[34] We identified two different groups of clinical trials based

on their recruitment method and found that this classification was useful in describing other important aspects of trial design and outcome. Eight of the clinical trials recruited trekkers as they ascended and then aimed to assess the same trekkers later on their expedition.[27, 29, 30, 33, 34, 36, 37, 43] We designated check details this type of trial “location-based.” The other nine trials, including the trial which was excluded from quantitative analysis, recruited people to the trial prior to embarking on an organized expedition(s) and we designated this type of trial “expedition-based.”[28, 31, 32, 35, 38-42] There are a number of key differences between the two different types of trial summarized in Table 2. Most importantly, location-based trials tended to be larger (median 160.5 vs 35) but have a higher dropout rate (median 52 vs 0.5). Expedition-based trials had a higher rate of ascent (mean 450 vs 2,800 m/d). All of the

studies used questionnaires to assess outcome. These were either administered by blinded researchers or self-administered. A number of assessment tools were used as shown in Table 1. The most commonly used assessment score was the Lake Louise Symptom Daporinad concentration score (LLS),[44] which was used in 10 studies (63%). Four studies (25%) used variations of the Acute Mountain Sickness score cerebral and respiratory domains (AMS-C and AMS-R) which are derived from the modified Environmental Systems Questionnaire.[45] Of the remaining clinical trials, one used the General High Altitude Questionnaire (GHAQ)[46] and one used a score developed for the clinical trial.[43] All of the scores were similar in that they were combined interval scores incorporating several symptom

domains and the diagnosis of AMS was made if a specific score was reached (often with the presence of headache mandatory). It is likely from the individual trial Bacterial neuraminidase reports that timing of assessment after arrival at altitude varied; however, they generally did not contain enough information on this factor to allow analysis. None of the study protocols were available for review. It was generally not possible to ascertain whether sequence generation, allocation, or blinding were satisfactory from the trial report since they were usually described briefly. However, no cause for concern about bias in any of these domains was found. All trials were therefore found to have low or unclear risk of bias in these domains. The main source of bias was found in the outcome data domain. As discussed above, studies which relied on location-based recruitment had a high dropout rate. We decided to perform a worst-case analysis of the missing data and exclude studies in which the worst-case analysis resulted in a change of result.