Cohort definitions that drift
Two reports on the same pathway include different patients because inclusion rules live in SQL nobody has read for two years. Clinical confidence collapses the first time the difference is noticed in a meeting.
Healthcare
Operational and clinical reporting with stable cohort definitions and pseudonymisation by design.
Healthcare analytics carries two constraints at once: the reporting has to be defensible to a clinical audience, and the underlying data is among the most sensitive a person has. Those constraints point the same way, towards strict definitions, strong lineage and access that can be evidenced.
- Cohort and pathway definitions agreed once and reused everywhere
- Identifiable data minimised in analytics environments by design
- Operational reporting fast enough to manage the service, not just review it
- Audit trail available without a special project
We build operational and clinical reporting on that basis. Pathway and waiting-time analysis with cohort rules written down, capacity and flow reporting that reflects how the service actually runs, and commercial and trial reporting for life sciences organisations where the audit trail is part of the deliverable.
Our scope is the data and reporting layer: acute and community providers, private healthcare groups, pharmaceutical and biotech commercial and manufacturing functions, medical devices, and health insurers. Clinical interpretation stays with clinicians; we make sure the numbers they interpret are right.
What goes wrong
Recurring failure modes in healthcare
Pseudonymisation applied too late
Identifiable data copied into analytics environments because it was easier, then relied upon. Retrofitting pseudonymisation after reports exist is far harder than designing it in.
Operational data that lags the decision
Capacity and flow reporting produced daily when the decision is made hourly, so the service manages on whiteboards and the dashboard becomes retrospective reporting.
Commercial and medical data mixed inappropriately
In life sciences, commercial analytics and medical or safety data require different controls. Combining them in one workspace creates compliance exposure that is hard to unwind.
Metrics
Metrics and their definition errors
Each metric with its definition and the error that most often distorts it. If our reading is wrong for your business, that disagreement is a productive starting point.
| Metric | Definition | The trap |
|---|---|---|
| Referral-to-treatment waiting time | Elapsed time from referral receipt to first definitive treatment, by pathway. | Clock pauses, re-referrals and patient-initiated delays must be modelled explicitly, or the distribution is wrong at exactly the tail that matters. |
| Bed occupancy | Occupied bed days over available bed days in the period. | Escalation and closed beds need a clear rule, and midnight census versus continuous measurement produce materially different figures. |
| Did-not-attend rate | Appointments where the patient did not attend, over appointments offered. | Cancellations by the provider must be excluded from the denominator, and short-notice cancellations handled as their own category. |
| Length of stay | Discharge date-time less admission date-time, by specialty and spell. | Spell versus episode grain is the usual error; transfers between specialties double count if the spell is not constructed properly. |
| Trial enrolment rate | Participants enrolled per site per month against target. | Screen failures and withdrawals need separate measures; folding them into one number hides the actual bottleneck. |
Use cases
What we typically build
Operational flow and capacity
Admissions, discharges, transfers, theatre utilisation and bed state at a frequency that matches operational decisions, with clear data-currency indicators.
Pathway and waiting-list analytics
Cohort rules implemented once in the semantic model with a documented definition per pathway, plus validation against the operational source of record.
Workforce and rota analytics
Establishment versus actual, agency and bank usage, absence patterns and cost per shift, linked to activity so productivity is examined fairly.
Life sciences commercial analytics
Territory and account performance, prescribing and procurement trends from licensed data sources, field activity and market access reporting, with compliance controls on what field teams may see.
Clinical trial operations
Enrolment, site performance, monitoring and data-query ageing reporting against protocol targets, with full audit trail.
Quality and safety reporting
Incident, infection and readmission analytics with correct denominators, trends tested for significance rather than reported as month-on-month movement.
Integration
Source systems we connect to
- Electronic patient record systems
- Patient administration systems
- Theatre and scheduling systems
- Laboratory information systems
- e-rostering and HR systems
- CTMS and EDC platforms
- Pharmacovigilance systems
- Licensed market and prescribing data
Regulatory context
Regulatory context by market
- UK: UK GDPR, DSPT assurance, NHS information governance expectations, Caldicott principles
- EU: GDPR Article 9 conditions for health data, EHDS developments, EU Clinical Trials Regulation
- Germany: Landesdatenschutz authorities plus sector requirements for Kliniken, KHZG digitisation reporting
- Switzerland: revFADP and cantonal health data rules
- Ireland: Health Information and Quality Authority expectations, DPC guidance on health data
- Life sciences: GxP data integrity (ALCOA+), 21 CFR Part 11 where US-facing, EFPIA transparency reporting
Questions
Healthcare: common questions
Yes, with the right design: in-region deployment, pseudonymisation at ingestion where identifiers are not needed analytically, separation of the re-identification key under tighter control, role-based access with evidence, and retention enforced in the platform. The architecture decisions should be reviewed by your information governance function before build, and we write them up specifically for that review.
We build the data and reporting layer and support validation evidence (specification, qualification documentation, change control and audit trail), working alongside your quality function, which owns the validated state. We are explicit about the boundary: we do not self-certify validation.
As a model feature rather than a report-level afterthought: suppression and rounding rules implemented in measures, applied consistently, with a documented threshold agreed with your information governance team. This prevents the common failure where a drill-through exposes a cell that the summary correctly suppressed.
Yes, with separate security boundaries and clear data-sharing rules. The usual design keeps conformed dimensions shared while facts and access are segregated, so group reporting is possible without creating an inappropriate flow of patient-level data between entities.
Services that apply here
Next step
Analytics work in healthcare
A 45-minute working session with an engineer rather than a sales call. You describe the problem; we set out what we would build, in what order, and what it is likely to cost.
Answered within one working day, by the engineer who would do the work