Applying Real-World Evidence Data for Measuring Pharmaceutical Digital Media Programs

Applying Real-World Evidence Data for Measuring Pharmaceutical Digital Media Programs
Ira Haimowitz, PhD, Vice President, Product Strategy, Crossix and Whitney Kemper, Sr. Director, Analytics Products, Crossix
Abstract:
Pharmaceutical marketers can now leverage advanced clinical metrics found in real-world evidence data to:
  1. Better understand the behaviors of audiences exposed to relevant media;
  2. Improve digital DTC campaign optimization; and
  3. Measure the direct-to-consumer incremental (exposed vs. control) impact on the patient journey.
This is demonstrated through specific case studies in the type 2 diabetes therapeutic category, showing enhanced insights and improved modeling accuracy.


Keywords: marketing optimization, real-world evidence, promotion evaluation, media measurement, health informatics, pharmaceutical marketing, digital media

Background
In recent years, improved technology and access to electronic health records (EHR) systems have enabled real-world evidence (RWE) data to be used for a range of healthcare applications. This data includes patient vitals captured during doctor visits, laboratory tests, and medical exam notes that describe symptoms and treatment responses in free text. The most common applications for RWE data are selecting institutions for clinical trial sites, modeling patients for clinical trial recruitment, and outcomes research measurement.1,2
 
More recently, RWE data has been applied to measuring direct-to-consumer (DTC) marketing programs for pharmaceutical brands. New audience quality measures for digital, television, point of care, and print campaigns based on RWE data have been developed.2,3 For example, a type 2 diabetes brand team can now evaluate what percentage of those reached during their media campaign have elevated A1C levels indicating high blood sugar.
 
As an extension, this article describes two advanced applications of RWE data for media measurement. In the first application, in-depth insights are highlighted across key milestones along the patient journey and the effect of digital media exposure on patient decisions and timing. In the second application, RWE metrics are aggregated at the media referral source level to be used for a predictive model that forecasts future media performance in driving new patient conversions to brand, which is often brands’ key media objective.
Digital Media Measurement and Optimization in Pharmaceutical Marketing
Digital marketing is a critical component of pharmaceutical marketers’ DTC advertising campaigns. According to Kantar Media, pharmaceutical spending on digital ads was $515 million in 2016, roughly 10% of total media spend.4 The analyses in this article focus on DTC digital display, mobile, and online video advertising that can be delivered through desktop, tablet, mobile web, or mobile app platforms. With this focus, there is still significant spend; a single branded digital display and online video media campaign may entail an investment that ranges from $1 million to $5 million.  Although this article focuses on digital display and video, there are similar measurement solutions that can be applied to evaluate paid search, website evaluation, and digital point of care, as well as healthcare professional digital media.
 
Over the past decade, the mix of digital publishers has become increasingly diverse and complex. Ten years ago, healthcare brand marketers usually funneled the bulk of their digital ad spend to endemic medical publishers that provide consumers with trusted sources with detailed health information by therapeutic category. Today, endemic publishers have also diversified themselves and now include niche patient disease-specific communities and websites that enable consumers to search for physicians, make appointments, and rate doctors’ services. Digital ad networks are more diverse sets of websites for purchasing display advertising.
 
However, the true rise in display advertising has come from programmatic digital advertising, in which automated targeting algorithms and online auctions determine which consumer is served an ad. With programmatic capabilities, healthcare advertising can be served to the most relevant consumers anywhere they visit the internet, even on news, sports, or entertainment websites. Informative overviews of programmatic techniques can be found at the Interactive Advertising Bureau.5 Specific providers have advocated that programmatic media is compatible with pharmaceutical privacy concerns and offers benefits of efficiency.6 An analysis of a representative sample of digital campaigns showed that over two-thirds of digital impressions are generated from programmatic media.7
 
