Identifying heterogeneous subgroups of systemic autoimmune diseases by applying a joint dimension reduction and clustering approach to immunomarkers

doi:10.21203/rs.3.rs-4311101/v1

Download PDF

Research Article

Identifying heterogeneous subgroups of systemic autoimmune diseases by applying a joint dimension reduction and clustering approach to immunomarkers

https://doi.org/10.21203/rs.3.rs-4311101/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 16 Sep, 2024

Read the published version in BioData Mining →

You are reading this latest preprint version

Background

The high complexity of systemic autoimmune diseases (SADs) has hindered precise management. This study aims to investigate heterogeneity in SADs.

Methods

We applied a joint cluster analysis, which jointed multiple correspondence analysis and k-means, to immunomarkers and measured the heterogeneity of clusters by examining differences in immunomarkers and clinical manifestations. The electronic health records of patients who received an antinuclear antibody test and were diagnosed with SADs, namely systemic lupus erythematosus (SLE), rheumatoid arthritis (RA), and Sjögren’s syndrome (SS), were retrieved between 2001 and 2016 from hospitals in Taiwan.

Results

With distinctive patterns of immunomarkers, a total of 11,923 patients with the three SADs were grouped into six clusters. None of the clusters was composed only of a single SADs, and these clusters demonstrated considerable differences in clinical manifestation. Both patients with SLE and SS had a more dispersed distribution in the six clusters. Among patients with SLE, the occurrence of renal compromise was higher in Clusters 3 and 6 (52% and 51%; p < 0.001) than in the other clusters. Cluster 3 also had a high proportion of patients with discoid lupus (60%; p < 0.001) than did Cluster 6 (39%). Patients with SS in Cluster 3 were the most distinctive because of the high occurrence of immunity disorders (63%) and other and unspecified benign neoplasm (58%) with statistical significance compared with the other clusters (all p values < 0.05).

Conclusions

The immunomarker-driven clustering method could recognise more clinically relevant subgroups of the SADsand would provide a more precise diagnosis basis.

autoimmune diseases

immune markers

cluster analysis

disease heterogeneity

Systemic autoimmune disease (SAD) is an umbrella term for autoimmune diseases that could affect all body systems and organs. SADs are typically divided into major categories, such as systemic lupus erythematosus (SLE), rheumatoid arthritis (RA), and Sjögren’s syndrome (SS). However, patients with specific SADs present with considerably varying manifestations. Regarding the initial presentation of SLE, a molar rash, pleural effusion, or even septic shock can be the first manifestation of SLE [1]. Moreover, severity and prognosis considerably vary among patients with the same SAD [2]. Some patients meet sufficient criteria and receive a clear diagnosis of SAD, whereas other patients fail to meet criteria required for the diagnosis [3]. The high complexity of SADs has hindered their precise management.

To provide more meticulous treatments for SADs, detailed identification of heterogeneous SAD subgroups would be the key first step for developing tailored healthcare strategies. Regrouping or reclassification of various well-known diseases has attracted considerable interest. Diabetes, among the most prevalent diseases, is classified into four categories, not only type 1 and type 2, based on the traditional classification [4]. Myocardial infarction, another major disease with a high prevalence, is classified into five types based on pathological, clinical, and prognostic differences, and treatment strategies vary for these five types [5]. An update of the classification scheme enhances clinical usefulness and establishes an accurate diagnosis [6]. Due to their complex nature, the current classification scheme for SADs does not meet the clinical need.

Redefining SADs is an ongoing task in the rheumatology community. European and American committees established new classification criteria for SLE in 2019 [7]. In addition to experts’ consensus, antinuclear antibody (ANA) testing and data-driven methods have been employed to refine the definition of SLE [7]. In fact, in the realm of disease diagnosis (e.g., autoimmune diseases, cancers), relying solely on individual biomarkers often proves inadequate. To counter this limitation, utilizing a biomarker panel consisting of multiple markers has shown substantial potential for enhancing diagnostic accuracy [8]. However, due to the intricate and implicit numerical patterns within such biomarker panels, the incorporation of machine learning and artificial intelligence technologies becomes pivotal in deciphering these specific disease patterns, thereby further amplifying diagnostic precision [9, 10]. On top of the approach, using numerous objective measurements (e.g., biomarker testing) with data-driven analytical methods has been advocated as a more adequate approach for redefining SADs. Using data-driven analytical methods, such as unsupervised clustering, studies have demonstrated that the heterogeneity of autoimmune diseases can be efficiently deconstructed to obtain clinically meaningful insights [11–17]. A study identified distinct groups of patients with anti-Ku syndrome developing different types of severe comorbidities using clinical features [16]. Another study classified patients with particular SADs by performing cluster analysis based on principal components (PCs) derived from a multiple correspondence analysis (MCA). Although dissimilarities were observed in the positive rates of autoantibodies and the frequencies of SADs between clusters, the association between clusters and clinical characteristics remains unclear [17].

This study investigated heterogeneity in SADs, attempting to build an algorithm to distinguish disease subgroups by applying a method that jointed dimension reduction and clustering to immunomarkers and evaluated heterogeneity based on clinical manifestations.

2.1. Study population and setting

This study retrieved electronic medical records from the Chang Gung Research Database between 2001 and 2019, containing clinical data from three medical centres and five regional hospitals in Taiwan [18]. We included patients receiving a principal diagnosis of SADs (Supplement Document 1) between 2001 and 2016 and had at least one ANA test.

