Three Studies Advance Precision Medicine and Outcome Prediction in Childhood Lupus Nephritis

A machine learning-driven framework accurately predicts mycophenolic acid (MPA) exposure and supports individualized dosing in childhood-onset systemic lupus erythematosus with lupus nephritis (cLN), according to a study published in Rheumatology. This model outperformed clinical variables alone and offering a practical strategy to overcome limitations of therapeutic drug monitoring.

Investigators, led by Baojing Liu, MD, of the First Affiliated Hospital in Guangdong, China, assembled a real-world pharmacokinetic database comprising 1376 treatment cycles from 154 children with LN after data cleaning from an initial 1492 follow-up records in 155 patients. The investigators developed 3 complementary scenario-based models: Scenario 1 classified whether a child's MPA exposure was below, within, or above the therapeutic range. Scenario 2 estimated the precise 12-hour area under the curve (AUC) value. Scenario 3 forecasted drug exposure after a dose adjustment.

Across 3 clinically relevant scenarios, random forest (RF), LightGBM, and XGBoost algorithms demonstrated robust discrimination and calibration, and a web-based decision-support tool was developed to facilitate implementation. In Scenario 1, focused on therapeutic window attainment, XGBoost achieved the strongest Class 1 discrimination (AUC = 0.81). However, because Class 2 prediction was clinically prioritized, LightGBM offered the most balanced performance (AUC = 0.67; precision = 0.64; recall = 0.66; F1 = 0.65).

For precise AUC estimation (Scenario 2), models incorporating concentration timepoints substantially improved accuracy over clinical indicators alone (R² = 0.20–0.28; mean absolute error [MAE] = 8.51–9.00; root mean square error = 11.80–12.40). Sequential inclusion of the plasma concentration of MPA at 0.5, 1.5, 4, and 9 hours progressively enhanced performance, with the concentration at 1.5 hours emerging as the single most informative timepoint. The optimal 3-point strategy combining clinical indicators with plasma concentration at 0.5, 1.5, and 4 hours achieved R² of 0.84, MAE of 4.16, and root mean square error of 6.14.

In Scenario 3, predicting exposure after dose adjustment, RF again performed best (R² = 0.51; MAE = 20.57), with most predictions falling within 30% of observed values. Shapley Additive Explanations analyses identified dose, weight, renal function, and concentration timepoints as dominant contributors. Height and weight showed negative correlations with exposure, whereas most other variables were positively associated.

The resulting web-based application operationalizes these optimized models for real-time clinical use. By integrating pharmacokinetic sampling strategies feasible within a single outpatient visit, this approach provides the largest real-world pediatric LN analysis to date and advances precision dosing to better balance efficacy and safety in MPA-treated children.

In separate research, anifrolumab significantly reduces urinary biomarkers of renal histologic activity compared with placebo in patients with lupus nephritis (LN), even those without complete renal response (CRR) status, supporting drug-induced attenuation of intrarenal inflammation and possibly damage accrual. These exploratory results from the phase 2 TULIP-LN trial (Clinicaltrials.gov, NCT02547922) were published in Arthritis and Rheumatology.

These proteomic analyses from the phase 2 TULIP-LN trial demonstrate earlier declines in CD163 and MCP-1 with anifrolumab, particularly with the intensified regimen (IR), compared with standard therapy alone, Andrea Fava, MD, of Johns Hopkins Rheumatology, Baltimore, Maryland, and colleagues reported. Among 147 randomized participants in the trial, 112 were included in the urine proteomics dataset (anifrolumab basic regimen [BR], n=35; IR, n=42; placebo, n=35). The basic anifrolumab dose was 300 mg intravenous every 4 weeks, whereas the intensified regimen was 3×900 mg then 300 mg thereafter.

At week 12, anifrolumab IR reduced urinary CD163 and MCP-1 versus placebo (false discovery rate [FDR] <0.1; adjusted P =.07 for both). IL-16 and visfatin were detectable in 8% and 10% of IR-treated patients, respectively, compared with 26% and 32% with placebo (P =.05 and P =.03). By week 48, CD163 and MCP-1 concentrations declined across all groups, indicating that standard therapy also can suppress inflammation over time, albeit with slower kinetics.

