Sex-Specific and Academic Program-Stratified Interplay of Anthropometric and Psychosocial Predictors of BMI Among University Students: A Multiple Linear Regression and ROC Curve Analysis

1. Introduction

Obesity represents one of the most severe public health challenges of the 21st century, characterized by a steadily increasing prevalence among young adults and associated with major cardiometabolic risks, developmental impairments, and diminished quality of life [1]. Although Body Mass Index (BMI) remains the universal clinical standard for classifying weight status, it possesses well-documented biological limitations. Specifically, BMI fails to differentiate between lean muscle mass and adipose tissue, nor does it accurately reflect the regional distribution of fat tissue [2]. Defined fundamentally as a disproportionate and excessive accumulation of body fat relative to height, age, and sex, obesity poses a direct threat to health, driving the development of metabolic syndrome and insulin resistance [3,4].

Previous literature highlights that university students often experience more pronounced weight gain compared to their non-academic peers [5]. Consequently, modern anthropometry seeks more reliable, rapid, and non-invasive alternative or complementary screening tools, such as Body Fat Percentage (BFP), Waist-to-Height Ratio (WHtR), Waist-to-Hip Ratio (WHR), and Body Adiposity Index (BAI) [6,7]. Integrating these markers within a sex-stratified approach is crucial given the pronounced sexual dimorphism and constitutional variations in fat distribution among young adults [8]. Furthermore, stratifying these indices across distinct academic programs offers valuable epidemiological insights [9].

While there are numerous modalities to quantify fat excess and its bodily distribution, establishing simple, low-cost, and non-invasive anthropometric indicators is essential [10]. Such tools enable robust clinical and epidemiological screening across diverse populations, simplify the identification of individuals at high risk for morbidity and mortality, and help determine priorities for both individual and community-based interventions regardless of race or country [11,12,13]. Despite racial, ethnic, and environmental variations in the risks associated with specific adiposity distributions, a global consensus on obesity classification remains vital [14]. Currently, however, this unification relies almost exclusively on BMI and waist circumference [15,16].

Importantly, a comprehensive understanding of obesity cannot be confined solely to its physical or metabolic dimensions [17]. Young adults frequently navigate intense social pressures regarding body image, which can trigger psychosocial vulnerabilities. Nonetheless, current literature yields conflicting results regarding the degree to which obesity directly impairs overall well-being [18,19,20]. Finding precise, non-invasive anthropometric predictors is therefore essential—not only because they are highly cost-effective and scalable, but also because they minimize the psychological burden and anxiety often associated with more invasive clinical evaluations.

The rationale of the present study lies in the necessity to simultaneously evaluate the diagnostic efficacy of modern anthropometric indicators relative to BMI, while objectively investigating whether weight status exerts a direct predictive influence on the psychosocial dimension of young adults. By leveraging the Satisfaction with Life Scale (SWLS) and the Multidimensional Scale of Perceived Social Support (MSPSS), this integrated approach aims to optimize clinical screening tools while actively destigmatizing the psychological stereotypes routinely attributed to individuals with excess body weight.

To address these gaps in the literature, the present study was structured around three primary objectives. First, we aimed to identify biological and anthropometric variations among students from the University of Oradea across three distinct academic specializations: Biomedical Sciences (BS), Computer Science and Engineering (CSE), and Social Sciences and Physiotherapy (SSP). Second, we sought to investigate sex-based differences regarding a comprehensive panel of health-related anthropometric indicators, including BAI, WHtR, WHR, BMI, and BFP—a specific combination that remains underrepresented in current student-population literature. Third, this study adopts a bio-psychosocial approach to analyze how these biometric indices, alongside psychosocial factors (life satisfaction and perceived social support), act as predictors of BMI, determining which variables carry the highest experimental and practical relevance for this cohort.

To achieve these objectives, the following specific research hypotheses were formulated and tested:

• Hypothesis 1 (H1): Body composition indicators and regional anthropometric indices (BFP, WHtR, BAI, WHR) exhibit a highly statistically significant predictive capacity (p < 0.05) in identifying BMI-defined obesity across both sexes and distinct academic programs.
• Hypothesis 2 (H2): Anthropometric indicators that evaluate central adiposity (such as WHtR) demonstrate superior diagnostic accuracy (higher AUC) compared to the traditional waist-to-hip ratio (WHR), showing greater predictive stability across sexes.
• Hypothesis 3 (H3): The optimal cut-off thresholds for body composition markers (BFP, BAI) and regional indices (WHtR, WHR) vary significantly by sex, directly reflecting natural biological dimorphism.
• Hypothesis 4 (H4): The psychosocial dimension, evaluated through SWLS scores, possesses significant predictive value regarding weight status, with obese subjects displaying predictably lower well-being scores.
• Hypothesis 5 (H5): MSPSS scores exert a statistically valid discriminative capacity regarding the presence or absence of obesity in both sexes, suggesting a direct dynamic relationship between social support networks and body mass index.

2. Materials and Methods

2.1. Participants

The total sample consisted of 445 university students from the University of Oradea, aged between 20 and 25 years. Participants were enrolled across three distinct academic programs: Biomedical Sciences (BS) (N = 144), Computer Science and Engineering (CSE) (N = 159), and Social Sciences and Physiotherapy (SSP) (N = 142). Regarding sex distribution, the cohort included 199 men (44.7%) and 246 women (55.3%). All subjects participated voluntarily and provided informed consent prior to data collection.

2.2. Psychosocial Measures

Satisfaction with Life Scale (SWLS): The Romanian version of the SWLS, originally developed by Diener et al. [21], was utilized to evaluate subjective well-being. This instrument conceptualizes life satisfaction as a conscious cognitive judgment, allowing individuals to evaluate their life circumstances against self-imposed criteria. The scale comprises five items rated on a 7-point Likert scale, where higher scores indicate greater life satisfaction. The SWLS has consistently demonstrated robust reliability and validity across extensive psychological research [22].

