Independent biomedical research

What can a biomarker
tell us about survivorship?

I built ONQIVA to make one research question inspectable: how does measured vitamin D relate to outcomes after cancer, and what would a transparent testing workflow need to prove before it could help a clinic?

Try the fictional workflowRead the NHANES analysis
515-person survivorship analysisALS recovery study conceptHands-on protein research
A WORKED RESEARCH NOTE

From a fitted coefficient to a result you can inspect.

This is one example of my work moving from a mathematical question, through public data, to a result with its limits attached.

EXPLORATORY · NHANES 2017–2018

Who was in the analysis?

Adults who reported a previous cancer diagnosis, had measured serum 25(OH)D, and were eligible for linked mortality follow-up.

515participants
35observed deaths
37 momaximum follow-up

In the descriptive groups, 65 people had 25(OH)D below 20 ng/mL and 450 had 20 ng/mL or higher. These counts are not a clinical threshold or a screening rule.

AGE- AND SEX-ADJUSTED COX MODEL
0.69hazard ratio per +10 ng/mL 25(OH)D

95% confidence interval: 0.51 to 0.94

WORKED COEFFICIENTexp(−0.367) ≈ 0.69The vitamin D coefficient is −0.367 for each 10 ng/mL increase in the fitted model.
Primary sourcesAnalysis code ↗CDC/NCHS NHANES files ↗Review paper (2023) ↗
A SEPARATE PRODUCT EXERCISE

Real analysis on one side. Fictional workflow on the other.

The clinic scenario below is a rule-based simulation. It does not use the NHANES model, patient records, or clinical outcomes.

CHOOSE A PATH

What brought you here?

These paths change the educational content, not a person’s risk or care plan.

FICTIONAL SCENARIO

Post-treatment care

Example profile · age 52 · five years after treatment

OUTDOOR ACTIVITYAbout 1 hour each week
BODY MASS INDEX31 kg/m²
DIETARY VITAMIN DLow reported intake
SUPPLEMENT USENone reported
Fictional clinic context

The clinic has a limited number of confirmatory tests this month. This exercise shows how a written scoring rule could be inspected before any real-world study.

Fictional, rule-based demonstration. It does not estimate a diagnosis or show clinical benefit.
FICTIONAL OUTPUT

See the rule, then question it.

Run the example to see how its written assumptions change a score and a capacity-limited allocation.

CHANGE ONE ASSUMPTION

100 fictional profiles.
20 available tests.

Change the number of available tests and run the example again. The comparison uses simulated outcomes.

TESTS IN EXAMPLE20 / 100
RULE-PRIORITIZEDNot run
RANDOM COMPARISONNot run

All 100 profiles and outcomes are simulated. The comparison does not establish that either approach improves care.

GENERAL NUTRITION EDUCATION

Food ideas that start with what you eat.

Share foods you enjoy and choose any preferences that matter to you. Skip anything you do not want to answer.

01

Your real foods and meals

OPTIONAL
0/280 characters

Use your own words. The planner looks for familiar food signals, keeps unfamiliar meal names, and shows what it understood.

Or tap a few foods you enjoy

Combine these choices with anything you type above.

02

Make it fit your routine

OPTIONAL
03

Foods to avoid

OPTIONAL

This list cannot check every allergy or cross-contact risk. Check ingredients; select “another allergy” to pause food suggestions.

04

Prepare for your next visit

OPTIONAL

Your answers stay on this page and clear when you reload. These are general food ideas, not treatment or supplement advice. The NHANES research model does not prescribe meals or doses.

Why these suggestions? Explore the science

Vitamin D is found in foods such as fatty fish and in some fortified products. This planner matches a small collection of food ideas to your choices. It does not measure vitamin D status or predict cancer outcomes.

ONQIVA’s research connects laboratory biology and mathematical modeling. Its exploratory NHANES findings are separate from this educational checklist.

NIH vitamin D food and nutrition information · Explore ONQIVA’s research
EXPLORATORY ANALYSIS · NHANES PUBLIC DATA

I rebuilt the Python analysis
with NHANES.

Explore measured vitamin D and short-term all-cause survival among adults who reported a prior cancer diagnosis. Association is not causation.

EXPLORATORY SURVIVAL ANALYSIS · OBSERVATIONAL

All-cause survival after NHANES examination

MODEL SCENARIO
ANALYTIC COHORT515self-reported cancer history
OBSERVED DEATHS35maximum 37 months
COX C-INDEX0.723same analysis sample
VITAMIN D HR0.693per +10 ng/mL; 95% CI 0.5087–0.9442
How to interpret this responsibly

The green and red curves are unadjusted Kaplan–Meier estimates. The dashed curve is S₀(t)exp(βX) from an age-, sex-, and vitamin-D-adjusted Cox model. The association may reflect confounding, reverse causation, cancer-type differences, treatment, or selection. It does not show that supplementation improves survival.

Reproducible source trail

Official CDC/NCHS files joined by SEQN. Serum total 25(OH)D was converted from nmol/L to ng/mL. Cancer history uses MCQ220; follow-up uses PERMTH_EXM and MORTSTAT.

