Six disciplines between a patient's scale and a nurse's call

Most of our customers bill Medicare and other payers for the monitoring their patients receive, using codes that pay for device set-up and patient education, for device supply in months with enough days of readings, and for clinicians' management time. My job is to make accurate billing easy without letting the codes design the product. Our clinic software shows each practice which patients have enough days recorded and how much management time has been documented, because the practice carries the compliance risk and needs to know. The minute count pauses whenever a record sits idle on the screen. I have declined features whose purpose was to fill the count, such as reminder sequences keyed to the days left in a billing period. Since 2023 clinicians can move stable patients onto reduced reading schedules, nearly a quarter of hypertension patients have moved to one, and we created a lower maintenance fee so that the practice is not penalised for the sensible choice.

I divide our roadmap early into two kinds of function. The first supports a clinician who can check the reasoning herself: reading displays, rules the clinician sets and can read, care-plan reminders. We build those under our ordinary software process with clinical review. The second asks the clinician to rely on the software's judgement, as our heart-failure risk score does, and we treat it as device software from the first prototype, built under design controls in our quality management system and taken to the FDA with prospective evidence. Lydia Habersham, who leads regulatory affairs and quality and reports to our chief executive, owns the regulatory conclusions. I own the decision to put a feature on the device path and to fund what that path requires. The FDA's thinking on decision support has shifted over the years, so we revisit each classification whenever its guidance changes rather than treating an earlier conclusion as settled.

An alert is a request for a nurse's time, and I manage it as a budget. I track alerts per 100 patients per week, monitoring minutes per patient per month and the share of alerts that lead to a change in care, and I report all three to customers. When our heart-failure programme reached 74 alerts per 100 patients a week in 2022, nurses at two customers admitted clearing them in batches. We replaced fixed population rules with each patient's own baseline and sorted alerts into three tiers, which brought the figure to 31 and monitoring time from 34 minutes to 21. Every rule remains visible to the clinician and adjustable for any patient. Dr Alma Teague, our chief medical officer, leads a monthly review of every admission that no alert preceded, and its findings go back into the defaults. I learnt the underlying lesson at Quarrystone, where an alert with sound logic failed because it arrived in the middle of a task.

Our patients' median age is 73, and many have arthritis, reduced vision, hearing loss or mild cognitive impairment. I treat accessibility as a count of steps and dependencies between opening the box and sending a first reading, measured in patients' homes rather than on our process maps. Moving to cellular devices that arrive linked to the patient's record took that count from about twenty to three and raised first-reading rates among patients over 75 from 51 to 87 per cent. About a third of kits are unpacked by a relative or neighbour, so the caregiver is designed for as a second user: up to two named caregivers per patient, each with a login showing only what the patient has agreed to share. Patients with hearing loss can choose a text message with a call-back time, a captioned call on a supplied tablet or a call to their caregiver. Instructions are printed as three pictures inside the lid.

Ardenwick acts as a business associate of the health systems and practices we serve, so our use of patient information is governed by HIPAA and by the business associate agreement each customer signs. In product terms that means three habits. Every new use of data is checked against those agreements before design begins, not after a feature is built. Family and caregiver access follows the patient's recorded choice and is visible to the clinician beside every contact. Model development uses data only as our agreements allow, with de-identification and customer permission where they are required, and we keep a register of which datasets trained each model version. In 2023 I recommended declining a manufacturer's offer to fund medication messages to our patients, for reasons I describe on my Board Readiness page. We also run a security review on every connected device and firmware release before it enters the kit, a practice I brought from Pellmore.

I do not allow a new condition or a new kind of alert onto general sale until a pilot has shown that it leads to care, and I ask for the measures to be agreed before the pilot starts rather than chosen afterwards. The measures are usually the share of alerts that changed treatment or follow-up, the share of patients still taking readings at twelve weeks, and an outcome the customer's clinicians care about, such as emergency visits. Our COPD pilot in 2024 failed the first two measures for a daily oximetry design, and we rebuilt the programme around weekly symptom check-ins before release. For regulated software the evidence is prospective and multi-site: the heart-failure risk score was studied in 2,900 patients at four health systems, and the study exposed weaker performance in advanced kidney disease that our retrospective data had hidden. We publish what we learn, including results that flatter us less than our sales materials would like.