Ethical approval and trial registration

The study protocol was approved by the Research Ethics Board (ref. no. AST/002/258, October 24, 2022) of Shahjalal University of Science and Technology (SUST). The procedures followed the ethical standards according to the 7^th revision of the Declaration of Helsinki [33]. Data were collected after obtaining written informed consent (S1 File in S1 Data) from each participant and approval from the university authorities. Privacy, anonymity, and confidentiality were ensured throughout data collection, analysis, and report writing.

The study involved no known risks, and no payments were made to participants. This study was registered on ClinicalTrials.gov (Identifier: NCT07047222) after completion of data collection, as the BCC module had been conducted prior to recognizing the requirement for prospective registration. All procedures, including participant recruitment, delivery of BCC sessions, and outcome assessment, were conducted according to a pre-specified protocol. The registration provides full transparency regarding the study protocol, outcome measures, and analysis plan. However, the authors confirm that all ongoing and related trials involving this BCC module are registered.

Description of BCC module

The BCC module was designed to provide information on menstrual disorders, particularly dysmenorrhea, and to promote awareness of lifestyle and health-related behaviors that may influence menstrual health outcomes. The program was conceptually grounded in the Transtheoretical Model (TTM), which recognizes behavior change as gradual and progressive process [32,34]. Accordingly, the module guided participants through stages of behavioral change, beginning with awareness during the pre-contemplation and contemplation phases, followed by preparation and initiation of healthier behaviors, and concluding with strategies to reinforce and maintain these practices. awareness, preparation, initiation, and maintenance of health-related practices [34]. The educational content therefore focus on improving knowledge, enhance motivation, and strengthen self-efficacy to facilitate healthier behavioral choices.

The module covered three main thematic areas: menstrual health education, lifestyle modification, and physical activity practices. Educational components included basic menstrual physiology, common menstrual disorders such as dysmenorrhea, and both modifiable and non-modifiable risk factors including poor dietary diversity, physical inactivity, food cravings (high-fat and high-sugar foods), overweight and obesity, psychological stress, insufficient sleep, excessive caffeine intake, early age at menarche, and family history. The sessions also emphasized healthy on dietary practices, including balanced and diverse food intake, regular meals, and adequate hydration and the importance of key micronutrients relevant to women’s health such as vitamin A, vitamin B-complex, vitamin C, iron and calcium. In addition, participants were introduced to the benefits of regular physical activity for pain reduction, stress management and overall well-being. Simple stretching exercises, breathing techniques, and selected yoga postures suitable for practice in dormitory or home settings were demonstrated, while discussions addressed sleep hygiene, stress management and healthy weight maintenance as supportive lifestyle practices.

The BCC module was delivered through structured and interactive educational sessions facilitated by two trained female educators using a standardized facilitator manual, presentation slides and participant handouts. The educator had received prior training in menstrual health from a public health expert and a gynecologist to ensure appropriate delivery of the content. The sessions were held in common areas or designated meeting rooms within the residential halls to ensure a comfortable and accessible learning environment. The teaching methods included multimedia presentations, guided discussions, practical demonstration of yoga and physical activities, distributions of illustrated informational pamphlets (S1 Fig in S1 Data). Interactive activities such as quizzes and group reflections were used to reinforce learning and encourage peer engagement. Attendance was recorded for each session, and regular coordination with the research team and supervision by the principal investigator ensured fidelity to the study protocol. Participants were encouraged to set personal goals related to diet, physical activity, hydration, sleep, and caffeine intake, while reflecting on potential barriers to adopt healthier behavior. Light refreshments, small gifts, and certificates of participation were provided to encourage engagement. The detailed structure and content of the BCC module are provided in S1 Appendix in S1 Data and the corresponding logic model is shown in S2 Appendix in S1 Data.

Study design

This study employed a matched cross-sectional comparative design to examine the association between attendance at a BCC module and dysmenorrhea prevalence among female university students. Random assignment to groups was not feasible due to fixed class schedules, residential hall arrangements, academic commitments, and voluntary participation in BCC sessions. Therefore, participants were classified based on their exposure to the BCC module. Students who voluntarily enrolled between May 1 and May 15, 2023 and completed all three BCC sessions were designated as the BCC-exposed group. The comparison group consisted of female students from the same universities who did not participate in any component of the BCC module during the study period, referred to as the BCC non-exposed (control) group. Control participants were identified from official dormitory and student enrollment records to ensure comparability in terms of academic environment, living conditions, and access to university health services. They were not approached to attend at the sessions and did not receive any BCC-related materials or follow-up communication during the study period.

