Assessment of Statistical Methodologies and Pitfalls of Dissertations Carried Out at National Cancer Institute, Cairo University

Abstract

<br /> <strong><span style="font-size: small;">Purpose: </span></strong><span style="font-family: Times New Roman,Times New Roman; font-size: small;"><span style="font-family: Times New Roman,Times New Roman; font-size: small;">To identify statistical errors and pitfalls in dissertations performed as part of the requirements for the Medical Doctorate (MD) degree at the National Cancer Institute (NCI), Cairo University (CU) to improve the quality of medical research. </span></span><strong><span style="font-size: small;">Methods: </span></strong><span style="font-family: Times New Roman,Times New Roman; font-size: small;"><span style="font-family: Times New Roman,Times New Roman; font-size: small;">A critical assessment of 62 MD dissertations conducted in 3 departments at NCI, CU, between 2009 and 2013 was carried out regarding statistical methodology and presentation of the results. To detect differences in study characteristics over time, grouping was into two periods; 2009-2010 and 2011-2013. </span></span><strong><span style="font-size: small;">Results: </span></strong><span style="font-family: Times New Roman,Times New Roman; font-size: small;"><span style="font-family: Times New Roman,Times New Roman; font-size: small;">Statistical methods </span></span><span style="font-family: Times New Roman,Times New Roman; font-size: small;"><span style="font-family: Times New Roman,Times New Roman; font-size: small;">were appropriate in only 13 studies (24.5%). The most common statistical tests applied were chi-square, log-rank, and Mann-Whitney tests. Four studies estimated sample size and/or power. Only 37.1% and 38.7% of dissertation results supported aims and answered the research questions, respectively. Most of results were misinterpreted (82.3%) with misuse of statistical terminology (77.4%). Tabular and graphical data display was independently informative in only 36 dissertations (58.1%) with accurate titles and labels in only 17 (27.4%). Statistical tests fulfilled the assumptions only in 29 studies; with evident misuse in 33. Ten dissertations reported non-significance regarding their primary outcome measure; the median power of the test was 35.5% (range: 6-60%). There was no significant change in the characteristics between the time periods. </span></span><strong><span style="font-size: small;">Conclusion: </span></strong><span style="font-family: Times New Roman,Times New Roman; font-size: small;"><span style="font-family: Times New Roman,Times New Roman; font-size: small;">MD dissertations at NCI have many epidemiological and statistical defects that may compromise the external validity of the results. It is recommended to involve a biostatistician from the very start to improve study design, sample size calculation, end points estimation and measures. </span></span>