We propose three distinct phases for the quality control. Firstly, each individual variable within a given record is examined to identify departures from the study protocol; secondly, each record is checked and excluded if it is ineligible or logically incoherent for analysis; lastly, the distributions of key characteristics in the whole dataset are examined for their plausibility.
Data for patients diagnosed with bladder cancer in England between 1991 and 2010 are used as an example to aid the interpretation of the differences in data quality. The effect of different aspects of data quality on survival estimates is discussed.
We recommend that the results of data quality procedures should be reported together with the findings from survival analysis, to facilitate their interpretation.