The BCC sessions were conducted between May 20 and June 30, 2023, at three public universities in Bangladesh: Patuakhali Science and Technology University (PSTU), Barisal University (BU), and Khulna University (KU). These universities were purposively selected to ensure geographic diversity and to represent female students from coastal and southern regions of the country, where access to menstrual health education and support services is often limited. Moreover, these institutions have a high number of female students residing in dormitories, which facilitated participant recruitment and program delivery. The sessions were delivered in small peer groups of approximately 20 participants, with each group attending three sessions, one of each domain of the BCC module. A total of 39 sessions were conducted, with two sessions delivered per day. Each session lasted 60–90 minutes and followed a standard structure to ensure consistency across groups and study locations.

Several measures were taken to minimize contamination between the study groups. BCC sessions were conducted in designated residential hall spaces with restricted access and scheduled separately from the routine academic activities. Attendance was strictly monitored, and educational materials (including printed pamphlet and visual aids illustrating yoga postures) were distributed exclusively to participating students. The participants were also requested not to share the training content with other students during the study period.

Following the completion of BCC sessions, participants in the BCC-exposed group were contacted periodically over a six-month follow-up period (July 15, 2023 to January 15, 2024). During this time, the educators, under the supervision of a research supervisor, conducted reinforcement visits at the second, fourth, and six months. These visits provided clarification on session content, addressed participants’ questions, and facilitated engagement with the practices discussed in the BCC sessions. The follow-up period was intended to support participants in applying the information covered in the three content domains of the BCC module and to enable descriptive monitoring of participant experiences, without implying causal effects on behavior or health outcomes.

Finally, the outcome assessment for both the BCC-exposed and matched control groups was conducted from February 1, 2024 to March 31, 2024. Data were collected at the individual level, allowing for a comparative evaluation of the association between BCC exposure and menstrual health-related outcomes, including dysmenorrhea prevalence.

Participants

Sample size estimation

The sample size for this study was calculated based on the anticipated prevalence of dysmenorrhea in the non-exposed group, which was 42% as reported in our previous study [30]. A 1:1 proportional sampling strategy was applied to ensure an equal number of participants in the BCC-exposed and non-exposed (control) group, maintaining a balanced comparison between the study arms. To detect a difference of 30% in dysmenorrhea prevalence between the BCC-exposed and non-exposed groups, a total of 452 participants were required, with 226 participants in each group. [35,36]. The proportional distribution of the sample included 56, 119 and 51 respondents for each independent group from PSTU, BU, and KU, respectively. The sample size was calculated based on a 95% confidence interval (CI) and 80% statistical power using the following equation (see S1 Text in S1 Data for detail of the sample size calculation):

Whereas, n = required sample size per group

(42% anticipated incidence of outcome in the non-exposed group)

(expected 30% relative difference from 42% in the BCC-exposed group)

(Z value for 95% confidence interval)

(Z value for 80% power)

(estimated variance for non-exposed group)

(estimated variance for BCC-exposed group)

Participants flow

Initially 498 participants were recruited from three public universities (PSTU, BU, and KU), evenly divided between the BCC-exposed and non-exposed groups (249 participants per group), accounting for an anticipated 10% non-response rate. After explaining the purpose of the study and obtaining informed consent, 249 students (62 from PSTU, 131 from BU, and 56 from KU) were enrolled in the BCC-exposed group. An equal number of students from the same universities were allocated to the non-exposed comparison group. Although all listed participants attended the first session, seven students (two, four, and one from PSTU, BU, and KU, respectively) were excluded due to absence during the second and third sessions of the BCC module. Consequently, the remaining 242 participants who completed all three BCC sessions were included in the follow-up phase. During the six-month follow-up, a total of 11 participants (3 BCC-exposed and 8 non-exposed) declined to participate in the outcome assessment. Therefore, 239 BCC-exposed and 241 non-exposed students completed the final outcome assessment. Data from eight participants were subsequently excluded due to incomplete or invalid responses, resulting in a final analytic sample of 234 BCC-exposed and 238 non-exposed participants. Fig 1 demonstrates the entire sampling process and participation flow.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. TREND flowchart of sampling process.

