Biological Mechanisms Linking COVID-19 Vaccination and Autoimmune Disorders: A Qualitative Analysis of Interpretive Consistency

1. Introduction

The global Covid-19 vaccination campaign was unprecedented in speed, scope, and scale. In the words of the Lancet Commission on lessons for the future from the Covid-19 pandemic, vaccines were “the single most important” technology for bringing the pandemic under control and enabling a “return to normalcy” [1]. At the same time, public and scientific claims about vaccine benefits have frequently relied on inferential strategies developed under conditions of incomplete clinical data, including the use of immunogenicity as a proxy for protection (See [2,3] and modeled counterfactual estimates of lives saved (See [4,5]), criticized for substituting mathematical modeling for direct empirical evidence [6].

Concerns about adverse effects following vaccination—particularly autoimmune reactions—have received comparatively less sustained attention. Despite longstanding recognition that vaccines can precipitate or unmask autoimmune processes, individuals with autoimmune disorders were largely excluded from initial Covid-19 vaccine trials. Nevertheless, professional societies encouraged vaccination in these populations early in the rollout, on the grounds of presumed vulnerability to severe Covid-19, even while acknowledging limited population-specific safety and efficacy data [7]. This configuration—restricted trial evidence, policy urgency, and the rapid extension of vaccination recommendations to heterogeneous clinical populations—underscores the importance of examining not only reports of autoimmune outcomes, but also interpretive practices concerning their causal implications.

In a completed scoping review of 109 peer-reviewed articles, we documented substantial and diverse reported associations between Covid-19 vaccination and autoimmune disorders, both among individuals with pre-existing autoimmune disease and among individuals without prior autoimmunity [8]. Importantly for the present analysis, a substantial subset of articles also proposed mechanistic explanations for how Covid-19 vaccines could trigger autoimmune responses. These included pathways such as molecular mimicry, bystander activation, epitope spreading, polyclonal activation, and immune stimulation linked to vaccine components—mechanisms that have been documented for decades in relation to post-vaccination autoimmunity in other contexts [9,10,11,12]. In addition, some authors have argued that such mechanisms may be relevant even for platforms not classically framed as “adjuvanted”, including mRNA vaccines, given their intrinsic immune-stimulatory properties [13,14].

This planned follow-up project was prompted by a recurring observation in the scoping review: a persistent disconnect between the evidence reported within articles and the interpretive conclusions those same articles advanced. Specifically, this disconnect appeared in two ways: first, between mechanistic plausibility and other evidence supporting a causal link between Covid-19 vaccination and autoimmune disorders, and authors’ conclusions about causality; and second, between the standards applied to causal claims about vaccine harms and those applied to causal claims about vaccine benefits. This pattern raised two narrower empirical questions about interpretive practice. First, when mechanistic plausibility is presented, how is it treated in authors’ conclusions—is the relationship between what is documented and what is concluded internally consistent? Second, how are standards of evidence for causality—traditionally articulated in vaccination research through the Bradford Hill criteria—applied across claims about vaccine harms and benefits?

Building on the scoping review mentioned earlier and a published protocol [15], this project addresses these questions through a two-phase qualitative coding design. Phase 1, reported in the present article, focuses on biological plausibility. It applies a structured three-tier typology (Consistent, Ambiguous, Inconsistent) to assess how the mechanistic explanations proposed by authors align with their interpretive conclusions. Phase 2, to be reported separately, will focus on epistemic integrity. It will apply a structured four-tier typology (High Integrity, Moderate Integrity, Low Integrity, Epistemically Neutral) to examine whether standards of evidence for vaccine benefits and harms are applied symmetrically within the same literature. Because the overall goal is interpretive, neither phase seeks to adjudicate the biomedical “truth” of vaccine-related claims, nor to substitute epidemiologic inference with rhetorical critique. Rather, the two phases together examine a rarely studied feature of vaccine safety research, and of medical evidence interpretation more broadly: how reported evidence is translated into causal conclusions and, ultimately, into clinical or policy-relevant guidance.

2. Methods

2.1. Study Design

This study reports Phase 1 of a two-phase qualitative, document-based analysis of interpretive practices in the biomedical literature on Covid-19 vaccination and autoimmune disorders. The broader project builds on a completed scoping review of 109 peer-reviewed articles [8], but differs from that review in analytic purpose. Whereas the scoping review asked what had been reported about autoimmune outcomes following Covid-19 vaccination, Phase 1 asks how authors interpret what they themselves report, including proposed or suggested mechanisms of action.

The purpose of Phase 1 was intentionally bounded and analytically self-contained: to assess whether authors’ conclusions about causality were internally consistent with the mechanistic and clinical material presented in the same article. Specifically, we examined alignment between reported mechanisms, relevant contextual elements such as temporality, and conclusions regarding causality or lack thereof. We also documented rhetorical strategies directly relevant to this alignment, including confirmation of causal relevance, disclaimers, narrative distancing, competing explanations, and qualifiers such as “rare”, “anecdotal”, or “safe”. These were not treated as evidence of bias, but as textual features shaping how mechanistic plausibility was incorporated into causal reasoning. The operationalization of interpretive consistency is described shortly.

Informed by a critical realist philosophy of science, mechanisms of action reported in the literature were examined as claims about underlying biological processes, while interpretive conclusions were examined as epistemic judgments made by authors on the basis of those claims. From this perspective, biological mechanisms are understood as real features of the world that may or may not be adequately captured in empirical reports, and whose interpretation requires careful attention to how evidence is presented and evaluated [16]. This orientation allowed us to treat biological mechanisms as more than rhetorical claims, without presuming that their presence alone established causality.

2.2. Data Source

The empirical corpus for the broader two-phase project consisted of the 109 scientific articles identified, screened, and included in the scoping review on Covid-19 vaccination and autoimmune disorders. That review focused on six selected autoimmune disorders: Graves’ disease (GD), Hashimoto’s thyroiditis (HT), multiple sclerosis (MS), rheumatoid arthritis (RA), systemic lupus erythematosus (SLE), and type 1 diabetes (T1D). The scoping-review corpus included a range of article types, including case reports, case series, and observational studies, and addressed autoimmune outcomes temporally associated with Covid-19 vaccination. These outcomes included flares or relapses among individuals with pre-existing autoimmune disease, new autoimmune manifestations among individuals with prior autoimmunity, and new-onset autoimmune disorders among individuals without known prior autoimmunity. All articles were available in full-text PDF format and had been retrieved through academic databases as part of the earlier review; no additional data sources were included in the present analysis.

Phase 1 comprised a narrower subset of this corpus. Because the present analysis addressed a distinct question from the scoping review, inclusion in the broader scoping-review dataset did not automatically qualify an article for Phase 1. Rather, articles were retained only when they explicitly discussed a plausible vaccine-linked autoimmune mechanism of action in relation to autoimmune disease onset, flare, relapse, or activation. Articles were excluded if they reported no mechanism, or if they addressed only short-term reactogenicity, general vaccine safety, metabolic deterioration, hyperglycaemic emergencies, or viral reactivation, unless the authors explicitly connected the reported outcome to autoimmune pathogenesis.

The purpose of this corpus refinement was not to determine whether the reported vaccine–autoimmunity associations were causal, but to delimit the articles suitable for an interpretive consistency analysis. To be retained, articles had to connect vaccination, autoimmune outcomes, and plausible immunological mechanisms in the authors’ own account. This boundary allowed the analysis to distinguish articles that advanced a mechanistic account of vaccine-associated autoimmunity from those that addressed immune disturbance, adverse events, or clinically significant post-vaccination outcomes in a more general sense. It also helped ensure that the Phase 1 corpus was defined by a stable analytic criterion rather than by the rhetorical usefulness or clinical importance of individual cases.

2.3. Coding Framework and Extraction

Articles were coded using a structured qualitative content analysis approach, following the predefined extraction categories specified in the published protocol. Because Phase 1 focused specifically on mechanisms of action and interpretive consistency, articles were re-charted to support this more specific analytic purpose. As part of this charting process, each article was systematically characterized by study type, first author institutional affiliation and country, funding sources and declared conflicts of interest, study population, vaccine type, and autoimmune condition(s) addressed. This re-characterization was conducted not to generate new descriptive findings, but to situate subsequent analyses within a clearly defined empirical corpus and ensure analytic transparency.

Following this initial charting, each article was examined in full and coded for the following Phase 1 analytic elements:

• Mechanisms of action, classified into immunological categories explicitly discussed by study authors, where applicable, including molecular mimicry, bystander activation, cytokine dysregulation, epitope spreading, polyclonal activation, and autoimmune/inflammatory syndrome induced by adjuvants (ASIA).
• Autoimmune outcomes, including flares, relapses, and new-onset autoimmune disorders, as well as the populations affected (individuals with pre-existing autoimmune disease versus individuals without prior autoimmunity).
• Interpretive conclusions regarding causality, including explicit assertions, denials, neutral positions, or hedging.
• Rhetorical framing strategies relevant to interpretation, such as disclaimers, narrative distancing, or qualifiers (e.g., “rare”, “anecdotal”, or “safe”).

For the purpose of Phase 1, references to genetic factors were not coded as mechanistic evidence, as they address the question of who may develop a condition rather than how the condition develops mechanistically. Where genetic explanations were invoked alongside proposed immunological mechanisms, they were recorded as part of the authors’ interpretive framing. This approach is consistent with immunological frameworks that distinguish triggering mechanisms from background susceptibility, as articulated in foundational work on vaccine-associated autoimmunity (e.g., [12]).

Mechanistic explanations were extracted verbatim and recorded alongside corresponding interpretive statements to facilitate direct comparison between the evidence authors reported and the conclusions they drew from that evidence. This approach allowed us to examine whether proposed mechanisms were treated as supporting causal plausibility, left indeterminate, or rhetorically neutralized in the discussion and conclusion sections.

2.4. Assessment of Interpretive Consistency

Interpretive consistency was assessed by examining how authors treated the biological mechanisms they reported in relation to their causal conclusions. Specifically, the analysis examined whether reported mechanisms were used to support causal interpretations, acknowledged but left unresolved, or effectively neutralized in conclusions that denied or distanced causal implication without engaging with the reported mechanisms.

The typology used to assess interpretive consistency was developed inductively from recurring patterns observed during the scoping review. Articles were classified as Consistent when authors proposed plausible biological mechanisms and allowed them to inform causal interpretation or clinical implications, for example by identifying vaccination as a possible trigger, stating that causality could not be excluded, or calling for caution, monitoring, individualized risk assessment, or reconsideration of vaccination protocols.; Ambiguous when mechanistic plausibility was acknowledged but causal interpretation remained unresolved; and Inconsistent when authors proposed or discussed mechanisms but explicitly dismissed, downplayed, or negated causal implication. These categories are capitalized throughout the analysis to indicate their use as analytically defined rather than everyday evaluative descriptors.

The assessment was descriptive and comparative. It did not seek to evaluate the biomedical validity of reported mechanisms or determine whether they were sufficient to establish causality. Rather, it examined whether authors’ interpretive conclusions coherently reflected the mechanistic content they themselves presented. This conservative classification approach was adopted to avoid extending interpretive judgment beyond what was warranted at this analytic stage. Particular attention was given to the treatment of temporality, especially in cases where autoimmune outcomes followed vaccination closely in time but were dismissed without explicit consideration of biological plausibility. The features listed in Table 1 are illustrative rather than exhaustive; articles were classified based on overall interpretive alignment rather than the presence or absence of any single feature.

