Skip to main content

The Clinical Documentation Conundrum

News  |  By  
   February 01, 2017

Healthcare leaders are looking at their options for making EHRs simpler to use and their data more interoperable, as well as making clinical documentation more useful and meaningful to clinicians. Healthcare leaders have had some success, but solutions aren't coming easily.

This article first appeared in the January/February 2017 issue of HealthLeaders magazine.

Electronic health records, in use at nearly all U.S. hospitals and by more than three-quarters of all physicians, achieve two major objectives for healthcare systems. They facilitate generation of a bill, and they provide documentation of providers' assessments, actions, and plans to treat patients. As the industry pivots from volume to value, the nature of clinical documentation is undergoing unprecedented change, and is driven by the way providers get paid.

This transition is challenging healthcare leadership in ways never before encountered. Changing documentation technology is often cited as a leading cause of physician burnout, in studies such as one by the Mayo Clinic in June 2016, in which 6,560 physicians in active clinical practice were surveyed.

While healthcare leaders and the industry look to evolve EHRs to make them simpler to use and their data more interoperable, and to integrate timely decision support into EHRs to make documentation more useful and meaningful to clinicians, at the same time, patterns of documentation well established under the fee-for-service model persist,confounding these moves toward simplicity.

Lessons from the traditional CDI dynamic
The practice of clinical documentation improvement actually predates the EHR and, when CDI's digital equivalent is used well, has proven to be complementary to a fee-for-service world. Prior to EHRs, physicians would document on paper, and then CDI specialists within a health system would pore over these notes, flagging areas where physicians had not been specific enough. This provided several benefits: alerting the system that a patient's condition was more severe than originally documented and flagging overlooked issues such as comorbidities, but also permitting billing at a higher level, which satisfied management's goal not to leave any money on the table.

"This kind of process was laborious and only partially successful on paper," says Michael Zaroukian, MD, vice president and chief medical information officer of Sparrow Health System, anchored by its 676-licensed-bed teaching hospital in East Lansing, Michigan. The CDI picture brightened when EHRs and accompanying decision support arrived in the form of Claro CDR software for CDI workflow, allowing Zaroukian's team of developers to "make that information available to various health professionals in our organization to decide, for example, which patients might need greater services at discharge to help either prevent readmission or to help them with an improvement in their quality of life," Zaroukian says.

"We did consider it to be part of our approach to making sure we had a better understanding of the total complexity of our care, so that we could then use it to inform other aspects of the tools we use to adhere to best practices, decrease unnecessary variation, and promote improvements in patient care, outcomes, and satisfaction."

In a 2015 HIMSS Analytics Stage 7 case study, Sparrow reported that these CDI efforts yielded $5 million in additional revenue from September 2013 to December 2014, due to increased case-mix index resulting from CDI and computer-assisted coding-supported workflows. Sparrow retired paper CDI queries and replaced them with in-basket EHR messages that made CDI query delivery and physician replies easier and more efficient.

Additionally, physician response rate to CDI queries increased from a baseline of 60% to 92% after implementation of EMR-supported CDI and CAC processes. These processes are now completely electronic, and contributed to Sparrow's CDI specialists being able to move from 80% to 100% review of Sparrow charts, without requiring additional specialists.

Sparrow leadership realized that technology would help with better documentation, so it augmented its Epic EHR with templates, which presented clinicians with cascading text prompts to achieve that greater specificity without a CDI specialist actually looking over their shoulder, Zaroukian says. "As we were getting ready for ICD-10, there was a recognition that just putting an EMR in place and just asking people to document better, using what we've educated them upon in the past and asking them to remember all that and to somehow take time from their busy world in a non-workflow-integrated way to improve their documentation, was not having any significant effect."

The heart of the process is that the text-prompts workflow presents critical information to the clinician at the point of care, but does not overwhelm the clinician.

The Sparrow initiative was facilitated by a physician advisory group, led by physicians who directed the organization's efforts to spearhead the implementation and refined the usability of documentation queries, Zaroukian says.