https://doi.org/10.1371/journal.pone.0349064.g001

Data collection

The data were collected on-site at the respective universities, in private rooms within dormitories or academic buildings to maintain confidentiality and minimize distractions. Within these settings, a standardized data collection approach was followed to ensure consistency across study sites. Data were gathered using structured questionnaires specifically designed for the context of Bangladesh, informed by a comprehensive review of relevant literature and information, education, and communication (IEC) materials on menstrual disorder management [25,30,37,38]. The questionnaires were initially developed in English and then translated into Bengali (S2 File in S1 Data). A reverse translation process was also conducted to verify the accuracy and maintain consistency with the original questionnaire. Data were collected using these questionnaires administered to respondents in both the BCC-exposed and non-exposed groups. A team comprising two supervisors and ten data collectors was recruited and trained in social science research methods, public health management, ethical principles, data collection tools, interviewing techniques, and the verbal multidimensional scoring system used to classify outcomes. Recognizing the sensitive nature of the topic, all data collectors and supervisors were female to create a comfortable environment for the participants to discuss menstrual-related issues. However, blinding of outcome assessors to the group allocation was not implemented during data collection. Before the actual data collection began, a pilot study was conducted with 45 participants to assess the validity, reliability and feasibility of the designed questionnaires.

Outcome variable

Dysmenorrhea was defined as any pain or discomfort associated with the menstrual cycle and assessed according to the scoring system of Andersch and Milsom [39]. This system categorized dysmenorrhea into four grades: grade 0, grade 1, grade 2, and grade 3, based on the severity of pain, interference with daily activities, and the dependence on analgesic use. For the purpose of the statistical analysis, dysmenorrhea was treated as a binary outcome variable. Respondents reporting no pain were categorized as “No = 0”, while those experiencing any level of pain (mild, moderate, or severe) were grouped as “Yes = 1” [40,41]. Dysmenorrhea severity was assessed using a standardized, interviewer-administered grading scale. All interviewers underwent comprehensive training on the scoring criteria and followed a structured, step-by-step assessment protocol covering all components of symptom severity before assigning the final grade. Before data collection, interviewers read the instructions to participants and clarified any unclear terms to ensure consistent understanding. This approach was used to minimize misclassification and enhance the reliability of the self-reported outcome measures. The detailed description of the outcome variable is provided in S2 Text in S1 Data.

Covariates

Appropriate selection of covariates is essential in PSM to improve comparability between exposure groups and reduce potential confounding. Covariates were selected based on prior literature, theoretical relevance, and their potential association with both BCC exposure and dysmenorrhea outcomes [42]. Physical activity was categorized as follows: i) sedentary (referring light-intensity physical activity, e.g., casual walking, light household chores; and activities with little or no exercise); (ii) active, (moderate-intensity activities such as brisk walking, light jogging, or moderate exercise, e.g., yoga, weight lifting etc. for at least 150–300 minutes per week); and (iii) athlete (vigorous-intensity activities such as running, jumping, swimming, climbing, fast cycling or competitive sports etc. for at least 75–150 minutes per week) based on World Health Organization’s (WHO) recommendations [43]. Obesity was assessed using body mass index (BMI, kg/m²) as a nutritional indicator. We classified BMI based on the WHO-recommended cut-off points for the Asian population [44], as follows: (i) underweight (< 18.5 kg/m²), (ii) normal weight (18.5–22.9 kg/m²), (iii) overweight (23.0–27.5 kg/m²), and (iv) obese (> 27.5 kg/m²). However, the overweight and obese categories were combined into an overweight/obese group to avoid zero cell counts in chi-square estimating. Dietary data were collected using a 24-hour recall method over five consecutive days, and dietary diversity score (DDS) was calculated as per Food and Agriculture Organization’s (FAO) guideline [45]. The average DDS (range: 0–10) was calculated by summing the five daily scores and dividing by five. Participants were then categorized as having high DDS (≥5) or low DDS (<5) based on their average score.

We extracted data on various lifestyle factors including cravings for high-fat and sweet foods (yes/no) [46], skipping breakfast one or more times in the last week (yes/no), sleep duration (< 7 hours per night, ≥ 7 hours per night), and caffeine consumption (infrequent, < 3 times per week; frequent, ≥ 3 times per week) [37]. Additional socio-demographic and nutritional covariates included family history of menstrual disorders (yes/no), age at menarche (years), and marital status (never married, ever married), father’s and mother’s education (below secondary education, secondary or higher education), and mother’s occupation (formal occupation, informal occupation). A comprehensive description of each covariate used in this study is presented in S3 Text in S1 Data.

