Measuring engagement and retention among prescribers and users

The account that looked healthy until it wasn't

A regional sales director at a medical device company noticed something odd in early 2025. One of her top hospital accounts still placed orders every month. Revenue looked flat, not falling. But the reorder intervals had quietly stretched from 21 days to 34 days, and the mix had shifted toward the cheapest SKU (stock keeping unit, a specific product variant). Six weeks later the account switched suppliers. The revenue drop showed up in Q3. The engagement drop had been visible in Q1.

That gap, between when behavior changes and when revenue changes, is what post-adoption measurement exists to close. Everything upstream is already settled here: the stages a cohort passes through are the ones the adoption funnelfunnelThe customer journey from awareness to purchase, typically Awareness, Interest, Consideration, Decision, Action, with prospects narrowing at each stage.View full definition → lesson maps, and what it cost to win the committee is the acquisition lesson's problem. You have the customer. The question is whether they are still using you, and how you would know before the purchase order stops.

Why engagement is measured differently here

In consumer software you count logins. Here the usage happens inside a clinical workflow you cannot see: a script written in an EHR you do not own, an implant chosen behind OR doors. So you proxy it, and before you count anything you have to say out loud what "active" means.

An active-use definition needs three parts: a behavior, a window, and an exclusion rule. Definitions that survive an audit look like this:

• A device account is active if it ran at least one case in two of the last three months. One month is noise; a full quarter hides a stop.
• A prescriber is active at two or more new-to-brand scripts in the trailing quarter, excluding samples and single-patient exceptions.
• A patient on a connected-device program is active with a reading in four of the last eight weeks. Chronic-care programs of the Livongo type live on exactly this distinction between enrolled members and engaged ones, because the payer or employer paying the fee is not the person taking the readings.

The exclusion rule is where dashboards rot. If your "engaged prescriber" count includes anyone who clicked a link in your own outreach email, you are measuring your email cadence, not their behavior. Strip rep-triggered and marketing-triggered events out of the numerator, or park them in a separate column labelled as such.

1. Reorder rate and reorder interval

For consumables (reagents, catheters, test kits, injectables), the cleanest signal is whether accounts keep buying and how fast.

Reorder rate = share of accounts that placed a repeat order within a defined window.

Worked example. You sold to 400 accounts in H1 2025. By the end of H2, 300 of them reordered.

Reorder rate = 300 / 400 = 75%.

But the average hides drift. Segment it by tier. If reorder rate among your top-quartile accounts fell from 90% to 78% while the overall number held at 75%, your best customers are leaving and new small ones are masking it.

Track the reorder interval too: the median days between orders per account. A lengthening interval is an early warning even when the account technically still "reorders."

2. Procedure share (share of a physician's cases)

For implants, surgical tools and interventional devices, the metric that matters is procedure share: of all the eligible procedures a physician performs, what fraction use your product?

A cardiologist who does 40 stent cases a month and uses your stent in 12 of them gives you a 30% procedure share. If that slips to 20% next quarter, you are being designed out of cases even if raw unit volume looks stable, because their total caseload grew.

You rarely observe this directly. You triangulate from your own units sold plus third-party procedure volume for that account or territory (IQVIA is the best-known vendor of such datasets; sourcing and judging that reference data is the benchmark lesson's job, not this one).

The privacy constraints bite harder than most teams expect. Prescriber-identifiable data is legal in most of the US, but the AMA's Physician Data Restriction Program lets individual doctors block their prescribing data from being shared with sales reps, so your prescriber-level persistence file has holes you cannot see. Patient-level data has to be de-identified under HIPAA (the US Health Insurance Portability and Accountability Act) before it reaches a marketing team, and in the EU health data is a special category under GDPRGDPREU regulation governing how organizations collect, store and use personal data, with fines tied to global revenue for breaches.View full definition → needing its own lawful basis. The enforcement is real: the FTC fined GoodRx $1.5 million in February 2023 and BetterHelp $7.8 million a month later over health data shared with advertising platforms, and in late 2022 HHS warned health organisations about third-party tracking pixels on patient-facing pages. If your retention dashboard is fed by a pixel on a patient portal, that is the exposure.

3. Formulary persistence (for drugs)

A formulary is the list of drugs a payer or hospital will reimburse or stock. Getting on it is acquisition. Staying on it, at a favorable tier (the position that sets the patient's copay), is retention.

Formulary persistence tracks whether you hold your positions across review cycles and whether prescribing continues once you are listed. Two related drug-specific metrics:

• New-to-brand (NBRx): prescriptions to patients starting your drug for the first time. An acquisition signal.
• Persistence / adherence: the share of patients still filling your drug after 6 or 12 months. Retention at the patient level, and the input the lifetime valuelifetime valueLifetime Value: the total revenue (or profit) a customer generates throughout their entire relationship with your business.View full definition → lesson's multi-stream model needs.

If NBRx is strong but 6-month persistence is weak, you have a leaky bucket: you win starts and lose them before the revenue compounds.

Turning engagement into a drift score

The point is to flag accounts drifting *before* revenue falls. Build an account health score from directional changes, not absolute levels.

