Pharma Forecasting Evolution: From PMSA 2019 to M4

If you attended the PMSA 2019 conference, you may have seen the presentation by Srihari Jaganathan, associate director of advanced analytics at UBC, Inc., about forecasting patient persistency rates. Titled "Simple Probability Models for Predicting Aggregate or Sparse Data: An Empirical Analysis of Projecting Patient Persistency," it won the Best Podium Presentation award.

If you missed it, here is a recap of the presentation, along with new information about Sri's triumph at the M4 competition. As you may know, the purpose of M4 is to identify the most accurate forecasting method(s) for different types of predictions. These competitions have attracted great interest in both the academic literature and among practitioners, attracting hundreds of entrants from countries all over the world.

More on M4 later. (Spoiler: Sri killed it.)

First, a bit of background on forecasting patient persistency, from his PMSA presentation.

Long-term data on patient persistency is critical for constructing patient flow models in pharmaceutical forecasting. But there's a problem with that. Persistency rates are typically available only for short durations such as 12 months or 24 months due to availability of data. Moreover, the data sometimes are available only at the aggregate level. (Example: persistency rates by month and product.)

A much longer duration of persistency rates is required in the analysis of patient flow models. Typically projecting patient persistency is achieved by using simple curve fitting techniques in spreadsheets on aggregate persistency data. This would sometimes provide sub-optimal and irrational projections.

Lee, Fader and Hardie (LFH) (Foresight, Issue 8 Fall 2007) proposed very effective and simple probability-based approaches such as shifted Beta Geometric and Beta discrete Weibull models to project patient persistency rates.

The main objective is to empirically analyze LFH models on persistency data from diverse disease states such as diabetes, epilepsy, osteoporosis, immunology, hypercholesterolemia and hypertension.

Guiding Principles for Modeling Customer Retention (or Any Analytical Problem!)

Aggregate Data
Models should account for sparse or aggregate data:
  • Data are often not available or sparse
  • Data is available only at aggregate level
  • Cannot access individual data
Missing Data/Heterogeneity
Most important variables will be missing in real world data; a model should be able to account for it:
  • Example: Disease Severity in RA is an important predictor of Adherence but is missing in claims data
  • Modeling should account for unobserved heterogeneity
Simple and Replicable
Models should be understandable to those with legitimate interest and should be replicable:
  • Can be easily programmed in standard software such as R or Excel
Projecting long-term persistency is critical to understanding patient flow modeling. Persistency rates are typically available only for short durations, usually 12 or 24 months, due to availability of data. A much longer duration of persistency rates is required in the analysis of patient flow models, usually five- or 10-year horizons.

Often, projecting persistency is achieved by using simple curve-fitting techniques, but results can be sub-optimal and irrational. That's why probability-based models, like the Beta-Geometric model, for projecting persistency leads to better accuracy. Beta-Geometric fits the validation data like a glove and gives the best forecast.

But, Geometric models assume a sequence of independent binary trials. Many expect that there should be some type of positive momentum or inertia over time when people are more familiar with the medications they're taking.

That leads us to modify the Geometric distribution to a beta discrete Weibull distribution or a k-latent class discrete Weibull distribution.

Modeling Development Process
  • Use only aggregate monthly data.
  • Consider each patient's refill decision at each month, but try to to explain the decision. Individual factors are usually uncaptures in aggregate data and unknown in the forecasted periods. The model is a mathematical approximation of the behavior.
  • Recognize that difference exist among patients.
Probability-based models (BG, BdW, and LCdW) clearly outperform Excel based trend models.
  • Having an accurate forecast of persistency can allow companies to plan future strategies accordingly
  • When using these models, we can extract additional diagnostics about the characteristics of the patients’ behavior
Forward thinking:
  • Patient-level variables might not be the causal factors for persistency, but they remain of interest to the analyst to understand how they are associated with persistency
  • Future research will explore the possibility of incorporating these observable covariates in the process of modeling

A Report from M4

The purpose of the annual M4 competition is to identify the most accurate forecasting method(s) for different types of predictions. These competitions have attracted great interest in both the academic literature and among practitioners, attracting hundreds of entrants from countries all over the world.

Several researchers have shown that combination-based forecasting methods are very effective in real world settings. Sri aimed to evaluate the effect, if any, of combination-based methods on the accuracy of forecasting when using both statistical and machine learning-based approaches. He proposed two types of combination-based forecasting approaches: evidence-based and optimization-based. He found that simple combinations of forecasting models performed competitively well.