2.5. Coding Process and Reliability

Each article was independently coded by two team members using standardized extraction templates developed for Phase 1. Coding focused on identifying interpretive consistency, ambiguity, or inconsistency, by comparing authors’ reported mechanisms, clinical observations, and causal language with the conclusions they drew from them. The analysis did not seek to adjudicate the biomedical truth value of authors’ claims or determine causality. Rather, coders were asked to justify each classification with reference to the article’s own text—for example, by showing how an article proposed a vaccine-related mechanism but later qualified, displaced, or neutralized its causal significance. Discrepancies between coders were resolved through full team discussion, and consensus decisions were documented.

Inter-rater reliability was assessed after coding an initial subset of articles, with a target agreement threshold of 80%, consistent with established qualitative research standards [17]. Where agreement fell below this threshold, coding guidelines were refined and applied retroactively to ensure consistency across the dataset. Reflexive memos were maintained throughout the coding process to document analytic decisions and enhance transparency [18].

2.6. Data Analysis and Presentation

Phase 1 data were analysed descriptively and thematically. The descriptive analysis summarized the types and frequency of mechanisms of action reported, as well as patterns of alignment or misalignment between mechanistic reporting and interpretive conclusions. Frequencies were reported as counts and percentages using the relevant analytic denominator for each calculation. Percentages were rounded to one decimal place using standard rounding conventions; therefore, totals may not always sum to 100.0% because of rounding.

The thematic analysis, informed by Braun and Clarke’s approach, examined how authors articulated mechanistic explanations, contextual evidence, causal language, and conclusions within the same article, with particular attention to recurring patterns of interpretive attenuation or alignment (Braun & Clarke, 2006). Verbatim quotations were used where necessary to illustrate interpretive patterns and show how mechanisms, uncertainty, causal claims, and rhetorical framing were presented in the literature.

Findings are presented in narrative and tabular form, with illustrative examples drawn from the coded articles. For readability, vaccine names are reported using brand or commonly recognized names, rather than the technical nomenclature used inconsistently across the included articles.

2.7. Statement on Reflexivity

The research team consisted of three investigators with complementary disciplinary and professional backgrounds spanning clinical medicine, biochemistry, pharmacy, medical sociology, and health policy analysis. Team members also brought lived experience of the Covid-19 policy response within Canadian public institutions. Consistent with established guidance on reflexivity in medical qualitative inquiry [19], these disclosures are offered not as personal or ideological positioning, but to enhance transparency and situate the knowledge claims advanced in this study. Reflexivity was maintained throughout the research process through memoing, team discussion, and explicit attention to the distinction between data and interpretation. The study also builds on a substantial body of individual and collaborative research by the lead investigator engaging medical, sociological, and policy aspects of Covid-19 [8,20,21,22,23]. This article reports the first phase of a study protocol published in advance of data collection [15].

2.8. Ethical Considerations

This study involved secondary analysis of publicly available scientific documents; Institutional Review Board approval was therefore not required.

3. Results

3.1. Article Characteristics

A total of 52 of the 109 articles included in the scoping review discussed biological mechanisms of action linking Covid-19 vaccination and autoimmune outcomes and were therefore selected for Phase 1 analysis. This section summarizes the key characteristics of this analytic corpus, including study design, country of first-author affiliation, autoimmune condition(s) addressed, vaccine type, and interpretive-consistency classification.

The analytic corpus was dominated by case-based literature. Most articles were case reports (36/52, 69.2%), followed by case series or case-based reviews (7/52, 13.5%), cohort studies or cohort reports (5/52, 9.6%), cross-sectional studies (3/52, 5.8%), and one retrospective observational study (1/52, 1.9%). First authors were affiliated primarily with institutions in the United States (10/52, 19.2%), Japan (9/52, 17.3%), and Italy (6/52, 11.5%), with additional contributions from researchers based in Austria, Belgium, Brazil, China, Colombia, France, Germany, Greece, Hong Kong (China), India, Iran, Israel, Korea, Peru, Puerto Rico, Saudi Arabia, Singapore, Spain, Taiwan, Thailand, Tunisia, and Turkey.

All six autoimmune conditions examined in the scoping review were represented in this subset. Most articles focused on a single condition (48/52, 92.3%), while a small minority addressed more than one condition or had a broader mixed autoimmune/autoinflammatory focus (4/52, 7.7%). The most frequently represented condition was systemic lupus erythematosus (21/52, 40.4%), followed by Graves’ disease (10/52, 19.2%), rheumatoid arthritis (8/52, 15.4%), multiple sclerosis (7/52, 13.5%), Type 1 diabetes/autoimmune diabetes (6/52, 11.5%), and Hashimoto’s thyroiditis (3/52, 5.8%). These frequencies include articles in which a condition was reported alone as well as articles in which it appeared alongside another autoimmune condition; therefore, condition-specific counts do not sum to 52.

In terms of autoimmune outcomes, a majority of articles reported new-onset autoimmune conditions, generally in individuals without documented prior autoimmunity (30/52, 57.7%). A substantial minority reported post-vaccination adverse events, relapses, flares, exacerbations, or new autoimmune manifestations in patients with pre-existing autoimmune disorders (17/52, 32.7%), while a smaller group reported mixed outcomes involving both new-onset and pre-existing autoimmune conditions (5/52, 9.6%).

Three Covid-19 vaccine platforms were represented: mRNA, viral vector, and inactivated virus. Most articles involved mRNA vaccines only (38/52, 73.1%), followed by mixed-platform (7/52, 13.5%), viral-vector vaccines (4/52, 7.7%), and inactivated-virus vaccines (3/52, 5.8%).

The most frequently reported vaccine brand was Pfizer-BioNTech (39/52, 75.0%), followed by Moderna (11/52, 21.2%), AstraZeneca (4/52, 7.7%), Johnson & Johnson (4/52, 7.7%) and Sinovac-CoronaVac (4/52, 7.7%). Less frequently reported brands included Sinopharm (1/52, 1.9%), Covaxin (1/52, 1.9%) and Covishield (1/52, 1.9%). One article did not report the vaccine brand (1/52, 1.9%), and one reported brand in all but a minority of study subjects (1/52, 1.9%). Finally, a small proportion of articles reported multiple vaccine brands (11/52, 21.2%); therefore, vaccine brand counts do not sum to 52.

3.2. Analysis of Interpretive Consistency

Applying the typology of interpretive consistency, a small number of articles were classified as Consistent (6/52, 11.5%), an equal proportion as Ambiguous (7/52, 13.5%), and most as Inconsistent (39/52, 75.0%). The following sections describe these three categories in greater detail, illustrate them with selected quotations, and explain the rationale underlying the classificatory process. Supplementary file 1 summarizes the 52 articles included in Phase 1, including basic article characteristics and a brief rationale for each interpretive-consistency classification.

Within each category, selected examples are presented alphabetically by autoimmune condition to provide a consistent organizational logic. Table 2 summarizes the main interpretive moves identified across the analysis, showing how similar evidentiary elements could support consistent interpretation, remain unresolved, or be attenuated depending on how authors carried them into their conclusions.

3.2.1. Articles with Consistent Interpretive Alignment with Reported Mechanisms

Across articles in this category, authors proposed biologically plausible immunological mechanisms linking Covid-19 vaccination to autoimmune outcomes and treated those mechanisms as relevant to causal interpretation. They also typically documented close temporal proximity between vaccination and symptom onset and situated their cases within prior literature on similar post-vaccination autoimmune phenomena.

Interpretive conclusions in these articles were aligned with the evidentiary material presented. In different ways, authors identified vaccination as a possible trigger, acknowledged that causality could not be excluded, or treated the temporal and mechanistic relationship as clinically meaningful despite uncertainty. These patterns were evident in the four articles discussed below, which address Graves’ disease, Graves’ ophthalmopathy, lupus nephritis, and systemic lupus erythematosus exacerbation. Across these articles, authors linked clinical onset or worsening to proposed mechanisms and recommended caution, closer monitoring, or further attention to autoimmune outcomes rather than dismissing causal relevance.

Example 1: Graves’ Disease Relapse and New Onset

Zettinig and Krebs (2022) reported two cases of Graves’ disease following the administration of Pfizer-BioNTech vaccines: one relapse after 17 years in a patient with prior Graves’ disease and one new-onset case in a patient with, as the authors noted, “no prior thyroid disease”. In the first case, a woman with prior Graves’ disease had been euthyroid with normal TSH-receptor antibody levels at yearly check-ups before receiving two doses of the vaccine. Several weeks after vaccination, she developed symptoms and laboratory evidence of recurrent Graves’ disease. In the second case, a man with no prior thyroid disease and previously documented euthyroidism developed hyperthyroidism after his first vaccine dose. Both cases also showed unusual sonographic and scintigraphic findings [24].

The authors situated their observations within existing literature on Graves’ disease after both Covid-19 infection and vaccination. They discussed autoimmune/inflammatory syndrome induced by adjuvants (ASIA), but did not restrict their interpretation to that mechanism, stating that “not only a random coincidence or ASIA could have induced autoimmune hyperreactivity in our patients, but also the vaccine itself could have triggered autoimmune thyroid disease”. They further concluded that their findings indicated that “in the ongoing phase 3 trials, thyroid autoimmunity should be in the focus of the investigators”.

Therefore, the article was classified as Consistent because the authors connected temporal sequence, clinical findings, prior literature, and plausible autoimmune mechanisms to the observed thyroid dysfunctions, and directed further research attention towards vaccination. Importantly, they explicitly held open the possibility that vaccination itself could have caused the autoimmune outcomes. The conclusion therefore preserved the causal relevance of the reported cases, maintaining alignment between their clinical observations, the posited mechanistic interpretation, and their concluding recommendation.

Example 2: Graves’ Ophthalmopathy Exacerbation

Patrizio et al. (2022) reported two cases of Graves’ ophthalmopathy recrudescence after administration of the Pfizer-BioNTech vaccine. In the first case, a woman with a history of Graves’ disease and Graves’ ophthalmopathy experienced worsening ophthalmopathy three days after receiving the second dose. In the second case, a man with prior Graves’ disease and Graves’ ophthalmopathy developed recrudescence of moderate-to-severe active Graves’ ophthalmopathy two weeks after vaccination. In both cases, thyroid function was normal, TSH-receptor antibodies were elevated, and the authors noted the absence of recent Covid-19 infection and other potentially triggering factors [25].

The authors proposed immune-mediated mechanisms, including autoimmune/inflammatory syndrome induced by adjuvants, polyethylene glycol (PEG) in lipid nanoparticles, and the self-adjuvant properties of mRNA. They noted that “PEG may act as adjuvant” and that mRNA can elicit cytokine-mediated immune responses. At the same time, their causal interpretation was qualified through the language of coincidence and predisposition: after stating that “coincidental occurrence of [Graves ophthalmopathy] and the vaccine administration [could not] be excluded”, they described vaccine adjuvants as capable of producing severe adverse events in “genetically susceptible and predisposed subjects”.

