APPROACH
IPQ engages with its clients to establish a collaborative approach to its model-building
- Listen and understand the client’s questions and how the answers are intended to be used
- Perform root-cause analysis to probe the client’s questions and go beyond surface level issues
- Review and prioritize the questions to address
IPQ typically accomplishes this using a two-phase model, adapted to specific client requirements.
Phase I: In collaboration with “the client”, IPQ will work to help identify and quantify both the critical “unmet clinical (or commercial) needs” and go beyond that to identify and quantify the even more critical “unstated, unmet needs”. We term this “root cause analysis” and work with the client to define, refine and prioritize requirements to address these gaps. IPQ then deconvolutes these “needs” to identify critical concepts and relationships necessary to address them and adds these to its proprietary model of the patient journey, as needed. This is instantiated within IPQ’s novel knowledge graph and identification of critical data sources, public and private, are undertaken.
Client data can be incorporated into this process. Use of IPQ’s knowledge graph identifies additional key concepts and relationships which are reviewed with the client to establish the “disease model” that incorporates the complexity of the real-world patient population, the need for refinement and stratification of existing disease/diagnosis categories and critical factors, e.g. conflicting guidelines, standards of care, reimbursement, that may exist. Phase I is typically done as a pilot study.
- Example: initial question: in multiple sclerosis, the neurologist’s question focuses on identifying the best treatment option, i.e. drug, for their patient. Problem to be addressed: The potentially more critical question concerns the adequacy of the diagnosis, i.e. 85% of all patients have relapsing remitting disease as a diagnosis (RRMS), which results from “ruling out other conditions”. This diagnosis requires further stratification into disease subtypes that should also consider the 5 year patient journey (on average) to diagnosis, personalized clinical presentation and intervening treatments for symptoms to provide stratification of RRMS and better analysis of potential treatments.
Phase II: IPQ will establish a unique federated data model whereby data sources, identified in Phase I, will be linked while maintaining individual data provenance. This enables further linking to national level public data sources in compliance with HIPAA and GDPR requirements. Critical evaluation of specific data fields within each data base, in support of the disease model, will be carried out to establish context and quality/confidence in existing data. This is critical to support quality analysis including AI/ML approaches as data fields with identical labels commonly do not contain data collected or computed in the same manner across multiple data sources. This can be a significant source of uncertainty affecting the analysis results and their potential interpretation because of limited content of contextual data maintained in EHR’s. These qualification factors are maintained in a parallel database so that later analyses can refer to these assessments prior to selection of data cohorts, etc. Missing data, potential biases in data and conflicting data are identified and incorporated into addressing the client’s original and expanded questions. In this manner, both the answers and their specific confidence levels or constraints can be established and reviewed with the client for further action, e.g. further data acquisition, etc.
