Drug repurposing

drug_repurposing.py assembles the data sets a drug repurposing study in an ALS registry starts from: who is in the motor neuron disease cohort and when they entered it, which medicines each participant was exposed to and what the evidence is, and how long each participant was followed to a defined data cut-off. The exposure criteria adapt the methodology of Reimer et al., Lancet Digital Health, 2026, a study of drug repurposing candidates for ALS in US Veterans Health Administration records. The module produces descriptive figures for generating hypotheses; it does not estimate the effect of any medicine.

Open the demo report Demo summary (CSV) Module source (drug_repurposing.py)

Live demo · Cohort · Exposure · Follow-up · Summary · Panels · Medicine names · Outputs · Running it · Limits

Live demo

The demo report below comes from the invented ALS cohort that ships with Registry Forge, run with a data cut-off of 31 October 2026. No real person is in it, and its figures say nothing about any medicine.

Back to top

Cohort

Every participant with a motor neuron disease code in the problem list is in the cohort. Cohort entry is the date of the earliest such code.

Vocabulary	Codes
ICD-10-CM	G12.20 (motor neuron disease, unspecified), G12.21 (ALS), G12.22 (progressive bulbar palsy), G12.23 (primary lateral sclerosis), G12.29 (other motor neuron disease)
ICD-9-CM	335.20 to 335.29
SNOMED CT	86044005 (ALS), 230258005 (motor neuron disease)
Mondo	MONDO:0004976 (ALS), MONDO:0019056 (motor neuron disease)

The codes are in COHORT_CODES at the top of the module; replace them to study a different disease. Participants with no medication record stay in the cohort and count as unexposed for every medicine, and the log reports how many cohort participants had any medication record. Reimer and colleagues left out patients with no medication record after cohort entry, as a sign that their care was not recorded; if your protocol does the same, apply that rule to the cohort file before analysis.

Back to top

Exposure

The default criterion, any, counts a medicine as an exposure when any record of it falls between 6 months before and 12 months after cohort entry. This suits participant-contributed records, where each medication row is an order, a list entry or a statement rather than a pharmacy fill.

The stricter Reimer criteria suit dispensing data with fill dates and days' supply:

Criterion	Rule
reimer-a	A dispensing within 12 months after cohort entry, with the prescription lasting at least 6 months
reimer-b	At least 2 dispensings between 6 months before and 12 months after cohort entry
reimer	Either A or B

Each exposure lists its evidence types: order (MedicationRequest), dispense, administration, reported_use (MedicationStatement) or document_list (a C-CDA medication list entry, which does not say whether the medicine was taken). An order is not evidence that a medicine was taken.

Back to top

Follow-up

Follow-up runs from cohort entry to the data cut-off given with --cutoff:

Situation	Follow-up ends	Counted as
Death on or before the cut-off (from FHIR or vital_status.csv)	at death	event
Confirmed alive on or after the cut-off	at the cut-off	censored
Last confirmed alive before the cut-off	at the last confirmed alive date	censored
No vital status	at the last dated record	censored

Nobody is assumed to be alive at the cut-off, and follow-up never runs to the date the script runs, so the same data and cut-off always give the same figures. Deaths recorded outside the EHR, for example from National Death Index linkage, go into vital_status.csv.

Back to top

Summary figures

For each medicine with at least --min-exposed exposed participants, the summary gives exposed and unexposed counts, deaths in each group, and Kaplan-Meier median survival in months measured from a landmark 12 months after cohort entry, among participants alive and under follow-up at the landmark. The landmark is there because exposure is assessed through month 12; measuring from entry would credit exposed participants with time they had to survive to be counted as exposed. "Not reached" means fewer than half of the group died during follow-up.

Suppression is applied to the whole row:

• If either group has fewer than k participants, every figure for that medicine is held back for cell size.
• Death counts and landmark counts are held back in pairs, because exposed plus unexposed always equals a cohort total.
• Landmark counts are held back when the number lost before the landmark is small.

The figures are unadjusted. Disease stage, site of onset, age, the reason a medicine was prescribed and many other factors differ between groups, so a difference between two medians says nothing about a medicine's effect.

Reimer and colleagues kept medicines with at least 30 exposed participants in a cohort of about 11,000. A registry of a few dozen to a few hundred participants needs a lower value; the default here is 5, and the value is never set below k. The log lists the medicines with the most exposed participants and marks which were kept.

