Methods

Study sites and participant selection procedures

We conducted an institutionally based retrospective cohort study between July 2017 and January 2018 at 60 public hospitals in China. Study participants were women who had given birth between 12 and 24 months earlier. First, we divided the country into three macrogeographic regions: eastern, central, and western. This is because there are huge socioeconomic differences and landscape variations among the three regions: Eastern China is the most developed and located near the eastern coastline of the country. Western China is the least developed and dominated by towering massifs and rolling plateaus. Central China is largely an agricultural area and generally less developed than the eastern coastal provinces but more developed than the western provinces. Second, we selected five provinces in each region (see Supplementary material, Figure S1), and appointed a senior physician or researcher as the provincial coordinator for each province. Third, we asked the provincial coordinators to provide a list of 10 hospitals with at least two each level-2 and level-3 general hospitals and two each level-2 and 2 level-3 Maternal and Child Hospitals, and this defined a sampling frame. A statistician randomly selected four hospitals from each province using computer-generated random numbers, randomly selecting a month between July 2015 and June 2016; and allocated it to each participating hospital as the initial month of investigation. The first 300 women who delivered a live birth in the selected month were eligible for this study. If the number of deliveries in the given month was less than 300, women who delivered in the following month were invited for interviews.

Sample size calculation

The sample size was calculated based on a prevalence (P) of unintended pregnancy of 8% among postpartum women in a given region. With a 0.01 uncertainty level (d), 95% confidential interval (CI), and 5% margin of error, and using the formula N =Z2 0.05P(1-P)/d2, we calculated a required sample size of 2828. As we used a cluster (hospital) sampling technique to select the participants (as proposed by Donner, Birkett, and Buck), the sample size calculated assumed that individual randomization could be inflated by a design effect (DE) to reach the required level of statistical power under cluster randomization: DE = 1 + (n-1)ρ, where n was the number of individuals per cluster and ρ was the intracluster correlation coefficient (ICC).(15) When the primary outcome variable is a rate, the coefficient of variation in the outcome can be used as an alternative measure to the ICC, and is denoted by κ(16) (κ is usually derived from previous studies). However, we did not find any relevant studies where investigators reported the value of κ. Based on experience, then, we set DE = 2, assuming a κ of 0.003 (using data from this study, we derived a κ of 0.001). We thus obtained a sample size of 5656 for each geographic region, or 283 participants for each hospital. We decided to recruit 300 participants at each hospital, which allowed for a 6% non-response rate.

Data collection

Data were collected with a structured questionnaire via face-to-face or telephone interviews. Two junior doctors and/or nurses selected as data collectors at each hospital were trained on the data collection procedures and the interview techniques. Recorded information included participants’ background characteristics, pregnancy history, immediate postpartum contraceptive use, resumption of sexual activity, return of menstrual cyclicity, and occurrence and time of subsequent pregnancy and/or induced abortion after childbirth. All of the participants were interviewed between August 2017 and January 2018.

Data processing and analysis

We defined long-acting reversible contraceptives (LARCs) as intrauterine devices (IUDs) and implants. Modern contraceptive methods were defined as using one or more of the following method(s): male and female sterilizations, male and female condoms, combined oral pills (COCs), injectables, implants, IUDs, emergency contraceptive pills (ECPs), and spermicides. Traditional methods were defined as the use of the lactational amenorrhea method (LAM), rhythm/calendar method, and withdrawal. We defined unmet need for postpartum contraception as women who resumed sexual intercourse after delivery but did not use any contraceptive methods or only used traditional methods.
Data were entered into EPI INFO version 3.1, and we used SAS 9.4 (SAS Institute Inc.) for data analysis. Frequencies, percentages, means +/- SD, and cross-tabulations were used to summarize descriptive statistics. Life-table analyses were used to calculate cumulative rates of contraceptive use, unintended pregnancy, and induced abortion during the first 24 months postpartum. We used the log-rank test to compare trends and Chi-squared test to compare rate differences between/among regions.

Ethical considerations

Ethical approval was obtained from the Institutional Review Board of Shanghai Institute of Planned Parenthood Research. Written (face-to-face interviews) and verbal (telephone interviews) consent was obtained from all the study participants, and they were assured that the information would remain confidential.