Phenotype and Lifestyle Intervention: Potential for Predicting and Improving Metabolic Outcomes?

Diagnosing prediabetes based on glycated hemoglobin A_1c (HbA_1c) tests has generated debate over the potential for misdiagnosis and resource allocation for “screen and treat” diabetes prevention initiatives (1). Both the HbA_1c and blood glucose test methods diagnose prediabetes based on intermediate test results that fall between normal and diabetes diagnostic range values. However, the blood glucose diagnostic tests can potentially characterize the pathophysiology associated with prediabetes and differentiate prediabetes phenotypes (2). Prediabetes phenotyping is intended to characterize the pathophysiology associated with prediabetes, which varies by the diagnostic method; a diagnosis of prediabetes may be based on any one or any combination of the following criteria: 1) impaired fasting glucose (IFG); 2) impaired glucose tolerance (IGT), determined by 2-h glucose tolerance test (GTT) for glucose response to a 75-g challenge; and 3) intermediate HbA_1c level.

A 2021 meta-analysis by Cai et al. (3), which included 129 studies and over 10 million participants, found that the diagnosis of prediabetes based on IGT was associated with a higher risk of all-cause mortality, coronary heart disease, and stroke than when diagnosed based on IFG. The authors concluded that the risk reduction effects attributed to lifestyle in trials that included IGT in the eligibility criteria are rated as uncertain regarding generalization to individuals diagnosed using IFG and HbA_1c.

Brannick and Dagogo-Jack (4) evaluated six prospective studies and identified the following predictors for conversion from prediabetes to type 2 diabetes: 1) baseline IFG and the 2-h GTT glucose values were positively associated with incident diabetes; 2) rates of progression from prediabetes to type 2 diabetes were exponential among study participants in the top quartile of baseline IFG but increased linearly with increasing 2-h oral GTT glucose levels; and 3) incident diabetes occurred at higher rates in populations of color, including Hispanic, Mexican-American, Pima, and Nauruan populations, than among race/ethnicity categories of northern European descent, such as Caucasian. Their review also examined “pitfalls” that could adversely affect the use of HbA_1c as an integrated measure of mean plasma glucose. Their concern focused on racial and ethnic differences in the relationship between blood glucose values and HbA_1c and suggested that HbA_1c levels for the diagnosis of prediabetes be confirmed with fasting and potentially 2-h GTT blood glucose measurements.

An article by Zhu et al. (5) in this issue of Diabetes Care examines whether body weight and cardiometabolic risk factors differed by prediabetes phenotype subgroup in a post hoc analysis of data from the PREVIEW (Prevention of Diabetes Through Lifestyle Interventions and Population Studies in Europe and Around the World) study. Previously published PREVIEW findings indicated that the incidence of diabetes was considerably lower than predicted and did not differ by lifestyle randomization group (6). Thus, in their current evaluation (5), differences in the lifestyle intervention were not considered, and all interventions were combined to evaluate phenotyping using available- and complete-case analysis. Participants (n = 1,510) had BMI ≥25 kg/m² and prediabetes (defined using fasting and GTT). The phenotype subcategorization identified 58% with the isolated IFG phenotype, 6% with the isolated IGT phenotype, and 36% with IFG+IGT. The subcategories, based on HbA_1c, included 73% with normal HbA_1c and 25% with intermediate HbA_1c. The 3-year incidence of diabetes varied considerably by phenotype from 2.6% to 7.9%, which warrants exploring the potential role of phenotyping in predicting and improving the lifestyle intervention outcomes for people with prediabetes.

Table 1 provides an overview of the PREVIEW post hoc analysis, designed to compare participant characteristics and outcomes, and it includes relevant prediabetes phenotype-specific findings from other studies. The article by Zhu et al. (5) provides a more detailed analysis of the phenotypes concerning sustained weight loss and changes in cardiometabolic risk biomarkers over the 3-year study period. A response profile approach with categorical time variable within a linear mixed-effects model framework with individual and site-specific random effects was used.

Study strengths include enrolling a relatively large participant sample from both sexes with a wide age range and from multiple sites. Study weaknesses (noted by the authors) include a high attrition rate at intervention cessation (creating potential selection bias) and the underrepresentation of men and high-risk groups. These weaknesses may limit generalization and translation of the study findings. Some readers could be confused by the study title and may mistakenly assume that the article examines the effects of the randomized lifestyle interventions. However, baseline phenotype is the primary predictor variable that is independent of randomized intervention. Comparing the impact of the type of intervention on outcomes by phenotypes was beyond the scope of the present post hoc analysis.

An “as-treated” analysis, with a gradient of intervention dosage or an adjustment for dosage, could provide some insights into the effects of the intervention by phenotypes on the outcomes studied. The authors fit a baseline-adjusted change score model for some outcomes (e.g., percent weight loss) in the data. However, if phenotypic characteristics are associated with baseline weight, there could be a spurious correlation between phenotypes and percent weight loss when adjusted for baseline weight (7). The statistical power for detecting differences, which may be clinically meaningful, could be increased by the following:

1. Fitting a more efficient semiparametric (spline) model with continuous time, which would maximize the power (8), even though the current models show significance.

2. Using a one-degree-of-freedom contrast test, which would gain more power (8) in testing a group by time interactions of a time trend model instead of a joint omnibus test (e.g., testing changes in percent body weight, systolic blood pressure, and triglycerides).

3. Using a pattern–mixture model or selection model–missing data approach would strengthen the estimates of the corresponding regression models (9).

4. Fitting a parsimonious model with limited variables in a Cox time-dependent model for adjusted diabetes incidence analysis given a small sample size, which may avoid any overfitting issues (10–12).

The PREVIEW study (5,6) post hoc analysis findings suggest that using GTT with the American Diabetes Association prediabetes diagnostic cut point criteria (13) can help advance understanding of the metabolic dysregulation that may distinguish each prediabetes phenotype (14). IGF is primarily characterized by hepatic insulin resistance, whereas IGT is associated with muscle insulin as well as diminished pancreatic β-cell insulin secretory response to consuming glucose as the proxy for dietary carbohydrates. However, little is known about the extent to which biological, psychosocial, and behavioral characteristics may differ by prediabetes diagnostic phenotype.

Priorities for future research include increasing the use of existing data (GTT and HbA_1c) from lifestyle intervention trials to investigate how prediabetes phenotypes are related to intervention outcomes. As noted by Brannick and Dagogo-Jack (4), research is also needed to address how hemoglobin variance may affect the diagnosis and characterization of treatment decisions by phenotype. However, the feasibility of obtaining GTTs and phenotyping participants in clinical and community settings is limited by staff and participant burden.

Randomized or pragmatic trials need to include analysis of the lifestyle intervention “as randomized (intention to treat)” and “as treated” by phenotype. Coupling prediction modeling with more comprehensive predictors could inform the development of useful diagnostic phenotyping tools with good measurement properties. Lessons learned from this research may provide insights into reducing prediabetes misdiagnosis and considering the prediabetes phenotype in the treatment plan in low-resource clinical and community settings in which HbA_1c is the only screening test.