Industries / Health / Healthcare / Home health
Managing medication
Helping people at home take the right medicines at the right time: checking the list of medicines, filling pill boxes, using dispensers and reminders, and stopping medicines that are no longer needed.
5
Milestones achieved, from 2002 to 2017.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
2002
Nurses screen for drug problems
A trial in two large US home health agencies had nurses screen new Medicare patients aged 65 and over for medicine problems, with a pharmacist and the doctor following up. Problems such as duplicate drugs were fixed in 71% of patients against 24% with usual care, though psychiatric drugs and painkillers showed no gain.
Credit
Sylvia Meredith, Wayne Ray and colleagues, Vanderbilt University and the Visiting Nurse Service of New York
2005
Pharmacist visits backfire
In the HOMER trial in England, 872 people over 80 got two pharmacist home visits to review their medicines after an emergency admission. They had more hospital readmissions within six months than the control group: 234 against 178 (rate ratio 1.30).
Lessona medicine review at home can do harm, so each way of doing one needs its own trial.
Credit
Richard Holland and colleagues, University of East Anglia
2010
Stopping medicines safely
In Israel, 70 older adults living at home, average age 82.8 and taking 7.7 medicines on average, were assessed with a step-by-step method for stopping medicines. Stopping was recommended for 58% of their drugs; only 2% of those stopped had to be restarted, and 88% of patients reported better health.
Lessonwith no control group, the study shows the method can be done, not that it works better.
Credit
Doron Garfinkel and Derelie Mangin
2017
Smart pill bottles fall short
The HeartStrong trial gave 1,003 heart attack survivors electronic pill bottles, lottery rewards and support from a friend or relative for 12 months, and compared them with 506 on usual care. Medicine-taking did not change significantly (0.46 against 0.42), and neither did heart-related readmission or death (hazard ratio 1.04).
Lessontracking pill-taking at home did not by itself lead to better health.
Credit
Kevin Volpp, David Asch and colleagues, University of Pennsylvania
2017
Drug review becomes a measure
From 1 January 2017, US home health agencies had to report whether the nurse reviewed all of a patient's medicines and contacted a doctor by midnight of the next day about serious issues. The measure meets the medication reconciliation requirement, the matching of medicine lists across settings, of the 2014 IMPACT Act.
Credit
CMS, with the contractors Abt Associates and RTI International
Watch for
Dispensers cut visits in Finland
Finnish home care clients who used an automatic medicine dispenser with a video-call link were compared with 3 matched non-users each. Users had 24% fewer in-person home care contacts, 72% more remote contacts, and 18% lower yearly costs, a net saving of about €3,200 per person per year after device costs.
Achieved whena peer-reviewed paper names the device, gives patient numbers and confirms the result, ideally with funding not tied to the device seller.
Claimed by
Mariann Lassenius and colleagues (Medaffcon, Tamro and the Wellbeing Services County of Central Finland), ISPOR Europe poster, November 2026
A robot reads labels
University of New Hampshire researchers say advanced versions of their Stretch 4-based home robot help users read prescription labels, as well as giving medicine reminders.
Achieved whenlabel-reading accuracy is published on a stated set of labels.
Claimed by
University of New Hampshire researchers, reported June 2026
Open
01
Phone checks a pill organiser
A phone app that photographs a filled weekly pill organiser (7 days, 4 times a day) and compares each compartment with a written schedule, flagging missing, extra or wrong items. It is tested only with coloured candies or placebo tablets, never real medicines.
The briefCredit · unclaimed02
Bottle-label reading benchmark
A public set of at least 500 made-up pharmacy labels, printed, stuck on empty bottles and photographed, each with a reference entry for drug, strength, dose and frequency. Vision AI models are scored field by field on turning the photos into a medicine list.
The briefCredit · unclaimed
Open 01 · the brief
Phone checks a pill organiser
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A phone app that photographs a filled weekly pill organiser (7 days, 4 times a day) and compares each compartment with a written schedule, flagging missing, extra or wrong items. It is tested only with coloured candies or placebo tablets, never real medicines.
Why it is still open
The closest work found is a US patent on imaging filled blister packs in pharmacies and hobby webcam sketches; no tested tool for home pill organisers turned up.
Built with what exists
a phone camera; an object-detection model such as Ultralytics YOLO11, or a vision-language model run on a laptop through llama.cpp (see the Note "A first stable number for local AI"); Label Studio for labelling photos.
First step
Buy 3 organiser brands and 6 kinds of candy, write 20 schedules, and photograph 40 correctly filled organisers.
Done looks like
On 200 staged organisers with planted errors, across 3 organiser brands and 2 lighting conditions, at least 95% of errors caught with no more than 1 false flag per 20 correct organisers, with the photos and scores published.
Open 02 · the brief
Bottle-label reading benchmark
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of at least 500 made-up pharmacy labels, printed, stuck on empty bottles and photographed, each with a reference entry for drug, strength, dose and frequency. Vision AI models are scored field by field on turning the photos into a medicine list.
Why it is still open
The closest recent benchmark, RxScribe Bench (September 2026), covers handwritten Indian prescriptions, not printed labels on curved bottles.
Built with what exists
a label generator using made-up names; a home printer; a phone camera; several vision AI models, including Claude for document reading (see the Note "Seven more flagged").
First step
Write the label template and generate 50 labels, including curved-surface and glare cases.
Done looks like
The dataset and scoring code released, and each model given an error rate for drug name, strength, dose and frequency, with wrong-drug errors reported separately.