Statistical analysis

Following data collection, propensity score matching (PSM) was applied to improve comparability between the BCC-exposed and non-exposed groups. The PSM approach estimates the conditional probability of exposure to the BCC module given a set of observed covariates [47] and is widely used in observational studies where random assignment is not feasible [48,49], helping to reduce potential confounding due to measured variables [50].

To implement this approach in the present study, propensity scores (PS) were first estimated for each participant using a logistic regression model with all the observed covariates implemented through the “psmatch2” command. Seven candidate propensity score models (Model 1–7), each incorporating different combinations of theoretically relevant covariates, were evaluated. The selection of candidate covariates was informed by prior literature and subject-matter knowledge rather than statistical significance of outcome. These alternate specifications were examined to assess the stability of covariate balance among the models.

Across all candidate models, the primary matching algorithm was held constant and implemented as 1:1 nearest-neighbor matching without replacement and a caliper width of 0.01 [51,52]. This approach paired each BCC-exposed participant with a non-exposed participant who had the closest propensity score within the specified caliper. Simulation studies indicate that tighter calipers reduce residual bias and improve match precision, while maintaining appropriate confidence interval coverage and type I error rates, particularly for binary outcomes and when covariates are binary or mixed in distribution [53,54]. Therefore, among the non-exposed participants, 238 fell within the common support region, indicating sufficient overlap in the propensity score distributions between the exposure groups. Applying the specified matching criteria resulted in 196 matched samples (98 BCC-exposed and 98 non-exposed participants), smaller than the initially planned sample size (Fig 1). The remaining 140 non-exposed participants, although within the common support region, were not selected due to the 1:1 matching procedure and the no-replacement rule, reflecting constraints of the matching algorithm rather than inadequate overlap between groups (see propensity score distribution in S2 Fig in S1 Data). This smaller matched sample was a deliberate consequence of applying strict matching criteria to ensure optimal covariate balance and strengthen internal validity. Although wider calipers could have increased the number of matched participants, they would have involved a trade-off between sample size and match precision. Given the smaller matched sample obtained through primary matching specification, additional sensitivity analyses were planned to examine the robustness of the results under alternative matching procedures. These analyses included nearest-neighbor matching with multiple neighbors, radius matching with wider calipers and kernel matching. Evaluating these alternative approaches ensured that the robustness of the findings could be assessed despite the reduced sample size in the primary matched dataset. Accordingly, a caliper width of 0.01 was selected for the primary analysis.

Following the matching procedure, covariate balance between BCC-exposed and non-exposed groups was carefully assessed. Balancing diagnostics included standardized mean differences (percentage bias), percentage reduction in bias, post-matching t-tests, and variance ratios (S1A Table in S1 Data) using the “pstest” command. Overall balance was summarized using Rubin’s B and Rubin’s R statistic (S1B Table in S1 Data). To enhance transparency and minimize the risk of specification searching, the criteria for selecting the propensity score model were defined a priori. Adequate balance was defined a priori as standardized bias within ±10% for all covariates [55,56], Rubin’s overall imbalance statistic (B) < 25% [57], and Rubin’s variance ratios (R) within the range of 0.5 to 2.0 [58]. Models were compared solely on the basis of these predefined balance diagnostics rather than outcome estimates or statistical significance, thereby reducing the risk of specification searching. The model meeting all pre-specified thresholds and demonstrating the most consistent reduction in residual imbalance across covariates was retained as primary propensity score specification.

After identifying the primary PSM model, the same set of covariates included in that model was used for adjustment in both pre- and post-matching analyses. Subsequently, the matched sample generated from the primary propensity score specification was used for post-matching outcome analysis. Before matching, bivariate and multivariable logistic regression models were fitted to examine crude and adjusted associations between BCC exposure and dysmenorrhea. After matching, conditional logistic regression (fixed-effect) was applied to the matched pairs to account for the matched study design using the “clogit” command. Bootstrap resampling was used to estimate the standard error (SE) and 95% confidence intervals for the propensity score adjusted estimates. A total of 500 bootstrap replicates were performed, which provided stable SEs and CIs, as additional replicates had minimal impact on their precision and coverage [59].