This framing softened the causal claim by distributing explanation between vaccine exposure, coincidence, and patient susceptibility, but did not displace vaccination as a possible trigger: the qualifications were followed by the author’s statement that “the close temporal relationship between [vaccination and autoimmune outcomes] raises questions about the potential immune stimulation elicited by the vaccine injection”. Finally, while prefacing their last paragraph with reference to “the tremendous benefits brought out for the public health by the SARS-CoV-2 vaccination”, the authors nevertheless concluded that “all the collected cases of autoimmune reactions, inclusive of autoimmune thyroid diseases and, now of GO [Graves’ ophthalmopathy], should raise health care providers’ caution of patients affected by autoimmune disease”.

The article was therefore classified as Consistent because its qualifying language did not reverse the direction of the mechanistic evidence linking vaccination to the observed autoimmune outcomes, nor did it prevent the authors from drawing a cautionary implication for patients with autoimmune disease.

This is important because the article contained several elements that could have been used to neutralize the reported association, including reference to individual predisposition, the impossibility of excluding coincidence, and the public-health benefits of vaccination. However, the authors did not allow these qualifications to erase the clinical meaning of the cases. They emphasized the close temporal relationship, the absence of alternative triggering factors, the immune-mediated pathways through which vaccination could plausibly contribute to Graves’ ophthalmopathy recrudescence, and the need for caution in patients with autoimmune disease. In this sense, the article maintained interpretive alignment between the clinical observations, proposed mechanisms, and concluding recommendation.

Example 3: Lupus Nephritis with Multi-Organ Involvement

Kim et al. (2022) reported new-onset class III lupus nephritis with multi-organ involvement after administration of the AstraZeneca vaccine. The patient had a prior history of rash and a positive antinuclear antibody test, but no autoimmune diagnosis had been established, and no further autoantibody testing had been performed. Before vaccination, she had no known Covid-19 infection, had reported no other medical disease, and was not taking medication. A general health check-up shortly before vaccination showed normal creatinine and negative urine findings. After receiving the second vaccine dose, the authors reported that the patient developed asthenia, lack of appetite, and foamy urine, followed by edema, fever, renal impairment, proteinuria, hematuria, positive autoantibodies, low complement, and biopsy-confirmed lupus nephritis with multi-organ involvement [26].

The authors proposed immunological mechanisms linking vaccination to the autoimmune outcome, including vaccine-induced T helper 1 (Th1) responses, expansion of CD8+ T cells, enhanced cytokine production, and cross-reactivity between antibodies against the SARS-CoV-2 spike protein and tissue antigens. They also situated the case within an emerging literature on post-vaccination glomerulonephritis, noting that the incidence of glomerulonephritis after Covid-19 vaccination had been increasing. Their mechanistic discussion therefore did not stand apart from the clinical case; rather, it was presented as a plausible explanation for the patient’s rapid development of lupus nephritis with systemic involvement after vaccination.

The article was therefore classified as Consistent because its causal interpretation was directly aligned with the mechanistic and clinical evidence presented. The authors did not merely state that the event followed vaccination, nor did they neutralize the association through rarity, coincidence, predisposition, or generalized reassurance about vaccine benefits. Instead, they concluded that the Covid-19 vaccine was a “key trigger that elicited an autoimmune response and the development of lupus nephritis” in the patient. In this case, temporality, biopsy-confirmed disease, multi-organ involvement, immunological mechanism, and causal conclusion were treated as mutually reinforcing rather than disconnected observations.

Example 4: Exacerbation of Systemic Lupus Erythematosus

Sugimoto et al. (2022) reported severe exacerbation of systemic lupus erythematosus after the Moderna vaccine in a patient with a 12-year history of systemic lupus erythematosus whose condition had been stable since 2017. The patient developed fever and worsening erythema two weeks after the first dose. After the second dose, she developed high fever, muscle pain, epistaxis, stomatitis, worsening rash, digital ulcers and gangrene, chest pain, hair loss, pleural effusion, leukopenia, thrombocytopenia, elevated creatine kinase and liver enzymes, markedly elevated ferritin, and low complement levels. She was diagnosed with severe systemic lupus erythematosus exacerbation, hospitalized, treated with methylprednisolone and prednisone, and subsequently improved [27].

The authors proposed a mechanism involving mRNA-induced type I interferon production, noting that externally injected mRNA induces interferon-alpha (IFN-α) and interferon-beta (IFN-β) and that type I interferon is considered especially important in systemic lupus erythematosus. They described this as the first report of systemic lupus erythematosus exacerbation “induced by the mRNA-1273 vaccine”. Although they observed that “treatment at the time of vaccination may affect the outcome”, this qualification did not displace vaccination as a relevant factor. Rather, the authors returned to vaccination as the source of clinical risk, concluding that patients with lupus and cutaneous symptoms “may be at risk” of disease exacerbation after mRNA vaccination.

The article was therefore classified as Consistent because the authors treated the proposed biological mechanism, temporal sequence, and clinical course as relevant to causal interpretation. Their qualification regarding treatment status introduced a possible modifying factor, but did not neutralize the vaccine-related interpretation. Unlike articles in the Inconsistent category, Sugimoto et al. did not redirect the case toward rarity, coincidence, presumed predisposition, or generalized reassurance about vaccination. Instead, they maintained alignment between the clinical evidence presented, the mechanism proposed, and the cautionary implication drawn for patients with lupus and cutaneous symptoms.

3.2.2. Articles with Ambiguous Interpretive Alignment with Reported Mechanisms

Across articles classified as Ambiguous, authors reported biologically plausible immunological mechanisms and documented autoimmune or immune-mediated outcomes following Covid-19 vaccination, often in close temporal proximity, but left the causal implications unresolved. In contrast to articles classified as Consistent, these articles did not clearly treat vaccination as a probable or clinically actionable trigger. At the same time, they did not fully neutralize mechanistic plausibility or deny the relevance of the reported temporal and biological evidence.

A recurrent feature of this category was interpretive suspension: mechanistic explanations, temporal sequence, clinical findings, and prior literature were presented alongside statements emphasizing uncertainty, coincidence, alternative explanations, or the need for further research. These moves did not necessarily dismiss the possibility of vaccine-related autoimmune or immune-mediated harm; rather, they left unclear how much causal weight should be assigned to the mechanisms and clinical patterns described.

The examples below, drawn from reports addressing Graves’ disease, Hashimoto’s thyroiditis, multiple sclerosis, and rheumatoid arthritis, illustrate several related forms of ambiguity. These included distributing causal interpretation across multiple possible triggers, acknowledging vaccine-associated harm while deferring causal interpretation, redirecting mechanistic concern into post-vaccination clinical awareness, and qualifying mechanistic plausibility through coincidence or other unresolved alternatives.

Across these examples, ambiguity was not limited to causal language. It also appeared in the practical implications drawn from the evidence. Authors recognized that autoimmune or immune-mediated events might occur after vaccination, but rarely translated that recognition into pre-vaccination caution, individualized risk assessment, or reconsideration of vaccination timing in potentially susceptible patients. The cases therefore show how mechanistic plausibility could be acknowledged while its causal and clinical implications remained unresolved.

Example 1: New-Onset Graves’ Disease in Patient with No Prior Autoimmunity

Hamouche et al. (2022) reported a case of new-onset Graves’ disease in a previously healthy 32-year-old man after Pfizer-BioNTech vaccination and confirmed SARS-CoV-2 infection. The patient developed Covid-like symptoms ten days after receiving the first vaccine dose, and infection was later confirmed by polymerase chain reaction testing. After recovery from the acute infection, he presented with palpitations, insomnia, tremor, irritability, diaphoresis, and dyspnea at rest. Graves’ disease was confirmed through thyroid function testing, positive thyroid-stimulating immunoglobulin, thyroid peroxidase antibodies, anti-thyroglobulin antibodies, and radioactive iodine uptake finding [28].

The authors discussed autoimmune/inflammatory syndrome induced by adjuvants as a plausible mechanism linking vaccination to autoimmune thyroid disease, postulating that “exposure to some vaccine adjuvants could lead to a cascade of immunological responses leading to autoantibodies production”, and observing that “ASIA was also described in other studies where the patients developed autoimmune thyroiditis after the administration of a human papillomavirus (HPV) vaccine and an influenza vaccine”. They also noted that, in their case, Graves’ disease symptoms occurred 22 days after the first Pfizer-BioNTech vaccine dose.

The ambiguity appeared in the way the authors distributed causal interpretation across several possibilities. Although vaccination was the first documented event in the sequence—preceding both the confirmed SARS-CoV-2 infection and the later diagnosis of Graves’ disease—the conclusion did not assign particular causal weight to this chronology. Instead, the authors described the case as “the first case report of a new-onset of GD in a young male adult following an administration of a Covid-19 vaccine and recovery from a mildly symptomatic Covid-19 infection”, and concluded that “whether the development of the GD in this patient was triggered by the Covid-19 infection, a side effect of the vaccine, both the infection and the vaccine, or purely coincidental is subject to further studies and analysis”.

The article was therefore classified as Ambiguous. The authors did not deny a possible vaccine-related autoimmune mechanism, and explicitly included vaccination as one possible trigger. However, by placing it alongside infection, combined exposure, and coincidence without adjudicating among them, or assigning interpretive weight to vaccination as the first documented exposure, its role was diluted and its causal relevance to the case remained unresolved. This ambiguity is especially notable because the patient was described as previously healthy, making the case clinically significant even though the authors refrained from drawing a clear causal conclusion or identifying actionable research, clinical, or policy implications.

Example 2: Multiple Sclerosis Relapse

Kataria et al. (2022) reported a case of multiple sclerosis relapse 18 days after the second dose of the Pfizer-BioNTech vaccine. The patient had been diagnosed with multiple sclerosis six years earlier, had been in remission, was compliant with interferon beta therapy, and had no new lesions on recent follow-up magnetic resonance imaging. After vaccination, she developed fatigue, tingling, numbness, stiffness, involuntary eye movements, and blurry vision. Clinical and radiological findings confirmed relapse, while Covid-19 polymerase chain reaction testing and infectious workup were negative [29].

The authors discussed several plausible pathways linking vaccination to autoimmune demyelination, including cytokine upregulation, epitope spreading, polyclonal activation, cross-reactivity involving spike-protein antibodies and myelin basic protein, and inflammatory responses involving ACE2 receptors in the blood-brain barrier or spinal neurons. They also acknowledged that vaccine-related multiple sclerosis activity had been reported previously, noting that a possible link had been suggested within 30 days after vaccination, given the possibility of vaccines “enhancing the transition from subclinical to clinical disease after stimulating the immune response”.

The ambiguity appeared in the conclusion. On the one hand, the authors stated that “our case provides evidence of vaccine-associated MS relapse”. On the other hand, they immediately added that “causality cannot be established” and that “more cases of MS relapse are needed to confirm vaccine-related etiology”. The article was therefore classified as Ambiguous because the authors acknowledged a vaccine-associated relapse and presented mechanistic and clinical evidence consistent with that interpretation, but deferred causal assessment to future studies rather than drawing a clear causal conclusion from the case itself.

Example 3: Rheumatoid Arthritis Flare with Epstein-Barr Virus Positivity

Nakamura et al. (2022) reported a severe flare of rheumatoid arthritis in a 60-year-old woman with a 20-year history of rheumatoid arthritis, long-term disease control on low-dose prednisolone, and diabetes. Ten days after receiving the first Moderna mRNA vaccine dose, she developed high fever and arthralgia, followed by polyarthritis, active synovitis on ultrasound, elevated inflammatory markers, and high clinical disease activity. The authors also reported positive Epstein-Barr virus markers and detectable Epstein-Barr virus DNA, while SARS-CoV-2 testing and other infectious investigations were negative. Rheumatoid arthritis activity improved after treatment with tocilizumab, and Epstein-Barr virus markers later became negative [30].