Demographics ↗Vitamin D ↗Cancer history ↗Mortality linkage ↗Model output JSON ↗
CROSS-SECTIONAL BENCHMARK · RESEARCH ONLY

Demographics contain some signal, with clear limits.

This real benchmark asks whether demographics available before a laboratory result can predict measured 25(OH)D below 20 ng/mL among the same cancer-history cohort.

01INGEST

CDC XPT + linked mortality

→
02ENGINEER

Impute, scale, one-hot encode

→
03TRAIN

Logistic + gradient boosting

→
04VALIDATE

Five held-out folds

→
05GATE

Calibration, subgroups, utility

INTERPRETABLE BASELINE

Logistic regression

0.652ROC-AUC
0.2263AVG PRECISION
0.1067BRIER ↓
NONLINEAR COMPARATOR

Histogram gradient boosting

0.6657ROC-AUC
0.2291AVG PRECISION
0.1118BRIER ↓
FEATURES USED

Age · sex · race/ethnicity · examination season · income-to-poverty ratio

VALIDATION DESIGN

Five-fold stratified cross-validation; all preprocessing refit inside each fold.

HONEST READOUT

Best ROC-AUC ≈ 0.67. Demographics contain signal, but not enough for clinical deployment. The next model needs richer predictors, calibration testing, and external validation.

THE EVIDENCE GAP ONQIVA IS BUILT TO STUDY

When does measured vitamin D add useful information?

Vitamin D is studied in bone health, immune signaling, cell growth, and survivorship. The value of testing may not be identical for every person or every cancer context. That is a modeling problem: identify where a measured biomarker improves risk stratification beyond information clinicians already have.

This analysis asks that question transparently. It compares models with and without vitamin D, examines uncertainty and subgroup performance, and asks whether better evidence could support more focused use of limited testing resources. A negative or mixed result is still valuable because it prevents weak associations from becoming confident clinical claims.

Does 25(OH)D improve prediction?For which survivor subgroups?Is the model calibrated?Would testing change a decision?
NIH vitamin D evidence ↗NCI research context ↗Screening evidence gap ↗
FOR TEENS & YOUNG ADULTS

Vitamin D literacy, without the hype.

Build healthy habits and understand the science. Do not turn a research association into a diagnosis or supplement plan.

01

Know your food sources

Fatty fish, egg yolks, and fortified foods such as some milks, plant beverages, and cereals can contribute vitamin D.

02

Enjoy outdoors safely

Outdoor activity supports overall wellbeing, but deliberately increasing UV exposure is not a safe vitamin D strategy. Protect your skin.

03

Skip the megadose mindset

More is not automatically better. Supplements can interact with health conditions and excessive intake can cause harm.

04

Ask when risk is real

A clinician can help when there is a prior low result, malabsorption, bone-health concern, relevant medication, or another genuine risk factor.

NIH guide for consumers ↗CDC outdoor safety ↗
TWO SEPARATE TRACKS

Keep each result in its proper context.

One analysis uses public survey data. The interactive clinic exercise uses fictional profiles and simulated outcomes.

PUBLIC DATA · EXPLORATORY

NHANES outcomes analysis

CDC/NCHS demographics, measured serum 25(OH)D, self-reported cancer history, and linked mortality follow-up. The analysis includes 515 adults and 35 observed deaths.

The result is observational, unweighted, and based on short follow-up. It is a research finding, not clinical evidence.

Read the methods
FICTIONAL DATA · RULE-BASED

Testing-allocation exercise

A separate demonstration asks how a clinic might compare priority rules when test capacity is limited. Its scores do not come from NHANES or from a trained clinical predictor.

A real evaluation would need a defined protocol, independent review, appropriate data, and external validation before clinical use could even be considered.

Try the example
THE NEXT RESEARCH STEPS

Earn each claim with evidence.

  1. 01IN THE REPO

    Reproduce the analysis

    Make the cohort, variables, exclusions, and uncertainty easy for another researcher to inspect.

  2. 02NEEDS REVIEW

    Test the assumptions

    Seek statistical and oncology feedback on the question, analysis choices, and what the current data cannot answer.

  3. 03PROPOSED

    Define a responsible next study

    Any clinical pilot would require partners, suitable data, governance, and validation. None is presented as underway.

FOUNDER & RESEARCH PATH

My work started on paper, then moved into code.

I worked through the model mathematics by hand, prepared the data, and then translated the analysis into Python. ONQIVA makes that work available for other people to examine, question, and improve.

See what I have done and what comes next
SCIENCE IN THE COMMUNITY

Research should be understandable—and shaped by the people it concerns.

See how ONQIVA plans to make study boundaries visible, teach the methods, and build future research questions with the community. This is an invitation for the next stage, not a claim that a formal partnership already exists.

Explore the community approach
SUPPORT THE NEXT STUDY

My research record now points to a testable next step.

ONQIVA is seeking financial supporters for a staged ALS motor-neuron recovery study. The funding case shows the work I have already done, the prediction to test, the milestones, and what support would pay for. I also welcome independent review of the separate NHANES survivorship analysis.

The ALS study is proposed; the published papers and public-data analysis are completed work.

Read the funding case