ESQUIZOFRENIA

Systemic lupus erythematosus (SLE) is a prototypic

autoimmune disease that affects more than 300,000

people in the US (1) and millions of people worldwide.

To ensure that there is a consistent definition of SLE for

the purposes of research and surveillance, classification

criteria for SLE are needed. The most widely used

classification criteria for SLE are those developed by the

American College of Rheumatology (ACR). These classification

criteria were published in 1982 (2) and were

revised by a committee in 1997 (3); according to the

revision, the item “positive LE preparation” was deleted,

and the criteria for an immunologic disorder were

changed to include anticardiolipin antibodies. The 1982

ACR criteria have been validated (4,5), but the 1997

revised criteria have not been validated.

Subsequently, multiple groups of investigators

used new statistical methodology to refine the criteria

for classification of SLE. Clough et al applied Bayes

theorem to data from patient and control populations

from the rheumatology department at the Cleveland

Clinic to develop weighted criteria for the diagnosis of

SLE (6). Costenbader et al formulated the Boston

Weighted Criteria system for the classification of SLE,

which was based on the Cleveland Clinic criteria but

included antiphospholipid antibodies (aPL) and renal

pathology (7). In addition, elements that might negate

the diagnosis, such as negative antinuclear antibodies

(ANAs), were subtracted from the criteria set. Some

criteria definitions were revised, such as arthritis requiring

an objective assessment of synovitis (7). The

weighted criteria were applied by Sanchez et al and were

shown to be more sensitive but less specific than the

ACR criteria (8).

An alternative statistical methodology, recursive

partitioning, was used by Edworthy et al (9). Recursive

partitioning or classification and regression tree (CART)

analysis is a computer-intensive method used to derive

a classification rule based on multiple candidate predictor

variables (10). The CART software package dichotomizes

variables based on all possible cut points.

The best discriminating cutoff is chosen for each variable.

Edworthy et al used the same data set as that used

for the 1982 ACR criteria but added 2 derived variables,

a standardized ANA variable and a “composite” complement

variable (9). In addition, analyses were performed

with the criteria for immunologic disorder divided

into components (anti–double-stranded DNA

[anti-dsDNA], anti-Sm, false-positive serologic test result

for syphilis) and the

...