The Cerdes research solution for obesity
Obesity,
answered.
The right question. The right minds. The right data.
A first-in-class data platform and a unique, expert-approved network, built to work as one. Bring us your hardest question about obesity and related metabolic conditions. In about 14 days you have an answer that our algorithms have tested against the data and our experts have qualified and approved.
~14 days
to your first Answer Pack
92%
faster than a traditional advisory process
60–75%
lower cost
Why obesity, why now
The fastest-growing area in medicine is also the poorest measured.
We have chosen to focus primarily on obesity and the metabolic conditions because this is the therapeutic are in which we have in-depth knowledge and experience – and where we are confident that our subject matter expertise will provide the highest valuable. Most of the current questions that matter cannot be answered by trials alone.
10.9%
Obesity may be sparsely captured in structured data. Only about one in nine Danish hospital patients with a recorded BMI of 25 or more carries an overweight or obesity diagnosis code. Similarly, only about one in four Swedish adults with a measured BMI of 30 or more had an obesity diagnosis on record when that BMI was first taken, and fewer than half had one five years later. Every burden estimate, payer dossier and comparative study rests on a case definition that someone has to validate first.
Gribsholt SB et al. Validity of ICD-10 diagnoses of overweight and obesity in Danish hospitals. Clin Epidemiol 2019;11:845–854.
Ritsinger V, Sundbom M, Dereke J, et al. Underdiagnosed and undertreated: obesity and its cardiometabolic burden in Swedish clinical practice. Insights from the AROS database. Diabetes Obes Metab 2026;28(8):6787–6800.
Questions
The pipeline is changing the questions
The race has moved beyond who loses the most weight. The questions now are about which patient, which dose and for how long: dose optimisation, maintenance dosing, combinations and tolerability in everyday care.
Definition
Obesity has been redefined
The 2025 Lancet Commission separated clinical from pre-clinical obesity, in terms that almost no routine dataset can apply directly. Defining who has obesity is now a research problem in its own right.
Evidence
The rules of evidence are changing
EU Joint Clinical Assessments began in 2025. The European Health Data Space applies in stages from 2027, with secondary use of health data from 2029. Regulators now ask whether a piece of real-world evidence is defensible.
Time
Every day of delay has a price
A day of delay in drug development costs $500,000–800,000 on average, and up to $1.4M in cardiometabolic areas. Saving one week preserves $3.5–9.8M.
What makes us different
Three unique capabilities. None works alone.
Others offer data tools, or they offer experts. Cerdes brings together a platform, an algorithm and an expert network, designed from the start to work together on every question.
The platform
Evidence, found and synthesised in days.
What it is. Our structured method for searching, appraising and summarising the existing research and data on a question. AI does the high-volume reading; the method decides what counts.
Why it matters. Most of an answer already exists, scattered across thousands of papers, registries and trial records. Finding and weighing it by hand takes months and is rarely repeatable.
Why it’s unique. Built for obesity alone, with the field’s landscape and definitions mapped in advance. Every AI step passes rigour gates and leaves a trail back to its source, and each answer strengthens the next.
The algorithm
Lumawarp: machine learning on a new mathematical foundation.
What it is. Lumawarp, the machine-learning engine from Lucidity Sciences, applied to obesity data.
Why it matters. Obesity data are high-dimensional, sparse and messy. Conventional methods are slow on data like this, or lose the signal.
Why it’s unique. A new geometric approach instead of more data and more compute: 382× faster than AutoML, with models over 1,000× smaller. It is light enough to run on a laptop behind your firewall, so analyses can be federated without patient data ever leaving its original location.
The expert network
Experts who qualify, control and approve.
What it is. A hand-picked core group of obesity specialists, with peer-to-peer reach into the world’s leading experts on any question.
Why it matters. Clinical, mechanistic and practical judgement is what a model cannot supply. No Answer Pack leaves until a named expert has approved it.
Why it’s unique. Real experts, not simulated ones. They do not just give an opinion on a call: they shape the question, check the work at every step and sign off, and their names stand behind the answer.
