Methodology — how we score your health

0. About this document

We publish our methodology because health AI without a documented methodology is just a guess. This page is written for two audiences: the clinical reviewers who ratify our work (M.D.s, naturopathic doctors, and condition specialists), and any patient, journalist, or research engine that wants to know exactly how the Phi Longevity score and confidence rating are produced.

Phi Longevity generates a numerical health score and a confidence rating (low / medium / high) from any combination of biomarkers a user uploads. This document explains:

1. What tests we require, recommend, or flag as missing for each of the four v1 condition tracks (General Wellness, Type 2 Diabetes / Pre-Diabetes, Lupus, Cancer Survivorship).
2. Where our confidence thresholds come from — every threshold is anchored to a published clinical guideline or peer-reviewed research, not invented.
3. How each test contributes to the score for the relevant clinical pillar (Metabolic, Cardiovascular, Hormonal, Inflammation & Immunity, Biological Age).
4. What's still in calibration — where the evidence is mixed and we welcome clinical input.

HIPAA note: all example biomarker values referenced on this page are synthetic, generated from clinical reference ranges. No patient PHI appears anywhere in this document.

1. The four condition tracks (v1)

When a user uploads their labs, they select a condition focus. v1 supports four tracks, each anchored to the primary clinical guideline body for that condition:

Future tracks (Phase 2+, post clinical-team feedback): Cardiovascular Disease primary prevention as its own track; Polycystic Ovary Syndrome (PCOS); Long COVID; Thyroid disease as its own track.

2. The five clinical pillars

The engine reports five scoring pillars. The composite Phi Score is a weighted average across them.

Pillar weighting is currently fixed across tracks. Whether weights should vary by condition (e.g., heavier Inflammation & Immunity in lupus) is an open clinical question — see §11.

3. Confidence framework

Every analysis returns a confidence level: low / medium / high. The v1 thresholds are count-based; v1.1 is moving to per-condition gating (see §3.5 below).

3.1 Current v1 thresholds

3.2 Why these specific cutoffs — honest answer

The 7 / 15 thresholds were chosen during initial engine development as round numbers approximating:

• Low (<7): a "spot check" — typically a partial panel. Insufficient for predictive inference across all five pillars.
• Medium (7–14): at least one full clinical panel (e.g., a CMP + lipid panel + HbA1c is ~12 biomarkers).
• High (≥15): a comprehensive panel including at least one extension into hormonal or inflammation testing.

The v1.1 evolution replaces this single global threshold with per-condition per-pillar confidence gating, anchored to the clinical literature for each track. The remainder of this document drafts those per-track frameworks.

3.3 Anchor: how published scoring systems handle "minimum data"

We cannot find a canonical paper that says "you need exactly N biomarkers to score someone." We can find canonical scoring systems that imply minimum-data requirements:

• ASCVD Pooled Cohort Equations (Goff et al., 2014) — 9 inputs. Below these, the published validity does not hold.
• Framingham Risk Score (Wilson et al., 1998) — 6 inputs minimum.
• QRISK3 (Hippisley-Cox et al., 2017) — 16 inputs at full specification.
• EULAR/ACR 2019 SLE Classification Criteria (Aringer et al.) — ANA is an entry criterion; without ANA ≥1:80, lupus classification is by definition not warranted regardless of other tests.
• 2018 ACC/AHA Cholesterol Guidelines (Grundy et al.) — classifies risk on ASCVD score + selective coronary calcium scoring + LDL targets, explicitly multi-input.

These are the natural anchors for our confidence framework — and they suggest per-condition minimum required sets that go beyond a global count.

4. General Wellness track

4.1 Required tests (must be present for ≥ low confidence)

4.2 Recommended (push to ≥ medium)

Adding these moves the user from "screening adequate" to "comprehensive baseline": Fasting Glucose (ADA 2024), Fasting Insulin (HOMA-IR per Wallace 2004), ApoB (ESC/EAS 2019), Lp(a) (ESC 2019), Total Cholesterol, CBC, Vitamin B12, Ferritin, Homocysteine, eGFR/Creatinine (KDIGO 2024), Body Fat %, HRV (Shaffer & Ginsberg 2017), RHR (Aune 2017), Sleep duration (NSF 2015).

4.3 High-confidence panel

Required 7 + recommended 14 = ~21 biomarkers ≈ comprehensive annual baseline + wearable signals. Enables predictive scoring across all five pillars without major data gap.

