Food Preference Patterns Associated with Body Mass Index: Findings from a Nationwide Web-Based Survey in Japan

1. Introduction

Obesity and its related disorders, including type 2 diabetes and cardiovascular diseases, constitute major public health challenges worldwide [1]. Obesity is driven by a complex interplay among biological susceptibility [2], social and environmental factors [3], and individual eating behaviors [4], among which food preference, defined here as the desire to eat particular foods, may serve as a key proximal determinant of food choice and energy intake.

Several instruments have been developed to assess food preferences, including the Leeds Food Preference Questionnaire [5] and the Macronutrient and Taste Preference Ranking Task [6], which has also been adapted for use in Asian populations. However, these questionnaires were not specifically designed to capture preferences for foods commonly consumed in Japan. Therefore, we recently developed the Japan Food Preference Questionnaire (JFPQ) as a clinically oriented, food-image-based questionnaire tailored to foods typically addressed in nutritional counseling for individuals with obesity and diabetes. The JFPQ classifies these foods into macronutrient- and sweetness-based categories and quantitatively assesses individual-level biases in food preferences [7]. In our previous small-scale study using the JFPQ, individuals with abdominal obesity showed stronger preferences for carbohydrate- and fat-rich foods than healthy volunteers, and these preference scores were associated with corresponding dietary intake assessed using a food frequency questionnaire (FFQ) [7]. These findings suggest that obesity may be associated with specific patterns of food preference, which may contribute to actual eating behavior linked to weight gain. However, obesity-related food preference characteristics at the population level remain unclear.

In this study, we examined the associations between JFPQ-derived scores and body mass index (BMI) in a large Japanese cohort, aiming to identify food preference characteristics associated with obesity.

2. Materials and Methods

2.1. Study Design and Participants

In this cross-sectional study, we conducted an anonymous, nationwide web-based survey from November 1 to November 5, 2024, using a framework adapted from previously reported methodology [8]. Individuals preregistered with a commercial panel were recruited using quota sampling to obtain 20,000 respondents aged 20 to 69 years from the general Japanese population. Quotas for age group, gender, and region of residence (Hokkaido/Tohoku, Kanto, Chubu, Kinki, and Chugoku/Shikoku/Kyushu/Okinawa) were set to reflect the population distribution of the 2020 Japanese census. Self-reported information, including gender, age, height, body weight, time since the last meal (<1, 1–2, 2–3, 3–4, 4–5, or ≥5 hours), and responses to the JFPQ, was collected through an online questionnaire.

Individuals with missing data on height or weight were excluded (n = 2,805; 1,107 men and 1,698 women). In addition, respondents who assigned identical preference ratings to all food items were excluded to enhance data quality (n = 1,700; 946 men and 754 women), because these response patterns were considered to indicate low response reliability. We defined obesity as a BMI ≥25.0 kg/m² based on the Japan Society for the Study of Obesity criteria, given the greater susceptibility of Japanese individuals to obesity-related diseases at lower BMI levels [9].

2.2. Assessment of Food Preference

Food preferences were assessed using the JFPQ, which comprises 25 food items commonly consumed in Japan and frequently addressed in nutritional counseling [7]. Participants rated their current desire to eat each food item using a 0–10 visual analogue scale with food images presented in randomized order to minimize order effects (Figure S1). As shown in Figure S1B, food items were categorized a priori into four macronutrient groups using predefined nutritional criteria: carbohydrate (≥70% of total energy derived from carbohydrates), fat (≥40% of total energy derived from fat), protein (≥30% of total energy derived from protein), and dietary fiber (≥1 g of fiber per 100 g and <100 kcal per 100 g of edible portion). The carbohydrate and fat groups were further subdivided into non-sweet (<10 g of sugars per 100 g of edible portion) or sweet (≥10 g of sugars per 100 g of edible portion) categories according to total sugar content (Figure S1). Preference scores for each nutrient category were calculated as a percentage of the maximum possible score, with higher values indicating a stronger preference [7].

