities between patients.
Many people with type 2 diabetes and their doctors do not think of weight loss surgery as a treatment option.
It is important to identify patients who will benefit the most following surgery to help all patients and doctors to discuss bariatric surgery and make informed decisions. Embedding prediction model in primary care will standardize the practice and may help with existing disparities.
Current models to predict type 2 diabetes reversal were developed in different countries with less ethnic diversity compared to the UK population and using information that is not normally collected by general practitioners (GPs); this means they may not be fair or work very well for UK patients.
First, we will see how well some of these existing models work in the UK population, particularly in different ethnic groups and socio-economic group patients. We may need to update the model to make sure it performs fairly in different groups. We will also work with patients and other interested people to create helpful information for patients and doctors to help decide whether weight loss surgery is right for them.
Aims and objectives:
1. Externally validate existing models to predict type 2 diabetes remission at short (1 year) and midterm follow-up at 3 and 5 years within Clinical Practice Research Datalink (CPRD) Aurum.
2. If any of the existing models perform well overall, the predictive performance will be evaluated in subgroups based on ethnicity and socio-economic status.
3. Update the best-performing model by recalibrating and potentially including additional variables available within primary care, to improve performance in different ethnic and socioeconomic (Index of Multiple Deprivation (IMD)) groups
Data sources: We will use CPRD Aurum, pseudo-anonymized electronic primary- care healthcare records to undertake external validation of available prediction models. Linkage to Hospital episode statistics (HES) admitted patient care (APC) will be required to further ensure the accuracy of bariatric surgery coding. We will use CPRD Gold for external validation of the updated prediction model.
Study population: Adults with type 2 diabetes and obesity who underwent bariatric surgery during the study period January 2010 - December 2021 will be included.
Outcome: Diabetes remission is defined as glycosylated hemoglobin (HbA1c) ≤ 6.5% and off glucose-lowering medications in the last 3 months.
Study design and methods
Based on predictors used, population used for developing the model, and performance, we identified four existing models- DiaRem, Ad-Diarem, DiaBetter and DiaRem2 models for validation.
Variables included in these prediction models are age, HbA1c, diabetes duration, diabetes medications, and insulin use.
Performance of prediction models will be assessed by discrimination (time-dependent C-statistics), calibration (time-to-event model and calibration curves created using pseudo values for observed outcomes).
Net benefit will be evaluated using decision curves.
Identification of a model with good performance will empower primary-care physicians to have informed consultation with patients seeking help for weight management. It will improve patient understanding of treatment options and access to bariatric surgery.
Model performance – We will examine the performance of existing models in UK population. Performance will be based on measures of discrimination (time-dependent C-statistic), and calibration (calibration-in-the-large, calibration slope, and calibration curves), as well as clinical utility (decision curves).
Diabetes remission will be defined as HbA1c ≤ 6.5%, off glucose-lowering medications in the last 3 months
We will predict short-term and long-term diabetes remission. Short-term diabetes remission will be defined as remission within 12 months and long-term diabetes remission will be defined as remission within 36 and 60 months. We will include HbA1c measures within 3 months of each endpoint (12, 36, and 60 months).
PUSHPA SINGH - Chief Investigator - University of Birmingham
Nicola Adderley - Corresponding Applicant - University of Birmingham
Jonathan Hazlehurst - Collaborator - University of Birmingham
Krishnarajah Nirantharakumar - Collaborator - University of Birmingham
Kym Snell - Collaborator - University of Birmingham
Shamil Haroon - Collaborator - University of Birmingham
Srikanth Bellary - Collaborator - Aston University, Birmingham
HES Admitted Patient Care;Patient Level Index of Multiple Deprivation