Estimation of Haplotypes at DRD2 May Have Produced Misleading Results

We are concerned that the results appearing to implicate an effect of DRD2 polymorphisms on the susceptibility to heroin dependence¹ may represent statistical artifacts rather than a true genetic effect. To a very large extent, the results are based on estimating the frequencies of haplotypes in cases and controls and testing for a difference, the frequencies being estimated using the MLOCUS program.² Likelihood ratio tests assuming a χ² distribution were reported to produce P values as low as 10⁻¹⁰. A Fisher exact test for 1 haplotype produced a P value of 10⁻²².

In case-control studies, haplotypes are not observed directly and their frequencies are only estimated. The authors of the MLOCUS program wrote that the use of the χ² distribution could lead to an inflated type I error rate and hence recommended using resampling methods to obtain statistical significance.² Apparently minor errors in the estimation of haplotype frequency can yield very misleading results. For example, for a sample of a few hundred subjects, if a haplotype frequency is estimated at 0.01 rather than a true value of 0.003, then a likelihood ratio test value on the order of 100 can be generated. Treating such a statistic as χ² will produce an infinitesimal, but incorrect, P value. Nor is it appropriate to apply the Fisher exact test to inferred frequencies rather than directly observed counts.

It seems to us that the results presented are inherently implausible. Taken at face value, they show that 3 haplotypes are present with a total frequency of almost 20% in heroin addicts but are not found at all in controls. This would imply that around a third of addicts would carry at least 1 copy of such a high-risk haplotype but that no controls would. It is hard to believe that a third of drug addicts carry some genetic risk factor that is not found in hundreds of controls—by implication a risk factor that would cause all who inherited it to go and become dependent on heroin. From the published information, it is difficult to see exactly what has gone wrong, but the inference of haplotypes is inherently a problematic endeavor and we strongly suspect that on this occasion the process may have produced misleading results.

Correspondence: Dr Curtis, Department of Adult Psychiatry, Royal London Hospital, Whitechapel, London E1 1BB, United Kingdom (dave@davecurtis.net).