Given the breadth of DTC digital channels and tactics now available, healthcare advertisers want to know if they are spending their money wisely, and there is an increasing premium on media measurement and optimization. In the early 2000s, digital ad measurement was primarily assessed in two ways:
  • Online primary research surveys of website visitors and media viewers: this was limited by long project durations where sufficient research sample accumulated, as well as its reliance on self-reported data rather than actual health transaction outcomes.
  • Online “engagement” measured as click-through rates of digital ads and subsequent website visitor behavior after clicks: this was limited because only a small percentage of media viewers actually click on ads, and it uses rough estimates linking website activity to health outcomes.
More recently the pharmaceutical industry has changed its approach to addressing this need.8 No longer is digital ad measurement based on analyzing click-throughs to websites. Instead, there are new methods that directly link media exposure to health metrics that are more closely aligned to brands’ key marketing objectives, like reaching a qualified audience and generating new patient starts on their brand. In this way, digital measurement is now compatible with that of other prominent DTC media channels like television.  
 
The health metrics that were evaluated for digital advertising, and all media channels, fall into three primary classes:
  • Audience quality: the percentage of a media-exposed consumer audience that is relevant from a health history perspective. This includes prior diagnosis with the relevant medical condition, previous treatment with medications in the relevant category, or prior results on relevant vitals or laboratory tests.
  • Intent to treat: the percentage of consumers exposed to media that proceeds to visit a primary care physician and/or a relevant specialist within a one-month time frame after exposure.
  • Conversion: the percentage of consumers that are exposed to media and later begin a new treatment for the advertised brand or within the relevant condition category (or sub-category) within three months after exposure.
Importantly, these metrics are evaluated net of a control group, matched by age, demographics, geography, and prior patient treatment history, in order to determine the incremental value driven by exposure to the campaign.
 
It is not sufficient to calculate these metrics in a one-time study at the end of a media campaign because a dynamic media channel, such as digital, requires ongoing monitoring and optimization. To optimize media spend, pharmaceutical marketers require two main components from their measurement system:
  • Granularity of Reporting: measurement of results not only at the overall digital campaign level, but separately for each publisher. Instead, metrics within each publisher at the level of placement groups were measured. These are sets of display or video ad locations that share commonalities of targeting, format (mobile or desktop), and content theme. Placement groups are ideal for enabling marketers to maintain their spend with a critical publisher and maximize their impact on quality reach and post-media behavior.
  • Frequency of Measurement: repeated evaluation and specific detection of trends, outliers, and market shifts. In digital measurement, all metrics were measured weekly as a standard.
Data Sources for Both Studies
For both studies, comprehensive distributed data network technology was leveraged, which includes clinical data (EHR, medical claims data, prescriptions, etc.), frequent shopper loyalty card data for OTC and packaged-goods purchases, consumer data (demographics, financials, interests, etc.), multi-channel media, and other data sets covering over 250 million U.S. consumers. A sample list of data attributes is shown in Table 1. Special attention was given to diagnoses from medical claims and lab results from EHR data.
Table 1: Data Attributes (U.S.)
Rx/Medical Claims/EHR Health Behaviors Shopping Consumer Media
250MM+ patients
Virtually all practitioners
Updated daily
CPG: 90MM+ households
Food and Drug Products
Updated daily
240MM+ U.S. adults
2,000+ different variables
Updated quarterly
Multi-channel
Consumer & HCP
Updated daily, weekly, monthly
Patient
Age, gender, geo

Rx
Date filled, product, quantity, refills

HCP
HCP/prescriber/specialty, location, office visit dates

Diagnoses
Diagnosis codes, lab orders & results

Payer & Cost

Pharmacy Type
Item
Date, product UPC, quantity, price

Store Type

Shopping Basket
Basket Size
Trips
Demographics

Financials

Interests & Hobbies

Media

Propensity to buy over certain channels – internet, mail, phone, cell phone, magazine, TV, etc.

Buying & Shopping Activity
Amount of spend in certain categories
TV
Digital
Print
POC
CRM
Email
Direct Mail
Sales Calls


Media data, in particular, was gathered using a tag-based implementation that tracks impressions from digital display, video, and mobile advertising. Media data is gathered at the overall campaign level, publisher level, and placement group level within publishers. Digital media impressions are matched to digital identities, which are anonymously matched to health data in a HIPAA-compliant way.
 