The Multidimensional Scale of Perceived Social Support (MSPSS): Perceived social support was assessed using the 12-item MSPSS [23]. This instrument captures a three-factor structure addressing specific sources of social support: family, friends, and significant others. Prior validation studies support high Cronbach’s alpha coefficients both for the individual subscales and the global score representing overall perceived social support [24].

2.3. Anthropometric Measurements and Working Procedure

To determine the target anthropometric indices (BMI, BAI, WHtR, WHR, and BFP), the following direct structural measurements were performed: standing height (H), body weight (W), waist circumference (WC), and hip circumference (HC).

Height was measured with an ADE wall taliometer with 1 mm precision (ADE® GmbH, Germany) with individuals lightly dressed, without shoes, standing erect, back straight, heels together and with feet slightly spread. Weight was measured with a digital device using bioelectrical impedance analysis (Omron BF-511; Omron Healthcare Co., Ltd., Kyoto, Japan).

WC and HC were measured using a flexible, non-stretchable anthropometric tape. WC was recorded in centimeters at the midpoint between the lower border of the rib cage and the iliac crest along the midaxillary line, with participants standing erect and measured at the end of a normal expiration. HC was measured in centimeters as the maximum circumference over the buttocks at the level of the widest diameter around the gluteal protuberance.

Body Mass Index (BMI) is the official clinical indicator utilized to evaluate whether an adult presents a healthy weight relative to height. This index is calculated by dividing body weight in kilograms by the square of height in meters [25]. The BMI metric provides a straightforward and practical approach to classify weight status, where a high value (BMI ≥ 30 kg/m²) defines obesity—a state closely linked to multiple metabolic comorbidities and increased mortality rates [26]. The calculation formula is as follows:

$$BMI={\displaystyle \frac{Weight \left(kg\right)}{Height{ \left(m\right)}^2}}$$

Bergman et al. proposed a new anthropometric measurement in 2011, known as the body adiposity index (BAI), in order to evaluate body composition by dividing an individual’s hip circumference in centimeters by their height in meters [27]. BAI serves as an alternative method for estimating body fat percentage. Unlike BMI, which calculates total body weight relative to height, BAI estimates adiposity levels using only hip circumference and height, with reference values varying substantially by sex. This index is often considered more precise than BMI for individuals with highly developed muscle mass, as it is not confounded by the heavy weight of lean muscle tissue. The following formula can be used to calculate BAI:

$$BAI={\displaystyle \frac{Hip circumference \left(cm\right)}{Height {\left(m\right)}^{\mathrm{1,5}}}}-18$$

WHtR measures body fat distribution and serves as an excellent predictor of metabolic and cardiovascular risks [28]. Studies indicate that this indicator appears to be the most reliable predictor for visceral fat deposition [29]. It is calculated by dividing waist circumference (in cm) by standing height (in cm), according to the following formula:

$$WHtR={\displaystyle \frac{Waist circumference \left(cm\right)}{Height \left(cm\right)}}$$

WHR is a straightforward indicator that measures body fat distribution and is calculated by dividing waist circumference (in cm) by hip circumference (in cm). A elevated value for this metric indicates an increased risk of cardiovascular and metabolic diseases [30]. This indicator also features distinct reference thresholds based on sex. The calculation formula is as follows:

$$WHR={\displaystyle \frac{Waist circumference \left(cm\right)}{Hip circumference \left(cm\right)}}$$

BFP estimated based on BMI, age, and sex utilizes a scientifically validated formula, such as the Deurenberg equation. At the same time, BFP serves as a better predictor of mortality in young adults than BMI [31]. A method for calculating an estimate of body fat percentage uses BMI [32].

The BFP formula for adult males is as follows:

$$BFP=\mathrm{1,20} x BMI+\mathrm{0,23} x Age-\mathrm{16,2}$$

The BFP formula for adult females is as follows:

$$BFP=\mathrm{1,20} x BMI+\mathrm{0,23} x Age-\mathrm{5,4}$$

2.4. Study Design and Procedure

This investigation was structured as a descriptive, cross-sectional study. Participants comprised university students from the University of Oradea, enrolled across three major academic programs: BS, CSE, and SSP, with ages ranging between 20 and 25 years.

Strict exclusion criteria were applied to ensure data integrity:

• Enrolment in academic programs other than the three selected specializations;
• Failure to state age or falling outside the designated 20–25 age bracket;
• Failure to indicate biological sex;
• Formally diagnosed chronic metabolic and/or cardiovascular diseases;
• Non-completion or partial completion of the psychosocial scales.

Data collection was carried out through direct, face-to-face clinical and anthropometric interactions. Immediately following the completion of the physical measurements, each participant was provided with the printout containing the psychosocial scales. To guarantee seamless data matching, the anthropometric recording sheets and the psychosocial questionnaires for each individual were physically stapled and coded together.

2.5. Ethical Considerations

This study strictly adhered to institutional and international research ethics guidelines. All subjects participated on a completely voluntary basis and were informed in advance regarding the scope, purposes, and design of the research.

To ensure absolute privacy, participants were explicitly instructed not to provide any identifying information, such as names or personal identification codes. The protocol emphasized that only biological sex and academic program affiliation were relevant to the target epidemiological analysis. All collected data were aggregated anonymously.

The study protocol was formally reviewed and approved by the Research Ethics Committee of the Faculty of Medicine and Pharmacy, University of Oradea. Furthermore, signed written informed consent was obtained from each individual participant prior to the initiation of any data collection or anthropometric measurement procedures.

2.6. Statistical Data Analysis

A non-probability, snowball sampling technique was employed to recruit the study population. Statistical analyses were performed using SPSS software. Descriptive statistics, including frequency distributions, were computed across academic programs and sex categories. To evaluate systematic differences across the three academic programs regarding the biometric indicators (BAI, WHtR, WHR, BMI, and BFP), a one-way analysis of variance (ANOVA) was executed, with academic specialization serving as the independent variable. Sub-analyses were further categorized and stratified by biological sex.