On the technology side, Sparrow's leadership team also implemented the Optum Enterprise CAC platform for two of its hospitals. This technology reads clinical documents and creates lists of supported diagnoses. These diagnoses contain links to the patient's EHR, highlighting key words and phrases in each document.

Sparrow leadership learned the following from the case study:

  • Successful implementation with desired results requires understanding and agreement around business processes and how health IT systems can support them. One of the key success factors was identifying and overcoming communication barriers to ensure understanding and promote buy-in.
  • Executive level support and involvement were critically important to a successful CDI program.
  • Remember the value of Plan-Do-Check-Act (PDCA) cycles of improvement and continue to get feedback and fine-tune systems and processes after implementation.
  • Communicating the advantages and expected benefits to affected caregivers throughout the implementation process creates positive engagement, anticipation, and commitment.
  • Involving frontline caregivers during system build and testing helps ensure successful implementation, improved processes, and high user satisfaction.

ICD-10 falls short
In the year since ICD-10 replaced ICD-9 in the United States, it has taken its place as one tool in the toolbox to help make documentation more consistent, but a tool that has fallen short of its original promise to assist with CDI. "Some people would say it's helping; other people would say it's hurting the documentation," says Dean F. Sittig, PhD, a biomedical informatics professor at University of Texas Science Center at Houston's Health School of Biomedical Informatics and a member of the UT-Memorial Hermann Center for Healthcare Quality and Safety in Houston. "Right now most of the documentation is in free text, so it's not using ICD-10. There's a lot of inconsistencies there and a lot of problems there."

Others studying the clinical documentation inconsistency problem agree physician complaints are real. "It bothers me when physicians say, 'I don't trust the medical record anymore,' " says William R. Hersh, MD, professor and chair of the department of medical informatics and clinical epidemiology in the School of Medicine at Oregon Health & Science University in Portland. "They don't know if this note's been copied and pasted from an earlier medical record or another patient's medical record. They don't know what the patient's story is, because the previous doctor had to click all these boxes to fill in these forms."

"It's not just about the technology. It's about how we communicate. I have physicians that tell me they don't really believe what's in the medical record anymore. I don't think there is turning back the clock, of going back to paper," he says.

The sheer amount of work it takes for physicians to check boxes and fill in structured clinical documentation fields in EHRs is becoming more and more well documented. Researchers from Weill Cornell Medical College and the Medical Group Management Association surveyed 394 physician practices nationwide in 2016 and found physicians and their staff averaged 15.1 hours per physician per week processing quality metrics—the equivalent of 785.2 hours per physician per year.

Hersh says the industry as a whole needs to address cost and burnout from clinical documentation, particularly as it moves to implement new models of care.

"Medicine is fundamentally an information science," Hersh says. "You see some of the news reports [such as in the September 2016 Annals of Internal Medicine that] physicians spend two-thirds of their time away from patients. The reality is that physicians always have spent a minority of their time with patients. Ten years ago it was writing notes in the chart. Today it's dealing with data and information. As we move to things like bundled payments, information becomes more important."

Although much of this does exist today in medicine, Hersh says the healthcare industry can still learn from other professions such as air traffic control and the military, where using precise language reduces variation and ambiguity in practice. "There's never going to be complete elimination of ambiguity," he says. "We do need to try to standardize as best we can, especially if we're going to see any kind of benefit from things like precision medicine."

SNOMED CT, a standard way of exchanging clinical documentation such as problem lists, has proven to be a more precise way of describing the clinical narrative in a structured way than ICD-10, and the good news is software developers are making progress at mapping natural language technology, taking speech or clinical notes as input, into specific SNOMED codes. In addition, ICD-10's eventual successor, ICD-11, will be derived from SNOMED, Hersh says.

"The best attribute of SNOMED is that it's extensible, so you can build more complex terms from simpler ones," Hersh says.

Another provider likens today's CDI to a kind of arms race in which algorithms suggest that more intricate diagnoses replace simpler diagnoses in order to maximize fee-for-service revenue.

