Clinical decision support · medication safety
Example cases · not real patient data · not for clinical decisions
Problem
Risk does not come only from two drugs interacting. The number of drugs, kidney and liver function, and the total burden on a single organ have to be assessed together.
In a patient on five or more drugs, the number of possible interactions grows quickly. Screening them by hand on every prescription is not realistic.
A prescription that is safe at an eGFR of 90 may need a dose adjustment at 30. The check has to look at the current lab value.
Drugs that are acceptable on their own can add up to a risk in the same organ. An interaction check that looks at two drugs at a time does not see this.
Alerts that ignore context fire so often that physicians learn to override them all. The alert that matters gets lost in the noise.
Pairwise combinations
For n drugs, the number of pairwise combinations is n(n−1)/2. Three-way and more complex effects are not included.
| Drugs | Pairs |
|---|---|
| 5 | 10 |
| 8 | 28 |
| 10 | 45 |
| 15 | 105 |
The World Health Organization estimates the global annual cost of medication errors at about USD 42 billion. WHO, Medication Without Harm (2017)
How it works
Data comes in from standard resources, is assessed by rules, and the result is shown to the physician with its reason. No step uses probabilistic prediction.
Patient, medication order, laboratory, diagnosis and allergy data arrive as HL7 FHIR R4 resources.
Active substance, dose, unit and route are mapped to shared codes. Missing information is flagged, not skipped.
The prescription is checked against versioned rules across 18 safety planes. The same input always gives the same result.
A finding is shown with its severity: which rule, which value, which threshold.
The physician decides. Accepting the alert, changing the order or overriding it with a reason is recorded.
Input, rule version, finding and the physician's decision are written to an append-only log.
18 safety planes
Each check has its own version and tests. Thresholds can be set to match the institution's policy.
Hospital system integration
The four questions hospital IT and biomedical engineering teams ask first: how it connects, how clients are authenticated, where it runs and what it logs.
Clinical data arrives over HL7 FHIR R4. HL7 v2 messages from systems that have not moved to FHIR are converted to FHIR R4.
Every request comes from an authenticated client and can reach only the data its role allows.
The system runs on the institution's own infrastructure; patient data does not leave the institution.
Application logs go to the institution's central log system; clinical decisions are also kept in a log that cannot be changed.
The matrix describes the integration architecture. Which interface is set up at your institution, and how, is agreed with your IT team during technical discovery before the pilot.
Verification
Every rule is linked to a clinical requirement and to the tests that verify it. A change that fails its tests is not released.
Regulation
PharmaDeux provides information used in drug therapy decisions, so it falls under Rule 11 of the EU Medical Device Regulation (MDR). It does not have a CE mark yet.
Intended use (draft)
To support the decision of a qualified healthcare professional by showing possible safety risks of a prescription, with their reasons. The system does not diagnose, does not recommend treatment and does not replace the physician's decision.
| Framework | Scope | Status |
|---|---|---|
EU MDR 2017/745 Medical Device Regulation · Annex VIII Rule 11 | Medical Device Regulation · Annex VIII Rule 11 | Class IIa target |
IEC 62304 Medical device software life cycle | Medical device software life cycle | Architecture aligned |
ISO 14971 Risk management for medical devices | Risk management for medical devices | Architecture aligned |
ISO 13485 Quality management system | Quality management system | On the roadmap |
IEC 62366-1 Usability engineering | Usability engineering | On the roadmap |
PharmaDeux is not on the market yet. For details, see Quality and MDR.
Clinical validation plan
PharmaDeux's performance will be measured on anonymised real cases against an independent panel of experts.
Primary and secondary endpoints and the data management plan are written, and the ethics application is submitted.
Anonymised past prescription and laboratory data are obtained from the partner institution in line with KVKK.
An independent panel of a clinical pharmacist and physicians, blind to the system, assesses the same cases.
The system's findings are compared with the panel's decisions. Sensitivity, specificity, positive predictive value and alert count are reported.
Results feed into the MDR clinical evaluation file and the design of a prospective pilot.
Partner institutions can take part in writing the protocol. Data is used only under ethics committee approval and a data processing agreement signed with the institution.
Pilot programme
To join a validation study on past cases, or to set up a pilot at your own institution, we can start with a non-binding Letter of Intent (LOI).