Response Heterogeneity to Lifestyle Intervention among Latino Adolescents

Introduction

Disparities in the prevalence of type 2 diabetes among Latino youth ¹ are the result of interactions between genetic factors and lifestyle behaviors ^{2, 3}. Insulin resistance, or decreased insulin sensitivity, is a central pathological mediator of type 2 diabetes ⁴ that is associated with obesity in adolescents and disproportionately impacts Latino youth as compared to their Black and White counterparts ⁵. If untreated, decreased insulin sensitivity may persist as hyperinsulinemia, glucose intolerance, and exacerbated insulin resistance, thereby increasing the risk for type 2 diabetes ⁶. Lifestyle intervention remains the cornerstone for preventing type 2 diabetes among high-risk adults ⁷, and pediatric studies have demonstrated increases in insulin sensitivity and glucose tolerance following lifestyle intervention ^{8, 9}. However, there is considerable heterogeneity in the response to intervention among youth with obesity ¹⁰ and few studies have rigorously evaluated this phenomenon in the pediatric population.

Precision medicine is an approach for tailoring medical interventions to individual traits to maximize prevention efforts, and this concept has recently been extended to lifestyle interventions ¹¹. The application of precision medicine to pediatric obesity has been limited ¹² but provides an important framework for understanding why some youth experience success with interventions and others do not¹³.

A major impediment to understanding heterogeneity in response to intervention is the need for alternative statistical methodology for use in clinical trials ¹⁴. Structural equation modeling techniques provide a platform for understanding response heterogeneity by identifying “latent classes” of individuals that change similarly over time. Latent class growth models and growth mixture models are commonly used to identify classes of responders and can be applied to interventions ¹⁵. Specifically, covariance pattern mixture modeling (CPMM) demonstrates better performance and circumvents estimation issues of other latent class methods with the sample sizes that are typical in intervention studies ¹⁶. With this context, the purpose of this study is to apply CPMM to characterize response heterogeneity following a 3-month lifestyle intervention among Latino youth with obesity⁸.

Methods

Results

A total of 90 Latino youth (age 15.4±0.9 y, BMI% 98.1±1.5%, female 56.7%) were included in the current analysis.

Table 1 depicts the log-likelihood, SA-BIC, classification likelihood criteria, bootstrapped likelihood ratio test, and entropy (a measure of class separation) for models with different numbers of response phenotypes. All measures suggested three response phenotypes. The 4-class model is not shown since it would not converge likely due to having many parameters relative to sample size. CPMM of three distinct insulin sensitivity response phenotypes exhibited the best fit (Figure 2). Phenotype 1 (PH1) demonstrated the most pronounced increase in insulin sensitivity up to one year follow up (linear 0.52±0.13, p<0.001; quadratic −0.03±0.01, p=0.001). Phenotype 2 (PH2) showed an overall negative but non-significant response that plateaus by one year (linear: −0.18±−0.095, p=0.059; quadratic: 0.01±0.006, p=0.094). Phenotype 3 (PH3) demonstrated significant short term increases in insulin sensitivity before returning to baseline levels by one year follow up (linear 0.08±0.03, p=0.002; quadratic −0.01±0.002, p=0.003). Overall effect sizes from T1 to T4 for PH1, PH2, and PH3 were 2.14, −0.63, and 0.30, respectively. No significant differences were found in program attendance (88.3±10.5%, 78.0±25.7%, 70.5±31.5%, p=0.075) or average heart rates recorded during physical activity sessions (155.4±9.7, 154.8±9.2, 157.5±9.5, beats/min p=0.471), between PH1, PH2, and PH3, respectively, suggesting that adherence to the intervention and fidelity of physical activity dosage was not different between phenotypes. Further, resting heart rate (64.8±3.7, 68.2±2.8, 73.5±2.6 beats/min, p=0.136) and estimated cardiorespiratory fitness (3,021.2±46.0, 2,962±34.5, 3,009.3±31.6 ml/min, p=0.495) after the 3-month lifestyle intervention period were not significantly different across phenotypes PH1, PH2, and PH3, respectively.

