Introduction

In a previous paper we discussed how Tudor Health data is being used to create synthetic data and Market Emulation Models (MEMs). To demonstrate some of the ways that MEMs are changing the nature of commercial insights generation and application, we have prepared these summary case studies.

The novelty of the MEMs approach means that the case studies must remain anonymous, but we have tried to setup the background and describe the details as much as possible so that you can see the impact of the approach. We have case studies in all the following categories:

1. Forecasting
– MEMs as a supplement or alternative to analogues.
– Integration and augmentation of existing forecasts.
– Scenario planning

2. Physician targeting
– In the USA
– Outside the USA

3. New product tracking
– Augmenting the ATU
– Rare disease first product

4. Market research and insights generation
– Application of projective techniques (choice models)
– Improved sample and questionnaire design

 

1. Forecasting

A key element of all MEMs based on longitudinal patient data is that they include details of the evolution of treatment practices in the target disease area. This characteristic allows for a number of forecasting applications – here are three:

 

1.1 An MEM used as part of an analogue-based forecast.

This client had acquired patient level longitudinal data to identify the diagnostic and treatment journey and decided to use the data to build a MEM that would give them a more complete market view. At the same time, they were updating their forecasts for the market and product, using their normal analogue approach.

They had selected three analogues; one which supported their market forecast, one used for their brand, and one for two products in the market that were losing exclusivity.

In the process of using their MEM to explore market characteristics we identified that there were two very distinct HCP segments prescribing their drug. Each HCP segment demonstrated quite different characteristics. It soon became clear that single analogue would not be optimal for forecasting both segments. By applying the MEM analysis to analogue selection, we were able to identify a different analogue to use for each of the two HCP segments, using the MEM to quantify the correct proportion of market that each applied to.

Internal confidence in the forecast went up as both projected and actual forecast accuracy have been shown to be more reliable than the previous analogue forecast.

1.3 Scenario planning using a MEM

This client has a Monte-Carlo forecast built on a substantial amount of secondary market data and most major assumptions are driven by primary market research. However, like most Monte Carlo forecasts, the scenarios were based on changes to the values of scenario input values and the distribution values for those assumptions were not modified.

After acquiring patient-level longitudinal data the client built a MEM and one of the goals was to improve forecast scenario planning.

The MEM demonstrated that two of the critical assumptions in the forecast had different distribution of values depending on the specific value in use (for example, one of the variables had a distribution that skewed to the right when set at a higher value, but was slightly skewed left when the value was below a certain threshold – all of which was identified from enhanced analysis of the MEM data and tested in the market).

The effect of this has been to modify the ways that scenarios are run on this forecast, where the MEM is used to both check the ranges for core assumptions and establish the most appropriate distribution for those values when running the Monte Carlo. In this way the availability of a MEM has improved forecast reliability as well as scenario flexibility for the commercial team.

 

2. MEM-based physician targeting

Physician targeting has been one of the most important features of commercial analytics for decades. Substantial efforts (and $$) are expended every year to improve the accuracy of this targeting – especially in the USA where doctor-level data are readily available.

These two case studies describe how a MEM was foundational in a targeting process for a company in the USA and how using MEMs made it possible to bring US-style targeting to the EU.

 

2.1. In the USA

This has been the most complex application of a MEM that we have seen so far (albeit in a short history of using MEMs built on longitudinal patient data).

HCP targeting in the USA is a highly sophisticated process, including the use of near- universal physician claims data that permits levels of analysis only dreamed of outside the USA. This said, there are disease areas where claims data have gaps that make the targeting less accurate, specifically those areas where specific test results forma core element of the target patient definition.

In this case the target patients started with one very broadly seen diagnosis but actually had a rare sub-condition that was largely determined by the presence of a specific genotype. Claims data showed that a genetic test had been completed but the results were not widely available and so the claims-based targeting system was insufficiently accurate.

A MEM was built using patient-level longitudinal data that included full coverage of the specific genotype (this is achieved by the patient data be EMR-equivalent data, enabling the extraction of structured data that normally reside in the unstructured physician notes). With this MEM in place the claims data were mapped using a modelling approach that created synthetic claims data from the MEM (what would the claims look like for this patient?) and then mapped to the actual claims data for the broad patient population.

Three things were found:
– The MEM-based synthetic claims data were an accurate model of the actual claims data.
– The target sub-cohort of patients were identified in the broad claims data.
– Target HCP lists were created that increased the number of target physicians that have relevant patients and improved the impact sales force and other marketing activities (the revised method is now being integrated into the omni-channel system at this client).
This on-going programme has modified the targeting approach and improved salesforce performance.

 

2.2 Outside the USA (EU)

Named physician data are not available for commercial purposes in the EU (or UK). Whilst salesforce and marketing targeting are conducted in the EU, they must be done using deidentified methods.

By building a MEM and integrating it with open-source data we have built an institutional model of a market that identifies the institutions most likely to have higher volumes of a target patient group (in this case a rare disease).