2.2. Patient inclusion algorithm

Figure 1 presents a flowchart of patient inclusion rules. We identified patients receiving at least two diagnoses of a specific SAD 30 days apart within 365 days [19]. Then, we included patients having at least one immunomarker result within 90 days before and 30 days after the first diagnosis date to minimise the bias from treatments that immunomarkers might be subject to. To have an adequate number of patients for each target SAD, we included only patients diagnosed with common SADs, namely SLE, SS, and RA, for the cluster analysis and excluded those with multiple SADs. Finally, patients having at least three years of follow-up from the first diagnosis were retained for a clinical implication analysis.

2.3. Cluster analysis

We converted the results of 33 immunomarkers (Supplement Document 2) into categorical variables. After preprocessing and imputing immunomarker data (Supplement Document 2), we performed a cluster analysis using MCA k-means. This method joints dimension reduction and cluster analysis by combining MCA with k-means in a unified framework. By contrast, the tandem approach that involves first determining PCs and then clustering patients using these PCs with clustering algorithms has an inherent problem where the first few PCs do not necessarily define the most informative subspace of a genuine cluster structure of the original dataset. Thus, we employed the joint approach to solve this problem.

Details of feature selection and parameter tuning are in Supplement Document 3.

2.4. Assessment of clinical implications

To evaluate the performance of the cluster analysis in aggregating patients into clinically significant subgroups, we compared the occurrence of clinical manifestations after SAD diagnosis between groups for patients who were followed for at least 3 years (Fig. 1) .The patients who had a follow-up period of of less than 3 years were excluded at this stage to trace the occurrence of occurrence of clinical manifestations after SAD. Clinical manifestations were categorised in two ways: (1) typical or severe manifestations observed in RA, SLE, and SS, defined in Supplement Table 1, and (2) Clinical Classifications Software (CCS) categories, considered to comprehensively screen for clinical manifestations that could be rare and associated with a specific group. Patients’ manifestations might be subject to medical treatments; because of the convention that treatments are conditional on diagnoses for the reimbursement purpose of the National Health Insurance (NHI), the temporality of manifestations and treatments is favourable in our study.

Details of the statistical analysis used in this study are described in Supplement Document 4. R scripts for the analyses above are available at https://github.com/DHLab-TSENG/Heterogeneity-in-SAD-paper.

3.1. Demographic characteristics and immunomarker tests

We profiled the data of patients with SS (n = 3,541), RA (n = 5,696), and SLE (n = 2,686) (Fig. 1 and Table 1). The patients with SLE (mean age = 33.9 ± 17.5 years) were younger than those with RA (51.0 ± 16.6) and SS (54.2 ± 14.2) (p < 0.001). More than 85% of the patients with SS and SLE were female, whereas the patients with RA had a slightly lower proportion (74%) in females. The proportion of patients with abnormal immunomarker results varied among the SADs (Table 1).

Table 1

Demographic characteristics and proportions of abnormal immunomarker test results in patients with SS, RA, and SLE (immunomarkers with normal test results for all the SADs were excluded)
Variable	Sjögren’s syndrome (n = 3,541)	Rheumatoid arthritis (n = 5,696)	Systemic lupus erythematosus (n = 2,686)	p-value
Age, mean (sd)	54·2 (14·2)	51·0 (16·6)	33·9 (17·5)	< 0·001
Sex, n (%)
Female	3,070 (86·70)	4,250 (74·61)	2,307 (85·89)	< 0·001
Male	471 (13·30)	1,446 (25·39)	379 (14·11)	< 0·001
Autoantibody (= positive), n (%)
ACAG	12 (0·34)	1 (0·02)	59 (2·20)	< 0·001
ACAM	3 (0·08)	1 (0·02)	3 (0·11)	0·188
ANA	2,049 (57·87)	1,159 (20·35)	2,335 (86·93)	< 0·001
AQP4-autoAb	2 (0·06)	0 (0)	0 (0)	0·094
Anti-BMZA	0 (0)	0 (0)	3 (0·11)	0·006
Anti-CENP	6 (0·17)	0 (0)	0 (0)	< 0·001
Anti-ICSA	0 (0)	1 (0·02)	2 (0·07)	0·164
Anti-Jo	1 (0·03)	0 (0)	0 (0)	0·306
Anti-RNP-Sm	18 (0·51)	8 (0·14)	92 (3·43)	< 0·001
Anti-RibP	0 (0)	0 (0)	2 (0·07)	0·032
Anti-SSA	496 (14·01)	37 (0·65)	155 (5·77)	< 0·001
Anti-SSB	290 (8·19)	18 (0·32)	79 (2·94)	< 0·001
Anti-Scl	3 (0·08)	1 (0·02)	5 (0·19)	0·031
Anti-THYG	28 (0·79)	10 (0·18)	10 (0·37)	< 0·001
Anti-TPO	62 (1·75)	30 (0·53)	34 (1·27)	< 0·001
Anti-TSHR	1 (0·03)	0 (0)	1 (0·04)	0·386
Anti-dsDNA	45 (1·27)	27 (0·47)	1,221 (45·46)	< 0·001
B2GP1G	0 (0)	0 (0)	1 (0·04)	0·179
CRYOFIBRI	83 (2·34)	55 (0·97)	68 (2·53)	< 0·001
CRYOID	215 (6·07)	141 (2·48)	208 (7·74)	< 0·001
DC IgG	1 (0·03)	0 (0)	19 (0·71)	< 0·001
FLC Kappa	5 (0·14)	1 (0·02)	5 (0·19)	0·031
FLC Lambda	4 (0·11)	0 (0)	5 (0·19)	0·009
GAD-Ab	0 (0)	1 (0·02)	0 (0)	0·579
RF	2,258 (63·77)	5,695 (99·98)	817 (30·42)	< 0·001
Complement (= abnormal), n (%)
C3	239 (6·75)	125 (2·19)	1,513 (56·33)	< 0·001
C4	72 (2·03)	124 (2·18)	1,006 (37·45)	< 0·001
Abbreviations (alphabetically ordered):
ACAG = anti-cardiolipin antibody IgG, ACAM = anti-cardiolipin antibody IgM,
ANA = anti-nuclear antibody, AQP4-autoAb = anti-aquaporin 4 antibodies,
anti-BMZA = basement membrane zone antibody, anti-CENP = anti-centromere antibody,
anti-ICSA = intracellular substance antibody, anti-Jo = anti-Jo 1 antibody,
anti-RNP-Sm = anti-ribonucleoprotein and anti-Smith antibody,
anti-RibP = anti-ribosomal P antibody, anti-SSA = anti-SSA/Ro antibody,
anti-SSB = anti-SSB/La antibody, anti-Scl = anti-Scl antibody,
anti-THYG = anti-thyroglobulin antibody, anti-TPO = anti-thyroid peroxidase antibody,
anti-TSHR = anti-TSH receptor antibody, anti-dsDNA = anti-double stranded DNA antibody,
B2GP1G = beta-2 glycoprotein 1 antibody IgG, CRYOFIBRI = cryofibrinogen identification,
CRYOID = cryoblobulin identification, DC IgG = direct coombs IgG,
FLC Kappa = free light chain kappa, FLC Lambda = free light chain lambda,
GAD-Ab = glutamic acid decarboxylase 65 antibody, RF = rheumatoid factor,
C3 = complement component C3, C4 = complement component C4

