PHLT 8520 Week 2 Design Selection Paper Example

Ranking Four Observational Designs Against a Trial Nobody Would Run: Choosing a Design for Night Rosters in Early Pregnancy and the Risk of Preterm Birth

Student Name

Walden University

PHLT 8520: Advanced Epidemiological Methods

Instructor Name

Month Day, Year

Ranking Four Observational Designs Against a Trial Nobody Would Run: Choosing a Design for Night Rosters in Early Pregnancy and the Risk of Preterm Birth

The Trial Nobody Would Run

The course question asks if working nights during the first half of pregnancy makes an early delivery more likely for women employed by a health system. The cleanest answer would come from a trial: at confirmation of pregnancy, employees in rostered jobs would be randomly assigned either to schedules that include night shifts between gestational weeks 4 and 20 or to day-only schedules, and all would be followed to delivery. Randomization would balance job grade, income, prior obstetric history and health on average, and follow-up from assignment would keep early losses and job changes inside the comparison.

No ethics board would approve that trial, and no employer could deliver it. Writing it out still matters, because it turns a vague preference for "a good design" into a list of specific departures that each observational design makes. Hernán and Robins (2016) describe this practice as specifying the target trial, then asking how closely the available data can emulate each element: eligibility, treatment strategies, assignment, start of follow-up, outcome and analysis.

Elements of the Target Protocol

Table 1

Target Trial Protocol for Night Rosters in Early Pregnancy

Note. Protocol written for this composite example.

Design One: Prospective Recruited Cohort

A prospective cohort would recruit employees at their first prenatal visit, record schedules weekly and follow them to delivery. It matches the protocol on start of follow-up and lets the team collect confounders it chooses. Its departures are cost and size. Preterm birth affects roughly one pregnancy in ten, and an employer cohort might enroll a few hundred pregnancies a year, so a modest effect would need many years of recruitment. Participation also depends on choice, and women who enroll may differ from those who decline in ways tied to both shift work and obstetric risk.

Design Two: Record-Linked Retrospective Cohort

A record-linked cohort would use data the health system already holds: timekeeping entries for every shift, human resources files for job title and grade, and the system's own obstetric records for confirmation date, ultrasound dating and delivery. It approximates the protocol closely on exposure measurement, because timekeeping records shifts as they happen, before anyone knows how a pregnancy will end. Danish researchers used the same kind of payroll data to study night work and preterm birth among more than 16,000 public employees (Specht et al., 2019). Its departures are confounders the records do not hold, such as household income beyond salary, and pregnancies that happen outside the system's own clinics.

Design Three: Case-Control With Recalled Schedules

A case-control design would sample employees who delivered preterm and a set of controls who delivered at term, then ask each about her schedule in early pregnancy. It is efficient for a rare outcome, and with careful control selection it can estimate the same ratio as a cohort (Wacholder et al., 1992). Here, though, the outcome is not rare enough to make efficiency decisive, and the design swaps recorded schedules for recalled ones. Mothers of preterm infants may search their memories for causes, so the recall error could differ by outcome, a departure the protocol cannot tolerate.

Design Four: Unit-Level Comparison

A fourth option would compare preterm rates across hospital units with heavy and light night staffing. It is cheap and uses published unit data. It departs from the protocol at the level of assignment itself: the unit, not the woman, carries the exposure, so a night-heavy unit could have higher preterm rates because of who works there, not because of what any one employee did. Individual schedules within units would be ignored entirely.

Ranking the Departures

The record-linked cohort ranks first, not because it is free of bias, but because each of its departures has a name, a likely direction and a way to be examined. Missing confounders can be probed with sensitivity analysis. Out-of-system deliveries can be counted from insurance claims and described. The prospective cohort ranks second; it would be cleaner on confounders but could not reach a useful size within a doctoral timeline. The case-control design ranks third, since differential recall would sit inside the exposure itself, and the unit comparison is set aside because no adjustment can recover individual exposure from unit averages.

Table 2

Departures From the Target Trial, by Design

Note. Rankings are judgments for the composite setting.

What the Chosen Design Must Still Prove

Choosing the record-linked cohort sets the agenda for the rest of the course. Its comparison group must be defended, since night workers differ from day workers in ways the records only partly capture. Its entry rules must be examined, because a pregnancy has to be confirmed in the system's clinics to appear at all. Its exposure record must be validated against a second source, and its estimate must be stress-tested against a confounder it could not measure. The protocol table becomes the checklist for those weeks: every later decision will be described as a step toward, or away from, one of its rows.

Two questions stay open. The first is whether follow-up should begin at confirmation or at week 4, since some pregnancies are confirmed after night shifts have already occurred. The second is how to treat women who move between schedules during the window. The emulation framework handles both by stating the strategy precisely, so the Week 4 proposal will define exposure across the whole window from week 4 through week 20 and treat roster changes as data rather than noise.

Summary

Four observational designs were compared against a written target trial of night rosters in early pregnancy. A record-linked retrospective cohort, built from timekeeping, human resources and obstetric records, departs least from that protocol in ways that can be stated and bounded, and it is chosen as the design for the remaining weeks. The departures it keeps become the agenda for the course.

References

Hernán, M. A., & Robins, J. M. (2016). Using big data to emulate a target trial when a randomized trial is not available. American Journal of Epidemiology, 183(8), 758-764. https://doi.org/10.1093/aje/kwv254

Specht, I. O., Hammer, P. E. C., Flachs, E. M., Begtrup, L. M., Larsen, A. D., Hougaard, K. S., Hansen, J., Hansen, Å. M., Kolstad, H. A., Rugulies, R., Garde, A. H., & Bonde, J. P. (2019). Night work during pregnancy and preterm birth: A large register-based cohort study. PLOS ONE, 14(4), Article e0215748. https://doi.org/10.1371/journal.pone.0215748

Wacholder, S., McLaughlin, J. K., Silverman, D. T., & Mandel, J. S. (1992). Selection of controls in case-control studies: I. Principles. American Journal of Epidemiology, 135(9), 1019-1028. https://doi.org/10.1093/oxfordjournals.aje.a116396