Pregnancy Interval and Conception Timing: A Statistical Review

Introduction

In "Pregnancy Outcome and the Time Required for Next Conception," Jain (1969) examines a large group of presumably fertile women to determine the average age of conception following a previous pregnancy, as well as the length of time between pregnancies in the absence of birth control. According to the study, the average length of time between pregnancies remains high among women under the age of 20, then levels out before spiking again in individual women as maternal age advances. Additionally, the outcome of the pregnancy was considered: women whose pregnancies ended in live birth, or in fetal death via natural or artificial means, also experienced shortened pregnancy intervals, though the intervals were higher among women under age 20 and gradually increased with advancing maternal age.

Hypothesis

The paper's hypothesis centers on an empirical exploration intended to add to the knowledge base regarding the length of the pregnancy interval by age of the woman and to help provide an inferred length of amenorrhea among these same women — all in the absence of contraception. The underlying assumption is that, in the absence of contraception, the natural pregnancy interval can be reliably determined.

Statistical Methodology

The methodology employed is demonstrative of a simple mean and variance report, in which the average interval was determined and then the variance among all participants was calculated. The mean was determined by adding all interval lengths in months across all participants and dividing by the number of participants. The mean was then calculated based on both maternal age and the outcome of the pregnancy — that is, live birth, termination of pregnancy, and infant death at birth, prior to one year of age, and after one year of age. The variance was determined by finding the square of the standard deviation, which is itself the square root of the variance.

The sample was drawn from 2,443 surveyed married women who recorded their pregnancy histories prior to beginning a family planning program in Taichung, Taiwan, in an effort known as the Intensive Fertility Survey (1962). Data associated with women who reported any contraceptive use — approximately 7% of participants — were eliminated from analysis (p. 422). The researchers further reduced the statistical dataset by removing the first pregnancy interval (the time from marriage to the end of the first pregnancy) and the last open pregnancy interval (from the end of the last pregnancy to the survey date), so as not to skew the data unnecessarily.

The study also identifies two significant statistical biases that could affect the accuracy of the data, and attempts to account for each. The first is a truncation bias, in which pooling data from women of all ages regardless of age-group placement was believed to bias the results. To address this, the researchers offered two methods of statistical analysis: one in which the interval determination was identical for all participants, and one in which age was measured at the beginning of pregnancy intervals for women in the under-20 and 20–24 age groups, and at the end of pregnancy intervals for women in the 25–29, 30–34, and 35–39 age groups (pp. 424–425).

The second potential bias is a memory bias, associated with participants' difficulty in accurately recalling pregnancy intervals from the past — particularly when reporting past fetal deaths. The researchers assumed this bias would be greater among older participants than younger ones (p. 423). Together, these two sources of bias represent meaningful threats to data validity in a study relying on retrospective self-report.

Conclusion and Critical Evaluation

This work is clearly dated (1969), but it is also important to note that the age of the dataset may be essential for the data collected, as finding a large population of married women who practice no contraception is exceedingly rare given the pervasive nature of contraceptive use worldwide today. A more recent study associated with pregnancy interval — looking specifically at the impact of pregnancy interval (intentional or otherwise) on pregnancy outcomes, such as fetal prematurity, low birth weight, or fetal death — utilized a much larger sample of 14,930 women. That work did not use multiple-pregnancy data (thereby eliminating memory bias); instead, it used the documented pregnancy interval from the termination of the previous pregnancy to the beginning of the next, based on due-date calculations after termination of the current pregnancy (Cecatti, Correa-Silva, Milanez, Morais, & Souza, 2008).

The data collection approach of the Jain study could clearly have used a similar method — that is, working from the current to the most recent past pregnancy based on medical records or current-pregnancy report — to arrive at a dataset with less unknown bias and reduced memory bias. Although the intention of the Jain work was to increase data availability by including multiple pregnancies via survey self-report, the inclusion of memory bias and self-report bias through this method is questionable.

The intention of the Cecatti et al. (2008) study was additionally to determine pregnancy outcomes — specifically, maternal and infant health as a function of pregnancy interval — rather than simply to determine the interval itself. The median inter-pregnancy interval reported by Cecatti et al. was 27 months, which again varied by age and other demographic factors. The findings of that article are also based only on pregnancies ending in live birth during the current pregnancy, and the interval is measured from the termination of the previous pregnancy to the end of the current pregnancy, minus the gestation period. While the mean pregnancy interval was not calculated (as it was not of interest to those researchers), the largest statistical group — 34.6% of participants — had pregnancy intervals of fewer than 18 months (Cecatti et al., 2008, p. 277), a figure only slightly higher than that found in the Jain study.

