Efficient Source Data Verification Using Statistical Acceptance Sampling: A Simulation Study

One approach to increase the efficiency of clinical trial monitoring is to replace 100% source data verification (SDV) by verification of samples of source data. An intuitive strategy for determining appropriate sampling plans (ie, sample sizes and the maximum tolerable number of transcription errors in the samples) is to use acceptance sampling methodology. Expanding upon earlier work in which the use of acceptance sampling strategies for sampling-based SDV was proposed, we describe an alternative acceptance sampling strategy that, instead of relying on sampling standards, evaluates all possible sampling plans algorithmically, thereby ensuring that selected sampling plans conform to prespecified criteria.

Empirical trial data guided the design of the proposed strategy. In addition, extensive simulations, also based on the empirical data, were performed to assess the performance in terms of workload reductions and the post-SDV error proportion of applying the proposed strategy.

13 different scenarios were simulated, but results of the default scenario show that the average pre-SDV error proportion per trial of .056 was reduced to .023 by inspecting only 40.5% of the case report form entries. Of the inspected data entries, almost half (18.0/40.5) was, on average, SDV-ed as part of the sampling process; remaining entries were inspected during full inspections after too many errors were observed in the samples.

Our results suggest that major reductions in workload can be achieved, while maintaining acceptable data quality levels. However, the results also indicate that the proposed strategy is conservative and further improvement is possible.