Introduction

Polycystic ovary syndrome (PCOS) is a prevalent and heterogeneous endocrine disorder affects 8% to 13% of reproductive-aged women globally, characterised by a constellation of interrelated reproductive abnormalities. Diagnosis is base on the presence of irregular menstrual cycles and/or hyperandrogenism, alongside either ultrasound evidence of polycystic ovarian morphology or elevated anti-müllerian hormone levels [1]. Additionally, PCOS is associated with the development of metabolic syndrome in up to 33% of affected women [2]. These combined reproductive and metabolic disturbances elevate the risk of major health complications over the lifespan, such as anovulatory infertility and type 2 diabetes [3], resulting in significant health and economic consequences.

While various theories have been proposed to elucidate the pathogenesis of PCOS, obesity emerges as a key pathophysiological component. Approximately 50% to 80% of women with PCOS are concurrently obese [4]. Moreover, evidence from Mendelian randomization analyses suggests a causal relationship between body mass index (BMI) and PCOS [5–7]. Obesity contributes to PCOS development through multiple mechanisms. For instance, the severity of obesity correlates with the extent of selective insulin resistance (IR). IR, along with consequent hyperinsulinemia, activates excessive ovarian androgen production, thereby accelerating PCOS development [8]. Additionally, obesity-related inflammation may exert potential effects on ovarian physiology due to dysregulated adipokine secretion, further impacting insulin sensitivity [9]. Collectively, it is well established that obesity exacerbates PCOS manifestations, while weight loss ameliorates the features of PCOS.

According to the international evidence-based guideline, first-line treatments for women with PCOS are lifestyle interventions, comprising dietary modification, exercise, and behavioural strategies [1]. Emphasis on prevention of weight gain and regular weight monitoring is crucial in PCOS management. Notably, achieving a targeted weight reduction of 5% to 10% demonstrated benefits in improving both reproductive and metabolic function indicators in women with PCOS [10]. However, existing clinical trials investigating the impact of weight loss in women with PCOS predominantly feature short to medium follow-up durations ranging from 4 weeks to 6 months, involving small sample sizes of 12–100 participants [11]. Despite effort to enhance adherence through more frequent visits in the intervention group, the rigorous lifestyle interventions employed, such as strict calorie restriction, often prove physiologically unrealistic and unsustainable [10]. Consequently, even in short-term follow-ups, considerable dropout rates are reported, potentially exaggerating treatment effects [11]. Furthermore, several medium- to long-term trials with follow-ups of at least 6 months reported dropout rates of 31% to 63% over six months, with a significant proportion of participants quitting the trial within the first 8 weeks [12–14]. It becomes evident that a single lifestyle prescription is unlikely to provide effective long-term management solutions.

Subsequently, growing attention has been directed toward maintaining adherence among women with PCOS undergoing lifestyle interventions and reducing high dropout rates in weight management. Identifying baseline characteristics linked with dropout is crucial for recognising individuals who could benefit from such interventions or who require alternative supports to achieve weight loss goals. While lower quality of life (QoL) score, higher baseline weight, elevated androstenedione level, elevated free and total testosterone level, and increased depressions score at baseline have been linked to higher dropout rates in prior studies [12–14], few investigations have explored dietary preferences, physical activity level (PAL), and body composition as predictors in overweight women with PCOS. Notably, these factors are robust predictors of weight loss maintenance in the general population, as supported by a systematic review [15]. The same review highlighted inconclusive value of depression scores and QoL due to the limited study numbers, warranting further investigation [15].

This exploratory analysis, conducted within a randomized controlled trial, aims to identify potential baseline characteristics associated with dropout during the lifestyle interventions in overweight women with PCOS. Baseline factors encompass menstrual characteristics, PCOS features, biochemical parameters, psychological scores, dietary and physical activity questionnaires, and objectively measured body composition, and exercise capacity.

