Your Waist Knows More Than MyFitnessPal: What 7,853 Americans’ Health Data Actually Shows

After my waist hit 100 cm and my triglycerides started climbing, I did something most people don’t do — I actually got the data.

I’m a behavioral designer, not a doctor. But I spent $1,500/month in AI subscriptions to download, clean, and analyze health data from 159 countries and 7,853 individual Americans from the CDC’s NHANES survey.

The single biggest finding? Your waist circumference predicts diabetes, gout, and metabolic disease 12x better than anything you eat. Not sugar. Not fiber. Not calories. A tape measure around your belly outperforms every food tracking app on the market.

This is Part 1 of a two-part series. This post covers what your body measurements and lifestyle habits actually predict — and what they don’t. Part 2: Gene Resilience explores how your ethnicity changes the entire health rulebook.

Important note: All findings in this article are observational correlations from cross-sectional data, not proven causal relationships. However, the waist circumference effect survives every control we tested: age, height, sex, ethnicity, income, and total calorie intake.

---

Author Credibility: Yu-kai Chou

Yu-kai Chou created the Octalysis Framework after studying gamification since 2003 — years before the term entered mainstream vocabulary. As a Human-Systems Architect & Behavioral Designer, his framework has been applied by LEGO, Microsoft, Porsche, Coca-Cola, Salesforce, and MrBeast, impacting over 1.5 Billion Users.

Chou has taught the Octalysis methodology at Harvard, Stanford, Yale, Tesla, Google, BCG, and IDEO.

His work has been cited by Harvard, Stanford, MIT, Forbes, Wall Street Journal, Wired, US Department of Energy, NIST, NSF, NCBI, US Department of Education, ClinicalTrials.gov, and Google Scholar — with 3,700+ more academic publications. Explore his books here.

1. Your Waist Knows 12x More Than MyFitnessPal

We ran a horse race: which variable best predicts your diabetes risk (HbA1c)?

We tested calories, sugar, fiber, sodium, saturated fat, protein percentage, fat percentage, BMI, waist circumference, age, and income. Here’s the ranking:

All dietary variables combined explain 0.5% of diabetes variance. Sleep explains 0.2%. Your waist circumference alone explains 5.7%. That’s a 12x difference.

Why Does Waist Dominate Everything Else?

Waist circumference is a direct proxy for visceral fat — the fat packed around your organs deep inside your abdomen. Visceral fat is metabolically far more dangerous than the subcutaneous fat you can pinch on your arms or thighs, or even your overall body weight (BMI).

Here’s the mechanism: visceral fat releases inflammatory cytokines and free fatty acids straight into the portal vein — the highway that feeds directly into your liver. This impairs your liver’s insulin sensitivity and drives up blood glucose. It’s a direct, mechanical pathway from belly fat to diabetes risk. No amount of sugar tracking captures this.

This is why waist beats BMI even though the two are highly correlated (r ≈ 0.85–0.90 in our data). BMI mixes up muscle, bone, and fat into one number. Waist circumference cuts through the noise and measures the one fat deposit that actually poisons your metabolism.

We verified this from multiple angles:

• Spearman rank correlations (non-parametric, no assumptions about data distribution): waist still wins decisively (ρ ≈ 0.32 vs. BMI’s 0.26)
• Multivariate regression (all five variables together): waist retains the largest standardized coefficient (β ≈ 0.28–0.30) and smallest p-value. BMI actually flips negative once waist is controlled for — a classic sign that central obesity is the real driver, and BMI was just riding its coattails
• Subgroup consistency: the pattern holds across age groups, sexes, and all ethnicities — including Asians, where diabetes risk starts at lower BMIs
• Edge cases: extreme calorie or sugar intakes (top 1%) and short sleep (<4 hours) don’t change the ranking

The clinical implication: a 10 cm increase in waist circumference is associated with roughly a 0.3–0.4 point rise in HbA1c — clinically meaningful. This dwarfs the negligible effects of sleep duration or dietary sugar. Measuring and reducing your waist size (via exercise + calorie deficit) is vastly more impactful for blood sugar control than obsessing over sugar grams or sleep tracking alone.