3.2. Clustering results and cluster characteristics

Figure 2A presents the results of the cluster analysis. The best model, with the highest average silhouette width, retained two PCs, resulting in six clusters. The first PC separated Clusters 1, 3, 4, and 6, whereas the second PC further opposed Clusters 2 and 5.

The proportion of patients with the three SADs was imbalanced among the clusters. The patients with RA, SLE, and SS mainly dominated Clusters 1, 3 and 6, and 4 and 5, respectively. The patients with SLE and SS were evenly distributed in Cluster 2 (Fig. 2B). Figure 2C depicts the proportion of patients with the SADs in the six clusters. RA accounted for a disproportionately high percentage (> 90%) in Cluster 1, whereas SLE and SS had a more dispersed distribution in the six clusters.

We identified a distinctive pattern by determining the proportion of abnormal immunomarker results between the clusters (Figs. 3A and 3B), and details are described in Supplementary Document 5.

3.3. High heterogeneity in clinical manifestations of SADs between the clusters

We determined the rate of clinical manifestations for the SADs grouped into six clusters. SS showed the most heterogeneous clinical manifestations between clusters (Fig. 4A top). Patients with SS in Cluster 3 were the most distinctive because of the high occurrence of immunity disorders (63%) and other and unspecified benign neoplasm (58%), compared with the other clusters (all p values < 0.05). More than half of the patients with SS in Clusters 1 (53%), 2 (54%), and 3 (53%) experienced inflammation or infection of the eye; however, Cluster 2 had other symptoms frequently observed in SS higher than the other clusters, such as spondylosis (47%; p < 0.05) (e.g., backache, lumbago, cervical and lumbosacral spondylosis without myelopathy) and upper respiratory infections (42%; p < 0.04) compared with Cluster 3 as well as higher upper respiratory disorders (48%) with Cluster 3 (p = 0.01) and Cluster 4 (p = 0.02). The patients with SS in Cluster 4 appeared to have less cutaneous symptoms (e.g., inflammatory condition of skin as well as skin and subcutaneous tissue infections) (p < 0.05) than the other clusters but had a higher risk of manifesting lower respiratory diseases as compared with Cluster 3 (p = 0.03) (Fig. 4A top).

The patients with SLE in different clusters also demonstrated highly diverse clinical manifestations, specifically in blood, renal, and cutaneous-related symptoms (Fig. 4 bottom). Both Clusters 3 and 6 exhibited significantly higher proportions of patients with the two manifestations, where more than half had deficiency and other anemia (50% and 65%, p < 0.001) and nephritis (52% and 51%, p < 0.001). Clusters 1, 2, and 3 significantly displayed higher proportions of patients with discoid lupus (55%, 67%, and 60% respectively); however, Cluster 2 had higher proportions of patients with inflammatory conditions of skin (61%) and skin and subcutaneous tissue infections (32%) than Cluster 1 (46%; p = 0.03 and 23%; p < 0.05). In contrast, Clusters 4 and 5 manifested less diversity than the other clusters.

Clinical manifestations in RA were relatively homogeneous (Fig. 4A middle). Clusters 1, 2, and 3 had a higher risk of being diagnosed with osteoarthritis than the other three clusters (p < 0.05). While Cluster 3 had a higher proportion of patients with upper respiratory infections (53%) than those in Cluster 5 (11%, p = 0.01). Furthermore, Cluster 4 had higher proportions of patients with both upper respiratory disease (42%; p < 0.05) and urinary tract infections (42%; p = 0.02) when compared with Cluster 5.

We implemented the same clinical implication analysis but without stratifying by SADs. The heterogeneities in clinical manifestations were still observable (Supplement Fig. 1).

We successfully deciphered the heterogeneity in common SADs and obtained six clinically meaningful clusters by applying the simultaneous dimension reduction and cluster analysis approach to immunomarker data. These clusters revealed considerable differences in the pattern of immunomarkers and the occurrence of clinical manifestations.

4.1. Heterogeneity in major SADs

The cluster analysis results revealed heterogeneity in the major SADs (Figs. 2B and 2C). Moreover, the level of heterogeneity considerably differed among the SADs. High heterogeneity was noted for SS. Patients with SS were majorly clustered into Clusters 1 and 2. Although fewer patients were grouped into Clusters 4 and 5, patients with SS dominated the two clusters. Similar heterogeneity was also identified for SLE. Patients with SLE were predominantly distributed in Clusters 2 and 3, but the proportions of the patients were both higher in Clusters 3 and 6. By contrast, RA revealed relatively low heterogeneity.