Each category of data attributes provides distinct benefits to digital measurement. Leveraging health data provides marketers with more insight into the consumers they reach through their campaigns and ensures that the right ads are reaching qualified audiences who are in line with the brand’s objectives. Shopping data can be used to indicate conditions where OTC products are being used as alternatives or as supplements to prescription therapy—like in the case of allergies. Additionally, many health conditions are often treated with concomitant diet modification (e.g. low salt for hypertension, sugar-free for diabetes, and gluten-free for celiac disease). Consumer data allows marketers to evaluate media targeting based on demographic objectives. For example, a contraceptive brand can target younger females, whereas an erectile dysfunction brand can target middle-aged males. This reduces media spend waste by only focusing on audiences that are clearly part of the intended audience. Media data for these analyses is accumulated through a tagging-based system, which captures time-stamped exposure to advertising at the most granular “placement” level. This data is collected in a HIPAA-compliant, privacy-safe manner, and is ultimately matched to consumer health transactions behind firewalls of data providers. Combined, these data sources provide a powerful foundation to optimizing digital media campaigns.
Patient Journey
The goal of pharmaceutical marketing is to influence a patient to fill a prescription for an advertised drug. However, there are many steps a patient takes before he or she decides (or does not decide) to take the intended action. This series of steps is oftentimes referred to as the patient journey and is a key way to understanding patient behavior and how certain milestones influence the decision to fill a prescription. By tracking these stages, a marketer can better recognize where progress is being made and where there are roadblocks to the patient getting on therapy.
 
Using the example of a type 2 diabetes therapy, there is a clear progression that must be followed for the patient to fill a prescription. First, a patient must take the initiative to visit an HCP to discuss his or her health and broach the topic of the advertised drug. For patients who are not already diagnosed with type 2 diabetes, an A1C test is usually administered. Depending on the results of that test, a patient may receive a new diagnosis of type 2 diabetes or prediabetes. Once a diagnosis has been made (or assuming a patient has already been diagnosed), the patient may be prescribed the relevant prescription or may get an earlier line therapy or a generic. In the latter case, there are follow-up visits, tests, and sometimes additional prescriptions.
 
Each touchpoint between the patient and HCP represents a step along the patient journey as well as an opportunity (or hindrance) to advance toward the brand marketer’s ultimate goal. The effect of marketing on each step of the patient journey can be measured by comparing the actual health behavior of patients exposed to advertising to the behavior of those in a control group who are not exposed to the same advertising. In doing so, the marketer can understand the impact of marketing on patients’ prescription filling behavior and where the marketing was effective (and ineffective) in moving the needle at each point in the progression of the patient journey. Additionally, marketing tactics can accelerate the timing for patients in taking their next steps along the treatment journey. This timing can be measured in days using transactional Rx, EHR, and medical claims data.
 
For example, a marketer may find that advertising is successful in getting patients to visit an HCP to discuss the possibility of them having diabetes, thereby increasing the rate of HCP visitation and the rate of patients receiving A1C tests and diagnoses. However, when it comes to the point at which an initial prescription is written (assuming the drug is a first-line therapy), patients may still receive prescriptions for competitor or generic drugs at the same rate as the control group. This would indicate that the advertising is effective in initiating a conversation with an HCP for new patients but not at influencing prescription-writing behavior. Had the marketer only considered the impact on conversion to their drug, they would have seen an increase in conversions but would have overlooked that this impact was happening due to more patients being diagnosed rather than an increasing share of prescriptions filled versus competitors.
 
The detailed descriptions of two research studies that use all of these data sets and employ the patient-level linkage between media exposure and post-media health behaviors including RWE data are below. These studies were also presented at the 2017 PMSA Annual Conference.9
APPLICATION 1: PATIENT JOURNEY POST-MEDIA EXPOSURE
Methodology
Over several type 2 diabetes-branded digital campaigns, key patient journey metrics were measured. Individuals exposed to digital advertisements were matched to EHR data with dates time-aligned relative to the dates of media exposure. Patient-level metrics related to doctor visits, lab tests, diagnoses, and ultimately treatment with a prescription were then calculated. These data preparation steps yielded a transactional data set of tens of thousands of patients, illustrated with a sample in Table 2.
Table 2: Patient-Level Behavioral Data from the Patient Journey Analysis*
Pre-Exposure Last A1C Result Pre-Exposure Last A1C Day Pre-Exposure T2D Diagnosis Day Post-Exposure PCP Visit Day Post-Exposure Endocrinologist Visit Day New Diabetes Diagnosis Day Rx Conversion Day Post Rx-Conversion A1C Day Post-Rx Conversion A1C Result
7.2 -26 n/a 6 125 125 126 128 7.2
8.4 -7 n/a 56 76 74 67 74 8.2
*Data illustrative