Materials and methods

Results

Baseline characteristics of women with PCOS

A total of sixty-nine women (61.06%) dropped out of the lifestyle interventions (Fig 1), with no significant differences in attrition rates observed between the network interactive platform group and the control group (p = 0.444). Key baseline comparisons between dropouts and completers (Table 1) showed that dropouts had significantly lower PAL (1.42 vs. 1.50, p = 0.002) and higher BMI (27.99 vs. 26.35 kg/m², p = 0.028), as well as a higher prevalence of acanthosis nigricans (38.2% vs. 15.9%, p = 0.019). Trends toward higher gravidity, weight, and body fat percentage were observed in dropouts (p = 0.056–0.076). No significant differences were observed in psychological measures or most dietary factors. Except for lower vitamin C intake in the dropout group (p = 0.044), no other dietary differences were detected.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Characteristics of study participants according to dropout status.

https://doi.org/10.1371/journal.pone.0351575.t001

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. CONSORT flowchart.

https://doi.org/10.1371/journal.pone.0351575.g001

Unadjusted univariable logistic regression and correlation between predictive factors

Univariate analysis identified BMI, body fat percentage, weight, and PAL as associated with dropout (p < 0.10). Due to strong collinearity among adiposity indicators (pairwise Pearson correlations: weight vs. BMI = 0.89, weight vs. body fat percentage = 0.92, BMI vs. body fat percentage = 0.94), only BMI, which showed the strongest univariate association and a more readily interpretable effect size, was retained for the multivariable analysis (Tables 2 and 3).

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Univariate (unadjusted) model of the determinants of dropout from lifestyle interventions.

https://doi.org/10.1371/journal.pone.0351575.t002

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Correlation matrix of the determinants derived from univariate models.

https://doi.org/10.1371/journal.pone.0351575.t003

Adjusted multivariable logistic regression on predictive factors

Multivariable logistic regression identified baseline PAL as the sole independent predictor of dropout after adjusting for other factors (Table 4). Each unit increase in PAL corresponded to a 98.5% lower likelihood of dropout (OR = 0.015, 95% CI: 0.001–0.352), equivalent to a 29.6% reduction per 0.3-unit increase. The model explained 19.5% of the variance (Nagelkerke R² = 0.195) and showed acceptable discrimination (AUC = 0.705, 95% CI: 0.605–0.804), with no evidence of miscalibration (Hosmer–Lemeshow p = 0.847) or multicollinearity (all VIF < 4)(Fig 2). Categorical analysis by PAL level was precluded due to the small number of participants in higher activity categories (n = 4 in dropouts, n = 5 in completers).

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Multivariate (adjusted) model of the determinants of dropout from lifestyle interventions.

https://doi.org/10.1371/journal.pone.0351575.t004

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Receiver operating characteristic (ROC) curve for physical activity level (PAL) and dropout decision. Area under the curve (AUC) = 0.705 (95% CI: 0.605-0.804).

https://doi.org/10.1371/journal.pone.0351575.g002

Discussion

This study identified baseline objectively measured PAL as the sole and strongest predictor of dropout among overweight women with PCOS undergoing lifestyle interventions. Contrary to our expectations, the network interactive platform did not improve adherence or reduce dropout rates. Notably, none of the previously reported predictors (psychological factors, androgen levels, quality of life) were replicated, underscoring the unique role of PAL in this population.

PAL is a dimensionless index reflecting total energy expenditure relative to basal metabolic rate (BMR) over 24 hours [21]. It is calculated as total energy expenditure (TEE) divided by BMR. Notably, BMR accounts for the largest component of TEE. While higher activity-induced energy expenditure(AEE) would intuitively lead to a higher PAL, overweight and obese individuals do not necessarily have lower PAL. In fact, their AEE may be similar or even higher than that of leaner individuals [22]. This apparent paradox occurs because BMR is largely determined by body size—heavier individuals expend more energy on fundamental metabolic functions (e.g., breathing, ion transport, enzyme turnover) [22].

