Free scoping read. A written assessment of your manuscript in two business days, at no cost.

Start one

Home›Services›Data, analytics & AI›Patient journey & insights

Statistical Analysis and Reporting

Patient Journey Analytics and Insights

Pubrica provides patient journey analytics that reconstruct real-world care pathways for pharmaceutical companies, healthcare providers, payers and research organisations. We apply machine learning to each stage of the care pathway, including onset, diagnosis, treatment, adherence and outcome, using EHR and EMR records, claims, wearables and patient-reported data. Our data scientists reconstruct the pathway from the data you hold and indicate which steps the evidence supports. Deliverables include pathway maps, predictive models, reports and dashboards, so that your teams can act on the pathway patients actually follow rather than the one described in the guideline.

  • Pathways reconstructed from several sources at once — EHR, EMR, claims, wearables and digital health apps — because no single source contains a whole journey
  • Predictive models for adherence, progression and drop-off, validated on held-out data rather than reported on the set they were fitted to
  • Treatment patterns and switching identified across structured and unstructured data, including the unmet need that shows up as a switch rather than as a complaint
  • Natural language processing on free text — clinical notes, surveys and patient forums, where much of the patient journey is recorded but not coded
  • Built under HIPAA and GDPR, with governance agreed before any patient-level data is transferred
What journey analytics delivers Included and excluded

Pathway reconstruction across sourcesIncluded

Adherence and drop-off modellingIncluded

Treatment pattern and switching analysisIncluded

Dashboards and reportingIncluded

Causation from observational dataAssociation only

Re-identifying individual patientsNever

Sources → linked → modelled → validated → acted on

Every journey engagement
  • Governance settled before data moves
  • Cohort definition written first
  • Models validated on held-out data
  • Association not reported as cause
  • Dashboards you keep
HIPAA and GDPRPlus ICMJE, STROBE and GCP for the reporting
3–10 weeksBy package · multi-geography quoted
EHR, claims, wearables, free textLinked, not analysed separately
ISO 9001:2015Quality management certified · cert. H3CR9TCU20251212INDQM10N

Service Scope

Eight Patient Journey Analytics Services

Each project is scoped on the order. Cohort definitions and clinical interpretation remain with your clinicians.

Journey mapping across linked sources Touchpoints tracked and analysed across EHR, EMR, claims, wearables and digital health apps, to reconstruct the care pathway that actually happened rather than the one the guideline describes.

From the pathway mapMedian 11.4 months from first symptom code to diagnosis, across four primary-care contacts. Two of the four carried a differential that was never followed up. That interval is the finding; the guideline assumes three months.

Predictive modelling of patient behaviour Models forecasting treatment adherence, disease progression and drop-off risk, so intervention can be targeted at the patients it would change rather than at everyone.

From the model reportAUC 0.91 reported on the training dataAUC 0.74 on a temporally held-out set, which is the number that means something. Calibration reported alongside discrimination, because a model that ranks well and is badly calibrated misleads at the threshold you would act on.

Disease burden and treatment patterns Analysis of therapy adoption, patient switching and unmet need across structured and unstructured data. Unmet need often appears first as a treatment switch.

From the pattern analysis“Switch rate 18%.”18% switch within 6 months, and 71% of those switch after a specific adverse event that appears in the notes rather than in the coded data. The number needed NLP on free text to mean anything.

Sentiment and intent analysis NLP applied to surveys, patient forums and social data to understand perception, satisfaction and expectation — with the sampling bias stated rather than ignored.

From the sentiment analysisForum data over-represents patients with a bad experience and under-represents those doing well. The report says so, weights where weighting is defensible, and does not present the result as a population estimate.

Clinical and real-world evidence integration Trial data harmonised with real-world data to examine outcomes, safety and effectiveness in populations the trial excluded, which is often the population of interest.

From the harmonisation noteTrial and RWD pooled into one analysis setHarmonised but analysed separately and compared. The trial population excluded the comorbidity that 41% of the real-world cohort has; pooling them would hide exactly the question you are asking.