The method relies on the underlying data capture method for longitudinal patient data that works directly with HCPs. Naming physicians is not permitted; but linking an anonymised patient to a specific institution is permitted so long as the patient is one of a cohort. Once these individual patient data have been integrated to create a synthetic dataset that is a MEM, we are able to project patient volume at the institutional level.

This is what we have done, and the effect has been to identify institutions that were not previously on the client target list. This is not as good as named physician targeting, but it is a very good surrogate that the client is now exploring for other countries and disease areas.

 

3. MEMs in New Product Tracking

The ATU is still the most dominate methodology applied to the tracking of launch activities. It is a tried and trusted method, and the first case looks at how an MEM has added value to an existing ATU programme.

The second case explores how a brand in a non-competitive environment (orphan drug in a rare disease) was very well tracked with significant additional insights and analytical values through using an MEM.

 

3.1 Using a MEM to augment an existing ATU

This case study started with a client building a MEM based on longitudinal patient data for market understanding and patient modelling. Their product launch was monitored using a conventional ATU.

Post-launch the updated MEM data showed a divergence in performance among HCPs that was not fully clear in the ATU. Examination of the MEM data suggested that a modification to the ATU would be appropriate, in effect increasing the granularity of one of the “likelihood to use” parameters.

The resulting increase in detail in the ATU has since been shown to accurately reflect the change in behaviour seen in the updated MEM data and the MEM and ATU data are now being presented in parallel.

 

3.2 Rare disease product launch tracked with an MEM

The ATU methodology is optimised for markets where there is a directly competitive product choice available to HCPs. This is not the case in rare diseases where there is an orphan drug. Whilst off-label products will be used to treat patients, the approved orphan drug is, by definition, the only directly labelled product for these patients and so the ATU method is not optimal.

In this case the client had longitudinal patient data that had been acquired for the purpose of completing market definition, dynamics, and structure. Leading up to launch it was clear that a MEM would be a powerful tool for the detailed launch planning as well as post-launch monitoring. A MEM was built using the existing and updated patient data.

The MEM was used to refine launch planning, including targeting and forecasting. Post-launch the use of the MEM transitioned into launch tracking, with different versions of the MEM created using different assumptions of likely HCP behaviour. This has proven to be an effective method of adjusting the post-launch activities, including implementing a patient communication programme that was not previously expected.

 

4. MEMs in market research and insights generation

MEMs are built using synthetic data. Based on a core of real patient-level data that are modelled to create the MEM, MEMs can have different versions depending on the way the raw data are modelled. When building a MEM, it is normal to test different methods in order to create a synthetic dataset that reflects as closely as possible the actual patient population and known parameters, but these are models and can therefore exist in different forms.

As such MEMs are being used in simulation exercises, like the application of choice models.

Also, MEMs emulate the total market and can identify detailed characteristics that influence the design of market research samples and the questions used in primary market research.

 

4.1 Applying a choice model using a MEM

Longitudinal patient data (the core of MEMs) includes aspects of the market that are dependent on time. In this case, HCPs were very loyal to their previous prescribing decision until the patient relapsed. At this point HCPs reconsidered their prescribing and, in some cases, would switch the patient from one brand to another. In effect, there was a switching gate keeper function attached to relapses.

The patient data had shown that the patients would experience a relapse once every 15 months but this average concealed that some would never relapse whilst others were experiencing a relapse every 4 months.

In addition to having a MEM built for the market, the client completed primary market research that used a brand choice model the output of which was preference share based on patient segment and opportunity to switch. The opportunity to switch was the relapse, and relapses varied by patient cohort.

It was therefore possible to build a version of the MEM that included different brand choice models for future behaviour based on the different scenarios of relapse rate, alternative product profiles and, as a result, arrive at a set of detailed scenario output options that could be used as input to the product planning process.

 

4.2 Improving sample and questionnaire design

This was perhaps our most straightforward MEM application to-date. Using details from the MEM for an established market it was possible to identify sampling elements that would not have previously been considered. Specifically, the MEM showed that there was not one specialist involved in treating, but two where the second specialty represented 15% of the patient prescribing decisions.

This was a very unexpected result and the primary research sample was adjusted to allow for a detailed exploration of the second specialty.

A similar situation occurred in another case where an existing MEM identified that target patients had a more complex comorbidity profile than expected. As a result a planned primary research project had a small set of comorbidity questions expanded to explore the greater complexity found through the MEM.  

 

In Conclusion

Tudor Health’s vision in becoming a leading provider of individual-level data through synthetic data and Market Emulation Models represents a bold step toward modernising commercial data utilization in biopharma. These cases illustrate some of the ways in which these datasets and MEMs have been utilised – so far. As the industry increasingly embraces non-traditional methodologies, the potential for innovation and competitive advantage becomes clear. Synthetic data, in conjunction with MEMs, has demonstrated the power to improve biopharma commercial decision- making and provide life science companies with a strategic edge in a data-driven world.