The pattern of immunomarkers differed among the six clusters. Patients with SLE and SS in Cluster 2 presented more closely than typical cases in Clusters 3 and 6 (typical SLE) and Clusters 4 and 5 (typical SS), respectively. Similarly, the patients with SS in Cluster 1 were similar to those with RA. The results implied that in specific clusters (e.g., Clusters 1 and 2), the SADs would manifest atypically with immune features that can be found in other SADs: patients with SS in Cluster 1 would manifest more RA-like features, whereas SS in Cluster 2 would manifest more SLE-like features. Thus, compared with the current diagnostic standards of SADs, the clustering-based categorization of the SADs would be a more representative and data-driven definition.

4.2. Association between immunomarkers and clinical manifestations

An SAD classification system considering heterogeneity, such as in the pattern of immunomarker presentation or the occurrence of manifestations, enables us to delineate various underlying pathological mechanisms for identical diseases classified by the current system. Our results revealed links between the particular patterns of immunomarker presence (Fig. 3) and the occurrence of clinical manifestations (Fig. 4).

4.2.1. Association between immunomarkers and clinical manifestations for SS

A meta-analysis revealed significant associations between primary SS and various malignancies; risks were especially unneglectable in non-Hodgkin lymphoma, leukemia, myeloma, and autoimmune disease malignancy [20]. Moreover, a cohort study revealed significant predictors of lymphoproliferative diseases, such as purpura/skin vasculitis, low C3 levels, and low C4 levels [21], which were also observed in our study (Cluster 3, Fig. 3A).

Xerophthalmia is a common ocular manifestation of SS. Clusters 4 and 6 revealed higher frequencies of dry eye (61% and 67%, respectively) and differences in anti-SSA and anti-SSB results compared to other clusters (Fig. 3A). A study reported that the presence of anti-SSA/Ro concurrently with anti-SSB/La would deteriorate the ocular manifestation of SS [22]. We observed a similar trend where xerophthalmia appeared to be more prevalent in Clusters 4 and 6 (having higher proportions of anti-SSA and anti-SSB positivity) than in Cluster 5 (only anti-SSA positivity).

We identified that Clusters 2, 4, and 5 had more patients with back pain problems (47%, 43%, and 46%, respectively). Although a few studies have reported that patients with SS were affected by spondyloarthritis, the connections between them and the pathology remained inconclusive [23, 24]. By contrast, Cluster 4 had more patients with lower respiratory tract disease. A previous study revealed that the lower respiratory tract can be affected by SS, including the lung parenchyma, mainly with interstitial lung disease, airways, vasculature, and pleura [25].

4.2.2. Association between immunomarkers and clinical manifestations for SLE

An association between immunomarkers and clinical manifestations was noted in Clusters 3 and 6. Renal compromise, anemia, and inflammatory condition of the skin, which manifests by patients with relapsing or active SLE, are of great concern to prognosis. The patients in Cluster 3 appeared to have a propensity for severe SLE comorbidities. Elevated ds-DNA and decreased C3 and C4 levels were correlated with severe disease activity [26, 27] and renal flares [28]. In addition, progressively simultaneous reductions in C3 and C4 levels and increasing ds-DNA levels may be helpful predictors for future relapses in serologically active clinically quiescent SLE [29].

A cross-sectional study reported that the expression of anti-dsDNA and anti-SSA was higher in patients with SLE without discoid lupus erythematosus (DLE−/SLE+) than in those with DLE (DLE+/SLE+) [30]. We observed a similar trend in our study: Cluster 6, which had lower cutaneous manifestations, had a higher proportion of patients with positive results for anti-SSA than Clusters 1, 2, and 5, which had higher cutaneous manifestations. Moreover, the contribution of anti-RNP-Sm to DLE remains unclear, showing contradictory results in recent studies [30–32].

DLE indicates that patients may undergo a benign lupus course and have a lower risk of renal involvement [30]. However, this trend was only observed in Clusters 1, 2, and 5; we observed a negative correlation – by performing a multiple logistic regression with the covariates of sex, anti-SSA, and anti-RNP-Sm – between discoid lupus and renal compromise for all the patients with SLE in the three clusters (OR_adj for discoid lupus = 0·47; 95% CI = 0·34–0·65).

4.2.3. Association between immunomarkers and clinical manifestations for RA

Although no difference in common comorbidities was noted between clusters for RA, several extraarticular manifestations displayed a higher occurrence in some clusters. The most common manifestations were upper respiratory infections and lower respiratory diseases in Cluster 3, and upper respiratory disease and urinary tract infections in Cluster 4. Lung involvement significantly contributes to mortality and morbidity in the extraarticular manifestations of RA. The involvements often correlate with treatment agents, longstanding diseases, infections, and health behaviours [25].

4.3. Limitations

The ACR/EULAR 2019 criteria for SLE require ANA positivity. Some SLE diagnoses in our study may not meet these updated criteria. Besides, ANA was chosen as the initial biomarker and patients without ANA test were excluded. Nevertheless, it's important to note that ANA is a nearly essential test for individuals with suspected autoimmune diseases, despite its suboptimal diagnostic accuracy in certain autoimmune conditions like RA. Consequently, while the entry criteria might raise concerns, they are unlikely to introduce bias into the study.