Custom models for insight generation Clustering, classification and deep learning built for the specific question — segmentation, risk stratification and pathway prediction — rather than a standard pipeline applied to a new dataset.

From the modelling planClustering with the number of clusters chosen by stability across resamples rather than by silhouette on one fit. Clusters that do not reproduce are not reported as segments.

Market access and HEOR support Patient insights prepared in a format that health economic models and reimbursement submissions can use, which differs from an internal insight presentation.

From the HEOR handoverResource utilisation by pathway stage, adherence by subgroup and time-to-treatment distributions — formatted as model inputs with their uncertainty, not as headline figures.

Dashboards and reporting Interactive dashboards for the stakeholders who will use the findings, built so a filtered view still says how many patients it rests on.

From the dashboard specFilter to a subgroup, see a rateEvery filtered view shows its denominator and greys out below a minimum cell size — for disclosure control, and because a 100% adherence rate on four patients is not a finding.

Samples are illustrative and anonymised. Hover any one of them to play it again.

Types of Patient Journey Analytics

Six Types of Patient Journey Analysis

These analyses overlap and are therefore offered as packages. The right package depends on how many data sources must be linked and whether free text is in scope.

Pathway reconstructionThe real care pathway assembled from linked sources, with intervals, touchpoints and the points where patients leave it.
Predictive modellingAdherence, progression and drop-off risk forecast, validated out of sample, with calibration reported alongside discrimination.
Treatment pattern and switching analysisTherapy adoption, line of therapy, switching and discontinuation — and what precedes each, which is where unmet need appears.
NLP on unstructured dataClinical notes, surveys, forums and free-text fields analysed for the information that was never coded, with sampling bias stated.
Segmentation and risk stratificationClustering and classification for patient segments, with cluster stability tested across resamples rather than asserted from one fit.
Dashboards and monitoringInteractive reporting with denominators visible and small cells suppressed, plus periodic refresh where the question recurs.

Data we work with. EHR and EMR extracts, claims and billing data, wearables and remote monitoring, patient surveys and PROs, digital health platforms, clinical trial data and published literature. What we will not do: attempt re-identification, report a filtered view without its denominator, or present an association as a cause.

Who we serve

Six Client Groups We Support with Patient Journey Analytics

Each client group uses the analysis to inform a different decision, such as a care pathway, a launch plan, a coverage policy or a trial protocol, and each needs a different output.

Pharmaceutical and biotech companiesUnmet need and real-world behaviour identified, and evidence generated in the form regulatory and payer submissions can use.
Healthcare providers and hospitalsPathways mapped to improve care, progression and outcome predicted, and adherence improved where the analysis shows it can be.
Payers and insurersUtilisation and cost-effectiveness patterns, high-risk populations identified for proactive intervention, and real-world support for value-based models.
Medical device and diagnostics companiesAdoption pathways and diagnostic utilisation understood, so development and support programmes match how the device is actually reached.
Contract research organisationsProtocol design informed by real-world data, recruitment and retention supported, and trial outcomes modelled before the trial runs.
Regulators and HTA agenciesEvidence for therapy evaluation and risk-benefit assessment, generated from real-world data with its limitations stated rather than minimised.

Compliance and Standards

Data Protection and Reporting Standards for Patient Journey Analytics

Patient journey analytics must comply with data protection law first and research reporting standards second, because data protection determines which data the analysis can use.

HIPAAUS

HIPAA (De-identification and safe harbour)

What may be done with US patient-level data, and what de-identification has to achieve before anything else happens.

GDPREU / UK

GDPR and UK GDPR (Lawful basis and processing)

Lawful basis, data minimisation and where processing may take place. This decides the project shape rather than being documented after it.

STROBEobservational

STROBE (For observational reporting)

How an observational study has to be reported. RECORD extends it for routinely collected health data, which is most of what this service uses.

TRIPOD+AI, 2024

TRIPOD+AI (BMJ, 2024)

Where a prediction model is developed or validated, this is the reporting standard — and it asks directly about validation and calibration.

