Have studies of urinary tract infection and preterm delivery used the most appropriate methods?

Authors

  • Marie S. O’Neill,

    Corresponding author
    1. Department of Epidemiology, School of Public Health, University of North Carolina at Chapel Hill,
    2. Department of Environmental Health, Harvard School of Public Health,
      Correspondence : Dr Marie O’Neill, Environmental Epidemiology Program, Harvard School of Public Health, 665 Huntington Ave., Bldg 1, 14th Floor, Boston, MA 02115, USA.
      E-mail: moneill@hsph.harvard.edu
    Search for more papers by this author
  • Irva Hertz-Picciotto,

    1. Department of Epidemiology, School of Public Health, University of North Carolina at Chapel Hill,
    2. Department of Epidemiology and Preventive Medicine, School of Medicine, University of California at Davis, and
    Search for more papers by this author
  • Lisa M. Pastore,

    1. Department of Epidemiology, School of Public Health, University of North Carolina at Chapel Hill,
    2. The Department of Obstetrics and Gynecology, School of Medicine, University of Virginia, Charlottesville, VA, USA
    Search for more papers by this author
  • Beth D. Weatherley

    1. Department of Epidemiology, School of Public Health, University of North Carolina at Chapel Hill,
    Search for more papers by this author

Correspondence : Dr Marie O’Neill, Environmental Epidemiology Program, Harvard School of Public Health, 665 Huntington Ave., Bldg 1, 14th Floor, Boston, MA 02115, USA.
E-mail: moneill@hsph.harvard.edu

Summary

Published investigations of the association between urinary tract infection (UTI) and preterm delivery used logistic regression or chi-squared tests. Because both exposure and outcome are time dependent, these methods were not optimal and did not account for person–time under observation, potentially an important feature given the variability of women's entry to prenatal care as well as of gestational lengths. Previous researchers probably classified as exposed some women whose UTI occurred after their pregnancies exceeded 37 weeks. We applied the previous analytical methods to 1990–93 births from two Durham, NC, USA, hospitals (n = 4053) and demonstrate survival methods as an alternative. Two logistic regression models were fitted with differing exposure definitions: model 1 in which exposed = UTI diagnosed after 20 weeks’ gestation; and model 2 in which exposed = UTI diagnosed between 20 weeks’ and 37 weeks’ gestation. Model 3 used proportional hazards regression with person–time after 20 weeks and before UTI diagnosis as unexposed, and person–time after diagnosis as exposed. Models were fit with and without five time-constant potential confounders. Model 1 yielded an adjusted odds ratio (OR) of 0.8 [95% confidence interval (CI) 0.5, 1.2], and model 2, which did not include UTI diagnoses after 37 weeks, an adjusted OR of 0.9 [95% CI 0.6, 1.4]. The Cox model hazard ratio (HR) for preterm delivery was 1.1 (adjusted) [95% CI 0.7, 1.7]. As these results indicated some bias, but not remarkable differences, we conducted a sensitivity analysis using 100 samples of 80% of the original data set, with replacement to determine how large the differences might be in other, similar data sets. The Cox method consistently produced higher effect estimates than either logistic model. The two samples with the greatest differences between the Cox and logistic model estimates yielded an OR of 1.47 [95% CI 0.95, 2.29] for model 1 vs. HR of 2.06 [95% CI 1.39, 3.06] for model 3, and an OR of 1.41 [95% CI 0.88, 2.25] for model 2 vs. HR of 1.79 [95% CI 1.17, 2.71] for model 3 respectively.

Previous published results on UTI and preterm delivery require cautious interpretation. Data on UTI timing should be gathered to allow appropriate analyses; survival methods account for person–time under observation and ensure that studied exposures precede effects.

Ancillary