A reusable predictor approach

Cassis develops predictive decision-support tools for difficult clinical and healthcare problems. The clinical work focuses on a practical question:

Can we identify when an individual patient is becoming more vulnerable before harm, deterioration or diagnostic failure becomes clinically obvious?

Current healthcare systems are often good at recording events after they occur. They are less effective at predicting what is likely to happen next for a specific patient. Cassis addresses that gap by developing patient-specific predictors for defined clinical decisions, including deterioration, medication-related harm, falls, diagnostic error, rehabilitation and heat-related risk. The underlying capability is robust developmental methodology for turning complex clinical uncertainty into a defined prediction problem, with a a usable decision support resource for clinicians.

The commercial problem

Preventable harm remains a major cost to healthcare systems, pharmaceutical companies, providers and patients. The problem is rarely a complete absence of information. More often, relevant information is:

• distributed across different sources;
• recognised too late;
• interpreted inconsistently;
• assessed against population averages rather than the patient’s own baseline;
• not converted into a forward-looking estimate of risk.

The objective is to estimate whether risk is changing within a defined future window, commonly the next 24 to 48 hours, while there is still time to review treatment, increase monitoring or intervene.

Why Cassis is different

Cassis begins with the decision, not the machine learning / AI technology. Each predictor is developed by defining:

1. the clinical problem and decision that needs to be improved;
2. the event or outcome that needs to be predicted;
3. the patient and clinical information available;
4. the desired clinical response;
5. the consequences of false reassurance and medical error.

This supports an important methodological shift comprising:

• individual prediction to replace population risk models;
• clinical foresight to replace retrospective detection;
• patient trajectory to capture continuity of data over time, not just isolated measurements;;
• generic alerts to decision-focused recommendations, with quantified risk banding to quantify uncertainty, not generic alerts.

The aim is to improve judgement under uncertainty.

A reusable predictor architecture

Although each application addresses a different clinical problem, the predictor architecture follows a common structure.A specific predictor estimates whether the patient is moving towards a defined adverse outcome or clinically important state.

Cassis has developed a portfolio of working predictor specifications, interactive simulations and clinical decision models, including:

• defined clinical use cases;
• model logic and mathematical specifications;
• interactive HTML simulators;
• synthetic patient cohorts;
• patient-specific baseline methods;
• clinician-facing decision outputs;
• risk visualisation;
• before-and-after intervention modelling;
• documentation for clinical, technical and commercial review..

The economic value lies in the reusable method. Once the core predictor architecture, data standards, patient-baseline logic, explainability and decision interface are established, new applications can be developed more efficiently. The process remains clinically specific, but the underlying system can be reused across different problems.

Investor proposition

Cassis is seeking commercialisation partners who can evaluate and commercialise one or more predictor applications. The value lies in:

• a repeatable development method;
• a portfolio of tested problem definitions;
• working simulations;
• clinically interpretable modelling;
• applications across several high-cost areas;
• multiple routes to licensing and partnership;
• a focus on preventable harm and decision quality;
• the potential to convert one validated use case into a broader platform.

Investment would support:

• selection and development of the lead commercial application;
• independent clinical and technical review;
• access to suitable datasets;
• software engineering;
• validation;
• regulatory planning;
• commercial development;
• partner acquisition.

For more information

Cassis welcomes enquiries from organisations interested in evaluating, licensing, funding or co-developing patient-specific predictive decision-support applications.

To request a private demonstration or discuss a feasibility study, please contact mike_tremblay@skythunder.net .