Comparison of baseline anthropometrics and type 2 diabetes risk factors across phenotypes are shown in Table 2. Overall, PH3 demonstrated the lowest insulin sensitivity (M±SD, 1.1±0.4) which was significantly lower than PH1 (1.8±1.1, F_2,82=21.5, p=0.028) and PH2 (2.4±1.2, F_2,82=21.5, p<0.001). No differences in insulin sensitivity were found between PH1 and PH2 (F_2,82=21.5, p=0.103). In addition, PH3 showed significantly higher BMI% (98.5±1.1% vs 97.6±1.7%, F_2,87=4.0, p=0.027), 2-hr glucose concentrations during OGTT (144.0±27.5 vs 120.7±18.0 mg/dl, F_2,87=7.8, p=0.001), and lower oral disposition index (4.5±2.8 vs 8.0±5.9, F_2,80=6.6, p=0.002) compared with PH2 with no significant differences between PH1 and PH3. Glucose area under the curve (AUC) was significantly lower in PH2 (15,373.7±1,992.3 mg-h/dl) compared to PH1 (17,337.4±2,619.1 mg-h/dl, F_2,81=9.0, p=0.038) and PH3 (17,692.1±2,380.0 mg-h/dl, F_2,81=9.0, p=<0.001) with no differences between PH1 and PH3. There were no significant differences in age, puberty, sex, exposure to gestational diabetes, or family history of type 2 diabetes across phenotypes. Additional baseline comparisons of cardiometabolic risk factors (lipids, liver enzymes, blood pressure percentiles) are presented in Supplementary Table S1.

Discussion

In order to inform precision approaches for diabetes prevention in high-risk youth, it is important to identify how individuals and/or groups respond to various prevention efforts. Using data from a completed trial that demonstrated significant increases in insulin sensitivity following lifestyle intervention, we were able to identify three distinct response phenotypes over the course of a year. PH1 showed the most robust response; PH2 did not respond to the lifestyle intervention; and, PH3 showed significant modest effects in the short-term but failed to sustain improvements by one year. These results support previous work on response heterogeneity of cardiometabolic risk factors among youth after lifestyle intervention, and further reinforce the need to develop and implement precision approaches for preventing type 2 diabetes among high-risk populations ¹⁰.

The complex pathophysiology of diabetes makes it challenging to identify potential predictors of response to an intervention ²⁷. Physiologically, insulin sensitivity is impacted by numerous factors, including variations in glucose absorption, the incretin effect, insulin secretion, body composition, inflammation and oxidative stress ^28-31. From a behavior standpoint, insulin sensitivity is affected by physical activity and dietary habits which were key behaviors targeted during the intervention ^{32, 33}. Given the multiplicity of factors that influence insulin sensitivity and the comprehensive nature of the intervention, it is impossible to identify whether the observed heterogeneity is the result of biological or behavioral factors. Twin studies in adults have demonstrated that insulin sensitivity is approximately 50% heritable, indicating potential for other factors to explain differential responses to intervention ³⁴. Global microarray analysis profiling of whole blood among Latino adolescents that completed a lifestyle intervention showed significant changes in genetic signatures of the insulin signaling pathway and were amplified by five times in responders versus non-responders of insulin sensitivity ³⁵. Therefore, it is likely that these genetic and epigenetic factors partially explain heterogeneity in response to intervention. Exercise is known to increase gene and protein expression of adiponectin (anti-inflammatory and insulin sensitizer), which is known to activate mitochondrial proteins that have been associated with exercise-induced changes in insulin sensitivity ³⁶. With this context, genetic variants of adiponectin receptor-1 (predominantly found in skeletal muscle) and mitochondrial intermediates, such as PGC1-a ³⁷ and SIRT-1 ³⁸, predicted changes in insulin sensitivity after lifestyle intervention in adults ³⁹. Further, a 20-week endurance training program showed greater improvements in glucose tolerance, insulin secretion, and beta-cell function among PPARG Pro12A1a allele (mitochondrial gene) carriers in sedentary adults without diabetes ⁴⁰. Data from the HART-D study in sedentary adults with type 2 diabetes showed that responders and non-responders of cardiorespiratory fitness to exercise showed no significant differences in changes in glucose regulation ⁴¹. These findings suggest that some individuals, perhaps those with underlying type 2 diabetes, may improve glucose regulation through non-aerobic pathways. Interestingly, the HERITAGE Family Study found that a genomic region around the leptin locus partially explained the insulin response following an exercise intervention ⁴². It is evident that the mechanisms by which heterogeneous responses to lifestyle intervention have yet to be elucidated, and –omics approaches may work in harmony with structural equation modeling techniques to explore response heterogeneity.