5. Type 2 Diabetes / Pre-Diabetes track

5.1 Required tests

5.2 Recommended (push to ≥ medium)

ApoB (small-dense LDL particle count, ESC/EAS 2019), eGFR/Creatinine (KDIGO 2024 CKD), ALT (NAFLD/MASLD screening per ADA 2024), Vitamin D (Pittas 2019 D2d), TSH (thyroid co-occurrence), hs-CRP (T2D inflammation).

5.3 High-confidence panel

Required 7 + recommended 6 + 2–3 of: Lp(a), Homocysteine, ≥3 historical HbA1c readings, HRV/RHR (autonomic / neuropathy early sign), Sleep duration (sleep apnea is highly prevalent in T2D).

6. Lupus (SLE) track

6.1 Required tests

6.2 Strongly recommended (push to ≥ medium)

dsDNA antibodies (EULAR/ACR 2019; disease activity correlation), C3 / C4 complement (low = active disease), anti-Sm antibodies (specific for SLE), anti-Ro (SSA) / anti-La (SSB) (Sjogren's overlap; pregnancy planning), ESR, Vitamin D (Yap & Morand 2015), Urinalysis with sediment.

Honest disclosure (per §3.5 Finding D): dsDNA, C3, C4, anti-Sm, anti-Ro, anti-La are recognized by the missing-tests layer but are not yet score-contributing in the production catalog. The next engine update adds them with EULAR-aligned ranges. We'd rather say this here than pretend otherwise.

6.3 High-confidence panel

Required 4 + ≥6 of the recommended 7, plus typically: anti-phospholipid panel (lupus anticoagulant, anti-cardiolipin, anti-β2-glycoprotein I), liver enzymes (drug monitoring), HbA1c + lipids (steroid-use risk profile). ~14–16 biomarker lupus-comprehensive panel.

6.4 Proposed hard-stop rule for v1.1

If conditionFocus = lupus AND ANA is missing, confidence should return insufficient_for_track rather than low. The UI surfaces "Add ANA to begin lupus-specific assessment". This is the kind of rule the clinical team is invited to ratify or reject.

7. Cancer Survivorship track

7.1 Required tests

CBC with differential (treatment-related cytopenias; NCCN Survivorship v2.2024), CMP (baseline organ function; cardiotoxic/hepatotoxic/nephrotoxic chemo exposure), HbA1c (cancer + cancer-treatment-related metabolic risk; steroids), LDL (anthracyclines, trastuzumab → premature CVD per Armenian 2017 + Lyon 2022), TSH (thyroid effects of radiation + checkpoint inhibitors), hs-CRP (chronic inflammation post-treatment + second-cancer risk).

7.2 Recommended (conditional on cancer type)

This is where cancer survivorship gets nuanced. Different primary cancers have different surveillance demands. Recommended additions include: PSA (prostate), CA-125 (ovarian), CA 19-9 (pancreatic / biliary), Mammography history (breast), BNP / NT-proBNP + ECG (anthracycline / trastuzumab / chest radiation exposure per ESC 2022 Cardio-Oncology), Vitamin D (bone-mineral density; chemo-induced osteoporosis), Calcium + Albumin (corrected calcium), eGFR/Creatinine (cisplatin / methotrexate / immunotherapy history), ALT/AST (hepatotoxic chemo + immune-mediated hepatitis from ICI), Iron studies, Ferritin (transfusion-related iron overload), Vitamin B12 (methotrexate; gastric history).

7.3 High-confidence panel

Required 6 + cancer-type-specific tumor markers + cardio-oncology panel (BNP, ECG, possibly echo) + bone-density markers + 2+ years of trend data.

8. Cross-cutting considerations

8.1 Sex-specific testing

Several biomarkers have sex-specific reference ranges: Testosterone / Free Testosterone, Estradiol / Progesterone (cycle-dependent in pre-menopausal female), FSH / LH (menopausal-status indicator), Body Fat %. The engine accepts an optional biologicalSex parameter. Whether menopausal-status self-report should be required for female-specific hormonal interpretation is an open clinical question.

8.2 Age stratification

Some recommendations vary by age (mammography ≥40; PSA 50–69; Testosterone symptomatic-only per Endocrine Society 2018). v1 does not yet stratify by age; v1.1 candidate.

