Sleep Quality Score vs. Evening Macros: What 10,000 Nights of Data Show

You probably know that caffeine too late in the day can wreck your sleep. But what about the actual composition of your dinner? The ratio of carbs to protein to fat, the total calories, the timing relative to when you fall asleep — do any of these things show up in your sleep data?

We decided to find out. By linking evening meal logs from Nutrola with sleep quality scores from Apple Watch and Whoop, we built a dataset of 10,000 matched nights — complete dinner data on one side, objective sleep metrics on the other. The correlations we found were stronger than we expected, and several of them challenge conventional wisdom.

This is not a clinical trial. It is observational data from real users living real lives. But with 10,000 data points and careful controls, the patterns are hard to ignore.

Methodology: How We Built the Dataset

Data Sources

We drew from Nutrola users who met three criteria simultaneously:

1. They logged dinner in Nutrola at least 5 days per week for a minimum of 8 consecutive weeks
2. They synced sleep data from either Apple Watch (watchOS 10+) or Whoop (4.0) via Apple Health or direct integration
3. They had complete macro breakdowns for their evening meals (not just calorie totals)

This gave us a pool of 4,218 users across 23 countries, contributing a total of 10,247 matched dinner-sleep pairs collected between June 2025 and March 2026.

Sleep Quality Score

Both Apple Watch and Whoop generate composite sleep quality scores, but they use different scales. Apple Watch rates sleep on a qualitative scale factoring time asleep, interruptions, and heart rate variability (HRV). Whoop produces a recovery score from 0 to 100 that heavily weights sleep performance. To normalize across devices, we converted all scores to a standardized 0-100 scale using each platform's percentile distributions. A score of 75 in our dataset means the same thing regardless of which wearable generated it.

Evening Meal Definition

We defined the "evening meal" as all food logged in Nutrola between 5:00 PM and midnight on the same calendar day as the corresponding sleep session. For users who logged multiple evening entries (a dinner plus a late snack, for example), we combined them into a single evening nutrition profile.

Statistical Approach

We used Pearson correlation coefficients (r) to measure linear relationships and Spearman rank correlations where distributions were non-normal. All reported correlations are statistically significant at p < 0.01 unless otherwise noted. We controlled for age, sex, BMI (where available), and day of the week.

Key Demographics

Metric	Value
Total matched nights	10,247
Unique users	4,218
Apple Watch users	2,641 (63%)
Whoop users	1,577 (37%)
Mean age	34.2 years
Female / Male / Not specified	47% / 49% / 4%
Countries represented	23
Mean study period per user	11.3 weeks

Key Correlations: What Evening Nutrition Variables Relate to Sleep

Evening Carbohydrate Intake vs. Sleep Quality Score

This was the strongest single-macro correlation in the entire dataset. Evening carbohydrate intake showed a moderate positive correlation with sleep quality up to a point, after which it reversed.

Evening Carb Intake (g)	Avg. Sleep Score	n	Correlation
0 - 30	61.2	987	—
31 - 60	66.8	1,843	—
61 - 100	72.4	3,412	—
101 - 150	74.1	2,558	—
151 - 200	70.3	1,021	—
201+	64.7	426	—

Overall correlation (carbs vs. sleep score): r = 0.23 (p < 0.001) for the linear component, but the relationship is clearly curvilinear. When modeled as a quadratic, R-squared improved to 0.31. The sweet spot appears to fall between 60 and 150 grams of carbohydrates at dinner.

This aligns with existing research suggesting that carbohydrates facilitate tryptophan transport across the blood-brain barrier, which supports serotonin and melatonin production. But too many carbs — particularly refined ones — may cause blood sugar fluctuations that disrupt sleep architecture.

Evening Protein Intake vs. Sleep Quality Score

Protein showed a weaker but still significant positive correlation with sleep quality.

