Evidence Coverage Index Methodology

Method 2.1.0

The method is designed to make every displayed count inspectable and versioned, preserve missingness, and prevent registry activity from being mistaken for treatment effectiveness.

1 · Scope

A registry-transparency index, not an evidence grade

Version 2.1.0 covers acne vulgaris, atopic dermatitis, psoriasis, rosacea and melasma. Its source is the ClinicalTrials.gov API. Only records identified by the source as interventional can enter a published condition set.

The index does not appraise study quality, calculate effect sizes, determine whether an intervention works, or match registry records to every journal publication. Those are separate research questions.

2 · Condition selection

Broad discovery followed by a visible condition screen

The official condition search retrieves broad matches from the registry.
Records not identified as interventional are removed from the included set.
The submitted Condition field is checked for a direct target term or named synonym.
For acne, record-wide exclusions apply when the submitted Condition field or brief title identifies scar treatment, post-acne pigmentation or erythema, or another named acne-labelled condition outside acne vulgaris. A generic acne label cannot override one of those exclusions.
Atopic dermatitis requires an explicit atopic dermatitis or atopic eczema label. Generic hand, contact, nummular and other non-atopic eczema labels are not included.
Psoriasis requires a direct cutaneous psoriasis term or named plaque synonym. Parapsoriasis and arthritis- or arthropathy-only labels are not included.
Melasma includes the standard synonym chloasma.

Every condition page shows the broad source count, interventional source matches, final included count and number not included by the condition check. The current version does not describe excluded records as clinically irrelevant; it describes them as outside the published scope.

The full JSON download includes automated record-level membership manifests for included records and records not included after the condition check. These manifests show the output of the versioned rule; they do not provide human-reviewed inclusion or exclusion reasons.

3 · Treatment research views

Two labels answer two different questions

Treatment summaries use included studies whose registered primary purpose is TREATMENT. The index publishes two parallel groupings because a standardized subject heading and a submitted intervention name are not interchangeable.

Standardized treatment terms

These groups use the U.S. National Library of Medicine’s MeSH terms in the ClinicalTrials.gov derived intervention browse module. A study is counted once under each MeSH term assigned to it and may appear under more than one term. Groups use the MeSH identifier when available, with a normalized term-label fallback.

The following non-specific browse terms are excluded because they describe a dosage form or control rather than a distinct treatment: capsules, creams, gels, ointments, placebos, saline solution, solutions, standard of care, tablets and vehicles. A study without a remaining MeSH treatment term is retained in the condition set and counted as lacking a standardized treatment term.

Registered intervention labels

These groups preserve the intervention type and exact name submitted to ClinicalTrials.gov. For matching only, Unicode and repeated whitespace are normalized and case is ignored. Other submitted names are displayed as aliases but do not merge groups. Consequently, brand names, generic names, doses and formulations may remain separate labels even when they refer to related products.

A control is excluded when its submitted name contains placebo, vehicle, sham, no intervention, or a control cream, gel, lotion, product or treatment. An anchored description rule is used only when the submitted name is empty or generic, so an active treatment is not excluded merely because its description mentions a placebo comparator. This is a deterministic text exclusion, not a clinical judgement. Within one label, an NCT record is counted once even if the label appears in multiple arms or intervention entries.

“Industry lead sponsor” means only that the registry classifies the study’s lead sponsor as INDUSTRY. It does not capture every collaborator, funder, author relationship or conflict of interest, and it is not used as a quality or bias score.

Complete coverage and detailed-summary limits

The total term and intervention-label counts cover every group found by these rules. The full JSON membership manifests also retain every label-to-study-ID mapping. To keep generation and browser payloads bounded, participant, outcome, momentum and other detailed summaries are materialized only for the 60 highest-volume standardized terms and 120 highest-volume registered intervention labels in each condition.

Groups are ranked first by completed studies with posted results, then by completed studies, then by all included studies, with the label used as the final tie-breaker. A label outside the detailed limit is therefore still included in the full count and audit manifest, but it does not have a treatment-summary object. The cap must not be interpreted as evidence that lower-volume labels are absent or unimportant.

