Panacea Index Logo

Command Palette

Search for a command to run...

Research Methodology

Paper Weight Calculation

Input Data Fields:

  • quality_score - Paper quality rating (1-100) (Used an LLM to grade based on things such as sample size, study duration, etc.)
  • citation_count - Number of times cited
  • is_clinical - Boolean for clinical trials
  • is_human_study - Boolean for human studies
  • is_animal_study - Boolean for animal studies
  • is_molecular_study - Boolean for molecular studies
  • publication_date - When paper was published

Weight Calculation Steps:

  • Base Weight: 1.0 + max(0, quality_score - 50) / 100
  • Citation Weight: 1 + log(citation_count + 1) / 10
  • Study Type Weight:
    • Clinical/Human studies: 2.0
    • Animal studies: 1.5
    • Molecular studies: 1.2
    • Other studies: 1.0
  • Recency Weight: exp(-years_old / 15.0) (exponential decay)
  • Final Weight: base × citation × study_type × recency

Example Weight Calculation:

Paper: High-quality human clinical trial
- quality_score: 85
- citation_count: 150
- is_human_study: true
- publication_date: 2020

Calculations:
- base_weight = 1.0 + (85-50)/100 = 1.35
- citation_weight = 1 + log(151)/10 = 1.51
- study_type_weight = 2.0 (human study)
- years_old = 2025-2020 = 5
- recency_weight = exp(-5/15) = 0.72

Final Weight = 1.35 × 1.51 × 2.0 × 0.72 = 2.94

Paper Score Calculation

Input Data Fields:

  • safety_score - Safety rating (1-100) -- (LLM Generated)
  • efficacy_score - Effectiveness rating (1-100) -- (LLM Generated)
  • quality_score - Study quality rating (1-100) -- (LLM Generated)

Score Normalization Steps:

  • Safety Score Normalization (40% weight):
    • Text "safe" → 0.8
    • Text with "concern/warning/risk" → 0.3
    • Numeric values → divide by 100
    • Fractions (e.g., "8/10") → calculate ratio
    • Default/missing → 0.5
  • Efficacy Score Normalization (40% weight):
    • Range 1-100 → divide by 100
    • Missing values → default to 50
  • Quality Score Normalization (20% weight):
    • Range 1-100 → divide by 100
    • Missing values → default to 50
  • Combined Score: (safety × 0.4) + (efficacy × 0.4) + (quality × 0.2)

Example Score Calculation:

Paper with mixed results:
- safety_score: "Generally safe with minor concerns"
- efficacy_score: 75
- quality_score: 80

Calculations:
- safety_normalized = 0.3 (contains "concerns")
- efficacy_normalized = 75/100 = 0.75
- quality_normalized = 80/100 = 0.80

Combined Score = (0.3 × 0.4) + (0.75 × 0.4) + (0.80 × 0.2)
= 0.12 + 0.30 + 0.16 = 0.58

Supplement Aggregation Algorithm

Input Data:

  • All papers for a supplement with calculated weights and scores
  • Paper-level: weight_score, normalized_score, safety_score, efficacy_score, quality_score

Aggregation Steps:

  • Weighted Averages:
    • Safety: Σ(safety_normalized × weight) / Σ(weight)
    • Efficacy: Σ(efficacy/100 × weight) / Σ(weight)
    • Quality: Σ(quality/100 × weight) / Σ(weight)
  • Overall Score: (safety × 0.4) + (efficacy × 0.4) + (quality × 0.2)
  • Consistency Score: max(0, 1.0 - (std_dev / mean)) of normalized scores
  • Confidence Level:
    • High: ≥20 studies AND ≥30 total weight
    • Medium: ≥10 studies AND ≥15 total weight
    • Low: ≥5 studies AND ≥5 total weight
    • Very Low: <5 studies OR <5 total weight

Example Supplement Aggregation:

Supplement: Vitamin D
Papers: 25 studies with various weights

Study Data:
- Paper 1: weight=2.94, safety=0.8, efficacy=0.75, quality=0.85
- Paper 2: weight=1.45, safety=0.9, efficacy=0.65, quality=0.70
- Paper 3: weight=3.12, safety=0.7, efficacy=0.80, quality=0.90
- ... (22 more papers)

Calculations:
- Total weight: 45.8
- Weighted safety avg: 0.78
- Weighted efficacy avg: 0.72
- Weighted quality avg: 0.81
- Overall score: (0.78×0.4) + (0.72×0.4) + (0.81×0.2) = 0.762
- Confidence: "high" (25 studies, 45.8 weight)

Final Aggregate:
- avg_safety_score: 0.78
- avg_efficacy_score: 0.72
- avg_quality_score: 0.81
- overall_score: 0.762
- confidence_level: "high"
- research_summary: "Overall: 7.6/10 | Safety: 7.8/10 | Efficacy: 7.2/10 | Quality: 8.1/10 | Confidence: high (25 studies)"