Prior to modeling predictive relationships, a preliminary bivariate correlation analysis was conducted on all biometric and psychosocial parameters to identify significant associations and to justify the structural framework of the subsequent regression analyses.

To evaluate the extent to which BAI, WHtR, WHR, BFP, SWLS, and MSPSS predict weight status, multiple linear regression analyses were performed, with BMI serving as the continuous dependent variable. These regression models were systematically stratified by academic program and biological sex. Multiple linear regression was specifically selected to account for potential multicollinearity and the interrelated nature of the anthropometric and psychosocial predictors in their shared relationship with BMI.

To evaluate the predictive capacity and diagnostic performance of the anthropometric indices (BMI, BAI, WHtR, WHR, and BFP), alongside the SWLS and MSPSS, in discriminating between obese and non-obese status, Receiver Operating Characteristic (ROC) curve analyses were performed, stratified by sex. The area under the curve (AUC) with a 95% confidence interval (CI) was calculated for each parameter for both male and female subgroups. The optimal cut-off values were determined using Youden's index, maximizing both sensitivity and specificity for each analyzed indicator and psychosocial scale. Statistical significance was set a priori at a p-value < 0.05.

3. Results

3.1. Descriptive Statistics and Distribution Analysis

In alignment with the primary research objectives, the preliminary investigation assessed the sample distribution across sexes and academic programs (Table 1). Furthermore, the homogeneity and structural distribution of the quantitative data were evaluated using Skewness and Kurtosis indicators. The resulting coefficients fell within acceptable universal thresholds, confirming the normality of the data distribution (Gaussian distribution) and justifying the subsequent application of parametric statistical tests (Table 2).

Table 2 presents descriptive statistics, including the mean (M), standard deviation (SD), and minimum and maximum values (Min. and Max.) for BAI, WHtR, WHR, BMI, and BFP, stratified by academic program.

The comparison of observed and expected frequencies based on academic program and biological sex revealed statistically significant differences between men and women within the BS program (χ² = 27.319, p < .0001) and the CSE program (χ² = 11.716, p < .0006). These recorded differences (Table 2) align with our expectations, reflecting a well-documented general gender-based tendency toward selecting specific academic disciplines. Conversely, for the SSP program, the frequency distribution by sex did not reach statistical significance (χ² = 3.308, p = .06), indicating a relative homogeneity in academic choices within this field.

3.2. Comparative Analysis of Anthropometric Health Indicators

The biological parameters were systematically analyzed by sex to identify potential variations across the different academic programs (Table 3).

Among male students, BMI demonstrated statistically significant differences across programs (F = 6.418, p < .002); the Scheffe post-hoc test indicated significantly higher mean differences for the BS program compared to both CSE and SSP. Among female students, BMI also showed significant variations (F = 3.518, p < .03), with post-hoc differences recorded exclusively between the BS and CSE programs.

For WHtR, a significant variation was observed in men (F = 6.283, p < .002), showing higher mean values in BS compared to CSE and SSP; however, no statistically significant coefficients were found among women (F = 2.834, p = .06). Similarly, WHR in male participants exhibited significant differences (F = 5.964, p < .003), driven by higher values in the BS group relative to CSE and SSP, whereas no significant differences emerged for female participants (F = 0.708, p = .49).

Regarding body composition, BFP in men (F = 7.369, p < .001) revealed higher mean values in the BS program compared to both CSE and SSP. In women, BFP differences were also significant (F = 3.718, p < .02), appearing solely between the BS and CSE programs.

Finally, BAI scores in men (F = 4.541, p < .01) were significantly higher in the BS program compared to CSE. For female participants, BAI (F = 4.106, p < .01) indicated significantly higher mean scores in BS compared to SSP, but not when compared to the CSE cohort.

The recorded biological parameters sustain a consistent pattern across the academic program comparisons, wherein the BS cohort systematically exhibits elevated values relative to the other disciplines, while no statistically relevant differences emerge when comparing the SSP and CSE cohorts.

3.3. Predictors of BMI Among Students

Table 4 illustrates the correlation matrix of the biological parameters across the three academic programs, revealing high correlation coefficients that strongly support their likelihood of serving as relevant predictors. Notably, across all three programs (BS, CSE, and SSP), WHR shows no statistical association with BAI, nor does BFP demonstrate an association with WHR.

However, simple statistical correlations cannot fully capture precise metabolic risk. Although these anthropometric indices quantify distinct physiological aspects of adiposity, they are frequently intercorrelated in clinical practice. Consequently, within this healthy young adult student cohort, these biological indicators do not show isolated pathognomonic associations with health status; rather, each index provides distinct, complementary diagnostic information regarding body composition and regional fat distribution.

In the final phase of the analysis, building upon the association patterns observed among the biological parameters (Table 4), we evaluated a comprehensive set of predictors to estimate their respective influence on BMI within each academic program. Preliminary parametric analyses confirmed data homogeneity across all three investigated disciplines.

Two psychosocial variables—life satisfaction (SWLS) and perceived social support (MSPSS)—were introduced into the regression equations based on the hypothesis that they could play a critical role in BMI variance. Consequently, it was postulated that WHtR, WHR, BAI, BFP, life satisfaction, and perceived social support serve as significant predictors in estimating BMI across the three target academic groups.

As demonstrated in Table 5, statistically significant differences emerged between the estimates generated by the regression equations and the mean outcomes for the BS cohort [F(6, 143) = 245.781, p < .001], the CSE cohort [F(6, 158) = 230.077, p < .001], and the SSP cohort [F(6, 141) = 295.821, p < .001].

