This retrospective, cross-sectional study used data from 6052 users who consented to participate between January 19 and 31, 2024. During this period, in-app messages were sent to all the users of the Asken app (Asken Inc), which is designed for dietary management and recording, requesting permission to use their stored data. The goal of the Asken app was to gather detailed information on dietary intake, enabling us to assess the participants’ nutritional habits.

At the same time, users who also used Pokémon Sleep (The Pokémon Company), a sleep-tracking app, were asked to provide their user IDs in their electronic informed consent. This linking of data was crucial for analyzing the relationship between the objective sleep data collected from the Pokémon Sleep app and the dietary data from the Asken app.

This study was conducted in accordance with the Declaration of Helsinki and approved by the Institutional Review Board of Sapporo Yurinokai Hospital, Japan (approval 030). Informed consent was obtained from all participants involved in the study. Participants used both apps concurrently from July 2023 to January 2024. Furthermore, data were extracted only from the overlapping period between the start and end dates of both apps, and average values were considered representative of each participant’s typical sleep habits and nutritional intake status. To ensure participant privacy and confidentiality, all data were deidentified before analysis, and no personally identifiable information was included in the dataset. The dataset was securely stored and accessible only to authorized researchers. Participants were not provided with monetary compensation for their participation in the study. However, they were fully informed about the study's purpose and significance, and their voluntary participation was obtained based on informed consent.

We calculated dietary intake patterns using the Asken smartphone app. Participants logged their meals by selecting food items and portion sizes from a database containing over 100,000 menu items. Participants entered the details of each meal into the app. For commercial products, the nutritional information was automatically recorded by scanning the barcode. The nutritional values were calculated based on the 2020 Standard Tables of Food Composition in Japan (eighth edition) [26]. The recording and calculation methods used by the participants are detailed in a previous study [12]. The Asken dietary recording app and traditional paper-based questionnaires have been reported to have a reasonably good correlation coefficient of 0.80 for nutrient intake [12,27,28].

In this retrospective, cross-sectional study, data were only included from days when participants reported consuming all three meals, as it was not possible to distinguish between missed recordings and actual skipped meals. Asken allows for the calculation of various macronutrients and dietary components, including total energy, protein, total fat (including saturated, monounsaturated, and polyunsaturated fats), carbohydrate, sodium, potassium, calcium, magnesium, phosphorus, iron, zinc, copper, manganese, vitamin A, vitamin D, vitamin E, vitamin K, vitamin B1, vitamin B2, niacin, vitamin B6, vitamin B12, folate, pantothenic acid, vitamin C, cholesterol, dietary fiber, and alcohol intake.

Based on previous research [29,30], we selected and analyzed the following components reportedly associated with sleep: total energy, protein, total fat (including saturated, monounsaturated, and polyunsaturated fats), carbohydrate, sodium, potassium, and dietary fiber intake, and calculated the sodium-to-potassium ratio. Macronutrients were energy adjusted by calculating the percentage of total energy intake. The absolute daily intake was calculated for other dietary components.

We calculated the TST, SL, and percentage of WASO (%WASO) derived from the Pokémon Sleep, which uses the Cole-Kripke algorithm for sleep-wake determination [31], as publicly disclosed [32]. Although accelerometer-based sleep determination integrated into smartphones lacks the ability to distinguish sleep stages [33,34], it provides a certain level of accuracy in determining sleep and wakefulness [22,23]. Although there may be a slight underestimation of SL, the overall TST and nocturnal awakenings showed good agreement with traditional actigraphy [34,35]. As previously mentioned, Natale et al [35] used the Cole-Kripke algorithm with smartphone accelerometers. Briefly, sleep-wake determination was performed using accelerometer data with 10-second epochs per minute. The decision was based on a 7-minute window of accelerometer data (from 4 min before to 3 min after the reference point [0 min]). A value greater than 1 indicates wakefulness and a value less than 1 indicates sleep [31]. SL was defined as the time from going to bed (start of the recording) to the first instance of being determined as sleep.

