Four decisions, each with a price someone else felt first

Making a cellular kit the default for patients who never paired a phone

When I joined Ardenwick in March 2021, our standard kit paired a Bluetooth scale and blood pressure cuff with the patient's own smartphone, or with a tablet we supplied. Our reports counted kits shipped, not readings received. When we looked, 32 per cent of patients had sent nothing within thirty days of delivery, and among those aged 75 or over the figure was 49 per cent. Those patients were enrolled, counted by our customers and invisible to their nurses.

The stakes were clinical before they were commercial. A heart-failure patient who never transmits a weight is not being monitored, whatever the enrolment list says, and the practice cannot bill for a programme the patient is not using.

I proposed making cellular devices, delivered already linked to the patient's record, the default for every new patient. The kit cost would rise from US$88 to US$154, plus a data charge of about US$2.10 a month. Our CFO, Grady Pelletier, accepted the clinical case but wanted a US$12 monthly surcharge for cellular kits. I argued against it because practices would then choose the cheaper kit for exactly the patients who needed the simpler one. Thaddeus Brannigan, our chief executive, backed the no-surcharge plan on condition that we held the year's operating budget, so we deferred integration of a new glucose meter by two quarters.

We switched the default between January and September 2022, kept the app for patients who asked for it, and replaced the instruction booklet with three pictures printed inside the lid.

Gross margin fell from 62 per cent in 2021 to 56 per cent in 2022, then recovered to 60 per cent by 2024 as volumes rose and set-up calls fell by 44 per cent. The share of patients sending a reading in their first thirty days rose from 68 to 90 per cent, and among those 75 or over from 51 to 87 per cent. Median time from delivery to first reading fell from nine days to three. First-reading rates now open every quarterly customer review.

I would do two things differently. I would have tested the kit with caregivers, who unpacked about a third of them, before the redesign rather than after it. And I would have measured which patients chose the app, because for two years we did not know.

Halving heart-failure alerts and reviewing every admission we did not flag

By late 2022 our heart-failure programme was producing 74 alerts a week for every 100 patients. Most came from fixed rules applied to every patient alike, such as a set weight gain in a day or a week. Nurses spent 34 minutes per patient each month, much of it clearing alerts that led nowhere. Two customers admitted their nurses now cleared alerts in batches, and an alert system nobody reads is worse than none, because everyone assumes it works.

My team proposed personal baselines. Each patient's weight and blood pressure would be compared with their own recent average, symptom answers would raise or lower priority, and alerts would fall into three tiers: call today, review this week, and information only. Every rule stayed visible to the clinician, who could change it for any patient. We tested the design on 2022 data covering 9,400 patients and 388 heart-failure hospitalisations. It would have flagged 11 of those hospitalisations earlier than the old rules and 6 later, by one to three days.

Those six were the trade-off, and I put them in the first paragraph of my paper. Our chief medical officer, Dr Alma Teague, supported the change on condition that her team examined every hospitalisation that no alert had preceded. Dr Theo Marchetti, who leads cardiology at Fernhollow Health System, one of our largest customers, objected. His patients were taught the fixed rules at discharge, and he did not want our software to contradict their teaching. We kept the old rules available as a customer setting. Nine of our 140 heart-failure customers, Fernhollow among them, kept them for the first year.

We released the new design in April 2023. Alerts fell to 31 per 100 patients a week and monitoring time to 21 minutes per patient a month. Heart-failure hospitalisations among enrolled patients were 17.2 per 100 patient-years in 2022 and 16.8 in 2024. I would not claim the redesign caused that difference, only that it did not make things worse. Fernhollow adopted the new rules in 2024 after reviewing our quarterly list of missed admissions.

I would start that review on release day; we began it three months late. I would also have rewritten the discharge teaching with Dr Marchetti's team before release, instead of leaving patients with rules our nurses no longer applied.

Taking the heart-failure risk score to the FDA instead of calling it decision support

In 2022 our data scientists built a model that estimated, from a heart-failure patient's weights, blood pressure, heart rate, symptom answers and medication changes, the likelihood of a worsening episode in the coming days. Our commercial team wanted to release it in early 2023 as an insights panel that listed the readings behind each score, and to present the whole as clinical decision support outside device regulation.

The stakes were larger than one feature. Releasing what was in substance a device without a device's evidence and controls would expose customers, patients and the company; the device route would cost at least two years of revenue.

Lydia Habersham, who leads regulatory affairs and quality, and I concluded that the score did not fit the exclusion. It analysed readings from connected devices, and a nurse could not independently review how it reached a result in the few minutes she had for each patient. I recommended that we treat it as a device. Nathaniel Strathby, our chief revenue officer, disagreed strongly: competitors were promising similar features, and we would be absent from a category we had helped to create. The board accepted my recommendation in November 2022, with a budget of US$7.4 million.

We ran a prospective study at four health systems with 2,900 patients through 2023 and 2024, built the software under our design controls and made our submission in November 2024. The FDA granted marketing authorisation in September 2025, including a predetermined change control plan for retraining. Two customers worth about US$3.6 million a year in combined revenue left before it arrived.

In the study the score identified 61 per cent of heart-failure hospitalisations at least five days ahead, with about one score alert per patient every seven weeks. It performed less well in patients with advanced kidney disease, and the labelling says so. By August 2026, 51 customers were using it for about 13,400 patients. Every product proposal now opens by saying whether a nurse could verify the output during an ordinary shift.

I would bring regulatory affairs into the room at the first prototype, not after the commercial plan was drafted; that sequence cost us four months of argument. And I would have given customers an honest date at the outset; two broken promises of a release later in the year cost more trust than the delay.

Holding a COPD launch for a nine-month pilot

In 2023 three large customers asked for a programme for chronic obstructive pulmonary disease, and our board wanted a second growth line. The sales plan assumed about US$11 million of annual recurring revenue by the end of 2025. The obvious design was what others sold: a connected pulse oximeter, daily readings and alerts below a saturation threshold.

I doubted it would help patients. Our clinical informaticists, two of them former respiratory therapists, pointed out that saturation in many people with COPD moves for reasons that do not call for a nurse, that exacerbations often announce themselves first through breathlessness and sputum, and that pulse oximeters can overestimate saturation in people with darker skin, a concern the FDA has been examining. Daily readings might recreate the alert burden we had just removed from heart failure.

I proposed a nine-month pilot with two health systems, with measures agreed in advance: the share of alerts that led to a change in care, the share of patients still taking readings at twelve weeks, and emergency visits. One investor director argued that the market would not wait and we could learn after launch. Thaddeus Brannigan sided with the pilot, and we bore the cost. The largest prospect, with about 2,500 eligible patients, signed with another vendor in 2024, and we gave up an estimated US$4.2 million of revenue that year.

The pilot ran from February to October 2024 with 640 patients. Only 9 per cent of oximetry alerts changed care, and 38 per cent of patients had stopped daily readings by week eight. We rebuilt the programme around a weekly symptom check-in, a written action plan agreed with each patient's clinician, and oximeter readings requested when symptoms changed rather than every morning.

We released it in March 2025. By December, 5,000 patients were enrolled, 31 per cent of alerts led to a change in care, and four in five patients were still answering the weekly check-in after twelve weeks. The programme is smaller than the sales plan assumed and, I believe, more useful.

I would agree a stopping rule and decision date with the board at the start, rather than defend the pilot afresh each quarter. And I would have asked patients from a pulmonary rehabilitation class into the first design session; in one afternoon they told us what the pilot took months to show.