shield_lock
JinDaGe
← Back to skills
extension
Category: Productivity & OfficeNo API key required

sr-patient-reported-outcomes

Handle patient-reported outcomes (PROs) in systematic reviews per Cochrane Handbook Ch 18. Use when included studies use patient-reported outcomes like quality of life, pain, function, or symptom scales. Includes COSMIN measurement property assessment, MID interpretation, and PRO selective reporting detection. Trigger on "patient-reported outcome", "PRO", "quality of life", "QoL", "COSMIN", "minimal important difference", "MID", "responder definition", "PROM", "EQ-5D", "SF-36", "patient-reported", "PRO selective reporting". Use ONLY when the review includes patient-reported outcomes.

personAuthor: TashanworldhubOpenAPI
open_in_newRepositorydownloadDownload package

Patient-Reported Outcomes (Cochrane Handbook Ch 18)

Your Role

Guide extraction, assessment, and interpretation of patient-reported outcomes in SRs. Assess measurement properties via COSMIN, interpret findings using MIDs, and detect selective PRO reporting.

Prerequisites

  • Included studies with PROs
  • Extraction data (from sr-extraction Block D)

Workflow

Step 1: Identify and Classify PROs

For each study, identify:

  • Which PRO instrument was used (e.g., EQ-5D-5L, SF-36 v2, PROMIS-29, VAS pain)
  • Who reported it (patient, clinician, proxy)
  • What construct it measures (QoL, pain, function, anxiety, satisfaction)
  • What recall period was used (24h, 7 days, 4 weeks)
  • Is the instrument validated in the study population?

DOCUMENT in structured table.

Step 2: Assess Measurement Properties via COSMIN

For each PRO instrument used, assess:

| Domain | What to check | |--------|---------------| | Content validity | Was the PRO developed with patient input? Does it cover relevant domains? | | Internal consistency | Cronbach's alpha ≥0.70? | | Test-retest reliability | ICC ≥0.70? | | Measurement error | SDC (smallest detectable change) reported? | | Construct validity | Hypotheses confirmed? Correlations with similar/different instruments as expected? | | Responsiveness | Can the PRO detect change over time? |

RUN: python3 scripts/cosmin_rob.py --instrument EQ-5D-5L --population sepsis

JUDGMENT: "The PRO instrument has [very good / adequate / doubtful / inadequate] measurement properties for [domain] in this population."

Do NOT pool PROs from instruments with inadequate measurement properties.

Step 3: Check MID (Minimal Important Difference)

For each PRO, identify the MID:

  • From literature (published MIDs for EQ-5D, SF-36, PROMIS)
  • From study (if study reports anchor-based MID)
  • Distribution-based estimate (0.5 × SD, 1 × SEM)

RUN: python3 scripts/mid_calculator.py --instrument EQ-5D-5L --method published

COMPARE:

  • If MD ≥ MID: "The effect exceeds the MID and is likely clinically meaningful."
  • If MD < MID: "The effect does not reach the MID, so its clinical importance is uncertain."
  • If MD ≥ MID but CI includes values < MID: "The effect may be clinically meaningful on average, but the CI includes values below the MID threshold."

Step 4: Detect Selective PRO Reporting

CHECK for each PRO:

  1. Was the PRO pre-specified in the trial registration?
  2. Was the PRO pre-specified in the study protocol?
  3. Are all pre-specified PROs reported?
  4. Are there any PROs reported that were NOT pre-specified?

FLAG: "The [instrument] was not pre-specified in the trial registration but was reported as a primary outcome — potential selective reporting."

Step 5: Synthesize PROs Separately

REPORT PRO outcomes SEPARATELY from clinician-reported or laboratory outcomes.

POOL only if:

  • Same instrument AND same recall period AND same metric
  • OR different instruments measuring same construct (use SMD)

INTERPRET: "A significant improvement in [PRO] was observed, but [it does/does not] reach the MID."

Scripts

scripts/cosmin_rob.py

COSMIN Risk of Bias tool (Mokkink 2018) — 9 boxes covering content validity, internal structure, reliability, measurement error, construct validity, and responsiveness. Returns per-domain ratings. Usage: python3 cosmin_rob.py --instrument EQ-5D-5L --population sepsis

scripts/mid_calculator.py

Calculates or retrieves MIDs for common PROs. Supports published-MID lookup, distribution-based (0.5SD, 1SEM), and anchor-based. Usage: python3 mid_calculator.py --instrument EQ-5D-5L --method published

scripts/pro_selective_reporting.py

Compares registered PROs against reported PROs. Fetches ClinicalTrials.gov via E-utilities. Usage: python3 pro_selective_reporting.py --nct NCT01234567

Assets

assets/pro-selection-guide.md

Prioritization rules for selecting which PRO to extract when multiple are reported.

assets/cosmin-checklist.md

Full COSMIN Risk of Bias checklist (9 domains, 3-5 items per domain).

assets/pro-glossary-em.md

Common PROs in EM/CC: EQ-5D-5L, SF-36, PROMIS-29/57, VAS, NRS, FACIT-F, HADS, IES-R. Included constructs, recall periods, MIDs, and validation populations.

Guardrails

  1. "Do NOT pool PROs from instruments with inadequate measurement properties."
  2. "PROs are at HIGH risk of selective reporting — always check protocol vs published."
  3. "Do NOT interpret a statistically significant PRO change as clinically meaningful without MID context."
  4. "Recall period matters — pain recalled over 24h is NOT the same as pain right now."
  5. "Patient-reported ≠ patient-important — some PROs may not matter to patients."
  6. "Separate PRO synthesis from clinical/lab outcomes — they measure different constructs."

Handoff

→ sr-extraction (PRO-tailored extraction), → sr-interpretation (MID-based interpretation), → sr-rob-update (selective PRO reporting)