4 · Calculations

Definitions used in the published figures

Displayed measure

Operational definition

Important limit

Completed

Overall registry status equals COMPLETED.

Does not establish quality or successful results.

Posted results

The root-level hasResults field is true.

Means registry summary results—not peer review or journal publication.

Enrollment

Registry enrollment count, kept separate as ACTUAL or ESTIMATED.

Counts are summed across records and are not guaranteed to represent unique people.

Age and sex or gender reported

A completed study with posted results contains an explicit age, sex or gender baseline measure.

Presence does not describe the balance of enrolled participants; protocol eligibility is a separate field.

Race and/or ethnicity reported

A completed study with posted results contains an explicit race or ethnicity baseline measure, excluding labels that say the field was not collected or available.

Presence does not establish adequate representation.

Skin phototype reported

A completed study with posted results contains a Fitzpatrick, phototype, melanin-index, colorimetry or related baseline term.

The registry has no dedicated phototype field; this is a versioned text rule.

Country count

Unique countries appearing in listed study locations.

Does not establish participant nationality, residence, ancestry or demographic composition.

Participant categories

Counts from completed studies with posted results are aggregated only when a matching baseline measure is a number or count of participants. An explicit total group is used when available; otherwise reported study groups are summed. Sex or gender can match a measure title or description. Race-only and ethnicity-only titles are kept separate; a title containing both is retained in a third “combined race and ethnicity” field and is never split into cleaner categories. Skin-phototype fields use the published Fitzpatrick, skin-tone, melanin and colour measurement terms. Common category spellings are normalized while unfamiliar submitted labels are preserved.

“Participant entries” is the sum of reported category counts, not a deduplicated number of people. Multiple matching submitted measures can contribute and are not deduplicated. Categories, measures and studies may overlap or may not be directly comparable.

Participant age

Mean age uses baseline measures titled “Age, Continuous,” reported as a mean in years or months. Months are converted to years, each group mean is weighted by its participant denominator, preferring a denominator submitted at the same result-table class before the module-level denominator. The combined value is rounded to one decimal. The displayed range uses only measures explicitly marked as a full range and reports the lowest lower limit and highest upper limit.

Studies without a usable mean, participant denominator or full range are excluded from that calculation and disclosed in the contributing-study counts.

Patient-relevant outcome domains

Outcome titles from completed studies with posted results are classified by a versioned text rule into visible signs or severity; symptoms; quality of life or daily impact; pigment, redness or skin-colour change; and harms or tolerability. A study may count once in each matching domain; if none match, it counts as “other.”

This shows which kinds of outcomes were named, not the result, effect size, importance to each patient or quality of the outcome measure.

Mapped patient improvement result

For each materialized treatment, up to 20 result records are selected first by whether the submitted intervention can be linked to a non-control, non-combination arm, then by whether outcome titles resemble a patient efficacy outcome, and finally by source update date. A deterministic parser looks first for a primary efficacy outcome, then a secondary outcome. It requires an explicit treatment-to-arm link, one numeric response or change value per group, an unambiguous time point and, for comparative language, a usable comparator. Up to six records are retained, placing safely mapped patient results first. Skinlogie calculates only the absolute difference between the submitted group values.

Safety, pharmacokinetic, biomarker, multi-category, combination-only, control-arm and ambiguously mapped outcomes are withheld. Single-arm change is labelled non-causal. Results are never pooled across records, and no threshold of clinical importance is inferred.

Six-year research momentum

Study starts are counted for the six complete calendar years before the source year. Recency-weighted points use weights 0 to 5 from oldest to newest. Direction compares the first three years with the most recent three.

Fewer than four starts is “insufficient.” “Increasing” requires at least two more recent starts and a ratio of at least 1.25; “decreasing” requires at least two fewer and a ratio no greater than 0.75. All other patterns are “similar.” This measures starts, not progress or success.