The categorical variables were expressed as numbers and percentages and compared using chi-square (𝛘²) test while the continuous variables were presented as means and standard deviations (SD) and compared using Student’s t-test. Crude odds ratios (COR) and adjusted odds ratios (AOR) are reported with corresponding confidence intervals. Analyses were conducted using Stata 17.0, with statistical significance defined at a two-sided α = 0.05. Significance levels were indicated as * p < 0.05, ** p < 0.01, and *** p < 0.001. Although the BCC module was delivered in group sessions, the individual participants were considered the unit of analysis.

Sensitivity analysis

Selection of primary propensity score specification

Among seven pre-specified propensity score specifications, only Model 2 satisfied all predefined balance criteria and was therefore retained as the primary propensity score specification (S1B Table in S1 Data). Under this model, all thirteen covariates achieved standardized bias within the ± 10% threshold, with a mean post-matching bias of 4.3%. Rubin’s B statistic (17.3%) and Rubin’s R (0.80) were also within the acceptable limits, indicating adequate overall balance. These balance metrics are further illustrated visually in Fig 2, which shows that propensity score matching substantially reduced covariate imbalances, bringing all standardized biases within the predefined ±10% range, representing improved comparability between the BCC-exposed and non-exposed groups.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Standardized bias plot of the covariates before and after propensity score matching.

The X-axis displays standardized percentage bias for each covariate, while the Y-axis lists the observed covariates. The standardized bias plot was used to assess covariate balance between the BCC-exposed and non-exposed groups before and after matching. The vertical reference line (red) on X-axis at ±10% standardized bias represents the accepted threshold for adequate covariate balance..

https://doi.org/10.1371/journal.pone.0349064.g002

In comparison, the remaining specifications did not fully satisfy the predefined thresholds. Specifically, several models exhibited residual standardized bias exceeding ±10% for one or more covariates, while others showed less favorable overall imbalance statistics, including Rubin’s R values approaching or exceeding 25% or variance ratios outside the acceptable 0.5–2.0 range (S1B Table in S1 Data). Additionally, Fig 3 demonstrates that after matching, the propensity score distributions of the BCC-exposed and non-exposed groups exhibit substantially greater overlap compared to before matching. This improved alignment further confirms enhanced comparability between groups and supports Model 2 as the primary propensity score specification for subsequent analysis.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Kernel density plots of propensity scores before and after matching.

X-axis: Propensity scores (range 0–1.0), and Y-axis: Kernel density (probability density of propensity scores). (A) Before matching: shows limited overlap between BCC-exposed and non-exposed (control) groups, (B) After matching: exhibits improved overlap in propensity scores, indicating better balance between groups.

https://doi.org/10.1371/journal.pone.0349064.g003

Participant characteristics

In this study, Fig 4 demonstrates the distribution of dysmenorrhea pain grades according to BCC exposure status before and after propensity score matching. Before matching, participants in the BCC-exposed group had a higher proportion reporting no pain (33.7% vs. 6.1%), whereas lower proportions reported mild (8.05% vs. 26.5%), moderate (6.1% vs. 12.08%), and severe pain (1.7% vs. 5.7%) compared with the non-exposed (control) group. These differences in pain severity distribution between groups were statistically significant (p < 0.001). In contrast, in the matched sample, the overall prevalence of dysmenorrhea was 69.4%, including 77.6% among non-exposed participants and 61.2% among BCC-exposed participants. Specifically, no pain was reported by 38.8% of BCC-exposed and 22.4% of non-exposed participants; mild pain by 30.6% vs. 42.9%; moderate pain by 22.4% vs. 23.5%; and severe pain by 8.2% vs. 11.2%. Although the BCC-exposed group had a higher proportion with no pain and the non-exposed group had higher pain levels, the differences in distribution were no longer statistically significant (p > 0.05).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Distribution of dysmenorrhea pain grades by BCC exposure before and after propensity score matching.

Prevalence (%): percentage of participants reporting different dysmenorrhea pain grades (No pain, Mild, Moderate, and Severe) in BCC-exposed and non-exposed (control groups). (A) Shows the distribution of dysmenorrhea pain grades in the study groups before matching (𝛘² = 168.01 and p < 0.001); (B) Demonstrates the distribution of dysmenorrhea pain grades in BCC-exposed and non-exposed (control) groups after matching (𝛘² = 6.8 and p > 0.05).

