Joy in repetition: the effect of advances in statistical software on the analysis of repeated measures studies in radiotherapy research

Introduction: As part of a move towards reflexivity, whereby practice can influence research and vice-versa, more longitudinal study designs are being utilised within radiotherapy research. Questions regarding change over time of an outcome and the assessment of predictive factors associated with the outcome can be answered by this approach. As subjects are measured repeatedly over time their responses are correlated, and failure to incorporate this fact into the data analysis may lead to erroneous conclusions and loss of statistical power (1). Conventional approaches to analyse correlated observations have been, to transform multiple observations into a single summary measure or to use a form of repeated-measures analysis of variance (rANOVA) (2). Both approaches have recognised strengths but also inherent weaknesses regarding inflexibility of the number, timing and variability of observations, an inability to estimate the way individual subjects change over time and difficulties when dealing with incomplete datasets (3). More sophisticated statistical methods, mixed-effects models and generalised estimating equations, overcome most of these issues but until recently have been too complex for most researchers to utilise. Novel software, such as SPSS version15, has now made them accessible without requiring statistical programming. Method: Archives of six journals - Radiotherapy & Oncology, International Journal of Radiation Oncology Biology Physics, Journal of Clinical Oncology, Radiography, Journal of Pain & Symptom Management and Journal of Radiotherapy in Practice - were searched for radiotherapy related repeated measure designs, published between 1999 and 2006. The level of statistical analysis was noted. Results: In 1999 12% of studies used sophisticated models, 75% used rANOVA and 13% summary measures. By 2006 the use of sophisticated models was 39%, with 59% rANOVA and only 2% summary measures. Discussion: Radiotherapy research has increasingly utilised mixe