Back to top

Medicine panels

The report always shows a panel of ALS clinical-interest medicines, whether or not they meet the minimum count. When fewer than k participants have a record of one, its participant count, year range, record count and evidence types are all held back for cell size.

Medicine	Brand names	ATC code	Status shown
Riluzole	Rilutek, Tiglutik, Exservan	N07XX02	FDA-approved for ALS (1995)
Edaravone	Radicava, Radicava ORS	N07XX14	FDA-approved for ALS (2017)
Tofersen	Qalsody	N07XX22	FDA-approved for SOD1 ALS (2023, accelerated approval)
Dextromethorphan-quinidine	Nuedexta	N07XX59	FDA-approved for pseudobulbar affect; of interest in ALS
Sodium phenylbutyrate-taurursodiol	Relyvrio, Albrioza	N07XX19	FDA-approved for ALS in 2022; withdrawn in 2024 after the PHOENIX trial

Medicines approved for ALS, and the withdrawn sodium phenylbutyrate-taurursodiol, are left out of the CURE ID review worksheet, because their use in ALS is not off-label. ATC codes were checked against the WHO ATC index in November 2026.

Back to top

Medicine names and ATC classes

Medication names are lowercased, stripped of dose and form, and mapped from brand to generic through BRAND_TO_GENERIC before the Anatomical Therapeutic Chemical (ATC) class is looked up in ATC_SEED. Combination products are recognized through COMBINATION_RULES: Nuedexta, or dextromethorphan and quinidine together, become dextromethorphan-quinidine, and Relyvrio, Albrioza, or phenylbutyrate with taurursodiol become sodium phenylbutyrate-taurursodiol. Dextromethorphan alone (an antitussive, R05DA09) and quinidine alone (an antiarrhythmic, C01BA01) are kept as separate medicines, as Reimer and colleagues reported them separately.

ATC classes in ATC_SEED

ATC class	Medicines
HMG-CoA reductase inhibitors (C10AA)	simvastatin, lovastatin, pravastatin, atorvastatin, rosuvastatin, pitavastatin, fluvastatin
PDE5 inhibitors (G04BE)	sildenafil, vardenafil, tadalafil, avanafil
Alpha-adrenoreceptor antagonists (G04CA, C02CA)	tamsulosin, terazosin, alfuzosin, silodosin, doxazosin
Centrally acting muscle relaxants (M03BX)	cyclobenzaprine, baclofen, tizanidine
Other nervous system drugs (N07XX)	riluzole, edaravone, tofersen, sodium phenylbutyrate-taurursodiol, dextromethorphan-quinidine
Opium alkaloid antitussive (R05DA09)	dextromethorphan
Class IA antiarrhythmic (C01BA01)	quinidine
Oxicam NSAID (M01AC06)	meloxicam
Vitamin D (A11CC05)	colecalciferol
Direct factor Xa inhibitor (B01AF01)	rivaroxaban
Thiazide diuretic (C03AA03)	hydrochlorothiazide
ACE inhibitor (C09AA03)	lisinopril
Anticholinergic for sialorrhea (A03AB02)	glycopyrrolate
5HT3 antagonist (A04AA01)	ondansetron
Opioid analgesics (N02AB03, N02AA, N07BC02)	fentanyl, morphine, oxycodone, hydromorphone, methadone
Gabapentinoid (N02BF01)	gabapentin
Decongestant (R01BA02)	pseudoephedrine
Mucolytic (R05CA03)	guaifenesin
H1 antihistamine (R06AE07)	cetirizine
Short-acting beta agonist (R03AC02)	salbutamol (albuterol)
Leukotriene receptor antagonist (R03DC03)	montelukast

Brand names in BRAND_TO_GENERIC

Generic	Brand names
riluzole	Rilutek, Tiglutik, Exservan
edaravone	Radicava, Radicava ORS
tofersen	Qalsody
dextromethorphan-quinidine	Nuedexta
sodium phenylbutyrate-taurursodiol	Relyvrio, Albrioza
statins	Lipitor, Crestor, Zocor, Mevacor, Altoprev, Pravachol, Livalo, Zypitamag, Lescol
PDE5 inhibitors	Viagra, Revatio, Levitra, Staxyn, Cialis, Adcirca, Stendra
alpha blockers	Flomax, Hytrin, Uroxatral, Rapaflo, Cardura
muscle relaxants	Flexeril, Amrix, Fexmid, Zanaflex, Lioresal, Gablofen, Ozobax
glycopyrrolate	Robinul, Cuvposa, Glycate, Dartisla
ondansetron	Zofran, Zuplenz
fentanyl	Sublimaze, Duragesic, Actiq, Fentora, Abstral, Subsys, Lazanda
gabapentin	Neurontin, Gralise, Horizant
cough and cold	Sudafed (pseudoephedrine), Mucinex and Robitussin (guaifenesin)
cetirizine	Zyrtec, Reactine
bronchodilators	Ventolin, Proventil, ProAir, AccuNeb (salbutamol); Xopenex (levalbuterol)
montelukast	Singulair
lisinopril	Prinivil, Zestril, Qbrelis
rivaroxaban	Xarelto
colecalciferol	Vitamin D, Vitamin D3, D3, cholecalciferol

