We reviewed observational studies that referred to any source of exposure to caffeine from food in pregnancy and to fetal mortality as the outcome (deaths occurring at twenty or more complete weeks of gestation).
Strategies to identify studies included an electronic search of the MEDLINE and LILACS bibliographic databases and other specialized libraries (Medcarib, Repidisca, Adolec), consultation with experts, and use of the "see related articles" feature in PubMed for articles meeting our eligibility criteria. The MEDLINE search strategy used as keywords "caffeine", "coffee", "tea", "cola", and "cacao" to define the relevant exposure and "fetal death", "stillbirth", "fetal demise", and "fetal loss" to define the outcome. The search strategy was limited to articles dealing with human subjects, published between January 1966 and September 2004, with no language restriction. To ensure completeness, the reference lists of all identified studies were reviewed. The search strategy in the LILACS database used the same keywords. Only publications in peer-reviewed journals were considered.
For each article that was included, an overall score of the methodological quality was given according to the checklist proposed by Downs & Black 12. This checklist considers aspects related to reporting, external and internal validity, bias, confounding, and power of the study to detect clinically important effects. The quality score was adapted for observational studies; items that did not apply were excluded, and we were able to retain 26 of the 31 original points.
For each individual report, letters to editors and responses by authors, and previous reviews which included commentaries on the articles were also reviewed. Methodological issues of the included studies were analyzed separately.