Each record in this dataset represents a single patient in the study; columns are a combination of patient journey elements and timestamps, expressed as days relative to the first media exposure at day 0. For example, the first row demonstrates that the patient had a 7.2 A1C level 26 days before digital media exposure, without a formal type 2 diabetes diagnosis historically. Six days after media exposure, the patient visited a primary care physician, and 125 days (about 4 months) after media exposure, the patient visited an endocrinologist. At the endocrinologist, the patient received a diagnosis, and a prescription (brand omitted), which was filled the next day (day 126). The lab test result came back on day 128 and showed an A1C of 7.2 again.
 
Patient-level metrics were then aggregated and summarized at a population segment level to determine the overall patient journey behavior for exposed individuals.
 
In analyzing the patient journey, distinct paths were identified by tracking relevant healthcare behavior across all consumers exposed to media. Certain paths were chosen for further analysis based on how frequently they appeared in the data and if they were relevant to the brand. Given its nature, EHR data may not always capture 100% of a patient’s longitudinal health behavior, which can lead to abbreviated or inconsistent paths for some patients. To prevent this from affecting the analysis, patient coverage and continuity eligibility rules were implemented for each patient to ensure that a complete, multi-stage path was being captured for each. Patients with insufficient data were removed from the analysis.
 
In order to measure the impact of the digital advertisements on this patient journey behavior, a control group was created to compare against the exposed treatment group. To do this, a group of individuals who were not exposed to the digital advertisements was identified. These individuals were similarly matched to EHR data. A matched pair control picking methodology was then used to identify patients who were similar to patients exposed to the digital advertising being analyzed. Patients were matched on both demographics (age, gender, geography) and healthcare characteristics (type 2 diabetes diagnosis, A1C level, Rx treatment, comorbidities).
 
After selecting this control group, the same patient level metrics were calculated for this population. Population rates were then compared to those of the exposed group in order to measure impact. In addition to measuring the impact on patient journey behavior, this same data was used to measure the impact on conversion to relevant drugs being advertised.
Key Learnings
Through analyses like this, it was discovered that the addition of RWE data can dramatically improve the assessment of media campaigns. For example, there was a significant lift (in some cases over 100%) across the patient journey metrics for the audience that was exposed to the relevant digital media versus those who were not exposed. Figure 1 illustrates the detailed metrics regarding patient visits to primary care physicians and endocrinologists after being exposed to a diabetes medication campaign.
Figure 1: Media Exposure Impact on Actual Patient Pathways by Physician Specialty


Consider for example, the leftmost branch of the Primary Care Visits tree of Figure 1. This indicates that those exposed to the digital media visited a PCP in 27 days, three days faster than those in the control group, which translates to a lift of 11.5% (note: the same timing difference was not detected for endocrinologists). Continuing down the tree, the exposed population received an A1C test at the PCP doctor visit 25.8% of the time compared to 18.7% of the time for the control group, a lift of 38%.  Reviewing the middle two branches of the Primary Care Visits tree, there were also sizable lifts for the exposed population receiving a new diagnosis of type 2 diabetes, starting a new treatment, and starting a treatment on the advertised brand. Similar lifts were seen in test taking and Rx treatment for visits to endocrinologists by the media exposed population, as shown in the Endocrinologist Visits tree on the right.
APPLICATION 2: MEDIA-SOURCE PREDICTIVE MODEL OF FUTURE CONVERSION
Methodology
Similar RWE data has also been applied to improve forecasting and determine which digital media publisher would most effectively drive new patients to fill a brand prescription. A leading type 2 diabetes brand ran a digital advertising campaign across eight media publishers, which in turn were divided into 30 placement groups for the purpose of optimization. The marketers wanted to predict which of the 30 media sources would generate the highest end-of-campaign, new-to-brand Rx conversion using mid-campaign information about the publishers and the audience quality data of the 30 media sources.
 