Descriptive analysis was conducted to summarize the basic characteristics of the participants, including sleep status and dietary nutrient intake. The multivariable regression analysis, isotemporal substitution model, and compositional data analysis were performed using the lmtest, deltacomp, Compositions, robCompositions, and ggplot2 R packages, respectively [25,36-38]. All regression models were adjusted for age, sex, and BMI.

First, multivariable regression analysis using the forced-entry method was used to determine the association among macronutrients, dietary components, and objective sleep variables (TST, SL, and %WASO). We conducted quartile categorization for total energy intake, energy-adjusted macronutrients (protein intake, total fat intake including saturated, monounsaturated, and polyunsaturated fats, and carbohydrate intake), sodium intake, potassium intake, dietary fiber intake, and sodium-to-potassium ratio. All the quartile range information for each variable is presented in Table S1 in Multimedia Appendix 1. Dummy variables were created for each nutrient based on the first quartile.

Second, to investigate the linear relationship between macronutrient intake and sleep parameters, we used an isotemporal substitution model and compositional data analysis [22-24]. In the isotemporal substitution model, when one macronutrient component is replaced with another on a one-to-one basis, the other components remain constant [22,23]. Compositional data analysis, in contrast, considers the overall composition and estimates the changes in sleep variables when the proportion of a specific macronutrient component is increased or decreased within the total composition [25]. Both methods have been used in studies that consider the interdependencies of factors contributing to a total composition of 100% (eg, dietary nutrient intake, 24-h physical/sedentary activity, and TST) [22,23,37,38].

We calculated the macronutrient intake (in grams) by multiplying each intake by its respective calorie contribution (ie, 4 kcal for carbohydrates and protein and 9 kcal for each type of fat), ensuring that the sum equals 1 (ie, total energy intake). Nutrient intake data were transformed into compositional data using an additive log-ratio transformation. Nutrient intake data were transformed into compositional data using an additive log-ratio transformation. This transformation was necessary to account for the relative proportions of macronutrients and facilitate subsequent analyses. The log transformation helps meet the assumptions required for compositional data analysis [25]. The variability in the data, in terms of the variability of each macronutrient intake relative to the variability of other components and the total variance of the whole composition, is described in Table S2 in Multimedia Appendix 1 through a variation matrix within each component. A value close to zero implied that the intake of the two macronutrient components was highly proportional.

The compositional data were then subjected to an isometric log-ratio (ilr) transformation to facilitate the application of the linear regression models. The ilr transformation was conducted using default and customized orthonormal bases for different nutrients. Linear regression models were constructed to evaluate the association between sleep parameters and the ilr-transformed nutrient composition after adjusting for sex, age, and BMI (Table S3 in Multimedia Appendix 1). The mean nutrient composition was calculated, and predictions for sleep parameters were made using a linear regression model. The compositional ratios of macronutrients among the participants are presented in Table S4 in Multimedia Appendix 1. Because of the isotemporal substitution model and compositional data analysis, it was not possible to replace values below the minimum composition. Therefore, in this study, we replaced values exceeding the upper limit of 6% (Table S4 in Multimedia Appendix 1).

To assess the impact of altering the nutrient composition, specific changes (eg, increasing monounsaturated fat and decreasing polyunsaturated fat) were applied, and new predictions were obtained. CIs for the differences in the predicted sleep parameters were computed using the residual standard error, model matrix, and critical value from the t-distribution. Hypothesis tests were conducted to examine the significance of the associations between individual nutrients and sleep parameters. Different ilr transformations were applied, each focusing on a specific nutrient (protein, carbohydrate, saturated fat, monounsaturated fat, or polyunsaturated fat), and the corresponding linear models were analyzed for significance.

Ramsey regression equation specification error test confirmed the linearity assumption for all models (TST, SL, and %WASO). Variance inflation factor values ranged from 1.10 to 4.65, indicating no multicollinearity. The Durbin-Watson test showed no autocorrelation in the residuals, and the Breusch-Pagan test confirmed homoscedasticity. Although the Shapiro-Wilk test showed significant P values, W-values between 0.996 and 0.997 indicated that the residuals were close to normally distributed.

All descriptive and statistical analyses were performed using R (version 4.4.0; R Foundation for Statistical Computing), with statistical significance assessed at the level of .05.