https://doi.org/10.1371/journal.pone.0349064.g004

Table 1 presents the characteristics of the study participants according to BCC exposure status before and after PSM analysis. Before matching, several participant characteristics differed between the BCC-exposed and non-exposed groups. Compared with the non-exposed group, participants in the BCC-exposed group were more frequently physically active or athletes (25.4% vs. 6.6%, p < 0.001), more likely to have normal BMI (36.2% vs. 17.6%, p < 0.001), higher dietary diversity (≥ 5 DDS: 29.2% vs. 18.01%, p < 0.001), and reported older age at menarche (13.4 ± 1.7 vs. 12.1 ± 1.5 years, p < 0.001). Significant differences were also observed for food craving, skipping breakfast, sleep duration, caffeine consumption, family history of menstrual disorders and father’s education (p < 0.001), as well as marital status, mother’s education and mother’s occupational status (p < 0.05), indicating imbalance in observed covariates between the groups. Following matching, the distribution of these characteristics was well balanced between the groups, with no statistically significant differences observed (all p > 0.05). This indicates improved balance in observed covariates in the matched sample used for subsequent analyses.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Participant characteristics according to BCC-exposure status before and after propensity score matching.

https://doi.org/10.1371/journal.pone.0349064.t001

Determinants of dysmenorrhea

Table 2 presents the association between lifestyle and dietary factors and the likelihood of experiencing dysmenorrhea according to BCC-exposure status, based on multivariable logistic regression analyses conducted before and after matching. Compared with sedentary participants, being physically active among BCC-exposed participants was associated with 77% and 94% lower odds of dysmenorrhea before matching (AOR = 0.23; 95% CI 0.07, 0.68; p < 0.01) and after matching (AOR = 0.06; 95% CI 0.03, 0.40; p < 0.01). Similarly, a high dietary diversity score (DDS ≥ 5) was associated with a substantially lower likelihood of dysmenorrhea among BCC-exposed participants, for both before (AOR = 0.06; 95% CI 0.02, 0.16; p < 0.001) and after PSM (AOR = 0.06; 95% CI 0.01, 0.23; p < 0.001).

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Factors associated with dysmenorrhea according to BCC exposure status before and after propensity score matching using multivariable logistic regression.

https://doi.org/10.1371/journal.pone.0349064.t002

Certain lifestyle factors, such as food craving for high fat and sweet foods (AOR = 3.0; 95% CI 1.1, 8.09; p < 0.05) and inadequate sleep (< 7 hours per night, AOR = 3.5; 95% CI 1.3, 9.1; p < 0.05), were also associated with higher odds of dysmenorrhea in the unmatched sample among BCC-exposed participants. However, these associations were attenuated and no longer statistically significant in the matched analysis, suggesting that confounding factors may have influenced the initial observations.

Association between BCC module exposure and dysmenorrhea

Table 3 exhibits the association between exposure to the BCC module, selected lifestyle factors, and dysmenorrhea before and after matching. Before matching, the odds ratio from the multivariable logistic regression model showed that participants who reported attending the BCC module had 87% lower odds of experiencing dysmenorrhea (AOR: 0.13, 95% CI: 0.06, 0.26; p < 0.001) compared with non-exposed participants. This association was adjusted for all covariates included in model 2, in addition to BCC module attendance. After PSM, the conditional (fixed-effect) logistic regression indicated a similar pattern, with BCC module exposure remaining significantly associated with a lower likelihood of dysmenorrhea (AOR: 0.13, 95% CI: 0.02, 0.79; p < 0.05), after controlling for the same set of covariates.

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Adjusted associations of BCC module exposure, key lifestyle factors, and dysmenorrhea before and after propensity score matching.

https://doi.org/10.1371/journal.pone.0349064.t003

Furthermore, participants who were physically active or engaged in athletic activities showed a lower likelihood of experiencing dysmenorrhea in both pre- and post-matching analyses. Similarly, having a high dietary diversity score (DDS ≥ 5) was associated with lower odds of dysmenorrhea across all regression models. Conversely, food cravings for high fat and sweet foods were associated with a higher likelihood of dysmenorrhea before and after PSM. The detailed results for all covariates are presented in S2 Table in S1 Data. To further assess the robustness of the associations, we conducted sensitivity analyses using the full four-grade dysmenorrhea severity scale, as summarized in S3 Table in S1 Data.

Sensitivity analysis

Robustness to estimation approaches.