The proportion of variance in weight status (the coefficient of multiple determination) explained by the joint action of WHtR, WHR, BAI, BFP, life satisfaction, and perceived social support yielded an R² = .915 for the BS program, indicating that these variables account for 91.5% of the BMI variance. Within the CSE program, the multiple coefficient of determination (R² = .901) demonstrated that the predictors contribute 90.1% to BMI variance, while for the SSP cohort, an R² = .929 indicated that the variables account for 92.9% of the variance in BMI.

Overall, the implemented models indicate a substantial and robust impact of the selected predictors on BMI. This study highlights that combining biological predictors with perceived social support and life satisfaction (acting alongside underlying lifestyle factors such as diet, physical activity, and hobbies) provides exceptional accuracy in predicting the risk of overweight and obesity.

The specific influence of each predictor across the academic programs demonstrates both positive and negative effects on BMI (Table 6). Notably, WHtR emerged as the most dominant positive predictor for BMI within the BS cohort (β = 1.725; t = 7.822; p < .001), the CSE cohort (β = 1.986; t = 11.172; p < .001), and the SSP cohort (β = 1.695; t = 11.259; p < .001). These elevated β coefficients strongly imply that this regional metric captures body composition dimensions closely aligned with BMI.

Conversely, WHR exhibited a strong inverse relationship with BMI in the BS program (β = −.735; t = −4.659; p < .001), as did BAI (β = −.472; t = −3.580; p < .001). A highly consistent pattern was observed within the CSE program for both WHR (β = −.939; t = −6.939; p < .001) and BAI (β = −.629; t = −5.630; p < .001). Similarly, the SSP program displayed comparable predictive values for WHR (β = −.695; t = −6.106; p < .001) and BAI (β = −.376; t = −4.008; p < .001), reinforcing the indication that a decrease in these two parameters mathematically corresponds to an increase in BMI within this model.

Regarding overall adiposity, BFP was identified as a statistically significant but considerably weaker positive predictor for the BS cohort (β = .142; t = 2.332; p < .02) and the CSE cohort (β = .136; t = 2.292; p < .02); however, it failed to reach statistical relevance within the SSP program (β = .050; t = 1.180; p = .24).

Perceived social support demonstrated a minor but statistically significant inverse effect on BMI within the BS group (β = −.056; t = −2.018; p < .04), whereas life satisfaction proved completely non-significant. Within the CSE program, perceived social support exhibited a slightly more pronounced and robust statistical relevance (β = −.064; t = −2.448; p < .01) compared to the BS cohort, while life satisfaction again remained irrelevant. In the case of the SSP program, neither of the two psychosocial predictors demonstrated statistical significance.

Although the global analysis by academic program (Table 6) suggests that BMI variations are driven exclusively by anthropometric predictors while psychosocial factors remain non-significant, this overarching model may mask critical sex-specific dynamics within each discipline. To uncover these potential nuances and prevent important intragroup differences from being overlooked, a subsequent, fully stratified analysis by both academic program and sex was performed (Table 7).

To identify the significant predictors of BMI, a multiple linear regression analysis was conducted, fully stratified by sex and academic program (Table 7).

Among men, the predictive model was identical across all three fields of study: BFP was the only predictor with a significant and positive impact (p < .001). Secondary anthropometric factors (BAI, WHtR, WHR) and psychosocial variables showed no predictive value for BMI in the male sample, indicating that for male university students, BMI variations are strictly a reflection of total adiposity.

Among women, the results revealed distinct and highly nuanced predictive models for BMI, heavily influenced by their specific academic profile. Within the CSE cohort, the model closely mimicked the male sample, with only BFP exhibiting a significant predictive effect on BMI (p < .001), while psychosocial and secondary physical indices remained non-significant.

Interestingly, the BS group was the only cohort where psychosocial factors emerged as critical predictors for BMI. In addition to BFP (β = 1.008, p < .001), both life satisfaction (β = −.007, p = .009) and total social support (β = −.008, p = .002) exerted a significant negative predictive effect. This indicates that, for female biomedical students, higher life satisfaction and a robust social support network are associated with lower BMI values, suggesting a vital protective/buffering role of psychological resources against academic weight gain.

Conversely, within the SSP cohort, alongside BFP (β = 1.013, p < .001), BAI emerged as a significant negative predictor (β = −.035, p = .024), indicating a unique anthropometric distribution pattern specific to this student subpopulation.

As illustrated in Figure 1a, for the male cohort, the ROC curve analysis indicated a perfect predictive capacity of Body Fat Percentage (BFP) in determining obesity, with an Area Under the Curve value of AUC = 1.000 (95% CI: 1.000 – 1.000, p < .001). By applying the Youden index (J = 1.000), the ideal discrimination point was identified at a threshold value of 24.38% (Table 8). At this cut-off point, the model demonstrated absolute diagnostic performance, yielding a sensitivity of 100% and a specificity of 100%. From a clinical perspective, this result demonstrates that the 24.38% BFP threshold allows for an impeccable separation of subjects, completely eliminating the risk of diagnostic error (both false negative and false positive cases) within the investigated sample.

This absolute AUC value of 1.000 can be explained by the fact that BFP and BMI are highly correlated variables, both directly or indirectly quantifying adiposity. In our male cohort, the 24.38% threshold coincided perfectly with the demarcation line for obesity established by BMI, thereby demonstrating an excellent internal validity of the collected data.

Regarding the female cohort, as shown in Figure 1b, BFP also demonstrated an exceptional discrimination capacity, with an Area Under the Curve value of AUC = 1.000 (95% CI: 0.999 – 1.000, p < .001). The calculation of the Youden index (J = 0.995) established the optimal cut-off threshold at a value of 35.43% (Table 8). At this demarcation point, the model recorded a maximum sensitivity of 100% and a near-absolute specificity of 99.5% (with a negligible false-positive rate of only 0.5%). Clinically, this result attests that the 35.43% threshold for BFP ensures a virtually infallible diagnostic accuracy in female obesity screening, completely eliminating the omission of positive cases, while the risk of erroneously classifying normal-weight subjects remains negligible.

