Data Fidelity and Latency: All things Clinical

Introduction

Data fidelity is the accuracy with which data quantifies and embodies the characteristics of the source. For example, stock ticker data may have a different level of accuracy and refresh rate at a Bloomberg terminal compared to an online web trading station. Signal transmission systems have been dealing with data fidelity challenges for ages. However, data fidelity has now been applied in areas such as cybersecurity with respect to the granularity of event data captured to deal with intrusion detection systems. Fidelity can also be defined as a function of data detailing granularity and accuracy.

Data fidelity in healthcare

This article outlines data fidelity and latency concepts in U.S. healthcare. A primary quality requirement for healthcare data is that it embodies all the required principal components to address expected goals to help streamline processes. Focusing on data fidelity and latency, with respect to clinical systems across different healthcare verticals, can assist in correlating independent factors and quantifying the complexity while ascribing to a particular healthcare vertical.

As we go through the fidelity and latency requirements for various applications of data, it becomes important to assess the data models and the skills required to build the systems. Let's begin by illustrating the categories of healthcare verticals with respect to fidelity and latency and diving into each.

This graph demonstrates the position of healthcare verticals from a fidelity and latency perspective.

EHRs

As providers interact with their EHRs, their systems need to provide data with very high fidelity to record the heart rate or blood pressure, and it needs to be qualitative and time-sensitive. Also such data needs to be refreshed often in response to any changes.

Such systems are extremely complex to build and take time to mature in their capabilities. The resources or entities at the data model level need to capture high fidelity data at a very fine granularity.

EHR attributes:

1. Extremely high fidelity
2. Very low latency and time-sensitive
3. Highest skill set requirements to build/maintain
4. Finest granularity in the data model
5. Slowly Changing Dimension (SCD) level 6 

Clinical Decision Support Systems

Like EHRs, clinical decision support systems (CDSS) have a lot of patient specificity. The intent across CDSS for acute and ambulatory care is to improve overall patient care. Acute is slightly more on fidelity and lesser on latency. An example may be an anemia work up in an outpatient setting versus a cardiovascular surgery procedure, but both check on platelet counts. Expectations are high for the utility of CDSS in acute care because acute care in hospitals and emergency rooms is the most intensive and expensive part of the healthcare system on a per patient basis.

Another example is in the onset of sepsis. Every hour that passes before sepsis is diagnosed and treated, the risk of mortality increases by 7.6 percent. Real time and contextual information is critical here for deeming a low latency requirement.

CDSS attributes:

1. High fidelity (inpatient slightly higher)
2. Low latency
3. Very high skill set requirements to build/maintain, real time and contextual
4. Finer granularity in the data model
5. SCD level 3 out of 6

Clinical operations and analytics

Clinical operations in hospitals foster an environment that is patient friendly and operationally consistent. Clinical operations software deals with issues pertaining to data capture and data reconciliation, but with a much lower fidelity than CDSS and EHRs. The reduction is acceptable as the software reduces clinical variability (uniform care) and enables a productive environment of care. Clinical analytics help organizations get ahead of demand while maintaining or improving patient outcomes and satisfaction. It helps manage the variation in patient demand enabling better planning and resource usage.

Clinical operations and analytics data enables success in the following areas:

1. Appropriate, timely care
2. A collaborative, goal-oriented approach to patient-centered care with a direct impact on a health system’s bottom line
3. Analyze the opportunity for improvement and identify the operational problem
4. Design and implement interventions and measure results.
5. Continuous process improvement

Clinical operations/analytics attributes:

1. High fidelity but lower than EHR/CDSS
2. Relaxed latency requirements than EHR/CDSS
3. High skill set required to replicate and version the data
4. Medium granularity required in the data model
5. SCD level 3 will suffice

Clinical analytics fall somewhere between revenue cycle operations (RCM) and clinical operations. It may analyze patterns around admission rates based on certain diagnosis to use real-time medical data to generate insights or make decisions. Fidelity-wise, it’s closer to EHR data than RCM data. 

Revenue Cycle Management operations and analytics

RCM operations place emphasis on claims, collections, and contract management with a laser focus on the dollar amount for net collections and zero account receivable (A/R) days. RCM analytics have much lower fidelity as its focus is more on coding productivity and billing operations to reduce average A/R days, rather than clinical activities. Data still needs to be highly relevant, but not as time-sensitive. 