"A lot of things that we do around billing, coding, and case-mix index are dependent on how you talk about a situation," says Sameer Badlani, MD, FACP, chief health information officer and vice president at Sutter Health, a 24-hospital system based in Sacramento, California. "If I were to write 'congestive heart failure,' I would get paid less versus if I say 'decompensated systolic heart failure,' which is a technical term for congestive heart failure. To a physician they mean the same thing. I would provide the same level of care, but there is a marked difference in the level of reimbursement that allows you to get paid better. So we're constantly playing this cat-and-mouse game where people are trying to document appropriate care in the best possible sequence of terms to get adequately compensated for the care they provide. And that is not the way we should be practicing medicine. If we were truly in a value-based system, you wouldn't care what I wrote in my documentation. You would care about the patient outcome."

Moreover, a fundamental challenge as clinical documentation pivots from volume to value is the continuing need to code in both old and new styles in order to satisfy requirements of Medicare and other payers, says Alistair Erskine, chief informatics officer at Geisinger Health System, the Danville, Pennsylvania-based system that serves more than 3 million residents throughout 45 counties in central and northeastern Pennsylvania and also in southern New Jersey with AtlantiCare.

"From 1984 on, evaluation and management coding has been a means for CMS and insurance companies to be able to pay doctors," Erskine says. "It's also been an enormous burden."

More and more, EHRs are helping physicians keep up with ever-growing clinical documentation demands, says Erskine. "When I write a note today, the computer really writes 75% of the note for me," he says. "It picks out certain templates, labs, and diagnostic imaging results. It may pull forward information that I had put in a note previously or that somebody else put in a note, and that becomes part of my new note."

While the computer-added information is a boon for analytics driving measures such as clinical quality reporting and population health initiatives, "the problem is we've lost a lot of the narrative in that note," Erskine says.

The ICD-10 transition also fell short of expectations, Erskine says. "It was dramatically overstated that somehow the doctors are going to be picking the right ICD-10-specific diagnoses, and we're going to have a much more robust and revealing medical record system," he says. "It wasn't that much of an issue for them before, and they're not really paying that much attention to it now."

In response, Geisinger has adopted a policy of not forcing its physicians to use a template when they write their notes. "If you want to just open up a [blank document] and start typing text, you're free to do so," Erskine says. "If you want to dictate your note and have that dictation turn into text, you're free to do so as well."

In addition to relying on ever-improving speech-to-text technology, Geisinger, in its role as an academic medical center, encourages its residents and medical students who encounter notes that do not make narrative sense to return the note to the author via the EHR's inbox function and ask the author to restate the note for better clarity, Erskine says.

Another approach that Geisinger and others have taken is to implement OpenNotes, a national movement for sharing the system's existing medical record notes with patients themselves, who are then encouraged to point out errors or additions, Erskine adds.

This refocusing, however, also calls into question the usefulness of some of what an EHR is supposedly measuring, he says. "Over the past couple of decades, we've gotten obsessed with measuring or using process metrics as opposed to outcome metrics," Erskine says. "There's actually a train collision ahead of us in about two years. We're moving away from fee-for-service as quickly as we can, all the while that we're not trying to lose our shirts in the process of doing so. However, the E&M coding and the billing practices, and the law and rules that exist today, are not going away in terms of what fee-for-service expects you to write. So you have this hybrid world where you're expected 100% of the time to do fee-for-service notes with E&M coding and all these different rules in terms of what you put in your clinical note."

Erskine says he has brought these documentation concerns to officials at CMS, but that officials there are reluctant to deemphasize process-oriented documentation in favor of outcomes-oriented documentation, in part because measuring outcomes remains an unfinished task for the industry, and because "CMS is constantly concerned, and rightly so, with fraud and abuse."

While copy-and-paste use in EHRs can serve a valuable function, misuse of copy and paste has also been an issue since EHR use became widespread. Erskine says EHR vendors such as Epic are helping to distinguish when material in the physician's note has been pasted from some other source through use of differently colored or italicized fonts with such text. "It helps illuminate a little bit where people have overused or where there isn't really that much new information, which in the previous iterations you couldn't really tell," he says.

