Tuberous sclerosis can be diagnosed very early – it is already possible to detect tumours in the foetus during pregnancy.

However, even in children diagnosed this early, treatment for epilepsy has typically been delayed until the first seizures occur.

After all, how can one treat a condition that hasn’t yet manifested?

Jóźwiak MD, PhD began treating differently

Four children (30%) showed no changes in their EEG and did not develop epilepsy. The remaining ten (70%) did. When changes appeared in the EEG, they were given antiepileptic drugs.

In four of these children, EEG readings normalized over time, and they exhibited no clinical symptoms of epilepsy. In six, seizures occurred but were easier to control.



Embarking on a clinical trial? As an academic, you recognize the intricacies involved.

Before we go deeper into the risks and solutions in clinical data analytics, let’s explore how we can partner on your next project.

Potential Risks: Misguided assumptions about the effects, Budget constraints, Misalignment with research goals, and more.

Our Approach: Together, we’ll leverage diverse strategies to mitigate these risks:

1. Real-World Data/Real-World Evidence (RWD/RWE): By analyzing data collected from actual patient experiences and clinical practices (RWD), and the evidence derived from this data (RWE), we can validate assumptions about effects, ensuring that the research is grounded in reality.
2. Meta-Analyses: Aggregating results from multiple studies provides a broader perspective, reducing the risk of misguided assumptions and aligning the research with existing evidence. This method combines the findings from various independent studies to identify common trends and insights.
3. Qualitative Studies: Engaging with qualitative insights, such as interviews and focus groups, allows for a deeper understanding of the context, ensuring that the research goals are relevant and achievable.

Potential Risks: Bad study design, Inappropriate study size, Methodological inconsistencies, etc.

Our Approach: Together, we’ll employ targeted services to address these challenges:

1. Sample Size Estimation (with Clinical Trial Simulations): By carefully estimating the study size and utilizing simulations, we’ll ensure that the design is statistically sound and powered to detect meaningful effects, reducing the risk of inappropriate study size.
2. Statistical Analysis Plan Development: Crafting a comprehensive statistical analysis plan will guide the methodology, minimizing inconsistencies and aligning with best practices.
3. Design Optimization: Working closely with you, we’ll review and refine the study design, ensuring that it aligns with your research goals and adheres to academic standards. Utilizing computational methods like optimization, we can compare different options, including adaptive designs, which allow for flexibility and adjustments during the trial, to find the best fit for your project.

Potential Risks: Human error, Data inconsistency, Unforeseen obstacles, and more.

Our Approach:: Our innovative and adaptive approach includes:

1. Utilizing ML-Based Classifiers: By employing machine learning classifiers, we’ll create algorithms that can detect and address challenges early in the data collection process. These classifiers can identify patterns and anomalies, often reducing their severity, ensuring quality and integrity in data collection. We’ll continuously monitor the study, ready to respond to any challenges that arise, ensuring that the project stays on track and aligned with your goals.
2. Leveraging Previously Created Tools (Simulators, Optimizers): Should the unexpected occur, we’ll draw on our collection of specialized tools, including simulators and optimizers, to make informed decisions on how to proceed. These tools have been developed to handle various scenarios in clinical trials, optimizing the study’s progress.

Potential Risks: Lack of documentation, Analytical errors, Misinterpretation, and beyond.

Our Approach: Our comprehensive and transparent approach includes:

1. Crafting Reproducible Reports and Sharing the Code: By creating clear and reproducible documentation, we’ll ensure that the analysis is transparent and verifiable, reducing the risk of misinterpretation and aligning with academic standards.
2. Employing Automated Data Tests: Utilizing automated tests, we’ll minimize analytical errors, ensuring that the data is accurate and consistent.
   
3. Collaborative Interpretation: Working closely with you, we’ll interpret the results, ensuring that they align with your research goals and adhere to the highest standards of academic rigor.

Our Approach: Our innovative and collaborative approach includes utilizing Explainable AI (XAI):


By employing XAI, a branch of artificial intelligence that provides clear and understandable explanations for complex AI-driven findings, we’ll demystify complex patterns. This approach allows us to uncover insights that might otherwise be missed, providing clear explanations that align with academic rigor. It ensures that the AI’s decision-making process can be transparently understood, aligning with the principles of scientific inquiry.

Our Approach: Our comprehensive and transparent approach includes:

1. Understanding Outcomes: We’ll work with you to interpret and articulate the results, ensuring that the findings are presented with clarity and academic rigor.
2. Preparing the Paper: Collaborating closely, we’ll assist in crafting the manuscript, aligning it with publication standards, and enhancing its potential for acceptance in high-impact journals.
3. Assisting with Reviewer Doubts: Should any questions or doubts arise from reviewers, we’ll be at your side, providing expert guidance and support in addressing them, ensuring a smooth publication process.

Without proper tools, it’s hard to consider different scenarios and hypotheses, especially with limited time and budget.

In the Epistop project, we faced the challenge of drug-resistant epilepsy, for which there is no one definition.

Some scientists defined it as not responding to drugs after 30 days; others said 45 days is a better measure. Instead of just “running a calculation” we prepared infrastructure for generating reproducible reports where changing one parameter recalculates everything within seconds and without the risk of error.

Even the smartest people are prone to human error and one undetected error can question the entire study. So… We made the trial more credible with software that helped to indicate mistakes and suspicious data values.

To understand the data and effectively answer questions at the end of a study, which can take years, we introduced versioning of generated reports in the Git repository. This was also where we conducted all project-related communication. Scientists can easily trace all report versions and add their comments. Months after generating a report, recalculations may be needed due to reviewers’ questions or subsequent projects. Changes to external tools and libraries can lead to variations in outcomes. We used Packrat – a tool for R code, which saves the calculation ecosystem for a given day, allowing recalculations to be performed in the same environment.