Build Medical Software From Evidence to Software Lesson 2 of 5 intermediate 7 min read

A Protocol Is a Memory System, Not a Patient

Clinical protocols reduce avoidable variation, but software must preserve their population boundaries, exceptions, provenance, and judgment points.

In This Lesson

Read with a defined objective.

View the complete course

Learning objectives

  • Define the applicability boundary of a clinical protocol.
  • Create a protocol map with explicit exceptions and judgment points.
  • Preserve source, version, and local adaptation metadata.

Prerequisites

  • A completed evidence card from Lesson 1.

Build Medical Software From Evidence to Software

Listen to this post

A Protocol Is a Memory System, Not a Patient

0:00 / 0:00
A precise clinical protocol grid bending around a complex patient silhouette without enclosing it
On this page5 sections

The protocol PDF is open. The patient fits the first three criteria and misses the fourth by one detail that the summary table never anticipated.

This is not a failure of evidence-based medicine. It is the reason clinical judgment exists.

A protocol is a memory and safety system. It carries a professional organization’s best current recommendation into a busy clinical environment. It reduces avoidable variation. It keeps a consequential threshold from depending on whether someone remembers the correct number at the end of a long call night.

But a protocol is still a model of reality.

The Applicability Boundary

Every protocol has an applicability boundary. Inside that boundary, the recommendation describes the intended population and clinical setting. Outside it, the same recommendation may become unsupported, incomplete, or wrong.

The boundary is often distributed across the document:

  • inclusion and exclusion criteria;
  • footnotes beneath a table;
  • definitions from an earlier section;
  • exceptions described only in prose;
  • assumptions about local resources;
  • cautions that do not fit into the main algorithm.

These details are easy to miss when a protocol is read quickly. They are even easier to erase when it is converted into software.

Decompose the Protocol Before You Simplify It

Do not begin by asking what the app should display. Begin by taking the protocol apart.

Create a protocol map with six categories:

  1. Inputs. What patient facts does the recommendation require?
  2. Decision points. Which facts change the branch?
  3. Outputs. What action, interval, classification, or recommendation follows?
  4. Exceptions. Which conditions interrupt the ordinary sequence?
  5. Sources. Which guideline section supports each rule?
  6. Judgment points. Where must a clinician interpret, individualize, or decline the recommendation?

The judgment points are not defects waiting to be automated. They are part of the design.

Protocols Need Provenance

A protocol without a source is advice. A protocol without a date is a future liability.

The software version should retain the sponsoring organization, document title, publication date, relevant section, local adaptation, and last clinical review date. If the institution modified the national recommendation, record that difference explicitly.

I have argued elsewhere that a protocol assistant needs retrieval, provenance, and source freshness. The same standard applies to a calculator or decision tree. The interface should never make a locally adapted rule look like universal evidence.

Conflicts Are Data

Professional organizations sometimes disagree. A local protocol may differ from a national guideline. A recommendation may change while the older document remains in circulation.

Do not resolve that conflict silently in code.

Record which source governs the application and why. If more than one pathway is defensible, represent the disagreement or narrow the application’s intended use. Hidden reconciliation creates false certainty.

Your Exercise

Select one protocol that matters in your practice. Build its protocol map.

Circle every exclusion. Mark every statement that requires interpretation. Add the source and review date beside every consequential threshold. Then write the sentence the eventual application must show when the patient falls outside the protocol.

The protocol organizes the recommendation. It does not contain the patient.

Share this article

Share X / Twitter Bluesky LinkedIn

Related articles

Clinical research papers narrowing toward a bedside monitor while uncertainty remains visible around the patient
Clinical Software

Evidence Is Not Yet a Clinical Instruction

Clinical evidence describes populations. Physician-developers must preserve its limits before translating it into protocol logic or software.

· 7 min read
evidence-based medicineclinical protocolsclinical software
A clinical guideline unfolding into an inspectable decision tree with thresholds, branches, and exception paths
Clinical Software

Make the Clinical Logic Visible Before You Code

Translate a clinical protocol into explicit inputs, branches, thresholds, exceptions, and failure states before choosing a programming language.

· 8 min read
clinical algorithmsdecision treesclinical protocols
A physician and patient evaluating a translucent software recommendation together at the bedside
Clinical Software

The Last Step Is Not Software. It Is Judgment.

Clinical software can execute protocol logic perfectly and still be wrong for the patient. The physician checkpoint must remain part of the architecture.

· 8 min read
clinical judgmentclinical decision supportautomation bias
Chukwuma Onyeije, MD, FACOG

Chukwuma Onyeije, MD, FACOG

Maternal-Fetal Medicine Specialist

MFM specialist at Atlanta Perinatal Associates. Founder of CodeCraftMD and OpenMFM.org. I write about building physician-owned AI tools, clinical software, and the case for doctors who code.