Analyzing Retrospective Data with Time-Varying Exposure: A Cautionary Tale of H₂ Blockers in ELBW Neonates

ABSTRACT

The objective of our study was to alert researchers to potential pitfalls involved in analysis of a potential association between risk factors and development of disease. We applied three different analytic methods-retrospective cohort study analyzed using logistic regression, case-control analysis, and survival analysis with time-dependent exposure-to retrospective data showing a correlation between histamine 2 (H₂) blocker exposure and the occurrence of necrotizing enterocolitis in extremely low birthweight neonates. Using case-control and survival analysis methods, there was an apparent increase in necrotizing enterocolitis following exposure to H₂ blocker treatment. The retrospective cohort method led to the opposite conclusion because it failed to account appropriately for exposure. We concluded that the choice of the analytic methodology used in a retrospective study is critical and should take into account the timing and characteristics of the risk factors and condition under investigation. Erroneous conclusions may be reached if inappropriate statistical methods are used.

Statistical analysis - epidemiology - necrotizing enterocolitis

REFERENCES

• 1 Stoll B J. Epidemiology of necrotizing enterocolitis.  Clin Perinatol. 1994;  21 205-218
• 2 Stoll B J, Kanto Jr W P, Glass R I, Nahmias A J, Brann Jr A W. Epidemiology of necrotizing enterocolitis: a case control study.  J Pediatr. 1980;  96 447-451
• 3 Patole S K, de Klerk N. Impact of standardised feeding regimens on incidence of neonatal necrotising enterocolitis: a systematic review and meta-analysis of observational studies.  Arch Dis Child. 2005;  90 F147-F151
• 4 Guillet R, Stoll B J, Cotten C M et al.. Association of H2-blocker therapy and higher incidence of necrotizing enterocolitis in very low birth weight infants.  Pediatrics. 2006;  117 e137-e142
• 5 Walsh M C, Kliegman R M. Necrotizing enterocolitis: treatment based on staging criteria.  Pediatr Clin North Am. 1986;  33 179-201
• 6 Hosmer D J, Lemeshow S. Applied Logistic Regression. 2nd ed. New York; John Wiley 2000
• 7 Breslow N, Day N. Statistical Methods in Cancer Research. Vol. 1. The Analysis of Case-Control Studies. Lyon, France; International Agency for Research on Cancer 1980
• 8 Cox D R. Regression models and life-tables.  J Roy Statist Soc Ser B (Methodological). 1972;  34 187-220
• 9 Spruance S L, Reid J E, Grace M, Samore M. Hazard ratio in clinical trials.  Antimicrob Agents Chemother. 2004;  48 2787-2792
• 10 Schwartzbaum J, Ahlbom A, Feychting M. Berkson's bias reviewed.  Eur J Epidemiol. 2003;  18 1109-1112