Evidence & Methodology

How the analysis works: four stages

1. 1
   Extraction (AI)

   A large language model (Claude, by Anthropic) reads the uploaded PDF or image and identifies every blood marker: name, value, unit, lab reference range, and any H/L flags printed by the lab. The AI handles the wide variation in lab report formats — Quest Diagnostics, LabCorp, NHS, and private labs all lay out results differently. The AI does not make any clinical judgement at this stage. It produces structured data only.

2. 2
   Threshold correction (deterministic)

   The extracted values are passed through a deterministic rules engine that applies evidence-based threshold overrides. This step handles two important problems: (1) lab reference ranges vary between labs and can be wider than current clinical guidelines recommend; (2) some markers require sex- or age-adjusted interpretation that the raw lab range does not reflect. For example, a TSH of 3.8 mIU/L may be within a lab's printed range but warrant borderline classification based on ATA guidelines. All threshold decisions are hardcoded, traceable to a specific guideline, and medically reviewed.

3. 3
   Ranking (deterministic)

   Each non-optimal marker is scored across six dimensions: severity of deviation, clinical impact, actionability with lifestyle/supplement interventions, expected response time to intervention, questionnaire context (sex, age, medications, diet), and lab flag weighting. The three markers with the highest composite scores become the Top 3 priorities. Markers that rank 4th or lower are shown in a secondary list with graded status labels (Slightly High, High, etc.) but are not accompanied by full protocol recommendations — the intent is to direct attention, not to overwhelm.

4. 4
   Recommendations (deterministic)

   Protocol items — diet, lifestyle, supplements, and expected response timelines — are drawn from a static database keyed to each marker's identity, severity tier, and direction (high or low). No AI generates these recommendations. Every item in the database was written with reference to published clinical guidelines and reviewed by Dr. Prahlad Rai Gupta. When a top-3 marker conflicts with the dietary advice for another top-3 marker (e.g., high iron and low calcium requiring different intake guidance), a cross-marker reconciliation step flags the conflict explicitly.

What is AI vs deterministic

The clinical outputs of FixFirst — what is flagged, how it is ranked, and what is recommended — are fully deterministic and not subject to the hallucination or inconsistency risks of generative AI. The AI only touches the extraction step.

How thresholds are chosen

Each marker's threshold values are sourced from the primary clinical guideline most specific to that marker, then reviewed by Dr. Prahlad Rai Gupta. When guidelines disagree, the more conservative threshold is used and the conflict is documented in the codebase with a comment citing both sources.

The primary guideline families used:

• ADA — American Diabetes Association Standards of Medical Care (HbA1c, fasting glucose, insulin)
• ATA — American Thyroid Association (TSH, T3, T4)
• ACC/AHA — American College of Cardiology / American Heart Association (LDL, HDL, triglycerides, cardiovascular risk)
• NICE CKS — UK National Institute for Health and Care Excellence Clinical Knowledge Summaries (ferritin, B12, vitamin D, thyroid)
• NIH ODS — National Institutes of Health Office of Dietary Supplements (vitamin D, B12, magnesium, iron)
• WHO — World Health Organization (haemoglobin, anaemia thresholds)
• ARUP / Mayo Clinic Laboratories — for reference range benchmarking where primary guidelines do not specify lab ranges

How age, sex, and context are incorporated

An optional questionnaire captures biological sex, age, medications (specifically statins and thyroid medication), and dietary pattern (plant-based or omnivore). These answers adjust interpretation in specific, documented ways:

• Biological sex — adjusts thresholds for haemoglobin (female 12.0 g/dL vs male 13.5 g/dL per WHO), ferritin (lower optimal range for females), HDL (female >50 mg/dL vs male >40 mg/dL), and uric acid.
• Age — adjusts eGFR interpretation (mild reduction is less concerning above age 65), TSH thresholds (ATA age-stratified upper limits), and vitamin B12 absorption context.
• Statins — triggers a CoQ10 co-depletion note and adjusts LDL interpretation to account for expected pharmacological reduction.
• Plant-based diet — elevates the contextual concern for vitamin B12, iron, and vitamin D deficiency and adds a dietary conflict note where applicable.

Context adjustments are applied at the threshold correction and ranking stages. They are not applied to the AI extraction step.

How conflicting guidelines are handled

When two authoritative sources set different thresholds for the same marker, FixFirst uses the following decision order:

1. The guideline most specific to the marker's primary clinical significance (e.g., ATA for TSH rather than a general internal medicine reference).
2. Where specificity is equal, the more conservative threshold (the one that flags more rather than fewer patients as borderline).
3. The conflict is documented as a comment in the codebase with both sources cited, so it can be audited and updated if guidelines change.

What FixFirst does not do

• Diagnose any disease or condition.
• Replace evaluation by a qualified clinician.
• Handle medical emergencies — if you have acute symptoms, contact emergency services.
• Prescribe medications or recommend changes to existing prescriptions.
• Account for your full medical history, imaging, biopsy results, or clinical examination findings.
• Provide personalised medical advice of any kind.
• Store your blood test data — all processing happens in memory and is discarded after your session.

FixFirst tells you which markers are worth your attention first, based on their values relative to evidence-based thresholds. What you do with that information — including whether and how to discuss it with a doctor — is a decision that remains entirely yours.

Update and versioning philosophy

Threshold values, ranking logic, and recommendation content are reviewed against their source guidelines on a quarterly basis. When a major guideline is updated — for example, a new edition of the ADA Standards of Care — the affected thresholds are reviewed and updated before the next content publication cycle.

Material changes to the ranking algorithm or threshold database are tested against a regression suite before deployment to ensure they do not silently change the output for previously consistent test cases.

Questions about the methodology or requests for the specific threshold values used for a given marker can be sent to fixfirstio@gmail.com.