Health Economics Metrics

Topic 40 of 76 · Health system operational economics

Earlier Intervention

If saved capacity lets a practitioner review diagnostic backlogs sooner, patients move from waiting list to active treatment faster — and treating earlier is usually cheaper and better than treating later, because untreated conditions progress.

Why it matters

Disease progression is the compounding interest of health care. A patient waiting with an untreated condition is not in a steady state: cancers stage-shift, heart failure decompensates, mild depression becomes severe. Intervening earlier therefore delivers a double dividend — better outcomes (more QALYs, treated from a healthier baseline) and often lower treatment costs (early-stage treatment is less intensive than late-stage rescue). This mechanism is what elevates "faster pathways" from an operational nicety to a clinical and economic imperative — and it is the deep reason cost of delay applies to clinical software.

The math

Value of earlier intervention (per patient) =
    [Cost_late − Cost_early]                       (treatment-cost offset)
  + [QALYs_early − QALYs_late] × λ                 (health gain × threshold)
  × P(progression during the delay)                 (probability weighting)

The probability weighting is essential: not every waiting patient progresses. Model the transition probability per unit time (from natural-history data), not the worst case. Then discount: costs avoided years away are worth less today (discounting) — and note most early intervention is cost-effective rather than cost-saving (see prevention economics).

Worked example

Diabetic retinopathy screening backlog: 4,000 patients, 6 months behind. AI-assisted grading triples throughput and clears the queue in 8 weeks. Natural history: ~2% of waiting patients/year progress to sight-threatening stages while un-reviewed.

Progression events avoided by ~4 months' acceleration:
  4,000 × 2% × (4/12) ≈ 27 patients

Per avoided progression:
  treatment offset (intravitreal therapy vs laser) ≈ £4,000
  QALY gain (vision preserved) ≈ 0.8 QALYs × £20,000 = £16,000

Value ≈ 27 × (4,000 + 16,000) ≈ £540,000 — from one backlog cleared once,
before counting the permanent throughput gain.

Software engineering connection

Two transfers. First, the obvious one: software that accelerates diagnostic and treatment pathways (triage, AI grading, results routing) monetizes via this exact model — and the model tells you which pathway to accelerate: the one with the steepest progression curve, not the longest queue. Second, the engineering mirror: defects progress too. A bug caught in design costs a conversation; in production it costs an incident; the "shift-left" cost curve (10–100× by stage) is a progression model, and the honest version carries the same caveat — early detection is usually cost-effective, not free money, because reviews and tests have real costs and most caught issues would never have progressed.

Pitfalls

  • Worst-case progression assumed for everyone — the probability weighting is the difference between analysis and advocacy.
  • Lead-time bias: finding disease earlier without changing outcomes looks like benefit but isn't; earlier effective intervention is the claim, not earlier detection alone (see screening economics).
  • Double counting with waiting-list and RTT claims built on the same acceleration — one pathway improvement, one set of benefits, allocated once.

Sources