Study Uncovers How Mask Efficiency and Cost Shape Post-Pandemic Adoption

A comprehensive public health investigation examining preventive practices in Bangladesh has revealed that community face mask adoption is governed by an intricate interplay of perceived filtration efficacy, household purchasing constraints, and geographic context rather than uniform behavioral compliance. Published in Discover Public Health, the study analyzed cross-sectional survey data alongside advanced machine learning techniques and compartmental epidemiological modeling to evaluate post-pandemic health choices across diverse residential environments. Researchers evaluated how stated attitudes, economic valuations, self-reported physical symptoms, and immunization histories shaped everyday protective decisions. While public endorsement of face masks remained exceptionally high throughout the study population, individual willingness to purchase specific protective coverings varied substantially depending on product pricing and perceived functional quality. The findings demonstrate that public health initiatives cannot treat protective masks as behaviorally interchangeable commodities, showing instead that sustained adoption requires nuanced intervention strategies that balance technical filtration performance with household economic affordability.

The investigation utilized a cross-sectional survey design incorporating retrospective recall components, administering face-to-face interviews to adults across seven purposively selected districts representing urban and rural landscapes in Bangladesh, including Dhaka, Narayanganj, Bhola, Faridpur, Barisal, Pabna, and Lakshmipur. Data collection occurred in a single administration per participant between October 12, 2024, and November 20, 2025, following the formal declaration by the World Health Organization that coronavirus disease no longer constituted a public health emergency of international concern. Out of 3,250 invited individuals, 3,100 completed the structured questionnaire, achieving an overall participation response rate of 95.4 percent. The sample comprised 55 percent male and 45 percent female respondents, with 48 percent residing in rural villages, 30 percent in regional urban centers, and 22 percent in the capital city. Occupational backgrounds spanned students at 30 percent, unemployed individuals at 27 percent, self-employed workers or business owners at 23 percent, and formal public or private sector employees at 19 percent, with 82 percent reporting monthly household incomes below 50,000 Bangladeshi Taka.

Initial evaluations demonstrated widespread normative support for protective face coverings across all surveyed demographic strata, although statistical significance testing revealed distinct analytical nuances. When asked about general protective habits, 59 percent of respondents reported that they always prioritized family safety by wearing masks, 72 percent affirmed that mask usage represented the correct course of action, and 86 percent maintained that wearing masks was never trivial or unnecessary. Furthermore, 88 percent affirmed that masks remained necessary even when others wore them or after receiving vaccinations. Pairwise chi-square analyses across 120 demographic and attitudinal combinations indicated that 85 percent reached nominal statistical significance at the one-percent level. However, calculating bias-corrected Cramér’s V effect sizes revealed that 53 of 56 associations were practically negligible to small, with a median value of approximately 0.07. Only residential environment paired with the perception of masking as a personal choice versus public responsibility demonstrated a moderate effect size of 0.35, illustrating that statistical significance in large samples often reflects statistical power rather than substantive practical magnitude.

Economic preferences for different mask varieties were evaluated using a double-bounded dichotomous-choice contingent valuation framework, establishing monetary valuation thresholds across three distinct product categories characterized by different filtration capabilities. Mask A represented high-efficiency respirators with approximately 95 percent filtration efficiency, Mask B denoted surgical and multi-layer cloth alternatives offering approximately 75 percent efficiency, and Mask C represented basic single-layer cloth coverings with approximately 30 percent efficiency. Stated willingness to pay proved highest for Mask A, yielding a sample mean of 83.91 Taka, a standard deviation of 62.97 Taka, and a median valuation of 100 Taka. Mask B generated an intermediate mean willingness to pay of 52.28 Taka, a standard deviation of 35.16 Taka, and a median of 60 Taka. Mask C recorded the lowest valuation, averaging 17.12 Taka with a standard deviation of 17.55 Taka and a median of 10 Taka. Linear regression across aggregate sample means demonstrated a strong direct correlation between perceived efficiency and willingness to pay, described by the equation y = 0.99x − 14.62 with a coefficient of determination of 0.97, reflecting an additional valuation of roughly one Taka per percentage point of filtration efficiency, while indicating an unwillingness to purchase masks with under fifteen percent efficiency.