ICMJE Recommendations (January 2026)

Where findings are published, including the declarations about data provenance and access.

GCPICH E6(R3)

Good Clinical Practice (ICH E6(R3), January 2025)

Where the work sits alongside a trial or feeds a regulatory submission.

ISO9001:2015

Quality management (cert. H3CR9TCU20251212INDQM10N)

How the work is run, recorded and reviewed — not a data protection claim, and not presented as one.

Responsible Use of AI

Governance of Patient Data in AI-Assisted Analysis

Patient journey analytics uses data about real people, often under a specific legal basis. Governance is therefore part of the analysis itself, and governance decisions are made by people, not by tools.

What we use it forWe use AI tools for pattern detection across large event sequences, for drafting transformation code, and for flagging where two data sources disagree about the same episode.
What we do not use it forWe do not use them for any processing that moves personal data outside the agreed environment, for inferring sensitive attributes that were not consented, or for conclusions that a person has not checked against the underlying records.
What goes in your declarationThe tool, the stage it was used at, and the human oversight applied — because journals, funders and the ICMJE January 2026 update all now expect it, and an undeclared tool found later is treated as a research integrity matter rather than a style preference.

Why we publish no accuracy figure of our own

Accuracy figures for AI tools are usually reported by the companies that sell them.

Pubrica therefore does not quote a self-generated accuracy figure, because it would carry the same conflict of interest. Instead, we tell you which tool was used on your project, at which stage, and what a human reviewer changed afterwards.

Methodology

How Our Patient Journey Analytics Process Works

No data is transferred until governance is agreed. The analysis then depends on how the cohort is defined, which is a clinical decision rather than a technical one.

Step one

Governance and cohort definition

The lawful basis, the consent position and where data may be processed are agreed first. The cohort definition is then written down before any data is queried, because a cohort defined after seeing the results would bias the findings.

Step two

Data collection and linkage

Patient data assembled from EMR and EHR, claims, wearables, surveys and digital platforms, and linked, with the match rate and the patients lost to linkage both reported.

Step three

Preparation and quality assessment

We harmonise coding, profile gaps and record the limitations of each source, because a gap in the source data remains a gap in the reconstructed journey.

Step four

Modelling and analysis

We carry out pathway reconstruction, predictive modelling, pattern analysis and NLP on free text, choosing methods for the question rather than applying a standard pipeline.

Step five

Validation

Models are tested on held-out data and, where possible, on temporally separate data, with calibration reported alongside discrimination. The out-of-sample performance is the figure we quote.

Step six

Reporting and handover

Findings are presented in clear reports and dashboards, with strategic implications kept separate from the findings so that you can accept one without adopting the other.

Request a Free Pathway Map for One Cohort

We reconstruct the pathway for one cohort with the touchpoints, drop-off points and data gaps marked, and add a note on which questions your data can and cannot answer. Two business days, no charge.

What you provide
The dataA de-identified extract, representative
The questionWhat decision this is meant to inform
The cohortHow you currently define it
The sourcesWhat else exists and could be linked
The governanceConsent, DPA, ethics approval
The horizonWhen the decision is made

Please confirm governance before sending any data. Patient-level data is subject to a legal regime, and that regime, not the project timeline, decides where and by whom it may be processed.

Our experts

Your Patient Journey Analytics Team

A named team of subject specialists stays with your project from scoping to delivery. Your project coordinator introduces each member before work begins, and the handover records which of them completed each stage.

Subject matter expertClinical or scientific expert matched to your topic

Reviews the scientific content for accuracy and confirms that each conclusion is supported by the evidence. Matched to your field rather than allocated from a rota.

OwnsScientific accuracy and interpretation
Reads againstThe study protocol and current literature
FieldMatched to your topic
BiostatisticianStatistical input where the work reports analysed data

Checks that statistical methods, tables and conclusions are consistent with the analysis plan and that the analysis can be reproduced.

