1 Why this article belongs here
Most of my writing focuses on healthcare data analytics, data engineering, and healthcare AI. At first glance, a short article about glucose and insulin may look more like a biology explainer than a data topic.
But GLP-1 medicines have changed the shape of the problem.
GLP-1s are no longer only a diabetes drug class. They now sit across obesity care, cardiovascular risk, pharmacy demand, PBS policy, private-pay access, food consumption, companion nutrition, and long-term real-world outcomes. The market signal is clear: Australia is entering an access and utilisation inflection point.
That means analysts will soon be asked harder questions:
- Who is eligible?
- Who is actually treated?
- Who starts, stops, switches, or restarts therapy?
- Which outcomes move first: weight, HbA1c, cardiovascular risk, medication burden, hospitalisation, food purchasing, or pharmacy spend?
- How should PBS exposure be modelled if indication, adherence, and persistence vary by cohort?
- What evidence is strong enough for policy, commercial, or clinical decisions?
To answer those questions well, we need a shared baseline. Before building dashboards, forecasting models, AI summarisation workflows, or claims-based cohorts, we need to understand what these medicines are acting on: the relationship between glucose, insulin, glucagon, appetite, and metabolic signalling.
This article is that starting point.
2 The metabolic baseline: what happens after eating?
Imagine a person has just eaten a meal or consumed glucose. Within minutes, carbohydrates are broken down into glucose, which enters the bloodstream.
In a simplified two-hour window, three things matter.
2.1 1. Glucose rises first
Blood glucose increases as absorbed carbohydrate enters circulation. In a healthy response, this rise is temporary. The body detects the increase and begins moving glucose into tissues where it can be used or stored.
However, for patients on GLP-1 therapy, this curve is fundamentally altered. GLP-1s significantly slow gastric emptying, meaning glucose enters the bloodstream much more slowly.
For analytics work, this matters because glucose is not a static value, and its “shape” changes under medication:
- The Delayed Curve: A “flat” glucose reading post-meal might not mean “low carb” or “not eating”; it may reflect a pharmacologic delay in absorption.
- Contextual Interpretation: A single reading has context: fasting or post-meal, medication timing, baseline insulin sensitivity, and weight-loss phase.
- Data Misinterpretation: If a model assumes a standard glucose-first response, it may misinterpret the suppressed glycaemic spikes of a high-dose GLP-1 patient as high adherence to a ketogenic diet rather than the effect of the drug.
2.2 2. Insulin follows the glucose signal
The pancreas releases insulin in response to rising glucose. Insulin helps move glucose from the bloodstream into muscle, liver, and other tissues.
A simple analogy is that insulin acts like a key that helps glucose enter cells. In reality, the biology is more complex, but the data principle is straightforward: insulin response is part of a system, not an isolated event.
For a data analyst, this is important because an outcome such as HbA1c or fasting glucose is only the visible surface. Behind it are multiple mechanisms:
- insulin secretion
- insulin sensitivity
- liver glucose output
- appetite and energy intake
- body weight
- medication persistence
- disease duration
If we model only one measurement, we risk missing the system behaviour.
2.3 3. The system returns toward baseline
In a healthy metabolic response, glucose and insulin return toward baseline after the meal. If this response becomes disrupted, glucose may stay elevated for longer, or the body may require more insulin to achieve the same glucose control.
This is where common metabolic problems appear:
- Insulin resistance: tissues respond less effectively to insulin, so the pancreas must produce more insulin to achieve the same effect.
- Insulin deficiency: the pancreas cannot produce enough insulin, so glucose remains elevated.
- Excess liver glucose output: the liver releases glucose when the body does not need it, contributing to high blood glucose.
- Weight-related metabolic strain: increased adiposity can worsen insulin resistance and cardiometabolic risk.
These are not just clinical concepts. They are also data-quality and model-design issues.
3 Where GLP-1 fits
GLP-1 is an incretin hormone released from the gut after eating. It helps coordinate the post-meal metabolic response.
However, thinking of GLP-1 as purely a “gut hormone” is a mistake for data analysis. Modern GLP-1 receptor agonists (GLP-1 RAs) act across a distributed system:
- The Pancreas: It increases glucose-dependent insulin release and reduces glucagon release.
- The Gut: It significantly slows gastric emptying, which flattens the post-ingestion glucose curve.
- The Brain: It acts on neuronal pathways in the hypothalamus and brainstem to regulate long-term energy balance. For many patients, this manifests as a reduction in “food noise”—the intrusive, constant thoughts about eating that drive calorie intake.
3.1 The GIP Distinction (Tirzepatide)
It is also critical for analysts to distinguish between pure GLP-1 agonists (like Semaglutide) and dual agonists like Tirzepatide (Mounjaro). Tirzepatide is a dual GIP (Glucose-dependent Insulinotropic Polypeptide) and GLP-1 receptor agonist.
