Eating behavior dimensions and 9-year weight loss maintenance: a sub-study of the Finnish Diabetes prevention study

The study addresses why many individuals fail to maintain weight loss after lifestyle interventions aimed at preventing type 2 diabetes mellitus (T2DM). Psychological dimensions of eating behavior—cognitive restraint of eating (including flexible and rigid components), disinhibition, and susceptibility to hunger—may be modifiable mediators influencing dietary choices and energy intake. Prior evidence links increased cognitive restraint, especially flexible restraint, to better short- and long-term weight outcomes, but data in people with impaired glucose tolerance (IGT) and long-term mediation evidence are limited. This sub-study of the Finnish Diabetes Prevention Study (DPS) investigated whether a lifestyle intervention affects eating behavior dimensions in adults with IGT and whether early (first-year) changes in these dimensions mediate long-term (9-year) changes in body weight.

Previous longitudinal interventions show increased cognitive restraint associates with successful weight loss and maintenance in people with overweight/obesity. Both rigid and flexible restraint have been linked to weight loss, with flexible restraint often showing stronger, more favorable associations. Reductions in disinhibition and susceptibility to hunger predict short-term weight loss. In IGT populations, the DELIGHT trial suggested that enhancing flexible control and reducing disinhibition benefited central adiposity and glycemia over 1 year. A DPP sub-study found that increased DEBQ-measured dietary restraint predicted better long-term weight loss (~2.8 years). Only one prior study explicitly tested eating behavior dimensions as mediators, finding flexible (but not rigid) restraint mediated 24-month weight outcomes in women with overweight/obesity after a 1-year program. Measures differ in what aspects of restraint they capture; TFEQ typically identifies individuals more successful at sustained control and weight management than the Restraint Scale. Thus, understanding which restraint components change during intervention and how they relate to long-term weight outcomes, especially in IGT, fills a gap.

Design: Sub-study of the Finnish Diabetes Prevention Study (DPS), a multicenter randomized controlled trial (parallel design) in individuals with IGT. This Kuopio center sub-study included 98 participants (38 men, 60 women) randomized to intensive lifestyle intervention (n = 51) or control (n = 47). Recruitment: Nov 1993–Jun 1997; follow-up spanned 9 years from baseline (through Nov 2006). Ethics approvals obtained; informed consent at baseline and post-intervention. Inclusion criteria: age 40–64 years, BMI > 25 kg/m², IGT based on mean of two 75 g OGTTs (WHO 1985). Baseline mean (SD): age 53.6 (7.4) years; BMI 31.3 (4.6) kg/m². Randomization stratified by center, sex, and baseline 2-h plasma glucose. Attrition: 24 did not complete 9 years; non-completers did not differ from completers on baseline characteristics. Intervention: Goals—≥5% weight loss from baseline; dietary targets: total fat <30% energy, saturated fat <10% energy, fiber ≥3.6 g/MJ (15 g/1000 kcal); physical activity ≥30 min/day moderate intensity. Delivery—seven individualized, 30–60 min sessions with a nutritionist in year 1; quarterly sessions thereafter for 3–6 years (median intervention duration 5 years in Kuopio). Counseling included tailored dietary feedback based on 3-day food records (four times yearly), education (risk factors, saturated fat, fiber, PA), problem-solving, goal setting, self-monitoring, weight tracking, and optional group sessions/lectures/contacts. Three participants received VLCD for 2–5 weeks. Physical activity guidance included optional supervised, tailored resistance training. Control: One baseline session (individual or group, 30–60 min) with general verbal and written lifestyle advice (weight reduction, PA increase, diet quality) without individualized counseling. Post-intervention follow-up: Annual nurse visits for all participants with same measurements as during intervention; no detailed diet/exercise counseling. Measures: Eating behavior via the Three Factor Eating Questionnaire (TFEQ; 51 items) yearly, producing scores for cognitive restraint (21 items), disinhibition (16), susceptibility to hunger (14). Cognitive restraint subdivided into flexible (7 items) and rigid (7) per Westenhoefer. Higher scores indicate greater levels of each construct. Internal consistency: total cognitive restraint α = 0.86, flexible α = 0.74, rigid α = 0.66, disinhibition α = 0.75, hunger α = 0.76. Body weight measured annually in light clothing to nearest 100 g. Statistical analysis: Descriptive stats; group comparisons via Mann–Whitney U, chi-square; correlations between 1-year changes in eating behavior and 9-year weight change via Pearson correlations. Primary longitudinal analyses used structural equation modeling (Mplus 8.4) with full information maximum likelihood (FIML) and robust SEs, testing Group × Time latent change scores (Wald tests) for TFEQ scales and weight. Effect sizes (Cohen’s d) computed for between- and within-group changes (year 1 and year 9), with thresholds: small ≥0.2, moderate ≥0.5, large ≥0.8. Mediation: Simple mediation SEMs tested whether 1-year changes in total, flexible, or rigid restraint mediated the effect of intervention (vs control) on 9-year weight change, controlling for sex, age, baseline restraint, and baseline weight. Indirect effects evaluated with bias-corrected bootstrap 95% CIs (5000 resamples); model fit assessed by χ² (p > 0.05), CFI ≥ 0.95, SRMR ≤ 0.08.

