Participant selection

We contacted 95 potential participants from the trial, of whom 19 (20.0%) could not be contacted or had moved away from the study area, and 76 (80.0%) were met in-person, consented, and enrolled into the pregnancy history survey between October 22, 2021, and November 18, 2021 (Fig. 1). Across all pregnancy outcomes, there were no significant differences in participant characteristics between the 76 enrolled participants and the 19 participants that could not be contacted (Additional file 1: Table S3). The proportion that could not be contacted varied by outcome: live birth (n = 3/29, 10.3%), stillbirth (n = 12/37, 32.4%), and neonatal death (n = 4/29, 13.8%) groups (Chi-squared test, p = 0.072). The 76 participants enrolled in the pregnancy history survey included women with at least one live birth that survived to 28 days (n = 26), stillbirth (n = 25), or live birth followed by neonatal death (n = 25), according to the prospective data, between January 4, 2015, and May 25, 2017. All participants in the trial were visited within 72 h after each of these pregnancy outcomes (median: 1 day).

Fig. 1

Participant flowchart and outcomes by data source

The characteristics of the 76 enrolled participants are presented in Table 1. There were no significant differences in participant characteristics by the three pregnancy outcome categories. However, women who experienced a stillbirth had lower literacy than those who experienced a live birth that survived to 28 days or live birth followed by neonatal death combined (illiteracy rate: live birth/neonatal death: n = 19/51, 37.3%, stillbirth: n = 3/25, 12.0%; p = 0.023).

Table 1 Participant characteristics by infant outcome from NOMS trial prospective dataNumber of pregnancy outcomes

In the prospective data, the 76 participants had a total of 122 pregnancy outcomes (median: 2, range: 1 to 4) between May 25, 2011, and May 25, 2017, during their follow-up in the NOMS trial (Table 2). In the pregnancy history survey, 76 participants reported having a total of 319 lifetime pregnancy outcomes (median: 4, range 1–11). Of the 319 pregnancy outcomes, 104 (median: 2, range 1–5) occurred within a time period for each participant spanning 30 days prior to the enrollment date and 30 days later than the last pregnancy surveillance or birth visit in the prospective data. Among the 76 participants, five had no pregnancy outcome dates in the pregnancy history survey within this time period. Of the 76 women, 56 (73.7%) reported the same number of pregnancy outcomes in both data sources, while 13 (17.1%) reported one fewer, 6 (7.9%) one more, and 1 (1.3%) two more than the pregnancy history in the time period defined. The time between the date of the 122 pregnancy outcomes in the prospective data and the date of the pregnancy history survey administration ranged from 4.4 years to 10.5 years (median 6.2).

Table 2 Pregnancy outcomes recorded in prospective data and pregnancy history surveyCompleteness of pregnancy outcome dates

All pregnancy outcome dates in the prospective data were fully recorded. In the pregnancy history survey, one-third of women (n = 29/76, 38.2%) were able to recall full dates for each of their pregnancy outcomes, including a specific day, month, and year, using the Nepali calendar. The other two-thirds (n = 47/76, 61.8%) recalled a partial date, either only the month and year or only the year, for at least one of their pregnancy outcomes. Among women who recalled at least one partial date, the number of pregnancy outcomes with a partial date ranged from 1 to 10. Of the 76 women, over half (n = 45/76, 59.2%) had at least one partial date due to non-recall of the day of birth (i.e., they were able to recall only the month and year), while ten percent (n = 8/76, 10.5%) had at least one partial date due to non-recall of the day and month (i.e., they were able to recall only the year). There were no observations with missing month but non-missing day and year; neither were there observations with missing year but non-missing day and month. In total, one-third (n = 108, 33.9%) of the 319 outcomes reported in the pregnancy history survey had a partial date. Of the 108 partial dates, 96 (88.9%) were missing only day and 12 (11.1%) both day and month. Among the 104 pregnancy outcomes from the pregnancy history survey that fell within the prospective observation period, and hence were available for matching, 41 (39.4%) of outcomes had a partial date, including most (n = 36, 87.8%) that were missing only day and 5 (12.2%) both day and month. Outcomes reported in the pregnancy history survey had different proportions of partial dates. Among the 319 outcomes, stillbirths (n = 25/31, 80.7%) and live births followed by neonatal death (n = 40/49, 81.6%) were more likely to have a partial date compared to live births that survived to 28 days (43/239, 18.0%) (Chi-squared test, p =  < 0.001). Among the 108 partial dates, stillbirths (n = 6/25, 24.0%) were more likely to have both missing day and month (rather than only missing day) compared to live births that survived (n = 2/43, 4.7%) and live births followed by neonatal death (n = n = 4/40, 10.0%) (Chi-squared test, p = 0.048).