The predictive value of PAL extends beyond PCOS. Lower objectively measured PAL is a strong predictor of all--cause mortality in chronic obstructive pulmonary disease [23], and higher PAL predicts successful long-term weight loss (2–5 years) in the general population [24,25]. However, evidence in endocrine disorders, particularly PCOS, remains limited. Of note, although we measured PAL both objectively (accelerometers) and subjectively (self-report), only the objective measure was associated with dropout. This discrepancy likely reflects common overestimation of physical activity in self-report, highlighting the need for accelerometer-based measures in clinical practice [26].

Unlike previous studies that identified psychological or biochemical predictors of dropout in women with PCOS, our findings revealed no such associations. Regarding psychological factors, our participants reported higher baseline PCOSQ scores than those in other studies using the Chi-PCOSQ [27], suggesting fewer PCOS-related negative moods and, consequently, smaller emotional differences between dropouts and completers. Regarding biochemical factors, we observed no significant differences in androgen levels between the two groups. This may be explained by ethnic variations in PCOS phenotypes: Western populations tend to present with more pronounced hyperandrogenism and hirsutism [12,13], whereas Asian women with PCOS more commonly exhibit metabolic syndrome and its related features [28]. In line with this, our participants had relatively low Ferriman Gallwey scores, further accounting for the divergence between our results and those of earlier investigations.

This study have several notable strengths. First, we assessed a comprehensive set of baseline predictors, including lifestyle characteristics often overlooked previously. This allowed us to adjust for confounders and identify objectively measured PAL, rather than baseline weight [12], as the sole dropout predictor. Second, this exploratory analysis stemmed from a randomized controlled trial, with strict inclusion criteria and standardised treatment protocols, ensured a homogeneous participant background and consistent measurements, strengthening causal inference.

A notable consideration is the high discontinuation rate observed in both the intervention and control groups, which is consistent with previous PCOS lifestyle trials [11]. Contrary to our expectations, the network interactive platform did not improve adherence or reduce dropout rates, despite a prior RCT reporting such benefits [29]. This lack of effect may be explained by habituation, whereby monthly reminders were perceived as intrusive and lost their effectiveness over time [30]. Moreover, unlike previous studies has using one-way text message reminders, our platform required mutual communication: participants had to regularly report their diet and physical activity. The time burden and stress associated with reporting unfulfilled monthly goals may have further undermined adherence. Therefore, alternative support strategies are needed. Future research should explore less intrusive reminder methods and optimal frequency thresholds to facilitate effective collaboration between providers and patients undergoing long-term lifestyle interventions.

Another consideration is the distribution of PAL categories in our sample. According to the 2004 FAO/WHO/UNU classification, PAL values of 1.40–1.69, 1.70–1.99, and 2.00–2.40 correspond to sedentary/light active, moderately active, and vigorous/very active lifestyles, respectively. Owing to the limited number of participants with higher PAL levels, we could not perform categorical analyses. Nevertheless, our continuous analysis clearly demonstrates that each 0.3-unit increase in PAL reduces dropout likelihood by approximately 30%. These findings underscores the value of objective PAL measurement over self-reported questionnaires, enabling the identification of at-risk individuals who may require additional support to improve long-term adherence.

Finally, as this is an exploratory analysis with a relatively small sample size and multiple variables examined, the p-values reported should be interpreted with caution. The hypothesis-generating nature of our study means that the observed associations, while promising, require confirmation in larger, well‑powered prospective studies. Readers are advised to view the statistical findings as supportive evidence rather than definitive conclusions.

Conclusions

In conclusion, unlike previous studies, we identified baseline objectively measured PAL as the sole and strongest predictor of dropout in overweight women with PCOS undergoing lifestyle interventions. This finding highlights that higher PAL, assessed objectively, is associated with better treatment adherence. Clinically, we recommend routine objective screening for PAL in women with PCOS. Those with lower PAL should receive additional support alongside lifestyle interventions to improve adherence and promote weight loss.

Supporting information

Acknowledgments

The authors thank all researchers and participants. The authors confirm that all ongoing and related trials for this intervention are registered. The study design has been registered and described in the Chinese Clinical Trial Registry (ChiCTR2000034263) as of June 20, 2020.