We in health information technology are optimistic that clinicians will update their clinical documentation practices so they use nomenclatures (e.g., Systematized Nomenclature of Medicine Clinical Terms [SNOMED CT], Logical Observation Identifiers Names and Codes [LOINC], RxNorm) in designated fields and then add more details within the free text portions of their electronic medical record documentation. While medical informaticists might expect to use natural language processing to extract more elements out of physician documentation, the American Medical Association (AMA) has worked to reduce documentation requirements starting January 1, 2021.1 Given the amount of time clinicians spend interacting with electronic medical records both inside and outside their clinical settings, the AMA’s documentation reduction will be welcomed by most physicians. How might clinicians use this shift in documentation requirements to improve their documentation practices, so they spend less time entering data while simultaneously entering higher-value notes from the perspective of other team members and patients?
New Documentation Requirements Starting January 1, 2021
This table summarizes the AMA documentation requirements for new and established office visits (Evaluation & Management [E/M] codes 99202-205, 99211-215). Clinical documentation is more likely to include:
- Number and complexity of problems – the requirements only ask that one or two problems be documented, regardless of evaluation and management billing code chosen
- Tests, documents and independent historians
- Independent test interpretation
- Discussion of management or test interpretation with an external qualified health care professional, and
- Risk of morbidity from additional diagnostic testing or treatment – prescription drug management meets the 99204/99214 threshold
Text-based natural language processing (NLP) algorithms attuned to the presence or absence of historical elements or physical exam findings within free-text clinical documentation are less likely to perform well after January 2021 when the documentation requirements change. The industry will need to monitor documentation changes over time to update their NLP algorithms appropriately. The AMA medical decision-making guidelines state that at least:
- One chronic condition with evidence of exacerbation, progression or side effects of treatment for a 99204 or 99214 E/M code, and
- One chronic condition with evidence of severe exacerbation, progression or side effects of treatment OR the condition poses a threat to life or bodily function for a 99205 or 99215 E/M code.
These documentation requirements are unlikely to support more intensive documentation within a visit. More importantly for some disease modelers, the documentation is unlikely to reinforce the need to document a patient’s functional status, quality-of-life or treatment preferences, much less the patient’s specific disease progression.
Clinicians do not document for the sole purpose of meeting billing requirements. The encounter documentation should reflect the rationale behind why a decision was made, especially if the decision is not consistent with a published guideline or may reflect competing priorities or an assessment of a patient’s willingness to comply with a suggested plan of action. The documentation should help other members of the care team and may serve as evidence in a malpractice trial many years in the future.
What might we expect a clinician to do within a 20-minute scheduled encounter in the clinic or a bedside visit in the hospital?2 From an electronic medical record perspective, clinicians might perform the following sequence of tasks:
- Before interacting with the patient
- Chart review with test results, clinical notes from other providers or hospitalizations, and the last encounter the clinician had with the patient
- Query for information with a health information exchange
- Problem list
- During the patient interaction
- Medication list and medication allergies
- Social history
- For savvy clinicians,
- Using online tools to help patients make more informed decisions about their health
- Check any online references to refine a diagnosis or treatment plan
- Review and negotiate the plan with the patient and their loved ones
- Entering orders during the interaction
- After the patient interaction
- If not already done, entering orders after reviewing any online references
- Complete the encounter documentation
Clinicians looking to optimize their practice workflow will minimize the time they spend on different screens within a patient’s electronic medical record.
For data gathering, they will look to those screens that provide them new information within a context for their understanding of that patient. Although some electronic medical record systems may highlight records from external facilities, providers are likely to focus on those conditions they have enough confidence in acting upon. Rather than focus on identifying hidden patterns in the patient’s chart, early iterations of natural language processing might identify diagnosis or treatment recommendations that were ignored or not followed up. The technology would highlight information for human providers to act upon (Meredith Broussard’s chapter on third-wave artificial intelligence describes human-in-the-loop systems to address edge cases that are impossible to address with software a priori in her book “Artificial Unintelligence: How Computers Misunderstand the World”).
Problem lists will only be updated to remind team members of conditions that might adversely affect another service’s diagnostic or treatment plans. The promise of higher reimbursement rates based on a higher risk adjustment factor (RAF) score may not be sufficient to drive more detailed diagnostic coding (e.g., clinicians will code “renal failure” instead of “renal failure stage IV” and “diabetes mellitus” instead of “diabetes with renal complications”). The Centers of Medicare and Medicaid Services expect clinicians to code each chronic medical problem with supporting documentation at least annually. Forward-thinking medical group leaders will identify common diagnosis codes that tip patients into higher RAF scores by specialty, helping clinicians look for data points that support those more severe diagnoses.
For decision support, clinicians may prefer to be interrupted before walking into an exam room (or going online for a telehealth encounter) so they can adjust their suggested visit agenda accordingly. Any suggested changes to a diagnostic or treatment plan might be best received during the “agenda-setting” phase of the visit rather than the “order-placing” phase. Disease management programs or digital health apps could be recommended as an extension of the provider’s office to help manage chronic disease. Limiting decision support to general suggestions (consider intensifying medication therapy, consider assessing the patient for depression, use a calorie-counting app) may be less attractive to health system leaders than making a specific medication recommendation to close a gap in care, but the workflow-concordant recommendation may be more likely to be accepted by a practicing clinician. Alerts that notify providers about adverse drug reactions or recent test results might be acceptable during an ordering workflow, but a reminder to get a tetanus vaccine will be viewed as irrelevant.
Health system leaders and payers may consider offering decision support after the encounter. The decision support could help identify the highest value medication within a specific medication class (even if that medication is outside the payer’s pharmacy benefit manager). The software could help identify high-value providers for laboratory, radiology, consults and durable medical equipment. Depending on the provider’s disposition, suggested referrals to an outside specialist for a second opinion may be valuable, especially before starting an expensive long-term medication or a procedure.
The AMA update to documentation requirements provides physicians and other members of the healthcare marketplace to reassess how to best consider steps within a clinical encounter. If nothing else, analytics using a clinician’s free-text documentation will need to be updated. Diagnoses, medications and procedures will continue to be coded within accepted taxonomies. The free-text section of a clinician’s documentation might no longer include a review of systems or 10-organ physical examination. Without specific guidance from any national body, providers may be reluctant to reallocate their documentation time to track a patient’s functional status or treatment preferences. Decision support may be better deployed if framed as agenda-setting, patient safety around ordering, and post-visit optimization, each with their own tactics.
1The AMA documentation update also includes revisions to the encounter time that might be used as an alternative to meeting the medical decision making requirements for evaluation and management codes. For those clinicians who bill based on encounter time, the new thresholds are:
|Code||Time (Minutes), pre-2021*||Time (Minutes), 2021 onwards||Code||Time (Minutes), pre-2021||Time (Minutes), 2021 onwards|
I interpret the new time requirements as an attempt to address the increase in 99214 codes for “routine” clinical encounters. Most clinicians are not interested in the marginal increase in reimbursement for longer encounter times and would prefer to meet the higher medical decision making criteria for higher value codes within a 15 or 20 minute appointment slot.
2The AMA documentation updates are restricted to outpatient encounters, but I am arguing that the workflow of an inpatient encounter are similar enough to an outpatient encounter to generate high-level generalities.