In the middle of these consumer digital campaigns, publisher/placement group-level regression models were built using the variables illustrated in Table 3.
Table 3. Regression Model Inputs
Placement Group Endemic Flag? Mid-Campaign Unique Reach Mid-Campaign Log (Unique Reach) Mid-Campaign Impression Frequency Mid-Campaign Treating in T2D Category Mid-Campaign Pre-Diabetes A1C Level Rate Mid-Campaign Diagnosed T2D Rate End Campaign Conversion-to-Brand Rate
Endemic Pub1 Desktop 1 123,027 5.09 5.2 12.40% 2.20% 18.40% 0.31%
Endemic Pub 1 Mobile 1 138,038 5.144.5 13.10% 3.00% 19.60% 0.24%
….
Programmatic 3 Demo Target 0 1,148,154 6.06 12.9 10.20% 1.60% 16.40% 0.17%
Programmatic 3 Behavior Target 0 1,318,257 6.12 21.9 10.80% 1.40% 14.30% 0.13%


Each row in the table represents a placement group within a media publisher that was serving digital display or online video media for this advertising campaign. Two placement groups for an endemic healthcare publisher and a non-endemic programmatic publisher are illustrated. The columns represent specific metrics for that placement group at the mid-campaign milestone; the rightmost column provides the percentage of consumers exposed to that media source that converted to the advertised brand by the end of the campaign. For example, the first row shows that the desktop audience of Endemic Publisher 1 had a unique reach of over 123,000 consumers (log = 5.09) and an exposure frequency of 5.2. By mid-campaign, 12.4% of the consumers exposed via this media source were already treating on Rx in the type 2 diabetes category, 2.2% of this audience had a prediabetes A1C level, and 18.4% were already diagnosed with type 2 diabetes. By the end of the campaign, 0.31% of exposed consumers for this media source converted to the advertised brand within three months after their first digital media exposure.
 
The objective of the modeling at the placement group level across all media sources was to determine a predictive relationship between the mid-campaign media and audience quality attributes, and the end-campaign conversion-to-brand. This relationship would enable media planners to optimize among the various sources mid-campaign to select those placement groups most likely to have high conversion rates to brand.
 
Two models were evaluated:
  • The first model used only publisher media delivery metrics and prescription history criteria of the media sources’ respective audiences.
  • The second model added additional RWE metrics, including the rate of taking the A1C lab test, uncontrolled threshold of the A1C level, and prior type 2 diabetes diagnosis.
Both models used a three-month lookahead rate of conversion to the advertised brand, modeled at the media placement group level. The two models were compared for goodness of fit as measured by an R-squared value. Performance was also evaluated for each model on a holdout sample of media sources.
Results
Comparison of the two predictive models (N=30) are summarized in Table 4.
Table 4: Impact of Real-World Evidence Metrics on Predicting End-Campaign Media Conversion
Baseline Predictive Model Enhanced Predictive Model
Inputs (mid-campaign) Treatment history on type 2 diabetes drugs
Media impression and frequency levels
Treatment history on type 2 diabetes drugs
Media impression and frequency levels
A1C lab test rate
Uncontrolled A1C rate Type 2 diabetes diagnosis rate
Outputs (end-campaign) Conversion to new to brand Rx Conversion to new to brand Rx
Goodness of Fit (R2) 0.56 0.69
Figure 2: Original Model Predictiveness of End-Campaign Media Population Conversion

Figure 3: Enhanced Model Predictiveness of End-Campaign Media Population Conversion with RWE Data