Yet another challenge is getting physicians to impose self-discipline on the length of various aspects of the documentation. "Problem lists get larger but tend not to get smaller," he says. "People have a tendency not to curate it and make sure it's nice and organized in a way that highlights the relevant things. Part of that is because, who really owns the problem list when it has 40 problems on it?"

Chances are difficulties such as this will require the joint efforts of healthcare organizations, physicians, and in particular, technology suppliers such as EHR vendors, Erskine says. "They're really trying to be ambassadors and reconcile and adjudicate those regulatory needs with what the clinicians are really asking for, and trying to do that in as elegant a way as possible, but it's going to be tough to do."

As one example, Erskine says Epic "is going to make at least provenance within their record more front and center, so if I hover over a term in the record that was copied from somewhere else, it will tell me who was the originator of that particular term."

"Presumably, if I'm communicating well because I'm writing good notes and if I'm coordinating well because I'm working with other institutions, and I'm managing the patient well—that's the measure of success, as opposed to what I've actually done and how many times I've ordered something and how many elements I've put into my clinical notes," Erskine says.

"More documentation expected from the physician and more focus on quantitative markers as surrogates for qualitative outcomes is our problem."

Badlani recommends one approach to make sense of the "note bloat" of today's EHRs in the coming value-based, team-based world of care is to rearrange the typical SOAP (subjective information, objective information, assessment, and plan) note into an APSO note, with the assessment and care plan coming first. "When I'm looking at another physician's note, I really want to see their assessment and plan," Badlani says. "And the reason you still maintain the subjective and objective elements at the bottom of the note is because you need them for compliance, for billing, and for coding standards, and then nobody looks at that except whoever is making sure you are billing appropriately."

As much as steps such as these can help, Badlani still sees the same documentation crisis as others do. "More documentation expected from the physician and more focus on quantitative markers as surrogates for qualitative outcomes is our problem," he says.

Sutter is looking to adopt from other healthcare systems in the clinical documentation picture. "In some organizations, they have standardized the way radiologists or pathologists document," Badlani says. "I, as a consumer, know exactly where in the document to expect what." Combining such standardization with a change from SOAP to APSO, would allow physicians to more quickly find what they are looking for within the EHR, he says.

"It's a really critical topic," Badlani says. "A significant portion of physician dissatisfaction comes from steps they have to do in the EHR, which don't really line up to any clinical benefit. The way we do documentation is one of them."

A second technology effort to help with documentation burdens combines the heads-up display found in Google Glass and similar augmented-reality wearables with a notion popularized shortly after the EHR was invented: the scribe. In this regard, in 2015 Sutter began using technology from Augmedix, a company that pairs Google Glass with scribes working from India who help ease the burden of documentation on the wearable-donning physician, presenting relevant information on the heads-up display and allowing the physician to keep their eyes and attention on the patient. "That's an example of innovation trying to leverage what we can in making this better," Badlani says. Sutter has also made a minority equity investment in Augmedix.

Erskine notes that trying to lean too heavily on scribes to help relieve physician documentation burnout has its own risks. "Scribes have tried to help address the issue but they introduced their own problems in terms of potentially overdocumenting what is actually being done," he says.

Still, all these efforts are a reaction to the failings of the meaningful use program to achieve its ultimate objective—improving outcomes, Badlani says.

"It perpetuated the concept that the mere presence of technology, trying to achieve meaningless measures, would make a difference," he says. "For the longest time, EHRs have not been seen as a clinical transaction tool. They have been seen as billing, coding, and documentation tools. We can't get people to do medication reconciliation on a regular basis in almost any organization consistently, because there are so many other tasks that don't add value to their clinical practice."

Erskine also sees a role for an emerging class of what he describes as "text analytics," which can help identify those clinicians doing a good job of writing notes—as opposed to "those who are clearly kind of using the system to just get through and not necessarily communicate the clinical information that's necessary. We've had conversations like that about how could we make it more usable and how could we make it easier.