The authors explicitly framed the case as vaccine-related. The title referred to “SARS-CoV-2 vaccination-related activation of rheumatoid arthritis”, and the abstract described a “severe flare of rheumatoid arthritis” following mRNA vaccination. In the discussion, the authors stated that “the pretreatment clinical features of this case were severe flare of RA and EBV positivity 10 days after the administration of the mRNA-type SARS-CoV-2 vaccine”, and later wrote that “we concluded that this was a case of SARS-CoV-2 vaccine-induced arthritis”. They proposed a plausible mechanism in which nucleic acids in the mRNA vaccine might be recognized by innate immune receptors, leading to acquired immune activation. They also discussed possible recognition of mRNA vaccine components by toll-like receptors, type I interferon production, and T-cell immune activation.

The ambiguity appeared in the authors’ handling of Epstein-Barr virus positivity and the incomplete causal pathway. On the one hand, they repeatedly acknowledged vaccine-relatedness and described the case as vaccine-induced arthritis. On the other hand, they stated that the mechanism by which vaccination caused arthritis was unclear, that the mechanism connecting vaccination with Epstein-Barr virus emergence was also difficult to identify, and that no studies had confirmed a link between innate immune activation and increased Epstein-Barr viral load. They also acknowledged that they had no pre-vaccination Epstein-Barr virus data and could not exclude the possibility of unknown viral involvement or the natural course of viral infection.

The article was therefore classified as Ambiguous. The authors did not merely identify mechanisms of action; they actively endorsed vaccine causation as a plausible explanation for the patient’s rheumatoid arthritis flare, even describing the case as “SARS-CoV-2 vaccine-induced arthritis”. At the same time, the discussion and conclusion presented an incomplete causal account in which vaccination, rheumatoid arthritis flare, Epstein-Barr virus positivity, and a long-Covid-like immune pattern were temporally and clinically linked, but not fully disentangled. The long-Covid comparison did not displace the vaccine as a possible trigger, especially since SARS-CoV-2 testing was negative, but it broadened the explanatory frame beyond vaccination alone. The ambiguity lay in this unresolved relationship between mRNA vaccine immune activation, Epstein-Barr virus reactivation, possible viral or post-viral immune pathways, and rheumatoid arthritis flare. The final implication was monitoring of Epstein-Barr virus status during rheumatoid arthritis treatment to clarify flare-associated factors after vaccination, rather than a clear causal conclusion, pre-vaccination risk assessment, or guidance about subsequent vaccination in similar patients.

Example 4: New-Onset Rheumatoid Arthritis in Patient with No Prior Autoimmunity

Watanabe et al. (2022) documented a case of new-onset rheumatoid arthritis in a previously healthy 53-year-old man four weeks after the second dose of the Pfizer-BioNTech vaccine. The patient developed left knee swelling and pain, bilateral shoulder pain, and morning stiffness. Laboratory testing showed marked leukocytosis and elevated C-reactive protein, both of which had been normal before the second vaccination, as well as elevated anti-cyclic citrullinated peptide antibody and rheumatoid factor. Magnetic resonance imaging showed diffuse knee effusion, and the patient was diagnosed with rheumatoid arthritis. Initial treatment with methotrexate and prednisolone, followed by etanercept, did not induce remission. Remission was achieved after treatment with methotrexate and tocilizumab [31].

The authors provided a detailed account linking vaccination to the autoimmune presentation. They argued that messenger RNA derived from Pfizer-BioNTech vaccine can induce type I interferon responses through toll-like receptors, retinoic acid-inducible gene I, and melanoma differentiation-associated gene 5, and reported that serum and synovial fluid concentrations of interleukin-6, tumour necrosis factor-alpha, and type I interferon were elevated during the active phase. They stated that “pro-inflammatory cytokine responses triggered by Covid-19 vaccination might be involved in the development of de novo RA”, and later that “BNT162b2 vaccination could be a trigger for RA development”. The mechanistic interpretation was further supported by treatment response: the authors noted that the patient “was successfully treated by the blockade of IL-6”, which they interpreted as suggesting a predominant role for that cytokine.

The ambiguity appeared in the way the causal interpretation remained suspended between mechanistic plausibility and coincidence. The authors did not dismiss vaccination as causally irrelevant; indeed, they proposed that vaccine-induced type I interferon responses might trigger arthrogenic cytokine responses leading to rheumatoid arthritis. However, after presenting this mechanistic and therapeutic support, they added that “we cannot exclude the possibility that the timing of RA development with regard to vaccination was coincidental” and concluded that large-scale epidemiological studies were needed “to establish its link”. The order of these interpretive moves is important: the article first developed a biologically plausible vaccine-trigger hypothesis, then left coincidence as the final unresolved qualifier. The article was therefore classified as Ambiguous because the authors identified vaccination as a possible trigger but ultimately left the causal meaning of the case unresolved.

3.2.3. Articles with Inconsistent Interpretive Alignment with Reported Mechanisms

Across articles classified as Inconsistent, authors reported biologically plausible immunological mechanisms linking Covid-19 vaccination to autoimmune outcomes and often documented additional contextual evidence, including close temporal proximity, absence of prior autoimmune disease, clinical severity, response to immunosuppressive treatment, or recurrence after additional vaccine exposure. However, these evidentiary elements were not carried forward consistently into the authors’ interpretive conclusions.

In contrast to articles classified as Ambiguous, where causal implications were left unresolved, articles in the Inconsistent category moved beyond uncertainty by introducing interpretive strategies that reduced or neutralized the causal and clinical significance of the mechanisms described. These strategies included susceptibility or predisposition framing, appeals to rarity, population-scale reasoning, reassurance about continued vaccination, benefit language not directly supported by the case evidence, redirection toward infection-related mechanisms, and narrow post-event surveillance recommendations. In some cases, authors explicitly acknowledged that vaccination may have triggered the autoimmune outcome, but then qualified this possibility in ways that shifted explanatory weight away from the vaccine and toward patient biology, background incidence, or generalized assumptions about vaccine benefit.

A recurrent feature of this category was the retrospective invocation of predisposition, genetic susceptibility, or pre-symptomatic disease. Such explanations were often introduced after the adverse event had occurred, without being independently demonstrated or operationalized as pre-vaccination risk criteria. Rather than clarifying causal reasoning, these formulations frequently functioned to contain the implications of the authors’ own evidence: the vaccine remained a possible trigger, but only in individuals redescribed after the fact as already vulnerable, predisposed, or in a latent disease phase.

Another recurrent pattern was the narrowing of practical implications. Several articles presented serious autoimmune outcomes and plausible mechanisms, yet concluded by emphasizing that the events were rare, transient, or treatable, and should therefore not discourage vaccination. Others recommended awareness or post-vaccination surveillance, without engaging whether the reported mechanisms and clinical events might warrant pre-vaccination caution, individualized risk assessment, or reconsideration of subsequent doses in similar patients. The inconsistency, therefore, did not lie in acknowledging uncertainty, nor in avoiding definitive causal claims from case reports or case series, but in the mismatch between the mechanistic and clinical evidence presented and the limited, reassuring, or deflective conclusions drawn from that evidence.

The examples below, drawn from reports addressing at least one example of each autoimmune condition examined in the study—Graves’ disease, Hashimoto’s thyroiditis, multiple sclerosis, rheumatoid arthritis, systemic lupus erythematosus, and Type 1 diabetes mellitus—illustrate several forms of this interpretive pattern. Across these articles, mechanistic plausibility was acknowledged, but its causal and clinical implications were softened, redirected, or neutralized in the authors’ conclusions.

Example 1: Graves’ Disease, New Onset and Relapse

Chee et al. (2022) reported 12 cases of Graves’ disease-related hyperthyroidism after Covid-19 mRNA vaccination. Six patients had new-onset Graves’ disease and six had relapse of previously well-controlled Graves’ disease. Patients developed hyperthyroidism within a relatively short interval after vaccination, with a median onset of 17 days and a range of 5 to 63 days. Five patients developed symptoms after the first dose and seven after the second dose. The authors also conducted a literature review and identified 21 additional cases of Graves’ disease occurring shortly after Covid-19 vaccination [32].

The authors discussed several plausible mechanisms through which Covid-19 mRNA vaccination could contribute to thyroid autoimmunity. They described molecular mimicry, activation of antigen-presenting cells, downstream T- and B-cell activation, and autoimmune/inflammatory syndrome induced by adjuvants. They also stated that the robust T helper 1 (Th1) cellular response induced by the vaccine “could be one of the possible mechanisms underpinning the unmasking of Graves’ disease in predisposed individuals or trigger relapse in a previously well-controlled individual with Graves’ disease”.

In discussing lipid nanoparticles and polyethylene glycol, they further noted that “the exposure could potentially induce an exaggerated immune response and precipitate the development of thyroid autoimmunity”, while adding that “ASIA remains poorly understood”. The reference to “predisposed individuals” qualified the causal interpretation by shifting part of the explanatory burden onto patient susceptibility, even though such predisposition was not independently demonstrated for any of the 12 cases but inferred largely from the occurrence of disease after vaccination.

The inconsistency appeared in the relationship between the authors’ mechanistic interpretation and their conclusion. On the one hand, they stated that their case series provided insight into individuals “in whom Graves’ disease was triggered by SARS-CoV-2 vaccination”. On the other hand, the conclusion shifted quickly from reported harm to vaccine continuation: “the SARS-CoV-2 vaccinations or boosters should not be delayed given the clear protection against severe disease conferred by the vaccine”. The authors further stated that, with timely diagnosis and treatment, “further doses of the SARS-CoV-2 vaccine can be safely administered” and unnecessary delay in completing the vaccination schedule minimized.

The article was therefore classified as Inconsistent because the conclusion did not maintain alignment with the weight of the clinical and mechanistic evidence presented. The authors reported 12 cases, identified 21 additional cases in the literature, described plausible mechanisms, and explicitly stated that Graves’ disease, whether new-onset or relapsed, were “triggered” by vaccination. Yet the interpretive conclusion redirected the practical implication toward avoiding delay in vaccination or boosters, while relegating the reported autoimmune outcomes to predisposition, awareness, timely diagnosis, and further research. The inconsistency lay in the way susceptibility framing, reassurance, and continuation of vaccination neutralized the clinical significance of the authors’ own evidence of vaccine-triggered Graves’ disease.

Example 2: Hashimoto’s Thyroiditis in Patient with No Prior Thyroid Disease

Lioulios et al. (2022) reported two cases of new-onset autoimmune thyroid disease after Pfizer-BioNTech vaccination in dialysis patients with no previous history of thyroid disease. The Hashimoto’s thyroiditis case involved a 67-year-old woman on continuous cyclic peritoneal dialysis for end-stage renal disease due to amyloid light-chain amyloidosis, previously treated with autologous bone marrow transplantation. She had no history of thyroid disease and had normal thyroid-stimulating hormone values during six-monthly laboratory evaluation. Fifty days after receiving two doses of the Pfizer-BioNTech vaccine, she presented with generalized weakness, fatigue, mood disorder, and hoarseness [33].

