Setting
This study uses data from women who adopted a modern family planning method from health facilities in Uttar Pradesh, India. The study is part of a five-year reproductive health quality improvement project in the India and Kenya. Ethical review and approval of all study documents was provided by the respective research institutions in each country and coordinating US-based university.
In India, the study was administered in nine peri-urban secondary level government health facilities across two districts in Uttar Pradesh. The study sites ranged from a 30-bedded Community Health Centers to a four bedded Primary Health Centers. All are government health facilities and provide free family planning services including pills, condoms and IUDs. Women who adopted sterilizations were dropped from this analysis. Facilities in India had low family planning case loads ranging from 7 to 166 cases per month based on health system data for the facilities from July-September 2017 (Ministry of Health and Family Welfare, Government of India, 2017).
Surveys
The family planning client survey included questions on demographics, birth history, and current family planning method, in addition to the PCFP indicators. Data were collected and stored on tablets using the SurveyCTO platform, and uploaded on the same day to a secure/encrypted server upon obtaining internet connection. Data collection was monitored through a range of quality assurance checks throughout the survey, including interview observations, high frequency checks, backchecks and spot checks by field supervisors.
A team of six female enumerators underwent a one-week training on the study topic, quantitative data collection methods, best practices for surveying, informed consent and recruitment, and the survey tool itself. Then, the team went to the field for piloting for one week in Kenya and two days in India.
Recruitment procedures and eligibility criteria are thoroughly described in the recent PCFP validation paper 3. To summarize, the eligibility criteria were women who had obtained a modern family planning method at the facility on the day of recruitment aged 18-49 years. This criteria excluded women coming in for a new pill pack due to the limited interaction that may be involved, but included women starting pills as a method or starting a new type of pills.
Survey data was collected between September 2016 and March 2017 in a phased manner across the nine selected facilities. Based on available government family planning service data, the target sample size was set at 88 women per facility. However, the actual user numbers were much lower than indicated in the available secondary data. Despite extending the data collection timeline and near-universal enrolment of all eligible women at each facility, a much lower family planning sample was achieved in India than initially anticipated. No refusals or drop-outs occurred during the survey. All interviews were conducted in Hindi, and in a secluded space within the facility. Per the recommendation of local partners, no incentive was given in India. At the end of the interview, women were requested for brief, follow-up phone survey at six weeks post baseline. All respondents agreed to the follow-up survey. A total of 225 women were interviewed in the baseline, of which 179 women were also followed up after 6 weeks (loss to follow-up was due to wrong numbers or women not answering the phone after multiple attempts).
Variable construction
Dependent variable: The primary outcome of interest in this analysis is whether women were still using the family planning method that they adopted at baseline at the follow-up interview. All women in the sample adopted a method at baseline. The follow-up question simply asked if they were still using the method they adopted at baseline (with no information about stops/starts). Women who had switched to another method (N=8) were dropped from the analysis. We use data from follow-up interviews conducted at 6 weeks.
Independent variable: Three main quality indicators are examined for their association with method continuation in this analysis. The first is a binary variable on whether a woman reported that her provider was involved the right amount (compared to the provider being involved too much or too little), henceforth referred to as the provider being sufficiently involved 5,8. The second quality indicator is a binary variable created from a question on whether the woman felt that her provider had no, slight, moderate, strong or extremely strong preference for what family planning method she adopted. The binary grouped moderate/strong/extremely strong preference together, with the comparison group of no/slight preference 6.
The third quality variable examined is a summary score of the Person-Centered Family Planning Scale (PCFP). The development and validation of the scale in India and Kenya is described in detail elsewhere 3. The two domains identified were “Autonomy, Respectful Care and Communication” (ARCC) and “Health Facility Environment” (HFE). Example items from the ARCC domain includes trust in her provider, whether the providers introduced themselves, whether the provider called the woman by her name, whether she received respectful care, and whether she was involved in her care. Sample questions from the HFE domain included whether the facility had water, was safe, free from bribes, and had clean facilities. The PCFP scale has 22 items in India, with 17 items in the ARCC subscale and 5 items in the HFE model. A summary PCFP score was created using the validated scale 3.
Socio-demographic covariates
We included the following variables: age, marital status, education, parity, household wealth, caste, and religion. The inclusion of these covariates was motivated by our theoretical framework and previous findings about factors associated with family planning continuation and women’s experiences of quality. Age was modeled as a continuous variable. Education was grouped into three groups (no school/primary, secondary/vocational, college or above). Parity was grouped into 4 groups: 1 child, 2 children, 3 children and 4 or more (no women had no births). A wealth quintile variable was constructed by making a quintile of women’s reports of their total household income. A variable for caste was included which was a binary with low caste groups (Scheduled Caste, Tribe and Other Backward Caste) compared to high caste groups (“General” category). A binary variable was created for religion, with the dominant religion (Hindu) compared to all other minority religions.
Analyses
First, we describe the three quality variables and socio-demographics by method continuation status (Table 1). We then explore the correlation between the three quality measures (Table 2). Next, three separate Penalized maximum likelihood regression models are run, first without controlling for potential confounders and then controlling for socio-demographic variables, looking at the association between the three quality measures and family planning continuation (Table 3a and 3b). Penalized maximum likelihood regression models (using firthlogit) were necessary because discontinuing was a relatively rare event, and the overall sample sizes were also fairly small. Finally, we explore if there are differences in the associations with family planning continuation by PCFP sub-scale, again using a penalized maximum likelihood regression (Table 4). All analyses are run using STATA 15.
<Place table 1 and table 2 here>