Background: Studies with favorable results have higher chances of being published than those with unfavorable results. This could introduce bias in systematic reviews. Objectives: We aimed to compare summary estimates calculated from published and unpublished data in a systematic review of diagnostic test accuracy. Methods: We gathered published data by searching MEDLINE and Embase for studies evaluating the diagnostic accuracy of Fraction of exhaled Nitric Oxide (FeNO), blood eosinophils or Immunoglobulin E (IgE) for detection of airway eosinophilia in adults with asthma (inception to August 2014). We obtained unpublished data by contacting authors of studies that did not report on the diagnostic accuracy of these markers, but seemed to have collected data from which estimates could be calculated, and performed random-effects meta-analysis of area under the receiver operating curve (AUC). Results: Analysing 2919 search results, we included 12 published studies reporting on the accuracy of FeNO, blood eosinophils or IgE (published data). We contacted the authors of two of these studies, and obtained additional diagnostic accuracy data of markers that had also been assessed, but for which no accuracy results were reported (unpublished data). Authors of another 14 articles were contacted because they could have data from which accuracy estimates could be calculated. Seven shared the requested data, and we also included one conference poster (unpublished data). Adding unpublished data increased sample size for FeNO and blood eosinophils by 28%, and for IgE by 103%. Summary AUC for FeNO was 0.75 (95%CI 0.72 to 0.78; 9 studies; 2263 patients) based on published data, vs 0.73 (0.68 to 0.78; 7 studies; 627 patients) for unpublished data. For blood eosinophils, these numbers were 0.77 (0.71 to 0.83; 7 studies; 1592 patients) vs 0.77 (0.71 to 0.83; 5 studies; 446 patients), and for IgE 0.64 (0.57 to 0.70; 2 studies; 300 patients) vs 0.61 (0.54 to 0.67; 4 studies; 310 patients). Conclusions: Many authors seem to be willing to share unpublished data. Adding unpublished data led to a considerable increase in precision, but barely affected summary estimates of accuracy.