Until now, people have been using spreadsheets to forecast persistency rates. For M4, Sri took a model that's not in healthcare domain and translated it to healthcare. He empirically showed that this model provides more accuracy in forecasting. Why is it important? Even a small shift in accuracy has huge implications. If one person misses one unit of medication per day, multiply that by 1,000 people on the same medication and that's huge.

M4 is a cutting edge leading competition attracting 200 entrants from 17 countries. Fifty of those entrants finished. Sri came in 2nd on monthly forecasting and 4th overall.

The software Sri created for this model, Foretell, has been downloaded some 4,000 times. Download it here.

Machine Learning Techniques Deliver Granular Insights to Enable Improved HCP Marketing

Companies have turned to analytics to help them drive more effective campaigns, and to rules-based programs to help reps decide when to execute different tactics. The challenge with most measurement programs is that they suffer from aggregation bias, washing out granularity as they estimate the overall response to a given tactic. They also can be challenged to provide critical insights around differences across specialties and geographies. When we look at rules-based execution aids, they are inflexible and do not tie to changing responses over time.

At the recent PMSA conference, Jane Portman, VP Health Analytics at Merkle, and Brian Demitros, VP of Analytics at Merkle, delivered a presentation designed to educate conference attendees on what's being done out there in terms of using machine learning and analytics to track the behavior of HCPs in response to past and current marketing efforts in order to predict future behavior, and also to gauge the efficacy of those efforts.

The goal for marketers is to find the optimal solution to evaluate and optimize the performance of new digital tools and a new addressable marketing ecosystem designed to provide continued support to sales reps within a multi-channel customer experience. Traditional measurement would not be able to get the level of insights that marketers need. We took a page out of our Digital Media Measurement practice to apply supervised machine learning techniques to the challenge to rapidly and consistently deliver granular insights to enable improved sales force execution and optimal multi-channel customer experience.

Why is it important? Because pharma marketers are facing common challenges in the industry:
  • Customers are saturated with promotional messages.
  • Outbound promotional engagement and access is declining.
  • Customers have channel preferences and propensity.
  • Value-add content and beyond-the-pill programs spark interest and brand engagement.
One key to addressing those challenges is to use a "marketing ecosystem" of various digital tools, including:
  • Digital leave behind
  • Targeted banners
  • Interactive sales aids
  • On-demand content
  • Virtual portals
  • Alerts/SMS
  • Rep-triggered EM
These digital tools require new methods to optimize the customer experience for multi-channel effectiveness, interaction effects of numerous tactics, and content-level optimization.

Meanwhile, existing or traditional marketing methods and approaches have their shortcomings, which machine learning can solve. The problem is, pharma has been slow to adopt it. According to 2018 data by Burtch Works LLC, healthcare and pharma has a 60% rate of machine learning use, which is lower than financial services, gaming, consulting, corporate, retail, general advertising and marketing, and technology. Why? Much of the reason involves pharma industry constraints, like medical legal approval, lack of CTAs or offers, limited content variation, and low campaign activity.

Key takeaways from the presentation?

Pharma marketing has changed to become more addressable. Given constraints in the industry, the analytics hasn't evolved at the same rate, but there's an opportunity in taking learnings and best practices from other industries and applying them to pharma. In the end, it's about reaching each individual HCP in the right way, at the right time, with the right message. Using machine learning, that goal is easier to achieve.

Targeting and Segmentation in Specialty Pharmaceuticals with Calibrated Messaging

At the PMSA 2019 conference, Mert Sahin, CMO of GE Healthcare and Ashish Patel from CareSet Systems, presented a session dealing with targeting and segmentation in specialty pharmaceuticals with calibrated messaging.

With capped resources for sales teams, modern sales force sizing and alignment decisions must artfully blend subjective knowledge captured from field experts with objective measurement gleaned from improved access to big data. Among the most critical exercises in planning is building a list of physician and account targets, leading to the prioritization challenge. How can sales management optimize both the physician’s and the field representative's time?

Here are some highlights:
  1. Among a vast portfolio of products, GE Healthcare manufactures injectable pharmaceuticals used to diagnose neurological disorders. Like other pharmaceutical firms, those diagnostic injectables are “buy and bill,” and face the same challenges, including:
    • Medicare datasets have no blackout markets or accounts.
    • Neurology, Cardiology, Oncology, Immunology and conditions prevalent in 65+ age group.
    • No proprietary identifiers. Physician/hospital NPI is backward/forward compatible.
    • First to obtain data from CMS, quarterly, with quarter lag.
    • Generating and publishing CMS PUFs like teaming.

