Some challenges in evaluating prostate screening: subjective adjudication of outcomes, and over-diagnosis
This seminar is presented in conjunction with the School of Mathematics and Statistics, University of Melbourne
The European Randomised Study of Prostate Cancer Screening has shown a substantial reduction in prostate cancer mortality, but a similar trial in the United States showed essentially no benefit. Because there is no “gold standard” for causes of death, we analysed the European data to find out if uncertainty in identifying prostate cancer deaths might have affected the results. Several latent class models were formulated to estimate the accuracy of outcome adjudicators and to determine if their accuracy varied, and ultimately to assess if adjudication errors rates might be differential between the study arms – and hence represent a potential bias.
The analysis revealed some discrepancies between adjudicators, but there was only limited evidence of asymmetry between them. There was no consistent evidence of differential accuracy by trial arm. We made corrections to the study results to allow for adjudication errors, but we concluded that bias arising from this issue was unlikely. Hence, other factors are probably responsible for the difference between the European and US findings.
In general terms, this analysis represents an interesting example of how one can assess potential bias in trial results when determination of the study end point is uncertain.
If time allows, we may also briefly discuss ongoing work on how to estimate the extent of overdiagnosis occurring in prostate screening.