Diagnosing and Treatment of Mixed Autoinflammation and Autoimmunity Disorders: Challenges, Solutions, and AI-Era Outlooks

1. The Special Niche of Mixed Autoinflammation and Autoimmune Disorders

In the earlier days in the field of immunology, autoinflammatory and autoimmune diseases were considered separate immune disorders. Therefore, concurrence of autoinflammation and autoimmune disorders was regarded as a distinct entity and could represents as “salad” type of disease mixture in the patient. Autoinflammation is defined by dysregulation of innate immunity characterized by an imbalanced cytokine profile and damaged tissues in patients without clear evidence of breakdown of immune tolerance (autoimmunity) [1,2]. Autoimmune diseases on the other hand entail the loss of immune tolerance, including expansion of activated autoreactive T cells, B cells, and autoantibodies that targets endogenous antigens or autoantigens that are detrimental to multiple organs [3]. Although autoimmunity can also induce inflammations, the resulting inflammations induced by autoimmunity is predominantly driven by adaptive immunity rather than innate immunity. Inflammation driven by autoinflammation on the other hand has the following traits: 1) genetic mutations/polymorphisms in innate immune pathway genes [4], 2) recurrent fever [5], 3) neutrophilia [5].

Since autoinflammation and autoimmunity often involves the activation of the immune system, characterized by elevated levels of cytokines and immune cell activation, identifying whether a patient has only autoimmunity and/or autoinflammation can be challenging. In particularly, patients with a known autoimmune diagnosis may not be elevated for concomitant autoinflammation in a timely manner [6,7,8,9]. This is due to the observation that autoimmune conditions, driven by activation of adaptive immunity, often simultaneously activate innate immunity, which is the main player for autoinflammation.

Conceptually, mixed autoimmune-autoinflammation disease phenotypes could be subdivided into two categories: 1) a combination of multiple autoimmune and autoinflammation diseases caused by separate environmental and/or genetic factors, or 2) concurrent autoimmune and autoinflammatory disease that stem from the same sets of genetic defects. The first category often requires separate efforts to diagnose and treat both autoinflammation and autoimmunity, whereas the second category is essentially one disease once the genetic cause(s) have been identified.

Dozens of patients who separately acquired both autoimmunity and autoinflammation have been described in clinical case reports [6,7,8,9], in which they were diagnosed with two or multiple autoimmune and autoinflammation diseases immune disorder. Since both autoimmune and autoinflammation must be accurately diagnosed and treated in order to achieve remission for both conditions, reaching to the correct diagnostic conclusion can easily entail multiple years of efforts, multiple rounds of diagnosis and courses of therapy for the clinicians to bring the patients to remission [9]. Manifestations of “salad-type” of autoimmune and autoinflammations can be seen in certain cases of co-existing autoinflammation and autoimmune diseases between rheumatoid arthritis (RA), familial cold autoinflammatory syndrome (FCAS), familial mediterranean fever (FMF), systemic lupus erythematosus (SLE), and Sjögren’s syndrome [6,7,8,9], (see Table 1). Beyond the published case reports, it is likely that additional patients with mixed autoimmune-autoinflammatory phenotypes reman under-recognized or misdiagnosed in route practice.

On the other hand, mixed disease in which genetic causes simultaneously underlie both autoimmunity and autoinflammation can be best illustrated by Omenn syndrome and Helios deficiency in the umbrella disease of inborn errors of immunity (IEI), where more than one in three patients can have concurrent autoinflammation and autoimmunity manifestations [10]. Clinically, IEI is regarded as a group of genetic disorders that jeopardize the maturation and function of the immune system, and approximately 3% of IEI patients have both autoinflammation and autoimmune manifestations [10]. Typical therapies of many subgroups of the disease resemble those used in “classical” autoinflammation and autoimmune disorders. However, since the root cause of IEI can be traced to genetic mutations, meaningful remission often require reconstruction the immune system by using allogenic hematopoietic stem cell transplantation (HSCT) or emerging combinations of cell therapy plus gene editing [11,12].

Therefore, mixed autoimmune-autoinflammation caused by genetic factors are often easier to root out the cause. For instance, Helios deficiency is caused by loss-of-function in the IKZF2 gene, which plays key roles in regulatory T cells and effector T cells [13]. On the other hand, the mixed autoimmune and autoinflammation triggered by both genetic and environmental factors can be substantially more challenging to diagnose and to bring into durable remission.