Powered by Lumawarp, from Lucidity Sciences
382×
faster than AutoML
>1,000×
smaller models
- Complex, high-dimensional analysis · Built for it
- Flexible, secure deployment · On-premise, behind your firewall
- Accelerated model development · Faster builds
Figures published by Lucidity Sciences. Details available on request.
Answers that are both computed and qualified.
The platform gathers what the literature and the evidence say; the algorithm finds what the data reveal. Together they brief the experts, so their time goes on judgement. The experts direct and check both, and their input is fed back in. Each makes the others better with every question.
“One cannot work without the other.”
The founding conviction behind Cerdes
The offering
One question in. One answer out.
You buy answers, not hours, panels or datasets. Each Cerdes Question is one well-framed obesity question, worked through our full engine and returned as a decision-grade Answer Pack, in the same four parts every time.
An example Cerdes Question
“Which patients are most likely to lose weight on GLP-1s, and who will not?”
Your question → Data platform × expert network → Answer Pack
01
What the field knows
A structured landscape of the literature and guidelines: what has been tested, what has not, and where the gaps are.
02
What the experts judge
Targeted input from the right clinicians and scientists, synthesized and approved by a named expert. Not a meeting transcript.
03
What the data can prove
Whether fit-for-purpose data exist, how far they can be trusted, and which obesity definition holds up in them.
04
What to do next
A clear recommendation: the decision, the study design or the evidence plan, with costs where a study is needed.
Every answer makes the next one faster. With your consent, each Answer Pack is captured in your Cerdes knowledge repository, so your next question starts from what is already known.
How we think about time
Most of the time is dead time.
When we line up a traditional process against ours at the same scale, the difference is not working harder. It is removing the white space and running in parallel what used to run in sequence.
A closer look: Ad Board In-A-Box
Same advisory objective, delivered through a different operating model. Traditional boards remain valuable for live debate and relationships. For most questions, this is faster, leaner and more decision-oriented.
Traditional advisory board
- Turnaround · 3–6 months
- Board size · 8–15 advisors
- Cost · €50k to €300k+
- Admin burden · Vendors, calendars, travel, venue
- Output · Varies with meeting dynamics
Ad Board In-A-Box
- Turnaround · ~2 weeks
- Expert input · Async-first, targeted capture
- Cost · A fraction: no travel, venue or AV
- Admin burden · Minimal, with one owner
- Output · Structured, tailored to your decision
COVID-19 vaccines were fast because the science existed before the question. Fifty years of mRNA research removed the discovery phase. We apply the same lesson to obesity: the data landscape, the validated phenotypes and the expert network are mapped before your question arrives, so the work starts on day one.
Illustrative comparisons based on advisory-board planning dynamics and published drug-development timelines. Sources: NIH/NIAID, FDA, NEJM, BIO Clinical Development Success Rates, Operation Warp Speed (PMC8514901).
How the work is done
The right tool for each job, checked by people who know the field.
What the engine makes possible
Prediction and forecasting
Obesity-specific models of population and individual risk, treatment response and impact.
Treatment-effect estimation
Machine learning combined with causal methods to remove confounding from real-world data.
Intervention simulation
Cost-sensitive exploration of which strategy works best, for whom.
Explainability and causal discovery
Advanced statistical and mathematical techniques to evaluate and refine causal models.
The right tool for each job
LLMs
Read at scale
Expert-guided extraction from unstructured literature and records, at a scale no review team can match.
Agentic logic
Orchestrate with rules
A deterministic, obesity-specific framework that makes the tools work together, through a plain-language interface.
ML and analytics
Find what the data reveal
Machine learning and causal methods that synthesise, evaluate and apply patterns from real-world data.
Real experts, not simulated ones.
Every evaluation, comment and recommendation in an Answer Pack comes from a named specialist who has read the work.
Six steps, shown on one question
Which patients are most likely to lose weight on GLP-1s?
[1] Landscape
What has already been established or explored within the field?
Rigorous pipelines for structured review of abstracts and full-text articles, using LLMs to extract the data. Each question type drives its own extraction schema.
[2] Assessment
What evidence has the current research programme established, and where are the gaps?
An evidence review benchmarked against research standards and clinical guidelines.