Registry research leaderboards

The condition page can order its materialized standardized terms by completed linked records with posted results, unique linked records, the six-year recency-weighted score, or the difference between recent and previous three-year starts. The growth ranking includes only terms classified as increasing: at least four starts across both periods, at least two more recent starts, and a recent count at least 1.25 times the previous count. Terms that do not qualify for the selected measure are counted in the disclosure but omitted from the ranked list. Equal values receive the same competition rank and are ordered alphabetically within that rank. The interface shows at most six terms from the first five ranks and reports any omitted ties; all summarized terms remain available in the treatment list below.

These are rankings only within the detailed term summaries, which are selected first by posted-result count, completed-record count, and linked-record volume. A standardized term may describe an active treatment, comparator, or combination component, and one record can contribute to several terms. The ranks do not compare effect, benefit, safety, certainty, quality, cost, or guideline preference.

These calculations describe registry activity, result visibility and reporting. A mapped result describes one submitted outcome, comparator and time point. It does not estimate a treatment-wide effect, rank products, assess certainty of evidence or grade study quality.

5 · Missingness

Zero and “not reported” are never treated as the same

A missing baseline race, ethnicity or skin-phototype field is recorded as not reported. It is not interpreted as evidence that a population was absent. Similarly, eligibility fields describe who could enroll; they do not describe who actually participated.

Separately submitted race, ethnicity, Fitzpatrick phototype and measured skin colour are stored as separate constructs. A combined race-and-ethnicity table remains explicitly combined rather than being reclassified. The index does not use one construct as a proxy for another.

6 · Automated updates

Full reconciliation on each new source snapshot

ClinicalTrials.gov normally refreshes its API Monday through Friday. The Skinlogie Worker checks the official source version after that update window. If the source timestamp is unchanged, no new index snapshot is created.

When the timestamp changes, all five condition searches are fetched again, every page token is followed, deterministic metrics are recalculated, and the normalized snapshot is stored with its source timestamp, method version and SHA-256 content hash. The preceding snapshot remains available for change calculation.

Stored D1 payloads are gzip encoded to remain below Cloudflare’s per-row size limit and decoded before validation. Compression does not change the public JSON, content hash or calculation method.

Full reconciliation runs in the scheduled Worker task rather than inside a browser request. Reader requests continue to serve the last validated snapshot and never start a long-running source reconciliation.

The current API moves forward when a new source snapshot is published. A citation can instead pin an exact stored dataset with ?snapshot=<SHA-256>. A pinned request returns that content-addressed snapshot or a not-found response; it does not silently substitute the current data.

An unexpected ClinicalTrials.gov API version fails closed and requires parser review. Failed updates do not replace the last successful snapshot. An expiring database lease prevents overlapping scheduled or explicitly invoked maintenance reconciliations.

7 · Known limitations

What remains incomplete in version 2.1.0

Condition fields and other registry content are submitted by sponsors or investigators and may be incomplete or inconsistent.
The deterministic condition rule may omit a relevant record with an unusual label or include a mislabeled record.
Automated record-level membership manifests are published, but the index does not yet provide human-reviewed inclusion or exclusion reasons or PubMed publication matching.
Baseline reporting flags detect named fields; they do not appraise category quality, sample size or representativeness.
MeSH treatment terms are standardized browse terms, while registered intervention groups depend on exact submitted labels. Neither is a curated product ontology, and related interventions may be split or grouped differently from a clinical review.
Detailed treatment summaries are limited to 60 standardized terms and 120 registered intervention labels per condition. Complete group counts and label-to-study membership remain available, but labels below the materialization limit do not have participant, outcome or momentum summaries.
Patient-outcome domains are automated classifications of outcome titles. They do not inspect outcome values or determine whether a change was beneficial, harmful or meaningful.
Patient-result summaries cover only the compact candidate set retained in a materialized treatment object. They are automated, conservative extractions from submitted registry tables—not systematic reviews. Empty output means the parser could not make a safe mapping, not that a treatment had no effect.
A study can appear in more than one condition set, so condition totals should not be added together.

References

How the evidence coverage index is built.