Chapter 13 · Research

From Reading Archive to Research Program

6 min read
In this chapterThe archive should expose uncertainty, not merely summarize consensusHighest-value unresolved problemsA decisive study design for candidate arteriolosclerosis criteriaEvidence milestones for the next releaseFinal research habit

The archive should expose uncertainty, not merely summarize consensus

A useful research knowledge base must help the reader discover where apparently compatible statements actually depend on different populations, measurements, or causal models. Its unit of work is therefore the claim, not the paper.

For each claim, this release records the supported interpretation, what is not established, evidence that supports or challenges it, a provisional appraisal, and the observation most likely to change confidence. This prevents review papers from silently accumulating authority and prevents negative evidence from disappearing.

Highest-value unresolved problems

Antemortem identification of arteriolosclerosis

Arteriolosclerosis lacks a broadly accepted in vivo diagnostic framework comparable to Boston criteria. ARTS is an important pathology-linked beginning, but transportability, calibration, fairness, lesion-level interpretation, and treatment sensitivity remain open. A next-generation model should combine continuous lesion location, diffusion, vascular physiology, clinical exposures, and acquisition metadata, while keeping a locked external pathology cohort untouched until the analysis plan is frozen.

Specificity of lobar microbleeds

The field needs lesion-level rather than only person-level validation. Each candidate CMB should be localized relative to cortical ribbon, juxtacortical white matter, sulci, veins, and vessels, then matched to serial histopathology. The decisive result is not merely another association between “lobar CMB count” and CAA grade; it is a map of which lesion morphologies and compartments arise from which vessel lesions.

Prehemorrhagic CAA

Criteria work best when hemorrhagic markers are already present. Preclinical diagnosis will require a different target: vessel dysfunction or molecular change before blood products appear. Longitudinal vascular reactivity, permeability, perfusion, diffusion, PVS, PET, and fluid measures should be evaluated for temporal ordering, not just cross-sectional group differences.

Mixed pathology and cognition

CAA, arteriolosclerosis, ADNC, LATE-NC, atherosclerosis, infarcts, and neurodegeneration commonly coexist. Standard regression adjustment can fail when variables are measured with different error, lie on causal pathways, or interact. Research should report total, direct, mediated, and interactive effects where assumptions permit, accompanied by causal diagrams and sensitivity analyses rather than a single “independent association.”

Pathology as an imperfect reference

Autopsy is the best available etiologic reference but is not error-free. Disease is patchy, tissue sampling is incomplete, routine grades vary, and MRI may precede death by years. Whole-hemisphere ex vivo MRI, spatially registered blocks, digital vessel quantification, and consensus regional sampling can reduce—but not eliminate—reference error.

A decisive study design for candidate arteriolosclerosis criteria

  1. Define the target condition and severity scale before examining imaging associations.
  2. Use multiregion, blinded pathology with explicit CAA, ADNC, LATE-NC, infarct, and atherosclerosis measurement.
  3. Predefine the eligible MRI-to-death interval and model interval continuously.
  4. Require and record sequence availability, field strength, resolution, and susceptibility technique.
  5. Rate images blinded to pathology and pathology blinded to images.
  6. Develop features only in a training cohort.
  7. Freeze preprocessing, features, thresholds, and missing-data handling.
  8. Evaluate discrimination, calibration, clinical utility, and subgroup performance in a locked external cohort.
  9. Perform lesion-level spatial validation for the features claiming etiologic specificity.
  10. Publish the full protocol, data dictionary, failure cases, container version, and negative analyses.

Evidence milestones for the next release

  • Complete full-text extraction for every essential source.
  • Dual-review QUADAS-3 appraisal of Boston and Edinburgh validation studies.
  • Dual-review PROBAST appraisal of ARTS development and validation reports.
  • Create at least 100 result-level evidence edges, with effect estimates and uncertainty when reported.
  • Add a contradiction ledger for every high-priority diagnostic and mechanistic claim.
  • Add prospective search alerts and retraction/correction checks.
  • Build the website only after the data schema survives the three pilot claim families: Boston transportability, lobar CMB specificity, and ARTS validation.

Final research habit

For every appealing hypothesis, write down the closest alternative explanation and the observation that would separate them. That habit—not the number of papers collected—is what turns the archive into a tool for connecting dots without inventing connections.