Cecatti et al. also note that little has changed with regard to family planning behavior over the preceding 20 years, meaning that few women report intentionally spacing pregnancies for maternal and infant health reasons (2008, p. 275). This observation does not, however, explain or eliminate a potential spike in contraceptive use or deliberate family planning that may have occurred between 1962 — when data were gathered for the Jain study — and the period 20 years prior to the 2008 article's data collection.

Overall, the methodology of the Jain article is valuable, with the noted exception above. Restricting the study to intervals between the current and the immediately preceding pregnancy would likely eliminate both the truncation bias associated with age grouping and the memory and self-report biases that Jain himself identifies. This restriction would necessarily reduce the number of events available for analysis, but in this writer's view it would be worth it for the elimination of bias. The remaining methodological approach — comparing results with and without the perceived bias — offers some allowance for researcher awareness of bias, which was likely incorporated after data analysis had begun, as the researchers noticed the potential for distortion in the collected dataset.

All other research providing similar information was based on outcomes for child, mother, or both with respect to health, or examined the impact of birth interval in special populations. In nearly all cases, those studies examined statistics between the most recent and the current birth in the same manner as Cecatti et al. (2008) rather than following the approach of Jain. It is therefore safe to conclude that this method of identifying pregnancy intervals — using only the current and immediately preceding pregnancies — is the accepted standard for data collection in this field, and one not followed in the Jain article.

References

Arafa, M. A., Alkhouly, A., & Youssef, M. E. (2004). Influence of inter-pregnancy interval on preterm delivery. Paediatric & Perinatal Epidemiology, 18(4), 248–252. doi:10.1111/j.1365-3016.2004.00564.x

Cecatti, J. G., Correa-Silva, E. B., Milanez, H., Morais, S. S., & Souza, J. P. (2008). The associations between inter-pregnancy interval and maternal and neonatal outcomes in Brazil. Maternal & Child Health Journal, 12(2), 275–281. doi:10.1007/s10995-007-0219-y

Gold, R., Connell, F. A., Heagerty, P., Cummings, P., Bezruchka, S., Davis, R., & Cawthon, M. (2005). Predicting time to subsequent pregnancy. Maternal & Child Health Journal, 9(3), 219–228. doi:10.1007/s10995-005-0005-7

In this issue. (2004). Paediatric & Perinatal Epidemiology, 18(4), 245. doi:10.1111/j.1365-3016.2004.00573.x

Jain, A. K. (1969). Pregnancy outcome and the time required for next conception. Population Studies, 23(3), 421–433.

Jun, Z. (2007). Partner change, birth interval and risk of pre-eclampsia: A paradoxical triangle. Paediatric & Perinatal Epidemiology, 21, 31–35. doi:10.1111/j.1365-3016.2007.00835.x

Klebanoff, M. A. (1999, February 25). The interval between pregnancies and the outcome of subsequent births. New England Journal of Medicine, pp. 643–644. doi:10.1056/NEJM199902253400809

O'Donnell, A. A., Raiko, A. A., Clegg, J. B., Weatherall, D. J., & Allen, S. J. (2006). +-Thalassaemia and pregnancy in a malaria endemic region of Papua New Guinea. British Journal of Haematology, 135(2), 235–241. doi:10.1111/j.1365-2141.2006.06274.x

Selo-Ojeme, D. O., & Tewari, R. R. (2006). Late pre-term (32–36 weeks) birth in a North London hospital. Journal of Obstetrics & Gynaecology, 26(7), 624–626. doi:10.1080/01443610600903321

Van Eijk, A. M., De Cock, K. M., Ayisi, J. C., Rosen, D. H., Otieno, J. A., Nahlen, B. L., & Steketee, R. W. (2004). Pregnancy interval and delivery outcome among HIV-seropositive and HIV-seronegative women in Kisumu, Kenya. Tropical Medicine & International Health, 9(1), 15–24. doi:10.1046/j.1365-3156.2003.01165.x

Appendix

Tables 2 and 3: Comparison of the two methods of determining age, from Jain (1969), pp. 424, 426.