This statistical model predicted future media-unit level conversion at the end of the campaign with a high model fit of R2=0.69. This was a lift over the R2=0.56 for a model that only used media-level treatment history. Likewise, the model that leveraged RWE data had a closer fit to the holdout sample. Additional details for the two models are illustrated in Figures 2 and 3.
Key Learnings
The addition of RWE variables to the regression forecasting model resulted in a significant increase in the goodness of fit in predicting end-of-campaign media-level conversion to the advertised brand. They were not quite as important in the model as prior treatment in category or brand but did show a significant contribution. The implication here is that healthcare digital advertisers assessing media performance in an advertising campaign should consider in-depth audience quality metrics, including whether their media sources are reaching patients with prior diagnoses and lab tests above key threshold values.
Conclusions and Future Work
These analytics approaches to evaluating consumer media campaigns are a significant advancement over previous approaches to campaign measurement in three primary regards:
  • Direct linkage of media exposure to health behaviors, rather than surveys or website clicks.
  • Real-world evidence data utilized for actual patient health history and post-media outcomes.
  • Predictive power and frequent reporting for campaign optimization.
Using these insights, marketers have discovered the multi-faceted benefits of digital DTC campaigns, including a deeper understanding of how media exposure impacts patient behavior. They have also optimized these campaigns across different media publishers much faster and with more confidence. These studies are each based on the commercialized digital measurement platform called Crossix DIFA™  (“Digital Impact for Advertisers”).

The authors are in the process of automating the predictive model analysis of media sources to a production capability. Then, any digital media campaign can be evaluated at a media source level (across publishers or placement groups) to get an early read on which media source will generate the most new patient prescriptions for the brand.  

Similar approaches are available not only for consumer-focused media campaigns, but HCP-focused campaigns as well. Increasingly, pharmaceutical marketers are turning to non-personal promotion to complement sales force efforts. Much of this non-personal promotion is digital, with doctors, nurses, and other providers as the audience of the display, video, search, and email advertising. To meet this marketplace need, similar measurements have been developed for healthcare professional-focused digital campaigns that allow real-time optimization at the publisher level. Future research will investigate relationships underlying the interaction of media campaigns across both patients and healthcare professionals.
About the Authors
Ira Haimowitz, PhD, serves as Vice President, Product Strategy, at Crossix. Ira helps accelerate new product innovation for Crossix’s range of healthcare targeting and measurement solutions. Ira also enhances client delivery as a subject matter expert. He has more than 20 years of pharmaceutical (Pfizer and Organon), CPG, agency, and consulting expertise. Ira has published multiple articles, including authoring the book Healthcare Relationship Marketing. Ira received both his Ph.D. in Computer Science and B.S. degree in Mathematics from MIT. Ira is a Past-President (2006) and longtime board member of PMSA.

Whitney Kemper, serves as Senior Director, Analytics Products at Crossix Solutions, which he joined in 2009. He leads the development of Crossix’s analytical frameworks as well as the implementation of new technology and data. Whitney’s extensive experience with data and marketing analytics allows him to navigate the challenges of integrating and gleaning meaning from the wide array of healthcare and consumer data that Crossix employs. He holds a B.A. in Economics, English, and East Asian Studies from Vanderbilt University.
References
1 Risser R, Saha S. Deep dive on real world evidence. Presentation at: PMSA Symposium on Real-World Data. San Diego. 2017 Jan.
2 Kumar K, Parikh J. Value proposition of diagnostic test data. Journal of the Pharmaceutical Management Science Association. Spring 2017, pp. 23-31.
3 Mittler J, et al. What does Crossix’s latest expansion of connected health data actually mean for pharma marketers? PM360. 2016 Dec 20. 
4 Lee J. Pharma’s $6 billion annual ad spend targeted by media publishers. Medical Marketing and Media. 2017 Aug 21. 
5 Interactive Advertising Bureau. Automation and programmatic [Internet]. 2017. Available from: https://www.iab.com/guidelines/programmatic-rtb/.
6 Neuner C. It’s time for the healthcare industry to take advantage of programmatic. Medical Marketing and Media. 2017 Jan 22.
7 Haimowitz I. Five things you should know, but don’t know about digital campaign measurement. Presentation at: Digital Pharma East Conference. Philadelphia. 2016 Oct.
8 Haimowitz I. Embracing the right metrics for digital success. PM360. 2016 June 20.
9 Haimowitz I, McKinley M. Enhancing pharmaceutical digital campaign measurement using real-world evidence data. Presentation at: PMSA Annual Conference. 2017 Apr.