How Important Data Get Lost
By Robert S. Gold, MD
For The Record
Vol. 26 No. 12 P. 6
We've all heard about severity-adjusted or risk-adjusted data streams. But how are these adjustments devised? What's the math?
On the whole, folks who are interested in medical statistics have uniform ways of calculating mortality or complication data: They use algorithms, which are lists of criteria. If the patient's record meets the criteria, then the patient falls into the algorithm. If they can't apply these across the board, then they're essentially playing with data. Let's examine this concept in its simplest form.
Take all of the deaths in a large population and the hundreds of thousands of Medicare inpatient cases reported each year, identify a particular ICD code, and examine the patient discharge status codes. Discharge statuses indicate whether the patient was discharged alive or dead (20). From this information, the gross death rate for a particular ICD code can be determined.
A Numbers Game
Consider ICD code 486: Pneumonia, organism unspecified. Run all of the discharges for a particular time period through a software program that can detect all of the principal diagnosis codes of 486 and discharge status codes to determine the total and the number with a discharge status of 20. Take the cases with a principal diagnosis code of 486 and a discharge status of 20 and divide the total by the number of cases with a principal diagnosis code of 486. The resulting percentage is the gross mortality rate for pneumonia 486 in the Medicare population.
Perform the same exercise with acute suppurative otitis media. Ascertain how many cases were reported with a principal diagnosis of 382.0x (three codes: 382.00, 382.01, and 382.02) and those with a discharge status of 20. Divide the total by all discharges with the principal diagnosis to determine the mortality rate.
Next, use computer power to examine every secondary diagnosis code starting with 001.x for cholera and redo the calculations all the way to 999.9 for other and unspecified complications of medical care, not elsewhere classified. Does any secondary diagnosis code affect the mortality rates for pneumonia or acute suppurative otitis media? Some secondary diagnoses will have no influence on mortality rate or be statistically insignificant, while others will fall into predefined ranges such as mild, moderate, severe, and extreme.
Compute the effects of all V, E, and ICD procedure codes on the principal diagnosis codes for pneumonia and acute suppurative otitis media. Create a model of the cumulative effects of ICD codes using either individual additive codes or group codes. No matter the method, it is expected that more complex cases will feature codes associated with higher mortality rates.
Factors such as age, gender, and environment can result in mortality rates varying across time periods. Still, over the course of a year, their influence is not as great. With that in mind, what steps can the health care industry take to influence outcomes?
Uses of Mortality Data
A national standard for treatment of the principal diagnosis can help determine whether a particular treatment strategy is beneficial. New medications or treatment plans may affect mortality rates, with statistical analyses helping to determine whether they survive or disappear. Integration of care, a new concept in which patients are treated for not only the principal diagnosis, but also every other diagnosis, may result in lower mortality rates for the principal diagnosis because other conditions are better controlled.
Suppose there's an additional diagnosis discovered during the original analysis that carries a high mortality rate. For example, what if a facility mandates that every patient who has hypoxia must be documented as having acute hypoxemic respiratory failure? Adding the code 518.81 will result in major complications or comorbidities and a 50% increase in reimbursement.
If it's been determined that 518.81 as a principal diagnosis carries a high mortality rate, what effect will that have on pneumonia, organism unspecified, and acute suppurative otitis media cases? What if they've experienced increased death rates, lengths of stay, and expenditures when acute respiratory failure was a secondary diagnosis? Add acute respiratory failure to all hypoxia cases and define hypoxia as a saturation under 92% on room air. As a result, 518.81 will be added to 40% of the cases, resulting in additional reimbursement, a more impressive case mix index, and kudos from the CFO—even though the diagnosis is incorrect and was unethically documented in the chart.
What happens to the pneumonia data? In all likelihood, deaths from pneumonia without 518.81 and the relative weight of the pneumonia diagnosis-related groups (DRGs) will decrease because costs, lengths of stay, and the death rate fell relative to the cases where documentation and coding was manipulated while those with 518.81 remained unchanged. Meanwhile, any hospital that didn't manipulate the data will receive less money and average shorter lengths of stay. It's a situation that would make any recovery audit contractor (RAC) drool.
What happens the following fiscal period when those cases documented with 518.81 when it wasn't present disappear because a RAC found that the patients didn't have acute respiratory failure? All additional reimbursement plus the original compensation must be forfeited.
This phenomenon can also occur on a grander scale. For example, pneumonia and acute suppurative otitis cases in which the patients have a fever, an elevated white count, and tachypnea may be documented "with SIRS." This changes the DRG to 871 for the pneumonia cases and 872 for the acute suppurative otitis media encounters, resulting in increased reimbursement and a shinier case mix index—but the patients don't have sepsis.
How does this documentation affect actual sepsis cases? If fabricated cases are added to legitimate care episodes, the condition's overall death rate, costs per patient, and lengths of stay decrease. As a result, RACs are on high alert for any misrepresentations in this area.
The health care industry has been doing this to itself. The massive overuse of acute kidney injury diagnoses, the excessive use of the code for acute respiratory distress syndrome for patients short of breath, and the proliferation of encephalopathy codes for patients either high on cocaine or appropriately sedated with medication have led to serious conditions being misrepresented.
Acute kidney injury used to be a major comorbidity. However, it lost that designation when it became apparent the condition had become a catchall for other diagnoses. The excessive use of the acute kidney injury code led to a decrease in costs and death rate, making it the only organ failure condition that isn't a major comorbidity.
At one time, status asthmaticus was viewed as a serious condition in children and associated with prolonged hospital stays and high costs. However, the addition of the acute respiratory distress syndrome code has resulted in status asthmaticus being regarded as no more serious than mild asthma. Take a condition with a 5% death rate and add one with a 50% to 70% death rate to that cohort of cases when that condition isn't present and the death rate artificially drops from 3% to 2% to 1%. In addition, length of stay drops from four days to 1/2 day.
As a result, organizations documenting correctly get hurt, statistics become skewed, and RACs rake in the cash. The reasons why organizations jumped onboard such tactics are immaterial. It's not the right way to treat medical data, and it's not the right way to treat physicians and hospitals. Don't document and code for dollars; document and code for accurate patient data.
— Robert S. Gold, MD, is cofounder of DCBA.