# Physio Study Appraisal

> Critically appraise a named or supplied physiotherapy, rehabilitation, or clinical study with design-appropriate methods. Use for a PDF, DOI/PMID, RCT, cohort, diagnostic/prognostic/measurement study, systematic review, meta-analysis, or guideline; and for sample, randomization, blinding, bias, statistics, effect size, clinical significance, MCID, harms, limitations, and applicability. Turkish triggers: çalışma değerlendirmesi, makale kritik analizi, kanıt kalitesi, PEDro, RoB 2, ROBINS-I, AMSTAR, QUADAS, AGREE. Do not confuse reporting checklists with risk-of-bias tools.

- Skill: `yigityildiz0/physio-study-appraisal` (Agent Skill, multi-file: 6 files)
- Install (CLI): `npx skillmds@latest add yigityildiz0/physio-study-appraisal`
- Raw SKILL.md: https://api.skillmd.com/api/skills/yigityildiz0/physio-study-appraisal/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Data & Analytics
- Author: yigityildiz0 (https://skillmd.com/u/yigityildiz0)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/yigityildiz0/physio-study-appraisal

---


# Physio Study Appraisal

## Mandatory first reads

Read [safety core](references/safety-core.md) and [design and appraisal map](references/design-and-appraisal.md). Appraising a study does not validate it for a specific patient. If the request contains urgent clinical features, stop the appraisal-to-treatment handoff and escalate first.

## Workflow

1. **Verify identity and status.** Confirm title, authors, year, journal, DOI/PMID, version, correction/retraction/expression of concern, registration, protocol, funding, and conflicts. If live verification is unavailable, label publication/correction/retraction/registration status **not live-verified**, appraise only the supplied packet, and list the exact checks still required; never infer current status from the PDF alone.
2. **Determine the actual design.** Infer design from methods—not the title or author label. Extract the causal/diagnostic/prognostic/measurement question and unit of allocation/analysis.
3. **Collect the full evidence packet.** Use full text, supplement, protocol, statistical analysis plan, registry, and linked reports when available. Label abstract-only, paywalled, missing-supplement, and offline-status limitations separately.
4. **Separate reporting from validity.** CONSORT, STROBE, PRISMA, STARD, TRIPOD, and RIGHT assess reporting. Apply a design-appropriate risk-of-bias or methodological tool separately.
5. **Audit the sample.** Source population, sampling, eligibility, representativeness, baseline imbalance, sample-size calculation, attrition, exclusions, and vulnerable populations omitted.
6. **Audit intervention/index and comparator.** Components, provider, standardization, study dose, adherence, fidelity, contamination, co-interventions, reference standard, and implementation context.
7. **Audit outcomes.** Prespecification, hierarchy, construct, instrument/version/language, validity, time point, assessor blinding, selective reporting, and multiplicity.
8. **Audit analysis.** Estimand, analysis population, missing-data assumptions, clustering/repeated measures, model assumptions, baseline adjustment, multiple testing, interaction tests, sensitivity analyses, and overfitting.
9. **Interpret effect and uncertainty.** Point estimate, 95% interval, absolute and relative effect, scale direction, clinical threshold, responder analysis, harms, burden, and precision. A p value is not a quality or benefit score.
10. **Apply the method tool transparently.** For every domain, show the source passage/location, rationale, judgment, and missing information. Do not make a confident judgment without the required material.
11. **Assess transportability.** State which people, setting, provider, dose, equipment, follow-up, and co-interventions match—and which do not.
12. **Give a bounded clinical implication.** Classify as supports, conditionally supports, does not change, or argues against practice. Do not prescribe from one paper alone.

## Clinical importance

When MIC/MCID is used, verify construct, instrument/version/language, population, baseline severity, follow-up, direction, anchor, method, and uncertainty. Distinguish MDC/SDC from MIC/MCID. A mean group difference is not an individual response rate.

## Harms

Check how adverse events were defined, solicited, adjudicated, timed, denominated, and reported; whether withdrawals were harm-related; and whether high-risk groups were excluded. **Not reported** is not **none occurred**.

## Output

1. One-sentence verdict: what the study shows and does not show
2. Study map
3. Main effects, uncertainty, clinical meaning, and harms
4. Design-appropriate domain judgments with evidence locations
5. Statistical audit
6. Reporting gaps versus validity threats
7. Applicability and excluded/high-risk groups
8. Limitations and missing information
9. Practice implication and required corroborating evidence
10. Publication status and direct sources

Never collapse reporting quality, risk of bias, certainty of a body of evidence, clinical importance, and applicability into one “scientific quality score.”