Assessing the link between hygienic material use during menstruation and self-reported reproductive tract infections among women in India: a propensity score matching approach

Data source

The present study used data from the NFHS-5, which was conducted between 2019 and 2021. NFHS-5, a nationally representative series of cross-sectional surveys, is designed to capture an array of vital information, including demographics, socioeconomics, maternal and child welfare, reproductive health, and family planning (Ministry of Health and Family Welfare, Government of India, 2021). Employing a two-stage stratified sampling method, a total of 664,972 households were selected for the survey, among which 636,699 were successfully interviewed, with a response rate of 98%. Additionally, within the households that were interviewed, a total of 747,176 women aged 15–49 were identified for individual interviews. Among them interviews were completed for 724,115 women aged 15–49, with a response rate of 97%. Among the total interviewed women, only 108,785 women were selected for the questions included in the state module. From those, ultimately 27,983 women aged 15–24 were chosen as a final sample for this study, as illustrated in Fig. 1.

Ethics statement

This study relies on publicly available, de-identified dataset that does not contain any personally identifiable information about the survey participants. As such, there was no requirement for ethical permission. The NFHS-5 data utilized in this study is publicly accessible through the official website of Demographic and Health Surveys (DHS) at https://dhsprogram.com/data/available-datasets.cfm. It can be obtained by submitting a formal request to the DHS.

Outcome variable

In NFHS-5, women who were part of the state module and who reported being sexually active, regardless of their marital status, were asked two following questions:

“During the last 12 months, have you had a bad-smelling abnormal genital discharge?”

“During the last 12 months, have you had a genital sore or ulcer?”

From the responses obtained for the above questions, we created a binary outcome variable, with two categories: (a) ‘women with RTIs’(coded as 1), i.e., women who reported either or both of the symptoms mentioned above (abnormal genital discharge and/or genital sore or ulcer), (b) The remaining women were categorised as ‘women without RTIs’(coded as 0).

Treatment variable

In the NFHS-5 survey, women aged 15–24 were asked about the methods they used to prevent bloodstains from becoming evident during their menstrual periods. The responses were categorized into two groups:

Hygienic Materials: This category included sanitary napkins, locally made napkins, tampons, and menstrual cups.

Unhygienic Materials: This category included cloth, using nothing (implying no use of any protective material), and other materials not falling into the hygienic category.

Based on the responses, we created a binary outcome variable with two categories:

Exclusive users of hygienic materials (coded as “1”): These are women who exclusively utilized hygienic materials such as sanitary napkins, locally made napkins, tampons, and/or menstrual cups.

Users of unhygienic materials (coded as “0”): This category includes women who used cloth, other materials, nothing (indicating no use of any protective material), or used a combination of both hygienic and unhygienic materials (Singh et al., 2022a; Singh et al., 2022b).

Throughout this article, we have consistently used the terms ’use of hygienic materials’ and ’exclusive use hygienic materials’ interchangeably to convey the same concept. In our context, both terms refer to the practice of relying solely on hygienic materials, such as sanitary napkins, locally made napkins, tampons, and menstrual cups, for menstrual protection, without combining them with unhygienic materials or practices.

Matching variables

In this study, a range of pertinent socioeconomic and demographic predictors identified in previous research on RTIs and MHM were selected as matching variables (Ghebremichael, Paintsil & Larsen, 2009; Anand, Singh & Unisa, 2015; Brown, Gause & Northern, 2016; Torondel et al., 2018; Wagenaar et al., 2018; Vishwakarma, Puri & Sharma, 2020; Das et al., 2021; Cho & Yang, 2023). These variables included: current age of respondents (in years), age at menarche (in years), years of schooling (in years), social groups (Scheduled Caste/Scheduled Tribe or SC/ST = 1, non-SC/ST = 0), religion (Hindu = 1, non-Hindu = 0), household wealth factor scores, exposure to mass media (exposure to at least one medium = 1, no mass media exposure = 0), discussion of menstrual hygiene (MH) with community health workers(CHW) (discussed = 1, not discussed = 0), working status (currently working = 1, not working = 0), bath during menstruation (take bath = 1, do not take bath = 0), consumption of alcohol (yes = 1, no = 0), place of residence (rural = 1, urban = 0), and region of residence (north, central, east, west, south, north-east).

