A database marker can identify records worth reviewing without being a definitive methamphetamine test. A 2025 study from an integrated safety-net health system examined that distinction by comparing combinations of diagnosis, laboratory, assessment and clinical-note markers with manual chart review.
What did the study count?
Among 510,885 patients seen in 2021, investigators identified 29,602 with at least one selected marker during 2018–2021. They sampled 814 charts for review, representing 713 unique patients, across eight marker-combination groups. The 814 reviews must not be presented as 814 independent people.
The markers included nonspecific stimulant diagnosis codes, clinical impressions, hospital urine screens, more specific outsourced testing, self-reported assessments and mentions in clinical notes. Extracting a structured code required less computational effort than searching and cleaning free text.
How much did predictive value vary?
Positive predictive values ranged from 0.21 for the hospital urine-screen-only group to 1.00 for a group combining clinical impression with a positive assessment or the more specific behavioral-health urine testing. Four groups exceeded 0.70. These values describe selected electronic definitions compared with the study’s chart-review reference.
The 0.21 result was not an estimate that 79% of the laboratory assays malfunctioned. A class-positive amphetamine screen can correspond to a prescribed amphetamine without establishing methamphetamine use. The record-classification question was narrower and drug-specific.
Does a predictive value of 1.00 mean perfect detection?
No. Positive predictive value asks how often a selected positive definition corresponds to the reference classification. It does not, by itself, establish how many people with methamphetamine use were missed, which would require a sensitivity assessment with an appropriate denominator.
The result also belongs to the sampled records and local documentation practices. A definition that performs well in one health system may behave differently where prescribing, prevalence, coding and assessment practices differ. The point estimate should not become a guarantee for every future record.
Was chart review a gold standard?
The authors expressly identified the absence of a gold standard for retrospective identification. Reviewers applied a standardized process, but the underlying records could not always establish intensity of use. When a clinician suspected use but the patient denied it, the investigators deferred to the patient report for classification.
That is a transparent operational choice rather than proof that either suspicion or denial was invariably correct. The authors used a probabilistic approach in recognition of the reference’s uncertainty. A forensic summary should preserve the reference process instead of describing every algorithm discrepancy as a known falsehood.
Why is free text particularly difficult?
The researchers searched notes for methamphetamine-related terms while removing negations, irrelevant matches and template material. A word can refer to current use, remote use, a family member, a denied history or a test order. Finding the word is not equivalent to establishing current personal exposure.
More elaborate extraction also does not automatically mean better performance. The study compared practical accuracy with the effort needed to obtain each marker, rather than declaring one technical approach universally superior.
What can the study support?
It supports careful selection and local evaluation of research case definitions. A marker-positive chart is neither a measured methamphetamine concentration nor a diagnosis of impairment. The useful unit is the explicitly defined record classification, with its time window, reference method and known uncertainties.
Source
Evaluating markers for methamphetamine use in electronic health records. Drug and Alcohol Dependence. 2025;276:112876. DOI: 10.1016/j.drugalcdep.2025.112876. Full text; bibliographic record.