Regarding WHtR, within the male cohort, the ROC curve analysis (Figure 1a) indicated a remarkable discriminative capacity in evaluating obesity, with an Area Under the Curve value of AUC = 0.980 (95% CI: 0.964 – 0.996, p < .001). By applying the Youden index (J = 0.862), the optimal cut-off point was determined at a value of 0.56 (Table 8). At this threshold, the model presented high diagnostic accuracy, recording a sensitivity of 91.7% and a specificity of 94.5% (corresponding to a minimal false-positive rate of only 5.5%). From a clinical perspective, the identified threshold of 0.56 for WHtR demonstrates a balanced and solid performance, successfully isolating cases of true obesity with precision and significantly reducing false alarms among male subjects. The fact that sensitivity (91.7%) and specificity (94.5%) are both above 90% shows that WHtR is a tool almost as robust as more complex body composition evaluation methods, carrying the massive advantage of being extremely simple to measure at the patient's bedside or in the office.

Within the female cohort (Figure 1b), WHtR also demonstrated a particularly high predictive capacity for identifying obesity, with an Area Under the Curve value of AUC = 0.969 (95% CI: 0.949 – 0.988, p < .001). Based on the Youden index (J = 0.880), the optimal cut-off value was established at 0.53 (Table 8). At this demarcation point, WHtR presented a remarkable and highly balanced diagnostic performance, with a sensitivity of 94.1% and a specificity of 93.9% (the false-positive rate being only 6.1%). From a clinical perspective, the threshold value of 0.53 proves to be an ideal screening tool among women, ensuring a minimal error rate and confirming the utility of WHtR as a simple, yet extremely rigorous anthropometric indicator in the evaluation of excessive adiposity.

The ROC curve analysis evaluated the predictive capacity of BAI in identifying obesity (defined according to BMI). For the male cohort (Figure 1a), the results highlighted an excellent diagnostic performance of BAI, with an Area Under the Curve value of AUC = 0.928 (95% CI: 0.887 – 0.969, p < .001). To establish the optimal discrimination point, the Youden index was utilized. Thus, at a cut-off threshold value of 26.91, BAI demonstrated a high sensitivity of 91.7% and a specificity of 77.9% (with a false-positive rate of 22.1%) (Table 8). From a clinical perspective, this 26.91 threshold represents an optimal balance for obesity screening: the test successfully detects 91.7% of subjects with true obesity, minimizing the risk of missing patients who require metabolic intervention, while the false alarm rate (22.1%) remains within a clinically acceptable interval.

Regarding the female cohort (Figure 1b), the ROC curve analysis revealed a remarkable predictive capacity of the Body Adiposity Index (BAI) in identifying obesity, with an Area Under the Curve value of AUC = 0.982 (95% CI: 0.969 – 0.995, p < .001). By applying the Youden index (J = 0.865), the optimal discrimination point was established at a value of 34.41 (Table 8). At this threshold, the test demonstrated an extremely high diagnostic accuracy, recording a sensitivity of 91.2% and a specificity of 95.3% (with a minimal false-positive rate of only 4.7%). From a clinical standpoint, the high specificity value confirms that the 34.41 threshold minimizes the risk of false alarms among women, while simultaneously providing an excellent capacity for the active detection of cases with true obesity.

Regarding WHR within the male cohort, the ROC curve analysis (Figure 1a) indicated a robust discriminative capacity in evaluating obesity, with an Area Under the Curve value of AUC = 0.890 (95% CI: 0.839–0.940, p < .001). By applying the Youden index (J = 0.702), the optimal cut-off point was established at a value of 0.88 (Table 8). At this threshold, the model recorded a high sensitivity of 91.7% and a specificity of 78.5% (corresponding to a false-positive rate of 21.5%). From a clinical standpoint, the threshold value of 0.88 for WHR proves to be an effective tool for screening central-type adiposity among male subjects, ensuring the identification of the vast majority of obesity cases while maintaining an acceptable capacity to exclude subjects without metabolic risk.

Within the female cohort (Figure 1b), WHR demonstrated a good predictive capacity for identifying obesity, yielding an Area Under the Curve value of AUC = 0.803 (95% CI: 0.734 – 0.872, p < .001). By calculating the Youden index (J = 0.489), the optimal cut-off value was established at 0.75 (Table 8). At this demarcation point, the model registered a high sensitivity of 91.4%, but associated with a lower specificity of 57.5% (corresponding to a false-positive rate of 42.5%). From a clinical perspective, although the 0.75 threshold proves highly efficient in the active detection of most positive cases (minimizing the risk of missing women with obesity), the low specificity suggests that WHR is more prone to false alarms among women, necessitating caution and subsequent correlation with other anthropometric measurements.

In the analysis of psychosocial variables for the male cohort (Figure 1a), the ROC curve demonstrated that the MSPSS does not constitute a valid or significant predictor for identifying obesity. The Area Under the Curve value was AUC = 0.482 (95% CI: 0.381 – 0.583), confirming a discriminative capacity equivalent to pure chance, with the result being completely devoid of statistical significance (p = .735). The formal calculation for the Youden index (J = −0.020) indicated a theoretical cut-off point at a value of 2.04, associated with a sensitivity of 72.2% and a specificity of 25.8% (Table 8). From a clinical perspective, these data reflect the fact that psychosocial resources and the support perceived by male subjects do not present a linear or direct causal relationship with weight status quantified via BMI, suggesting that the perception of social support evolves independently of anthropometric composition within the investigated sample.