"I wish I could give you a crisp answer and say we do this, and that way it's all working itself out. But as long as humans are involved, I think we'll still have to learn together in terms of how to make that happen," Erskine says.

There is also a sense that clinical documentation improvement is only one of many imperatives facing physicians. "Clinical documentation improvement for the sake of clinical care becomes one of the 20 things that the healthcare administration has to put in front of the physician," says Badlani.

Overall, U.S. healthcare is still at the beginning of these transitions, says Robert M. Wachter, MD, professor and chairman of the department of medicine at the University of California, San Francisco, and also author of the 2015 book The Digital Doctor: Hope, Hype, and Harm at the Dawn of Medicine's Computer Age, which delved deeply into clinical documentation issues in EHRs.

"My experience would tell me, that the places that do a big CDI initiative are largely thinking about their balance sheet," Wachter says. Instead, they should look for variations, determine costs of care, and ask why their care is more expensive than care elsewhere, he adds. "You need a higher level of documentation. You need to be more confident that these two patients really are requiring the same treatment, and why did it cost twice as much at this hospital we own than that hospital that we own."

"It's easy to copy yesterday's note and make it today's, and critics said you should just ban that. You shouldn't because there's maybe two-thirds of yesterday's note that is today's."

An example Wachter encountered during the writing of his book involved an epidemic Medicare saw four years ago of a rare African diagnosis of kwashiorkor, a severe form of malnutrition not common in the United States. "As soon as you see that, you know what's happened, that some CDI consultant came through and said, 'It looks like this patient has a low albumin, which is a blood protein level,' " Wachter says. "He says, 'Why don't you call that kwashiorkor, because it pays a hell of a lot better than 'malnourished' or 'wasted.' "

While CMS eventually got wise to this type of exaggeration, Wachter is concerned that as machine learning enters the toolbox of CDI specialists and management, such exaggerations could become yet another arms race between cash-starved healthcare systems. "As the money gets tighter, the premium on trying to figure out how can we, within the bounds of legality, get the doctor to write down the words that trigger the highest payment and provide the highest case-mix severity of illness—that's only going to grow," Wachter says. He also shares that he personally has an interest in tackling this, serving on the board of a company—Accuity Delivery Systems—that creates this kind of CDI technology. "It's just an arms race, but it is a race we all have to run," he says.

Meanwhile, at UCSF, Wachter and other executives are "trying to train our house staff on how to write a good note in the electronic era," he says. "It's easy to copy yesterday's note and make it today's, and critics said you should just ban that. You shouldn't because there's maybe two-thirds of yesterday's note that is today's."

Wachter also challenges the notion that ultimately all clinical documentation will be outcome measurement, not process measurement. "We don't understand outcome measurement well enough," he says. "It may be that the outcomes of interest aren't going to happen for a year or two, controlling someone's cholesterol or blood pressure or smoking. The balance here is how do you put some constraints on the number of measures and the amount of time that people have to spend documenting, and then pragmatically ask people to only record so much stuff in that box."

Team care is changing the documentation process
One aspect that seems to get lost in many discussions around the EHR documentation burden is that the nature of physician documentation is itself changing, from the physician documenting solo late into the night, to one where a care team can gradually shoulder more and more of the documentation burden from the physician.

"I really do think that we are going to evolve more rapidly to team-based care," says Darren Dworkin, senior vice president of enterprise information systems and chief information officer at Cedars-Sinai Health System, anchored by its 886-licensed-bed medical center in Los Angeles. "As healthcare systems better adapt to that team-based care, you might see things like individuals that are teeing up orders by protocol, so physicians are just simply reviewing them and approving, based on some clinical decision support or other tools that are presented to them."

When incentives at a provider align more closely, such decision support tools can liberate physicians from some of the EHR's documentation drudgery, says John Mattison, chief medical information officer at Kaiser Permanente, the integrated health system and payer that serves 10.6 million members in eight states and the District of Columbia.

"We put the member first, and that was a complete game changer in terms of what we did with the product," Mattison says.

