WeCureUs

WeCureUs Module Design Methodology

Purpose and scope

This document describes how WeCureUs designs its questionnaire modules, extracts structured data from contributed clinical records, links the two, and governs all output. It is written for researchers, clinicians, and institutions evaluating the platform's scientific basis.

Grounding in established clinical frameworks

Every module is designed against the standardized instruments and clinical standards used in MS research and care, and against the documented gaps those instruments leave. The frameworks consulted, and how each informed question design:

The principle is consistent: where an instrument produces a severity score, WeCureUs captures the shape, pattern, trigger, and consequence the score omits. The underlying symptom inventory finds that seven of the ten highest-priority effectively invisible MS symptoms remain uncaptured even when the full catalog of validated instruments is applied, and that two, the MS hug and barometric sensitivity, have no standard-instrument presence at all.

The nine-criteria research-quality framework

Every question is audited against nine criteria before it ships:

  1. Clarity. One construct, plain language, no double-barreled items.
  2. Option completeness. A full plausible range, including a genuine negative or not-applicable path wherever the question could not apply.
  3. Option distinctness. Options are mutually exclusive, with no two that mean the same thing.
  4. Presupposition. No question assumes a symptom before a prevalence anchor has established it.
  5. Priming. Nothing shown before the options anchors toward a particular answer.
  6. Construct validity. Wording and options measure the intended construct, not a neighbor (time versus effort, frequency versus severity).
  7. Longitudinal suitability. Re-asked questions use current-state or fixed-window wording so answers stay comparable.
  8. Response-type appropriateness. The response type fits the construct (single select, multi select, numeric integer, year, month-year, partial date, postal code, free text, boolean).
  9. Ordering. Within a domain, questions move general to specific without priming later items.

These nine criteria are not invented whole cloth; they synthesize several established traditions in health measurement and survey science. Clarity, the avoidance of double-barreled items, and response-type and recall appropriateness follow the United States Food and Drug Administration's Patient-Reported Outcome (PRO) guidance (2009) and the COSMIN (Consensus-based Standards for the selection of health Measurement INstruments) standards. Construct validity and longitudinal suitability draw on COSMIN's definitions of construct validity, reliability, and responsiveness, and on classical test theory. Option completeness and distinctness (a full, mutually exclusive, and exhaustive response set), the avoidance of presupposition and priming, and within-domain ordering follow long-established survey-methodology and questionnaire-design principles and the cognitive-interviewing tradition used to pretest items. No single framework governs all nine; they are a synthesis applied uniformly across every question. Item response theory, by contrast, informs scoring and scaling rather than the item-authoring criteria above, and is not invoked here.

Multi-item measurement

Several constructs are measured by more than one item by design. A single item captures one facet of an experience and carries the measurement error of one wording choice. Multiple items addressing the same construct from different angles, frequency, severity, time-of-day, trigger, and functional consequence, capture its full structure and yield more reliable aggregate estimates than any single item can. Deliberate mirror questions across modules also enable within-participant cross-module analysis.

Taxonomy extraction from contributed records and free-text responses

Structured extraction operates on contributed records today, with a second pathway for free-text module responses on the roadmap. Participants can contribute radiology reports, laboratory results, and ancestry summaries as pasted text. These are converted into structured fields through a versioned extraction pipeline against controlled vocabularies maintained per record type: lesion location and supportive imaging features for radiology, LOINC-coded analytes for laboratory results, and population and admixture components for ancestry. The free-text response fields within the modules are designed to feed that same controlled-vocabulary extraction process, so that a phrase written in a module answer and a phrase parsed from a contributed document would be normalized into the same structured fields. Bounded fields use a predefined controlled vocabulary. Open-ended fields use an inductive approach, in which contributed phrasings are collected and promoted into the vocabulary as patterns emerge. Taxonomies are versioned so results are reproducible and historical records can be re-processed as a vocabulary grows. Source documents are de-identified before storage. Every extracted field is typed, stored in the same schema as questionnaire responses, and inherits the same k-anonymity suppression at the output layer.

The intersectional dataset

The defining property of the dataset is that structured questionnaire responses, extracted taxonomy fields, and contributed clinical records are linked under the same anonymous participant identifier and are cross-queryable. A worked example: a participant answers questions about their history of Epstein-Barr virus and mononucleosis testing, and also contributes the actual EBV serology result from their own records. The dataset then holds, for that participant, both the self-reported recollection and the objective laboratory value. The relationship between self-report and clinical measurement, examined across a cohort, is itself a research signal that no existing MS dataset can produce, because none links patient-reported experience to contributed clinical records at scale.

Platform agility: One Shots and Twofers

WeCureUs can deploy a single question, a One Shot, or a question pair, a Twofer, to the community with no module overhead and roughly a thirty-second participant burden. A researcher with a hypothesis not yet ready for a full study can propose one; if it passes community value review, it deploys within days, and the response data indicates whether a signal is present before the cost of a formal study is incurred. The illustrative case is the Bjornevik 2022 Science paper establishing EBV as a near-necessary precondition for MS: a question about EBV testing history could have reached thousands of people with MS within a week of publication, where a committee-governed registry measures that turnaround in months or years. Researcher-proposed questions are explicitly welcome through this community value process.

Privacy architecture

All output is aggregate-only. The minimum cohort size is five participants: no query result is returned unless at least five participants match, and this k-anonymity threshold is enforced at the output layer before any result reaches a researcher. The database that holds individual responses and the system that produces research output are structurally separate and share no query path, so no individual record can be surfaced. Account contact information such as email and phone number is held in a separate system that is never joined to the research dataset; responses are linked only to a pseudonymous code. Extracted taxonomy fields are subject to the same suppression as questionnaire responses.

Response rates and denominators

Every result carries two counts: the number of participants who answered the question or contributed the record, and the size of the cohort-filtered population from which they are drawn. Reporting both separates two findings that a single count conflates: a small answered count can mean a small cohort, or it can mean a low response rate within a larger cohort, and those support different conclusions. The population count is the denominator for the response rate. When the group that did not answer is itself below the minimum cohort size, the exact population is withheld and only the answered count is shown, so the denominator can never be used to isolate the small non-answering group.

Laboratory result values

Beyond recording which laboratory tests a participant contributed, the platform makes the clinically meaningful results of the key MS-related tests queryable in aggregate. Qualitative results are canonicalized to a controlled positive, negative, or equivocal at the time the record is contributed, never held as free text: Epstein-Barr VCA IgG and EBNA IgG serostatus, CSF oligoclonal bands, and antinuclear antibody. Numeric markers are returned only as k-anonymous banded ranges, never as an exact value or a mean: the vitamin D level, the CSF IgG index, and the kappa/lambda free light chain ratio. Because these results are linked to the same participant identifier as the questionnaire responses and the other contributed records, the relationship between an objective laboratory result and self-reported experience, examined across a cohort, becomes a research signal in its own right.