In the HCUP inpatient databases, the first listed diagnosis (DX1) is the principal diagnosis. In the HCUP outpatient databases, the first listed diagnosis (DX1) may not be the principal diagnosis; it may just be the first listed diagnosis on the record.

In practice, ICD-9-CM diagnoses are represented by 3- to 5-character codes with explicit decimals. In the HCUP data files, ICD-9-CM diagnoses are represented as 5-character alphanumeric codes with implicit decimals. Examples are given below (alphanumeric codes are enclosed in quotation marks):

For proper handling of diagnosis codes:

• Alphanumeric diagnosis codes must be left-justified so that there are 2 spaces following a 3-character diagnosis code and 1 space following a 4-character diagnosis code. For example - '481 '.
• Trailing blanks should never be zero-padded (filled with zeroes so that all 5 characters are filled for codes that should be 3 or 4 characters long). For example - '481 ' should not be changed to '48100'.
• Leading zeroes must be preserved; they are significant.

The original value of the first listed diagnosis (DX1), whether blank or coded, is retained in the first position of the diagnosis vector. Starting at the first secondary diagnosis (DX2), the diagnoses are shifted during HCUP processing to eliminate blank secondary diagnoses. For example, if DX2 and DX4 contain nonmissing diagnoses and DX3 is blank, then the value of DX4 is shifted into DX3. Secondary diagnoses are never shifted into the first listed position (DX1).

Prior to 2003, E-codes are included in the diagnosis array (DXn). Beginning in 2003, any separately reported E-codes and any E-codes encountered in the diagnosis array are placed in a separate array specific to E codes (ECODEn).

Diagnoses are compared to a list of ICD-9-CM codes valid for the discharge date. Anticipation of or lags in response to official ICD-9-CM coding changes are permitted for discharges occurring within a window of time around the official ICD-9-CM coding changes (usually October 1). In the data prior to 1998, a six months window (three months before and three months after) is allowed. Beginning in the 1998 data, a year window (six months before and six months after) is allowed. For example, the code for Single Liveborn changed from "V300 " to "V3000" as of October 1, 1989. Under HCUP validation procedures, "V300 " is classified as valid for discharges on December 31, 1989, and "V3000" is classified as valid for discharges on July 1, 1989. If the diagnosis is not left justified, contains intermittent blanks, or is zero filled, then the diagnosis will be invalid.

Diagnoses are compared to the sex of the patient (EDX03 beginning in the 1998 data and ED1nn prior to 1998) and the patient's age (EAGE04 and EAGE05 beginning in the 1998 data and ED3nn and ED4nn prior to 1998) for checking the internal consistency of the record.

How invalid and inconsistent codes are handled varies by data year.

• Beginning in the 1998 data, invalid and inconsistent diagnoses are masked directly. Validity flags are not included on the HCUP record. Clinical Classifications Software (CCS) data elements are coded with respect to the diagnosis.

• Prior to 1998 data, invalid and inconsistent diagnoses are retained on the record. Validity flags (DXVn) indicate invalid, inconsistent diagnosis codes. Clinical Classifications Software (CCS) data elements use the former name (DCCHPRn). The CCS was formerly known as the Clinical Classifications for Health Policy Research (CCHPR). The diagnosis related data elements are coded as follows:

The validity flags (DXVn) need to be used in connection with any analysis of the diagnoses (DXn).

The maximum number of diagnoses reported varies by state. HCUP retains all diagnosis fields provided by the data source.

Since on the NIS the number of diagnoses coded on the discharge (NDX) can be greater than the number of diagnoses available on the inpatient record, caution needs to be taken when using NDX to loop through the diagnoses. A counter for the loop should not extend past 15. Programming code such as the following example SAS statement is needed to take this into account:

DO I = 1 to MIN(15,NDX);Followed by code to process all diagnoses.END;

Beginning in data year 2009, hospitals in Arizona are required to report E codes for an injury diagnosis in the range 800-999. In addition, the number of E codes that could be retained on a discharge record increased.

California does not require the reporting of E-codes in the range E870-E879 (misadventures and abnormal reactions).

Incomplete External Cause of Injury code (E code) information in 2005 and 2006
In 2005 and 2006, the Oklahoma inpatient data provided to HCUP did not include their field for a separately collected External Cause of Injury code (E code). Some E codes were reported with the diagnosis, but the exclusion of this data field means that the E code reporting is incomplete.

Oklahoma requires E codes to be reported for only the initial treatment of the injury. They are not required for transfers or any secondary treatments. E codes are required whenever there is any diagnosis (primary, secondary) of an injury, poisoning, or adverse effect (ICD-9-CM codes 800-999) and it is the initial treatment for that condition.

South Carolina does not provide to HCUP E-codes in the range E870-E876 (misadventures and abnormal reactions).

West Virginia hospitals were not required to submit E-codes. Thus, these fields were almost 99% missing in the 2007 files.