We assumed that the immunomarker results for patients not specifically ordered by a physician were within the reference range, except for RF in RA and ANA in SLE and SS cases. Physicians use expertise to determine test necessity for diagnosis and tailor treatments in clinical settings. This assumption, while introducing bias, seemed plausible because SAD symptoms strongly correlate with autoantibody presence, helping us address missing data mechanisms. Additionally, the plausibility and efficacy of imputing missing values with normal values have been reported [13]. Notably, ANA seronegativity in SLE cases is rare, and post-diagnostic ANA seropositivity is common in primary SS [14][15]. Therefore, we imputed RF and ANA antibodies in a disease-specific manner. Moreover, some diagnostically valuable tests not covered by NHI (e.g., anti-citrullinated peptide antibody) were excluded due to high missing rates.

Other limitations include coding bias when using the diagnosis codes to identify clinical manifestations. We used an inclusion algorithm (codes appearing at least twice within a specified time) to reduce misclassification. Major SADs are not limited to RA, SLE, and SS, and patient with multiple SADs is not uncommon – constituting nearly a quarter (24.9%) of SAD patients in our database. While the prevalence of certain other SADs, like systemic sclerosis (SSc), is relatively lower in our database, we intend to continue collecting data on patients with various SADs and those with multiple SADs for future in-depth analyses.

Finally, our study primarily focused on Han Taiwanese patients, which limits the generalizability to other ethnicities.

We developed an immunomarker–driven clustering method for major SADs based on large-scale data. High heterogeneity was identified for the conventional SADs among these immunomarker-based clusters. Moreover, clinical manifestations correlated well with the immunomarker-driven clustering, implying that the clustering method could recognise more clinically relevant subgroups. The data-driven classification for the SADs would provide a more objective and precise diagnosis and lead to more adequate treatment.

Heterogeneity in systemic autoimmune diseases is identified by clustering method.
These clusters revealed considerable differences in the pattern of immunomarkers.
The clinical manifestations correlated well with the immunomarker-driven clusters.
Identifying heterogeneity helps precise diagnosis, leading to precise disease management.

Ethics approval and consent to participate

This study was approved by the Institutional Review Board of Chang Gung Medical Foundation (202001298B0), and the requirement for patient consent was waived.

Conflict of Interest

The authors declare no competing interests.

Funding

This research was supported by the National Science and Technology Council, Taiwan under Grants NSTC 111-2320-B-182A-002-MY2 and 111-2628-E-A49-026-MY3; Chang Gung Memorial Hospital under Grants CORPG2P0071 and CMRPD3K0011; the Higher Education Sprout Project of the National Yang Ming Chiao Tung University and Ministry of Education (MOE), Taiwan under Grants CGMH-NYCU-113-CORPG2P0071.

Author Contribution

C-W. C and Y-J. T had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. C-W. C analysed/interpreted the data and performed experiments. C-W. C, H-Y. W, and Y-J. T designed the study and wrote the paper. W-L. L, W-T. L, and J-R. Y reviewed/edited the manuscript for important intellectual content and provided administrative, technical, or material support. Y-J. T obtained funding and supervised the study.

Availability of data and materials

All data used in the analyses is restricted due to the regulation of the law on the protection of patients’ data in accordance with local and institutional legal requirements; de-identified participant data will be made available upon reasonable request to the corresponding author.