Contingent valuation price-acceptance curves underscored sharp price elasticity across all three categories, revealing that purchase acceptance fell consistently as bidding increments escalated. Mask C demonstrated near-universal acceptance at nominal prices, securing 97.4 percent approval at five Taka and 70.3 percent at ten Taka, but dropped to 45.7 percent at 25 Taka and 11.6 percent at 60 Taka, establishing a 50 percent acceptance threshold near 22 Taka. Mask B peaked at 91.6 percent acceptance at 30 Taka before descending to 60.0 percent at 60 Taka and 19.8 percent at 100 Taka, corresponding to a 50 percent threshold around 70 Taka. Mask A reached 80.1 percent acceptance at 50 Taka and 68.9 percent at 100 Taka, falling to 22.5 percent at 150 Taka with a median threshold near 125 Taka. Both higher-tier masks exhibited anomalously lower acceptance at their lowest starting bids of 20 and 30 Taka, creating localized non-monotonic curve segments. Demographically, Mask A was favored among younger adults under 25 years old, university students, and higher-earning households, whereas Mask C was predominantly selected by capital city residents and individuals aged 41 to 55.

To isolate independent predictors of uptake, researchers fitted a multivariable logistic regression model evaluating willingness to purchase high-efficiency Mask A against comprehensive sociodemographic, health, and attitudinal indicators. The belief that mask-wearing reduces disease transmission emerged as the strongest independent correlate of willingness to buy, exhibiting an adjusted odds ratio of 4.68 with a 95 percent confidence interval spanning 3.28 to 6.66. Residence in regional urban areas relative to rural villages increased purchase odds to 2.16, while university and college education levels demonstrated elevated odds ratios of 1.94 and 1.93, respectively. Monthly household income above 80,000 Taka produced an odds ratio of 1.84, prior COVID-19 infection showed an odds ratio of 1.45, male gender recorded 1.40, and viewing mask use as a civic responsibility yielded 1.35. Conversely, living in the capital city Dhaka markedly lowered purchase willingness to an adjusted odds ratio of 0.47 relative to rural villages. Chronic medical conditions were associated with reduced odds of 0.77, and each one-unit increase in composite worry score corresponded to an odds ratio of 0.91, while age and vaccination status showed no statistically significant adjusted associations.

Complementing traditional inferential statistics, supervised machine learning models utilizing random forest and extra trees classifiers were optimized through five-fold cross-validation on an eighty percent training partition to capture complex nonlinear dependencies and rank variable importance. Evaluating out-of-sample test data, permutation feature importance for Mask A confirmed that area of residence was the single most influential determinant of predictive discrimination, producing an average receiver operating characteristic area reduction of 0.096 when permuted. Composite worry scores represented the second most influential predictor with an importance metric of 0.046, followed by educational attainment at 0.028, conviction regarding mask effectiveness at 0.026, age at 0.020, and gender alongside income within an intermediate tier between 0.018 and 0.020. Conversely, self-reported chronic disease and previous COVID-19 infection contributed negligibly to predictive accuracy at 0.006 and 0.001, respectively. These machine learning findings confirmed that geographic context, structural socioeconomic standing, and cognitive risk perceptions exerted far greater predictive leverage over purchasing choices than documented clinical history alone.

Comparative machine learning modeling across all three protective products established that each mask catered to distinct behavioral archetypes within the population. Cross-validated receiver operating characteristic analyses demonstrated robust out-of-sample discrimination across all fitted algorithms, recording area under the curve metrics of 0.821 for Mask A, 0.840 for Mask B, and 0.804 for Mask C. Pairwise dominance matrices based on individual-level predicted purchase probabilities indicated that Mask A surpassed Mask B for 77.0 percent of respondents and exceeded Mask C for 58.5 percent, while Mask C surpassed Mask B for 56.2 percent, yielding a macro-level preference ordering of A over C over B. Subgroup recommendation profiling demonstrated that Mask A dominated across all income categories among participants under 25 years old and among upper-income older adults. Conversely, Mask C proved optimal for middle-aged and lower-income cohorts aged 26 to 55 years. Notably, Mask B never achieved highest predicted uptake in any demographic stratum despite yielding the highest classifier area under the curve, confirming that statistical classification discrimination measures a fundamentally different property than population-level product appeal.