In this paper, we will be discussing about the strategies in diagnosing and treating mixed types of autoinflammation and autoimmune diseases with the currently available medications. We also provide an outlook on utilizing AI/ML (artificial intelligence and machine learning) methods for developing diagnostic and prognostic models for autoinflammation and autoimmune diseases.

2. Diagnosis of Mixed Autoimmune-Autoinflammation Diseases

Patients with mixed autoimmune-autoinflammation necessitate more careful diagnostic process, as conventional diagnosis may miss either the autoimmune or the autoinflammation part of the disease. In published clinical practices, the diagnostic workup of autoinflammation diseases generally includes several steps: 1) observation of recurrent fever and/or flares and exclusion of noninflammatory disorders, 2) deep phenotyping and detailed family history investigation, and 3) targeted genetic testing [14,15]. Similarly, the diagnosis of autoimmune diseases involve: 1) review of history physical exams, 2) histological and baseline serological test, 3) disease-specific autoantibody tests, and 4) genetic, cytometric, and cytokine markers [16,17]. These stepwise diagnosis approaches represent workflows that were more frequently adopted in the 2010s in many nations, when the more expensive biomarker and cross-validations tests were not preferred as the first-line investigations.

In addition to imaging tools and endoscopic procedures, lab-based tests are now prescribed at an earlier phase due to their relative affordability. For instance, in the case reports by Savic, et al. [6] of patients with mixed autoimmune and autoinflammatory rheumatoid arthritis, all patients displayed classical signatures of autoimmune type of RA, with typical autoimmune RA markers of ACPA positivity (Anticitrullinated peptide antibodies), RF (rheumatoid factor), and elevated CRP (C-reactive protein) [18]. Moreover, all the patients also had neutrophilia, a hallmark of autoinflammation [19], and some patients had elevated cytokine levels, such as IL-1beta, TNF, and IL-6 [6,20]. Similarly, for SLE patients, positivity for autoantibodies is typically considered to be a key diagnostic marker. Nevertheless, in clinical observation, a patient with mixed autoimmune-autoinflammation can be positive for certain autoantibody, such as anti-nuclear antibody and anti-U1 RNP antibody but not positive for other autoantibodies, including anti-dsDNA, Smith, SS-A, SS-B [8], so the diagnostic marker panel needs to be as comprehensive as possible in order to avoid false negative diagnosis.

While conventional stepwise diagnostic workflows may appear cost-effective, when encountering a patient with a more complex mixed autoimmune-autoinflammation disorder, multiplex biomarkers panels are often required so that the end result can differentiate between a mixed versus single autoimmunity or autoinflammation phenotype. Retrospectively, elevation of these innate immune system-related cytokines could be seen as supporting evidence for the presence of autoinflammation, but it should be noted that IL-6 [21], IL-1beta [22], and TNF [23] elevation can also present in purely autoimmune form of RA, as crosstalk between innate and adaptive immune system allows autoimmune responses to activate innate system as well.

In order to validate the coexistence of autoimmunity and autoinflammation, genetic tests can serve as a supplementary proof for identifying the underlying drivers of these cytokines. Clinicians treating mixed autoinflammation-autoimmune disease often perform genetic tests in these patients, especially for genes like TNFRSF1A, NLRP3, MEFV and NOD2, which regulate the release of proinflammatory cytokines IL-1beta, IL-6 and TNF, and most importantly, are highly correlated with hereditary autoinflammation diseases [6]. These genetic testing results have eventually helped the clinicians to reach a more accurate diagnosis.

3. Current Standard Treatment of Mixed Autoimmune-Autoinflammation

Arguably, the difficulty in accurately diagnosing and identifying the subtypes of the autoimmune and autoinflammatory disease is often the most challenging part, compared to selecting the treatment strategy. Upon recognizing the mixed nature of the immunological disorder in the patients, the next step is to identify proper treatment regimens.