Every food tracking app, every macro calculator, every “sugar is poison” documentary. (For more on how fitness apps use behavioral design to keep you tracking, see my analysis.) All fighting over a signal that’s 12 times weaker than a tape measure around your belly.

Buy a tape measure. For men, the danger cutoff is 102 cm (or 90 cm if you’re Asian). That one number tells you more about your health than any food diary.

An even better metric than raw waist circumference may be your waist-to-height ratio (WHtR). A 2013 meta-analysis of 512,000+ people found WHtR superior to BMI for detecting diabetes, metabolic syndrome, cardiovascular disease, and mortality. The rule is simple: keep your waist under half your height.

In our NHANES data, WHtR slightly outperforms raw waist for predicting HbA1c in 3 out of 4 ethnic groups. It corrects for the obvious objection that taller people naturally have bigger waists. People over 178cm have average waists of 107cm but actually healthier blood sugar than people under 162cm with 98cm waists. WHtR catches what raw waist misses.

When I later expanded this analysis to 9 different health outcomes (adding inflammation, liver function, gout risk, insulin resistance, triglycerides, HDL cholesterol, and kidney function), waist circumference won 5 out of 9 categories. Age won 2 (diabetes and kidneys, which deteriorate with time regardless of lifestyle). Income won 1 (depression). BMI won 1 (inflammation). Sleep won zero.

---

2. Sugar Doesn’t Cause Diabetes, Obesity, or High BP

We controlled for calories. Same calorie band (1,600-2,400 per day). Then we compared people who eat a lot of sugar against people who eat almost none. Across all four ethnicities:

In every ethnic group, the high-sugar people had the same or better health markers than the low-sugar people.

For Mexican-Americans, the high-sugar group had an HbA1c 0.32 points LOWER (Cohen’s d = 0.24, a meaningful effect).

The data seems to suggest that sugar’s effect is almost entirely a proxy for total calorie intake and overall diet quality. Hold calories constant and sugar doesn’t predict anything.

I’m not saying go chug soda. But if your total calories are reasonable, obsessing over sugar grams is measuring the wrong thing. This connects to what I teach about gamification in healthcare: the systems we build to track health need to measure what actually matters.

---

3. No Dietary Variable Predicts Pre-Diabetes. Not One.

We looked at 5,208 American adults and asked a simple question: what distinguishes people with pre-diabetic blood sugar (HbA1c ≥ 5.7) from healthy people?

Sugar: d = 0.00. Zero.

Fiber: 0.01. Saturated fat: 0.01. Sodium: 0.01. The entire macro-tracking industry is built on signals that predict diabetes with the statistical power of random noise.

Age, waist size, and BMI. At least in this dataset, those are the only variables with meaningful predictive power.

---

4. Fiber’s “Miracle” Mostly Disappears When You Match Calories

This is the finding that surprised me most, because I expected fiber to be the standout variable.

Compare high-fiber (≥25g) vs. low-fiber (<10g) people without controlling for anything, and the high-fiber group looks dramatically healthier: lower BMI, lower blood sugar, lower blood pressure.

But high-fiber people also eat 1,100-1,800 MORE calories per day. They eat more of everything. They’re not healthier because of fiber. They’re different people with different diets entirely.

Match calories (2,000-2,800 band) and compare again:

HbA1c? Identical in both groups once calories are matched.

Fiber isn’t useless. A Cohen’s d of 0.20 for BMI in Whites is real, if small. But the data seems to suggest that much of fiber’s apparent benefit is a confounding variable. High-fiber eaters are different people with different overall diets.

---

5. Sleep’s Real Impact Is on Your Mind, Not Your Blood Sugar

NHANES includes a PHQ-9 depression questionnaire alongside the health data. When I merged it with the sleep data, this is the finding that surprised me most.

Short sleepers are 3x more likely to be clinically depressed (20.6% vs 6.3%). The Cohen’s d is 0.41, a medium-to-large effect. That makes short sleep the strongest lifestyle-related predictor of depression in the entire dataset, and the largest effect size I found for any single variable predicting any health outcome in this analysis.