8.3 Wearable signal integration

HRV, RHR, Sleep, and Steps from connected wearables (Fitbit, Oura, Garmin, WHOOP) contribute primarily to the Biological Age pillar, with secondary contributions to Cardiovascular (RHR, HRV) and Metabolic (Sleep — sleep deprivation impairs glucose tolerance). If wearable data is absent, the Biological Age pillar's confidence is reduced and the user is invited to connect a device.

9. Calibration methodology (synthetic-data validation)

To validate our thresholds, we run a synthetic biomarker calibration harness:

1. For each condition track, generate scenarios with N biomarkers (1–25) drawn from clinical reference ranges.
2. Vary biomarker values across optimal / borderline / out-of-range.
3. Score each scenario via the production computeScoreForBiomarkers endpoint.
4. Record (track, biomarker_set, value_profile) → (score, confidence, pillar_breakdown, missing_tests).
5. Aggregate across hundreds of scenarios → identify the empirical relationship between biomarker count, mix, and resulting confidence.

HIPAA compliance: all synthetic biomarker values are generated from clinical reference ranges. No real patient data is used. No real user UID is touched. The harness operates entirely on synthetic substrate per the project's HIPAA hard rule.

10. Limitations and known gaps

1. v1 confidence thresholds are coarse. A single global cutoff (7 / 15) doesn't reflect per-condition reality. The remainder of this document specifies per-track logic that v1.1 encodes.
2. No primary-diagnostic-test gating yet. Lupus without ANA shouldn't be "low confidence" — it should be "not assessable for lupus track." v1.1 fix proposed in §6.4.
3. Cancer survivorship is undifferentiated. Cancer-type sub-tracks are a v1.1 candidate.
4. Age stratification not yet implemented. USPSTF / NCCN / Endocrine recommendations vary by age band.
5. Confidence scoring is binary per pillar. No partial credit for stale data (e.g., LDL from 3 years ago). v1.2 candidate.
6. No comorbidity-aware adjustments yet. A patient with both T2D and CKD has different optimal LDL targets than T2D alone.
7. Race/ethnicity weighting is intentionally not included. Recent guidelines have moved away from race-adjusted GFR. We follow that.
8. All calibration in v1 is synthetic. Real-world performance pending v1.2 from beta cohort.
9. Lupus serology markers (dsDNA, C3, C4, anti-Sm, anti-Ro, anti-La) are not yet score-contributing. They are correctly recognized by the missing-tests / recommendation layer (ANA-gating verified live in calibration — see §3.5 Finding C). The next engine update closes this gap with EULAR-aligned ranges.
10. score = null boundary. When fewer than 3 biomarkers are scored, the engine returns null plus a CTA recommending specific tests. Whether this null-boundary is appropriate or whether we should attempt a coarse score below 3 is a v1.1 candidate question.

11. Open clinical questions we're working on

These are the questions we are actively seeking clinical-team input on. Each is a real decision in the engine — not theoretical.

1. Challenge the required-test lists per track. Are any tests missing? Any we'd remove?
2. Challenge the confidence thresholds. Is "≥7 = medium, ≥15 = high" defensible? Or is the per-track logic in §4–7 the right floor?
3. Ratify or reject the lupus ANA-gating proposal (§6.4) — should a missing primary diagnostic test downgrade confidence below "low"?
4. Ratify or reject the cancer sub-track proposal — should top-5 cancers (breast, prostate, colorectal, lung, melanoma) each get type-specific panels?
5. Identify citations to substitute or add. Some v1 anchors may not be your preferred references.
6. Flag clinical edge cases we're missing. Pediatrics? Pregnancy? Long COVID? Bariatric-post-op? Add tracks in v2 if you signal need.
7. Ratify the lupus serology range plan. The next engine update adds dsDNA, C3, C4, anti-Sm, anti-Ro (SSA), anti-La (SSB) to the scored catalog with EULAR-aligned ranges. A rheumatologist's read on the exact thresholds is the highest-leverage input we can get on lupus.
8. Read the proposed gating rules. §3.5, §6.4, §7. Each is a yes/no/edit decision.

If you're a clinician interested in reviewing this work, contact the founders via the team page.

