Claims Data – The Good, The Bad and The Ugly


By Robert Rowley, MD, co-founder and Chief Medical Officer of Flow Health
Twitter: @RRowleyMD

As we move towards building a universal patient-centered data platform in health IT, several sources of data are useful. Data coming from transitions of care, clinical summary documents (C-CDAs) which can be shared between healthcare providers taking care of a patient, and claims data from insurance payers – these are all sources of data that can build the universal health record.

With all the focus in interoperability discussions around sharing data found in clinical Electronic Health Records (EHRs) and sharing them in some way between disparate systems, we have lost sight of the potentially important role of health plan data. It might serve well to consider the pluses and minuses of this data, in order to best understand where it might fit in.

The good
Health plans have been keeping data based on insurance claims that they have paid for many years. Their historical archives will therefore pre-date the information found in clinical EHR systems in hospitals and doctors’ offices, since EHRs are a relatively recent arrival on the healthcare scene. Therefore, one advantage of health plan data is that it has longitudinal depth.

Another advantage of this type of data is that is collects information from all the different parties that have submitted bills for a given patient’s care – all the different doctors, hospitals, laboratories, imaging centers, pharmacies, etc. Thus, a second advantage is that health plan data is multi-sourced, irrespective of what each biller uses to send claims. Such data can, then, help determine who is the care team taking care of a given patient, so that a patient-centered health record can show the involved providers in a hub-and-spoke fashion.

The kind of information sent on a bill to a health plan is limited in scope, but still valuable. Clinical notes, vital signs, allergy lists, etc., are not things that appear on bills. What does appear, however, are procedure codes (CPT codes) and their associated diagnosis codes (ICD9 codes). Prior to 2012, up to 4 different ICD9 codes could be associated with each CPT code in a bill (the 4010 standard), but since then the new standard (the 5010 standard) allows up to 12 different diagnosis codes to be associated with each procedure code line.

There is considerable incentive for medical providers to use as many diagnosis codes as are appropriate on their bills. Medicare’s HMO offering, Medicare Advantage, will pay contracted private insurers a per-member-per-month premium that is adjusted on how “sick” their enrollees are. This is determined by HCC codes, which weight the level of acuity of a given patient based on their health conditions, and is determined by ICD9 codes found on their bills. Therefore, for those participating in Medicare Advantage plans, there is considerable incentive to be as complete as possible in capturing all the patient’s diagnoses in their billing every single year. Thus, another advantage to health plan data is that the Problem List for a patient, especially when it is incentivized in an environment of maximizing HCC coding, is robust (as well as multi-sourced).

The bad
Many important elements of a patient’s record are not captured in bills. There has been an attempt to capture certain clinical items used for Clinical Quality Measures – such as blood pressure ranges for diabetic patients, or other similar clinical data points – using claims. Medicare has introduced CPT-II and other HCPCS codes which are zero-dollar codes intended to submit clinical information to Medicare and other payers, for use in certain pay-for-performance programs such as PQRS.

Such data could be helpful in building a clinical record, but the use of these codes in billing is spotty at best. After all, they are non-payable codes, and represent a coding burden to providers. In addition, there are other non-claims-based methods of submitting quality measures to PQRS and Meaningful Use, so the absolute need to use these is not compelling.

Thus, a major weakness of claims data is its incompleteness, particularly when it comes to reporting data important for Clinical Quality Measures.

The other weakness from health plan data is that patients often will switch health plans frequently over time. It is rare for an individual to keep the same health insurance their whole life. Therefore, in order to build a complete patient data record over time, data would need to be obtained from each different health plan that they were covered under, since no one plan would likely provide the whole story.

The ugly
Health plan data may be accurate when it comes to identifying who has taken care of the patient, when certain procedures were done, when the patient was hospitalized, and when medications were dispensed. But diagnosis data can be inaccurate, largely based on source.

Community physicians in their offices are often directly involved in declaring the diagnoses on their claims. This is especially true for those involved in HCC environments. Diagnosis data from outpatient claims, therefore, are the most likely to be accurate.

In hospitals, billing is generally done by billing staff, and diagnoses are extracted from examining the record rather than by the physicians themselves. A degradation of diagnosis-coding accuracy is inherent in this arrangement. In addition, hospital bills are often bundled, or paid globally based on Diagnosis Related Groups (DRGs), so individual ICD9 codes are often not relevant.

Therefore, problem lists built from hospital-based claims can be quite inaccurate. As famously chronicled in 2009 by e-Patient Dave’s efforts to put his medical records on a PHR, he found widespread inaccuracies in his problem lists as recorded by the hospital’s system. Granted, such systems have improved in the past 5 years, but the point is that hospital diagnosis data can harbor many inaccuracies.

Perhaps the worst source of diagnosis data is from laboratory claims. Often, a diagnosis is placed on a lab order form simply to make sure the needed test is paid. The concurrence between the diagnosis submitted on a lab order and the diagnoses in a physician’s EHRs is less than 100%.

Using claims data
Insurance companies, in their efforts to keep their enrollees healthy and avoid unnecessary costs, spend considerable efforts identifying “gaps in care.” These can be from lapses in filling of maintenance medications, to suggesting medications that are appropriate for the conditions the patient seems to have but is not on (such as asthma controllers, or ACE/ARBs for diabetics). These kinds of alerts are sent to physicians daily, at considerable cost, often by fax. And entirely based on claims data. And often ignored by physicians as an “annoyance” that does not fit well within their already-overburdened workflows.

Clearly, insurance companies look to the data within their own data silos, and try to make the most of it. For all the good, bad and ugly of it, it remains actionable data.

Perhaps the best way to look at claims data, as a potential source for building universal patient records, is as a supplement to data from other sources. There are distinct advantages to insurance-based information, and there are distinct shortcomings as well.

Why would insurance companies share any of their data? Two reasons:

  • Like health providers and healthcare clearing houses, health plans are HIPAA Covered Entities. They can share their data for the purposes of healthcare, governed by the rules of HIPAA. As a covered entity, just like healthcare providers, if a patient requests their records, then they must be provided within 30 days (per the HIPAA Final Rule of September 2013). Though this is currently seldom done (most requests for data are directed at care providers rather than health plans), it is possible that such requests might increase in the future.
  • Insurance companies already spend large sums on their “gaps in care alerts” efforts, which struggle with low effectiveness. Partnering with outside sources of data can allow them to more effectively identify the right patients needing the right alerts, and present them to the right doctor taking care of them in a way that integrates with (rather than distracts from) the ordinary workflows of healthcare.

Claims-based data from health plans represent a source of data that can supplement a universal health record. It is part of the larger puzzle, and serves a useful role.

About the Author: Dr. Rowley is Co-Founder and Chief Medical Officer of Flow Health. Previously, Dr. Rowley served as Chief Medical Officer of Practice Fusion. This article was original published on the Flow Health blog and it is republished here with permission.