Methodology

Every data source. Every calculation. In language a steering committee can follow.

We watch what HCPs think, what they do, and what shows up in claims. We score the change. We compare the HCPs your medical team engaged against the ones they did not. The gap is your contribution. The rest of this page explains exactly how, with no hand waving.

Step 1

The four data layers we gather

Four input streams feed the model. Each has a different source, a different refresh cadence, and a different level of detail (HCP, account, or region). We are explicit about what each one cannot tell us.

None of these streams alone proves contribution. Combined and aligned to a strategic imperative, they form the evidence chain.

HCP perspectives

What clinicians believe about the disease, the biomarker, the guidelines, and your asset. Captured from surveys and MSL call notes, coded against the Mayne contribution ladder (Awareness, Understanding, Acceptance, Adoption, Advocacy).

Source: panel surveys + MSL CRM. Grain: HCP. Refresh: quarterly. Limit: self-report bias is real, which is why we triangulate against behaviour.

HCP behaviours

What clinicians actually do. Prescribing volume, biomarker test ordering, referral patterns, line-of-therapy choices, and adherence to guideline-recommended sequencing. Pulled from claims and Rx data.

Source: open + closed claims, Rx. Grain: NPI / account. Refresh: monthly with claims lag. Limit: claims lag 60 to 90 days; off-label use can be invisible.

Biomarker data

IHC and NGS results joined at NPI or account level. Lets us see who was tested, what the result was, and whether the treatment that followed matched the result. Sourced from lab feeds and reference partners.

Source: Foundation Medicine, regional lab feeds. Grain: patient, rolled to account. Refresh: monthly. Limit: incomplete capture in community settings.

Engagement data

Every NBA-guided medical touch. Who was engaged, when, by which channel, against which strategic imperative, targeting which dimension. This is the mechanism we test contribution against.

Source: MSL CRM with NBA tags. Grain: HCP touch. Refresh: weekly. Limit: only NBA-guided touches are isolated; ad hoc engagement is harder to attribute.

Step 2

Turning perspectives into a number: the ladder scoring engine

The Mayne contribution ladder has five rungs: Awareness, Understanding, Acceptance, Adoption, Advocacy. Every HCP receives a score from 1 to 5 on every dimension of every strategic imperative.

Scores are not opinions. They are derived from coded evidence: specific survey responses, specific phrases or assertions in MSL call notes, congress activity, publication patterns. Two coders, blind to engagement status, score each piece of evidence. Disagreements go to adjudication.

Coverage is the share of the HCP universe for which we have at least one piece of evidence on that dimension in the period. Coverage is shown next to every score. A score with thin coverage is flagged, not hidden.

Worked example

Rung 1
Awareness
Rung 2
Understanding
Rung 3
Acceptance
Rung 4
Adoption
Rung 5
Advocacy
T-DXd, dimension: "HER2-low reclassification on central review"
Quarter: Q2. HCP universe: 1,820.
  • mean = 3.4 on the 1 to 5 ladder
  • coverage = 62% (1,128 HCPs with coded evidence)
  • baseline = 2.7 (prior quarter)
  • delta = +0.7 rungs

A 0.7 rung shift at 62% coverage is a defensible signal. The same shift at 18% coverage would be reported as "insufficient evidence" rather than a number.

Step 3

The headline number: contribution, not attribution

Contribution analysis follows Mayne (2012). It does not claim that medical engagement caused the change in a counterfactual sense. It claims that engagement plausibly contributed, through a coherent and evidence-supported causal theory, demonstrated by correlated ladder movement.

That distinction matters. Counterfactual claims invite legal and compliance risk and collapse under scrutiny. Contribution claims are auditable, defensible, and survive a critical reviewer.

  1. 1
    Baseline score
    Mean ladder score per dimension in the pre-period.
  2. 2
    Current score
    Mean ladder score per dimension in the post-period.
  3. 3
    Score change
    current minus baseline, per dimension, per HCP.
  4. 4
    Engaged cohort
    HCPs with ≥1 NBA-guided engagement on that dimension in the period.
  5. 5
    Universe cohort
    Matched HCPs (same specialty, decile, geography) with no NBA engagement.
  6. 6
    Engagement delta
    score change (engaged) minus score change (universe).
  7. 7
    Contribution signal
    The delta, paired with coverage % and direction of fit to the strategic imperative.
Worked example
engaged_delta = +0.82 rungs
universe_delta = +0.31 rungs
contribution = +0.51 rungs at 67% coverage

Read as: HCPs your team engaged moved 0.51 rungs further than comparable HCPs they did not, on this dimension, in this period. That movement is consistent with the strategic imperative. That is the contribution claim.

