Clinical Data System
Case Study 03

Project Health Equality

Data-Backed Playbook for Offline Screening Lift

Overview

We built the measurement infrastructure before we could prove the outcome—linking media timing to offline mammography utilization across 3 geographies, with a 90-day lag built to reflect the real screening journey, not standard attribution windows

Goal

Drive real-world mammography utilization among Black women despite disparities in access and conversion—and build a repeatable measurement framework so future BCAM* cycles have evidence to plan from, not assumptions.

Team

Product, analytics & optimization, data science, omnichannel

Role

Owned the activation strategy across experience design, go-to-market, and measurement, aligning campaign execution with a repeatable framework for evaluating real-world screening behavior.

We linked media timing to screening utilization across geographies using Hologic’s Unifi network data to estimate incremental lift and separate “noise” from signal.
1 / 4

Problem

Completed mammograms happen offline, while access, trust, and capacity constraints can erase digital intent before it becomes screening behavior

The challenge was not awareness alone—it was proving whether BCAM* media drove completed mammograms across 3 geographies despite a 90-day conversion lag, unstable local baselines, and offline care pathways. That mattered because Black women face 38% higher breast cancer mortality despite 5% lower incidence1.

Problem 01

The true outcome was offline

Standard web metrics could not prove screening lift

  • Completed mammograms happened in clinics, not on websites, so the outcome had to be measured through offline utilization data rather than digital conversion events.1, 2
  • CDC notes that mortality differences reflect inequities across the full continuum, including screening, follow-up, and treatment.2
  • Standard attribution couldn't answer the question. We had to build the measurement method—which is why the framework is the durable product, not the campaign.
Problem 02

The behavior was delayed

Screening often converted weeks later, not immediately

  • Black women were 40% less likely than White women to receive diagnostic imaging follow-up within 15 days after an abnormal screening mammogram.3
  • When biopsy was recommended, Black women waited 1.75x as long to obtain a tissue diagnosis.3
Problem 03

Baselines were noisy

Lift detection was unstable because local conditions also changed outcomes

  • In a cohort of 26,476 women, disparities appeared across the screening continuum, not just at the initial screen.4
  • Offline screening outcomes were also affected by follow-up access, treatment coordination, and facility resources, which made local baselines harder to interpret.2, 5
Problem 04

The equity stakes were high

The outcome mattered because the disparity was large

  • The measurement stakes were high because the health stakes were high. Black women face 38% higher breast cancer mortality despite 5% lower incidence1 — which meant proving lift wasn't just analytics rigor. It was the evidence base for where to invest next.
  • The disparity is longstanding; CDC previously reported Black women were 40% more likely to die of breast cancer than White women.6

*Breast Cancer Awareness Month (BCAM); 1. American Cancer Society, Breast Cancer Facts & Figures 2024–2025; 2. CDC MMWR, Racial Disparities in Breast Cancer Severity and Survival; 3. Manik et al., 2024 review, Racial Disparities and Strategies for Improving Equity in Breast Imaging; 4. Ganguly et al., 2023, Racial disparities in the screening mammography continuum; 5. Karliner et al., 2018, Mammography facilities serving vulnerable women; 6. CDC Vital Signs, Breast Cancer

Converting awareness into completed mammograms is hard—because the barriers aren’t digital
1 / 3

Solution

We built a 3-geo offline-lift framework with a 90-day lag to estimate whether BCAM media drove completed mammograms

The solution was a practical measurement playbook, not a perfect attribution model. We aligned pre-, during-, and post-flight periods to campaign timing, added a 90-day lag to reflect the real screening journey, and used Unifi exam volumes to estimate directional lift by market.

Framework 01

Multi-geo lift detection

Hologic's Unifi network provided anonymized mammography exam counts across Houston, Atlanta, and Miami—allowing us to compare screening volume trends during and after campaign flight periods against pre-campaign baselines.

Framework 02

90-day lag window

Instead of standard 7- or 30-day attribution, we built a 90-day post-campaign window to capture the full scheduling-to-screening behavior cycle, acknowledging that real-world screening journeys don't happen instantly.

Framework 03

Baseline comparison approach

Pre-flight screening volumes established local baselines for each geography, so we could detect directional lift—even if local conditions (facility capacity, competing health events) introduced noise into the signal.

Framework 04

Confounder tracking

We documented external factors (local health campaigns, facility closures, pandemic disruptions) alongside exam volume data to help separate media-driven lift from environmental noise—critical for building a repeatable playbook.

Measurement Framework

Campaign Inputs

Geo selection

Flight timing

Message penetration

Unifi exam volumes

Lift Model

Pre / during / post periods

90-day lag window

Baseline comparison

Significance check

Decision Outputs

Lift by market

Confounder readout

Geo prioritization

Next-cycle playbook

This framework made the next cycle plannable—not just reportable. Geo selection, flighting, and investment decisions now have an evidence base instead of assumptions.

The approach was grounded in offline reality
1 / 3

Scale

Houston showed a +4.7% statistically significant lag effect, while Miami was positive at +2.8% and Atlanta appeared confounded at -5.5%

The results mattered less as a campaign score and more as a market readout. Houston's lag-only lift pointed to scheduling behavior; Miami's sustained lift suggested awareness plus follow-through; Atlanta's -5.5% correlated with COVID-driven utilization spikes—not media failure. Knowing the difference is what makes future geo selection defensible.

GeoEst. message penetrationWindow evaluatedΔ Exams (est.)Lift vs baselineSignificance / readout
Houston27-32%Lag only+4,563+4.7%Post-flight lift consistent with scheduling delay
Atlanta33-38%During + Lag-10k-5.5%Likely confounded
Miami23-27%During + Lag+3,600+2.8%Sustained lift suggests awareness + follow-through

← Swipe to see more →

Signals captured

The strongest signal appeared where baseline patterns and operational constraints allowed conversion—which made local context as important as media performance.

The framework separated signal from noise—which matters when baselines are unstable and confounders are everywhere
1 / 4

Intelligence

The analysis became a repeatable BCAM measurement + optimization playbook built around capacity, baseline stability, and care-pathway readiness

The most durable product outcome was not the Houston result—it was the playbook. We translated the lift analysis into a geo selection framework, flighting strategy, and reporting template—identifying the conditions under which awareness converts: capacity, appointment access, provider alignment, and a stable geo baseline.

01

Impact Factors

Capacity constraint · baseline volatility · competing campaigns · journey friction

02

Conversion Conditions

Capacity available · appointment access · reminder / retargeting · provider alignment · geo baseline stable

03

Playbook Outputs

Geo selection framework · flighting strategy · reporting template · stronger quasi-experimental measurement roadmap

This reframed the work from campaign reporting into a decision product—future plans optimized for outcome conditions, not just impressions.

The playbook turned one-time measurement into a repeatable decision framework
1 / 4