2.3. Statistical Analysis

All of the values are presented as the means ± standard deviations (SDs) or the number of subjects (%). Differences in preference scores by gender were assessed using the Wilcoxon rank–sum test. Differences in preference scores across age categories and time-since-last-meal categories were assessed using the Kruskal–Wallis test, and trends across ordered categories were evaluated using the Jonckheere–Terpstra test. Multiple linear regression analyses were conducted separately by gender, with BMI as the dependent variable and each preference score as the independent variable, adjusting for age and time since the last meal. To account for multiple comparisons, P values were adjusted using the Benjamini–Hochberg false discovery rate procedure within each predefined family of tests, including macronutrient categories and individual food items. Differences in food preference scores according to BMI category (18.5 to <25.0 kg/m² vs. ≥25.0 kg/m²) and time since the last meal (<3 h vs. ≥3 h) were assessed using two-way ANCOVA adjusted for age and gender, with BMI category, time since the last meal, and their interaction included as fixed factors. Least-squares means and 95% confidence intervals (CIs) were estimated from the ANCOVA models and are presented in the corresponding figures. Post hoc simple main-effect comparisons were performed to compare time-since-last-meal groups within each BMI category and BMI categories within each time-since-last-meal stratum, with P values adjusted using the Holm method. For all statistical analyses, two-sided P values <0.05 were considered statistically significant (JMP version 18; SAS Institute Inc.).

3. Results

3.1. Characteristics of the Study Participants and Food Preference Patterns Among Normal-Weight Individuals

The participant flow for the present study is shown in Figure S2, and Table 1 summarizes the characteristics of the 15,495 participants included in the final analysis (7,971 men and 7,524 women). The mean BMI was 23.1 ± 3.8 kg/m² in men and 20.8 ± 3.5 kg/m² in women. By BMI category, 554 men (3.6% of the total study population) and 1,848 women (11.9%) had a BMI <18.5 kg/m², 5,433 men (35.1%) and 4,898 women (31.6%) had a BMI of 18.5 to <25.0 kg/m², 1,624 men (10.5%) and 594 women (3.8%) had a BMI of 25.0 to <30.0 kg/m², and 360 men (2.3%) and 184 women (1.2%) had a BMI ≥30.0 kg/m².

Among normal-weight individuals (BMI 18.5 to <25.0 kg/m²), food preference scores differed significantly between men and women (Table S1). Women had higher preference scores for sweet fat and dietary fiber groups, whereas men had significantly higher scores for all other food categories. Significant age-related differences in food preferences were also observed across most nutrient groups (Table S2). Overall, older age was associated with higher preference scores for dietary fiber and lower preference scores for sweet carbohydrate and both non-sweet and sweet fat in both men and women. Time since the last meal was also associated with differences in preference scores across all nutrient groups (Table S3). A longer time since the last meal was associated with higher preference scores for non-sweet carbohydrate and protein in both men and women, lower preference scores for sweet carbohydrate and sweet fat in men, and higher preference scores for dietary fiber in women.

3.2. Associations Between BMI and Food Preference Scores

In the initial analysis, individuals with a BMI <18.5 kg/m² were excluded to focus on obesity-related food preference patterns across the normal-to-higher BMI range. The standardized β coefficients for the associations between BMI and macronutrient-based preference scores adjusted for age and time since the last meal are shown in Figure 1. In both men (Figure 1A) and women (Figure 1B), BMI was positively associated with the fat group, particularly non-sweet fat, and inversely associated with the dietary fiber group.

Figure 2 shows the associations between preference scores for individual food items and BMI in men (Figure 2A) and women (Figure 2B). Higher BMI was associated with higher preference scores for several non-sweet fat items, including deep-fried foods, hamburgers, and high-fat ramen noodles, in both genders. Among the sweet carbohydrate items, only soft drinks (juice) were positively associated with BMI in both genders. In contrast, higher BMI was associated with lower preference scores for multiple types of dietary fiber–rich foods, particularly among women. Notably, the preference for vegetables showed the strongest inverse association with BMI for both genders.

3.3. Differences in Food Preference Scores According to Obesity Status and Meal Timing

Next, we compared the macronutrient-based preference scores between individuals with a BMI ranging from 18.5 to <25.0 kg/m² (normal weight) and those with a BMI ≥25.0 kg/m² (obesity), stratified by the time since the last meal and adjusted for age and gender. For this analysis, time since the last meal was dichotomized at 3 h, approximately the median value in the study population, into <3 h (normal weight, n = 5,253; obesity, n = 1,326) and ≥3 h (normal weight, n = 5,078; obesity, n = 1,436). This cutoff was also supported by physiological evidence that postprandial glycemic dips occurring 2–3 h after a meal predict subsequent hunger and energy intake, indicating a transition from postprandial satiety to re-emerging appetite [10]. As shown in Figure 3, the interaction between obesity status, defined by BMI category (18.5 to <25.0 vs. ≥25.0 kg/m²), and dichotomized time since the last meal (<3 h vs. ≥3 h) was significant for all nutrient groups. In normal-weight individuals, non-sweet carbohydrate, non-sweet fat, and protein scores were lower at <3 h than at ≥3 h, whereas no such within-group differences were observed in individuals with obesity (Figure 3A, 3C, 3E). At <3 h, these scores were higher in individuals with obesity than in normal-weight individuals, with no differences at ≥3 h. In contrast, sweet carbohydrate (Figure 3B) and sweet fat (Figure 3D) scores were lower at ≥3 h than at <3 h in both BMI categories but were higher in individuals with obesity at <3 h. In terms of dietary fiber, scores were lower in individuals with obesity than in normal-weight individuals at ≥3 h (Figure 3F).