Laboratory testing showed exceptionally elevated thyroid-stimulating hormone, reduced free thyroxine and free triiodothyronine, and strongly positive antibodies against thyroglobulin and thyroid peroxidase. Ultrasonography showed an enlarged hypoechoic thyroid gland with echogenic septations and hypervascularity. She was diagnosed with Hashimoto’s thyroiditis and rhabdomyolysis, treated with levothyroxine, became asymptomatic five days later, and had thyroid hormone levels return to the reference range at follow-up.

The authors proposed several mechanisms through which Covid-19 vaccination could plausibly contribute to autoimmune thyroid disease. They discussed autoimmune/inflammatory syndrome induced by adjuvants and molecular mimicry, describing molecular mimicry between the spike protein encoded by Pfizer-BioNTech vaccine and thyroid antigens as “a more convincing idea”. They also discussed cytokine dysregulation and inflammatory pathways associated with SARS-CoV-2 immune responses, as well as immune dysfunction and relative immunosuppression in end-stage renal disease. In this context, altered adaptive immunity, impaired antibody production, and diminished regulatory T cells were presented as potentially relevant to an extended immune response and cross-reactivity to non-pathogenic antigens.

The authors also acknowledged that onset occurred relatively long after vaccination, but suggested that end-stage renal disease may have affected the interval between vaccination and disease manifestation. The article therefore did not merely report temporal association; it offered a mechanistic account connecting vaccination, immune dysregulation, dialysis-related immune vulnerability, and autoimmune thyroid disease.

The article was classified as Inconsistent because the conclusion did not maintain alignment with the mechanistic and clinical evidence presented. The authors documented new-onset Hashimoto’s thyroiditis after vaccination in a patient with no prior thyroid disease, discussed several mechanisms by which vaccination could contribute to autoimmune thyroid dysfunction, and, in discussing the two cases together, stated that the rarity of Graves’ disease in hemodialysis patients, post-vaccination development, and subsequent relapse after SARS-CoV-2 infection, “reinforces the causative relationship between the two events”. Yet in the next interpretive move, they stated that “none of the above proposed mechanisms have been proven to associate vaccination against SARS-CoV-2 with such disturbances”, and concluded that the cases did not aim to discourage vaccination but to “provide awareness and encourage further investigation”.

The inconsistency lay precisely in this rhetorical move: mechanisms and temporal-clinical evidence were presented as causally meaningful, but their implications were then sharply attenuated through the claim that proof was lacking and through reassurance that vaccination should not be discouraged. The practical implication was reduced to awareness and future investigation, without clarifying what clinical action should follow for similar patients before vaccination, after the first dose, or before subsequent doses.

Example 3: New-Onset Multiple Sclerosis After Recent mRNA Covid-19 Vaccination

Toljan et al. (2022) reported five cases of new-onset multiple sclerosis following recent exposure to mRNA Covid-19 vaccines. Symptoms occurred between one day and five weeks after vaccination. The patients ranged from 29 to 47 years of age, and none had a reported family history of autoimmune conditions. The cases included neurological symptoms such as weakness, numbness, paresthesia, gait disturbance, urinary symptoms, optic or sensory involvement, and radiological findings consistent with demyelinating disease. Several patients required high-dose corticosteroids, and one required plasmapheresis, indicating that the reported events were clinically significant rather than merely transient or mild post-vaccination symptoms [34].

The authors proposed a plausible immunological mechanism through which mRNA vaccination could have contributed to the emergence of multiple sclerosis. They stated: “The basis of mRNA vaccines includes stimulation of T and B cell response through exposure of dendritic cells to exogenous mRNA. This exaggerated immune response on a background of an already altered immune response in patients susceptible to MS may play a role in unmasking of disease after vaccination”. They further acknowledged that “our cases are possibly supportive of such hypothesis, as we demonstrate a temporal association between vaccine and a new diagnosis of MS with COVID-19 mRNA vaccination”. Thus, the article did not merely report temporal proximity; it also identified a mechanism by which vaccination-induced immune activation could plausibly contribute to the first clinical manifestation of disease.

The interpretive inconsistency appeared in how this mechanistic and clinical evidence was subsequently qualified. In their conclusion, the authors wrote: “The causality of vaccination and onset of MS cannot be determined. Based on these cases we cannot conclude whether vaccination represents a trigger in an otherwise predisposed or pre-symptomatic MS phase versus a purely spurious result as a consequence of vaccination in a very large proportion of the population, where incident cases occur independent of vaccination.” This formulation offered three alternatives: vaccination as a trigger in a predisposed person, vaccination as a trigger in a pre-symptomatic phase, or a spurious association produced by mass vaccination. Each possibility shifted interpretive attention away from vaccination as a possible contributor to harm and toward either the patient’s presumed underlying biology, an undefined latent disease state, or background incidence.

The appeal to predisposition and pre-symptomatic disease is especially important because the case series did not demonstrate these conditions independently. The patients were diagnosed after vaccination, and the article did not establish a pre-vaccination clinical baseline capable of confirming that they were already in a latent or preclinical multiple sclerosis phase. Nor did it operationalize what “pre-symptomatic MS phase” means in a way that allowed this explanation to be distinguished from a vaccine-associated first clinical event. As will be shown in Nelson et al., susceptibility language functioned retrospectively: after harm had occurred, the patient was redescribed as biologically predisposed or already on the path to disease, even though this status was not demonstrated before vaccination.

The same pattern continued in the authors’ use of rarity and benefit language. They stated that “although these cases represent interesting findings, they remain relatively rare occurrences,” and then shifted directly to vaccination reassurance, arguing that “the overall benefits of Covid-19 vaccination…cannot be understated”. The reported cases were therefore acknowledged as clinically “interesting”, but their practical significance was contained through rarity and subordinated to a general benefit claim. The conclusion then stated that these findings “should not dissuade vaccine use in the general population or even in the MS population.” This final move is central to the inconsistency: the article reported five new multiple sclerosis diagnoses after mRNA vaccination, proposed a plausible immune-mediated mechanism, and acknowledged a temporal association, yet concluded by directing readers not to alter vaccine use.

The article was therefore classified as Inconsistent because the conclusion did not maintain alignment with the clinical and mechanistic evidence presented. The inconsistency lay in the cumulative interpretive effect of the authors’ qualifications: predisposition, pre-symptomatic disease, spurious association, rarity, generalized benefit, and reassurance against dissuading vaccination. Together, these moves neutralized the significance of the authors’ own evidence by relocating causal weight away from the vaccine and toward the patient, background incidence, or population-level benefit.

Example 4: Rheumatoid Arthritis with Chronic Eosinophilic Pneumonia

Morikawa et al. (2022) reported a case of rheumatoid arthritis disease activity with eosinophilic infiltration after Pfizer-BioNTech vaccination in an 88-year-old woman with a 20-year history of chronic eosinophilic pneumonia. Although she had elevated rheumatoid factor at the time of her earlier chronic eosinophilic pneumonia diagnosis, she had no reported joint symptoms or relapse for 16 years. Three days after receiving the Covid-19 vaccine, she developed fever, dyspnea, productive cough, malaise, and worsening joint symptoms. Laboratory testing showed elevated rheumatoid factor, anti-cyclic citrullinated peptide antibody, C-reactive protein, interleukin-6 (IL-6), and peripheral eosinophil count, while musculoskeletal ultrasonography confirmed active synovitis. The patient required methylprednisolone pulse therapy, after which respiratory and joint symptoms improved [35].

The authors proposed cytokine dysregulation as a plausible mechanism connecting vaccination with the patient’s rheumatoid and eosinophilic reactions. They explicitly distinguished the case from Covid-19 infection, stating: “In the late phase of Covid-19, the cytokine storm is the main cause of lung injury, and several immune factors, such as IL-6, JAK1/2, and GM-CSF, were reportedly common signaling cascades, like those in RA. These mediators have also been reported in Covid-19 patients from Wuhan, China. In our case, the immune response was caused by SARS-CoV-2 vaccination rather than infection. Previous clinical studies have not reported on cytokine release syndrome (CRS). However, the cytokine reactions against SARS-CoV-2 spike proteins were likely similar to those against Covid-19 infection”. The authors later added: “Therefore, in our case, BNT162b2 vaccination might have induced RA-related cytokine production and triggered RA disease initiation”.

The interpretive inconsistency appeared in the movement from this vaccine-specific mechanistic account to the conclusion. After stating that the immune response was caused by vaccination rather than infection, and after proposing that Pfizer-BioNTech vaccination may have induced rheumatoid arthritis-related cytokine production and triggered disease initiation, the authors concluded more diffusely: “Although the causality of the correlation between vaccination and the development of these reactions is not established, this reaction occurred transiently and was completely recovered. Both innate immune and Th cell adaptive reactions to Covid-19 might be a trigger of RA, subsequently inducing activation of disease activity of [chronic eosinophilia pneumonia]”. This wording partially redirected causal attention from the vaccine to Covid-19 immune reactions more generally, even though the reported case involved vaccination rather than documented infection.

This redirection matters because the article itself presented the clinical event as vaccine-proximate and biologically plausible. The patient developed worsening respiratory and joint symptoms shortly after vaccination, with laboratory and imaging evidence supporting both eosinophilic and rheumatoid inflammatory activity. The authors also described a plausible cytokine-mediated pathway and explicitly stated that the immune response was vaccine-mediated. Yet the conclusion weakened the interpretive force of this evidence by emphasizing that causality was not established, that the reaction was transient and completely recovered, and that Covid-19-related immune responses might trigger rheumatoid arthritis.

The article was therefore classified as Inconsistent because the conclusion did not fully align with the mechanistic and clinical evidence presented. The inconsistency lay in the shift from a vaccine-specific mechanistic explanation to a more generalized Covid-19 immune-reaction framing, combined with emphasis on transient recovery. Together, these moves softened the clinical significance of the authors’ own evidence that vaccination may have triggered rheumatoid arthritis disease initiation and activated chronic eosinophilic pneumonia disease activity. This clinical significance was not minor: the patient had a long history of stable chronic eosinophilic pneumonia, no prior joint symptoms, developed respiratory and rheumatoid symptoms after vaccination, and required methylprednisolone pulse therapy before improvement. The conclusion therefore narrowed the implications of a clinically serious vaccine-proximate event by emphasizing recovery rather than considering what the case might imply for risk assessment, monitoring, or subsequent vaccination decisions in similar patients.

Example 5: Systemic Lupus Erythematosus and Alopecia Areata

Gamonal et al. (2022) reported a case of new-onset systemic lupus erythematosus and alopecia areata after heterologous Covid-19 vaccination in a previously healthy 27-year-old woman with no family history of autoimmune disease. The patient developed bullous, exulcerated, and crusted lesions affecting the face, chest, arms, legs, and oral mucosa, accompanied by fever and fatigue, three weeks after receiving the second dose of the AstraZeneca vaccine. SARS-CoV-2 testing was negative. Subsequent clinical, serological, and histological findings supported a diagnosis of systemic lupus erythematosus. After receiving a Pfizer-BioNTech booster, she developed diffuse alopecia after 15 days, and biopsy findings confirmed alopecia areata [36].

The authors treated the temporal relationship as biologically meaningful. They wrote that the patient’s initial systemic lupus erythematosus presentation and hair loss occurred “immediately after SARS-CoV-2 vaccination, suggesting a pathophysiological association among them”. They also stated that “no drug is completely harmless” and that this principle “can also be applied to vaccines”, which “could be potential triggers for autoimmune diseases”. For alopecia areata specifically, they noted that “the close temporal context”, normalization of laboratory parameters related to systemic lupus erythematosus, and “absence of other trigger factors” supported “the possibility that the BNT162b2 vaccination caused AA in our case”, although they added that systemic lupus erythematosus might have played a synergistic role. The authors further cited cross-reactivity involving the SARS-CoV-2 spike protein as a plausible mechanism, while noting concerns about autoimmune responses following production of antibodies to SARS-CoV-2 spike glycoproteins.

