Machine learning models automate classification of penicillin adverse drug reaction labels.

There is a growing interest in the appropriate evaluation of penicillin adverse drug reaction (ADR) labels. We have developed machine learning models for classifying penicillin ADR labels using free-text reaction descriptions, and here report external and practical validation. The models performed comparably with expert criteria for the categorisation of allergy or intolerance and identification of high-risk allergies. These models have practical applications in detecting individuals suitable for penicillin ADR evaluation. Implementation studies are required.

Immunoglobulin repertoire restriction characterizes the serological responses of patients with predominantly antibody deficiency.

BACKGROUND: Predominantly antibody deficiency (PAD) is the most common category of inborn errors of immunity and is underpinned by impaired generation of appropriate antibody diversity and quantity. In the clinic, responses are interrogated by assessment of vaccination responses, which is central to many PAD diagnoses. However, the composition of the generated antibody repertoire is concealed from traditional quantitative measures of serological responses. Leveraging modern mass spectrometry-based proteomics (MS-proteomics), it is possible to elaborate the molecular features of specific antibody repertoires, which may address current limitations of diagnostic vaccinology. OBJECTIVES: We sought to evaluate serum antibody responses in patients with PAD following vaccination with a neo-antigen (severe acute respiratory syndrome coronavirus-2 vaccination) using MS-proteomics. METHODS: Following severe acute respiratory syndrome coronavirus-2 vaccination, serological responses in individuals with PAD and healthy controls (HCs) were assessed by anti-S1 subunit ELISA and neutralization assays. Purified anti-S1 subunit IgG and IgM was profiled by MS-proteomics for IGHV subfamily usage and somatic hypermutation analysis. RESULTS: Twelve patients with PAD who were vaccine-responsive were recruited with 11 matched vaccinated HCs. Neutralization and end point anti-S1 titers were lower in PAD. All subjects with PAD demonstrated restricted anti-S1 IgG antibody repertoires, with usage of <5 IGHV subfamilies (median: 3; range 2-4), compared to ≥5 for the 11 HC subjects (P < .001). IGHV3-7 utilization was far less common in patients with PAD than in HCs (2 of 12 vs 10 of 11; P = .001). Amino acid substitutions due to somatic hypermutation per subfamily did not differ between groups. Anti-S1 IgM was present in 64% and 50% of HC and PAD cohorts, respectively, and did not differ significantly between HCs and patients with PAD. CONCLUSIONS: This study demonstrates the breadth of anti-S1 antibodies elicited by vaccination at the proteome level and identifies stereotypical restriction of IGHV utilization in the IgG repertoire in patients with PAD compared with HC subjects. Despite uniformly pauci-clonal antibody repertoires some patients with PAD generated potent serological responses, highlighting a possible limitation of traditional serological techniques. These findings suggest that IgG repertoire restriction is a key feature of antibody repertoires in PAD.

The choice of anti-LEDGF/DFS70 assay matters: a comparative study of six assays.

Lens-epithelial derived growth factor (LEDGF/DFS70) autoantibodies result in the commonly observed dense fine speckled (DFS) pattern by anti-nuclear antibody (ANA) assay. However, there is no consensus approach for confirmation of this autoantibody specificity. To evaluate current approaches, we examined inter-assay agreement between six anti-LEDGF/DFS70 assays. A total of 395 consecutive sera samples from routine ANA diagnostics were obtained, tested by routine ANA, anti-ENA line immunoblot assay (LIA) and anti-dsDNA assay and with six anti-DFS/LEDGF assays: the EuroLine-LIA (Euro-LIA), Medical and Biological Laboratories ELISA (MBL-ELISA), Phadia-EliA (EliA), QUANTA Flash CLIA, EuroImmun ELISA (Euro-ELISA) and Immco-Diagnostics HEp-2 ELITE/DFS-Knockout (HEp-2KO). Of 395 sera, 108 tested positive by at least one assay. Despite general good concordance between all assays across the cohort (Gwet's AC1=0.89), within the target DFS-ANA pattern group inter-assay agreement was poor (AC1=0.59). Euro-LIA, CLIA and MBL-ELISA assays were most concordant, but CLIA and Euro-LIA were also most likely to identify discordant positive results. EliA and Euro-ELISA had poorer agreement, which could be attributable to ill-matched cut-offs between assays. HEp-2KO was frequently discordant with all other assays tested. Euro-LIA, CLIA and MBL-ELISA were most concordant at manufacturer's specifications and are suited for use in clinical laboratories. Modified assay thresholds are required to ensure comparative results for Euro-ELISA and EliA. HEp-2KO assay is frequently discordant with all other assays, making it less suited for routine diagnostics. The study highlights the importance of considering inter-assay variability when developing a diagnostic strategy for anti-LEDGF/DFS70 autoantibodies in clinical laboratories.

Automation of penicillin adverse drug reaction categorisation and risk stratification with machine learning natural language processing.