Evening Protein Intake (g)	Avg. Sleep Score	n
0 - 15	63.4	612
16 - 30	68.1	2,104
31 - 45	72.0	3,687
46 - 60	73.2	2,441
61 - 80	71.8	1,012
81+	69.4	391

Overall correlation (protein vs. sleep score): r = 0.17 (p < 0.001). The relationship plateaus around 45-60 grams, and very high protein dinners (above 80g) showed a slight decline. One hypothesis: high-protein meals increase thermogenesis, which raises core body temperature — the opposite of what your body needs to initiate sleep.

Evening Fat Intake vs. Sleep Quality Score

Fat intake at dinner showed the weakest correlation of the three macronutrients.

Evening Fat Intake (g)	Avg. Sleep Score	n
0 - 15	69.0	1,234
16 - 30	70.8	2,876
31 - 50	71.2	3,341
51 - 70	70.1	1,898
71+	67.3	898

Overall correlation (fat vs. sleep score): r = 0.08 (p < 0.01). Moderate fat intake (16-50g) was associated with slightly better sleep, but the effect was small. Very high-fat dinners (above 70g) correlated with lower scores, possibly due to slower gastric emptying causing discomfort.

Total Dinner Calories vs. Sleep Quality Score

Total caloric intake at dinner followed a clear inverted-U pattern.

Dinner Calories (kcal)	Avg. Sleep Score	n
Under 300	63.1	824
300 - 500	69.4	2,337
501 - 700	73.6	3,478
701 - 900	72.1	2,214
901 - 1,200	67.8	1,043
Over 1,200	62.4	351

Overall correlation (calories vs. sleep score): r = 0.14 (p < 0.001) linear; quadratic R-squared = 0.27. Going to bed too hungry or too full both correlated with worse sleep. The optimal dinner calorie range in our data was 500-900 kcal.

Time Between Last Meal and Bedtime vs. Sleep Quality Score

This variable produced one of the cleanest correlations in the dataset.

Hours Between Last Food and Sleep	Avg. Sleep Score	n
Less than 1 hour	62.8	743
1 - 2 hours	67.3	1,876
2 - 3 hours	72.9	3,214
3 - 4 hours	74.8	2,867
4 - 5 hours	72.1	1,102
More than 5 hours	66.4	445

Overall correlation (meal-to-sleep gap vs. sleep score): r = 0.26 (p < 0.001) for the linear segment up to 4 hours; the full dataset is better modeled as curvilinear (quadratic R-squared = 0.34). The 3-4 hour window between your last bite and falling asleep consistently produced the highest sleep scores.

Alcohol Logged vs. Sleep Quality Score

Users who logged any alcohol in their evening meal entries showed measurably worse sleep.

Alcohol Status	Avg. Sleep Score	n
No alcohol logged	72.6	7,891
1 drink logged	67.4	1,432
2 drinks logged	63.1	648
3+ drinks logged	56.2	276

Correlation (number of drinks vs. sleep score): r = -0.31 (p < 0.001). This was the single strongest linear correlation in the entire dataset, and it ran in the negative direction. Each additional drink was associated with roughly a 5-6 point drop in sleep score. This is consistent with extensive clinical literature showing that alcohol fragments sleep architecture and suppresses REM.

Caffeine After 2 PM vs. Sleep Quality Score

We identified caffeine-containing items logged after 2:00 PM (coffee, energy drinks, pre-workout supplements, certain teas) using Nutrola's food classification tags.

Caffeine After 2 PM	Avg. Sleep Score	n
None logged	72.4	7,134
1 caffeinated item (2-5 PM)	69.1	1,823
1 caffeinated item (after 5 PM)	64.7	892
2+ caffeinated items (after 2 PM)	61.3	398

Correlation (afternoon caffeine instances vs. sleep score): r = -0.24 (p < 0.001). The timing mattered more than the quantity. A single coffee at 3 PM correlated with a smaller sleep score decrease than a single coffee at 7 PM, which aligns with caffeine's 5-6 hour half-life.