Unlike diabetes, depression shows a full U-curve. Oversleeping (>8.5 hours) also increases depression risk to 9.0% (d = 0.23). For diabetes, oversleeping had zero effect. For your mind, both extremes are associated with problems.

The ethnic breakdown is where it gets really interesting:

White and Mexican Americans are the most vulnerable to sleep-related depression. Short sleep nearly triples their depression rates. Asian Americans are the most resilient: even with short sleep, only 8.2% are depressed, lower than any other group’s normal sleep rate. This connects to Finding 11: East Asian cultures sleep the least globally but seem to handle it better, both physically and mentally.

There’s another dimension to sleep’s impact that doesn’t show up in the NHANES cross-sectional data. An 8-week randomized controlled trial (Wang et al., 2018) put dieters into two groups: same calorie restriction, but one group slept an hour less per night.

Both groups lost the same 3.3 kg on the scale. But the well-rested group lost 83% fat and 17% muscle. The sleep-deprived group lost 58% fat and 39% muscle. Same weight loss, completely different body composition underneath. Sleep doesn’t change how much weight you lose. It changes what you lose. (I explored the extreme end of sleep manipulation in my Lich Pact experiment.)

One more finding from the depression data: income is the strongest predictor of depression (R² = 0.033), stronger than BMI, waist, sleep, or any dietary variable. For diabetes, waist circumference is king. For depression, it’s money. The mental health equivalent of “measure your waist” might be “check your bank account.”

The combination of short sleep and depression is particularly concerning: people with both have an HbA1c of 6.26 and BMI of 34.2. That’s 0.45 HbA1c points worse than the healthy-sleep, not-depressed baseline. Short sleep and depression appear to compound each other’s metabolic damage.

---

6. Sleep Hits Your Lungs and Joints, Not Your Liver

With the expanded NHANES dataset (now including lab work, self-reported conditions, and inflammation markers for ~8,000 adults), I could finally test sleep against everything. Not just diabetes and depression, but 16 different health outcomes.

Here’s what short sleep (under 6 hours) does compared to 7-8 hours, ranked by effect size:

The pattern is striking. Sleep seems to hit three systems hard: the brain (depression), the lungs and airways (COPD, asthma), and the inflammatory/autoimmune system (CRP, arthritis). It barely touches your liver, kidneys, thyroid, or uric acid levels.

This makes biological sense. Sleep is when the body regulates cortisol, inflammatory cytokines, and immune function. Short sleep disrupts the immune-inflammatory axis, which may explain why respiratory and joint conditions are so affected while organ-specific lab values aren’t.

The cardiovascular findings (stroke +4.6pp, heart attack +3.6pp) are notable too. These are low base-rate events, so a 4.6 percentage point increase in stroke prevalence among short sleepers is proportionally large (more than double).

---

7. It’s Not When You Sleep, It’s How Consistently

Here’s something I did not expect from the NHANES sleep data: what time you go to bed has almost zero predictive power for diabetes. The R² is 0.0002. People going to bed at 9 PM actually have worse HbA1c (5.89) than the midnight crowd (5.75). The most likely explanation? Early bedtimes skew older, and age is the strongest diabetes predictor we found.

But bedtime is 10x stronger than sleep duration for predicting depression (R² = 0.020 vs 0.002). That’s a small number in absolute terms, but the direction is clear: when you sleep matters more for your mood than how long you sleep.

The data seems to suggest the real mechanism is something researchers call “social jetlag,” which is the gap between your weekend and weekday bedtimes. Think of it as chronic jet lag you impose on yourself every Monday morning. Here’s what we found in the NHANES depression data:

People with more than 2 hours of social jetlag are 2.4x more likely to be depressed than people who keep a consistent schedule. That’s a clean dose-response curve, which is exactly what you want to see if a relationship is real rather than noise.

External data strengthens this pattern. A UK Biobank study (Knutson 2018, n=433,268) found that “definite evening types” had 10% higher mortality and 94% more psychological disorders compared to morning types, even with the same sleep duration. The problem wasn’t sleeping less. It was the chronic misalignment between their biological clocks and their social obligations (work schedules, school drop-offs, morning meetings).

