Aim: This study aimed to construct prognostic mathematical models utilizing multifactorial regression analysis to assess the risk of developing drug-induced neutralizing antibodies in patients with rheumatoid arthritis, psoriatic arthritis, and ankylosing spondylitis treated with tumor necrosis factor alpha blockers.

Materials and methods: Over a four-year period, we enrolled 213 patients in this study and divided them into three groups: the rheumatoid arthritis group (n=121), the ankylosing spondylitis group (n=50), and the psoriatic arthritis group (n=42). The study included also a group of healthy controls consisting of 31 healthy subjects who matched the patient groups in age, sex, body mass index, and conditions typical for rheumatology patients. Prognostic mathematical models based on statistically significant factors determined through univariate correlation and regression analyses incorporated patient medical history and serological and immunological data.

Results: The study encompassed all 213 patients and 31 healthy controls. Data analysis was conducted at 12 and 24 months after commencing treatment. During this follow-up, the patients exhibited the highest percentage of antidrug antibodies. At 6 months, 6.57% of patients had confirmed neutralizing antibodies, which increased to 12.69% at 12 months and 17.72% at 24 months. Multivariate logistic regression analysis revealed that factors such as age over 55 years, excess weight, smoking, and absence of methotrexate treatment at a dose less than 7.5 mg per week had the highest predictive value.

Conclusions: Investigating clinical and biological markers with predictive value for individual patients’ therapeutic responses is a complex task. This complexity arises from the interplay of at least three distinct parameters: the patient’s disease state, drug bioavailability, and pathophysiological changes within the patient’s body, all of which are influenced by various factors.

anti-drug antibodies, prognostic model

The increasing prevalence of rheumatoid arthritis (RA), psoriatic arthritis (PsA), and ankylosing spondylitis (AS) has led to a growing reliance on tumor necrosis factor alpha (TNF-α) blockers for the management of these conditions. While these drugs have demonstrated efficacy in ameliorating symptoms and improving quality of life for patients, the emergence of drug-induced neutralizing antibodies poses a significant concern. These antibodies can compromise the effectiveness of TNF-α blockers, leading to diminished therapeutic outcomes and potential complications in patients undergoing treatment.

In light of this, there is a pressing need to develop robust prognostic models that can assess the risk of patients developing drug-induced neutralizing antibodies. To address this concern, our study employed multifactorial regression analysis to construct mathematical models for evaluating the risk in patients with RA, PsA, and AS. This involves an in-depth examination of various demographic, clinical, and immunological factors, identified through univariate correlation and regression analyses, to ensure the models are based on statistically significant variables.^[1–3]

Existing literature underscores the importance of considering patient medical history, serological, and immunological data in constructing predictive models for drug-induced neutralizing antibodies.^[4] Despite the acknowledged significance of these factors, there remains a dearth of comprehensive studies that integrate them into a unified prognostic framework. Our research seeks to bridge this gap by providing a thorough exploration of the interplay between demographic, clinical, and immunological factors, emphasizing their collective impact on the development of neutralizing antibodies in patients treated with TNF-α blockers.

This study aims to contribute not only prognostic models but also a deeper understanding of the intricate relationships between patient characteristics and the risk of developing neutralizing antibodies. By addressing this gap in the literature, we strive to enhance the precision of risk assessment and, subsequently, the effectiveness of therapeutic interventions in patients undergoing TNF-α blocker treatment.

This prospective, longitudinal, open-label study spanned a four-year period, involving a comprehensive assessment of 213 patients diagnosed with rheumatoid arthritis, psoriatic arthritis, and ankylosing spondylitis. The study also included a control group consisting of 31 healthy individuals. The patient groups were further divided into two categories based on the type of biological agent used: adalimumab and etanercept.