[3] Data
What secondary data can support the evidence, and is primary data collection needed?
Real-world data source identification and a feasibility assessment framework, including the obesity definition that holds up in each source.
[4] Expertise
Which experts can help us understand the current clinical picture and how the data were generated?
A peer-to-peer expert network: the right person answering the right question at the right time. Key opinion leaders are briefed in a structured way, so their time goes on judgement.
[5] Design
What are the best design choices for this study?
Guided study design supported by expert insight, real-world data, and regulatory and statistical templates. Protocols come back review-ready and annotated.
[6] Analysis
What targeted answers do the data reliably reveal?
Prediction and forecasting, treatment-effect estimation, cost-sensitive intervention simulation, and explainability and causal checks, with secure on-premise deployment options.
The steps are not a fixed sequence. Each question leans on the steps it needs.
Naming the gap honestly
“Lipids were routinely measured but not tested.“
A landscape cannot predict who responds, because the predictor signals are rarely published in a usable form. What it can do is show what the field has proposed, what it has tested, and what it has left untouched. That map of gaps is often the most valuable part of the answer.
From a Cerdes Landscape of candidate predictors of weight loss on GLP-1 therapy.
Rigour by design
Fast is only useful if it is right.
Language models are fluent, and fluency is not accuracy. Every high-throughput AI task on the Cerdes platform passes through checks a reviewer would recognize, and every number traces back to its source.
Inter-rater gates
A statistically meaningful sample of every AI task is reviewed by people. Agreement must clear a threshold before work proceeds.
Planted known answers
Known answers are seeded into the work. More than 5% disagreement triggers a review of definitions.
Provenance
Every extracted value carries an audit trail back to the source.
Multiple models and runs
Independent passes with different models expose non-deterministic output and measure cross-model reliability.
Sensitivity analyses
Abstract against full text, and variations in parameters and definitions.
Red team
Adversarial review pressure-tests results as a matter of course.
Expert review
Clinicians and scientists add the clinical, mechanistic and practical interpretation. AI accelerates the work, and people sign it off.
Questions we answer
Different questions. One engine.
Each format below is a type of Cerdes Question. They run through the same engine and change only which steps carry the weight.
| Format | You ask | Your Answer Pack adds | Typical time |
|---|---|---|---|
| Landscape Start here | “What does the field know, and where are the gaps?” | Gap map of tested and untested variables. | 1–2 weeks |
| Ad Board In-A-Box | “What do the best clinicians think?” | Structured expert positions, with consensus and dissent. | ~2 weeks |
| Scientific Advice In-A-Box | “Will the regulator or HTA body accept our plan?” | Precedent analysis, mock panel, briefing draft and EU JCA/PICO readiness. | 3–6 weeks |
| Data Feasibility | “Is there data that can answer this, and can we trust it?” | Real-world data sources identified and assessed, with an obesity definition that holds up in them. | 2–4 weeks |
| Study Design & Protocol | “How should we design the study?” | A review-ready, annotated protocol, including RFP support. | 1–2 weeks |
| Advanced Analysis | “What do the data actually reveal?” | Results from the Cerdes engine: prediction, treatment effect and simulation, with explainability and causal checks. | 8–12 weeks |
The expert network
Every answer is qualified, controlled and approved by real human experts.
This is the part no one else offers. Our experts do not just give an opinion on a call. They shape the question, check the work at every step and sign off the answer. Their names stand behind it.
Before the work
Qualify
Experts validate and sharpen every question, so the platform answers the right one, not just the one that was asked.
Day 1
During the work
Control
Experts check the evidence, the data and the analysis, and add the clinical, mechanistic and practical interpretation a model cannot.
Every step
After the work
Approve
No Answer Pack leaves Cerdes until a named expert has reviewed and approved it.
Before delivery
Every Answer Pack carries its approval trail
Which patients are most likely to lose weight on GLP-1-based therapy?
Expert approved · TheExpertGroup
Qualified
Core-group lead
Question framed, scope agreed, the right specialists named.
Controlled
Specialists + core group
Landscape, data and analysis checked against clinical reality.