BACKGROUND: The penicillin adverse drug reaction (ADR) label is common in electronic health records (EHRs). However, there is significant misclassification between allergy and intolerance within the EHR and most patients can be delabelled after an immunologic assessment. Machine learning natural language processing may be able to assist with the categorisation and risk stratification of penicillin ADRs. OBJECTIVE: The aim of this study was to use text entered into an EHR to derive and evaluate machine learning models to classify penicillin ADRs and assess the risk of true allergy. METHODS: Machine learning natural language processing was applied to free-text penicillin ADR data extracted from a public health system EHR. The model was developed by training on labelled dataset. ADR entries were split into training and testing datasets and used to develop and test a variety of machine learning models. These were compared to categorisation with a simple algorithm using keyword search. RESULTS: The best performing model for the classification of penicillin ADRs as being consistent with allergy or intolerance was the artificial neural network (AUC 0.994, sensitivity 0.99, specificity 0.96). The artificial neural network also achieved the highest AUC in the classification of high- or low-risk of true allergy (AUC 0.988, sensitivity 0.99, specificity 0.99). All ADR labels were able to be classified using these machine learning models, whereas a small proportion were unclassifiable using the simple algorithm as they contained no keywords. CONCLUSION: Machine learning natural language processing performed similarly to expert criteria in classifying and risk stratifying penicillin ADRs labels. These models outperformed simpler algorithms in their ability to interpret free-text data contained in the EHR. The automated evaluation of penicillin ADR labels may allow real-time risk stratification to facilitate delabelling and improve the specificity of prescribing alerts.

Basophil reactivity to BNT162b2 is mediated by PEGylated lipid nanoparticles in patients with PEG allergy.

BACKGROUND: The mechanisms underpinning allergic reactions to the BNT162b2 (Pfizer) COVID-19 vaccine remain unknown, with polyethylene glycol (PEG) contained in the lipid nanoparticle suspected as being the cause. OBJECTIVE: Our aim was to evaluate the performance of skin testing and basophil activation testing to PEG, polysorbate 80, and the BNT162b2 (Pfizer) and AZD1222 (AstraZeneca) COVID-19 vaccines in patients with a history of PEG allergy. METHODS: Three known individuals with PEG allergy and 3 healthy controls were recruited and evaluated for hypersensitivity to the BNT162b2 and AZD1222 vaccines, and to related compounds by skin testing and basophil activation, as measured by CD63 upregulation using flow cytometry. RESULTS: We found that the BNT162b2 vaccine induced positive skin test results in patients with PEG allergy, whereas the result of traditional PEG skin testing was negative in 2 of 3 patients. One patient was found to be cosensitized to both the BNT162b2 and AZD1222 vaccines because of cross-reactive PEG and polysorbate allergy. The BNT162b2 vaccine, but not PEG alone, induced dose-dependent activation of all patients' basophils ex vivo. Similar basophil activation could be induced by PEGylated liposomal doxorubicin, suggesting that PEGylated lipids within nanoparticles, but not PEG in its native state, are able to efficiently induce degranulation. CONCLUSIONS: Our findings implicate PEG, as covalently modified and arranged on the vaccine lipid nanoparticle, as a potential trigger of anaphylaxis in response to BNT162b2, and highlight shortcomings of current skin testing protocols for allergy to PEGylated liposomal drugs.

Virtual Global Transplant Laboratory Standard Operating Protocol for Donor Alloantigen-specific Interferon-gamma ELISPOT Assay.

The quantification of frequency of IFN-γ-producing T cells responding to donor alloantigen using the IFN-γ enzyme linked immunosorbent spot (ELISPOT) holds potential for pretransplant and posttransplant immunological risk stratification. The effectiveness of this assay, and the ability to compare results generated by different studies, is dependent on the utilization of a standardized operating procedure (SOP). Key factors in assay standardization include the identification of primary and secondary antibody pairs, and the reading of the ELISPOT plate with a standardized automated algorithm. Here, we describe in detail, an SOP that should provide low coefficient of variation results. For multicenter trials, it is recommended that groups perform the ELISPOT assays locally but use a centralized ELISPOT reading facility, as this has been shown to be beneficial in reducing coefficient of variation between laboratories even when the SOP is strictly adhered to.

Peripheral natural killer cell and allo-stimulated T-cell function in kidney transplant recipients associate with cancer risk and immunosuppression-related complications.

Reducing immunosuppression has been proposed as a means of preventing cancer in kidney transplant recipients but this can precipitate graft rejection. Here we tested whether anti-tumor natural killer (NK) cell and allo-responsive T-cell function in kidney transplant recipients may predict cancer risk and define risk of rejection. NK cell function was measured by the release of lactate dehydrogenase and T-cell allo-response by interferon-γ quantification using a panel of reactive T-cell enzyme-linked immunospot (ELISPOT) in 56 kidney transplant recipients with current or past cancer and 26 kidney transplant recipients without cancer. NK function was significantly impaired and the allo-response was significantly lower in kidney transplant recipients with cancer. With prospective follow-up, kidney transplant recipients with poor NK cell function had a hazard ratio of 2.1 (95% confidence interval 0.97-5.00) for the combined end point of metastatic cancer, cancer-related death, or septic death. Kidney transplant recipients with low interferon-γ release were also more likely to reach this combined end point. Thus, posttransplant monitoring of allo-immunity and NK cell function is useful for assessing the risk of over immunosuppression for the development of malignancy and/or death from cancer or sepsis.