Medicines not in ATC_SEED are kept under "Not in ATC seed mapping". The log lists the most frequent unmatched names, which is the list to work from when you extend the two dictionaries for your own data.

Back to top

Outputs

File	Holds	Where run_all.py puts it
drug_repurposing_cohort.csv	One row per participant and medicine: pseudonymous ID, cohort entry, age, sex, race, ethnicity, exposure, criterion met, evidence types, follow-up end and its basis, vital status at the cut-off. Calendar dates are kept	restricted/drug_repurposing/
drug_repurposing_summary.csv	One row per medicine with ATC class, ALS status, counts and landmark medians, with suppression applied	restricted/drug_repurposing/ and reports/
drug_repurposing_report.html	The report, counts only	restricted/drug_repurposing/ and reports/
cure_id_review_worksheet.csv	One row per exposed participant and medicine, in the shape of a CURE ID case report, for clinical review. It is a local worksheet, not a CURE ID submission	restricted/drug_repurposing/
off_label_snippets.md	Example source records for common ALS symptom medicines, with years only	restricted/drug_repurposing/

--also-write-identified adds a cohort file with registry participant IDs; treat it as protected health information (PHI).

The cohort file is the input to a prespecified analysis, for example a Cox model or propensity score matching with lifelines or statsmodels. Reimer and colleagues matched with a caliper of 0.2 standard deviations of the logit of the propensity score and up to three controls for each exposed participant.

Back to top

Running it

run_all.py runs the module when --cutoff is given. To run it on its own:

python pipeline/drug_repurposing.py --bundle RUN/dashboard_data.json \
    --out RUN/restricted/drug_repurposing --cutoff 2026-10-31 \
    --vital-status RUN/vital_status.csv --min-exposed 5 --k 5

Option	Default	Meaning
--cutoff	latest record date, with a warning	Data cut-off
--vital-status	none	Deaths and alive confirmations from outside the records
--min-exposed	5	Fewest exposed participants for a medicine to be summarized (never below k)
--k	5	Small-cell threshold
--landmark-months	12	Landmark for the survival medians
--criterion	any	any, reimer-a, reimer-b or reimer
--cohort-name	ALS registry cohort	Name in the report header

The exposure windows, the cohort codes, the ATC seed, the brand names and the CURE ID age bands and race options are plain Python dictionaries and constants near the top of the module.

Back to top

Limits

• Confounding by indication. A participant prescribed a PDE5 inhibitor may be healthier than average, and one prescribed an opioid may have more advanced disease. Matching reduces this but does not remove it.
• Diagnosis accuracy. A coded motor neuron disease diagnosis has not been checked against clinical criteria. Reimer and colleagues ran a sensitivity analysis restricted to participants treated with riluzole, whose indication is specific to ALS; the same check suits a registry.
• Missing records. Participant-contributed records can miss medicines filled outside connected health systems and deaths not yet linked.
• No dose. Dose is not taken from the bundle, so dose-response analysis needs an extension of the medication parser.
• Many comparisons. Screening many medicines raises the chance of false signals; report how you correct for this.

The cohort file is the starting point for a prespecified analysis, which needs a protocol stating the comparison, time zero, eligibility, covariates and how missing records and outside care are handled. Findings from one registry generate hypotheses; randomized trials test them.

Back to top

Reference

Reimer RJ, Soper B, Wilson JL, Goncalves AR, Cadena J, Suarez P, Gryshuk AL, Osborne TF, Grimes KV, Ray P. Identification of drug repurposing candidates for amyotrophic lateral sclerosis using electronic health records: a retrospective cohort study. Lancet Digital Health. 2026. doi:10.1016/j.landig.2025.100963