The data needs to be of the highest quality as it enables all contract, collection, and claim management; It is necessary for overall process streamlining and revenue realization. RCM locates lost revenue opportunities where a provider is administering a service and not capturing it, or not getting paid for it. The fidelity/latency chasm between RCM operations and analytics does merit each having their own details. 

RCM operations:

1. Process of streaming the data to the clearinghouse/payers and realizing the contract, claims, and collections workflows is important to keep it accurate to the dollar and it is highly time sensitive, deeming it high fidelity and low latency
2. Keeping in sync with payer rosters and clearinghouse metadata, helps to keep workflows moving, and has low latency requirements
3. Adoption of alternative payment models has made coding and reimbursement much more complex than fee-for-service revenue cycle resulting in the added emphasis on operations. 

RCM analytics:

1. Analyzing the coding/collections productivity or the A/R aging reports does not need very high fidelity and slightly higher latency is sufficient
2. Also determining days to bill, or a transaction detail report for types of transactions based on financial class or transaction type, requires higher latency data.

RCM operations/analytics attributes:

1. Medium fidelity with operations ahead of analytics
2. Latency is higher than other categories with more operations analytics
3. Skill set required to replicate and author the data for proper management of workflows
4. Medium granularity required in the data model, denormalized tables are acceptable
5. SCD level 2 will suffice

Clinical trials and research

The data collection from clinical trials and research happen over a period of time as part of experiments required for research. This data needs to be relatively clean, however fidelity should be maintained well within the prescribed time period. Clinical trial data stemming from trials performed for drug discovery is furthest away from the EHR on the basis of latency; however, the fidelity is much closer, as the data still needs to be clinically relevant and statistically significant to establish the efficacy of the drug.

Though the latency may be much greater to establish drug efficacy, the fidelity loss needs to be minimal to establish the veracity of the data and validity of the trial.

Clinical trials research attributes:

1. Low data fidelity compared to previous verticals
2. Higher latency than previous categories
3. Relatively high skill sets not required
4. Coarse granularity suffices in the data model as looking for specific outputs to determine success criteria
5. SCD level 2 will suffice

Population health management and value-based care

The slicing and dicing of data in value-based care (VBC) is paramount for healthcare providers to provide optimal care to patients. It provides the right payer level views for insurance carriers who may participate in different care protocols to ensure better quality of care for the patient while ensuring cost efficiencies. 

A range of reporting and information mechanisms are possible for the patient, healthcare providers and payers that assist each in their own respective domains. ,Population health and its analytics may be the lowest with respect to fidelity and highest on latency. And this may be acceptable as it does not impact the patient-centric care protocols or the analytics, as the data is sufficient and demonstrates the required value propositions for a value-based care approach. It also implies that the data granularity is the coarsest here in comparison to other verticals or categories. 

Population health and VBC attributes:

1. Lowest fidelity compared to other verticals
2. Highest latency
3. High skill set not required
4. Coarsest granularity suffices in the data model, highly denormalized datasets for measures and quality gaps suffice
5. SCD level 2/1 will suffice

Conclusion

The necessary skill sets to build, manage, and maintain data architectures get more complex as the need for higher data fidelity increases. Though data velocity, volume, and variety may have their respective effects, they are managed by technology drivers like asynchronous processing and improved I/O throughput mechanisms. The skill/engineering required ramps up considerably as we head to higher fidelity and lower latency verticals. 

Verticals like EHR and CDS inpatient warrant higher data fidelity and also much finer granularity with SCD levels at 6. As we walk towards lower fidelity zones such as revenue cycle and clinical analytics, data entities may be much coarser and possibly even more denormalized. It may be evident that the data model complexity for each of these verticals may grow exponentially as seen in Figure (A). It also implies that a higher level of expertise may be required to manage these models effectively. It may also be noted that such models should encompass business processes and workflows that run the given customer’s business and its related activities. The skill set required for data fidelity and latency is illustrated below.

Data fidelity and latency has a pronounced effect on the data complexity, data architecture, and data models and the level of expertise needed to create, manage, or slice and dice the data to achieve any meaningful insights. The possibility of quantifying data fidelity within a domain such as healthcare requires a scientific determination of its value on a scale calling for a dedicated academic exercise.

Data fidelity and latency or its impact on the data model may be likened to the efficacy of a vaccine. Measuring data fidelity in any data related project could clearly quantify its success. One rule of thumb would be to correlate the data fidelity and latency to the business use cases and determine how well it did to solve the different business scenarios.