# Simple account drift flag (illustrative)
def drift_flag(reorder_interval_change, procedure_share_change, sku_mix_shift):
    score = 0
    if reorder_interval_change > 0.15:   # interval stretched >15%
        score += 2
    if procedure_share_change < -0.10:   # lost >10 pts of share
        score += 3
    if sku_mix_shift == "down":          # trading to cheaper SKU
        score += 1
    return "AT RISK" if score >= 3 else "OK"

Deliberately crude. Two things then decide whether it earns its keep. First, audit it: of the accounts flagged AT RISK two quarters ago, what share actually churned or lost share? Below roughly half and you are spending rep hours on false alarms. Second, do not attach compensation to the flag. Tie a bonus to "active accounts" and someone will place a small order to reset the reorder clock, which turns your leading indicator into a lagging one.

The plumbing matters as much as the logic. A customer data platformcustomer data platformSoftware that unifies customer data from every source into one persistent profile that marketing, sales and service teams can act on.View full definition → like Segment (now part of Twilio, and it sells this tooling) exists to unify identity across touchpoints, and here you deliberately cripple half of that: patient identifiers stay out of the marketing stackmarketing stackThe connected set of software tools a marketing team uses to plan, run, measure and automate campaigns across channels.View full definition → and join only in de-identified or aggregate form. Conversational tools like Drift capture intent on a reorder portal, which is useful engagement data and also a promotional surface subject to the same claim review as a printed brochure. Doximity, where the large majority of US physicians hold a profile, sells pharma and device brands access to those physicians and reports back campaign-level engagement; that aggregate-in, aggregate-out shape is what compliant prescriber engagement reporting usually looks like.

Reference ranges for retention

Two numbers worth carrying, treated as approximate ranges rather than targets (the benchmark lesson handles how to judge a gap):

• Medication persistence at 12 months for many chronic oral therapies is commonly cited around 40% to 60%, and lower for asymptomatic conditions. See the CDC on medication adherence. Half your starts can be gone within a year, which makes persistence programs a marketing responsibility, not just a clinical one.
• Net Promoter Score (NPS), the 0 to 10 "would you recommend" question, is used with prescribers and procurement contacts, but sample sizes are small and switching costs are high, so a mediocre NPSNPSNet Promoter Score (NPS) measures customer loyalty by asking how likely customers are to recommend a brand, then subtracting detractors from promoters.View full definition → can sit happily on top of very sticky revenue.

🎬 [VIDEO: "Customer Retention & Cohort AnalysisCohort AnalysisCohort analysis groups users by a shared starting trait or time (such as signup month) and tracks their behavior over time to reveal retention and lifecycle patterns.View full definition → Explained" - youtube.com - a clear, sector-neutral primer on cohort retention curves you can mapmapUsing software to automate repetitive marketing tasks and campaigns, enabling personalisation at scale across channels like email, web, and social.View full definition → onto reorder and persistence data]

Cohorts: the honest way to read retention

Average retention lies. Cohort analysis does not. Group accounts (or patients) by when they started, then track each group's survival.

Example. Take every account acquired in Q1 2025 and measure what fraction is still ordering at month 3, 6, 9, 12. Do the same for Q2 and Q3. If each newer cohort retains worse at month 6, your acquisition quality is degrading, perhaps because you are winning price-sensitive accounts that never intended to stay. Blended averages would hide this for a year.

The same technique works for drug persistence: cohort patients by start month and plot the survival curve.

What drift costs, and the three ways teams get it wrong

Retention curves feed the value model built in the lifetime value lesson, and the sensitivity there is steep enough that catching drift early usually beats winning another logo. What that lesson cannot do for you is fix a broken measurement culture. Three failure modes recur.

Churn asymmetry. A reagent account that skips a quarter has almost certainly churned. A capital equipment account that orders nothing for four quarters is behaving normally, because the disposables and the service contract are the recurring part. Applying one churn window across a mixed portfolio produces a number nobody in the room believes, and once finance stops believing it, the drift flag loses its budget.

Engagement theatre. Portal logins, webinar attendance and email opens are easy to move and easy to report. None of them is procedure share. A team that hits its engagement targets for three quarters while procedure share slides has optimised the proxy and lost the behavior. Keep at least one hard behavioral metric per product line that marketing cannot influence directly.

Firewall breaches. Patient-level persistence data from a support program or a connected device is collected with patient authorisation for care and program administration. Routing it back to target the prescribing physician is the kind of second-order use that ends in a consent decree. Separately, engagement generated by rep meals and honoraria becomes a reported transfer of value under the US Sunshine Act above a low dollar threshold, published and searchable in Open Payments, so "we increased physician engagement" is a claim your competitors can read.

Putting it together

1. Pick the right stickiness metric per product type: reorder for consumables, procedure share for devices, formulary and patient persistence for drugs.
2. Write the active-use definition down, including the exclusion rule, and get it reviewed once by legal and once by the analytics owner.
3. Track changes and intervals, not just levels, and segment by account tier.
4. Build cohorts so averages cannot hide degradation, then hand the survival curves to whoever owns the value model.

Key takeaways

• Behavior drifts before revenue drops. Lengthening reorder intervals, falling procedure share and slipping formulary tiers lead; revenue lags by a quarter or two.
• Define active use before you count it. A behavior, a window, and an exclusion rule that removes engagement your own outreach created.
• Match the metric to the product, and do not apply one churn window across capital, consumables and drugs.
• Aggregate in, aggregate out. De-identification, physician data opt-outs and pixel exposure decide what you are allowed to measure; the FTC's 2023 health-data cases show the price of getting it wrong.
• Audit the drift flag and never bonus it. If fewer than half the flagged accounts deteriorate, retire it; if pay depends on it, someone will reset the clock with a token order.