Evaluating the performance of maximum likelihood estimation in a discriminant function framework to account for non-detectable exposure measurements in matched case-control studies