Health Innovation Adoption District (3D)

A Blender-modelled hospital district where a new technology spreads building by building: Bass diffusion against an agent-based model.

Switch to 2D

Presets

Playback

Who influences whom

Model

Camera

Layers

Quality

Model check

The science

What is modelled

How a new health technology (for example a remote-monitoring service or a new imaging workflow) spreads through a set of clinics. Each building is a clinic agent; in the 2D page it is a square. Each clinic is an agent that either uses the technology or does not. A clinic adopts because of external influence (trial publications, regulators, a campaign) at rate p, or because of imitation of clinics it is in contact with, at rate q times the share of its contacts that have already adopted. Adoption is permanent. This is a model of how adoption spreads, not of whether the technology works.

Equations

  • Bass (1969) mean-field law: dF/dt = (p + q F)(1 - F), with the closed form F(t) = (1 - e-(p+q)t) / (1 + (q/p) e-(p+q)t).
  • Peak adoption rate at t* = ln(q/p)/(p+q) (only if q > p; otherwise adoption just decelerates). The time to reach half of all clinics is t50 = -ln[0.5/(1 + 0.5 q/p)]/(p+q).
  • Agent model: in a time step dt a clinic that has not adopted does so with probability 1 - exp(-(p + q a/k) dt), where a of its k contacts have adopted. If the contacts are a random sample of everyone, the expected value of a/k is F and the model reproduces the Bass curve; if contacts are only nearby clinics, imitation spreads as a slow wave and the curve is later and lower at the start.
  • The yellow dashed reference curve is the mean-field equation integrated with Runge-Kutta 4 (with the campaign included when the boost is above zero).

Checks run in the page

  • RK4 against the closed-form Bass curve.
  • Numerically found peak of the adoption rate against ln(q/p)/(p+q) for three parameter sets.
  • A 4000-clinic well-mixed agent run (mean of five seeds) against the Bass curve.
  • Irreversibility and completeness: every clinic adopts eventually, and the external plus imitation counts add up to the total.

Honest simplifications

The parameters are illustrative: published Bass fits for other products often have p around 0.01 to 0.05 and q around 0.3 to 0.5 per year, but a real health technology depends on evidence, cost, reimbursement, training and regulation, none of which is modelled. All clinics are identical, there is no price, no competition, no abandonment and no market-size growth, contact networks are idealised (a regular grid with optional random long links), and the grid has only 324 clinics so single runs are noisy. The model says nothing about patient outcomes.

What to try

  • Keep the network well mixed and compare the white curve with the yellow one; then switch to local contacts and watch the delay.
  • Add a few long links (small-world): adoption speeds up much more than the 10 percent of rewired links suggests.
  • Set q below p: the S-curve disappears.
  • Switch on the campaign (boost 3 to 10) in year 3 and see how long it takes imitation to take over.
Loading model...
Loading Blender model...
Adoption S-curve
New adopters per year
External push versus imitation
Phase plot: rate versus share
Made with 🟠 Blender · Three.js