Premium
Methods and risk of bias in molecular marker prognosis studies in oral squamous cell carcinoma
Author(s) -
SHahinas J,
Hysi D
Publication year - 2018
Publication title -
oral diseases
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.953
H-Index - 87
eISSN - 1601-0825
pISSN - 1354-523X
DOI - 10.1111/odi.12753
Subject(s) - medicine , confounding , oncology , proportional hazards model , multivariate statistics , data extraction , multivariate analysis , disease , medline , machine learning , computer science , political science , law
Background We investigated methods and risk of bias, focusing on research design, aim, prognostic factors, outcome, and statistical analysis in molecular marker prognosis studies of oral squamous cell carcinoma. Material and Methods We used a database search strategy to indentify relevant articles published in English in 2016. We developed a data extraction form to assess and extract information on methods of molecular marker prognosis studies in oral squamous cell carcinoma, based on methodological recommendations for prognosis studies. We used the Quality in Prognosis Studies tool to assess the risk of bias in six domains. Results Thirty‐six papers were retrieved for full text review: 35 were replication prognosis factor studies and one was a model development based only on molecular markers to stratify patient's risk. Retrospective cohort was the design used in most studies (91%). Despite recommendations against dichotomizing continuous prognostic variables, this was observed in the majority of cases. A substantial number of studies (60%) conducted survival analysis, COX regression, and Kaplan–Meier. Prognostic variables included in the multivariate model were often preselected based on the results of univariable analysis. Risk of bias was assessed high for confounding, statistical analysis and reporting domains in 46% and 49% of studies, respectively. Conclusions The prognosis studies analyzed here can be considered phase II explanatory studies. The next step is to construct and validate models, which can be applied for use in the clinical practice, to guide patient management or build explanatory models that can help better understand the causative role in the disease process of these markers.