OBJECTIVES: Studies with small sample sizes are generally unreliable and should be interpreted considering the disease context and available evidence. It is accepted practice to remove such studies during synthesis in reviews with substantial evidence. We analyzed the use of a sample size filter to allow removing these studies during the search strategy phase.

METHODS: The sample size is generally reported in the published abstract. A filter consisting of truncated numbers in proximity with keywords like patients, controls, adults, pediatrics, males, females, men, etc., and phrases like 'n:,' 'n=*' was developed. These filters were combined with disease terms of five oncology indications to establish reproducibility. The search was focused on epidemiology, as the sample size is an essential criterion for such reviews. Lastly, the filter was validated against published literature reviews on the same five indications that used sample size as a restriction but did not limit their search to study size.

RESULTS: The search resulted in 1489, 1612, 2108, 2431, and 1903 numbers needed to screen (NNS) for mesothelioma, melanoma, follicular lymphoma, gastric, and endometrial cancer, respectively. After applying the filter, the NNS was reduced by 33-48%. Despite the reduction of NNS, the search retrieved some non-relevant studies, such as studies mentioning years, percentages, specific numbered biomarkers, or the performance status of patients. We observed that it was challenging to retrieve older studies as previous reporting standards didn't recommend mentioning the sample size in abstracts. However, this gap can be covered through manual searching or restricting this filter's use in reviews to retrieve the latest evidence. The filter's sensitivity was 78-91% (mesothelioma: 87%, melanoma: 91%, follicular lymphoma: 84%, gastric cancer: 78%, and endometrial cancer: 82%).

CONCLUSIONS: The search filter would need further refinement and testing. However, it significantly reduces the NNS and can be considered in targeted reviews with sample size restriction.