OwnsStatistical accuracy
Reads againstICH E9 and the statistical analysis plan
ToolsR, SAS, Stata, Python
Data specialistData handling, cleaning and documentation

Prepares, cleans and documents the data so that every step from raw data to final output can be traced and reproduced.

OwnsData handling and documentation
Reads againstThe data management plan and applicable data standards
OutputDocumented, reproducible datasets
Second reviewerIndependent check of code and outputs

Independently checks code, tables and outputs against the analysis plan. Discrepancies are logged and resolved before delivery.

OwnsIndependent quality control
Reads againstThe analysis plan and outputs
Signed off byProject lead

Sample work

Sample Work: Adherence Model, Before and After Validation

Step through the changes one at a time. The model is unchanged; what changes is how its performance is reported, and only the validated figures are a sound basis for decisions.

6 changes, each explained

As first reported. A 12-month adherence prediction model on 18,000 linked patients.

  • AUC 0.91
  • “Validated by cross-validation”
  • Prior adherence among the strongest predictors
  • Accuracy 89%
  • No calibration reported
  • Cohort defined during the analysis

Validated. Same model, same data, reported so a decision can rest on it.

  • AUC 0.74 on a temporally held-out year — the number to plan with
  • Preprocessing fitted inside the folds — the 0.91 included leakage
  • Prior adherence dropped — not available at the point of decision
  • Accuracy is 89% because 88% adhere; PPV and recall reported instead
  • Calibration plot included; the model over-predicts in the top decile
  • Cohort definition written and frozen before querying

Figures illustrative; the therapy area, cohort size and values are changed. The pattern is the one we see most often.

Illustrative · therapy area and values changedAdvanced package · TRIPOD+AI reporting

Benefits

Benefits of Real-World Patient Journey Analysis

Each of these is checkable in your own report and dashboards. None is a causal claim, because observational data does not support one.

01 Measurement of care intervalsThe time from first symptom to diagnosis, and where that time is spent, is often the finding that changes what an organisation does. It is rarely known until the pathway is reconstructed.
02 Drop-off points identifiedKnowing that adherence falls is of limited use. Knowing that it falls at month four, in a particular subgroup, after a particular event, allows targeted action.
03 Analysis of free-text recordsClinical notes carry what the coded data does not. A switching pattern that looks arbitrary in claims data often has an obvious cause in the notes.
04 Validated model performanceOut-of-sample validation and calibration are reported in full. An over-optimistic model is worse than no model, because resources are committed on the basis of it.
05 Association kept separate from causeObservational data supports association. Where a causal question is being asked, the report says what design would answer it rather than implying this one has.
06 Governance settled firstThe lawful basis and the processing location are agreed before any data is transferred, which avoids delays later in the project.

Packages

Patient Journey Analytics Service Packages

Each package is priced on the number of data sources to be linked, whether free text is in scope and the number of geographies involved. We therefore provide the free pathway map first.

Basic

Basic package

Small research teams and early-stage studies wanting a foundational view

Quoted per projectPriced on cohort size

Delivery 3–4 weeks

  • ✓Data extraction from EMR/EHR, surveys and published sources
  • ✓Patient demographics and treatment pattern analysis
  • ✓Standard visualisation dashboards
  • ✓Summary report of the key journey findings
  • ✓Optional: custom data cleaning, advanced visualisation, regulatory alignment

Suited to questions that a single data source can answer.

View a sample Request a quote
Advanced

Advanced package

Pharma, biotech and providers needing predictive analysis

Quoted per projectPriced on sources linked

Delivery 5–7 weeks

  • ✓Multi-source integration — EHR, wearables, claims, social and survey data
  • ✓Predictive modelling of outcomes, validated out of sample
  • ✓Sentiment and intent analysis from patient-reported data
  • ✓Detailed journey mapping with touchpoints and drop-off analysis
  • ✓Strategic insights report
  • ✓Optional: platform integration, therapeutic-area benchmarking, stakeholder decks

Suited to projects that need several linked data sources.