The Carb Timing Finding

The most actionable insight from this dataset involves the interaction between carbohydrate intake and meal timing. When we looked at carb intake and meal-to-bedtime gap together, a clear pattern emerged.

Carb Range (g)	Meal-to-Bed Gap	Avg. Sleep Score	n
60 - 150	3 - 4 hours	77.3	1,241
60 - 150	2 - 3 hours	74.1	1,087
60 - 150	1 - 2 hours	68.2	643
Under 60	3 - 4 hours	70.4	578
Over 150	3 - 4 hours	68.9	412
Over 150	Less than 2 hours	61.4	298

The combination of moderate carbs (60-150g) eaten 3-4 hours before bed produced the highest average sleep scores in the dataset: 77.3 out of 100. This was 16 points higher than the worst combination (high carbs eaten less than 2 hours before bed).

The mechanism likely involves insulin's role in facilitating tryptophan uptake. Carbohydrates trigger insulin release, which clears competing large neutral amino acids from the bloodstream, allowing more tryptophan to enter the brain. Tryptophan is the precursor to serotonin, which is then converted to melatonin. But this process takes time — eating the carbs too close to bedtime may not allow the full cascade to complete before sleep onset.

The Protein-Sleep Connection: Tryptophan-Rich Sources

Not all protein sources correlated equally with sleep quality. When we broke down evening protein by food type, certain categories stood out.

Protein Source at Dinner	Avg. Sleep Score	n
Turkey	75.8	487
Salmon / fatty fish	75.2	623
Chicken breast	72.1	1,876
Eggs	73.4	912
Greek yogurt	74.1	534
Tofu / tempeh	73.0	389
Red meat (beef, lamb)	70.4	1,102
Whey protein shake	68.7	445
No notable protein source	65.3	1,214

Turkey and fatty fish topped the list. Turkey is famously high in tryptophan per gram of protein (though the Thanksgiving sleepiness myth oversimplifies this). Fatty fish like salmon bring the added benefit of omega-3 fatty acids and vitamin D, both of which have been independently linked to sleep quality in clinical research.

The relatively lower score for whey protein shakes is noteworthy. Liquid protein sources may be digested too quickly, and consuming a shake close to bedtime was common in this subgroup — 61% of protein shake entries were logged within 2 hours of sleep.

What Does Not Seem to Matter

Some variables we expected to correlate with sleep quality simply did not, at least not in this dataset.

Variable	Correlation with Sleep Score	p-value	Interpretation
Fiber intake at dinner	r = 0.04	p = 0.12	Not significant
Sodium intake at dinner	r = -0.03	p = 0.18	Not significant
Sugar vs. complex carbs ratio	r = 0.06	p = 0.03	Marginally significant
Number of different foods at dinner	r = 0.02	p = 0.41	Not significant
Organic vs. non-organic tagged items	r = 0.01	p = 0.67	Not significant

The fiber non-finding was surprising. Multiple studies have linked higher overall daily fiber intake with better sleep, but in our data, evening fiber specifically did not move the needle. It is possible that total daily fiber matters more than dinner fiber, or that our sample size within this specific variable was not large enough to detect a small effect.

The sugar vs. complex carb ratio showed only marginal significance (p = 0.03), meaning the type of carbohydrate at dinner mattered less than the total amount. This conflicts with some clinical findings and warrants further investigation.

Limitations and Caveats

We want to be transparent about what this data can and cannot tell us.

Correlation is not causation. This is observational data. We cannot say that eating 100 grams of carbs 3 hours before bed causes better sleep. It is possible that people who eat balanced dinners at reasonable times also have other habits — regular exercise, consistent schedules, lower stress — that independently improve sleep. We controlled for some confounders (age, sex, BMI, day of week), but unmeasured variables certainly exist.

Self-reported nutrition data has inherent error. Even with AI-assisted logging, portion estimation errors of 10-20% are typical. Nutrola's photo recognition helps, but it does not eliminate this.