Broadly speaking, medications used in mixed autoinflammation and autoimmunity can be categorized into five groups: 1) broad anti-inflammatory and symptom controllers, such as corticosteroids and non-steroidal anti-inflammatory drugs (NSAIDs), 2) innate immune and inflammasome targeted agents, 3) conventional adaptive immune disease-modifying antirheumatic drugs (DMARDs), 4) targeted biologic DMARDs, such as anti-TNFs, anti-IFNs, vedolizumab, B-cell depleting agents, dupilumab, and 5) direct T-cell modulators.

Some of the cytokine modulating drugs, such as IL-1 blockade therapy, can address not only autoinflammation but also selected suspected autoimmune conditions. IL-1 is a key driver in autoinflammatory disease, expressed by monocytes, macrophages and neutrophils [24], and anti-IL-1 therapies have shown efficacy against monogenic and some polygenic autoinflammatory diseases [24]. A clinical case report [6] has shown that in cases of mixed autoimmune-autoinflammatory, seropositive (positive for autoantibodies), rheumatoid arthritis patients have poor responses to anti-rheumatic drugs (DMARDs) but show excellent response to colchicine or IL-1 pathway blockers. Genetic analysis has shown that responders to IL-1 therapies have mutations or SNPs occurring in autoinflammatory pathways. For instance, a patient with NLRP3 SNP (c973C>T or pArg325Trp), an autoinflammation-related gene responsible for cryopyrin associated periodic syndrome (CAPS), has peripheral blood mononuclear cells (PBMCs) that produce higher levels of IL-1β, and treatment with IL-1β-targeted drug anakinra was associated with complete response [6].

In adult-onset Still’s disease (AOSD), both autoinflammatory and autoimmune disorder phenotypes are present. While the process of ASOD involves extensive release of IL-1β and IFN-γ, which contributes to necrosis of keratinocytes due to innate system including neutrophils and macrophages, ASOD patients can also have elevated IL-4 and IFN-γ producing T cells and reduced level of Tregs which also correlates with disease severity [25]. Drugs targeting IL-1, IL-6, and IL-18 have high partial and complete response rates [26,27,28]. When the efficacy of this anti-interleukin therapy is poor in certain refractory AOSD cases, B-cell depleting drug rituximab was able to be effective and well tolerated [25,29], suggesting these cases are also autoimmune-related.

4. Severe Case Treatment: Depletion and Reconstruction of Immunity

In patients with highly aggressive autoimmune disease who fail conventional biologic or small-molecule therapy, a more radical strategy is to ablate the existing immune repertoire and allow it to rebuild from hematopoietic stem cells, effectively attempting an ‘immune system reset. One of the prototypical example of immune “depletion and reconstruction” in severe autoimmunity is the immunoablation–autologous HSCT protocol developed at The Ottawa Hospital for patients with early, aggressive multiple sclerosis (MS). In this phase 2 (single-arm Canadian trial), near-complete immune-ablation with busulfan, cyclophosphamide and rabbit anti-thymocyte globulin was followed by reinfusion of immune-cell–depleted autologous CD34⁺ hematopoietic stem cells, with the explicit goal of eradicating autoreactive lymphocyte clones and allowing de novo immune reconstitution [30,31]. Among 24 patients, the 3-year MS activity-free survival (no relapses, no new MRI lesions, and no disability progression off disease-modifying therapy) was approximately 70%, and extended follow-up up to 13 years showed complete suppression of inflammatory disease activity on clinical and MRI measures, with about one-third of patients experiencing sustained improvement in disability scores despite having highly aggressive disease at baseline. These data, together with larger observational cohorts of autologous HSCT in MS showing multi-year progression-free survival in a substantial fraction of carefully selected patients, support the concept that profound but time-limited ablation of both innate and adaptive immune compartments, followed by hematopoietic reconstitution, can functionally “reset” pathogenic immune networks in a subset of severe autoimmune disorders [32] .

Reconstructing the immune system may happen either spontaneously or with the need of transplantation of hematopoietic stem cells. The former case typically applies to those patients whose bone marrow has experienced less toxicity from the ablation drugs, whereas those who experienced substantial bone marrow toxicity and those who have strong genetic risk of disease more often need HSCT [33]. For instance, patients with poor prognosis for multiple sclerosis treated with autologous CD34 HSCT showed nearly 70% achieving activity-free survival at 3 years after transplantation [34]. Autologous transplantation is generally preferred when the patients have severe autoimmune disease, but their genetic risk factors are minor so that their immune system can be reeducated [33,35,36], whereas autoinflammation patients with strong genetic linkage more often need allogenic HSCT in order to replace the mutant hematopoietic cells, even at the risk of graft versus host disease [37].