Similarly, the ROC curve analysis for the MSPSS among the female cohort (Figure 1b) indicated the absence of a consistent predictive capacity relative to weight status. The model recorded an AUC value of 0.563 (95% CI: 0.453 – 0.672), with the result being completely statistical non-significant (p = .241). By applying the Youden index (J = 0.010), a theoretical cut-off point was obtained at a value of 2.12, characterized by a sensitivity of 76.5% and a low specificity of 24.5% (Table 8). From a clinical and psychosocial standpoint, these findings reconfirm that the perceived quality of support received from family, friends, or significant others does not represent a determinant factor or a direct predictor of obesity in women, both dimensions (the psychosocial and the anthropometric ones) manifesting independently within the investigated cohort.

Regarding the evaluation of psychometric variables for the male cohort, the ROC curve analysis (Figure 1a) demonstrated that the SWLS does not represent a valid predictor for identifying obesity. The Area Under the Curve value was AUC = 0.416 (95% CI: 0.319 – 0.513), with the result being completely statistical non-significant (p = .116) (Table 8). Although the formal application of the Youden index (J = −0.007) indicated a theoretical cut-off point at a value of 17.50 (associated with a sensitivity of 88.9% and a specificity of 10.4%), the overall performance of the model is inferior to pure chance (random classification). From a clinical and psychological perspective, these data suggest that the level of life satisfaction perceived by male subjects does not correlate linearly or predictively with weight status (defined by BMI) within the investigated sample, the dynamics between psychological well-being and obesity most likely being mediated by other complex factors.

Similarly, the ROC curve analysis for the SWLS among the female cohort reconfirmed the absence of any predictive value for weight status (Figure 1b). The model recorded an AUC value of 0.357 (95% CI: 0.264 – 0.450), with the result being completely devoid of statistical significance (p = .264). The formal calculation of the Youden index (J = −0.031) generated a theoretical cut-off threshold at a value of 20.50, associated with a sensitivity of 79.4% and an extremely reduced specificity of 17.5% (Table 8). From a psychosomatic and clinical standpoint, these data demonstrate that the level of life satisfaction in women does not evolve in a linear or predictive manner relative to BMI-defined obesity. This finding highlights the relative independence of the subjective perception of life quality from rigorous anthropometric indicators within the investigated female cohort.

4. Discussion

Hypotheses H1, H2, and H3 were fully confirmed by the obtained results. In contrast, hypotheses H4 and H5 were refuted, both through the multilinear regression analysis within the involved academic programs and through the sex-stratified ROC analysis, demonstrating that the SWLS and MSPSS possess no predictive value within the investigated sample.

A major finding of this study is the dominance of the WHtR index as the primary predictor across all academic programs. This association is widely analyzed in the specialized literature, where Ashwell and Gunn [33] emphasize that while BMI quantifies total body mass, WHtR provides superior precision in evaluating central adiposity.

In a study conducted in Switzerland, the authors, utilizing multinomial logistic regressions, demonstrated that the probability of belonging to the overweight category (BMI ≥ 25.0 kg/m²) decreased with increasing height for both sexes across all contemporary datasets [34]. They explained that this negative association proved to be constant; only among recruits measured in the 1870s was the association positive, a period when an increase in height was linked to a higher BMI. This negative association appears not only for BMI but also for WHtR and WHR [34].

The standardized coefficients obtained in our study confirm that, among university students, an increase in body mass is almost intrinsically linked to the expansion of abdominal circumference.

However, due to the complex relationship between risk factors at the individual level, an increasing number of studies support the premise that BFP is a better indicator of obesity than BMI, as the latter fails to take into account actual body composition [35].

BMI is a simple formula based solely on height and weight. Nevertheless, body weight encompasses fat, muscle, bone, and water. A significant percentage of individuals with a normal BMI (18.5 – 24.9 kg/m²) present metabolic syndrome, and combining BFP with BMI can serve as an important and useful indicator for evaluating the risk of metabolic diseases [36]. Other studies have demonstrated that the sensitivity of BMI proved to be low, such that more than half (51%) of patients with an abnormal BFP were not identified as obese when using BMI [37]. Consequently, BMI should not be considered the sole criterion for obesity, particularly in patients with a BMI value below 30 kg/m². As early as the year 2000, the World Health Organization recognized the discrepancy between BMI and BFP in the Asian population, recommending that the BMI threshold for obesity in Asians be reduced to 27.5 kg/m² [38]. Although some research has shown a significant positive correlation of BMI with BFP and height [39], other studies have demonstrated that BMI cannot make a clear distinction between fat mass and muscle mass, with most obesity indices representing merely markers for central obesity, such as neck circumference (NC), hip circumference (HC), WHR, and BFP [40].

In the case of Body Fat Percentage (BFP), although it exerts a significant positive impact within the BS and CSE groups, it ceases to be a valid predictor for students enrolled in the SSP program. This discrepancy can be corroborated with the study by Nevill et al. [41], which suggests that in individuals with higher levels of physical activity (as is often the case for physiotherapy students), BMI tends to reflect muscular mass rather than adipose tissue, thereby reducing the predictive power of BFP over BMI. Sex-based differences regarding body fat percentage are unanimously reported by researchers, but variations also exist across different ethnic populations. For instance, most recently, it was communicated that predicted BFP at the obesity threshold (BMI = 23.0 kg/m²) was 21.5 – 24.4% for men and 33.1% for women; for class I obesity (BMI = 25.0 kg/m²), these values were 24.1 – 27.0% and 35.7%, respectively, thresholds that were validated within the South Korean population [42].

Furthermore, in the effort to demonstrate the importance of non-invasive indices for identifying obesity, a highly interesting study conducted by Anderson and colleagues [43] must be highlighted. Specifically, normal-weight obesity (NWO) describes individuals who present a normal body mass index but possess an abnormal amount of body fat. Identifying these individuals can represent a particularly crucial element in the prevention of obesity and cardiometabolic risk (which serve as risk factors for cardiovascular mortality); therefore, determining feasible methods to measure body fat is essential, as BMI can mask obesity in young adults [43,44]. Normal-weight obesity has also been identified among university students, particularly in women, leading authors to recommend focusing attention toward lifestyle adjustments aimed at preventing cardiometabolic risk within this population category [45]. It is vital that prevention strategies concentrate on monitoring body composition and promoting healthy habits [46,47].

