sanaitio

Building trust
in healthcare AI.

Trust by Design  ·  AI medical device consulting

sanaitio helps medtech companies develop, validate, and deploy AI-based medical devices that earn clinical adoption and regulatory approval.

Get in touch ⟶
01 / Problem
The adoption gap

Approved,
but not used.

The market is large. The clinical impact is not. Most approved AI medical devices never reach routine use, and most lack the evidence to support it.

950+
FDA-approved AI medical devices
$70B
Estimated global market
<1%
Studies reporting patient outcomes
95%
FDA approvals rely on regulatory equivalence
95%
Studies missing demographic data to assess bias

Four recurring obstacles

Lack of trust

The top adoption barrier across three audiences: 28% of healthcare professionals, 50% of hospital representatives, and 59% of AI developers.

Regulatory complexity

The AI Act, MDR/IVDR, and GDPR overlap. Most companies lack the in-house capacity to navigate all three at once.

Inadequate strategies

Without an early, structured development plan, projects accumulate costly delays and produce devices that work technically but fail in clinical practice.

Data and algorithmic risks

Biased training data, model drift, and non-representative validation cohorts erode real-world performance and create regulatory and patient-safety exposure.

Sources: European Commission (2025); Joshi et al. (2025); Muralidharan et al. (2024); Liu et al. (2025).

02 / Trust by Design
Our approach

Start with
the end in mind.

Trust by Design adapts Quality by Design (ICH Q8/Q9/Q10), the framework long established in pharmaceutical CMC, to AI-based medical devices. The core idea translates directly: trust must be built in by design, not tested in at the end.

Every phase is informed by a predefined clinical target, a structured risk assessment, and a control strategy that spans the full product lifecycle.

03 / Lifecycle
Total product lifecycle

Seven phases,
one strategy.

From product definition to post-market surveillance. Each phase contributes evidence to the next; the control strategy carries through.

01

Objectives

  • Target Product Profile
  • Users, workflow, population
  • Risk, safety, class
02

Causality graphs

  • Causal DAG construction
  • Variables and confounders
  • Bayesian network
03

Data strategy

  • Data quality assurance
  • Optimal sampling design
  • Representative cohorts
04

Model training & validation

  • Causal AI methods
  • Stress testing
  • Robustness
05

Clinical validation

  • Protocol with prior information
  • Adaptive Bayesian designs
  • Real-world evidence
06

Regulatory dossier

  • Strategy justification
  • Performance evidence
  • Interpretability
07

Deployment & monitoring

  • Local calibration
  • Drift detection
  • Quality control
04 / Methodology
From causal graphs to clinical decisions

Optimize the whole
development path.

Our methodology optimizes the entire development pathway, not just clinical validation. The objective is real adoption: a device used by physicians, funded by hospitals, and trusted by patients. Every modeling, data, and design choice is evaluated against that endpoint.

At the heart of the approach sits a causal graph: a structured map of how patients, workflow, decision thresholds, environment, and bias jointly determine the clinical outcome. Making these relationships explicit is what tells us, in every phase, which data to collect, which studies to run, and which design choices actually move the result.

01

Start with structure

We map the workflow as a causal graph: every actor, process, and variable that influences the outcome. The graph makes assumptions explicit and surfaces:

  • Confounders affecting performance
  • Variables required for valid study design
  • Hidden bias pathways
02

Add quantification

The causal graph becomes a Bayesian network by attaching probability distributions to each node, drawn from data, literature, or expert input. This enables:

  • Simulation of design and threshold choices
  • Explicit uncertainty quantification
  • Optimal validation protocols
03

Decide under uncertainty

The Bayesian network informs development decisions before resources are committed. Each option is evaluated under the full uncertainty of the model:

  • Data collection priorities and cohort selection
  • Validation strategy proportionate to risk
  • Trial design: sample size, adaptive elements, endpoints
  • Trade-offs across scientific, regulatory, and operational constraints
05 / Services
What we deliver

How we help.

01

Clinical validation strategy

Design and execute clinical validation using causal modeling. From workflow mapping through Bayesian protocol design to regulatory documentation (AI Act, MDR/IVDR).

02

Upstream development strategy

Before you build the model, build the map. The Bayesian network as a prospective simulation tool surfaces success factors, adoption risks, data needs, and regulatory constraints at the earliest design stage.

03

Regulatory navigation

Translate the AI Act, MDR/IVDR, and GDPR into concrete technical and clinical actions: implementation guidance tailored to your device, not generic checklists.

04

Deployment and monitoring

Trust does not end at certification. We design the post-market plan: user training, local calibration, drift detection, quality control, and feedback loops that keep the device safe and effective in routine use.

06 / Why choose us
  1. 01

    Trust by Design,
    end to end.

    Our approach aligns project goals with compliance and adoption from day one, shortening time to market.

  2. 02

    Bayesian
    methodology.

    Uncertainty quantified, not hidden. A prerequisite for clinical and financial trust.

  3. 03

    Pharmaceutical
    foundations.

    Decades of practice qualifying drugs, processes, and measurement systems. The same rigor, applied to AI.

  4. 04

    One offering,
    four disciplines.

    AI, advanced statistics, causal inference, and regulatory science delivered as a single, integrated service.

  5. 05

    Concrete
    implementation.

    Tailored guidance for your device, not generic compliance checklists.

07 / Team
Founders & team

Behind sanaitio.

Bruno Boulanger

Bruno Boulanger

Co-founder & Scientific Advisor

30+ years in applied statistics for the pharmaceutical and medical device industries (Eli Lilly, UCB). Founder of Arlenda (2003). USP Committee on Statistics, 2010-2024. 130+ peer-reviewed publications.

LinkedIn
Nils Boulanger

Nils Boulanger

Co-founder & CTO

Data scientist and AI engineer. Designs and validates deep-learning systems for medical imaging deployed in real clinical workflows. Leads technical implementation at Sanaitio.

LinkedIn
Gaëlle Martin

Gaëlle Martin

Head of QA

20+ years in QA leadership across biotech, pharma, and consultancy. Designs QMS aligned with GMP, GCP, GAMP5, ISO 9001, and ISO 13485. Quality oversight for computerized system validation and GxP projects.

LinkedIn

Bruno and Nils also lead Trilenda, sister company specializing in CMC statistics and software for biopharma.

08 / Contact
Get in touch

From algorithm
to adoption.

A device in development, a validation path to plan, or just want to explore where Trust by Design fits your roadmap. Send a note.