A cross-sectional, multicenter observational approach was employed for data collection. After obtaining informed consent, patients were interviewed, examined, and their information was collected and entered into files for subsequent statistical analysis. Self-assessment questionnaires were voluntarily completed by patients, and blood samples were taken for the assessment of cytokines (TNF-α and IL-6) and the detection of drug-induced neutralizing antibodies. Some patients, particularly those treated with adalimumab, were additionally tested for drug bioavailability six months after treatment initiation.

The process of constructing prognostic models for the emergence of drug-induced neutralizing antibodies in patients with RA, PsA, and AS treated with TNF-α blockers within the analyzed patient group followed these steps:

• Identifying a sample of 213 patients to isolate independent, statistically significant prognostic factors for the emergence of drug-induced neutralizing antibodies in these patients.
• Analyzing all factors proven to be statistically significant in univariate analysis through multivariate analysis, employing a regression model with a stepwise selection procedure.
• Ensuring the proportionality of all covariates included in the regression model.
• Investigating the existence of individual data instances with notably deviating values for the model-influencing covariates.
• Repeating a multifactorial analysis involving all 213 patients.
• Evaluating the goodness-of-fit of the model, which assesses its adequacy in describing the resulting variable.

The significant variables identified through unifactorial regression analysis or correlation analysis were tested to establish a predictive mathematical model for the emergence of drug-induced neutralizing antibodies in patients with RA, PsA, and AS treated with TNF-α blockers. These variables included gender, age, education, excess weight, smoking, alcohol intake, concomitant pathology, dosage regimen of methotrexate below 10 mg per week, disease severity, ESR ≥28 mm, CRP ≥10 mg/L, VAS for pain as assessed by the patient, scales and activity indices such as HAQ-DI, DAS-28 for patients with RA, BASFI, ASDAS, BASDAI for patients with AS, and DAPSA for patients with PsA. Additionally, the levels of TNF-α and IL-6, as determined through ROC analysis, played a role in distinguishing patients from healthy controls, as well as the bioavailability of adalimumab exceeding the specified laboratory threshold value.

The equation used to calculate the risk likelihood estimate is as follows:

Z – linear combination

Z = Vo+B1o1+B2o2+B3o3+... In Hoch

E – Non-zero number – 2.71.

Vo – represents the constant unique to each mathematical model of probabilities

B1o1 – constant of the first indicator

B2o2 – constant of the second indicator, and so forth

High risk of drug-induced neutralizing antibodies in patients with RA, PsA, and AS treated with TNF-α blockers was associated with factors including age over 55 years, BMI over 25, smoking, methotrexate dosage below 10 mg per week, and disease duration exceeding 12 years, as shown in Table 3.

Our mathematical models are based on patient history information and the medical documentation provided by patients at their initial appointments with the physician. These models encompass a comprehensive range of sequential serological and immunological parameters.

A heightened risk of developing drug-induced neutralizing antibodies in patients with RA, PsA, and AS treated with TNF-α blockers is associated with specific criteria, including:

• Elevated ESR ≥28 mm
• Elevated CRP ≥10 mg/L
• Significantly different levels of TNF-α and IL-6 compared to the control group of individuals corresponding to the respective disease
• Evaluation of adalimumab drug bioavailability conducted prior to treatment initiation (Table 4).

These criteria aid in identifying patients with a higher risk of developing drug-induced neutralizing antibodies.

A simplified predictive mathematical model, information on disease limitation, morning stiffness, and disease activity indexes were obtained in Steps 3 and 4 by multivariate regression analysis (Table 5).

This model is preferable because it only collects information from the patient without the need for clinical laboratory tests.

The following mathematical model was obtained with the serological indicators included, drug bioavailability of adalimumab (model for RA patients only), and activity indices (Table 6).