Ablation of the immune cells precedes the reconstitution of the immune system. Treating autoimmunity can require more intense ablation than treating autoinflammation. As recommended by the European Society for Blood and Marrow Transplantation [36], ablation of the immune system for treating autoimmunity can involve usage of anti-thymocyte globulin, anti-B-cell antibody rituximab, and chemotherapy that together would deplete the majority of factions of lymphoid and myeloid lineage cells from both adaptive and innate immune systems. Theoretically, depleting both innate and adaptive immune cells allows for removal of pathogenic cells that are responsible for autoimmunity or autoinflammation. Removal of T cells not only eliminates the autoreactive clones but also reduces the risk of immune rejection if source of HSCT is allogenic. B-cell depletion removes autoreactive clones, and eradication of myeloid cells removes cells that are highly inflammatory and are responsible for autoinflammation. Therefore, such methodology can be particularly important for patients with severe disease states from multiple genetic root causes [11].

When biologics and small molecule drugs are not efficacious, more aggressive immune system reset therapy, such as hematopoietic stem cell transplantation, B-cell targeted therapy and T-cell targeted therapy, may be necessary to rescue the patients’ life. When acting alone, many of these therapies have high relapse rate, such as cyclophosphamide (chemotherapy, 40-60%), belimumab (anti-BAFF, 60%-70%), rituximab and obinutuzumab (anti-CD20, 50%-60% and 60-70%, respectively) [38], due to incomplete depletion of pathogenic cell populations or due to the fact that inhibition is reversible. Due to these reasons, a substantial portion of patients ultimately experience relapse, highlighting the need for more effective ablation regimen to achieve a true “immune rest”.

B-cell depletion therapy marks one of the increasingly promising avenues for treating autoimmune disease even for patients with strong genetic predisposition for autoimmune disease that are strongly B-cell driven, even when the disease is genetically related. For instance, even with the high relapse rate of rituximab, its therapeutic outcome indicates there are no strong prognostic factors (genetic and environmental) that can be used to predict the efficacy of traditional IgG rituximab in treating SLE [39]. On the other hand, the mechanism of relapse for B-cell targeted therapy can be attributed to incomplete depletion of autoreactive B cells. If the coverage of autoreactive B cells were sufficiently comprehensive, relapse rates would be expected to decrease substantially.

Clinical data shows that complete versus incomplete depletion of CD19+ B cell by rituximab results in relapse rates of 17% and 60% (p=0.02), respectively [40]. This observation gives rise to the drug development strategy of targeting B cell surface markers that span all stages of the B cells lifespan, such as the combination of CD19 and BCMA that covers B-cell stage from pro-B cell to plasma cells [41]. Relapse in many B-cell-depleting monospecific CAR-T-treated autoimmune patients may experience late relapse after 18-24 months as an indication of non-permanent reset of immune system [38]. To address this issue, multi-specific CAR-T constructs have been tested in clinical trials to address antigen escape phenomenon seen in monospecific CAR-T. For instance, a phase I result of a CD19 x BCMA CAR-T [42] has reported that 10 out of 12 refractory SLE patients achieved stringent complete remission defined by medication-free DORIS CR and complete renal response (CRR). CAR-T therapy in general has excellent efficacy for autoimmune diseases such as SLE [38,43], systemic sclerosis (SSc) and idiopathic inflammatory myopathies (IIM) [44]. Since mechanistic research and clinical outcome of CAR-T are mostly sourced and extrapolated from oncology research, more clinical data are needed in order to dissect and address the relapse patterns seen in autoimmune disease clinical landscape.

T-cell depletion therapies, such as anti-CD52 alemtuzumab, have demonstrated significant efficacy in autoimmune diseases with predominantly T-cell driven pathogenesis like multiple sclerosis, achieving annual relapse rate (ARR) of 17% from year 3-13 [45] that are similar to B-cell depletion therapies like ofatumumab (9% [46]), but higher than standard therapy like IFN-β-1a (29% [47]). Compared to approved B-cell depletion therapies, alemtuzumab is known to cause substantial secondary autoimmunity such as thyroid disease (17%-26% by different clinical trials) [48]. For this reason, safer T cell depleting reagents like rabbit anti-thymocyte globulin (ATG) are more widely used and recommended [33,35,36].

