Preterm Delivery Rates During Coronavirus Lockdown

A journal club I attend recently reviewed two articles published in Obstetrics & Gynecology (the “Green journal”) that reported rates of preterm delivery during “lockdowns” imposed last year due to the coronavirus epidemic.  One of the articles was from Australia and observed a decrease in preterm births.

The paper (https://journals.lww.com/greenjournal/Fulltext/2021/03000/Prematurity_Rates_During_the_Coronavirus_Disease.3.aspx) examined monthly preterm delivery rates occurring at the three Monash Health maternity hospitals from January 2018 through September 2020.  These hospitals serve a large proportion of the population in the surrounding area.  The authors did an interrupted time-series analysis using auto-regressive integrated moving average model (ARIMA).  A picture says a thousand words so see Figure 1 from article below.

“Interrupted time-series” means the regression model includes a term that indicates whether the data point was captured during the pre-intervention or post-intervention time period. ARIMA models use prior months’ preterm delivery rates to predict subsequent preterm delivery rates (seen in the line in Figure 1). The ARIMA model can take into account seasonal effects that might occur (a classic example is flu hospitalizations.) Another thing the ARIMA model can do is show what kinds of changes might happen after an intervention is implemented. For example, right after the intervention is put in place, preterm delivery rates might drop pretty steeply (a “step” change). Or, the general trend in preterm delivery rates (slope of the line) might change. If enough data points are obtained, one could evaluate whether an initial benefit is seen, followed by reversion back to the previous level and/or trend.

The lockdown period in Australia was July 8, 2020 through September 28, 2020.  Figure 1 suggests:

1. The general trend prior to the lockdown was a decrease in preterm delivery rates.
2. There may have been a step change (sudden decrease) in preterm delivery rates after the lockdown was initiated.
3. The trend in decreasing preterm delivery rates appears to have accelerated (the slope of the line after the lockdown is steeper and in the same direction as the pre-lockdown trend).

Given the above, should we “lock up” pregnant women so that they don’t give birth prematurely?  Before we adopt such a strategy, consider the following:

1. We only have three data points after the “intervention” (in this case the “intervention” wasn’t designed to prevent preterm birth).  If the lockdown had an effect on preterm delivery, we would expect preterm delivery rates to return to pre-lockdown levels.  Observing this would have bolstered the authors’ conclusions.
2. The monthly preterm delivery rates vary by a fair amount during the pre-lockdown period, from roughly 7.5% to 11.5% (a relative difference of about 50%).  The preterm delivery rates are within this range, although it’s true it’s the lower end of the range.
3. The pre-lockdown trend in preterm delivery rates will depend on the time period you include as your pre-intervention time period.  For example, if the pre-lockdown time period started in May 2019, the trend (slope of the line) in preterm deliveries likely would have looked more similar to that for the lockdown period.

The lockdown was short lived.  In such a case, from an epidemiological perspective you would want to ensure that you restrict your analysis to people at risk of the outcome.  For example, if a woman is 38 weeks pregnant at the start of the lockdown, she would never have delivered preterm during the lockdown.  Similarly, if she is 10 weeks pregnant, she may have a spontaneous abortion but she will not deliver preterm.  Since the data came from hospital-based records, the authors could have constructed two cohorts, one pre-lockdown and one during lockdown, who were at risk based on gestational age.

This requires thinking about how the lockdown would affect a pregnant woman.  In reality, I suspect its effect varied from one woman to the other, depending on whether she is an “essential” worker among other factors, so I have to suspend my disbelief for the sake of this example.  But suppose you decided that the gestational ages that would be most sensitive to the lockdown were beginning of the 6^th month to end of the 8^th month (22-35 weeks of pregnancy).  Suppose further that for the lockdown to have an effect, the pregnant woman would have to experience it for at least two weeks.  That means the highest gestational age to include would be 33 weeks (to allow for two weeks of exposure prior to week 35).

You would assemble your cohort by including women 22-33 weeks pregnant at the start of lockdown (July 8, 2020).  During the lockdown period you would add women as they became “eligible” due to arriving at 22 weeks and having the opportunity to experience at least two weeks of lockdown, until September 14 (two weeks prior to the end of lockdown).  Instead of calculating monthly rates, you would calculate the rate of preterm delivery in this cohort, including preterm deliveries that occurred after September 28.  You would follow a similar procedure for the pre-lockdown cohort. 

When designing a retrospective study, it’s still important to consider the exposure time of a cohort at risk.  This is part of the inclusion and exclusion criteria for the study.