Statistical analysis

We begin by presenting a summary of the sample characteristics of the women (aged 15–24 years) included in our study. Subsequently, we calculate the prevalence of RTIs over the past year for these sampled women. We employed a Propensity Score Matching (PSM) approach to assess the impact of hygienic material use on RTI prevalence among women in India (Rosenbaum & Rubin, 1985; Dimewiki, 2019; Kane et al., 2020). For the analysis, we utilized the psmatch2 command in Stata16 statistical software (StataCorp, 2019).

When randomized controlled trials, often regarded as the gold standard for assessing treatment effectiveness, are not feasible, and we are working with cross-sectional, observational, or non-experimental data, as in this study, quasi-experimental methods like PSM can serve as valuable tools for evaluating treatment effects (Rosenbaum & Rubin, 1985; Dimewiki, 2019; Kane et al., 2020). PSM replicates the randomization process of randomized controlled trials, enabling robust causal inference, particularly when researchers have limited control over treatment assignment. By systematically matching individuals with similar probabilities of receiving treatment, PSM effectively addresses selection bias and creates comparable treatment and control groups.

In this study, we utilized the 1:1 nearest neighbor matching method with a caliper of 0.2 standard deviations of the logit of the propensity score. This approach was chosen based on recommendations from Rosenbaum & Rubin (1985) and Austin (2011) for its effectiveness in addressing imbalance and reducing bias by matching propensity scores between the treatment and control groups (Rosenbaum & Rubin, 1985; Caliendo & Kopeinig, 2005; Austin, 2011; Greifer, 2022).

Additionally, there are multiple reasons behind choosing 1:1 nearest neighbour matching over other matching methods (e.g., k:1 nearest matching, kernel or local linear matching etc.). First, 1:1 nearest neighbour matching is relatively straightforward to implement. It involves pairing each treated unit with one control unit that is closest in terms of covariate values (Greifer, 2022). Second, it requires fewer computational resources compared to more complex matching methods, making it practical for large datasets such as NFHS, which is why several studies utilizing the NFHS dataset have consistently favoured nearest neighbour matching over other complex matching methods (Dixit, Dwivedi & Ram, 2013; Dixit, Gupta & Dwivedi, 2018; Wilkinson, Mamas & Kontopantelis, 2022). Third, this matching is more interpretable because each treated unit is directly matched to one control unit (Stuart & Rubin, 2007). This pairing can simplify the analysis and make it easier to explain to a non-technical audience. Furthermore, recent research by Greifer (2022) lends support to the preference for 1:1 nearest neighbor matching over k:1 matching, as an increase in k in the latter method can potentially exacerbate imbalance (Greifer, 2022). Also, nearest neighbor matching is better than kernel or local linear matching methods due to the latter occasionally including observations that are poor matches (Caliendo & Kopeinig, 2005).

The matching quality and balance were assessed by calculating the mean and median bias before and after matching using the pstest command in Stata 16. Sensitivity analysis was conducted using the Mantel-Haenszel bounds proposed by Becker to estimate the significance levels at different levels of hidden bias (Becker & Caliendo, 2007). The details of PSM, balancing test, and sensitivity analysis are provided in the appendix (see Appendix S1). For all analyses and estimations, the significance level was established at 0.05, corresponding to a confidence level of 95%.

Sample characteristics of the respondents

Table 1 represents the weighted percentage of women by background characteristics. The majority experienced their first menstruation between the ages of 13–15 years. About 47% of women were OBC, and most were Hindu. Only a few women had discussed MH with CHW in the past three months. About 16% of women reported no exposure to mass media. More than two-thirds of the women resided in rural areas. Half of the women reported exclusive use of hygienic materials during menstruation, while approximately 4% reported experiencing symptoms of RTIs.

Prevalence of RTIs by use of hygienic materials

The results of the bivariate analysis demonstrated a significant association between the utilization of hygienic materials during menstruation and the prevalence of reproductive tract infections (RTIs) among women. Specifically, only three out of 100 women (3.2% (95% CI [2.8–3.6])) who used hygienic materials during menstruation reported symptoms of RTIs. In contrast, about 5% (4.9% (95% CI [4.4–5.5])) of women who did not use hygienic materials during menstruation reported symptoms of RTIs (see Fig. 2).

