Methods
Data
Data were from the 2019-21 National Family Health Survey (NFHS-5) of
India. The NFHS-5 is a nationally representative survey that provides
sociodemographic and health information of reproductive-aged women in
India.16 Our sample comprised 528,816 ever married
women aged 20-49 years. We used publicly available anonymized data that
met the definition of the NIH exempt human subject research (exemption
4). Therefore, ethics committee approval was not required for this
study.
Measures
Child marriage and Adolescent Childbearing Exposure Variable :
Based on the NFHS-5 report of respondent’s age at first marriage and
detailed record of births, we identified women who were married before
the age of 18 years and who gave birth by the age of 19 years. Our
exposure variable was a categorical variable that took four mutually
exclusive categories – i) married as adult – did not give birth in
adolescence (reference category), ii) married as adult – gave birth in
adolescence, iii) married as child – did not give birth in adolescence,
and iv) married as child – gave birth in adolescence.
Hysterectomy Outcome Variables. In the NFHS-5, women were asked
if they had undergone an operation to remove their uterus. Those who
answered yes to this question were determined to have a hysterectomy. As
such, our outcome variable is a binary variable indicating whether a
women did or did not undergo a hysterectomy. Women who reported to have
hysterectomy by age 19 years (N=257 – 0.05% of the total sample,
1.23% of all hysterectomies) were excluded in line with our goal of
examining the role of marriage before age 18 years and childbearing by
age 19 years on likelihood of hysterectomy between age 20 to 49 years.
Those who had a hysterectomy were further asked how many years ago the
procedure was performed. Subtracting this period from the respondent’s
current age (in years), we calculated the age at which the hysterectomy
was performed. This age construct was used for survival analyses.
Respondents who had a hysterectomy were also asked about the reason for
the hysterectomy. These options included: i) excessive menstrual
bleeding and/or pain, ii) fibroids/cysts, iii) uterine rapture, iv)
cancer, and v) other (uterine prolapse, sever postpartum hemorrhage,
cervical discharge, and other). We separately examined the likelihood of
hysterectomy due to each of these causes.
Statistical analysis
We first estimated percentage of women who underwent a hysterectomy by
the four categories of child marriage and adolescent childbearing. We
performed adjusted Wald tests to examine whether the percentages varied
across the groups.
Next, we estimated binomial logistic regression models to obtain the
odds ratios of hysterectomy for child marriage and adolescent
childbearing categories compared to the reference category of married as
adult and did not give birth in adolescence. We fitted the models for
all cause hysterectomy as well as for five cause-specific hysterectomy
(e.g., hysterectomy because of uterine rapture) outcomes. Of note, when
one gynecologic problem specific hysterectomy was assessed, hysterectomy
due to other problems were excluded from the analysis. For example, in
the model where hysterectomy due to uterine rapture was the outcome
variable, hysterectomy due to all other causes, i.e., excessive
menstrual bleeding, fibroids/cysts, cancer, and other were excluded.
We estimated the models both with and without covariates. In the
multivariable specification, we accounted for sociodemographic
attributes including age, educational attainment, religion, caste.
urban/rural residence, and household wealth index quintiles. Further, we
accounted for women’s body mass index (BMI) categories and parity (i.e.,
number of children born), which are regarded as risk factors for
hysterectomy. Lastly, to account for state-specific differences in
women’s health issues, we controlled for state of residence fixed
effects. These covariates were included in the model to enhance the
internal validity of our estimates of the relationship between
hysterectomy and child marriage and adolescent childbearing.
Next, to mitigate the influence of socioeconomic heterogeneities on the
relationship, we estimated the models by sub-groups of household wealth,
women’s educational attainment, urban/rural residence, and geographic
regions (North, Central, East, Northeast, West, and South). All models
were estimated using the complex survey weights of the NFHS-5. These
results are presented in the Supplementary Materials file.
Lastly, we estimated non-parametric Kaplan-Meier (K-M) survivor
functions for the event of having a hysterectomy for the four
child-marriage and adolescent-childbearing groups. We performed log-rank
tests to examine the equality of survivor functions across the groups.
Statistical analyses were performed using Stata 18.0 software.