Prompt Guidances

Modular expert refinements injected into prompts at classification time. Global entries apply to all prompts; field-specific entries apply to that field's system prompt section and shaky prompt.

New Guidance Package
accuracy-improvements active
Initial guidances for the 5 highest-disagreement fields plus one global rule. Designed to push consensus rate above 95%.
2026-05-01 15:39
Target Guidance Text
intervention_digital Digital is 'yes' ONLY when digital technology IS the intervention or its primary delivery mechanism (telehealth therapy, mHealth app, web-based program, VR therapy, remote monitoring as intervention). Electronic data collection (e-questionnaires, REDCap, online surveys), phone follow-ups, and EHR documentation are 'no'.
intervention_education Education is 'yes' ONLY when it is a PRIMARY intervention arm or a structured named component. Brief medication counseling, standard discharge instructions, informed consent education, and routine information leaflets within another intervention are 'no'. Test: 'Would removing the educational component fundamentally change the intervention?'
intervention_palliative Palliative is 'yes' when the intervention IS palliative care or can only exist in a palliative context (dignity therapy, palliative consultations, hospice programs). A generic drug (paracetamol, gabapentin, an opioid) tested in palliative patients is 'no' because the drug itself is not palliative-specific. Ask: 'Could this intervention exist outside palliative care?' If yes → 'no'.
intervention_screening Screening is 'yes' ONLY when the study TESTS a screening approach as the intervention. Using validated instruments (PHQ-9, ESAS, HADS) to MEASURE outcomes is 'no' — that is outcome measurement, not a screening intervention. Eligibility screening for study enrollment is also 'no'.
pharmacologic_sub When intervention_pharmacologic is 'no', this field is 'N/A'. The values 'acupuncture' and 'music therapy' appear in the allowed list for legacy reasons but are NON-pharmacologic — prefer 'N/A' over those values when the intervention is non-pharmacologic. Use 'other' for approved standard drugs (paracetamol, NSAIDs, antiemetics) that don't fit a more specific category.
setting_place Hospital = inpatient or outpatient hospital care. A primary care clinic, GP office, or community health center physically located within a hospital campus is 'other' (the CARE MODEL determines classification, not the building). Multiple settings → choose the PRIMARY setting where most intervention delivery occurs.
global Intervention fields are INDEPENDENT — a study can have multiple intervention_* values set to 'yes' simultaneously. Do not assume mutual exclusivity. Apply each intervention rule independently.
Add entry
complex_intervention_rules active
Defines when an intervention counts as complex based on MRC framework. 10 disagreements found.
2026-04-22 14:09
Target Guidance Text
intervention_complex A complex intervention has MULTIPLE INTERACTING COMPONENTS that work together. Based on MRC framework: Classify as "yes" ONLY when: - The intervention explicitly combines 2+ different types (e.g., pharmacologic + psychological, education + behavioral + organizational) - The components interact with each other (not just co-administered) - The study describes the intervention as "complex", "multicomponent", or "multifaceted" Classify as "no" when: - A single intervention type is used, even if it has multiple sessions or steps - Pharmacologic intervention alone (even if it involves titration, monitoring) - A single behavioral/psychological intervention with multiple modules (e.g., CBT with 8 sessions) - Standard care comparison arms - Adding patient education to a drug trial does NOT automatically make it complex Key test: Could you describe the intervention as a SINGLE type? If yes → "no". Do the components fundamentally depend on each other? If no → "no".
Add entry
digital_intervention_rules active
Distinguishes digital health interventions from studies that merely use digital data collection. 5 disagreements.
2026-04-22 14:09
Target Guidance Text
intervention_digital A digital health intervention means the TECHNOLOGY IS the intervention or a core delivery mechanism. Classify as "yes": - App-based interventions (self-management apps, symptom tracking apps that provide feedback) - Telehealth/telemedicine as the core intervention model - Web-based therapy programs (e.g., online CBT) - Remote monitoring systems that trigger clinical actions - Virtual reality therapy - Digital decision support tools for patients Classify as "no": - Studies that merely use electronic data collection (e.g., tablet-based questionnaires) - Phone calls for follow-up or data collection (unless phone-based coaching IS the intervention) - Electronic health records as part of routine care - Using a computer to randomize patients - Video recording consultations for research purposes Key test: Remove the digital component — does the intervention fundamentally change or become impossible? If yes → "yes". If the study could work with paper forms instead → "no".
Add entry
non_cancer_spec_rules active
Clarifies when to use early_palliative_care vs N/A for non_cancer_spec. 9 disagreements found.
2026-04-22 14:09
Target Guidance Text
non_cancer_spec Only applies when cancer_status = "non-cancer". If cancer_status = "Cancer", always use "N/A". Use "early palliative care" ONLY when: - The study explicitly targets EARLY integration of palliative care (not late/end-of-life) - The study is disease-agnostic (not focused on one specific disease) - The population is defined by palliative care needs, not a specific diagnosis Use "N/A" when: - The study is about palliative care in general but with mixed or cancer populations - The study targets a specific disease that fits another category (cardiac, respiratory, etc.) - The study does not specify the stage or timing of palliative care involvement Prioritize specific disease categories: If the study is about COPD patients receiving palliative care → "respiratory", NOT "early palliative care". Dementia vs cognitive impaired: Use "dementia" for diagnosed dementia (Alzheimer, vascular, Lewy body). Use "cognitive impaired" for mild cognitive impairment, delirium, or unspecified cognitive dysfunction.
Add entry
outcome_instrument_rules active
Comprehensive instrument lists for symptom, QoL, dignity outcomes. 6 disagreements on outcome_symptom.