Step 4

From ladder shift to real-world outcomes

A ladder shift is only interesting if it shows up in the world. We connect behaviour score changes to outcomes visible in claims and biomarker data using four standard views.

Patient flow Sankey

Diagnosis through 3L+, with absolute patient counts scaled to claims. Proportions update live as you move simulator levers, so you can see flow under projected conditions.

Time-to-next-treatment curves

Kaplan-Meier style cohort curves. Patients are censored at last observed claim. Cohorts split by biomarker status and treatment line so attrition is comparable like for like.

Biomarker concordance

Joining lab results with treatment claims at NPI level surfaces two key gaps: positive-untreated (eligible patients missed) and negative-treated (off-label or pre-result decisions).

Geographic gap map

Testing rate and time-to-2L by region or HCP decile. Lets you spot disparities a national average hides and target the next-best-action where the gap is widest.

Step 5

The what-if simulator: how projections are built

The simulator lets you move levers (biomarker testing rate, share of voice, retention from 1L to 2L) and see the projected impact on incremental claims and revenue.

Each lever maps to an elasticity coefficient calibrated on the historic claims base using a regularised regression. We hold all other levers constant when projecting a single lever, so the user can reason about one variable at a time.

Every slider shows three things in plain English: the elasticity, the confidence interval, and the one or two assumptions that matter most. If you cannot explain a projection in a sentence, we do not surface it.

Lever example
Biomarker testing rate +10 percentage points
Elasticity0.62 claims per tested patient
Projected incremental claims+1,840 / quarter
90% confidence interval+1,210 to +2,470
Key assumptionTest-positive rate holds at the observed Q2 level.
Step 6

What the model does not claim, on purpose

No counterfactual causality

We never claim 'engagement caused X'. We claim engagement plausibly contributed to X through a documented causal chain. This is a feature, not a hedge.

No single-touch attribution

We do not assign credit to the last touch, the first touch, or any single touch. The unit of analysis is cumulative engagement against a dimension over a period.

Coverage thresholds

Below the coverage threshold we set with the client (typically 25 to 30%), a dimension is reported as 'insufficient evidence' rather than a number. Thin signals are not surfaced.

Honest data gaps

Missing biomarker data, off-label use, and small-cell suppression in regional cuts are flagged in the relevant view. We tell you when a result is shaped by what we cannot see.

Audit trail

Every score change traces back to the coded evidence behind it. MLR, compliance, and the steering committee can drill from headline number to source survey item or call note in two clicks.

No black-box models

Every coefficient, every threshold, and every cohort definition is documented and reproducible. If we cannot explain how a number was produced, we do not ship it.

Appendix

Glossary

Mayne contribution ladder
Five-rung scale (Awareness, Understanding, Acceptance, Adoption, Advocacy) used to score where each HCP sits on a given dimension.
Dimension
A single, specific belief or behaviour we score against. Many dimensions roll up into a strategic imperative.
Strategic imperative
A medical strategy goal (e.g. 'close the HER2-low testing gap'). Each one decomposes into dimensions and care gaps.
NBA (Next Best Action)
An algorithmically guided recommendation telling an MSL which HCP to engage, when, on which topic.
Engagement delta
Score change in the engaged cohort minus score change in the matched universe cohort. The contribution signal.
Coverage
Share of the HCP universe for which we have coded evidence on a dimension in the period. Used to weight confidence.
Concordance
Whether the treatment given matches what the biomarker result would recommend. Surfaces positive-untreated and negative-treated gaps.
Contribution vs attribution
Contribution: plausible causal chain, evidence supported. Attribution: counterfactual proof a specific touch caused the outcome. We do the first, never the second.

See the methodology in action

Sign in to explore the live model with the Dato-DXd and T-DXd reference data loaded.

Sign in to explore