The inconsistency appeared in the conclusion, where the clinical and mechanistic significance of the reported events was redirected into a general reassurance about vaccination. After acknowledging temporal association, absence of other triggers, and possible vaccine-triggered autoimmunity, the authors concluded that “the possibility of skin manifestations or worsening of auto-immune diseases should not discourage vaccination, as the benefits outweigh the risks”. This benefit-risk claim was not directly demonstrated by the case report itself; rather, it functioned as a general public-health reassurance after the article had presented two autoimmune outcomes temporally associated with vaccination in the same patient. The article was therefore classified as Inconsistent because the conclusion subordinated the case-specific evidence of vaccine-associated autoimmune disease to a generalized assertion that vaccination should not be discouraged, thereby neutralizing the clinical implications of the authors’ own reported findings.

Example 6: Systemic Lupus Erythematosus with Lupus Nephritis

Nelson et al. (2022) reported a case of new-onset systemic lupus erythematosus in a previously healthy 14-year-old male, occurring two days after his third dose of the Pfizer-BioNTech vaccine. The patient had no reported prior medical history or family history of autoimmune disease. Following vaccination, he developed a facial rash that did not respond to topical steroids and later experienced hair loss, chest pain, photophobia, and joint symptoms. Subsequent laboratory testing and clinical evaluation supported a diagnosis of systemic lupus erythematosus, and kidney biopsy confirmed class V lupus nephritis. The patient required treatment with hydroxychloroquine, prednisone, mycophenolate mofetil, and losartan [37].

The authors identified several biologically plausible mechanisms through which vaccination could have contributed to the onset of systemic lupus erythematosus. These included cytokine dysregulation through type I interferon, molecular mimicry involving the SARS-CoV-2 spike protein, and direct activation of B cells. Taken together, these mechanisms provided a plausible biological basis for interpreting the reported event as potentially vaccine-related.

The interpretive inconsistency appeared in how the authors moved from this clinical and mechanistic evidence to their conclusion. On the one hand, they acknowledged the possibility of vaccine-associated autoimmune harm, stating: “However, Covid-19 vaccination has been linked with rare autoimmune-mediated adverse events, and thus it's important to be aware of potential associations given the limited data on rare adverse events with these vaccinations”. On the other hand, the discussion immediately placed the case within a population-scale frame: “This association does not prove causality. Indeed, there have been billions of Covid-19 vaccine doses given worldwide so some medical events will inevitably occur after vaccination”. This formulation did more than acknowledge uncertainty; it made the reported event appear statistically expected, rather than clinically significant in its own right.

A similar minimizing effect appeared in the authors’ use of rarity and susceptibility. After presenting a severe autoimmune outcome in a previously healthy child and discussing plausible mechanisms, the authors stated that “SLE is clearly a rare complication and may only occur in a genetically susceptible patient”. The conclusion followed the same pattern: “Large epidemiologic studies are needed to assess whether this is more than an association, but it would clearly be a rare complication. It is possible that the vaccination led to SLE in a genetically susceptible individual”. This wording did not deny a possible vaccine role; indeed, it explicitly allowed that vaccination may have led to systemic lupus erythematosus. However, that possibility was immediately qualified through rarity and presumed genetic susceptibility, even though such susceptibility was not demonstrated in the case itself. Nor was it specified how such susceptibility could have been identified before vaccination or distinguished from a vaccine-associated first clinical presentation after the fact.

The article was therefore classified as Inconsistent because the conclusion did not maintain alignment with the clinical and mechanistic significance of the evidence presented. The authors reported a serious new-onset autoimmune disorder in a previously healthy child, beginning two days after vaccination, with biopsy-confirmed lupus nephritis and the need for immunosuppressive treatment. They also identified plausible pathways through which vaccination could contribute to lupus onset. Yet the emphasis repeatedly shifted toward uncertainty, rarity, population-scale reassurance, and presumed susceptibility. The inconsistency lay in the use of these moves to soften the implications of the authors’ own evidence of a severe, temporally proximate, biologically plausible vaccine-associated autoimmune event.

Example 7: Type 1 Diabetes Mellitus in Previously Healthy Patient

Sakurai et al. (2022) reported a case of new-onset Type 1 diabetes mellitus in a previously healthy 36-year-old woman after the first dose of the Pfizer-BioNTech vaccine. The patient had no personal history of diabetes, allergy, or autoimmune disease, and no family history of autoimmune disease or diabetes. Three days after vaccination, she developed thirst, polydipsia, polyuria, palpitations, loss of appetite, and fatigue. Ten days after vaccination, she presented to hospital with diabetic ketoacidosis and was diagnosed with Type 1 diabetes. The authors reported marked hyperglycemia, metabolic acidosis, ketonuria, low C-peptide, and negative islet-related autoantibodies. The relatively low glycated hemoglobin level despite severe hyperglycemia suggested rapid onset [38].

The authors discussed several biologically plausible mechanisms through which vaccination could have contributed to Type 1 diabetes onset, including innate immune activation, molecular mimicry, and bystander activation. They also raised the possibility of fulminant Type 1 diabetes, noting that “The serum C-peptide level decreased to 0.13 ng/mL 14 days after admission, suggesting a possibility of fulminant Type 1 diabetes”. They later stated: “Another possible pathogenesis of the present case may therefore be fulminant Type 1 diabetes triggered by Covid-19 RNA-based vaccine”. Thus, the article did not merely report temporal association; it explicitly identified vaccine-triggered fulminant Type 1 diabetes as one possible pathogenesis.

The interpretive inconsistency appeared in the way this possibility was qualified and then only weakly translated into clinical implications. The authors first acknowledged the temporal relationship, but immediately introduced coincidence as an alternative explanation: “The hyperglycemic symptoms occurred 3 days after the first administration of mRNA Covid-19 vaccines in our case. Therefore, we cannot deny the possibility that the onset of Type 1 diabetes just coincided with the timing of the Covid-19 vaccination”.

As in other articles in this category, the problem was not the mere presence of a competing interpretation. Coincidence is a legitimate possibility in a single case report. The issue is that the article presented several features supporting causal plausibility—rapid onset, diabetic ketoacidosis, possible fulminant Type 1 diabetes, absence of prior diabetes or autoimmunity, and proposed mechanisms—yet the conclusion gave these features limited interpretive weight, stating that: “The present case suggests that Type 1 diabetes should be added to the list of the possible adverse effects of Covid-19 vaccination and should be surveyed carefully after Covid-19 vaccination, even in subjects without prior histories of diabetes”. This statement did acknowledge possible vaccine-related harm, but translated the finding into post-vaccination surveillance only. It did not engage the implications of a severe, acute, insulin-dependent autoimmune outcome occurring in a person without prior diabetes, nor did it consider whether the proposed mechanisms might warrant pre-vaccination risk assessment, heightened clinical caution, or reconsideration of vaccination practice in any subgroup.

The article was therefore classified as Inconsistent. The authors reported a serious new-onset autoimmune condition shortly after vaccination and discussed plausible mechanisms, including fulminant Type 1 diabetes triggered by Covid-19 RNA-based vaccination. Yet the interpretive conclusion was limited to adding Type 1 diabetes to a list of possible adverse effects and surveying carefully after vaccination. The inconsistency lay in the mismatch between the severity and mechanistic plausibility of the reported case and the narrowness of the clinical implication drawn from it. In this sense, the article did not dismiss vaccine-related harm outright, but weakened its significance through hedging and post-event surveillance rather than engaging the broader causal and clinical implications of its own evidence.

Example 8: New-Onset Type 1 Diabetes Mellitus, Mostly in Patients Without Prior Autoimmunity

Aydoğan et al. (2022) reported a case series of four patients diagnosed with new-onset Type 1 diabetes mellitus after Covid-19 vaccination. Three patients had no personal or family history of autoimmune disease, while one had a medical history of vitiligo and Hashimoto’s thyroiditis. All four had positive glutamate decarboxylase 65 (GAD65) antibodies, supporting autoimmune diabetes. Symptoms developed within two to eight weeks after vaccination. Clinical presentations included fatigue, dry mouth, polyuria, polydipsia, weight loss, blurred vision, vaginal candidiasis, hyperglycemia, and, in one case, diabetic ketoacidosis. Three patients experienced a rapid reduction or disappearance of insulin requirement during follow-up, while the fourth required ongoing but reduced insulin therapy. Although one patient had previously received CoronaVac, all four cases were reported by the authors as occurring after Pfizer-BioNTech vaccine exposure [13].

The authors provided a detailed discussion of possible vaccine-related autoimmune diabetes. They situated their cases within prior literature on Type 1 diabetes after other vaccines and after Covid-19 vaccination, and discussed ASIA as a framework for understanding vaccine-associated autoimmune phenomena. They also proposed several mechanisms through which vaccination could plausibly contribute to new-onset Type 1 diabetes. These included molecular mimicry, based on possible cross-reactivity between SARS-CoV-2 proteins and human target proteins including GAD65; ASIA and adjuvant-related immune stimulation; self-adjuvant features of mRNA; polyethylene glycol lipid conjugates; and immune stimulation in individuals with an “individual predisposition to autoimmunity”. The authors therefore did not merely report temporal association; they developed a substantive literature-based and mechanistic account of how Covid-19 vaccination could contribute to autoimmune islet injury.

The article was classified as Inconsistent because the conclusion did not maintain alignment with the breadth and force of the evidence the authors themselves presented. Its importance as an example lies precisely in the fact that vaccine-related harm was taken seriously in the body of the article. The authors reported four new-onset Type 1 diabetes cases, documented positive GAD65 antibodies in all patients, reviewed prior vaccine-associated autoimmune diabetes literature, proposed several mechanisms, and stated that Pfizer-BioNTech vaccination “may trigger Type 1 diabetes”. The use of “individual predisposition” reflected the broader pattern noted earlier: susceptibility was invoked as a possible explanation without being operationalized as a pre-vaccination risk marker. Moreover, although three of the four patients were reported to have no personal or family history of autoimmune disease, family history itself would not establish the specific predisposition invoked or clarify how it contributed to the post-vaccination autoimmune outcome.

The inconsistency appeared in the final interpretive move: after this extensive account, the conclusion stated only that “more data are needed” to clarify the role of Covid-19 vaccines in Type 1 diabetes and other autoimmune diseases, reaffirmed that the vaccination programme is the “cornerstone” of the fight against the pandemic, and translated the clinical implication into post-onset awareness that Type 1 diabetes may occur as a “rare phenomenon” of ASIA syndrome. Thus, the article’s own evidence was not denied, but its practical significance was narrowed. A case series of new-onset autoimmune diabetes, mostly in persons without prior autoimmunity, was ultimately converted into a recommendation for after-the-fact clinical evaluation rather than pre-vaccination caution, individualized risk assessment, or consideration of implications for subsequent doses.