This study examined the relationship between nutritional intake and sleep using multivariate regression analysis, isotemporal substitution models, and compositional data analysis. Overall, the findings are consistent with those of previous research on individual nutrients and sleep [3-6,16-21]. As shown in Figure 2, greater total energy, fat, and sodium intakes were associated with shorter TST, whereas greater protein and dietary fiber intakes were linked to longer TST. Furthermore, the main findings of this study, based on compositional data analysis considering interdependencies, demonstrated that increasing protein while reducing other factors is associated with an elevated TST, whereas an increase in the polyunsaturated fat with a reduced TST (Table 2). Regarding SL and WASO, monounsaturated fat and polyunsaturated fat exhibited opposing relationships (Table 2). The differences in the compositional data analysis results, compared with individual macronutrients (eg, carbohydrates), likely reflect the strong interdependent relationships among various nutrients.

Dietary fiber intake was consistently associated with improved SL and %WASO. Additionally, greater sodium intake was correlated with shorter TST, whereas greater potassium intake was associated with shorter SL. Notably, polyunsaturated fats demonstrated potential benefits for SL and %WASO compared with other types of fat. Studies suggest that dietary fiber can affect sleep by influencing microbiota [39,40]. When dietary fibers are fermented by microbiota in the large intestine, they produce short-chain fatty acids, such as acetate, propionate, and butyrate, which enhance serotonin release. Furthermore, administering tributyrin, a compound made of three molecules of butyric acid and glycerol, to mice resulted in a 50% increase in nonrapid eye movement sleep for 4 h [40]. The results of this study support those of previous studies [39,40].

Previous studies have established that greater total energy intake and poor dietary quality are associated with lower subjective sleep quality [7,41]. Specifically, higher protein intake correlates with improved sleep, whereas a higher intake of fatty acids is linked to worse sleep [41]. This study also found variations in the effects of lipid intake on sleep, depending on the type (eg, saturated, monounsaturated, or polyunsaturated fats). The mechanism by which proteins affect sleep following acute feeding may involve tryptophan; tyrosine; and the synthesis of brain neurotransmitters, such as serotonin, melatonin, and dopamine [42]. Although no significant differences were observed in SL or %WASO, a greater protein intake was positively associated with longer TST.

However, the specific mechanisms underlying the relationship between monounsaturated fat, polyunsaturated fat, and sleep remain unclear. A meta-analysis found an association between a greater intake of hexadecenoic acid, a saturated fatty acid, and subjective difficulty falling asleep [43]. Conversely, difficulties falling asleep were linked to a lower intake of dodecanoic acid, a monounsaturated fat, and butanoic acid, a saturated fatty acid [43]. Additionally, a reduced intake of dodecanoic acid was associated with subjective difficulties in maintaining sleep [43]. An epidemiological study involving adults evaluated the relationship between two polyunsaturated fatty acids, ω-6 and ω-3, and found that ω-6 fatty acids were related to the risk of sleep disorder, based on self-reported data, in an inverse U-shaped manner [19]. The ω-6:ω-3 ratio was positively and linearly associated with the risk of sleep disorders [19]. The underlying mechanisms involving ω-3 fatty acids and sleep are discussed, noting that a deficiency in ω-3 intake may affect cortical neuron oscillatory activity and sleep-wake patterns [20].

Additionally, the timing of fat intake may influence sleep; one observational study based on polysomnography suggested that consuming high-fat foods close to bedtime may be associated with increased WASO [44]. Continued consumption of a high-fat diet has been partially understood to reduce physical activity and increase rapid eye movements and sleep fragmentation through dopaminergic dysregulation [45]. Research involving mouse models and pregnant women has suggested a potential positive association between monounsaturated fat intake and sleep [46,47]. However, other studies have indicated a negative association between monounsaturated fat and sleep, whereas polyunsaturated fat has shown a positive association with sleep [48-50]. Previous studies have often focused on dietary fiber levels and comparisons between high- and low-saturated-fat diets [45,48]. Future studies should explore the impact of unsaturated fats, including monounsaturated and polyunsaturated fats.