The pursuit of identifying predictors of a favorable clinical response to TNF-α blockers in patients with inflammatory joint diseases, such as RA, AS, and PsA, has been acknowledged in numerous scientific studies.^[5–8] According to Hyrich et al.^[5], they found a correlation between higher HAQ scores and reduced therapy efficacy, while the concomitant use of NSAIDs and MTX intake was associated with improved clinical responses in patients with RA. Interestingly, this study also revealed that smokers exhibited a less favorable response to therapy, while factors like patient age, disease duration, and previous DMARD use did not significantly impact therapy effectiveness.^[5] However, it is worth noting that despite discussing risk factors, these authors did not consolidate them into predictive mathematical models.

In our research, we have constructed several mathematical models aimed at predicting the development of drug-induced neutralizing antibodies in patients with RA, PsA, and AS undergoing treatment with TNF-α blockers. These models incorporate a wide array of demographic, serological, immunological, and disease activity index data. It is important to recognize that in practical clinical settings, physicians may not always be able to assess cytokine levels, drug bioavailability, or antibodies to the medication used. Furthermore, we did not include sex as a predictive element in our models, as our studies indicated that sex does not hold predictive value. Interestingly, these findings differ from those reported by Arends et al.^[9], who suggested that younger age, male sex, higher ASDAS values, lower ESR and CRP values, and elevated PtGADA values hold prognostic value for the effectiveness of TNF-α blocker treatment.

In our evaluation of the prognostic value of serum TNF-α and IL-6 levels, we sought to create specific predictive models for each disease, incorporating only these two cytokines. While the medical literature recommends the study of the Cytokine Activity Index, which includes an array of cytokines, such as TNF-α, IL-1, IL-6, GM-CSF, IL-4, IL-5, IL-13, and others^{[10, 11]}, practical considerations led us to propose predictive mathematical models that focus on the two cytokines most relevant to our study.

Our analysis considered both groups of patients, those with and without serum-neutralizing antibodies treated with TNF-α blockers, as a unified cohort. This approach was adopted because, despite variances in the etiology of RA, AS, and PsA, the ultimate outcome of treatment, as indirectly measured by the presence of neutralizing antibodies, exhibited no substantial difference at 12 and 24 months from the initiation of therapy. These findings diverge from those of Maneiro et al.^[12], who suggested that factors like younger age, male sex, high baseline BASDAI, low baseline ESR, low baseline CRP, and HLA-B27 positivity are predictive of a better response to TNF-antagonist treatment in AS patients but not in those with PsA.

Each of the prognostic mathematical models that integrated pharmacovigilance data related to adalimumab’s bioavailability yielded highly significant results. This underscores the findings of Jani et al.^{[13, 14]}, who reported that adalimumab’s bioavailability is the most reliable predictor of changes in DAS-28 at the 12-month mark in RA patients monitored over a year. They found a strong regression coefficient of 0.060 (95% CI 0.015, 0.10, p=0.009). These authors, like us, concluded that pharmacological assessments of patients receiving TNF-α blockers offer clinical utility, even in the absence of drug-induced antibodies.

Despite the numerous predictive factors identified for response to biological therapies, consensus among researchers is only shared for a select few. Predicting outcomes in severe rheumatic diseases, indicated by a high HAQ score and prior treatment failure, remains a challenging endeavor, as a high baseline DAS-28 predicts a stronger DAS-28 in a follow-up, irrespective of the type of treatment received. Conversely, younger age, smoking cessation, alcohol abstinence, lower body weight, and a negative serological status prove to be reliable predictors of TNF-α blocker treatment efficacy.

The pursuit of clinical and biological markers with predictive value for therapeutic responses in individual patients remains a formidable task. The complexity arises from the fact that treatment responses depend on at least three distinct parameters: the disease’s state, drug bioavailability, and the pathophysiological changes occurring in the affected organism. Moreover, each of these parameters is influenced by a myriad of other factors.

This study was approved by the Ethics Committee of the Medical University of Plovdiv (approval No. DP- P-8278 /02/13/2018)

The authors have no support to report.

The authors have no funding to report.

The authors have declared that no competing interests exist.