3.4. Associations Between BMI and Food Preference Scores in Women with BMI <25.0 kg/m²

Given that a substantial proportion of women in our cohort had a BMI <18.5 kg/m² (n = 1,848; 24.6% of women), and in light of growing attention in Japan to health issues related to low BMI, recently conceptualized as Female Underweight/Undernutrition Syndrome (FUS) [11], we additionally examined the associations between BMI and food preference scores among women with a BMI <25.0 kg/m², focusing on normal-weight and underweight individuals. The positive association between BMI and preference scores for the fat group, particularly non-sweet fat, was consistent with that observed in the initial analysis of individuals with a BMI ≥18.5 kg/m², whereas no significant association was observed for dietary fiber (Figure S3A and S3B). Notably, among women with a BMI <25.0 kg/m², BMI was also positively associated with the non-sweet carbohydrate group (Figure S3A), particularly with preference for white rice and udon/soba noodles at the individual food-item level (Figure S3B), suggesting that women with lower BMI had lower preferences for these staple carbohydrate foods commonly consumed in Japan.

4. Discussion

In this large-scale web-based survey, we demonstrated that higher BMI was associated with a greater preference for fat, particularly non-sweet fat, and a lower preference for dietary fiber, especially vegetables. Moreover, whereas macronutrient-based food preference scores differed according to time since the last meal in normal-weight individuals, such differences were attenuated in individuals with obesity, who showed higher preference scores even within 3 h after a meal.

Several interpretations of our findings merit consideration. First, the pronounced preference for non-sweet fatty foods among individuals with higher BMI is consistent with evidence that savory, energy-dense foods promote excess energy intake in modern food environments [12]. The item-level patterns (e.g., deep-fried foods, hamburgers, and high-fat ramen noodles versus vegetables) suggest a gradient by energy density rather than a uniform shift across macronutrient categories. Furthermore, fat-rich foods may strongly activate reward-related neural pathways, reinforcing palatability-driven intake and potentially creating a feedback loop between food preference and eating behavior [13,14]. Within the sweet carbohydrate group, only preference for soft drinks (juice) was positively associated with BMI in both genders. This observation may be attributable to the widespread accessibility of soft drinks, which can facilitate the translation of preference into actual intake. In addition, liquid carbohydrates may produce weaker satiety than solid foods, partly because of the absence of mastication and more rapid gastric emptying, thereby attenuating compensatory reductions in subsequent energy intake [15]. Furthermore, a dose-response meta-analysis of prospective cohort studies reported that greater consumption of sugar-sweetened beverages was associated with a higher risk of obesity, type 2 diabetes, and all-cause mortality [16]. Collectively, these findings suggest that preference for soft drinks should be considered with caution, as it may contribute to weight gain through both environmental and physiological mechanisms that promote overconsumption and a positive energy balance.

It should be noted that the JFPQ assesses food preference rather than actual dietary habits. Although dietary intake was not assessed in the present study, food preference may not necessarily correspond to actual intake, as food choices and the amount consumed can be influenced by factors such as health consciousness, self-restraint, and food availability. In our previous study, JFPQ-based preference scores for carbohydrate- and fat-rich foods were associated with dietary intake assessed using an FFQ among individuals with abdominal obesity, whereas such associations were limited in healthy volunteers, in whom dietary intake was associated only with preference scores for the dietary fiber group [7]. Thus, the associations between BMI and JFPQ-quantified food preferences observed in the present study may be meaningful because these findings suggest that obesity is linked to subjective desire to eat specific foods per se, which may be particularly relevant to obesity-related eating behavior.