From a psychosocial perspective, perceived social support emerged as a significant negative predictor for BMI within the BS and CSE groups. This result aligns with the research of Uchino [48], which demonstrated that adequate social support can function as a behavioral regulation mechanism, reducing the likelihood of adopting unhealthy dietary habits triggered by stress (emotional eating) that would ultimately lead to an increased BMI. The fact that this effect is absent within the SSP cohort might indicate the presence of other mediating factors, such as self-efficacy or the specific educational environment, which attenuate the relationship between social support and BMI.

Finally, the implemented model shows that the β coefficients are highest within the CSE group, suggesting that among students from this technical profile, the link between BMI and anthropometric indicators is much more rigid. The sedentary lifestyle characteristic of this profile (frequently correlated with prolonged desk work) could mean that any increase in body mass is deposited almost mathematically in the form of abdominal adipose tissue.

The psychosocial variables are not only statistically non-significant, but their practical relevance is nearly non-existent. Students from the physiotherapy program demonstrate a significantly higher level of somatic and body literacy. They are capable of maintaining a healthy BMI through targeted physical activity and technical knowledge regarding nutrition, thereby rendering them relatively "immune" to social or emotional influences over their body weight.

The constant presented in Table 6 represents the theoretical value of BMI if all other predictors within the model were fixed at zero. The higher constant observed in the CSE cohort (24.618) indicates a higher baseline "starting point" for BMI within this academic profile compared to the SSP group (14.041). This aligns with previous studies conducted on university student samples, which have reported lower BMI cut-off values, specifically 22.09 kg/m² for men and 23.73 kg/m² for women [49].

The results of the multiple regression analysis reveal a distinct dynamic of BMI predictors based on gender and academic major. While BMI is exclusively predicted by BFP among male students (regardless of major) and female CSE students, the model becomes significantly more complex for female students in BS and SSP.

A particularly valuable finding involves the female sample from BS, where psychosocial dimensions act as significant negative predictors of BMI. Specifically, a high level of life satisfaction (p = .009) and robust total social support (p = .002) are associated with lower BMI values. This phenomenon is highly consistent with the literature regarding the "buffering effect" of psychological resources against academic stress. As demonstrated previously, chronic stress alters eating patterns, often leading to overeating and a preference for energy-dense foods [50]. In university settings, a strong social support network and an optimistic outlook on life mitigate these effects, reducing emotional eating behaviors and stress-induced cortisol secretion—mechanisms heavily documented in BMI fluctuations among female students [51].

On the other hand, among female students in SSP, the significant negative association between BAI and BMI (p = .024), under BFP control, underscores a known mathematical and anatomical discrepancy. Studies evaluating surrogate measures of body composition emphasize that because BMI relies strictly on total body mass (weight) and height, it fails to differentiate between muscle and fat distribution [52]. Conversely, BAI relies on hip circumference. In cohorts with distinct physical activity profiles, such as physiotherapy students, variations in gluteal muscle mass versus adipose tissue can cause BAI and BMI to adjust inversely within a simultaneous multiple regression model [27].

For both sexes (men: 0.56; women: 0.53), the cut-off thresholds obtained in our study for WHtR are slightly higher than the general threshold of 0.50 frequently recommended in the international literature for global cardiometabolic risk [53]. We consider this to be perfectly expected and highly valuable, as it indicates that for the strict diagnosis of obesity (directly correlated with an elevated BMI), a higher, specific threshold is required to effectively eliminate false alarms.

It is crucial to highlight that the AUC value for WHR (0.890) is slightly lower than that obtained for WHtR (0.980). This discrepancy reflects a clear consensus in modern anthropometry: WHtR tends to be a stronger and more stable predictor of generalized obesity compared to WHR, because height remains fixed, whereas hip circumference can vary independently of adipose mass due to skeletal or muscular structure. Furthermore, WHR demonstrated a weaker performance among women (AUC = 0.803, specificity = 57.5%) compared to men (AUC = 0.890, specificity = 78.5%), which is fundamentally driven by the natural gynoid fat distribution pattern characteristic of female biology (localized around the hips and thighs). When a female individual gains weight, hip circumference frequently increases in parallel with waist circumference. Consequently, the mathematical ratio between them (WHR) may remain relatively constant or low (e.g., around 0.75), thereby masking true obesity as classified by BMI. For this reason, our results firmly demonstrate that WHtR represents a much more reliable and faithful indicator for female populations than WHR. Similar findings regarding the paramount importance of WHtR in obesity diagnosis were also reported in a study conducted on a university student population in Mexico [54].

Overall, the anthropometric predictors (BAI, BFP, WHtR, WHR) demonstrated remarkable diagnostic performance (with AUC values ranging from 0.803 to 1.000), confirming an excellent screening capacity and tight physiological links with BMI.

Conversely, hypotheses H4 and H5 were robustly refuted, as both the regression modeling and the ROC curve analyses demonstrated that the SWLS and MSPSS possess no predictive value within our sample. In the case of the MSPSS, a state of total independence was established; an AUC value so close to 0.50 (0.482 for men and 0.563 for women) provides strong mathematical evidence of independence between these variables. The social support network of male students (perceived through friends or family) neither influences nor is directly influenced by the presence or absence of obesity. Similarly, among women, the perceived quality of support received from family, friends, or significant others does not act as a determinant factor or a direct predictor of obesity, with both the psychosocial and anthropometric dimensions manifesting completely independently within the investigated cohort.