12. References (full bibliography)

Type 2 Diabetes

• American Diabetes Association. Standards of Care in Diabetes—2024. Diabetes Care 2024;47(Suppl 1).
• Wallace TM, Levy JC, Matthews DR. Use and abuse of HOMA modeling. Diabetes Care 2004;27(6):1487-1495. PMID 15161807
• Pittas AG, Dawson-Hughes B, Sheehan P, et al. Vitamin D Supplementation and Prevention of Type 2 Diabetes (D2d). N Engl J Med 2019;381(6):520-530. PMID 31173679

Lupus / SLE

• Aringer M, Costenbader K, Daikh D, et al. 2019 European League Against Rheumatism / American College of Rheumatology Classification Criteria for Systemic Lupus Erythematosus. Arthritis Rheumatol 2019;71(9):1400-1412. PMID 31385462
• Fanouriakis A, Kostopoulou M, Andersen J, et al. EULAR recommendations for the management of systemic lupus erythematosus: 2023 update. Ann Rheum Dis 2024;83(1):15-29. PMID 37827694

Cancer Survivorship

• Sanft T, Day AT, Goldman M, et al. NCCN Guidelines Insights: Survivorship, Version 2.2024. J Natl Compr Canc Netw 2024;22(10):648-658.
• Armenian SH, Lacchetti C, Barac A, et al. Prevention and Monitoring of Cardiac Dysfunction in Survivors of Adult Cancers: ASCO Clinical Practice Guideline. J Clin Oncol 2017;35(8):893-911. PMID 27918725
• Lyon AR, López-Fernández T, Couch LS, et al. 2022 ESC Guidelines on Cardio-Oncology. Eur Heart J 2022;43(41):4229-4361. PMID 36017568

Cardiovascular / General

• Goff DC Jr, Lloyd-Jones DM, Bennett G, et al. 2013 ACC/AHA Guideline on the Assessment of Cardiovascular Risk. Circulation 2014;129(25 Suppl 2):S49-S73. (ASCVD Pooled Cohort Equations)
• Wilson PWF, D'Agostino RB, Levy D, et al. Prediction of coronary heart disease using risk factor categories. Circulation 1998;97(18):1837-1847. (Framingham)
• Hippisley-Cox J, Coupland C, Brindle P. Development and validation of QRISK3 risk prediction algorithms. BMJ 2017;357:j2099.
• Arnett DK, Blumenthal RS, Albert MA, et al. 2019 ACC/AHA Guideline on the Primary Prevention of Cardiovascular Disease. Circulation 2019;140(11):e596-e646.
• Grundy SM, Stone NJ, Bailey AL, et al. 2018 AHA/ACC Guideline on the Management of Blood Cholesterol. Circulation 2019;139(25):e1082-e1143.
• Mach F, Baigent C, Catapano AL, et al. 2019 ESC/EAS Guidelines for the management of dyslipidaemias. Eur Heart J 2020;41(1):111-188.
• Greenland P, Alpert JS, Beller GA, et al. 2010 ACCF/AHA Guideline for Assessment of Cardiovascular Risk in Asymptomatic Adults. Circulation 2010;122(25):e584-e636.
• Ridker PM, Cannon CP, Morrow D, et al. C-reactive protein levels and outcomes after statin therapy. N Engl J Med 2005;352(1):20-28.

Renal

• Stevens PE, Ahmed SB, Carrero JJ, et al. KDIGO 2024 Clinical Practice Guideline for the Evaluation and Management of CKD. Kidney Int 2024;105(4S):S117-S314.

Thyroid

• Jonklaas J, Bianco AC, Bauer AJ, et al. Guidelines for the treatment of hypothyroidism. Thyroid 2014;24(12):1670-1751.

Vitamin D

• Demay MB, Pittas AG, Bikle DD, et al. Vitamin D for the Prevention of Disease: An Endocrine Society Clinical Practice Guideline. J Clin Endocrinol Metab 2024.
• Note: Holick MF et al. 2011 Endocrine Society Vitamin D guideline has been retired and is superseded by Demay 2024.

Wearables / Biological Age

• Shaffer F, Ginsberg JP. An Overview of Heart Rate Variability Metrics and Norms. Front Public Health 2017;5:258.
• Aune D, Sen A, ó'Hartaigh B, et al. Resting heart rate and the risk of cardiovascular disease, total cancer, and all-cause mortality. Nutr Metab Cardiovasc Dis 2017;27(6):504-517.
• Hirshkowitz M, Whiton K, Albert SM, et al. National Sleep Foundation's sleep time duration recommendations. Sleep Health 2015;1(1):40-43.