番茄社区

WASHINGTON -- Lupus nephritis (LN) diagnosis could improve greatly if early results with a multimarker urine test are confirmed in larger studies, research presented here suggested.

Measurement of 12 biomarkers in urine, yielding a numerical score for LN activity, led to an area under the receiver operating characteristic curve (AUC) of 0.90, whereas presence of proteinuria -- the usual means for detecting LN -- had an AUC of just 0.61, not much better than random chance, according to Andrea Fava, MD, of Johns Hopkins University in Baltimore.

Prediction of histologically active LN was also highly accurate by other measures, he reported at the American College of Rheumatology's (ACR) annual meeting:

• Sensitivity: 81%
• Specificity: 90%
• Positive predictive value: 87%
• Negative predictive value: 86%
• Overall accuracy: 86%

Other blood tests currently in use didn't come close to these numbers when applied to the same patient set. Levels of C3 complement, for example, yielded overall accuracy of 29%, while a test for anti-double-stranded DNA was 28% accurate. Proteinuria, defined as levels greater than 500 mg/day, was somewhat better, with 100% sensitivity and 53% positive predictive value, but its overall performance was dragged down by a total lack of specificity (0%) -- yielding accuracy of 53%.

Improvement in diagnostic performance is desperately needed, Fava told ACR attendees, because -- as the above figures show -- current tests are terribly inaccurate. He noted that an estimated 50% of all LN cases are initially asymptomatic and are usually detected through routine proteinuria screening. But, he said, proteinuria is a "lagging indicator" that "cannot distinguish treatable inflammation from chronic [irreversible] damage."

Moreover, resolution of proteinuria is a poor indicator of actual renal recovery. A large swath of LN cases in which proteinuria normalizes remain histologically active. By the same token, one study found that in histological remission still dump urinary protein.

And biopsies aren't always accurate. Typically, Fava said, initial biopsies are poor predictors of future morbidity. Most patients need a second biopsy to get a truer picture of flare risk and renal damage progression.

To find a better in vitro test for LN, Fava and colleagues analyzed samples from 179 patients with histologically and clinically confirmed LN, including 78 with highly active disease (NIH Activity Index >2) and 101 with less active LN (≤2). Clinical outcomes for these patients were known for up to 7 years following biopsy-confirmed diagnosis. The researchers measured some 1,200 urine proteins initially and used machine learning to winnow them down to 12 that, when combined in a proprietary algorithm, correlated most closely with LN activity and renal survival. Diagnostic performance was then evaluated in a validation set of samples.

Some of the 12 proteins included CD163, cathepsin-S, P-cadherin, complement C5a, interleukins 10 and 16, and catalase. CD163 alone contributed 20% to the overall accuracy.

Other encouraging findings with the test included that its numerical score decreased in patients whose proliferative LN responded to treatment, and that results predicted renal survival and estimated glomerular filtration rate (eGFR) declines over periods of up to 7 years. A positive test result ("high score") was associated with an approximately 70% mean decrease in eGFR at 84 months, while a negative "low" score meant almost no change for most patients. This was true regardless of whether patients were still showing proteinuria >500 mg/day at month 12 after biopsy.

Limitations to the study included the small patient cohort used to derive and validate the panel, such that its performance in subgroups (e.g., stratified by age, sex, race/ethnicity, comorbidities, etc.) could not be evaluated. Also, the extent to which the validation set was independent from the initial test set was not reported.

Comment