2026-04-22 14:09
Target Guidance Text
outcome_dignity Classify as "yes" if outcomes measure dignity, spirituality, meaning, or existential well-being: - PDI (Patient Dignity Inventory) - FACIT-Sp (Functional Assessment of Chronic Illness Therapy - Spiritual Well-being) - DEM-II (Dignity-related Existential Module) - MLQ (Meaning in Life Questionnaire) - SAIL (Spiritual Attitude and Involvement List) - Dignity Impact Scale - Schedule for Meaning in Life Evaluation (SMiLE) - Demoralization Scale - Existential Distress Scale
outcome_qol Classify as "yes" if outcomes use validated Quality of Life instruments: - EORTC QLQ-C30 (and modules: QLQ-LC13, QLQ-BR23, etc.) - FACT-G (Functional Assessment of Cancer Therapy - General) and subscales (FACT-L, FACT-B, etc.) - EQ-5D, EQ-5D-5L (EuroQol) - SF-36, SF-12 (Short Form Health Survey) - WHOQOL-BREF - McGill Quality of Life Questionnaire - FACIT-Pal (palliative care specific) - POS/IPOS quality of life domains Note: If a study uses BOTH symptom AND QoL instruments, classify BOTH outcome_symptom AND outcome_qol as "yes".
outcome_symptom Classify as "yes" if ANY primary or secondary outcome measures symptoms using validated instruments or direct symptom assessment. Common SYMPTOM instruments (→ outcome_symptom = "yes"): - Pain: NRS, VAS, BPI (Brief Pain Inventory), McGill Pain Questionnaire - Multiple symptoms: ESAS (Edmonton Symptom Assessment System/Scale), MSAS (Memorial Symptom Assessment Scale), IPOS (Integrated Palliative Outcome Scale) - Depression: PHQ-9, PHQ-8, BDI (Beck Depression Inventory) - Anxiety: GAD-7, STAI (State-Trait Anxiety Inventory) - Combined: HADS (Hospital Anxiety and Depression Scale) - Dyspnea: mMRC, Borg Scale, Dyspnea-12 - Fatigue: BFI (Brief Fatigue Inventory), FACIT-F - Nausea: MASCC Antiemesis Tool - Distress: Distress Thermometer, POS (Palliative Outcome Scale) NOT symptom instruments (these are QoL → outcome_qol): - EORTC QLQ-C30, FACT-G, EQ-5D, SF-36, WHOQOL
Add entry
palliative_intervention_rules active
Distinguishes palliative care setting from palliative care intervention. 8 disagreements found.
2026-04-22 14:09
Target Guidance Text
intervention_palliative Distinguishes palliative care as SETTING from palliative care as INTERVENTION. Classify as "yes" ONLY when the intervention IS palliative care itself: - Early palliative care consultation/integration programs - Palliative care team involvement as the experimental arm - Specialist palliative care models (e.g., hospital palliative care team vs usual care) - Interventions designed to improve palliative care delivery Classify as "no" when: - The study happens to be conducted with palliative care patients but tests something else (e.g., a drug, a psychological intervention, a device) - The intervention is a specific therapy delivered IN a palliative care setting - The study tests an intervention that COULD exist outside palliative care Key test: Is "palliative care" the INTERVENTION being tested, or is it the SETTING where another intervention is tested? If setting → "no".
Add entry
screening_intervention_rules active
Distinguishes screening as intervention from screening as outcome assessment. 4 disagreements.
2026-04-22 14:09
Target Guidance Text
intervention_screening intervention_screening = "yes" ONLY when SCREENING IS the intervention being tested. Classify as "yes": - Testing a systematic screening program (e.g., screening for depression in palliative care patients) - Comparing screening tool A vs screening tool B - Evaluating the impact of implementing a screening protocol - Primary purpose = SCREENING Classify as "no": - Using screening tools to ASSESS outcomes (that is an outcome measure, not an intervention) - Screening for eligibility before randomization - Baseline assessments using validated instruments - Using screening to identify patients for enrollment Key test: Is the ACT of screening the thing being experimentally tested? If yes → "yes". If screening is just how you measure results → "no".
Add entry
setting_patient_rules active
Clarifies inpatient vs outpatient vs mixed classification. 6 disagreements found.
2026-04-22 14:09
Target Guidance Text
setting_patient Classify based on the patient care status DURING the intervention: "inpatient": Patient is admitted/hospitalized during the intervention. Must involve overnight stay or continuous monitoring. "outpatient": Patient comes for visits but goes home. Includes clinic visits, day procedures, home-based care. "mixed": Study explicitly includes BOTH inpatient AND outpatient phases, or recruits from both settings. Default to "outpatient" when not explicitly stated — most clinical trials are outpatient. Do NOT classify as "mixed" just because some patients might be hospitalized during the study for unrelated reasons. Do NOT classify as "inpatient" just because the study site is a hospital — the question is about the patient status, not the building.
Add entry
setting_place_rules active
Clarifies hospital vs other for setting_place classification based on disagreement analysis (21 disagreements)
2026-04-22 14:09
Target Guidance Text
setting_place setting_place refers to WHERE the intervention is DELIVERED to the patient, NOT where the study site/facility is located. Key rules: - A study conducted at a hospital but delivering outpatient clinic visits = "other" (not "hospital") - Only "hospital" if the patient must be hospitalized/admitted as part of the intervention (inpatient ward, ICU, hospital-based infusion) - Use "other" as default when ambiguous, for generic outpatient clinics, community settings, or multi-site studies - Do NOT default to "hospital" just because the trial is registered at a hospital or medical center Specific mappings: - Outpatient clinic at hospital → "other" - Day hospital / ambulatory care → "other" - Inpatient ward / admitted patients → "hospital" - Home visits / telehealth from home → "home care" - Nursing home / long-term care / residential care → "nursing home" - Hospice unit / hospice facility → "hospice"
Add entry