  2. Market access can leverage affiliations by building data-driven plans and tactics:
    • Identify local key opinion leaders (KOLs) who utilize the brand.
    • Expose all organizational affiliations between the target and hospitals, medical groups and other physicians.
    • Invite KOL to local payer account conversations.

  3. Three targeting strategies:
    • Mass marketing: Targets the whole market and ignores segments. The products focus on common customer needs.
    • Segmented: Targets several market segments within the same market, products are designed and targeted at each segment, requires separate marketing plans.
    • Concentrated: Focuses on smaller niche segments or customers, aims to achieve a strong fit within niche.

  4. Putting it all together: How do we use all of the data collected?
    • Capture ROI with your favorite CRM.
    • Maximize ROI with natural language generator.
    • Dynamic content enables niche messaging at scale.
    • Contextualize each target for the representatives.
    • Hyper-localize messaging for market access.

    It is a concrete example in the field of neurology, revealing a method and segmentation framework to help organize physicians to optimal lead allotments. Results and impacts expected include a demonstration of targeting refinement to build messaging, KOL identification by local territory, and an example of how KOLs can power tactical market access in active negotiation with local payers.

Watch Out for the Newbie: Utilizing Real World Data to Develop Leading KPIs

Biologics is the fastest-growing sector and driving innovation in the industry. Several biologics launch every year with unique attributes and it's a crucial task for other biologics companies to measure the initial impact of these new products in the market and develop a quick strategy to maintain and increase their market share. This was the topic of a popular session at the recent PMSA conference. Harpreet Singh and Jiaran Wang gave a presentation, Watch Out for the Newbie: Utilizing Real World Data to Develop Leading KPIs in Prep for a New Product Launch. It had a lot of people talking at the conference. In case you missed it, here's a rundown of the highlights, from Singh.

Novel drug approvals are increasing each year. Significantly.

There's been a 20% annual growth in novel drug approvals from 2008 to 2018, from 24 new drugs approved in 2008 to 59 last year. The result? Among other things, a more competitive landscape.

With many more entrants into an already crowded market, incumbent teams need to build a responsive strategy to ensure that their product's value proposition to patients, providers and payers is not lost. There's a small window to monitor and modify actions or your market share is going to blow away.

So, how do you build a responsive strategy? The name of the game is tracking the right KPIs. That's going to be critical to ensuring your success.

Most of the KPIs are at the national level where we may not see a real impact in the initial months, so it’s important to develop leading KPIs to get an early look and impact in the market, especially for sales and marketing teams.

It's about:
  • Understanding the impact of the entrant to your business.
  • Understanding the impact and trend of market events like regulatory changes, and changes in the access landscape for entrants and incumbents alike.
  • Leveraging learnings for future launches.
  • Proactively mitigating risks and quickly responding to challenges.
  • Revealing new opportunities and areas for improvement.
Real-world data is rich in information and traditionally used more for medical/HEOR studies and patient journey analysis.

The upshot? It's all about leveraging the right KPIs.

Leveraging insightful leading and lagging KPIs to build a responsive strategy

Here's a hypothetical scenario to illustrate what we mean by that:

New Drug, launched last month, indicated for conditions A and C.
Brand X, the market leader, indicated for conditions A, B and C, launched 7-9 years back and has 40% of the market share.
Brand Y, indicated for conditions A, B and C, launched 7-9 years back and has 32% of the market share.
Brand Z indicated for conditions A, B and C, launched 3-5 years back and has 28% of the market share.

Traditional KPIs are usually based on sales data, in terms of total prescriptions or number of units sold. So as we watch New Drug's performance over a three-month period compared to Brands X, Y and Z, we see a sharp upward trajectory — maybe people are talking about it, maybe there's buzz — and a slight decline for X, Y and Z. But we need more than just that. We need insights before making any strategic decisions.

It means creating indication-based KPIs based on patient diagnosis history. Only then can we measure the true uptake. You can see which patients with which conditions are taking the new drug, and create action plans in response.

With the real-world data, we can identify an important segment of the market and profile these patients and doctors who may be an early adopter and potential future advocate of the launch drug. Looking at the switching patterns, days on therapy and add-on drugs, the patient journey can provide crucial actionable insights.