An anifrolumab effect was observed even among patients without clinical renal response. Among nonresponders, anifrolumab BR (adjusted P =.04) and IR (adjusted P <.01) significantly reduced CD163 and MCP-1 from baseline to week 12 versus placebo. Similar findings were observed with combined BR+IR treatment (adjusted P <.01). These biomarker improvements were independent of conventional clinical response criteria.

Multiplex proteomic profiling of 107 proteins detected in ≥75% of samples further underscored anifrolumab's anti-inflammatory effects. At week 12, several markers of intrarenal inflammation, including B2M (log2 fold change, −1.6), myoglobin (−1.6), MMP3 (−1.1), BAFF (−0.7), and CD163 (−0.6), were higher in placebo-treated patients vs anifrolumab IR (all FDR <0.1). Only TRAIL-R3 was increased with IR vs placebo (log2 fold change, 0.5; FDR <0.1).

CD163 and MCP-1 correlated with urinary protein to creatinine ratio (UPCR) and renal activity, reflecting the intrarenal inflammatory milieu. Their earlier reduction with anifrolumab, particularly IR, suggests modulation of myeloid-driven renal inflammation beyond standard therapy alone, reinforcing the biologic rationale for type I interferon pathway inhibition in LN.

A third study found that failure to achieve modified primary efficacy renal response (mPERR) at 24 months independently predicts progression to advanced chronic kidney disease (CKD) in childhood-onset lupus nephritis (cLN). Attaining mPERR is associated with superior long-term kidney survival and fewer relapses, according to a study published in Kidney Reports.

Eugene Yu-Hin Chan, MBBS, MD, of The Chinese University of Hong Kong, and colleagues studied a cohort of 107 Chinese children with biopsy-proven cLN. Those with 24-month mPERR nonresponse had a more than 4-fold increased risk of progression to CKD stage 4 to 5, kidney failure, and/or death (CKD 4-5D/T) (adjusted HR, 4.39; 95% CI, 1.04-18.4; P =.04). Baseline estimated glomerular filtration rate (eGFR) less than 60 mL/min/1.73 m² also predicted advanced CKD or death (adjusted HR, 6.07; 95% CI, 1.21–30.39; P =.03). mPERR was defined as a UPCR less than 0.7 mg/mg and an eGFR at least 60 mL/min/1.73 m² or within 20% of baseline value.

Median patient age at LN diagnosis was 13.7 years, and 91% had proliferative LN. Eighteen patients had baseline eGFR less than 60 mL/min/1.73 m². Of the cohort, 87% were female. Over a median 12.0 years or 1287.9 patient-years, 8 individuals experienced the composite renal outcome.

At 24 months after biopsy, 87% achieved mPERR and 76% achieved modified complete renal response (mCRR), defined as UPCR less than 0.5 mg/mg and eGFR of at least 90 mL/min/1.73 m² or within 10% of baseline value. Only 87% of mPERR responders also met mCRR criteria.

In Kaplan-Meier analysis, higher rates of freedom from CKD 4-5D/T or death were observed among mPERR responders (100%, 98.8%, 98.8%, and 97.2%) compared with nonresponders (92.9%, 85.7%, 78.6%, and 78.6%; log-rank P =.013) at 1, 3, 5, and 10 years, respectively. Although mCRR achievement trended toward improved kidney survival, it did not reach statistical significance in multivariable modeling.

Relapse analyses reinforced the strong prognostic value of 24-month mPERR. Ten-year relapse-free survival was 57.2% for mPERR responders vs 21.4% for nonresponders. Recurrent relapses were more frequent in mPERR nonresponders (71% vs 22%; P = 0.001).

Neither sex, lupus nephritis class, crescents, thrombotic microangiopathy, age at diagnosis, treatment era, baseline proteinuria, nor SLE Disease Activity Index 2000 score influenced kidney survival. Although stringent criteria of complete proteinuria resolution (UPCR of 0.2 mg/mg or less) and eGFR of at least 90 mL/min per 1.73 m² were achieved in 55%, these criteria did not discriminate CKD 4-5D/T-free survival.