Cojocaru M, Cojocaru IM, Silosi I, Vrabie CD. Manifestations of systemic lupus erythematosus. Maedica. 2011;6:330–6. /pmc/articles/PMC3391953/.
Arnaud L, Tektonidou MG. Long-term outcomes in systemic lupus erythematosus: trends over time and major contributors. Rheumatology. 2020;59:v29–38. https://doi.org/10.1093/rheumatology/keaa382.
Mosca M, Tani C, Talarico R, Bombardieri S. Undifferentiated connective tissue diseases (UCTD): Simplified systemic autoimmune diseases. Autoimmun Rev. 2011;10:256–8. https://doi.org/10.1016/j.autrev.2010.09.013.
A.D.A.P P, Committee. 2. Classification and Diagnosis of Diabetes: Standards of Medical Care in Diabetes—2022, Diabetes Care. 45 (2022) S17–S38. https://doi.org/10.2337/dc22-s002.
Thygesen K, Alpert JS, Jaffe AS, Simoons ML, Chaitman BR, White HD, Chairpersons AFM, Subcommittee B, Katus HA, Apple FS, Lindahl B, Morrow DA, Subcommittee E, Clemmensen PM, Johanson P, Hod H, Subcommittee I, Underwood R, Bax JJ, Bonow RO, Pinto F, Gibbons RJ, Subcommittee C, Fox KA, Atar D, Newby LK, Galvani M, Hamm CW, Subcommittee I, Uretsky BF, Steg PG, Wijns W, Bassand J-P, Menasché P, Ravkilde J, T.&, Subcommittee R, Ohman EM, Antman EM, Wallentin LC, Armstrong PW, Subcommittee HF, Januzzi JL, Nieminen MS, Gheorghiade M, Filippatos G, Subcommittee E, Luepker RV, Fortmann SP, Rosamond WD, Levy D, Wood D, Subcommittee GP, Smith SC, Hu D, Lopez-Sendon J-L, Robertson RM, Weaver D, Tendera M, Bove AA, Parkhomenko AN, Vasilieva EJ, Mendis S, CPG) EC, Baumgartner H, Ceconi C, Dean V, Deaton C, Fagard R, Funck-Brentano C, Hasdai D, Hoes A, Kirchhof P, Knuuti J, Kolh P, McDonagh T, Moulin C, Popescu BA, Reiner Ž, Sechtem U, Sirnes PA, Torbicki A, Vahanian A, Windecker S, Reviewers D, Morais J, Aguiar C, Almahmeed W, Arnar DO, Barili F, Bloch KD, Bolger AF, Bøtker HE, Bozkurt B, Bugiardini R, Cannon C, de Lemos J, Eberli FR, Escobar E, Hlatky M, James S, Kern KB, Moliterno DJ, Mueller C, Neskovic AN, Pieske BM, Schulman SP, Storey RF. K.A. Taubert, P. Vranckx, D.R. Wagner, Third universal definition of myocardial infarction, in: Eur Heart J, 2012: pp. 2551–2567. https://doi.org/10.1093/eurheartj/ehs184.
Thygesen K, Alpert JS, Jaffe AS, Chaitman BR, Bax JJ, Morrow DA, White HD, Group ESD, Mickley H, Crea F, de Werf FV, Bucciarelli-Ducci C, Katus HA, Pinto FJ, Antman EM, Hamm CW, Caterina RD, Januzzi JL, Apple FS, Garcia MAA, Underwood SR, Canty JM, Lyon AR, Devereaux PJ, Zamorano JL, Lindahl B, Weintraub WS, Newby LK, Virmani R, Vranckx P, Cutlip D, Gibbons RJ, Smith SC, Atar D, Luepker RV, Robertson RM, Bonow RO, Steg PG, O’Gara PT, Fox KAA, Hasdai D, Aboyans V, Achenbach S, Agewall S, Alexander T, Avezum A, Barbato E, Bassand J-P, Bates E, Bittl JA, Breithardt G, Bueno H, Bugiardini R, Cohen MG, Dangas G, de Lemos JA, Delgado V, Filippatos G, Fry E, Granger CB, Halvorsen S, Hlatky MA, Ibanez B, James S, Kastrati A, Leclercq C, Mahaffey KW, Mehta L, Müller C, Patrono C, Piepoli MF, Piñeiro D, Roffi M, Rubboli A, Sharma S, Simpson IA, Tendera M, Valgimigli M, van der Wal AC, Windecker S, Chettibi M, Hayrapetyan H, Roithinger FX, Aliyev F, Sujayeva V, Claeys MJ, Smajić E, Kala P, Iversen KK, Hefny EE, Marandi T, Porela P, Antov S, Gilard M, Blankenberg S, Davlouros P, Gudnason T, Alcalai R, Colivicchi F, Elezi S, Baitova G, Zakke I, Gustiene O, Beissel J, Dingli P, Grosu A, Damman P, Juliebø V, Legutko J, Morais J, Tatu-Chitoiu G, Yakovlev A, Zavatta M, Nedeljkovic M, Radsel P, Sionis A. Eur Heart J. 2019;40:237–69. https://doi.org/10.1093/eurheartj/ehy462. T. Jemberg, L. Abid, A. Abaci, A. Parkhomenko, S. Corbett, Fourth universal definition of myocardial infarction (2018).
Aringer M, Costenbader K, Daikh D, Brinks R, Mosca M, Ramsey-Goldman R, Smolen JS, Wofsy D, Boumpas DT, Kamen DL, Jayne D, Cervera R, Costedoat-Chalumeau N, Diamond B, Gladman DD, Hahn B, Hiepe F, Jacobsen S, Khanna D, Lerstrøm K, Massarotti E, McCune J, Ruiz-Irastorza G, Sanchez-Guerrero J, Schneider M, Urowitz M, Bertsias G, Hoyer BF, Leuchten N, Tani C, Tedeschi SK, Touma Z, Schmajuk G, Anic B, Assan F, Chan TM, Clarke AE, Crow MK, Czirják L, Doria A, Graninger W, Halda-Kiss B, Hasni S, Izmirly PM, Jung M, Kumánovics G, Mariette X, Padjen I, Pego-Reigosa JM, Romero-Diaz J, Fernández ÍR-F, Seror R, Stummvoll GH, Tanaka Y, Tektonidou MG, Vasconcelos C, Vital EM, Wallace DJ, Yavuz S, Meroni PL, Fritzler MJ, Naden R, Dörner T. Johnson, 2019 European League Against Rheumatism/American College of Rheumatology classification criteria for systemic lupus erythematosus. Ann Rheum Dis. 2019;78:1151. https://doi.org/10.1136/annrheumdis-2018-214819.
Wen Y-H, Chang P-Y, Hsu C-M, Wang H-Y, Chiu C-T, Lu J-J. Cancer screening through a multi-analyte serum biomarker panel during health check-up examinations: Results from a 12-year experience. Clin Chim Acta. 2015;450:273–6. https://doi.org/10.1016/j.cca.2015.09.004.
Wang H-Y, Chen C-H, Shi S, Chung C-R, Wen Y-H, Wu M-H, Lebowitz MS, Zhou J, Lu J-J. Learn Algorithms Cancers. 2020;12:1442. https://doi.org/10.3390/cancers12061442. Improving Multi-Tumor Biomarker Health Check-Up Tests with Machine.
Wu X, Wang H-Y, Shi P, Sun R, Wang X, Luo Z, Zeng F, Lebowitz MS, Lin W-Y, Lu J-J, Scherer R, Price O, Wang Z, Zhou J, Wang Y. Long short-term memory model – A deep learning approach for medical data with irregularity in cancer predication with tumor markers. Comput Biol Med. 2022;144:105362. https://doi.org/10.1016/j.compbiomed.2022.105362.
Mossotto E, Ashton JJ, Coelho T, Beattie RM, MacArthur BD, Ennis S. Classification of Paediatric Inflammatory Bowel Disease using Machine Learning. Sci Rep-Uk. 2017;7:2427. https://doi.org/10.1038/s41598-017-02606-2.
Orange DE, Agius P, DiCarlo EF, Robine N, Geiger H, Szymonifka J, McNamara M, Cummings R, Andersen KM, Mirza S, Figgie M, Ivashkiv LB, Pernis AB, Jiang CS, Frank MO, Darnell RB, Lingampali N, Robinson WH, Gravallese E. the A.M.P. in R.A. and L. Network, V.P. Bykerk, S.M. Goodman, L.T. Donlin, Identification of Three Rheumatoid Arthritis Disease Subtypes by Machine Learning Integration of Synovial Histologic Features and RNA Sequencing Data, Arthritis Rheumatol. 70 (2018) 690–701. https://doi.org/10.1002/art.40428.