View a sample Request a quote
Premium

Premium package

Large pharmaceutical companies, CROs and global healthcare organisations

Quoted per projectPriced on geographies and refresh

Delivery 8–10 weeks

  • ✓End-to-end mapping across geographies and therapeutic areas
  • ✓Advanced ML — NLP, image recognition and deep learning
  • ✓Real-time dashboards with interactive analytics
  • ✓Comparative analysis across treatment regimens
  • ✓Comprehensive reporting with strategic recommendations
  • ✓Optional: API integration, monthly or quarterly refresh, expert consultation

Each market has its own governance requirements, so please name all geographies at scoping.

View a sample Request a quote

Send one de-identified cohort extract and we will map the pathway free of charge, with a note on what your data can and cannot answer. Ongoing monitoring is quoted against a schedule.

Quality Assurance

Our Service Commitments

Each commitment below states what we check, its limits and the remedy if the work does not meet the level on your order.

The Pubrica guarantee

Only Out-of-Sample Performance Is Reported

Every model is reported on data it was not fitted to, with calibration alongside discrimination and the threshold you would actually act on, because resources are committed on the basis of these figures.

01

Governance before data

Lawful basis, consent position and processing location are agreed first, because they determine how the project is designed.

02

Cohort defined before querying

Written and frozen in advance, so the definition is a study decision rather than a result.

03

Association is not reported as cause

Where the question is causal, the report says what design would answer it instead of implying this one has.

04

Free rework within the agreed scope, or a refund

Where the work does not meet the level on your order we do it again at no charge within the agreed scope, and where you would rather not, we refund it.

Client Testimonials

“The work uncovered patterns in behaviour and adherence we had not seen. The drop-off point was four months in, which changed where we put the support programme.”
Dr. KlimChief medical officer

USA

Patient journey FAQs

Patient Journey Analytics FAQs

If your question is not answered here, ask an expert — no quote required.

It makes the analysis feasible at the volumes involved. Linking millions of records across sources, reconstructing sequences and reading free text cannot be done manually at scale.

It does not change what can be inferred. These are observational data, so the results show association and sequence, not causation.

Electronic health records, clinical trial data, wearables and mobile apps, patient surveys and feedback, claims and billing data, and digital health platforms.

The linkage is usually the hard part. Match rates are never 100%, and the patients who fail to link are rarely a random sample — so the report states the rate and who was lost.

Pharmaceutical and biotech companies, hospitals and providers, medical device manufacturers, public health organisations, payers and insurers, CROs and HTA agencies.

The output differs between these groups. A hospital needs the intervals in its own pathway; a market access team needs model inputs with their uncertainty attached.

Understanding disease progression as it actually occurs, locating where adherence fails, informing R&D and trial design with real-world behaviour, and optimising treatment pathways against evidence rather than assumption.

A frequent and useful finding is a previously unmeasured interval, typically the time from first symptom to diagnosis and where that time is spent.

The lawful basis, the consent position and where processing may occur are agreed before any data is transferred. Data is de-identified to the applicable standard, with controlled access, encrypted transfer and no third-party tools.

Dashboards suppress small cells and always show denominators, both for disclosure control and because a rate calculated on four patients is not a reliable finding.

We can tell you what precedes it, how consistently, and in whom. That is association, and it is often enough to act on.

Observational data cannot establish cause. Where that is the real question, the report states which study design would answer it.

Related services

Related Data Analytics Services

This service reconstructs the patient pathway. The services below prepare the data, extend the modelling or turn the findings into a submission.

Request a Free Pathway Map for One De-identified Cohort

We mark the touchpoints, drop-off points and data gaps, and add a note on which questions your data can and cannot answer. Two business days, no charge.

Guides and articles in this section

Show all 1
Get a free quote

Regulations

Check your project against this rule

  1. 1Your work
  2. 2Your details
  3. 3Send
What do you need?

Material you send is held under signed confidentiality, with access limited to the assigned team and encrypted transfer. It is not used to train anything and is not passed to third-party tools. Ask for the agreement before you send anything you would rather cover first.