Pregnancy outcome date match

We graphed all pregnancy outcomes from the complete dataset (n = 226) in Fig. 2, separated for clarity by the three pregnancy outcomes groups selected for inclusion in the validity study. Comparison of dates from the 122 and 104 pregnancy outcomes in the complete dataset resulted in 202 comparisons for date matching (Table 3). Of these, 30 (14.9%) dates matched exactly in the prospective and pregnancy history survey, while another 32 dates matched within ± 30 days, together constituting the 62 matched pregnancy outcomes in the matched dataset (n = 124 total). The differences between matched pregnancy outcome dates (prospective data minus pregnancy history survey) ranged from -28 to 17 days (Additional file 1: Fig. S2). The proportion of date matches by outcome was higher for live births that survived to 28 days (n = 43/65, 66.2%) than stillbirths (n = 9/25, 36.0%) or live births followed by neonatal death (n = 10/32, 31.3%) (p = 0.001). Stillbirth, neonatal death, higher parity, and delivery at a health facility were associated with occurrence of a non-matched pregnancy outcome (Table 4). The likelihood of matching did not vary by time from the pregnancy outcome in the prospective data to the date of the pregnancy history survey administration (non-match: mean 6.2 years vs. match: mean 6.4 years, t-test p-value: 0.649). Pregnancy outcomes that did not match within ± 30 days from both data sources are displayed in Additional file 1: Fig. S1. Results from a sensitivity analysis comparing pregnancy outcomes in both sources among outcomes matched by date with wider ranges (± 60 days, ± 100 days, etc.) is given in Additional file 1: Table S5.

Fig. 2

Pregnancy outcomes from prospective data and pregnancy history survey. The graphs below display pregnancy outcomes from the prospective data and pregnancy history survey that occurred within the prospective follow-up period for each participant plus 30 days before and after their observation. Specifically, this period was defined as the time from 30 days before the participant’s enrollment date in the prospective study and 30 days after either the last birth visit or last pregnancy surveillance visit (whichever was later) in the prospective study. This yielded 226 pregnancy outcomes, including 122 from the prospective data and 104 from the pregnancy history survey. We display pregnancy outcomes that did not match by date with ± 30 days, including those inside and outside the prospective follow-up range defined above, in Additional file 1: Fig. S1. The 76 participants enrolled in the pregnancy history survey included women with at least one live birth that survived to 28 days (n = 26), stillbirth (n = 25), or live birth followed by neonatal death (n = 25), according to the prospective data, between January 4, 2015, and May 25, 2017

Table 3 Measures of agreement and validity for infant outcomes for prospective data and pregnancy history surveyTable 4 Adjusted odds ratios and 95% confidence intervals of non-matched pregnancy outcome* for participant characteristics using logistic regression with generalized estimating equations~Infant sex and name match

Among 62 pregnancy outcomes with a matched date, most (n = 55/62, 88.7%) had the same sex in both sources, although 3 (4.8%) boys and 3 (4.8%) girls had no sex recalled in the pregnancy history survey. Among the 76 selected pregnancy outcomes, one-fifth (n = 15/76, 19.7%) had a name match between the pregnancy history survey and the prospective data. In the pregnancy survey history data, among the non-matched names (n = 61/76, 80.3%), over half (n = 37/61, 60.7%) were listed as either “baby” or blank, while the remainder (n = 24/61, 39.3%) had a name. Similarly, in the prospective data, the non-matched names were mostly (n = 52/61, 85.2%) “baby boy _” or “baby girl _” with many fewer having a full name (n = 9/61, 14.8%).

Stillbirth and neonatal outcome counts and measures of validity

Among the 226 outcomes in the complete dataset, a similar number of live births that survived to 28 days (n = 63/65, 96.9%) were reported in both sources, but fewer stillbirths (n = 15/25, 60%) and live births followed by neonatal death (n = 26/32, 81.3%) were reported in the pregnancy history survey compared to the prospective data (Table 2).

Among the date matched dataset of 124 outcomes, only half (n = 5/9, 55.6%) of stillbirths were correctly classified; the misclassified were reported as live births followed by neonatal death. The neonatal deaths were mostly correctly classified (n = 9/10, 90.0%); the one misclassified was reported as a stillbirth. Associations between misclassification and participant characteristics are given in Additional file 1: Table S4.

Age at death for the 10 neonatal deaths among the 62 matches ranged from 0.02 to 27.1 days (median 2.2 days) according to the prospective data. The difference in dates of death for the 9 correctly classified neonatal deaths ranged from − 2.0 days to 1.0 day (pregnancy history survey minus prospective data).

Among the 60 prospective non-date matched outcomes, the most likely errors, according to the criteria outlined in Additional file 1: Fig. S1, were date error or omission (n = 45, 75.0%), misclassification, or omission (n = 10, 16.7%), and date error, misclassification, or omission (n = 5, 8.3%). There were no prospective non-date matched outcomes that could be definitively classified as omissions.