A 2025 study (Wang et al., J Clin Sleep Med) of 18,129 adults found U-shaped mortality curves for both bedtime and wake time. The optimal window was roughly 11 PM to 6 AM. Going to bed before 10 PM or after midnight both carried higher mortality risk, as did waking before 5 AM or after 8 AM.

The practical takeaway from all of this: the data seems to suggest that consistency matters more than the specific hour. If you naturally go to bed at midnight and wake at 7:30 AM every day (including weekends), that appears to be better for your mental health than going to bed at 10 PM on weekdays and 1 AM on weekends. Your brain can adapt to almost any schedule. What it can’t handle is a schedule that keeps shifting.

---

8. Your Waist Predicts Gout Risk Better Than Your Diet

When we asked Grok Heavy (xAI’s reasoning model) to find the most surprising correlation in our dataset, it discovered something we’d missed: uric acid correlates with waist circumference at r = 0.338 — stronger than waist’s link to HbA1c (r = 0.232).

Most people blame gout on red meat, shellfish, and beer. The data says otherwise. Diet variables (sugar, protein, saturated fat, calories) show near-zero correlation with uric acid levels. Waist size predicts it 7x better than anything you eat.

People in the largest waist quartile have uric acid levels 1.31 mg/dL higher than the smallest quartile. That gap translates to dramatically higher lifetime risk of gout flares, kidney stones, and cardiovascular events.

The “but taller people just have bigger waists” objection doesn’t hold up. After statistically removing height’s effect, the waist-uric acid correlation barely budges (partial r = 0.308). Raw waist actually beats the height-adjusted waist-to-height ratio (r = 0.247) for predicting uric acid. The absolute measurement captures what matters: visceral fat.

In our Asian subgroup, the pattern is identical (r = 0.343). The 90 cm male / 80 cm female Asian thresholds aren’t just diabetes markers — they’re gout markers too.

---

How We Did This (For The Data Nerds)

Data sources (all publicly available; replicate at will):

• NHANES 2017-March 2020 (Pre-Pandemic): Individual-level health data on ~6,300 Americans, covering demographics, diet (24-hour recall), body measurements, blood pressure, HbA1c, and cholesterol. Download from the CDC NHANES page.
• Global Burden of Disease (GBD) via Our World in Data: Country-level dietary composition, life expectancy, and risk factors for 159 countries. Download at ourworldindata.org.
• GDP per capita (PPP): From Our World in Data (World Bank data). Download at ourworldindata.org/grapher/gdp-per-capita-worldbank.
• NCD-RisC: Country-level BMI and obesity trend data. Available at ncdrisc.org.

Tools used:

• NHANES Sleep Disorders Questionnaire (P_SLQ): Self-reported weekday and weekend sleep times for ~10,000 Americans, merged with health data above.
• Biochemistry Profile (NHANES P_BIOPRO): Lab data for 10,000+ Americans including kidney function (creatinine), liver enzymes (ALT, AST), uric acid, and more.
• Medical Conditions Questionnaire (P_MCQ): Self-reported cancer, heart attack, stroke, asthma, arthritis, COPD, liver disease, and thyroid conditions.
• Inflammation Marker (P_HSCRP): High-sensitivity C-reactive protein for systemic inflammation.
• Insulin + Glucose (P_INS, P_GLU): Used to compute HOMA-IR insulin resistance scores.
• Lipid Panel (P_TRIGLY, P_HDL): Triglycerides and HDL cholesterol for cardiovascular risk.
• NHANES PHQ-9 Depression Screener (P_DPQ): 9-item clinical depression questionnaire for ~9,000 Americans. Scores 0-27; 10+ = moderate depression.
• World Population Review / Sleep Cycle: App-based average sleep duration for 50 countries.
• World Bank / IDF Diabetes Atlas: Country-level diabetes prevalence (ages 20-79).
• PNAS 2025 (Heine et al.): Cross-national sleep and health study, 71 countries, 353 national sleep estimates.
• Three Claude Opus agents ($200/month each), which designed analysis strategies, wrote Python code, verified results, and caught errors
• SuperGrok Heavy ($300/month) for hypothesis generation and pattern scanning. We found its specific numbers couldn’t be trusted without verification.
• Perplexity Pro ($200/month) for research, finding datasets, and fact-checking claims
• Python for all final computation. Every number in this post was computed directly, not estimated by any AI. (I describe how I use AI agents as research tools on my projects page.)

