February 2017
Industry Perspectives: Documenting Death
By Erica E. Remer, MD, FACEP, CCDS
For The Record
Vol. 29 No. 2 P. 5
What are the odds that your next-door neighbor is coming home from the hospital after a bout of pneumonia? It depends. Is it the 22-year-old with asthma or the 86-year-old grandfather in septic shock with comorbidities of congestive heart failure, metastatic lung cancer on home oxygen, and dehydration?
Intuitively, you know there is a higher likelihood that grandpa will die, but statistics also bear that out. That is what risk modeling does, and that is how risk-adjusted mortality rates work.
Although mortality is one of the easiest metrics to understand, all risk adjustment works the same. The basic principle is observed over expected—that is, what happened divided by how likely was it to have happened in this type of patient with his or her specific constellation of demographics, diagnoses, conditions, and procedures.
Let's take mortality as an example. Observed is binary—did the patient die or not? The question is really, "How expected was the death?" The model relies on examining a historical population with known outcomes and determining which factors made death more or less likely. To reach that goal, health care organizations must accurately and completely capture the diagnoses, which are then entered into an algorithm that spits out a relative probability, or how likely the patient was to have died.
To minimize confounding from hospital-specific attributes, there is a case mix index–related fudge factor in the modeling applied to the "expected" denominator. This explains why hospitals are eager to capture diagnoses on all patients in the cohort, not just expired ones.
Risk-adjusted mortality rate, or index, is calculated for a compilation of a patient population over a given period of time. In essence, the rate can predict the relative risk of death if patients are admitted into hospital A vs hospital B. The general risk-adjusted mortality index measures the risk of death for all patients (excluding neonates). There are condition-specific, 30-day risk-standardized mortality rates, which are part of the clinical care domain for the hospital value-based purchasing program.
A final point of consideration is present on admission (POA) status. Patients who have nonacute deterioration ending in demise often share common final pathways such as coma, respiratory failure, and cardiac arrest. Hence (with rare exceptions), only conditions that are POA are calculated into the model. Even if not POA, it is still important to capture these serious conditions because they may help establish the diagnosis-related group (DRG) as being with a major complication or comorbidity (MCC).
Details, Details, Details
Which conditions make a difference? Let's begin with a general overview of documentation in mortality cases, followed by details on some specific conditions.
A comprehensive documentation effort should address several key issues, most notably integrity. Documentation should not be a fishing-for-dollars expedition or an attempt to exaggerate conditions to make a patient appear sicker. For example, don't query for hyponatremia with a sodium of 133 just because fluids and electrolytes are on the mortality model.
The top priority should be ensuring there was excellent patient care, not just documentation that gives the perception of quality care. If there is a doubt, there must be a system in place to refer to for a quality review.
It's important to capture the correct inpatient admission type. There are models that risk adjust, for example, for Admit Status = Emergency/Trauma Center or Transfer From Acute.
Demographics are usually immutable, but organizations that use a dummy birthdate for John Does should correct it prior to submission to the mortality model. A 22-year-old who dies after a craniotomy for multiple significant trauma whose pretend birthdate makes him appear to be 101 is not accurately risk-adjusted because "Male, 18 <= Age < 31" is one of the variables.
Capturing accurate DRGs can be tricky. If there are two appropriately documented conditions and both can meet the definition of principal diagnosis, there are options. It may completely change the risk model and relevant conditions. However, it's best not to query for and introduce a new diagnosis when there is no evidence that the health care providers entertained it during the encounter.
In particular, if a patient dies of an infectious disease and the team missed the sepsis boat, it is dangerous to query the patient into a DRG 871. This may unintentionally disrupt core measures. It's important to refer such a case to Quality for education—missing sepsis is bad clinical practice. Similarly, if the care team never acknowledged or worked up an abnormal troponin, it's not advisable to query for type 2 myocardial infarction de novo.
Is the patient in the highest DRG in a triplet (ie, with an MCC)? Check documentation for the possibility of an uncaptured MCC or CC, similar to what occurs during clinical documentation improvement reviews. To avoid being exposed to a retrospective clinical validation denial, do not be overly aggressive with acute kidney injury as a sole CC.
In almost all cases of demise, the patient was sick, which should be reflected in the severity of illness (SOI) and risk of mortality (ROM). In fact, there are mortality models that have the admit ROM as the only risk-adjustment factor. As in all clinical documentation improvement programs, patients should look as sick in the EMR as they did in real life.
Difference Makers
Now that the admit status is correct, the patient is in the right DRG with all the documented CCs/MCCs captured—hopefully at a SOI/ROM 4/4—which conditions make a difference?
Under Vizient (formerly University Health System Consortium) Mortality Modeling, which will be used here as the exemplar, there are small variations between the academic and community hospital set, but not enough to warrant separate exploration.