Barturen G, Babaei S, Català-Moll F, Martínez‐Bueno M, Makowska Z, Martorell‐Marugán J, Carmona‐Sáez P, Toro‐Domínguez D, Carnero‐Montoro E, Teruel M, Kerick M, Acosta‐Herrera M, Lann LL, Jamin C, Rodríguez‐Ubreva J, García‐Gómez A, Kageyama J, Buttgereit A, Hayat S, Mueller J, Lesche R, Hernandez‐Fuentes M, Juarez M, Rowley T, White I, Marañón C, Anjos TG, Varela N, Aguilar‐Quesada R, Garrancho FJ, López‐Berrio A, Maresca MR, Navarro‐Linares H, Almeida I, Azevedo N, Brandão M, Campar A, Faria R, Farinha F, Marinho A, Neves E, Tavares A, Vasconcelos C, Trombetta E, Montanelli G, Vigone B, Alvarez‐Errico D, Li T, Thiagaran D, Alonso RB, Martínez AC, Genre F, Mejías RL, Gonzalez‐Gay MA, Remuzgo S, Garcia BU, Cervera R, Espinosa G, Rodríguez‐Pintó I, Langhe ED, Cremer J, Lories R, Belz D, Hunzelmann N, Baerlecken N, Kniesch K, Witte T, Lehner M, Stummvoll G, Zauner M, Aguirre‐Zamorano MA, Barbarroja N, Castro‐Villegas MC, Collantes‐Estevez E, Ramon E, Quintero ID, Escudero‐Contreras A, Roldán MCF, Gómez YJ, Moleón IJ, Lopez‐Pedrera R, Ortega‐Castro R, Ortego N, Raya E, Artusi C, Gerosa M, Meroni PL, Schioppo T, Groof AD, Ducreux J, Lauwerys B, Maudoux A, Cornec D, Devauchelle‐Pensec V, Jousse‐Joulin S, Jouve P, Rouvière B, Saraux A, Simon Q, Alvarez M, Chizzolini C, Dufour A, Wynar D, Balog A, Bocskai M, Deák M, Dulic S, Kádár G, Kovács L, Cheng Q, Gerl V, Hiepe F, Khodadadi L, Thiel S, Rinaldis E, Rao S, Benschop RJ, Chamberlain C, Dow ER, Ioannou Y, Laigle L, Marovac J, Wojcik J, Renaudineau Y, Borghi MO, Frostegård J. J. Martín, L. Beretta, E. Ballestar, F. McDonald, J. Pers, M.E. Alarcón‐Riquelme, Integrative Analysis Reveals a Molecular Stratification of Systemic Autoimmune Diseases, Arthritis Rheumatol. 73 (2021) 1073–85. https://doi.org/10.1002/art.41610.
Moritz CP, Paul S, Stoevesandt O, Tholance Y, Camdessanché J-P, Antoine J-C. Autoantigenomics: Holistic characterization of autoantigen repertoires for a better understanding of autoimmune diseases, Autoimmun Rev. 19 (2020) 102450. https://doi.org/10.1016/j.autrev.2019.102450.
Stafford IS, Kellermann M, Mossotto E, Beattie RM, MacArthur BD, Ennis S. A systematic review of the applications of artificial intelligence and machine learning in autoimmune diseases. Npj Digit Med. 2020;3:30. https://doi.org/10.1038/s41746-020-0229-3.
Spielmann L, Nespola B, Séverac F, Andres E, Kessler R, Guffroy A, Poindron V, Martin T, Geny B, Sibilia J, Meyer A. Anti-Ku syndrome with elevated CK and anti-Ku syndrome with anti-dsDNA are two distinct entities with different outcomes. Ann Rheum Dis. 2019;78:1101. https://doi.org/10.1136/annrheumdis-2018-214439.
Molano-González N, Rojas M, Monsalve DM, Pacheco Y, Acosta-Ampudia Y, Rodríguez Y, Rodríguez-Jimenez M, Ramírez-Santana C, Anaya J-M. Cluster analysis of autoimmune rheumatic diseases based on autoantibodies. New insights for polyautoimmunity. J Autoimmun. 2019;98:24–32. https://doi.org/10.1016/j.jaut.2018.11.002.
Shao S, Chan Y, Yang YK, Lin S, Hung M, Chien R, Lai C, Lai EC. The Chang Gung Research Database—A multi-institutional electronic medical records database for real‐world epidemiological studies in Taiwan. Pharmacoepidem Drug Safe. 2019;28:593–600. https://doi.org/10.1002/pds.4713.
Tseng Y-J, Chiu H-J, Chen CJ. dxpr: an R package for generating analysis-ready data from electronic health records—diagnoses and procedures. Peerj Comput Sci. 2021;7:e520. https://doi.org/10.7717/peerj-cs.520.
Liang Y, Yang Z, Qin B, Zhong R. Primary Sjögren’s syndrome and malignancy risk: a systematic review and meta-analysis. Ann Rheum Dis. 2014;73:1151. https://doi.org/10.1136/annrheumdis-2013-203305.
Theander E, Henriksson G, Ljungberg O, Mandl T, Manthorpe R, Jacobsson LTH. Lymphoma and other malignancies in primary Sjögren’s syndrome: a cohort study on cancer incidence and lymphoma predictors. Ann Rheum Dis. 2006;65:796. https://doi.org/10.1136/ard.2005.041186.
Baer AN, DeMarco MM, Shiboski SC, Lam MY, Challacombe S, Daniels TE, Dong Y, Greenspan JS, Kirkham BW, Lanfranchi HE, Schiødt M, Srinivasan M, Umehara H, Vivino FB, Vollenweider CF, Zhao Y, Criswell LA, Shiboski CH. S.I.C.C.A. (SICCA) R. Groups, The SSB-positive/SSA-negative antibody profile is not associated with key phenotypic features of Sjögren’s syndrome. Ann Rheum Dis. 2015;74:1557. https://doi.org/10.1136/annrheumdis-2014-206683.
Jarrot P-A, Arcani R, Darmon O, Roudier J, Cauchois R, Mazodier K, Jean R, Balandraud N, Kaplanski G. Axial Articular Manifestations in Primary Sjögren Syndrome: Association With Spondyloarthritis. J Rheumatol. 2020;48:1037–46. https://doi.org/10.3899/jrheum.200189.
Eren R, Can M, Alibaz-Öner F, Yilmaz-Oner S, Yilmazer B, Cefle A, Direskeneli H, Atagündüz P. Prevalence of inflammatory back pain and radiologic sacroiliitis is increased in patients with primary Sjögren’s syndrome. Pan Afr Med J. 2018;30:98. https://doi.org/10.11604/pamj.2018.30.98.15588.
Luppi F, Sebastiani M, Sverzellati N, Cavazza A, Salvarani C, Manfredi A. Lung complications of Sjogren syndrome. Eur Respir Rev. 2020;29:200021. https://doi.org/10.1183/16000617.0021-2020.
Bombardier C, Gladman DD, Urowitz MB, Caron D, Chang CH, Austin A, Bell A, Bloch DA, Corey PN, Decker JL, Esdaile J, Fries JF, Ginzler EM, Goldsmith CH, Hochberg MC, Jones JV, Riche NGHL, Liang MH, Lockshin MD, Muenz LR, Sackett DL, Schur PH. Derivation of the sledai. A disease activity index for lupus patients. Arthritis Rheum. 1992;35:630–40. https://doi.org/10.1002/art.1780350606.
Gladman DD, Ibañez D, Urowitz MB. Systemic lupus erythematosus disease activity index 2000. J Rheumatol. 2002;29:288–91. jrheum.org/content/29/2/288.long.
Narayanan K, Marwaha V, Shanmuganandan K, Shankar S. Correlation between systemic lupus erythematosus disease activity index, C3, C4 and anti-dsDNA antibodies. Med J Armed Forces India. 2010;66:102–7. https://doi.org/10.1016/s0377-1237(10)80118-2.
Miyawaki Y, Sada K, Asano Y, Hayashi K, Yamamura Y, Hiramatsu S, Ohashi K, Morishita M, Watanabe H, Matsumoto Y, Kawabata T, Wada J. Progressive reduction of serum complement levels: a risk factor for relapse in patients with hypocomplementemia in systemic lupus erythematosus. Lupus. 2018;27:2093–100. https://doi.org/10.1177/0961203318804892.
Chong BF, Tseng L, Lee T, Vasquez R, Li QZ, Zhang S, Karp DR, Olsen NJ, Mohan C. IgG and IgM Autoantibody Differences in Discoid and Systemic Lupus Patients. J Invest Dermatol. 2012;132:2770–9. https://doi.org/10.1038/jid.2012.207.
Vasquez R, Tseng L, Victor S, Zhang S, Chong BF. Autoantibody and Clinical Profiles in Patients With Discoid Lupus and Borderline Systemic Lupus. Arch Dermatol. 2012;148:651–5. https://doi.org/10.1001/archdermatol.2011.3249.
Kim A, O’Brien J, Tseng L-C, Zhang S, Chong BF. Autoantibodies and Disease Activity in Patients With Discoid Lupus Erythematosus. Jama Dermatol. 2014;150:651–4. https://doi.org/10.1001/jamadermatol.2013.8354.