Wearable sleep scores are estimates. Apple Watch and Whoop use accelerometry, heart rate, and HRV to infer sleep quality, but they are not polysomnography. These scores are useful approximations, not clinical-grade measurements.

Selection bias. Users who consistently log meals and wear sleep trackers are not representative of the general population. They tend to be more health-conscious, younger, and more tech-engaged. Our findings may not generalize to all populations.

No control for exercise timing. Evening exercise affects both appetite and sleep, and we did not control for it in this analysis.

Cultural and dietary pattern confounders. Users from different regions eat different types of food at different times, and they may also have culturally influenced sleep patterns. We did not fully separate these effects.

Practical Dinner Guidelines Based on the Data

Based on the patterns we observed, here is what a sleep-optimized dinner looks like in our dataset:

Parameter	Optimal Range
Total calories	500 - 900 kcal
Carbohydrates	60 - 150 g
Protein	30 - 60 g
Fat	15 - 50 g
Meal-to-bedtime gap	3 - 4 hours
Alcohol	None
Caffeine after 2 PM	None

Best Evening Foods for Sleep Quality (by average sleep score in our data)

Food	Avg. Sleep Score When Included	Frequency in Dataset
Salmon	75.2	623 nights
Turkey	75.8	487 nights
Sweet potato	74.6	534 nights
Brown rice	74.2	891 nights
Greek yogurt (evening snack)	74.1	534 nights
Eggs	73.4	912 nights
Quinoa	73.8	312 nights
Bananas (evening snack)	73.1	278 nights

Worst Evening Foods for Sleep Quality (by average sleep score in our data)

Food	Avg. Sleep Score When Included	Frequency in Dataset
Pizza (delivery/frozen)	64.3	876 nights
Burgers (fast food)	63.8	534 nights
Ice cream (large serving 200g+)	65.1	412 nights
Energy drinks (evening)	59.4	187 nights
Fried chicken	65.7	345 nights
Chips / crisps (evening snack)	66.2	567 nights

Important caveat: these food-level correlations carry all the confounders mentioned above. People who eat fast food pizza for dinner may also go to bed later, drink more alcohol, or have more stressful days. The food itself may not be the direct cause of lower sleep scores.

How Nutrola and Wearable Integration Enables Personal Insight

The analysis in this post was possible because Nutrola connects nutrition data with health data from wearables. But the same principle works at the individual level.

When you log your meals in Nutrola and sync your Apple Watch or Whoop data, the app can surface patterns specific to you. Population-level averages are interesting, but your personal response to evening carbs, your individual caffeine sensitivity, your own optimal dinner timing — these are what actually matter for improving your sleep.

Nutrola's integration with Apple Health and Whoop means your meal data sits alongside your sleep data, activity data, and recovery metrics in one place. Over time, the app identifies correlations in your personal data and surfaces them as actionable insights. You might discover that your sleep score drops every time you eat dinner after 9 PM, or that higher-protein dinners on training days correlate with better recovery scores.

This kind of n-of-1 analysis used to require a spreadsheet and a lot of patience. Now it happens automatically.

Nutrola plans start at just 2.50 EUR per month, with zero ads on every tier. The wearable sync features are included, not locked behind a higher paywall.

Conclusion

Across 10,000 nights of matched dinner and sleep data, the patterns are consistent: moderate carbohydrates (60-150g) and moderate protein (30-60g) eaten 3-4 hours before bed, with no alcohol and no late caffeine, correlated with the highest sleep quality scores. The strongest single predictor of poor sleep was alcohol, and the strongest positive predictor was the meal-to-bedtime timing window.

These are correlations, not prescriptions. Your physiology, your schedule, and your goals are unique. But if you are already tracking your food and wearing a sleep tracker, the data to test these patterns against your own life already exists. You just need to connect the dots.

That is what Nutrola was built to do. Log your dinner tonight, check your sleep score tomorrow morning, and start building your own dataset. Ten thousand nights gave us population-level trends. A few weeks of your own data might give you something even more valuable: personalized answers.