4. Discussion

Phase 1 of the larger project found a recurrent disconnect between the reporting of biological mechanisms linking Covid-19 vaccination to autoimmune outcomes and the interpretation of those mechanisms in authors’ conclusions. Of the 109 articles included in the parent scoping review, 52 reported mechanisms of action and were therefore included in this analysis. In these articles, authors treated the mechanisms as biologically plausible and often drew on immunological pathways recognized in the literature on vaccine-associated autoimmunity, in addition to temporal patterns, clinical severity, prior literature, or other contextual evidence relevant to possible vaccine-associated autoimmune outcomes. Most conclusions, however, did not allow that evidence to bear proportionately on causal interpretation or clinical implications.

The dominant interpretive pattern, therefore, was not evidentiary absence—articles were included precisely because they contained mechanistic evidence—but interpretive attenuation. While mechanistic and clinical evidence was presented as relevant to vaccine-associated autoimmune outcomes, its causal and clinical significance was generally displaced, narrowed, or neutralized within the same article. Mechanisms were reported but not carried forward; susceptibility was invoked but rarely translated into protection; rarity was asserted, often before incidence was known; monitoring was recommended while prevention remained largely unavailable; and benefit language entered as a stabilizing background frame. The result was a literature in which autoimmune harms could be documented, mechanistically explained, and even causally entertained, while still being prevented from substantially altering the clinical or policy-facing conclusion.

This finding requires careful interpretation. Classifying an article as Inconsistent does not mean that case reports, case series, or observational studies can establish causality on their own, nor that every acknowledgement of uncertainty constitutes minimization. Rather, the issue addressed in this phase of the broader study was more specific: whether authors’ conclusions were internally aligned with the clinical and mechanistic evidence they themselves presented. In this sense, inconsistency refers to a mismatch between reported evidence—including, in most cases, explicit acknowledgement of a plausible causal role for vaccination—and conclusions that displaced, minimized, or neutralized the causal significance of that evidence.

The distinction between Ambiguous and Inconsistent articles is therefore central. Ambiguous articles left causal interpretation unresolved: authors reported mechanisms and temporal associations but did not assign clear causal weight to them. Inconsistent articles went further, presenting mechanisms and supportive contextual evidence while drawing conclusions that downplayed, displaced, or negated the causal implications of that evidence. The presence of Consistent and Ambiguous categories is important because it shows that the typology did not classify uncertainty itself as inconsistency but instead helped identify whether mechanisms were treated as relevant to causal reasoning, left unresolved, or neutralized in the conclusion.

Within the Inconsistent category, the neutralization of vaccine-associated harm occurred through a recurring cluster of interpretive mechanisms that can be grouped into five broad themes: epidemiological containment, post-event clinical narrowing, reassurance closure, deferral to future knowledge, and patient-centred redirection. The following paragraphs describe how each mechanism worked and how, in different ways, they limited the causal or clinical significance of vaccine-associated autoimmune harm while allowing authors to acknowledge biological plausibility.

First, epidemiological containment occurred when authors shifted from the clinical significance of the reported case to population-level frames such as rarity, background incidence, mass vaccination, or coincidence. Rarity often functioned less as a demonstrated epidemiological finding than as a stabilizing interpretive frame: events were described as rare even when authors also stated that further epidemiological research was needed to determine frequency, association, or under-identification. Similarly, appeals to millions or billions of administered doses changed the scale of analysis. At the individual level, the reported outcome could be disabling, serious, or life-altering; at the population level, it became a barely traceable event. This shift did not refute the clinical or mechanistic evidence presented in the article, but it made the harm appear less consequential by relocating it into a population-reassurance frame.

Second, post-event clinical narrowing occurred when authors translated reported harm into surveillance, awareness, monitoring, screening, early recognition, or treatment after vaccination. These recommendations were not necessarily meaningless; clinician awareness and early treatment may be clinically useful. However, in many Inconsistent articles, such recommendations were narrower than the evidence warranted. Severe or new-onset autoimmune disease, close temporal sequence, mechanistic plausibility, and recurrence after exposure were often translated into “be aware” rather than into pre-vaccination caution, individualized risk assessment, deferral, exemption, or reconsideration of subsequent doses. This created an after-the-fact clinical logic: vaccinate first, then monitor for injury. The mechanism appeared scientifically but disappeared practically.

Third, reassurance closure occurred when authors acknowledged plausible vaccine-associated harm but closed the interpretive space through benefit-over-risk statements, non-discouragement language, or claims that vaccination should continue unchanged because it is “well-established” that “Covid vaccines have saved millions of lives”. In these cases, the harm was not necessarily denied. It was recognized but immediately subordinated to a broader pro-vaccination frame. For our purpose, the issue was not whether a benefit claim was true in the general population sense, but rather how these rhetorical devices shaped the clinical and policy implications of the evidence presented. We argue that when broad benefit or non-discouragement language follows detailed harms evidence, it can prevent that evidence from generating stronger clinical or policy implications — readers are instructed to acknowledge adverse events and at the same time naturalize the pro-vaccination frame such that it can never be challenged.

Fourth, deferral to future knowledge occurred when authors called for further research, larger epidemiological studies, case accumulation, or mechanistic clarification in ways that suspended clinical implication. Calls for further research are appropriate, especially when causality is uncertain. The interpretive problem arises when future research becomes a way to defer action despite the article’s own presentation of temporality, plausible mechanisms, clinical severity, and similar cases. In such cases, patient harm may be converted into a research opportunity while the immediate clinical question—what should follow for patients facing subsequent exposure or similar risk—is left unanswered.

Fifth, patient-centred redirection occurred when explanatory weight shifted from the vaccine to the patient’s presumed vulnerability. This included appeals to genetic susceptibility, predisposition, family history, latent disease, occult pathology, pre-symptomatic disease, or an “overactive” immune system. The problem was not that susceptibility is biologically irrelevant, but that it was often invoked retrospectively, after harm occurred, without having been established, operationalized, or used prospectively to guide vaccination decisions. In such cases, the vaccine could remain a possible trigger, but the causal burden was redistributed toward the patient, who was redescribed as already susceptible, latent, or on the path to disease. This framing diluted the clinical implications of vaccine-triggered harm because it recast the event as the manifestation of a patient-specific vulnerability rather than as a safety signal requiring broader caution, risk stratification, or guidance about future exposure.

Because patient-centred redirection was especially common and theoretically consequential, the recurrent use of predisposition and susceptibility deserves particular attention. The issue is not simply that susceptibility was invoked without being demonstrated, but that biological susceptibility was often treated as if it were equivalent to genetic predisposition. This equivalence is not self-evident. As noted by Lewontin, for complex phenotypes appeals to “genetic susceptibility” are empirically empty unless the relevant developmental and environmental conditions are specified [39]. His distinction between the analysis of variance and the analysis of causes is useful here: a statistical association between genetic variation and phenotypic variation in a particular population does not explain the causes of a trait in an individual, nor does it establish a stable genetic tendency across environments. Similarly, heritability estimates do not measure how much genes cause a disease in a person, but only how much genetic variation statistically accounts for phenotypic variation in a specific population, under a specific range and sequence of environments [40]. This point is especially important for complex disorders, where vulnerability may emerge through developmental, intergenerational, or life-course exposures rather than through identifiable genetic variants [41].

In the articles analyzed here, however, “susceptibility”—often “genetic susceptibility”—was usually invoked retrospectively. After an autoimmune event occurred, the patient was redescribed as genetically susceptible, predisposed, or already in a latent or preclinical disease phase. Yet such susceptibility was rarely measured, established, operationalized, or used prospectively to guide vaccination decisions. In other words, “genetic susceptibility” was used as if it explained the observed autoimmune outcome when, in the articles analyzed here, it had not been demonstrated as a cause, a risk marker, or even a clinically identifiable pre-vaccination condition. In this context, susceptibility functioned less as an empirically established explanation than as an interpretive device: the vaccine remained a possible trigger, but only in individuals who were retrospectively recast as already vulnerable. This shifted causal weight away from the vaccine and toward the patient, without clarifying how such patients could have been identified, counselled, monitored, or protected before exposure.

Taken together, these five interpretive mechanisms did not operate simply by denying vaccine-associated autoimmune harm. Rather, they allowed harm to remain visible while limiting what could follow from it. Epidemiological containment relocated case-level injury into population reassurance; post-event clinical narrowing translated possible harm into surveillance after exposure; reassurance closure subordinated harm to broad benefit claims; deferral to future knowledge postponed clinical implication; and patient-centred redirection shifted causal weight toward presumed vulnerability. Across these moves, vaccine-associated autoimmune events could be acknowledged as biologically plausible while their implications for causal interpretation, clinical judgment, individualized risk assessment, or vaccination recommendations were narrowed or suspended.

The overall pattern raises questions not only about vaccine safety interpretation, but also about causal reasoning, pharmacovigilance, case-based evidence, and public health ethics. It therefore speaks to a gap across several adjacent literatures, rather than to a single established field of “interpretive consistency” in vaccine safety research.

The first relevant body of literature concerns causal inference and the role of mechanisms, clinical observation, and study design in medicine. Bradford Hill’s classic account of association and causation identified temporality, biological plausibility, coherence, analogy, dose-response, and other considerations as relevant to causal judgment, while explicitly cautioning against rigid rules of proof [42]. His reflections on controlled trials point in the same direction: controlled trials are valuable, but they are not the only route to medical knowledge, and statistical design should support, not replace, skilled clinical observation and careful reasoning from the evidence at hand [43]. Hill’s framework is especially relevant here because it does not reduce causal reasoning to a single method, criterion, or evidentiary hierarchy. The present study did not apply Bradford Hill criteria formally to each article, but it addressed a related question: whether authors allowed temporality, plausibility, coherence, clinical context, and mechanistic reasoning to bear on their own interpretive conclusions.

More recent philosophical and methodological work similarly emphasizes that causal claims in the health sciences are not supported by one kind of evidence alone. Russo and Williamson argue that health sciences infer causal relations from mixed evidence, including both mechanistic knowledge and probabilistic dependencies. Their point is not that there are separate mechanistic and probabilistic kinds of causality, but that different types of evidence—mechanistic and probabilistic, among others—support a single causal claim [44]. Clarke and colleagues develop a related argument for treating evidence of mechanisms alongside evidence of correlation, especially in assessing causal claims, interpreting interventions, and generalizing from populations to patients [45]. Howick, Glasziou, and Aronson add an important qualification: mechanistic reasoning can mislead when it is incomplete, speculative, or fails to account for complexity, but high-quality mechanistic reasoning can be useful when the links in the chain are explicit and supported [46].

This body of work clarifies what is at stake in the present analysis. Mechanisms do not establish causality by themselves, but neither should they be reported as biologically plausible and then treated as having little or no bearing on the causal or clinical conclusion. However, in articles classified as Inconsistent, and to a lesser extent in those classified as Ambiguous, authors often moved in the opposite direction: mechanisms were presented as biologically plausible and relevant to the reported autoimmune outcomes, sometimes in considerable detail, but their bearing on causal interpretation, clinical judgment, or practical recommendations was left unresolved, narrowed, or attenuated.

The second relevant body of literature concerns pharmacovigilance and adverse drug reaction assessment. Edwards and Aronson define adverse drug reactions in terms of harmful or unpleasant reactions related to medicinal products that may predict hazard from future administration and warrant prevention, treatment, dose alteration, or withdrawal [47]. They also distinguish adverse events from adverse effects, emphasizing that not every event occurring after an intervention can be attributed causally to it, but that timing, pattern recognition, investigations, dechallenge, and rechallenge, all contribute to causality assessment [47]. This framework provides an important contrast to the interpretive patterns found in the present analysis. Many articles acknowledged temporal proximity, clinical severity, immunological plausibility, treatment response, or absence of prior disease, yet often stopped short of considering whether the reported event might predict hazard from future exposure or warrant prevention, monitoring, treatment modification, or avoidance of future exposure.