- Weight outcomes: First-year weight loss was greater in the intervention vs control group (−5.2 vs −1.2 kg; p < 0.001). At 9 years, the intervention group maintained greater weight loss (−3.0 kg) compared with control (+0.1 kg; p = 0.046). A significant time × group interaction for weight from baseline to year 9 (Wald χ² = 17.164, p = 0.046) supported differential trajectories. - Proportion achieving ≥5% weight loss at 9 years: 43.6% intervention vs 23.5% control (p = 0.07). - Eating behavior changes: Significant time × group interactions from baseline to 9 years for total cognitive restraint (p = 0.002) and rigid restraint (p = 0.004); flexible restraint showed a trend (p = 0.093). First-year increases favored intervention: total restraint +4.6 vs +1.7 points (p < 0.001), flexible +1.7 vs +0.9 (p = 0.018), rigid +1.6 vs +0.5 (p = 0.001). At 9 years, between-group differences persisted: total restraint +2.6 vs +0.1 (p = 0.002), rigid +1.0 vs +0.4 (p = 0.004). No significant time × group interaction for disinhibition (p = 0.48); hunger showed a borderline trend (p = 0.05), with negligible between-group effects at years 1 and 9. - Correlations: 1-year increases in restraint inversely correlated with 9-year weight change: total restraint r = −0.34 (p = 0.004), rigid r = −0.32 (p = 0.007), flexible r = −0.23 (p = 0.058). - Mediation (indirect effects significant with 95% CIs excluding zero): • Total cognitive restraint: indirect effect = 1.406 kg (95% CI 0.380, 3.194); direct effect c′ = −0.489 kg (p = 0.008); total effect c = 2.517 kg (p = 0.046). Model fit: χ² = 3.890, df = 2, p = 0.143; CFI = 0.992; SRMR = 0.042. • Flexible restraint: indirect effect = 0.590 kg (95% CI 0.002, 2.064); direct effect c′ = −0.817 kg (p = 0.057); total effect c = 2.606 kg (p = 0.041). Model fit: χ² = 1.334, df = 2, p = 0.513; CFI = 1.000; SRMR = 0.022. • Rigid restraint: indirect effect = 1.025 kg (95% CI 0.317, 2.298); direct effect c′ = −1.038 kg (p = 0.008); total effect c = 2.313 kg (p = 0.074). Model fit: χ² = 2.893, df = 2, p = 0.235; CFI = 0.996; SRMR = 0.039. - Effect sizes: Between-group effect sizes favored intervention for restraint measures at years 1 and 9 (e.g., total restraint d ≈ 0.64 at year 1; 0.61 at year 9) and for weight (small d). Within-group changes in restraint were large at year 1 and moderate at year 9 in the intervention group, but small/negligible in controls.

An intensive, individualized lifestyle intervention led to substantial early increases in cognitive restraint (total, flexible, rigid) and weight loss, with differences persisting up to nine years. Early changes in restraint statistically mediated long-term weight outcomes, suggesting that enhancing cognitive control over eating during the initial intervention phase contributes to durable weight loss maintenance in people with IGT. Findings align with prior work linking increased restraint—particularly flexible restraint—to better weight outcomes and extend evidence by demonstrating long-term mediation effects over nine years in an IGT cohort. Disinhibition and hunger did not show meaningful between-group differences, indicating restraint changes were the primary psychological pathway captured here. The authors note that not all forms of restraint are beneficial; TFEQ-identified restraint is associated with successful, health-motivated control, contrasting with other restraint measures. The intervention’s components—goal setting, self-monitoring, problem solving, tailored feedback, and optional supervised activity—likely enhanced self-regulation skills, which may underlie improvements in cognitive restraint and, in turn, weight maintenance. The results also resonate with broader DPS findings of sustained diabetes risk reduction and cardiometabolic benefits. Future programs should integrate lifestyle changes with psychological skill training (self-regulation, motivation, self-efficacy) and leverage technology while ensuring accessibility for vulnerable groups to avoid widening health disparities.

Intensive, individually tailored professional lifestyle counselling in adults with overweight and IGT produced durable increases in cognitive restraint of eating and sustained weight loss over nine years. Mediation analyses indicate that early increases in cognitive restraint (total, flexible, rigid) play a role in long-term weight loss maintenance, potentially contributing to reduced T2DM risk. Future research should identify and optimize specific behavior change techniques that strengthen self-regulation, test generalizability across ages and populations, and evaluate scalable, inclusive delivery methods (e.g., digital health) that remain acceptable and effective in lower socioeconomic groups.

- Modest sample size and varying intervention duration due to staggered recruitment and early termination of intervention phase. - Volunteer sample likely more health-conscious and motivated than the general population, limiting generalizability. - Predominantly middle-aged participants; results may differ in younger or older populations. - In mediation models, changes in eating behavior and weight partially overlapped temporally (first-year mediator vs nine-year outcome), leaving potential for bidirectionality (weight change may also influence restraint changes).