An application of decision theory to patient screening for an autologous tumour vaccine trial

Patients are eligible for accrual onto a phase I autologous tumour vaccine clinical trial if their resected and dissociated tumour achieves a minimum viable cell count. Because tumour pre‐processing and cell count determination are expensive, there has been developed a screening procedure based on tumour mass to screen out those tumours unlikely to yield sufficient viable cells. If θ is the ratio of the expected benefit of an accrual onto the study to the cost of tumour pre‐processing and cell counting, then we maximize long‐run benefit by pre‐processing and counting only those tumours whose masses exceed a cutoff m c , such that Pr {sufficient tumour cells mass = m c } = 1/θ. We derive algorithms for estimating m c and evaluate them under a variety of assumptions concerning the cell count/mass relationship. These include explicit equations for m c under parametric assumptions as well as more general algorithms based on non‐parametric smoothing techniques. We show that when θ deviates substantially from 2, these methods outperform simple inverse interpolation.