5. The Ongoing Trend and the Future: Remapping Understanding and AI

Advancement of bioinformatics and artificial intelligence (AI) have enabled identification of genes that are responsible for immune disorders. With these tools, in contemporary immunology, there are growing support [1,49,50] for the theorization of “autoimmune and autoinflammatory diseases as a continuum”, where autoinflammation and autoimmune diseases could be redefined by their genetic and cellular level involvement with innate and adaptive immune systems [51]. This theory provides a foundation for application of method like multi-omics and flow cytometry for more promising diagnosis and identification biomarkers.

To genetically blueprints the immune disorder landscape under the continuum theory, many researchers have attempted to genetically explain immune disorders. For instance, Fominykh and collogues [51] have conducted a Genome-Wide Association Study (GWAS) for 15 immune disorders and proposed a four-factor system, including factor 1 polygenic autoimmune (e.g., IL-12 related pathways, TCR signaling), factor 2 autoimmune-mixed (zero gene sets no unique pathway), factor 3 mixed pattern (16 gene sets including enriched T-cell, B-cell, JAK, IL-18, and innate immunity pathways), and factor 4 polygenic autoinflammatory (41 gene sets with enrichment for IL-12/IL-23, IL-3, IL-4, IL-18, IL-24, T-helper, and IFN pathways). The factor 3 mixed pattern is a particular designation for mixed patterns of autoinflammation and autoimmune diseases [51]. Interestingly, one of the gene sets highly associated with the factor 3 group, named GOBP_INNATE_IMMUNE_RESPONSE, contains genes NLRP3, MEFV and NOD2, whose pathogenic mutations were identified in earlier clinical case studies due to their key roles in controlling inflammasome and oxidative stress (Table 1) [6,8,9]. Despite the progress, since there are hundreds of genes in the same category, more in-depth analysis will be needed for establishing a sufficiently narrow enough gene panel that can be clinically translatable. Autoimmune diseases also have their groups of genes that serve as risk factors, such as those in the HLA locus 6p21, CTLA-4, and PTPN22 [52,53].

Similar to GWAS, when combined with biostatistics and machine learning algorithms, flow cytometry holds promise in generating clinically meaningful diagnostic biomarker panels to significantly improve the accuracy of the diagnosis to address unmet needs in accurate diagnosis. In the diagnosis of SLE, a combination of relative mean fluorescence intensity (rMFI) of CD169 on monocytes, CD177 on neutrophils, and CD317 on B cells in a binary logistic regression model has achieved an excellent area-under-the-curve (AUC) of 0.9243 [54]. Interestingly, the flow cytometry model (sensitivity = 82.1%, specificity = 97.3%) also significantly outperformed anti-dsDNA (AUC=0.7650, sensitivity = 53.0%, specificity = 100%) [54], indicating that the flow method can address the false negative issues seen commonly in anti-dsDNA testing [55]. In addition, there are also growing applications of using flow cytometry to subtype systemic autoinflammation. For example, the usage of DNT, B220+, HLA-DR and CD25 panel can help identify subgroups of autoimmune lymphoproliferative syndrome (ALPS) patients with undefined autoimmune or autoinflammatory disorders [56].

One of the other techniques which has revolutionized the identification, classification, and treatment follow-up of autoinflammatory diseases — particularly those for which no conventional diagnostic test previously existed — is the NanoString-based Interferon Response Gene (IRG) scoring approach [57]. Unlike conventional qPCR-based interferon assays, which are typically limited to 5–6 target genes and suffer from poor inter-center standardization, NanoString technology uses molecular fluorescent barcodes and a digital optical scanner to simultaneously and directly count up to several hundred unique target gene sequences in a single hybridization reaction without requiring amplification, enabling highly reproducible, batch-to-batch comparable results suitable for longitudinal clinical monitoring [57] . This platform was leveraged to develop and validate a standardized 28-gene IRG score — calculated either as a sum of Z-scores relative to healthy controls or as a geometric mean of normalized gene counts — to reliably distinguish patients with presumed IFN-driven pathogenesis, such as Mendelian autoinflammatory interferonopathies CANDLE (Chronic Atypical Neutrophilic Dermatosis with Lipodystrophy and Elevated temperature) and SAVI (STING-Associated Vasculopathy with onset in Infancy), from patients with IL-1-mediated autoinflammatory disease (NOMID), where IRG scores are expected to be low [57]