Spatial variation in the prevalence of RTIs and use of hygienic materials

Figure 3 reveals, the prevalence of RTIs was not homogeneously distributed across the Indian states. Among the larger states, highest prevalence of RTIs among women were observed in West Bengal (9.3%) followed by Himachal Pradesh, Rajasthan, Uttar Pradesh, and Gujarat (more than 4%). Among the smaller states, highest prevalence of RTIs was observed in Meghalaya, followed by Arunachal Pradesh, Sikkim, and Tripura. Across these above-mentioned states, the prevalence of RTIs was far more than national average of 4%. Among the larger Indian states, lowest prevalence of RTI was identified in Odisha (1.3%), followed by Uttarakhand and Maharashtra (less than 2%).

Among the UTs, the highest prevalence of RTIs among women was reported from Chandigarh (5.9%), followed by Delhi (3.7%). In contrast, the lowest prevalence of RTI was recorded in women from Puducherry (0.9%).

Figure 4 illustrates the spatial distribution of hygienic material use among women. Approximately 80% of women from Tamil Nadu, followed by Telangana and Haryana, reported use of hygienic materials during menstruation. Conversely, the lowest rates of hygienic material use were reported by women from Uttar Pradesh (30.8%), followed by Chhattisgarh (31.0%), Bihar (34.2%), and Madhya Pradesh (35.5%).

Among the UTs, Chandigarh (94.6%) exhibited the highest use of hygienic materials, closely followed by Andaman & Nicobar Islands (89.2%). In contrast, the lowest use of hygienic material was reported by women from Ladakh and Jammu & Kashmir (below 47%).

Impact of hygienic material use on RTIs

The results presented in Table 2 provide estimates of the impact of hygienic material use during menstruation on RTIs among women in India.

Before matching, the initial estimates from the unmatched sample indicated that in the control group (women who did not use hygienic materials during menstruation), 48 per 1,000 women reported symptoms of RTIs. In contrast, in the treated group (women who used hygienic materials during menstruation), a lower proportion of women, specifically 31 per 1,000, reported symptoms of RTIs. These initial estimates from the unmatched sample suggest that there was a notable difference in the prevalence of RTIs between the two groups of women. To provide a more accurate assessment of the impact of hygienic material use on RTI prevalence, we employed PSM.

After matching, the ATT estimates indicated that among women not using hygienic materials (control group), approximately 41 per 1,000 women reported symptoms of RTIs (ATT: 0.0416). In contrast, among the women using hygienic materials (treated group), 31 per 1,000 women reported symptoms of RTIs (ATT: 0.0315). These ATT estimates suggested that even after the matching process, there remained a significant difference of 10 per 1,000 RTI cases between women who used hygienic materials and those who did not. This implies that the use of hygienic materials is associated with a lower prevalence of RTIs among women.

Verification of estimates

Quality of matching

Figure 5 displays the distributions of propensity scores between two groups of women: those who reported using hygienic materials during menstruation (treated group) and those who did not (control group). In this figure, control group women’s propensity scores are illustrated as bars below a vertical line, whereas treated women’s scores are presented as bars above the line. Through propensity score matching, it is observed that the distributions of propensity scores in the treatment and control groups become comparable, as depicted in the figure. This concordance suggests that the characteristics of the treated and untreated women were strikingly similar.

Balancing test

Figure 6 indicates the standardized % bias of each matching variable was dissimilar prior to matching. Notably, the standardized % bias was observed to be most pronounced in the variables of respondents’ household wealth status, years of schooling, place of residence, and mass media exposure. However, after matching the differences in standardized % bias of nearly all covariates were sufficiently reduced. For instance, after the matching the % bias was reduced (see % reduction bias) by more than 80% in almost each matching variable (see Appendix S3). Also, after the matching, the % bias reduced to less than 10% for all variables, which indicates a good matching (Singh, 2016). Furthermore, after the matching procedure, the p-values resulting from t-tests mostly exceeded 0.05, suggesting the absence of statistically significant differences between the groups. However, two variables—age and wealth factor score—still exhibited p-values >0.05, indicating statistically significant differences between the groups.

Moreover, after matching, the variance ratios for each matching variable approached a value of 1, including age and wealth factor score. This signifies a favourable balance between the control and treated groups (Zhang et al., 2019) (see variance ratios in Appendix 3). In simpler terms, it implies that our matching process successfully brought the groups into closer alignment, even for variables that initially showed substantial differences.