In the case of the SWLS, a clear independence of subjective well-being was observed in both men and women. The obtained data reveal that, among male participants, the global perception of life quality is not directly or exclusively dictated by the body mass index. A male student may present a BMI classified within the obesity range yet report high life satisfaction (and vice versa). In contrast to other studies, this finding suggests that within our male cohort, obesity did not dramatically impair the overall life satisfaction score, pointing toward potential psychological resilience or a distinct perception of body image in relation to personal happiness. Similarly, a relative independence of the subjective perception of quality of life relative to rigorous anthropometric indicators was highlighted within the investigated female cohort.

Regarding the cut-off thresholds, the theoretical inflection point was higher for women (20.50) than for men (17.50). This difference reflects the fact that, generally, the female individuals in this sample tend to report slightly higher or differently distributed raw satisfaction scores.

Overall, the psychological/psychosocial predictors (SWLS, MSPSS) demonstrated no predictive value (with AUC values close to 0.50 and p > .05). While physical and anthropometric parameters can successfully substitute or complement one another in obesity assessment, the psychological well-being (life satisfaction) and relational resources (perceived social support) of these young adults are not automatically or linearly determined by their body mass index.

Through these two combined analyses, we have solidly demonstrated that obesity and life satisfaction operate as mutually independent dimensions within our chosen sample of students from the University of Oradea. Weight status (obese vs. non-obese) does not automatically trigger a predictable decline in perceived happiness, thereby refuting the widespread stereotype that individuals with an elevated BMI necessarily suffer from dramatically diminished life satisfaction.

4.1. Practical Recommendations for Clinicians and Educational Institutions

Based on the empirical findings of this study, several actionable recommendations can be formulated for both clinical practice and academic environments:

• Implementation of Rapid Screening Protocols: It is highly recommended to integrate WHtR into university and school healthcare services as a primary screening metric. WHtR represents a simple, cost-effective, and substantially more reliable tool than WHR, enabling healthcare providers to rapidly identify young adults exhibiting elevated metabolic risks.
• Utilization of Sex-Specific Diagnostic Thresholds: Anthropometric evaluations in youth populations must strictly utilize differentiated cut-off thresholds rather than generalized values. Within the context of the investigated cohort, the optimal clinical demarcation points are established at 0.56 for men and 0.53 for women regarding WHtR, and at 24.38% for men and 35.43% for women regarding BFP.
• Adoption of a Holistic and Non-Stigmatizing Approach: Given that obesity status did not predict a decline in either life satisfaction or perceived social support, nutritional and metabolic intervention programs must remain focused strictly on physical health and biometric biomarkers. Clinical frameworks should actively avoid intervention strategies built upon the erroneous premise that young adults with obesity universally suffer from a global psychological decline.
• Deployment of Complementary Independent Evaluations: In the clinical management of obesity, psychological dimensions (such as SWLS) and relational resources (such as MSPSS) should be evaluated independently as potential therapeutic coping assets to support treatment compliance, rather than being treated as automatic or direct consequences of an elevated body mass index.

4.2. Limitations of the Study

Despite its methodological strengths and the novelty of its stratified approach, this study presents several limitations that should be acknowledged:

• Cross-Sectional Research Design: Because data were collected at a single point in time, this study cannot establish definitive temporal or long-term causal relationships between changes in body composition and the evolution of psychosocial factors.
• Specificity of the Sample Cohort: The investigation was conducted on a specific group of young adults (university students). Consequently, the direct extrapolation of the exact calculated cut-off thresholds to the general population or to other distinct age groups (such as children or the elderly) must be approached with caution.
• Self-Reported Nature of Psychometric Data: The SWLS and MSPSS scales are psychometric instruments based entirely on the subjective perception of the participants. Therefore, the potential influence of social desirability bias on the collected responses cannot be completely ruled out.

5. Conclusions

The present research offers a comprehensive, bidimensional perspective on obesity among young adults, highlighting a distinct dissociation between biometric/anthropometric markers and psychosocial variables. The major conclusions of this study are outlined as follows:

• Absolute Accuracy of Body Composition Metrics: BFP was confirmed as a highly robust and stable predictor of BMI-defined obesity across both sexes, demonstrating perfect or near-perfect diagnostic performance. The identified cut-off thresholds successfully capture biological constitutional dimorphisms and can serve as rigorous, population-specific clinical benchmarks.
• Superiority of Modern Anthropometric Indices: The study demonstrates that WHtR represents a highly reliable and robust screening tool, significantly outperforming the traditional WHR, particularly among female participants. The calculated optimal cut-off points underscore the utility of WHtR as a rapid, cost-effective, and non-invasive indicator for evaluating excessive adiposity.
• Validation of the Body Adiposity Index: BAI represents an excellent predictor for both cohorts, displaying superior diagnostic performance in women compared to men. This index is validated as a viable alternative screening modality in clinical settings where direct body weight measurement is challenging or unfeasible.
• Independence of the Psychosocial Dimension: In sharp contrast to the physical and biometric parameters, the investigated psychological and relational factors—namely MSPSS and SWLS—demonstrated no predictive capacity or statistical significance relative to obesity status.
• Clinical Implications and Destigmatization of Stereotypes: The total absence of predictive value from the SWLS and MSPSS scales constitutes a major conceptual conclusion of this study. These data firmly demonstrate that, within the investigated cohort, an elevated weight status does not automatically trigger an impairment in perceived happiness or a degradation of social support networks. This functional independence implies the presence of complex psychological resilience mechanisms among young adults, indicating that obesity does not follow a linear association with a decline in the subjective quality of life.
• Sex- and Academic Program-Specific Multi-Predictive Frameworks: The predictive models for BMI differ fundamentally based on sex and academic program. While BMI reflects a purely compositional dimension (driven strictly by BFP) for the male population and female students in technical fields, BMI variations are significantly shaded by non-compositional factors among female students in biomedical and social/physical therapy programs. Specifically, psychosocial resources, namely life satisfaction and total social support, act as significant protective factors against an elevated BMI exclusively among female BS students. These findings highlight the critical need for holistic, tailored, and gender-specific approaches when evaluating and monitoring the health and body composition status of young university students.