Forecasting: Patient flow modeling comes of age

Judith Kulich, associate principal, and Emily Jin, manager, at ZS Associates talk to eyeforpharma about how patient flow modeling can inform forecasting

Judith Kulich, associate principal, and Emily Jin, manager, at ZS Associates talk to eyeforpharma about how patient flow modeling can inform forecasting

Patient flow, or system dynamics, forecasting is a common approach for modeling situations in which patient potential is affected by movement between disease states.

The technique is used frequently in oncology, diabetes, RA, and other markets with lines of therapy dynamics.

(For more background on patient flow models, see Patient flow analysis and forecasting.)


E4P: How do you assess the maturity level of patient flow modeling, which in oncology and HIV, for instance, is related to the therapeutic history of patients and survival characteristics? How do you assess the maturity level for therapies and diseases that involve different drug courses based on the severity of disease? 

ZS: There have been significant improvements in recent years with increasing availability of patient-level data in many markets. With monthly granularity, patient flow models can incorporate oncology-specific dynamics, such as survival rates and relapse rates based on the types of data that are available from clinical trials. Similarly, line of therapy models can be developed based on severity (e.g., patient progression) and previous regimens (e.g., in markets with add/switch dynamics or failure-dependent indications). Overall, the maturity level of patient flow models has less to do with technical capabilities and is really driven by availability of data sources that meet the level of granularity needed to develop robust input assumptions. 


How are patient flow models being utilized to assess future patient segments in shifting patient populations? Can you provide an insight into techniques, variables, and also patients characteristics, like age? 

Patient flow models can capture movement at the patient segment level and/or between segments (e.g., through aging, increasing severity, etc.). In contrast, agent-based modeling offers the advantage of capturing these dynamics at the individual level for greater granularity, as well as incorporating patient-specific events (e.g., acute hospital-based events).


What is most critical when it comes to working on patient treatment pathways in patient flow modeling? Which variables need to be considered for optimal planning and forecasting? How can one ensure the segments are identified in an appropriate manner and that they optimally contribute to forecasting? 

Trying to model every single dynamic can present challenges in the complexity of the model and generating assumptions for every single factor that affects the decision process. With this in mind, it is critical to find the right balance between accuracy/granularity vs. transparency and confidence in the results. While it may often seem that every nuance needs to be incorporated in a model, ultimately more inputs implies more assumptions, which can imply greater uncertainty in the outputs. 

Ultimately, the model design and decisions should be based on forecasting needs and strategic considerations specific to the product. An important consideration to remember is that the patient flow model and the forecasting model do not need to be identicala patient flow model can be used to evaluate market dynamics and scenarios, while the forecasting model provides annual revenues.


Oncology is described as probably the most complex healthcare market. How do you assess this complex scenario from a modeling perspective?

Generally, patient flow modeling (and even forecasting) will be specific to one tumor type (e.g., lung, breast, hematological, etc.), to reduce complexity. This also allows flexibility to capture the nuances of these different markets across epidemiology dynamics and competitive activity. For products with multiple tumor types, the forecaster can still forecast separately for each tumor type, then aggregate revenues.


There is a need for a forecasting model that can help view oncology from a single perspective. How can this need be addressed? 

Across an organization, it is not unreasonable for different business perspectives to drive different modelseach model with its own level of granularity based on the business questions being addressed. Trying to meet every need in one model can generate an unnecessary level of complexity. For example, we typically recommend using a patient flow model to understand market dynamics and evaluate strategic or promotional scenarios, while using a traditional patient-based forecasting approach and model for revenue generation. One model can be used to inform the other model to drive consistency in assumptions and market potential, for example, deriving drug-treated patients by line of therapy.


What is the best way to account for the complex ways the patient source develops and also to introduce flexibility to accommodate changes in the eligible population? 

Every patient flow model is typically customized to the therapeutic area, market dynamics, product profile, broader strategy, and specific business questions. Typically, models are designed with flexibility for inputs to be updated frequently as new data becomes available. This reflects the forecasting philosophy that forecasts are ongoing, living documents used to guide decision-making prior to launch, but with the expectation of being updated as new information becomes available.  


Can you highlight best practices for patient flow forecasting, use of patient-level data, and innovations with agent-based modeling that support a transparent, manageable, and credible patient flow model? 

The most important point to keep in mind for both patient flow and agent-based modeling is to design the model to address the business questions at hand, based on available data. Its similar for general patient-based forecast: The outputs are only as robust and defensible as the inputs.

Learn more about patient flow modeling at the Pharma Forecasting Excellence Summit in Boston from October 5-6, 2010.

Since you're here...
... and value our content, you should sign-up to our newsletter. Sign up here

comments powered by Disqus