Previous studies have identified higher risk phenotypes that have exhibited less favorable responses to lifestyle interventions. In the Tuebingen (TULIP) Study of adults with prediabetes, a higher risk phenotype defined by low insulin sensitivity with non-alcoholic fatty liver disease showed the least improvements in 2-hr glucose concentrations compared to a lower risk phenotype following a 9-month lifestyle intervention ⁴³. The Diabetes Prevention Program (DPP) showed that low insulin secretion and insulin sensitivity at baseline were predictive of greater incidence of diabetes during long-term follow up ⁴⁴. The DPP also demonstrated an attenuated response of insulin sensitivity among individuals with increased visceral and liver fat⁴⁵. Similarly, our analysis revealed a higher risk phenotype, PH3, which was the most insulin resistant at baseline compared to the other phenotypes and only modestly increased insulin sensitivity in the short-term. On the other hand, it is particularly interesting that PH2 did not respond to the intervention given that PH2 started with similar levels of insulin sensitivity as the most robust responders (PH1). The only factor that significantly differentiated PH1 from PH2 at baseline was increased glucose AUC (17,337.4±2,619.1 vs 15,737.7±1,992.3 mg-h/dl, p=0.038, respectively). Thus independent of insulin sensitivity, an increased glucose response to the OGTT may predict pronounced increases in insulin sensitivity following lifestyle intervention among Latino youth with obesity. However, these results should be followed up in larger samples and highlight the need for intensive phenotyping at baseline to identify predictors of response to lifestyle interventions. Additionally, more aggressive and sustained interventions may yield greater physiologic effects. Whether variations in response to lifestyle intervention are explained by distinct biological processes is an interesting notion that warrants future investigation.

Although PH1 exhibited the most favorable response (effect size=2.14), only 17.8% youth in our sample were in this group. Lifestyle intervention remains the cornerstone approach to preventing diabetes in high-risk populations, but the optimal combination of intervention targets (e.g., nutrition, physical activity/inactivity, sleep, etc.) that influence type 2 diabetes risk remain largely unknown. Furthermore, social determinants of health are also operational and perhaps even more so among racial and ethnic minority groups ⁴⁶. As such, ecological factors outside of intervention targets may contribute to response heterogeneity ⁴⁷. Thus, it is important for future studies to consider the sociocultural context of the priority population and consider a more comprehensive array of factors that may help predict response to lifestyle intervention.

The current study advances the science by identifying distinct response phenotypes of an important type 2 diabetes biomarker among Latino adolescents with obesity following lifestyle intervention. We used robust statistical methods (i.e., CPMM) to analyze longitudinal data from youth enrolled in a lifestyle intervention as part of a randomized control trial, which is novel to the field of pediatric obesity. CPMMs circumvent issues that other latent class analyses do not; therefore, this study introduces an appealing analytical technique to the field of pediatric obesity for examining heterogeneity of longitudinal data. Further, our study focuses on a vulnerable population at high risk of developing type 2 diabetes ⁴⁸. We acknowledge that the current study is not without limitations. Insulin sensitivity was the primary physiologic outcome in the parent study and was the sole parameter of analysis in the current study. This outcome measure was used as the primary outcome of the study since it is proximally related to T2D risk and regression over time. Insulin sensitivity was estimated using a surrogate measure (WBISI); however, this measure has been validated in both youth with obesity and adults with the gold-standard hyperinsulinemic euglycemic clamp ^{21, 22}. Our sample size for each phenotype was relatively small which may have influenced study results, and a larger sample size may more clearly illustrate differences in risk profiles across phenotypes. Further, physical activity and eating patterns were not included in the analysis. Thus, it remains plausible that changes in physical activity and nutrition explained the observed heterogeneity. Lastly, the current study included a distinct population of Latino youth with obesity, which limits generalizability.

A 3-month diabetes prevention program that included physical activity, nutrition education, and behavioral skills training induced a heterogeneous response in terms of changes in insulin sensitivity among Latino adolescents with obesity. The insulin response to glucose among Latino youth with obesity is variable with some youth showing a more malleable response to lifestyle intervention than others . Future research is warranted to understand the physiologic, genetic, psychosocial, environmental, and behavioral predictors of response to lifestyle intervention to inform precision medicine practitioners that serve high-risk youth populations.

Supplementary Material