"To the point about doctors feeling like they're just billing agents, typing in things to get appropriate billing, the second mantra was that billing would be a collateral derivative, but not drive it. And so I built a team of natural language processing folks so we have real-time natural language processing going on with what's being entered, and the principle there was that we right code everything with the automation of the machine learning associated with natural language processing. So our physicians are liberated to document what's in the interest of documenting quality of care and quality outcomes. Then we derive and support the generation of an appropriate code as a secondary collateral benefit, not as the driving force."

Yet another effort underway would make the EHRs and other repositories of information more interoperable not only on a data file exchange level but on the level of semantics or meaning, so that information flowing in from a different care setting could flow more seamlessly into the patient's primary clinical record and narrative.

"Semantic interoperability is, I think, what we need to aim for," Hersh says. "The aim for version three of HL7 was to achieve semantic interoperability to move everything into some lingua franca that anyone can understand. That proved to be more difficult than people anticipated."

As a middle ground, Hersh points to HL7's Fast Healthcare Interoperability Resources, an emerging standard now being incorporated into EHRs.

FHIR is likely to provide that sort of integration in the short term, Hersh says. "It has structures to start to solve the semantic interoperability problem on a simpler level," he says. "If we can get things like prescriptions and documents with standardized names like progress notes, discharge summaries, operative reports, some standard metadata around them, dates, creators, patients, etc., that will actually go a pretty long way."

Test results, patient reminders, and alerts could flow automatically into the EHR if the EHR could recognize the format of the information seeking inclusion, and if governance at healthcare organizations permitted it, or is incentivized by value-based payment structures, according to Thomas H. Payne, MD, FACMI, chairman of the board of the American Medical Informatics Association, which last year produced EHR 2020, a healthcare informatics road map. Payne is also an attending physician at the University of Washington Medical Center and medical director of information technology services at the University of Washington Medicine in Seattle.

"To change the way we reimburse care from E&M to alternatives at a faster rate than we're already undertaking would require substantive changes in government regulation," Payne says.

Payne also says other forms of standardization in the way clinical documentation is created in the United States could help reduce both the burden and the variation in care targeted by value-based care. Considering the oft-repeated scenario of a healthcare system's physicians gathering to define their own order sets for just that system, "in this age of maturity of our system, you wonder why each organization needs to begin with a blank screen, and the answer is they shouldn't have to begin there," Payne says. "That kind of variation is helpful when we are able to incorporate the preferences or special circumstances of the patient, or the strengths and weaknesses of an individual practitioner. That sort of variation fits into usability."

But Payne favors standardization of order sets, problem lists, and other team-oriented care in the same way that he praises the standardized flight deck instruments and displays that have helped the aviation industry achieve a good safety record.

And as in other industries, it seems inevitable that this sort of standardization in healthcare will be achieved, one way or another, through extensive use of computer algorithms.

"The ultimate endgame seems, to me, to be almost unmistakable," Wachter says. "It has to be that as a doctor and a patient have a conversation, voice recognition picks it up, machine learning sifts through it, parses it in the way it needs to be parsed for the note template, and sifts through it to find things it needs to find to measure the things it needs to measure." Then algorithms would prompt clinicians to ask only important questions not otherwise answered. But such a world is likely still a decade away, says Wachter.

For now, healthcare leaders realize that investing in clinical informatics expertise is an essential strategy to driving CDI, Wachter says.

"You can't get this right if you think of this as you're buying a machine, turning it on, and you're done," he says. "The process of getting it right is it's constantly being updated and you need people who are bridges between the IT and the clinical people to make this work."

Lack of clinical informatics resources in healthcare could slow this progress, says Hersh.

"We always need to resist the temptation to just sort of throw technology at things," he says. "People who have knowledge and experience in informatics know that there are things you need to do, and ways you need to look at things, to make the best use of technology."


Scott Mace is the former senior technology editor for HealthLeaders Media. He is now the senior editor, custom content at H3.Group.

Get the latest on healthcare leadership in your inbox.