Accepted author manuscript, 13 KB, Word document
Final published version
Research output: Contribution to Journal/Magazine › Meeting abstract › peer-review
Research output: Contribution to Journal/Magazine › Meeting abstract › peer-review
}
TY - JOUR
T1 - Internal pilot sample size re-estimation in paired comparative diagnostic accuracy trials with a binary response
AU - McCray, Gareth
AU - Titman, Andrew
AU - Ghaneh, Paula
AU - Lancaster, Gillian
PY - 2017/5/8
Y1 - 2017/5/8
N2 - The sample size required to power a trial to a nominal level in a pairedcomparative diagnostic accuracy trial, i.e. Trials in which the diagnosticaccuracy of two testing procedures are compared relative to a goldstandard, depends on the correlation between the two diagnostic testsbeing compared. The lower the correlation between the tests thehigher the sample size required, the higher the correlation betweenthe tests the lower the sample size required. A priori, we usually do notknow the correlation between the two tests and thus cannot determinethe exact sample size. Furthermore, the correlation between two testsis a quantity for which 1) it is difficult to make an accurate intuitive estimateand, 2) it is unlikely estimates exist in the literature, particularly ifone of the tests is new, as is very likely to be the case.One option, suggested in the literature, is to use the implied samplesize for the maximal negative correlation between the two tests,thus, giving the largest possible sample size. However, this overlyconservative technique is highly likely to be wasteful of resourcesand unnecessarily burdensome on trial participants - as the trial islikely to be overpowered and recruit many more participants thanneeded. A more accurate estimate of the sample size can be determinedat a planned interim analysis point where the sample size isre-estimated - thereby incorporating an internal pilot study into thetrial design, with the intention of producing an accurate estimate ofthe correlation between the tests into the trial.MethodsThis paper discusses a sample size estimation and re-estimationmethod based on the maximum likelihood estimates, under an impliedmultinomial model, of the observed values of correlation betweenthe two tests and, if required, prevalence, at a plannedinterim. The method is illustrated by comparing the accuracy of twoprocedures for the detection of pancreatic cancer, one procedureusing the standard battery of tests, and the other using the standardbattery with the addition of a PET/CT scan all relative to the goldstandard of a cell biopsy. Simulation of the proposed method arealso conducted to determine robustness in various conditions.ResultsThe results show that the type I error rate of the overall experimentis stable using our suggested method and that the type II error rateis close to or above nominal. Furthermore, the instances in which thetype II error rate is above nominal are in the situations where thelowest sample size is required, meaning a lower impact on the actualnumber of participants recruited.ConclusionWe recommend a paired comparative diagnostic accuracy trial whichused an internal pilot study to re-estimate the sample size at the interim.This design would use a maximum likelihood estimate, under amultinomial model, of the correlation between the two tests beingcompared for diagnostic accuracy, in order to more effectively estimatethe number of participants required to power the trial to at least thenominal level.
AB - The sample size required to power a trial to a nominal level in a pairedcomparative diagnostic accuracy trial, i.e. Trials in which the diagnosticaccuracy of two testing procedures are compared relative to a goldstandard, depends on the correlation between the two diagnostic testsbeing compared. The lower the correlation between the tests thehigher the sample size required, the higher the correlation betweenthe tests the lower the sample size required. A priori, we usually do notknow the correlation between the two tests and thus cannot determinethe exact sample size. Furthermore, the correlation between two testsis a quantity for which 1) it is difficult to make an accurate intuitive estimateand, 2) it is unlikely estimates exist in the literature, particularly ifone of the tests is new, as is very likely to be the case.One option, suggested in the literature, is to use the implied samplesize for the maximal negative correlation between the two tests,thus, giving the largest possible sample size. However, this overlyconservative technique is highly likely to be wasteful of resourcesand unnecessarily burdensome on trial participants - as the trial islikely to be overpowered and recruit many more participants thanneeded. A more accurate estimate of the sample size can be determinedat a planned interim analysis point where the sample size isre-estimated - thereby incorporating an internal pilot study into thetrial design, with the intention of producing an accurate estimate ofthe correlation between the tests into the trial.MethodsThis paper discusses a sample size estimation and re-estimationmethod based on the maximum likelihood estimates, under an impliedmultinomial model, of the observed values of correlation betweenthe two tests and, if required, prevalence, at a plannedinterim. The method is illustrated by comparing the accuracy of twoprocedures for the detection of pancreatic cancer, one procedureusing the standard battery of tests, and the other using the standardbattery with the addition of a PET/CT scan all relative to the goldstandard of a cell biopsy. Simulation of the proposed method arealso conducted to determine robustness in various conditions.ResultsThe results show that the type I error rate of the overall experimentis stable using our suggested method and that the type II error rateis close to or above nominal. Furthermore, the instances in which thetype II error rate is above nominal are in the situations where thelowest sample size is required, meaning a lower impact on the actualnumber of participants recruited.ConclusionWe recommend a paired comparative diagnostic accuracy trial whichused an internal pilot study to re-estimate the sample size at the interim.This design would use a maximum likelihood estimate, under amultinomial model, of the correlation between the two tests beingcompared for diagnostic accuracy, in order to more effectively estimatethe number of participants required to power the trial to at least thenominal level.
U2 - 10.1186/s13063-017-1902-y
DO - 10.1186/s13063-017-1902-y
M3 - Meeting abstract
VL - 18
JO - Trials
JF - Trials
SN - 1745-6215
IS - Suppl. 1
M1 - 200
ER -