Health Economics Metrics

Topic 67 of 76 · Consumer health apps and devices

Engagement Metrics

Engagement metrics measure how much users actually use a health app: DAU/MAU stickiness, session frequency and duration, feature usage. In digital health, engagement is not vanity — it is dose: the exposure through which any clinical effect must flow.

Why it matters

A drug that stays in the bottle heals nobody; an app that stays uninstalled or unopened is the same failure mode. Every health-economic claim for a consumer health product multiplies through engagement — efficacy shown in trials was measured at some usage level, and real-world value scales with how close deployment usage gets to that level. Standard product benchmarks: DAU/MAU around 20% is considered healthy for mobile apps generally, >25% exceptional; health apps often run lower.

The math

Stickiness (DAU/MAU) = daily active users / monthly active users × 100
Session metrics      = sessions/user/period; avg duration = total time / sessions
Feature engagement   = users performing key action / active users

Dose-response framing (the health-economics upgrade):
  realized effect ≈ trial effect × f(actual usage / trial usage)
  where f comes from dose-response analysis — see the "effective
  engagement" concept in adherence-and-persistence.md: enough usage
  to achieve the intended outcome, which may be modest and finite

Worked example

A blood-pressure app's pivotal study showed a 6 mmHg systolic reduction among users logging ≥4 readings/week. In deployment across 50,000 registered users:

MAU 20,000 (40%); of those, logging ≥4×/week: 7,000
Effective-dose users = 7,000 / 50,000 = 14% of the registered base

Population-level effect ≈ trial effect delivered to 14%, not 100%:
any economic model quoting "50,000 users × 6 mmHg" overstates ~7×.
Honest model: 7,000 × full effect + partial credit (from dose-response
data, if any) for the 13,000 sub-threshold users.

This multiplication — through the engagement funnel to the effective dose — is the single most common place digital-health economics inflate.

Software engineering connection

Engineers own the engagement funnel, which makes them owners of a clinical variable: onboarding friction, notification strategy, load time, and offline resilience all move the dose delivered. Two design implications: instrument the clinically meaningful action (readings logged, lessons completed), not opens — DAU built on notification-bounce sessions is dose-fraud; and treat engagement targets as sufficiency targets, not maximization — an app that achieves its outcome in 5 minutes/week and gets out of the way is clinically ideal and metrically "poor" (see effective engagement in adherence and persistence). Value the engagement work itself via the population-effect model above: a 2-point gain in effective-dose share is a quantifiable QALY line.

Pitfalls

  • Engagement as outcome: usage is a means; the outcome is the PROM or clinical endpoint.
  • Averages over bimodal usage: health-app populations split into devoted users and ghosts; means describe nobody — cohort it.
  • Dark-pattern dose inflation: streaks and guilt notifications lift metrics and can harm the anxious populations health apps serve; clinical products carry clinical ethics.
  • Vendor-benchmark provenance: most published engagement benchmarks come from analytics vendors, not peer review; calibrate against your own trials.

Sources