To explore the broader epidemiological dynamics of disease transmission and behavioral feedback, the authors formulated a compartmental epidemiological model integrating evolutionary game theory replicator dynamics across five distinct population states: susceptible, masked, susceptible-infected, masked-infected, and recovered individuals. Configured with transmission rate beta at 0.8333 and recovery rate gamma at 0.3333, theoretical simulations predicted that maximum mask utilization would emerge when product cost and protective efficiency were both minimal, even though this parameter space yielded a high final epidemic size. In contrast, empirical survey observations demonstrated that real-world mask adoption peaked when efficiency was high and cost remained low. Furthermore, while the theoretical formulation assumed gradual declines in adoption as costs escalated, survey data demonstrated an abrupt abandonment of protective equipment once prices crossed moderate affordability thresholds. This critical discrepancy underscores that conventional epidemiological models underestimate the deterrent power of direct out-of-pocket expenses in resource-constrained developing economies, requiring the integration of empirical demand elasticity curves to improve forecasting precision.

Analysis of self-reported health complaints experienced during the seven days preceding the survey illustrated that community-level symptom burden remained multidimensional in the post-pandemic era. Although 65.4 percent of participants reported no acute symptoms, symptomatic respondents reported a median of two concurrent health problems. Fever was the most prevalent complaint at 12.9 percent, followed closely by cough at 12.1 percent and headache at 10.5 percent, while secondary manifestations included generalized tiredness at 7.4 percent, musculoskeletal pain at 7.1 percent, and upper nasal congestion at 6.8 percent. Age stratification revealed substantial variation across life stages: individuals under 25 years old experienced primarily acute febrile and upper respiratory complaints, including fever at 16.3 percent and cough at 15.6 percent. Conversely, respondents aged 56 years and older exhibited heightened constitutional and musculoskeletal complaints, with tiredness reaching 15.8 percent and joint pain reaching 15.3 percent. Overall, the proportion of participants experiencing at least one symptom increased from 29.9 percent among adults aged 26 to 40 up to 41.5 percent in older adults, while COVID-19 vaccination coverage reached 94.6 percent across the sample, with primary two-dose completion concentrated among youths and third-dose booster uptake concentrated among older age cohorts.

The authors identified several methodological boundaries that contextualize the interpretation and generalizability of their conclusions. Because the investigation relied on a cross-sectional framework with retrospective recall, observed statistical associations cannot confirm temporal sequence or establish direct causal mechanisms. All behavioral practices and medical symptoms were self-reported without independent clinical examination, laboratory verification, or administrative auditing, introducing potential vulnerability to recall inaccuracies and social desirability biases, particularly given established community norms supporting facial coverings. Furthermore, purposive sampling across seven geographic districts limits claims of national representativeness, and machine learning models were assessed through internal cross-validation rather than testing on external geographic cohorts. Consequently, researchers caution against treating predictive outputs as definitive forecasts of individual behavior. Nonetheless, by uniting contingent valuation economics, compartmental transmission modeling, and explainable predictive algorithms, the study provides valuable empirical evidence demonstrating that sustained post-pandemic protection in resource-constrained environments depends upon targeted public subsidies, localized risk communication, and distribution frameworks aligned with household purchasing realities.

Subject of Research: Socioeconomic determinants of mask adoption, valuation, and post-pandemic symptom burden in Bangladesh

Article Title: Mask adoption, symptom burden, and machine learning insights from Bangladesh during and after COVID-19

Article References: Kabir, K. M. A., Shuvo, M. F. R., & Zobayer, A. (2026). Mask adoption, symptom burden, and machine learning insights from Bangladesh during and after COVID-19. Discover Public Health, 23(1), Article 1803. https://doi.org/10.1186/s12982-026-03303-0

Image Credits: AI Generated

DOI: 10.1186/s12982-026-03303-0

Keywords: face masks, COVID-19, Bangladesh, public health, machine learning, contingent valuation, willingness to pay, symptom burden, epidemic modeling, game theory, random forest, preventive behavior, vaccination, respiratory symptoms, socioeconomic factors, cross-sectional study