Use of analysis methods in original articles on risk factors and health outcomes published in the European Respiratory Journal, 2010–2013

Analysis methods usedDetailsPublication year
Total number of studies retrievedEpidemiological original studies  looking at risk factor–disease  associations49 (100)40 (100)44 (100)22 (100)155 (100)
Simple methods
 Statistical significance testsChi-squared test, Fisher's exact test,  Student's t-test, ANOVA47 (96)37 (93)40 (91)20 (91)144 (93)
 Simple regression methodsLogistic, linear, Poisson and  Cox regressions33 (67)27 (68)34 (77)18 (82)112 (72)
 Regression methods for repeated   outcomesGEEs, linear mixed models7 (14)5 (13)8 (18)1 (5)21 (14)
 Regressions adjusted for   confounders22 (45)15 (38)20 (45)14 (64)71 (46)
Multivariate methodsPrincipal component analysis, cluster  analysis, factor analysis, co-inertia  analysis4 (8)2 (5)1 (2)0 (0)7 (5)
Causal multivariate methodsCross-lag modelling, path analysis,  Bayesian networks0 (0)0 (0)1 (2)0 (0)1 (1)
  • Data are presented as n (%) studies investigating risk factor–outcome associations. We performed an OVID Medline search, which targeted only the European Respiratory Journal, and only original research articles from observational epidemiological studies (i.e. we excluded meta-analyses, reviews, historical articles, consensus development conferences, guidelines, randomised controlled trials, clinical trials, comment, biographies and editorial). The term “epidemiolo*” was added as research term across all fields. We carefully examined the methods section of each article, removed duplicates, and retained only studies that looked at associations between risk factor(s) and outcome(s). GEE: generalised estimating equation.