23Purpose: We compare published non-Cochrane reviews of nursing with their pre-registered 24 protocols on PROSPERO to quantify the prevalence of differences and the extent to which the 25 differences were explained.
26Methods: We searched for protocols and their corresponding reviews in PROSPERO's nursing 27 group that were "completed and published" from inception to September 8 th , 2019. Two 28 authors independently identified differences and classified the difference as none, partial, or 29 complete, and determined if the existed differences had been explained. Frequency (n), 30 percentage (%), median, and inter-quartile ranges were used to analyze the extent of differences 31 and explanations.
32
Results:We identified 22 pre-registered protocols and their reviews. All 22 pairs (100%) 33 exhibited differences. Eighteen pairs (82%) showed differences in at least six methodological 34 sections, while 21 pairs (95%) involved completed difference in at least one section. The 35 median number of differences per review was 8.00 (upper quartile = 6.00, lower quartile = 36 9.75). The differences involved all 13 compared methodology-related sections. Only 5 (3%) of 37 all differences were explained in the systematic reviews.
38
Conclusions:We observed widespread differences between non-Cochrane reviews of nursing 39 and their protocols recorded in PROSPERO, with relatively few explanations for the changes.
40Measures including establishing a new item in the reporting guideline of systematic reviews to 41 guide reporting and explaining the reasons for differences between protocols and systematic 42 reviews or even requiring authors to do so at the Journal's author guideline are recommended 43 to improve transparency. 44 45 3 46 126 protocol or acknowledgment and explanation of the alterations in the published review.130Differences were judged and classified using an internal guideline that was developed, 131 independently pilot-tested (in n=10 pairs), and revised by two authors (K.Y.H. and B.M.) (S1 132 Appendix). The categories used were: no difference (when the systematic review and protocol 133 matched or when a section was either not applicable or not reported), partial difference (when 134 a section was either not fully specified in the systematic review or the protocol, or when minor 135 differences were observed between parts of sections) , and complete difference (when aspects 136 of a section were modified in the systematic review or when a section was omitted from either 137 the protocol or the review). If judgments differed on a section, we chose the ''worst case'' 138 (e.g., when we observed a partial and a complete difference, we classified the compared section 139 as having a complete difference). If there were more than two partial differences concerning 140 the one section, we also classified it as having a complete difference. 141 Statistical methods 142 Statistical analysis was performed with Microsoft Excel 2016 software. Frequency (n), 143 percentage (%), median and Inter-quartile range were used ...