Table 4 represents additional sensitivity analyses using alternative estimation approaches to evaluate the robustness of the observed differences in dysmenorrhea prevalence between BCC-exposed and non-exposed participants. In the matched sample, the adjusted difference in dysmenorrhea prevalence estimated using IPWRA (doubly robust) was −0.23 (95% CI: −0.33 to −0.13; p < 0.001; Table 4), indicating a 23–percentage-point lower frequency of dysmenorrhea among BCC-exposed participants compared with matched non-exposed after adjusting for observed covariates. Estimates derived from RA (ATT = −0.23; 95% CI −0.33, −0.13; p < 0.001) and IPW (ATT = −0.23; 95% CI −0.34, −0.12; p < 0.001) were consistent with the doubly robust estimate, demonstrating stability across estimation methods.

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Adjusted differences in dysmenorrhea prevalence between BCC-exposed and non-exposed participants (ATT estimates).

https://doi.org/10.1371/journal.pone.0349064.t004

Additional analyses incorporating propensity score matching with one or two neighbors as well as nearest-neighbor matching with and without bias adjustment produced ATT estimates similar to the primary doubly robust approach (see details in S5 Table in S1 Data), supporting the robustness of observed differences to alternative estimation methods.

Strengths and limitations

This study has several methodological strengths that enhance the credibility of the observed findings. It examined associations within a structured, theory-informed BCC framework implemented in a real-world university setting. The data quality in this study is considered high, with trained female supervisors and data collectors engaged to foster to create a comfortable environment for discussing menstruation-related issues and to minimize bias in understanding technical terms. Additionally, the analysis incorporated key covariates identified in the systematic reviews by Mitsuhashi et al. and Greaves et al. as relevant to behavior adherence [37,71], improving adjustment for measured confounders. Propensity score matching resulted in improved balanced across observed covariates, and multiple analytic approaches produced directionally similar results, indicating stability of the observed associations across statistical methods. Participants were followed up for six months, a standard duration for assessing intermediate-term behavioral patterns, which allowed sufficient time to integrate new habits compared with shorter follow-ups [71,76]. The BCC module was guided by the theory of change framework emphasizing awareness, motivation, skill building and self-efficacy. While this framework provides conceptual justification for examining behavioral associations, the present design does not allow determination of whether these components directly produced the report outcomes.

Despite these strengths, several important limitations must be emphasized. First, baseline prevalence of dysmenorrhea was not measured prior to BCC exposure, preventing assessment of temporal sequencing and causal inference. Although, propensity score matching improves comparability on measured covariates, it cannot replace baseline outcome measurements. Therefore, the observed differences may reflect pre-existing variation between groups, and self-selection bias and regression to the mean remain plausible alternative explanations. Thus, the study should be interpreted as a cross-sectional comparison between exposed and non-exposed groups rather than as an evaluation of intervention effectiveness. Second, the non-randomized design introduces potential selection bias. Despite improved balance on observed characteristics through matching, residual confounding from unmeasured or unknown factors cannot be excluded. Third, the strict caliper applied in propensity score matching resulted in a smaller matched sample (98 pairs), which may have reduced statistical power. However, consistent results across sensitivity analyses suggesting that the reduced sample did not materially compromise the robustness of the findings. Fourth, contamination between groups may have affected internal validity, as exposed and non-exposed participants were recruited from the same university and dormitory settings, allowing potential peer interactions and indirect exposure to BCC-related information. No formal monitoring or empirical verification of non-exposure was conducted. Therefore, the extent of contamination cannot be quantified. Any indirect exposure could have attenuated differences between groups. Sensitivity analyses explored hypothetical contamination scenarios, but these remain assumption based. Fifth, dysmenorrhea was assessed using self-reported Andersch-Milsom multidimensional scoring system, which has not been formally validated in Bangladeshi population. Formal inter-rater reliability was not calculated and outcome assessors were not blinded to the group allocation, which may introduce potential measurement bias. Nevertheless, structured questionnaires and predefined scoring criteria were used to reduce subjective variation. Even though the attendance at BCC sessions was documented, these records do not capture the intensity of participant engagement or to the extent to which recommended behaviors were implemented. Furthermore, detailed quantitative dietary intake data were unavailable. Although, dietary data were collected using a five-day 24-hour recall based on FAO guidelines, recall bias may have affected accuracy and the representation of true dietary habits. Finally, the study was conducted among female university students in a specific setting, which may limit the generalizability of the findings to other populations or contexts. Overall, the findings indicate an association between BCC module exposure and reported dysmenorrhea patterns. However, given the absence of baseline outcome data, potential self-selection, non-random design, and possible contamination, the result should be interpreted as associative rather causal.