Predictive modeling for incident and prevalent diabetes risk evaluation

Abstract

With half of individuals with diabetes undiagnosed worldwide and a projected 55% increase of the population with diabetes by 2035, the identification of undiagnosed and high-risk individuals is imperative. Multivariable diabetes risk prediction models have gained popularity during the past two decades. These have been shown to predict incident or prevalent diabetes through a simple and affordable risk scoring system accurately. Their development requires cohort or cross-sectional type studies with a variable combination, number and definition of included risk factors, with their performance chiefly measured by discrimination and calibration. Models can be used in clinical and public health settings. However, the impact of their use on outcomes in real-world settings needs to be evaluated before widespread implementation.

Keywords:

• diabetes
• incident
• prevalent
• risk prediction
• screening

Financial & competing interests disclosure

K Masconi was supported by a scholarship from the South African National Research Foundation and the Carl & Emily Fuchs Foundation. The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.

No writing assistance was utilized in the production of this manuscript.

Key issues

• A fast increasing number of multivariable models to predict prevalent undiagnosed or incident diabetes have been developed, but only a few have been tested in diverse settings.

• It is not always obvious from published studies to ascertain how the development of existing models has addressed the critical methodological challenges, which may affect the performance of the models both on the derivation sample, and in subsequent external validation studies accurately.

• The complexity of existing models varies substantially and in the absence of head-to-head comparison studies, it will be very difficult in most settings to choose the most appropriate model.

• Existing models are mostly based on glycemia-defined diabetes and may not be all valid in the context of the recommendations for using HbA1c as well for diabetes diagnosis.

• Existing models overwhelmingly originate from developed countries and have seldom been tested in developing countries that may derive the most benefits from the introduction of those models in routine care.

• Studies to assess the impact of adopting diabetes prediction models in routine care are a very recent development, and little is known on the effect of introducing diabetes prediction models in routine care on the behavior of healthcare providers and the outcomes of care.