This point is reinforced by pharmacovigilance definitions of signal detection. Edwards and Aronson define a signal as reported information on a possible causal relation between an adverse event and a drug when that relation is previously unknown or incompletely documented, noting that the number of reports required depends on seriousness and information quality [47]. In this light, the interpretive handling of case reports and case series becomes consequential. If plausible mechanisms and temporal patterns are repeatedly documented but then neutralized in conclusions, potential safety signals may be weakened at the interpretive stage even before formal epidemiologic assessment occurs. Pharmacovigilance depends not only on collecting adverse event reports, but also on keeping causal inquiry open when the clinical and mechanistic evidence warrants it.

The third relevant body of literature concerns the epistemic value of case reports and case series. Because the analytic corpus was dominated by case-based literature, critics may correctly note that such evidence cannot establish incidence, relative risk, or population-level causality. This limitation is real. However, it does not make case reports irrelevant to vaccine safety science. As Vandenbroucke argues, case reports and case series permit the discovery of new diseases, unexpected drug effects, and mechanisms, and remain important precisely because they are sensitive to novelty, noting that case reporting remains important for detecting adverse and beneficial drug effects and has prompted withdrawals of medicines from the market [48]. The present findings should be read in that context. The study goal was not to assess whether case reports demonstrated causality, but whether authors interpreted case-based evidence coherently and preserve its signal-generating value.

This point is also relevant to the use of “rare” as a qualifier. Rare events may be difficult to study epidemiologically, but this is precisely why case reports, mechanistic reasoning, and careful causal interpretation matter. A rare event can still be clinically serious, mechanistically informative, and ethically significant, particularly when exposure occurs at population scale. Rarity may provide epidemiologic context, but it does not resolve the interpretive question raised by an individual case or case series: whether the mechanism and clinical course presented by the authors were reflected coherently in their conclusions.

The fourth relevant body of literature concerns uncertainty and public health ethics. Jasanoff’s concept of “technologies of humility” is useful here because it resists both false certainty and endless deferral to more research. She argues that real-world problems are complex, that science provides only part of the picture, and that uncertainty, ignorance, and indeterminacy are always present. Her call for humility includes attention to ambiguity, vulnerability, the distribution of risks and benefits, and the limits of scientific knowledge [49]. Applied to the present findings, this suggests that uncertainty about whether vaccination contributed to autoimmune harm should not automatically be converted into reassurance that such harm is unlikely, clinically insignificant, or outweighed by presumed benefits. A humble interpretation of possible vaccine-associated autoimmune outcomes would neither overstate causality nor close down concern prematurely. It would ask what follows clinically and ethically from serious post-vaccination events accompanied by plausible mechanisms.

Kass’s framework for public health ethics offers another way to situate the implications, provided that claims about public benefit are treated as claims requiring evidence rather than as assumptions. She argues that public health interventions should reduce morbidity or mortality, that data must substantiate claims that a programme will achieve its goals, that burdens must be identified and minimized, and that benefits and burdens must be balanced fairly [50]. In such interventions, the ethical question is not only whether benefits are expected or asserted at the population level, but whether those benefits are demonstrated with evidence adequate to the intervention, population, and outcome under consideration. It is also necessary to ask how burdens are identified, minimized, distributed, communicated, and made subject to meaningful informed consent. The findings of this study suggest that autoimmune harms were formally acknowledged in the literature, yet their practical implications were often narrowed to awareness, surveillance, future research, or reassurance.

This concern is especially important when medical interventions are administered to very large populations, including clinically heterogeneous groups and, most importantly, healthy individuals who may never experience the negative outcome that the intervention is designed to protect them from. Therefore, even when an adverse outcome is uncommon, its interpretation matters. A benefit-risk argument cannot be assumed in advance as the frame through which all harms are interpreted; it must be supported by evidence relevant to the populations and outcomes under consideration. Nor does an asserted population benefit, by itself, resolve the ethical question raised by exposing individuals to unwanted medical intervention, particularly when the intervention may carry risks for some recipients. Otherwise, the conclusion risks placing the cart before the horse: possible harms must meet high evidentiary thresholds before they are treated as clinically meaningful, while benefits are treated as background assumptions requiring less case-specific support. This asymmetry is not the primary object of Phase 1, but it is clearly foreshadowed by the findings and provides the conceptual bridge to Phase 2.

Interpretive consistency is therefore a necessary foundation for epistemic integrity. Before assessing whether authors apply symmetrical standards to claims of vaccine-related harms and benefits, it is first necessary to determine whether they apply coherent standards within the harm analysis itself. Phase 1 suggests that this first condition was frequently not met. Mechanisms were reported, but their causal and clinical implications were often attenuated. Phase 2 will examine whether similar asymmetries appear more explicitly in how harm and benefit claims are evaluated within the same literature.

Limitations and Strengths

This study has several limitations. First, it did not assess the biomedical validity of each proposed mechanism and did not determine whether the reported autoimmune outcomes were caused by Covid-19 vaccination. However, establishing biomedical validity was not the purpose of this analysis, which was interpretive rather than adjudicative: to examine whether authors’ conclusions aligned with the mechanisms, clinical observations, and causal language they themselves presented, not whether those mechanisms were sufficient to establish causality. By suspending the question of ultimate biomedical causality, the analysis focused on a prior and more basic problem in vaccine safety research: whether authors interpreted their own evidence consistently. In this sense, the study does not claim that the reported mechanisms demonstrate vaccine causation; it indicated instead that mechanisms considered plausible enough to report were not consistently allowed to bear on the authors’ conclusions.

Second, the study did not quantify risk, estimate incidence, or compare vaccinated and unvaccinated populations. Many of the included articles were case reports or case series, and such designs cannot establish population-level causal effects. However, this limitation should not be confused with evidentiary irrelevance. While the findings cannot be read as estimates of the frequency of vaccine-associated autoimmune outcomes, they should be read as a cautionary observation about the tendency to discount case-based evidence even when it contains temporality, clinical severity, and biologically plausible mechanisms. As the preceding discussion noted, case reports and case series cannot substitute for population-level studies, but they remain essential for early signal detection and for keeping causal inquiry open when the clinical and mechanistic evidence warrants it.

Third, the analysis was limited to articles from the parent scoping review that explicitly reported mechanisms of action. Articles that reported autoimmune outcomes without proposing mechanisms were excluded. However, this restriction reflects the logic of the study: the purpose was not to estimate how often autoimmune outcomes were reported in the broader vaccine safety literature, but to examine how authors handled mechanisms when they themselves treated those mechanisms as relevant enough to discuss. The findings therefore speak to interpretive practice within mechanistically engaged literature, not to the frequency of autoimmune outcomes across the full universe of Covid-19 vaccine safety research. This is an analytic rather than frequency-based claim: once a recurring interpretive pattern is identified within the relevant corpus, its significance lies in the logic of the pattern, not in the need to exhaust every possible instance.

Fourth, the classification process necessarily involved judgment. Although the typology was structured, applied systematically, and designed to distinguish uncertainty from inconsistency, some cases remained borderline. For example, articles that acknowledged possible vaccine-related harm but translated it only into post-vaccination surveillance required careful interpretation. However, judgment is not a limitation unique to qualitative research. All research depends on classificatory decisions: what to count, how to define variables, where to set thresholds, which observations to include or exclude, and how to distinguish signal from noise. As Bowker and Star argue, classification systems are not neutral containers for reality but infrastructures that shape what becomes visible, comparable, and actionable [51]. The relevant question is therefore not whether judgment was involved — it always is — but whether the criteria were explicit, consistently applied, and analytically appropriate to the research question.

Finally, because the study relies on published articles, it cannot assess unpublished adverse events, editorial decisions, peer-review dynamics, institutional pressures, or broader political conditions that may have shaped what was reported and how. This is an important limitation, because published articles are not transparent windows into scientific practice; they are the end product of research, writing, review, revision, editorial judgment, and publication norms. However, these questions fall outside the scope of Phase 1, which focused on the internal alignment between reported mechanisms and authors’ conclusions. They are more directly relevant to Phase 2, which will examine epistemic integrity in the same literature, including whether evidentiary standards are applied symmetrically to claims about vaccine-related harms and benefits.

These limitations notwithstanding, the study has several strengths. First, it builds on a defined corpus of 109 peer-reviewed articles identified in a completed scoping review, rather than on an ad hoc selection of examples. This strengthens the transparency and coherence of the analytic foundation.

Second, it focuses on an underexamined step in scientific reasoning: the movement from reported mechanisms and clinical observations to authors’ interpretive conclusions. Existing literature addresses adverse event reporting, pharmacovigilance, mechanisms, case reports, and causal inference, but to our knowledge, relatively little attention has been paid to how mechanisms are incorporated—or not—into published conclusions about vaccine-related harms.

Third, the study uses a conservative analytic frame. It does not claim that mechanisms demonstrate causality, nor does it treat caution as inconsistency. By distinguishing Consistent, Ambiguous, and Inconsistent interpretation, it allows for uncertainty while identifying cases where uncertainty becomes attenuation, displacement, or neutralization.

Fourth, the analysis is clinically relevant. It shows that interpretive practices may affect how possible safety signals are perceived, discursively treated, and translated into meaningful clinical implications. This is especially important in the context of autoimmune outcomes, where mechanisms may be complex, onset may vary, and affected individuals may not have been identifiable as high risk before vaccination.

Finally, although the empirical focus of this study is Covid-19 vaccination and autoimmune disorders, its implications extend beyond this specific corpus. The analysis identifies a broader problem in medical and public health research: how evidence of possible harm is interpreted, qualified, and translated into clinical or policy conclusions. If medical research is to inform practice and policy responsibly, it must not only collect empirical evidence, but also interpret that evidence consistently, assess harms and benefits with comparable standards, and engage openly with the ethical implications of uncertainty. In this sense, the study contributes not only to vaccine safety research, but also to broader debates about causal reasoning, evidence interpretation, and epistemic integrity in medicine and public health.

5. Conclusions

Our analysis showed that biological plausibility was often present in the literature on Covid-19 vaccination and autoimmune disorders but was frequently interpretively attenuated. Most articles that reported mechanisms of action were classified as Inconsistent because their conclusions did not maintain alignment with the mechanistic and clinical evidence presented. These findings establish the empirical foundation for Phase 2, which will examine whether similar asymmetries appear in the evidentiary standards applied to vaccine-related harms and benefits. More broadly, the findings suggest that rigorous vaccine safety science requires not only the documentation of adverse events and mechanisms, but also a critical appraisal of interpretive practices that allow such evidence to bear appropriately on causal reasoning, clinical judgment, and empirically and ethically informed decision-making.

Although grounded in the Covid-19 vaccine–autoimmunity literature, this analysis speaks to a broader problem in medical research: evidence does not interpret itself. The movement from observation to causal conclusion requires judgment, and that judgment must be applied consistently if clinical and public health recommendations are to be empirically and ethically defensible. A literature that records mechanisms of harm but repeatedly limits their causal significance does not lack evidence; it lacks a consistent way of allowing that evidence to matter.