Most of the risk-adjustment variables are groups of conditions such as "adrenal disorders" composed of multiple E codes or "liver disease," including conditions found in both the infectious disease and diseases of the digestive organs sections. Few variables are solitary ICD-10 codes, such as acute tubular necrosis or end-stage renal disease. A relatively small set of condition groupings represent the lion's share of the risk-adjusting variables. They are repeatedly found in multiple risk models, eg, vent on admission day, metastatic cancer, and CC fluid and electrolyte disorders.
Unlike some other risk-adjustment methodologies, unspecified codes are often included in the groups, such as E46, Unspecified protein-calorie malnutrition in CC Malnutrition and I50.9, Heart failure, unspecified in CC Congestive Heart Failure. For efficiency's sake, address as many deficiencies as possible in one action. For example, if some type of respiratory failure is suspected, query to elicit the most specific diagnosis and ICD-10 code and not just the unspecified version, even though the latter would be sufficient to trigger the mortality model. There are many other reasons to eschew "unspecified"; never let a teachable moment go to waste.
Don't include too many conditions in the postmortality review query. Providers will view this as an unethical attempt to inflate the SOI and may decide to not agree with any of them. Pick the top three highest risk-adjusting conditions (attend to capturing the MCC first, if necessary) and save the rest for a future skirmish.
Some disorders result in a predictable constellation of conditions or signs/symptoms, each of which should be cataloged and coded. The archetype of this is liver disease. The causative hepatitis or cirrhosis triggers the liver disease category, but hepatic encephalopathy and coagulopathy due to liver disease should also be diagnosed and captured.
For now, hepatic encephalopathy should be coded with the "without coma" version of the liver disease plus G93.49, Other encephalopathy. (The Maintenance Committee is working on its own unique code.) This is in accordance with the Coding Clinic Clarification 3Q 2016, p. 42, which instructs to use a type 2 diabetes mellitus w/ hypoglycemia w/o coma plus G93.41, Metabolic encephalopathy for "encephalopathy due to diabetic hypoglycemia." Because the underlying conditions do not invariably result in encephalopathy, there must be an additional code to describe that manifestation and how profoundly ill the patient is.
The appropriate coagulopathy code for the linked coagulation defect is D68.4, Acquired coagulation factor deficiency. This is important because documentation resulting in the R79.1, Abnormal coagulation profile, does not trigger the model.
Be on the lookout for the following specific conditions (as applicable and with integrity):
• Low blood pressure: Was it really hypotension or shock instead?
• Fluid and electrolytes disorders: Includes E86.9, Volume depletion, unspecified; dehydration; acid-base diagnoses; fluid overload; hypo-/hypernatremia/osmolality; hypo-/hyperkalemia.
• Metastatic cancer (any current active malignancy).
• Severe brain/spinal conditions: These are triggered by most types of encephalopathy, anoxic brain damage, cerebral edema, brain compression, and brain death, but not coma.
• Coagulopathy variable groups: There are several that risk adjust, and providers have no idea that they are consequential, so they often don't include them in the History and Physical assessment and plan. The reason these conditions are considered significant is that they render the patient more likely to die. Recognizing them early with POA-Y designation could improve the patient's outcome.
Certain D68 coagulation defect codes and particular types of thrombocytopenia trigger coagulopathy and CC coagulopathy. D68.32, Hemorrhagic disorder due to extrinsic circulating anticoagulants, refers to a drug-induced condition of excessive anticoagulation complicating a bleeding condition. The provider needs to make a link between the coagulopathy and the medication, as well as the bleeding and the coagulopathy. This is fortunate in that it's difficult to get providers to use that ICD-10 code verbiage.
• Code status: It is imperative to address code status for clinical reasons. Additionally, the status of DNR and/or palliative care is found in certain mortality models. This supports addressing the issue early, specifically in the emergency department. Remind providers that DNR does not mean Do Not Record; who is more expected to die than this type of patient? Regardless of whether you transition the patient at or after admission, you still want to be sure appropriate risk-adjusting conditions are documented and captured, even if an active decision to not treat due to code status has been made.
As with all worthwhile clinical documentation improvement activities, the last action is to close the loop with feedback and education. Querying for conditions that impact the risk-adjusted mortality model serves as a teaching moment for the individual provider. Information regarding lessons learned from mortality reviews should be disseminated to the entire medical staff to help improve their documentation of all patient encounters.
Keep in mind that the ultimate goal of accurate documentation and coding is not to explain why a death occurred unexpectedly, but rather to ensure excellent clinical care and the prevention of avoidable deaths.
— Erica E. Remer, MD, FACEP, CCDS, is founder and president of Erica Remer, MD, Inc, which offers clinical documentation and ICD-10 consulting services.