No competing interests reported.

SumitSupplementarymaterials.docx

Download PDF

Journal Publication

published 16 Sep, 2024

Read the published version in BioData Mining →

Editorial decision: Revision requested
15 Jul, 2024
Reviews received at journal
10 Jul, 2024
Reviews received at journal
28 Jun, 2024
Reviewers agreed at journal
24 Jun, 2024
Reviewers agreed at journal
03 Jun, 2024
Reviewers invited by journal
29 Apr, 2024
Submission checks completed at journal
24 Apr, 2024
Editor assigned by journal
24 Apr, 2024
First submitted to journal
23 Apr, 2024

You are reading this latest preprint version

Identifying heterogeneous subgroups of systemic autoimmune diseases by applying a joint dimension reduction and clustering approach to immunomarkers

Status:

Journal Publication

Version 1

Abstract

Figures

1. Introduction

2. Method

2.1. Study population and setting

2.2. Patient inclusion algorithm

2.3. Cluster analysis

2.4. Assessment of clinical implications

3. Results

3.1. Demographic characteristics and immunomarker tests

3.2. Clustering results and cluster characteristics

3.3. High heterogeneity in clinical manifestations of SADs between the clusters

4. Discussion

4.1. Heterogeneity in major SADs

4.2. Association between immunomarkers and clinical manifestations

4.2.1. Association between immunomarkers and clinical manifestations for SS

4.2.2. Association between immunomarkers and clinical manifestations for SLE

4.2.3. Association between immunomarkers and clinical manifestations for RA

4.3. Limitations

5. Conclusion

Summary Table

Declarations

Conflict of Interest

Funding

Author Contribution

Availability of data and materials

References

Additional Declarations

Supplementary Files

Status:

Journal Publication

Version 1