The application of NanoString-based IRG scoring has since been extended beyond the monogenic interferonopathies to broader autoimmune and autoinflammatory conditions, including juvenile dermatomyositis (JDM) — an autoimmune disease with a prominent interferon signature for which no validated molecular diagnostic test had previously been available. Kim and colleagues[58]. Systematically evaluated the 28-gene NanoString IRG score in 57 JDM patients stratified by myositis-specific autoantibody (MSA) subgroup and compared them to CANDLE, SAVI, NOMID, and healthy controls. IRG scores in JDM patients were significantly higher than in healthy controls and NOMID (the IL-1-mediated negative control), but overall lower than CANDLE or SAVI, suggesting that JDM occupies an intermediate position on the autoimmune-autoinflammatory continuum with a substantial IFN-driven component[58] .Importantly, principal component analysis revealed that the IRG gene expression pattern in JDM overlapped more closely with SAVI than CANDLE, implicating the STING pathway and IFN-β as potential contributors to JDM pathogenesis — particularly in the anti-MDA5 autoantibody-positive JDM subgroup, whose IRG scores were notably similar to SAVI patients [58]. In anti-TIF1 autoantibody-positive JDM, the IRG score correlated strongly with skin disease activity [58].

Building on this foundational work, our group and collaborators have further demonstrated the longitudinal clinical utility of the NanoString IRG score as a dynamic biomarker for disease activity and treatment response in JDM [59]. In a prospective cohort study of 43 JDM patients enrolled in the SickKids CARD biobank with 87 longitudinal biosamples — including 18 treatment-naïve subjects sampled at diagnosis — a strong correlation between IRG score and the modified Disease Activity Score (mDAS) was demonstrated (Spearman rs=0.78, P ≤ 0.001 overall; rs = 0.63, P = 0.005 in treatment-naïve patients), with this relationship maintained longitudinally by linear mixed model regression (β = 0.004, P < 0.001) [59]Critically, the IRG score exhibited a higher standardized response mean (SRM 0.9) than conventional biomarkers including CMAS (0.6), CHAQ (0.5), and muscle enzymes (0.6), and normalized more rapidly following treatment initiation than the mDAS itself, indicating that the NanoString IRG score is more sensitive to early immunological changes than current standard tools Treatment-naïve patients at disease onset exhibited the highest IRG scores (mean 1085 in JDM with muscle disease), and the score declined rapidly and reliably with effective therapy, underscoring its potential as a guide for therapeutic escalation, de-escalation, and early flare detection [59]Taken together, these studies establish NanoString-based IRG scoring as a transformative molecular diagnostic and monitoring tool across the autoinflammatory-autoimmune spectrum, capable of simultaneously classifying disease mechanistic subtype (IFN-mediated versus IL-1-mediated), stratifying patients by disease severity and autoantibody profile, and dynamically tracking treatment response in conditions where no validated biomarker had previously existed.

Besides identifying markers for diagnosis of a single given disease, flow cytometry could also be used for categorizing diseases based on their similarities at the cellular levels, potentially offering translational insights for development of drugs that have higher target specificity as well as a broader applicable scope of indications. For instance, Tchitchek, et al. [60] performed deep phenotyping with 13 flow cytometry panels from blood samples of 443 patients with autoimmune or autoinflammatory diseases. LAG3+ and ICOS+ T cells within the regulatory T cells (Tregs) compartment, key players acting as checkpoint inhibitor for MHC-II to inhibit T cell activation, were highlighted as particularly important [60]. LAG3 is an immune checkpoint receptor that binds to MHC-II for inhibiting T cell activation, although the function of LAG3 (lymphocyte-activation gene 3) in Treg remains largely unknown [60,61]. ICOS (inducible T cell co-stimulator) is an important co-stimulator molecule for T cell activation, and its downregulation on Treg can suggestion can suggest decreased ability of the Treg to preserve immune tolerance and prevent autoimmunity [60,62].