GIP has distinct effects on adipose tissue (fat storage) and glucagon that pure GLP-1s do not. This distinction explains why Tirzepatide data often shows different weight loss trajectories, lipid profiles, and side effect patterns compared to Ozempic or Wegovy. If a data model lumps them together without a “molecule” or “class” flag, it will miss these metabolic nuances.
This is why GLP-1 medicines can affect multiple observable data streams at once.
They may influence:
- HbA1c
- weight
- prescribing patterns
- diabetes medication de-escalation
- cardiovascular risk markers
- gastrointestinal adverse events
- discontinuation and restart behaviour
- food volume and food category consumption
- pharmacy revenue and PBS/private-pay spend
- companion nutrition demand
The important point is that GLP-1 treatment is not a single-metric intervention. It changes a network.
4 Why this matters for GLP-1 analytics
When a category becomes commercially and clinically important, dashboards often appear before the data model is mature. GLP-1s are at risk of exactly that.
A simple utilisation chart may show prescription growth. But it may not explain whether that growth is driven by:
- diabetes treatment
- obesity treatment
- private prescribing
- PBS access
- supply recovery
- switching between products
- oral formulations
- telehealth acquisition
- pharmacy channel concentration
- continuation versus short-term trial use
4.1 The “Compounding” Shadow Market
In the Australian context, analysts must also account for a massive “grey market” that is often invisible to standard aggregators. Due to global shortages of TGA-approved Ozempic and Wegovy, significant volumes of GLP-1s are moving through compounding pharmacies.
This data does not appear in standard wholesale or PBS pharmacy reports. A “Metabolic Baseline” for analytics that ignores this shadow market will have a massive blind spot, undercounting the total number of patients on therapy and potentially miscalculating the market size and clinical impact.
Without the metabolic baseline, it is easy to over-read the signal.
For example, a decrease in HbA1c may reflect direct glycaemic improvement, but it may also be mediated by weight loss, reduced food intake, better adherence, or changes in other diabetes medicines. A fall in pharmacy sales after initiation may reflect discontinuation, dose interruption, supply shortage, or patients moving between private and subsidised channels.
The biology tells us what relationships to expect. The data model tells us whether we can observe them.
5 A practical analytics frame
For future GLP-1 work, I would separate the data model into five layers. Note for Australian analysts: Joining these layers is a significant challenge. Clinical “renal risk” or “BMI” data is often siloed in hospital EMRs, while “pharmacy spend” sits in claims data. A robust model requires bridging these siloes.
| Layer | What to capture | Why it matters |
|---|---|---|
| Patient baseline | BMI, diabetes status, HbA1c, CVD history, renal risk, age, sex, socioeconomic access | Defines eligibility, expected benefit, and equity questions |
| Treatment pathway | product, dose, start date, stop date, restart, switch, prescriber, pharmacy channel | Distinguishes adoption from persistence and identifies “micro-dosing” (spacing out doses) |
| Outcome measures | weight, HbA1c, lipids, blood pressure, adverse events, hospitalisation | Links utilisation to clinical value |
| Access and cost | PBS/private status, co-payment, stock availability, substitution, private price | Explains who can stay on therapy |
| Behaviour and support | nutrition support, protein/fibre supplementation, digital programs, follow-up cadence | Distinguishes adherence (taking dose on time) from long-term maintenance |
This frame is deliberately broader than a medicine table. GLP-1s behave like a cross-system intervention, so the data architecture needs to be cross-system as well.
6 What healthcare AI needs to get right
GLP-1s are also a good test case for healthcare AI.
An AI system summarising GLP-1 evidence, classifying patient cohorts, or monitoring market signals must avoid treating every mention of “Ozempic”, “Wegovy”, “Mounjaro”, “semaglutide”, or “tirzepatide” as equivalent.
The model needs context:
- indication: diabetes, obesity, cardiovascular risk reduction, or off-label use
- product and molecule
- dose and formulation
- country and regulator
- PBS/private access status
- persistence and discontinuation
- trial evidence versus real-world evidence
- clinical outcome versus commercial signal
This is why domain modelling comes before automation. If the ontology is weak, the AI summary will sound confident while mixing distinct concepts.
7 Closing thought
GLP-1s are becoming one of the most important healthcare data categories to watch over the next decade.
The opportunity is not only to count prescriptions or estimate market size. The real work is to connect biology, access, adherence, outcomes, cost, and behaviour into a data model that can support better decisions.
That starts with the simple post-meal curve: glucose rises, insulin responds, and the body tries to return to balance.
Everything else in GLP-1 analytics builds from there.