Health Economics Metrics

Topic 68 of 76 · Consumer health apps and devices

Retention and Churn

Retention measures what fraction of a user cohort is still active N days after starting (D1/D7/D30 curves); churn is its complement. The brutal digital-health baseline: roughly 90% of health-app users abandon within 30 days — digital health D30 retention runs ~3–4% against an all-app average of ~6%.

Why it matters

Eysenbach named this in 2005: the law of attrition — losing users at high rates is an intrinsic, structural property of eHealth interventions, not an implementation bug, with attrition in eHealth trials routinely exceeding 50%. The economic consequence is total: retention defines the treatment window within which any benefit can be delivered, and the unit economics — CAC paid per user who stays 12 days delivers neither LTV nor QALYs. Any economic model for a consumer health product that doesn't weight benefits by the retention curve is describing a product that doesn't exist.

The math

Retention_Dn = users active on day n / cohort size × 100
Churn rate   = users lost in period / users at period start × 100

Benefit weighting (the health-economics move):
  expected benefit per acquired user = Σ_t retention(t) × benefit rate(t)
  ≈ area under the retention curve × per-time benefit
  — NOT trial benefit × 100% of acquired users

Cost per retained-at-D30 user = CAC / D30 retention
  (at 4% D30, a £5 CAC is really £125 per retained user)

Worked example

A mental-health app: trial showed 0.02 QALYs gained per user completing 8 weeks. Deployment cohort of 100,000 downloads, retention D7 25%, D30 8%, week-8 4%:

Completers          = 100,000 × 0.04 = 4,000
QALYs delivered     = 4,000 × 0.02 = 80  (not 100,000 × 0.02 = 2,000)
At £20,000/QALY     = £1.6M of health value (not £40M)

Per-download health value = £16 — the number that should set what a
payer will pay per download, and it is 4% of the naive claim.
Retention improvement case: moving week-8 completion 4% → 6% adds
40 QALYs/year ≈ £800k — retention engineering IS health production.

Software engineering connection

Retention is the metric where product engineering most directly manufactures health value, per the arithmetic above. The practices that move it are ordinary: onboarding time-to-first-value, re-engagement design, performance, and crucially planned dose completion — a program with a defined end (8 weeks, then graduation) should measure completion, not perpetual DAU, aligning the metric with the clinical model instead of the ad-funded attention model. Survival analysis is the right toolkit (the same Kaplan-Meier math as life-years gained); segment curves by acquisition channel, since channel mix changes retention more than most features do.

Pitfalls

  • Intention-to-treat laundering in reverse: trials report completers; deployment economics must count everyone acquired (Eysenbach's core warning).
  • Retention theater: notification-driven "active" users who never perform the therapeutic action (see engagement metrics).
  • Comparing curves across definitions: "active" defined as open vs meaningful action shifts D30 by multiples.
  • Ignoring who churns: if the sickest churn fastest, per-user benefits fall as retention improves among the healthy — pair curves with case-mix (see reach and equity).

Sources