Skip navigation

Preventing Teenage Pregnancy


If you were set the task of designing a programme to prevent teenage pregnancies, where would you start? Most of us would now start with a search for a systematic review on the subject. But suppose that review included both randomised and observational studies: would it make a difference?

A scintillating examination of this question [1] comes up with a straight, if hard, answer. Observational studies conclude that interventions work. Randomised trials demonstrate that they do not.


Eleven computerised databases were searched, eight journals hand searched, and reference lists reviewed of all relevant papers. Included were randomised and observational studies on adolescents 18 years or less and evaluating primary prevention programmes with outcomes of start of sexual intercourse, birth control use or responsible sexual behaviour.


There were 25 key reports with 30 studies; 13 were randomised and 17 were observational. Of the 25 reports, 13 were published in journals, three were chapters in textbooks, six were doctoral theses, one a masters thesis and two were unpublished.

For females, observational studies had statistically significant results for intervention over control for all four outcomes (Table). Randomised studies were not significantly different for any outcome.

Table: Interventions for adolescent pregnancy prevention - randomised versus observational studies

  Odds ratio (intervention versus control; 95% CI)
  Starting intercourse Pregnancy Responsible sexual behaviour Birth control use
Randomised trials (N 7-9) 1.1 (0.90 to 1.32) 1.1 (0.91 to 1.27) 1.0 (0.75 to 1.36) 1.0 (0.64 to 1.54)
Observational studies (N 6-11) 0.64 (0.44 to 0.93) 0.74 (0.56 to 0.98) 1.3 (1.1 to 1.5) 1.4 (1.2 to 1.6)
Randomised trials (N 3-4) 0.81 (0.35 to 1.90) 0.97 (0.62 to 1.51) 0.94 (0.55 to 1.60) 0.91 (0.71 to 1.18)
Observational studies (N 2-6) 0.71 (0.52 to 0.98) 0.85 (0.68 to 1.06) 1.2 (1.04 to 1.4) 0.82 (0.35 to 1.91)
For starting intercourse and pregnancy, odd ratios less than 1 indicate a desirable effect of intervention. For responsible sexual behaviour and birth control use odds ratios greater than 1 indicate a desirable effect of intervention. Shaded results are statistically better than control. N indicates number of trials included in analysis.

For males, observational studies had statistically significant results for intervention over control for two of four outcomes (Table). Randomised studies were not significantly different for any outcome.


There are at least three important lessons in this paper.

For those interested in reducing teenage pregnancy the message is bleak. The highest quality studies show that interventions do not work. Perhaps a really clever person might dredge some comfort from individual studies, but it is hard to see where that comfort will come from.

Yet again we learn the hard lesson that randomisation is everything. One example as to why observational studies were positive is given in the paper, where investigators assigned adolescents declining participation in the intervention group to the control group. It is entirely likely that adolescents receiving the intervention were more receptive to the message, and those in the control group were not. Magically statistical significance results from people doing much as they would have done anyway. The lesson is that including observational studies in reviews may lead us to make policy based on incorrect assumptions. "One of these [2] received wide public attention in the United Kingdom and was considered key data for public health policy".

The final lesson is about the amount of unpublished material. Only 13 of the 25 reports were in journals. Many were as dissertations, even the RCTs, and some were not small. This is good evidence that, in some areas at least, unpublished material might be important. Bandolier found it more convincing than a ramble through the publication bias literature [3].


  1. GH Guyatt et al. Randomized trials versus observational studies in adolescent pregnancy prevention. Journal of Clinical Epidemiology 2000 53: 167-174.
  2. R Dickson et al. Preventing and reducing the adverse effects of unintended teenage pregnancies. Effective Health Care 1997 3: 1-12.
  3. A Thornton, P Lee. Publication bias in meta-analysis: its causes and consequences. Journal of Clinical Epidemiology 2000 53: 207-216.
previous or next story in this issue