Asking too much of administrative data?

Abstract

We read with interest the recent work by Best and colleagues, who linked the Department of Veterans Affairs (VA) National Surgical Quality Improvement Program (NSQIP) database with the Patient Treatment File (PTF), the VA’s uniform hospital discharge abstract database. They used this linked database to evaluate the accuracy with which numerous ICD-9-CM-coded preoperative risk factors and postoperative complications are reported in the PTF, and, by implication, in other administrative data sets of the same type. Few previous analyses of this topic have been as comprehensive and as potentially useful to researchers and quality improvement professionals who are eager for more efficient methods to identify adverse outcomes of hospital care. But, the usefulness of Best and colleagues’ work is limited by several key problems in how they mapped NSQIP variables to ICD-9-CM. Although ICD-9-CM is the most widely used scheme for classifying inpatient diagnoses in the United States, its application requires knowledge, experience, and careful reliance on coding resources, such as Coding Clinics for ICD-9CM. For example, “codes that describe symptoms and signs. . .are acceptable for reporting purposes when a related definitive diagnosis has not been established (confirmed) by the physician. . .symptoms and signs that are integral to the disease process should not be assigned as additional codes.” So, “weight loss” (mapped to 783.2), “dyspnea” (mapped to 786.0x), and “impaired sensorium” (mapped to 780.02, 780.09) are not supposed to be coded, unless they are either idiopathic or related to an underlying diagnosis that does not typically cause that symptom. One should not even try to ascertain signs and symptoms from hospital discharge data; any effort to do so is doomed to failure. Second, “abnormal findings (laboratory, x-ray, pathologic, and other diagnostic results) are not coded and reported unless the physician indicates their clinical significance”; only “conditions that affect patient care in terms of requiring clinical evaluation; or therapeutic treatment; or diagnostic procedures; or extended length of hospital stay; or increased nursing care and/or monitoring” should be coded. This provision precludes coding of minor complications that might not affect patient care, such as “superficial wound infection,” and many complications of uncertain clinical significance, such as “failure to wean from ventilator 48 hrs” and “prolonged ileus.” So, Best and colleagues’ finding that these complications were reported with poor sensitivity is neither surprising nor disturbing. Third, ICD-9-CM includes both diagnosis and procedure codes. Coders are instructed to code all “significant” procedures, including any procedure that “is surgical in nature. . .carries an anesthetic risk. . .carries a procedural risk. . .requires specialized training.” Procedural risk encompasses “any procedure that has a recognized risk of inducing functional impairment, physiologic disturbance, or possible trauma. . ..” Best and colleagues failed to consider that procedure codes can sometimes be used to identify patients with a condition of interest more accurately than diagnosis codes alone. For example, the use and duration of mechanical ventilation are indicated by procedure codes 96.70 to 96.72, in addition to associated codes for endotracheal reintubation (96.04) or tracheostomy (31.1 to 31.29). Severe renal failure leads to hemodialysis, a costly procedure that can be accurately ascertained with procedure codes 39.95 and 38.95. Fourth, coders are instructed to apply only the single most appropriate code for each diagnosis, unless multiple coding is needed “to fully describe a single condition that affects multiple body systems” or “to more fully describe” a bacterial infection, late effect, complication, or obstetric diagnosis. For example, if the etiology of a patient’s coma is known, then the diagnosis code describing that type of coma (eg, diabetic coma, 250.2x to 250.3x; hypoglycemic coma, 251.0; hepatic coma, 572.2) should be applied instead of a general code for coma (780.01). If the etiology of a patient’s limb pain (729.5) or gangrene (785.4) is known, then the specific diagnosis code with symptom-based modifiers (eg, atherosclerosis of the extremities with intermittent claudication, 440.21; rest pain, 440.22; or gangrene, 440.24)