Randomized trials versus observational studies in adolescent pregnancy prevention
Additional Document Info
The objective of this study is to compare the results of randomized trials and observational studies of interventions to prevent adolescent pregnancy. We identified published and unpublished reports through computerized searches of CATLINE, CINAHL, CONFERENCE PAPERS INDEX, DISSERTATION ABSTRACTS ONLINE, EMBASE, ERIC, MEDLINE, NTIS, POPLINE, PsycINFO, and SOCIOLOGICAL ABSTRACTS; manual searches of eight relevant journals; reference lists from primary articles; and contact with content experts. We included randomized trials and observational studies that evaluated the impact of primary prevention interventions including sex education classes, school-based clinics, free-standing clinics, physician/nurse practitioner practice-based service, improved access, and community-based programs on four outcomes: sexual intercourse, birth control use, responsible sexual behavior, or pregnancy in adolescents. One investigator abstracted the data and a second conducted a detailed review of the abstraction. We identified 13 randomized trials and 17 observational studies. We generated estimates of the impact of the interventions separately for males and females for all four outcomes for both observational studies and randomized trials. For six of the eight outcomes the summary odds ratios for the observational studies showed a significant intervention benefit (P<0.05) while the randomized trials did not show a benefit for any outcome in either females or males. The difference between the results of the observational studies and randomized trials was statistically significant in two of the eight outcomes (P<0.05 for initiation of intercourse and pregnancy in females). Observational studies yield systematically greater estimates of treatment effects than randomized trials of adolescent pregnancy prevention interventions. Public policy or individual patient treatment decisions should be based on observational studies only when randomized trials are unavailable and only with careful consideration of possible biases.