Not asked questions — Immunology

Assays and endpoints that actually predict protection, safety signals you can trust, and analysis designs that survive heterogeneity.

Order 7 science-backed questions

Level-1 consensus (abstract)

Shared ground

Fit-for-purpose assays, clinically meaningful endpoints, adequate power, and vigilant safety monitoring are prerequisites for credible inference. Pre-specification and transparent reporting reduce bias.

Edges and limits

Population heterogeneity, prior infection, exposure gradients, and waning immunity complicate generalization. Variant escape, assay drift, and background rates can invert conclusions without robustness checks.

Why Immunology has not asked questions

Blind-spot generators

Single-assay proxies for protection, endpoints misaligned with decision stakes, unmeasured exposure risk, and safety reviews without observed-vs-expected context conceal failure modes.

What our service does

We co-create seven decision-ready questions per initiative, each bound to assumptions, evidence routes (neutralization tiers, T-cell panels, test-negative designs, MaxSPRT), and decision rules, then align clinical, lab, and safety stakeholders.

Which assay stack best predicts protection across variants?

Combine live/pseudovirus neutralization (ID₅₀/ID₈₀ standardized to WHO IU/mL) with binding IgG/IgA, functional ADCC/ADCP, and T-cell ELISpot/ICS panels (CD4/CD8 cytokine polyfunction). Calibrate inter-lab with reference sera; report limits of detection/quantitation and batch effects; model as a multivariate correlate rather than a single titer.

Entities: neutralization titer, binding titer, ADCC, ELISpot, ICS, WHO standard. Predicates: standardize, calibrate, harmonize, model. Constraints: biosafety level, panel cost, inter-lab drift.

How do we define correlates of protection under waning and imprinting?

Use time-varying Cox models linking titers and T-cell polyfunction to infection/severe endpoints with spline-based waning; include prior exposure as an effect modifier; report threshold bands (credible intervals) not single cutoffs.

Entities: correlate, hazard, spline, prior exposure. Predicates: stratify, interact, threshold, update. Constraints: assay noise, interval censoring, variant mix.

Which endpoints are decision-robust: infection, symptomatic disease, or severe outcomes?

Prioritize clinically anchored endpoints (hospitalization/ICU/death) for policy; use infection/symptomatic as surveillance diagnostics. Guard against misclassification with confirmatory testing windows and competing-risk handling.

Entities: endpoint, severity, competing risk. Predicates: anchor, confirm, reconcile. Constraints: testing cadence, healthcare-seeking bias.

How do we control confounding in real-world effectiveness?

Adopt test-negative or target-trial emulation with propensity scores/IPTW, high-granularity exposure proxies (occupation, household size), and negative-control outcomes/exposures. Validate with sensitivity to unmeasured confounding (E-value, tipping-point analysis).

Entities: propensity score, IPTW, target trial, negative control. Predicates: emulate, balance, stress-test. Constraints: data completeness, selection bias.

Where are heterogeneous treatment effects material?

Pre-specify moderators (age, immunocompromise, comorbidity, prior infection) and fit hierarchical models with partial pooling; report subgroup credible intervals and policy-relevant minima (worst-case protection).

Entities: moderator, partial pooling, credible interval. Predicates: prespecify, borrow strength, bound. Constraints: sparse strata, multiplicity.

Durability and booster timing — what curve governs waning?

Estimate joint antibody/T-cell decay with biexponential or mixed-effects models; link to protection via mechanistic mapping; trigger boosters when projected protection crosses a pre-declared floor for high-risk strata.

Entities: half-life, biexponential, protection floor. Predicates: project, trigger, stratify. Constraints: assay harmonization, variant drift.

What safety signal detection plan keeps risk honest?

Publish background incidence rates; use observed/expected with sensitivity bands, self-controlled case series for within-person control, and sequential monitoring (MaxSPRT/CuSum) with DSMB-defined stopping rules.

Entities: background rate, O/E, SCCS, MaxSPRT, DSMB. Predicates: calibrate, surveil, adjudicate. Constraints: latency, under-reporting, multiplicity.

How do we guarantee inter-lab comparability?

Run external quality assessment (EQA) panels, include calibrators in each batch, publish Bland–Altman bias and LoD/LoQ; version assay protocols and lock deviations via change control.

Entities: EQA, calibrator, LoD/LoQ, Bland–Altman. Predicates: harmonize, audit, lock. Constraints: reagent availability, lot variation.

Where does mucosal immunity change conclusions?

Measure secretory IgA and tissue-resident T-cells via standardized mucosal sampling; model mucosal/systemic joint effects for acquisition vs severity; report sample adequacy and site effects.

Entities: sIgA, TRM, lavage, swab. Predicates: sample, co-model, qualify. Constraints: protocol tolerability, assay sensitivity.

How do we analyze across hospitals without sharing raw PHI?

Use federated analysis with harmonized schemas (CDM), secure aggregation, and differential privacy on diagnostics; pre-register queries and publish audit trails for governance.

Entities: CDM, secure aggregation, DP. Predicates: federate, harmonize, attest. Constraints: jurisdiction, re-identification risk.