An Exposure-Time Model’s Interval Coverage Was Not a Probability of Cannabis Impairment
An exposure-time model’s interval coverage describes how often its time estimates encompassed known exposure times in a validation dataset. It is not the probability that a person was impaired, even when the inputs are plasma THC and its metabolite.
The models estimated time since exposure
Huestis and colleagues developed one model using plasma THC and another using the plasma carboxy-THC-to-THC ratio. They examined the models against data from nine published and unpublished studies involving different exposure conditions and cannabis-use histories.
The reported intervals contained actual exposure times for 235 of 261 samples in one evaluation and 232 of 260 in the other, approximately 90.0% and 89.2%. These denominators count specimens, not hundreds of independent participants.
Coverage and individual certainty differ
A model can perform usefully across a dataset while remaining uncertain for a particular specimen. The investigators reported timing tendencies: predictions overestimated elapsed time immediately after smoking and underestimated it later.
Nor does an interval described as 95% automatically demonstrate 95% empirical coverage in every population. The observed coverage, the conditions of validation and the direction of error all belong in the interpretation.
The endpoint was not driving impairment
The target was exposure timing, not a simulator score, crash outcome or clinical diagnosis of impairment. A time estimate therefore cannot be relabeled as an impairment probability.
Application also requires the correct specimen and compatible conditions. Plasma inputs, a metabolite ratio and the exposure setting should remain explicit. A mathematically produced estimate can be more precise in appearance than the evidence supports if those requirements are ignored.
Related: Why the blood fraction matters for THC interpretation.
Research cited
Huestis MA; Henningfield JE; Cone EJ; Journal of Analytical Toxicology; 1992. Blood Cannabinoids. II. Models for the Prediction of Time of Marijuana Exposure from Plasma Concentrations of THC and THCCOOH.