Cytokines and chemokines are often regarded as key players in autoimmunity and autoinflammation, and chemokines can also play significant roles, especially when serving as diagnostic markers. For instance, in SLE patients, levels of type 1 interferon-regulated chemokines are elevated relative to healthy controls [63,64]. A biomarker validation study with 267 patient samples helped to identify a composite score derived from CXCL10/IP-10, CCL2/MCP-1, and CCL19/MIP-3b, which was able to demonstrate a competitive advantage over traditional assays, such as at flare onset vs pre-flare (vs anti-dsDNA, complement), assessment of improvement (C4), and prediction of future flares (anti-dsDNA, C3, C4, ESR) [65]. In addition, the significance of chemokine as clinical biomarker has also been investigated in Behçet’s disease [66], multiple sclerosis and rheumatoid arthritis [67].

When powered with AI and machine learning, cytokines and chemokines can become particularly promising biomarker building blocks for diagnostic and prognostic models. In general, AI is widely recognized as the future for many subject matters. Many encouraging modeling works now hold promises to diagnose and predict progression of immune disorders (see Table 2). In this context, mixed autoimmunity–autoinflammation represents an especially compelling test bed for AI-enabled cytokine and chemokine–based models, where multi-analyte panels and diverse algorithms (e.g., random forests, gradient boosting, SVMs, and multimodal architectures) can be trained to disentangle innate–adaptive signatures and more accurately classify, subtype, and predict outcomes across this continuum of immune-mediated disease [68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85]. At least over 20 publications have AI or machine-learning models that achieved statistically and mathematically meaningful performance in diagnosis and prognosis of autoimmune diseases [86]. Here are numerous reviews that are available for the reader’s reference [86,87,88,89].

Compared to unimodal or monomodal AI/ML models, multi-modal AI brings a step further towards precision medicine that leverages more comprehensive input from patient health information. Traditionally, clinical integration of multi-dimensional data relies on low-throughput manual synthesis. Recently, unimodal AI models have been gaining significant popularity but have several insufficiencies: 1) information loss or representation gaps, 2) lack of cross-functional or complementary validation, and 3) lack of generalizability to patients at different disease stages. Multiple submodules or AI modules are gaining popularity by addressing limitations of monomodal AI models. They take considerations of higher dimensions of input from the patient, such as radiology, pathology, serological, cytometric, genomics, metabolomics, wearable device output, clinical notes, electronic health record, etc. [90,91]. Inherently, when output from one model is inconclusive (e.g., serological level of autoantibodies), other modalities (e.g., genomic, family medical history, cytometric profiling), can help to disambiguate [90,91]. Compared to monomodal AI, multi-modal AI can also have: better calibration (alignment of prediction vs observation, such as for Kaplan-Meier analysis) [92,93], and more accurate stratification of disease subtype or heterogeneity [90,91,92,93]. For instance, an AI scoring system called Genetic progression score (GPS) built upon GWAS and electronic health record data has yielded 25-1000% more accurate risk prediction than 20 existing models for autoimmune diseases [94]. Multi-omics AI framework have been built for molecular stratification for SLE, primary Sjögren’s syndrome, and RA [95].

While all of the abovementioned research advances are encouraging, it should be noted that many challenges remain for establishing diagnostic and prognostic biomarkers. Many infections and oncology conditions can also cause elevated levels of cytokines and chemokines [96,97,98], which may serve as confounding factors for interpreting inflammatory flares. Variation in patient parameters, sample processing procedures, and analytical specifications may affect data quality and therefore requires rigorous quality control. Critical evaluation must be done by data and modeling scientists before feeding data into training AI models to avoid “garbage in, garbage out” scenario caused by low-quality inputs fed into the training process. Moreover, the next generation multi-modal AI model paradigm requires large volumes of training data and will likely needs more strenuous efforts in validation and attainment of regulatory approval.

6. Conclusions

Treatment of mixed autoimmune and autoinflammation diseases has gained significant progress as available treatment options from biologics and reconstruction of the immune system have becomes increasing available and sufficiently efficacious. On the other hand, timely and cost-effectively achieving accurate diagnosis and pharmacogenomic precision in treatment planning remains challenging. Emerging biomarker panels and AI/machine learning tools hold promise to empower precision medicine by integrating multi-dimensional information from the patient into clinically actionable diagnostic and therapeutic decisions.