F. Van Workum1, G. Hannink2, C. J. Van Der Velde4, H. J. Bonenkamp1, I. D. Nagtegaal3, M. M. Rovers2, C. Rosman1 1Radboud University Medical Center,Surgery,Nijmegen, GELDERLAND, Netherlands 2Radboud University Medical Center,Evidence Based Surgery,Nijmegen, GELDERLAND, Netherlands 3Radboud University Medical Center,Pathology,Nijmegen, GELDERLAND, Netherlands 4Leiden University Medical Center,Surgery,Leiden, NOORD HOLLAND, Netherlands
Introduction:
Learning curves are often observed after introduction of innovative surgical techniques, but there is currently no robust data suggesting that learning curves can influence outcome of high quality surgical randomized controlled trials (RCTs).
Methods:
Individual patient data was acquired from the Dutch D1-D2 trial, in which 1078 patients were randomised between D1 gastrectomy (old intervention) or D2 gastrectomy (innovative intervention). This RCT concluded that postoperative complications and mortality were significantly higher in the D2 group and that this did not support implementation of D2 resection into practice. Data from centres that included at least 10 consecutive cases (the minimum to perform meaningful trend analysis) were pooled for individual consecutive case numbers. Weighted moving average analysis was performed for the main outcome parameters and incidence graphs showing trends in outcome were plotted.
Results:
The incidence of postoperative death was 6% in the D1 group and no trend was observed during the trial, but in the D2 group, the incidence of postoperative death decreased from 10% to 3%. The incidence of postoperative complications increased from 19% to 20% in the D1 group (no significant trend). However, the incidence of postoperative complications decreased from 42% to 25% in the D2 group.
Conclusion:
This study showed significantly improving trends in the D2 group (innovative intervention) but not in the D1 group (old intervention), reflecting learning curve bias. Learning curve bias can significantly influence high quality RCT results and lead to misinterpretation of trial results. Incorporation of trend analysis in RCT reporting can assist clinicians with the interpretation of trial outcome data. Methodology to incorporate this into the design of RCTs is proposed.