Approved
Named expert sign-off
Synthesis approved and the expert input retained for next time.
How an expert engagement runs
Step 1
Need identification
The client brief is captured and the question qualified.
Step 2
Early engagement
The core group refines the agenda and the questions experts must answer.
Step 3
Identification
The exact specialists are found peer-to-peer, backed by a literature search.
Step 4
Preparation
Experts are onboarded and briefed with what the platform has found.
Step 5
Full engagement
Input captured live, in writing, by chat or one-to-one, and synthesised question by question.
Step 6
Wrap-up
Synthesis approved, and every input retained and fed back into the platform.
TheExpertGroup
About 20 hand-picked, interdisciplinary experts in obesity, metabolism, real-world evidence, epidemiology, regulatory science and AI. They own the quality of every answer.
Global Expert Network
The specialists each question needs, such as a bariatric surgeon or a paediatric endocrinologist, reached peer-to-peer by the core group. No static roster.
Why peer-to-peer matters: a leading clinician answers a colleague’s call. Through the core group, Cerdes found a world-leading specialist for an orphan-drug RFP in 5 hours.
Why Cerdes
Others have data tools or in-house experts. We have the data tools and the real clinical experts (those who are engaging with, and treating the patients), working as one.
| Consultancies | Expert networks | CROs / RWE vendors | AI-only tools | Cerdes | |
|---|---|---|---|---|---|
| Speed | Weeks to months | 1–4 weeks | Months | Hours, unvalidated | Days to weeks, validated |
| Expert quality | High, expensive | Variable, one-to-one | Hired in | None | Curated, interdisciplinary, peer-reached |
| Analytics engine | Standard tools | None | Standard statistics | Generic models | First-in-class AI/ML engine |
| Expert approval | Consultant review | Opinion only | Internal review | None | Qualified, controlled and approved by named experts |
| Quality assurance | Ad hoc | None | Standard SOPs | None | Rigour gates, red team, provenance |
| Knowledge retained | Leaves with the team | Lost after the call | Per project | Partial | Compounding repository |
Machine
A first-in-class engine that makes new kinds of research possible.
Minds
Experts who qualify, control and approve, reached peer-to-peer within hours.
Together
Designed as one system, so each half makes the other better.
Rigour
Every number traces back to its source.
Memory
Every answer enriches the repository, while others start from zero.
Working with us
Two ways in. One engine behind them.
One question
Cerdes Question
Any format from the question menu, run on the data platform, approved by the expert network and returned as a four-part Answer Pack. Studies are scoped separately.
- What the field knows
- What the experts judge
- What the data can prove
- What to do next
From ~2 weeks · priced per question
For programmes
Cerdes Membership
An annual quota of Cerdes Questions for teams with a continuous obesity evidence agenda.
- Priority response
- A dedicated core-group lead
- Your own searchable knowledge repository
- The Obesity Signal briefing
Annual · tiered by question volume
We price against the value of time, not consulting hours. In a cardiometabolic programme, one week saved is worth more than a year of membership.
Independence and integrity
Our credibility is the product.
Scientists sign off
AI accelerates the work. Named experts approve every answer before it is delivered.
Evidence first
Commissioned work buys an honest answer, never a preferred conclusion. We name the gaps, too.
Disclosed and fair
Experts are paid fair-market value, and every engagement is disclosed.
Your data, your consent
Knowledge enters the repository only with your consent, under GDPR-compliant governance.
Let’s start
Bring us your hardest obesity question.
Tell us the decision you need to make. We will come back with how Cerdes would answer it, which format fits, and how long it will take.
The Obesity Signal
A short, regular briefing that turns obesity pipeline news into the evidence questions it raises.
Sources Cost of delay: Tufts CSDD, 2024. Study start-up time: Veeva Systems, 2020. Diagnosis coding: Gribsholt SB et al., Clin Epidemiol 2019;11:845–854. Clinical obesity: Lancet Diabetes & Endocrinology Commission, 2025. European Health Data Space: Regulation (EU) 2025/327. Vaccine timeline: NIH/NIAID, FDA, NEJM, Operation Warp Speed (PMC8514901).