How real-world evidence, AI will enable better care

Advances in analytics conducted on data in EHRs, linked to insurance claims databases, will help improve therapies.

During the prescribing process, a physician’s ability to accurately and reliably match the most appropriate therapy option with the patient at hand is limited by the amount and quality of clinical information that is available.

While randomized clinical trials are considered the gold standard for establishing the safety and efficacy of prescription medications, and the clinical findings verified from the trial inform the label on the approved therapy, the overall utility of such trials is limited by a variety of factors.

One particular limitation is the use of strict inclusion and exclusion criteria during trial enrollment. While the resulting homogeneous patient population provides some benefits for researchers during the initial trial, the clinical findings are not broadly applicable to patient subgroups who were not represented in the trial, or to patients who are managing one or more chronic conditions or comorbidities. Thus, additional studies are needed to further refine the medication’s full prescribing information — and broaden physicians’ understanding of the therapy — over time after it has become commercialized.


In recent decades, the gap between randomized clinical trial findings and real-world prescribing conditions has widened, because most of today’s chronic diseases have multiple approved therapies available, and many adults are managing one or more chronic conditions or comorbidities.

Studies that use advanced techniques to develop real-world evidence (RWE) based on routinely collected data, in both electronic health records and insurance claims databases, can help to bridge this gap. But such studies must use state-of-the-art modeling and data-analytics techniques if they are to produce data-driven insights that are accurate and reliable enough to inform treatment.

In pursuit of evidence-based medicine
To support evidence-based medicine, ongoing studies are routinely carried out — after the drug has entered the marketplace and is being prescribed to patients on a routine basis. Such studies involve complex analysis of real-world data and aim to help stakeholders throughout the pharma and healthcare arenas to develop broader and deeper clinical findings related to how a given therapy performs in specific patient sub-populations over time.

These types of studies are critical, as the findings allow the medical community to better understand the clinical and safety profile of the medication across a much broader and more diverse patient population (compared to those enrolled in the trial). Given time and resource constraints, it is simply not possible or practical for any randomized clinical trial to study every possible combination of therapy, on every patient sub-population, while also factoring an infinite combination of co-morbidities.

Instead, physicians are left to rely on their best judgment when prescribing therapies for patients who do not match the criteria of those enrolled in the actual trial. These studies help physicians to select the most appropriate therapy option from a crowded therapeutic class for specific patients who may be managing multiple chronic conditions or comorbidities. But well-designed RWE studies can help to bridge the information gaps and provide relevant clinical insights to inform prescribing within specific disease states, therapy classes, and patient populations.

Ensuring that the RWD is accurate for secondary use
Fortunately, ongoing advances in two data-analytics techniques — subgroup analytics and comparative effectiveness — enable in-depth clinical studies to be designed and carried out using real-world data that is collected routinely in EHR systems and insurance claims databases and using the study findings to create RWE that is valid and accurate enough to inform treatment.

The ability to accurately and reliably leverage the massive amount of data that is routinely collected throughout the healthcare arena and use it to amplify the clinical findings of the randomized clinical trial creates enormous opportunities for improved prescribing and improved health outcomes for patients. The resulting insights create direct patient benefit as well as financial dividends for the healthcare system and the payer community by improving care.

The higher the validity of an RWE study, the more appropriate it is to modify care and tailor therapy based on its findings. To achieve high validity, explicit effort and expertise is needed to design studies that will provide the required accuracy and generalizability, and to devise ways to address some of the shortcomings that are inherent in today’s available real-world data.

While mountains of data are collected every day in the EHR systems and claims databases that are a ubiquitous part of in today’s healthcare experience, much of the information available not necessarily accurate enough for secondary use. For instance, the quality and accuracy of EHR data is sufficient to capture details of a patient’s encounter with his or her physician, and claims data is sufficient to determine whether the encounter should be reimbursed by the health plan. But the information may be neither accurate nor clean enough for secondary use in a clinical study unless additional effort is undertaken at the outset of the study.

Using a study-design paradigm that combines deep phenotyping with linked outcomes, stakeholders can develop advanced RWE-based studies that yield clinically relevant findings to inform prescribing decisions. Data related to a specific patient phenotype can be found in the EHR system, as that source is rich in both structured data in the patient’s chart and unstructured narrative information (such as notes and comments the physician has added to the patient’s chart) related to the patient’s ongoing health journey.

However, EHR data is often incomplete and is rarely consistently detailed enough for study purposes — for instance, when a patient sees a physician or specialist who is in a different practice that is not affiliated with the patient’s main EHR system or has an emergency department or hospital encounter that produces clinical or treatment data that is also not married into the patient’s EHR system.

To help bridge this gap, additional information on health outcomes can be found in the patient’s insurance claims or billing record. Such databases provide a rich source of information for all healthcare-related incidents and interventions the patient has experienced that were submitted to the health plan for reimbursement. By linking outcomes data found in claims databases with deep phenotyping information found in EHR data, today’s advanced RWE studies are able to yield high-validity, highly accurate clinical findings.

What’s at stake
Studies based on advanced RWE enable better clinical care and enable physicians to prescribe the most appropriate therapy among several competing options that may be available in the same therapy class — based on improved knowledge of how that therapy performs in different patient subgroups that may have not been adequately represented in the underlying trial. The ability to reliably connect rich phenotyping information that can be found in the EHR data, with additional outcomes-related insights that can be found in the claims database, creates many opportunities to provide more tailored prescribing and improve overall health outcomes.

But this design and execution process brings a variety of data-analytics challenges in terms of processing, accuracy determination, alignment, privacy and data security. To address some of these needs, artificial intelligence (AI) can be applied, to extract the relevant information needed for the clinical study and create the most appropriate subset of patient records that meets the RWE study criteria. Then, the model can run small-scale tests to confirm the accuracy of the protocol standards before the larger-scale study is conducted.

This process allows the study designers to access the most relevant and accurate data — specifically assembling data that is “fit for purpose.” For example, if a given study protocol calls for 80 percent accuracy, the use of deep phenotyping to collate real-world data from the most appropriate patients (those that meet the study criteria) would allow for improved overall accuracy. Marrying this more robust source data with linked outcomes data from the claims database, then using an AI model augmented by natural language processing techniques, enables a strong feedback loop that verifies the required level of accuracy.

When advanced RWE studies can demonstrate that a specific therapy has a higher effectiveness in a particular patient sub-group, patients in that subgroup have access to that option as the standard of care. The approach described here — broader use of advanced RWE studies that are explicitly designed to include deep phenotyping and linked outcomes using AI to enable comparative effectiveness — provides a path forward in healthcare.

This approach creates not just an incremental improvement in healthcare but a disruptive change for the better to inform the standard of care for patients and physicians to improve prescribing. Doing so ensures the best treatment possible, minimizes disease progression, and promotes optimal health outcomes. Patients deserve nothing less, and the approach has broader business and financial implications as well, because deeper clinical insights based on actual treatment data support helps to inform drug-pricing, formulary-placement, and reimbursement decision making.

Dan Riskin is the founder and CEO of Verantos, which specializes in providing high-validity, real-world evidence for clinical, regulatory and reimbursement purposes.