Methodology:

• All effect sizes reported as Cohen’s d (0.2 = small, 0.5 = medium, 0.8 = large) or R²
• Where we controlled for variables (GDP, calories, income), we used matched-band analysis rather than regression residuals. Simpler, more transparent, harder to manipulate.
• We tested every finding across all 4 ethnicities in NHANES (Asian, White, Black, Mexican-American). If a finding only held for one group, we said so.
• Total compute cost: roughly $1,500 in AI subscriptions and ~8 hours of human oversight

What we DID NOT do:

• This is observational, cross-sectional data. We can’t prove causation.
• NHANES dietary data comes from a single 24-hour recall. People underreport.
• Country-level data can’t tell you about individuals (ecological fallacy).
• We lack exercise, smoking, and healthcare access data for the NHANES individuals. (Sleep data was available and is analyzed in Finding 11.)
• These findings generate hypotheses. They don’t settle debates. But they’re based on more data than most health advice on Instagram.

---

7 Actionable Lifestyle Lessons

What the data actually supports:

1. Measure your waist, not your macros. Your waist circumference predicts diabetes risk 12x better than everything in your food diary combined. For men: stay under 102 cm (90 cm if you’re Asian). For women: under 88 cm (80 cm if you’re Asian).

Or use the even simpler rule: keep your waist under half your height (waist-to-height ratio below 0.50). Measure monthly at the same spot and track the trend. This is the single most important health number you’re probably not tracking.

2. If you’re Asian, your “healthy” BMI threshold is 23, not 25. Nearly half of Asians at BMI 25 are already pre-diabetic. Get your HbA1c tested. Don’t rely on BMI alone.

3. Stop obsessing over sugar. When calories are held constant, sugar doesn’t predict BMI, diabetes risk, or blood pressure in any ethnic group. Focus on total calorie quality, not demonizing one nutrient.

4. Eat more vegetables. Way more. The countries that live longest eat 150-220 kg of vegetables per person per year. Americans eat 122 kg. This was the most consistent dietary signal across the entire 159-country dataset.

5. Calories don’t matter as much as you think (in wealthy countries). Korea eats 3,375 calories a day and lives to 84. The US eats 3,847 and lives to 77. Among rich countries, total calorie intake has zero correlation with life expectancy.

6. Eat diversely. Countries where no single food group exceeds 40% of calories consistently outlive countries with monotonous diets. Don’t let any one food group dominate your plate.

7. Don’t skip the blood work. BMI alone misses 68% of metabolically unhealthy thin Asians and falsely alarms perfectly healthy obese people. HbA1c, blood pressure, and cholesterol tell you what’s actually happening inside.

---

The Bottom Line

I wrote this because I couldn’t find anyone who actually ran the numbers instead of repeating what they heard from someone who heard it from a study they never read. Maybe I’m wrong about some of this. The data has real limitations that I tried to be transparent about. But at minimum, these numbers deserve to be part of the conversation.

If you want to crunch the data yourself, all the sources are linked above. I’d love to see what you find.

Next up: Part 2 — Gene Resilience: Why Your Ethnicity Changes the Health Rulebook. We look at how money fixes health for some races but not others, why 68% of “thin” Asian-Americans are already metabolically sick, and the ethnic vulnerability map that emerged from 8,000 medical exams.

— Yu-kai Chou

P.S. My current BMI is 29.8 and my waist is 100 cm. Ten centimeters over the Asian danger cutoff. Finding #1 was personal. I’m working on it.

WOULD YOU LIKE YU-KAI CHOU TO WORK WITH YOUR ORGANIZATION?