The federal government publishes a running count of hospital visits linked to stimulant use. It is built entirely from billing paperwork. A separate federal project that also reads what clinicians actually wrote in the chart finds a large number of cases the billing paperwork never captured, and newly updated documentation puts numbers on the gap for the first time.
In emergency department data from 2020, the National Center for Health Statistics identified 21,220 encounters with documented methamphetamine use. Of those, 16,943 were found only in clinical notes. Medical codes alone would have identified roughly one in five.
That gap is not a scandal. It is a known limitation of how American health data gets assembled, and NCHS has been working on it for years with funding from the Office of the Secretary Patient-Centered Outcomes Research Trust Fund. But it has consequences for anyone trying to understand how much stimulant-related illness moves through emergency departments, because the number most people can look up is the smaller one.
Two Ways to Find the Same Patient
The National Hospital Care Survey collects inpatient and emergency department records from participating hospitals. Researchers built what the documentation calls the Stimulant Algorithm to identify encounters with documented evidence of therapeutic stimulant use, non-therapeutic stimulant use, non-therapeutic co-use of stimulants and opioids, and whether a drug screen was performed.
The algorithm has two parts, and the distinction between them is the whole story. The medical code component searches for diagnosis, medication, laboratory, procedure, and service codes. The clinical note component uses machine learning and natural language processing to read free text, matching predefined search terms to the clinical context in which they appear. The Stimulant Use Dataset documentation, updated in May 2026, sets out both components and the search-term lists that are hosted in a public CDC code repository.
Why would a case appear in a note but not in a code? Coding follows billing requirements and available documentation, so a substance mention that does not drive reimbursement may never be coded. A clinician may record that a patient reported using methamphetamine while being treated for chest pain, and the coded diagnosis becomes chest pain. The documentation also notes that codes for prescription stimulant misuse are sparse and often do not distinguish misuse at all.
The pattern is not uniform across drugs. For cocaine in the 2020 emergency department data, medical codes did most of the work, identifying 47,042 encounters that notes missed, while notes alone added 11,829. For methamphetamine, the ratio inverted almost completely: codes alone found 607 encounters, notes alone found 16,943, and both components agreed on 3,670. Across all stimulant use in the emergency department that year, the algorithm identified 133,865 encounters, of which 37,117 came only from notes.
The Dashboard Running on Claims Alone
Here is where the practical difference shows up. The CDC's public drug use hospital data pages present monthly figures on hospital encounters involving stimulants, opioids, benzodiazepines, cannabis, and all drugs. Those figures come from UB-04 administrative claims data covering January 1, 2024, through December 31, 2025, from 81 hospitals submitting inpatient data and 70 submitting emergency department data. The categories are defined by ICD-10-CM diagnosis codes alone.
The agency is transparent about the limits. It states that the data are preliminary, that they are not nationally representative, that estimates are based on the first ten diagnosis codes available, and that months with fewer than 30 encounters in the denominator are suppressed.
Claims data alone is precisely the input the note-reading component was designed to supplement. The richer method has been applied to 2020 and 2021 survey data, and those restricted datasets are available now through the NCHS Research Data Center rather than as a public download.
None of this means the published figures are wrong. It means they are a floor rather than a full count, and readers encountering a statistic on stimulant hospitalizations should understand which method produced it.
The Note Reading Method Has Its Own Limits
Two caveats belong next to those numbers, and neither is small.
First, clinical notes are submitted by only a subset of participating hospitals, mostly those sending electronic health record data, and the notes must contain enough detail to be analyzable. The note component processed 9.0 percent of all 2020 encounters and 9.8 percent of 2021 encounters. The comparison above therefore understates what notes could add if they were available everywhere, which strengthens the undercount argument rather than weakening it, but it also means the note-derived counts cannot be read as a national total.
Second, the algorithm cannot pin down timing. Electronic records do not reliably separate current from past substance use, so the algorithm captures documented use at any point in a patient's clinical history as reflected in that year's data, not necessarily use during the visit itself. Anyone reading these counts as a measure of acute presentations would be reading them wrong.
Validation is also mixed rather than settled. A 2024 NCHS report tested the sibling opioid algorithms against manual abstraction of a stratified sample of 900 hospital encounters from the 2016 survey. That validation report found that the opioid involvement algorithm performed well, with an F-score of 0.95, while the opioid overdose algorithm scored 0.48. Judged by Matthews' correlation coefficient, a stricter measure, performance across all four algorithms ranged from 0.33 to 0.57. Combined code and text methods are a measurement improvement, not a solved problem.
Undercounting Shapes Where Money and Treatment Go
Surveillance numbers are not abstract. They inform where treatment funding is directed, how hospitals staff for substance-related presentations, how emergency departments plan for behavioral health boarding, and how states argue for federal support. An undercount in one drug category relative to another can quietly shape those decisions for years, because the following year's allocation is usually built from the previous year's measured burden.
Stimulants matter here specifically because the overdose picture has shifted. MedicalDaily has reported on the sustained decline in overdose deaths and on the growing role of stimulants alongside synthetic opioids. Unlike opioid use disorder, methamphetamine and cocaine use disorders have no FDA-approved medication treatment, so the response depends almost entirely on behavioral treatment capacity, which is funded on the basis of measured need. A category with no drug to prescribe is also a category with fewer built-in advocates in the data.
For families, the useful takeaway is narrower and more immediate. If a relative is hospitalized and substance use is part of the picture, what gets written in the note and what gets coded are not the same thing, and neither is guaranteed to follow the patient to the next provider. Asking the care team whether a substance use treatment referral is part of the discharge plan is more reliable than assuming the record carries it forward.
Anyone concerned about their own or a loved one's substance use can contact the SAMHSA National Helpline at 1-800-662-4357, which is free and operates around the clock.
Several things remain unknown. NCHS has not published equivalent code and note comparisons for the years covered by the current public dashboard, so the size of the gap in recent data cannot be determined. The hospitals contributing to that dashboard are not a national sample, so the monthly figures cannot be scaled to the country. NCHS lists new files and documentation on its survey documentation page as they are posted.
Key Questions Answered
What is the Stimulant Algorithm? A National Center for Health Statistics tool that identifies hospital encounters with documented stimulant use, using both medical codes and natural language processing of clinical notes. It was funded through the Office of the Secretary Patient-Centered Outcomes Research Trust Fund.
How large is the gap between the two methods? In 2020 emergency department data, 16,943 of 21,220 methamphetamine encounters were found only in clinical notes. For cocaine, the pattern reversed, with codes identifying far more than notes did.
Why do billing codes miss cases? Coding follows billing requirements and documentation. A substance mentioned in a clinical note may never become a coded diagnosis if it was not the reason for treatment.
Are the CDC's published hospital drug figures wrong? No. They are drawn from administrative claims data and should be read as a floor. The agency states that they are preliminary and not nationally representative.
How current are the public figures? They cover January 1, 2024, through December 31, 2025, from 81 hospitals submitting inpatient data and 70 submitting emergency department data.
What are the limits of the note-reading method? Clinical notes were analyzable for about 9 percent of encounters, and the algorithm cannot distinguish current from past use. A 2024 validation of the related opioid algorithms produced mixed performance results.
Where can someone get help with substance use? The SAMHSA National Helpline at 1-800-662-4357 is free and available 24/7.