Among normal-weight individuals, preference scores for non-sweet carbohydrates, non-sweet fats, and protein were significantly lower in the postprandial state (<3 h) than in the fasting-like state (≥3 h), likely reflecting satiety and a transient reduction in cravings for savory and energy-dense foods after a meal. Notably, such differences in preference scores according to time since the last meal were attenuated among individuals with obesity, and obesity-associated differences in preference scores were more pronounced within 3 h after a meal across all nutrient-based categories except dietary fiber. These findings suggest that obesity may be characterized by altered feeding-state-related patterns of macronutrient preference. Given that the period 2–3 h after a meal represents a transitional phase during which postprandial satiety may begin to decline and appetite may re-emerge [10], the higher preference scores observed in individuals with obesity even within 3 h after a meal may reflect impaired postprandial suppression of food desire. This pattern might reflect disrupted satiety signaling, including reduced glucagon-like peptide-1 (GLP-1) responsiveness [17] and impaired postprandial ghrelin regulation [18,19]. In addition, obesity-associated eating behaviors, such as rapid eating, as reported in previous studies [8,20], may further impair postprandial satiation and thereby contribute to increased subsequent food intake [21].

The inverse association between BMI and preference scores for the dietary fiber group, particularly vegetables, is noteworthy. Previous epidemiological studies have shown that lower intake of dietary fiber and vegetables is associated with obesity and weight gain [22,23]. In general, this association may be partly explained by social and environmental factors, including socioeconomic disadvantage, which can limit access to healthier foods such as vegetables and fruits while promoting greater consumption of lower-cost, energy-dense foods [24,25]. However, an important finding of the present study is that individuals with obesity showed a lower preference for these dietary fiber-rich foods themselves, highlighting an association at the level of desire to eat. Moreover, preference for the dietary fiber group was significantly lower in individuals with obesity than in normal-weight individuals, especially when more than 3 hours had elapsed since the last meal. This pattern may indicate that a lower preference for low-energy-density, fiber-rich foods becomes more evident under conditions in which appetite is more likely to re-emerge, potentially shifting food preference toward more energy-dense alternatives and contributing to excess energy intake. Importantly, low intake of dietary fiber-rich foods, such as vegetables, has been suggested to increase obesity risk through alterations in the gut microbiota composition [26], which may reduce short-chain fatty acid production and disrupt the gut–brain signaling involved in appetite regulation and food reward [27,28]. In addition, preference for vegetables can be shaped not only by innate taste predispositions but also by repeated exposure and the home food environment early in life [29,30,31]. Thus, lower preferences for vegetables and dietary fiber may reflect the cumulative effects of insufficient consumption of these foods during childhood on later food preference formation. Further studies are needed to examine the associations of social determinants of health, including educational attainment and household income, with food preferences and obesity.

The additional analysis among women with a BMI <25.0 kg/m² also provides exploratory insight into low BMI in women, which has recently received increasing attention in Japan as　FUS. FUS has been proposed as a health concept encompassing diverse physical and psychological disorders related to underweight or undernutrition, especially in premenopausal women, including low muscle mass and reduced muscle strength, menstrual irregularities, abnormalities in bone metabolism, eating disorders, and depressive symptoms [11]. Notably, in the present cohort, nearly one in four women had a BMI <18.5 kg/m², underscoring the relevance of this issue in Japanese women. In this context, lower preference for staple carbohydrate foods commonly consumed in Japan, such as rice and udon/soba noodles, may represent a distinctive food preference pattern associated with lower BMI in women. A similar pattern was not observed among men with a BMI <25.0 kg/m² (data not shown). This interpretation is consistent with previous observations in young Japanese women showing that underweight women with a desire for thinness consumed less cereal and rice than normal-weight women [32], suggesting that avoidance of staple carbohydrate foods may characterize a subgroup of women with low BMI and potentially FUS-related undernutrition. Further studies are needed to clarify whether reduced preference for staple non-sweet carbohydrate foods contributes to underweight and undernutrition or reflects broader dietary, psychosocial, or health-related factors in women.

Our findings should be interpreted with caution. First, the cross-sectional design precludes causal inference, and the self-reported nature of the data may have introduced reporting bias, including social desirability bias, as well as misclassification. Second, the web-based sampling approach may have contributed to selection bias. Third, although BMI was significantly associated with multiple food preference scores, the standardized β coefficients were modest in magnitude. Nevertheless, even small effect sizes may be informative at the population level, particularly for multifactorial traits such as obesity. Moreover, despite the inherently subjective nature and potential variability of self-reported preference scores, the consistent associations observed across related food categories support the robustness of the findings.

5. Conclusions

Findings from our large cohort of the Japanese population demonstrate that obesity is associated with distinctive food preference patterns, that is, patterns of desire to consume specific foods, which are potentially linked to eating behavior. Our findings highlight food preference profiling as a valuable approach to guide personalized nutritional counseling for weight management and to identify individuals at risk of future weight gain, with potential implications for preventing obesity-related adverse health outcomes.