Sodium and potassium have been reported to impact health individually [51,52], and a high sodium-to-potassium ratio (indicating relatively high sodium intake) has been associated with higher all-cause mortality [53]. Sodium intake is known to be correlated with sleep, indicating that greater sodium intake is associated with shorter subjective TST and worse sleep quality [30,50]. This relationship is attributable to the potential effect of sodium on nocturnal urination [51,52]. However, in this study, a greater intake of sodium alone was associated with a shorter TST, although no significant differences were observed in %WASO. Moreover, greater sodium intake was associated with shorter SL. Greater potassium intake was associated with shorter SL and reduced %WASO. All sleep variables were adversely related when considering the balance between sodium and potassium, which indicates an unhealthy dietary pattern (ie, relatively greater sodium intake). In this study, sodium intake was used as a basis for analysis, while the sodium-to-potassium ratio was also evaluated (Figure 2). Therefore, the results linking sleep outcomes with the sodium-to-potassium ratio are considered reasonable, as they provide a more comprehensive perspective on the impact of both minerals on sleep.

Recent studies suggested that changes in the gut microbiota may influence sleep and nutrient absorption [39,54,55]. Disrupted gut flora in patients with celiac disease has been linked to sleep disturbances [56]. Future research is warranted especially focusing on patients with celiac disease. Additionally, digital health tools offer the potential for health monitoring and early diagnosis [57].

In this study, we investigated the association between daily nutrient intake and sleep using isotemporal substitution models and compositional data analyses to account for the interdependencies [25]. These approaches enabled us to assess how substituting macronutrients and considering the overall composition ratio affect sleep outcomes. Nonetheless, this study had several limitations. First, this cross-sectional study does not provide causal explanations. As noted above, previous studies have reported bidirectional relationships between these factors [58]. Additionally, the isotemporal substitution model and compositional data analysis are merely statistical inferences [25]. The findings of this study underscore the need for future interventions and longitudinal studies to establish causal effects definitively.

Second, this study may have included a highly health-conscious group that uses both health management apps simultaneously. Therefore, although values around the second and third quartiles, as shown in Table S1 in Multimedia Appendix 1, correspond to Japanese macronutrient standards [59] and suggest the possibility of representing the general population, caution is needed when generalizing the findings to broader populations, owing to the inclusion of this health-conscious group.

Third, the sleep assessment method used in this study involved applying gamified elements. Some users may have exhibited sleep patterns different from their habitual patterns to achieve specific goals within the game. Moreover, because users have their own smartphones, the reliability of accelerometer measurements varies across different device models. In the future, integrating alternative methods, such as tappigraphy for sleep-wake rhythms, actigraphy, or in-home electroencephalogram devices could provide more objective evaluations [13-15].

Fourth, retrospective studies like this one are inherently prone to recall bias, as participants may inaccurately report their dietary intake and sleep patterns based on memory. It should also be noted that dietary intake was not cross-checked, and this study did not separately account for weekdays and weekends in assessing overall nutrient intake. Finally, because this study was retrospective, we could not adequately capture confounding factors related to sleep and nutrition. We collected basic demographic information, such as age, sex, height, weight, and physical activity level. However, other confounding factors, such as alcohol and smoking habits, employment status (including shifts and night shifts), marital status, cohabitation status, medical history, and medication use, were not considered. Therefore, since we could not fully eliminate concerns regarding internal validity, the presence of selection bias cannot be ruled out. Although this is an observational study with a large sample size of real-world data, it is essential to exercise caution in interpreting the results owing to these limitations.

We found that a greater intake of polyunsaturated fats was associated with better SL and reduced WASO, while saturated and monounsaturated fats were correlated with longer SL and wakefulness. Dietary fiber intake was positively associated with improved sleep quality, and a greater sodium-to-potassium ratio was linked to worse sleep outcomes. Furthermore, our compositional data analysis, which accounted for the interdependencies between nutrients, suggested that increasing the protein intake in place of other nutrients could improve the TST, and similar effects were observed for polyunsaturated fats on SL and WASO. Although this study is cross-sectional, the results highlight the complex potential role of dietary factors in sleep regulation and suggest the possibility of dietary interventions to enhance sleep health.