4.1. Main Findings
The current study addressed a major gap in the research literature on sleep and nutrition, by investigating the relationship between sleep duration and overall dietary patterns in a nationally representative sample of UK adults. Addressing this research gap was important, since previous research has focused on describing the relationship between sleep duration and individual nutrients or foods [7
]. Such a reductionist approach often ignores complex interconnections between the various food groups, thereby potentially undermining more relevant associations and rendering a public health campaign centered on sleep and diet challenging. In the present study, we identified an inverse relationship between long sleep duration on weekdays and a healthy dietary pattern characterized predominantly by higher intakes of ‘fruit’, ‘salad and other raw vegetables’, ‘tea, coffee, and water’, ‘vegetables not raw’, ‘oily fish’, ‘high-fiber breakfast’, and ‘nuts and seeds’. This association remained significant even after adjustment for relevant confounders including BMI, mean daily energy intake, and sleep duration on weekends. In terms of public health implications, our findings may have important value, given that a decline of 0.45 units in the healthy dietary pattern score can be translated as a mean daily increase of approximately 500 kcal in energy intake, accompanied by a 22 g increase in the intake of ‘chips, fried, roasted potatoes, and potatoes’ at the expense of a 2-portion reduction in daily fruit intake.
Our findings contrasted with a previous analysis of the NDNS RP, wherein sleep duration was not found to be related to diet [9
]. This may be ascribed to the single-nutrient/food approach, adopted by the previous analysis [9
]. Our study, however, complements the observations made by Dashti who found that normal sleep duration was associated with more favorable dietary behavior [25
]. Our findings were also consistent with previous observations made by Mossavar-Rahmani and colleagues, who found that long sleepers had lower intake of caffeine in a sample of 16,415 Hispanic/Latino participants living in the US [26
]. Similar observations have been made by Grandner, who reported lower intake of theobromine, a metabolite of caffeine, and marker of tea and coffee intake in long sleepers [27
]. Contrary to Grandner and colleagues [27
], however, we did not observe a u-shaped relationship between sleep duration and energy intake in the univariate analysis. Similarly, we did not observe an inverse association between short sleep duration and the healthy dietary pattern. This contrasts with previous research, which has demonstrated associations between short sleep duration and reduced fruit [28
] and/or vegetable intake [28
]. In the Shanghai Women′s Health Study, which included 68,832 Chinese women, an inverse association between short sleep duration (<6 h) and tea and fruit intake was also found [30
]. In our study, the lack of findings of an association between short sleep duration and dietary patterns may be explained by differences in the cut-offs used for defining the sleep duration tertiles, as well as the lack of objective measures of sleep. In the former case, redefining the tertiles using 6 h or 9 h as the cut-off for the lower or upper tertile, respectively, did not alter the findings. Concerning measures of sleep, we were unable to differentiate between short sleep duration associated with circadian misalignment and sleep disturbance, versus short sleep that is not associated with circadian misalignment and sleep disturbance. The latter is important, considering that the combination of short sleep and circadian misalignment is associated with more adverse health outcomes [31
]. Moreover, growing evidence suggests that sleep duration, sleep quality, and social jetlag may be modulated by an individual’s circadian type or so-called chronotype [32
]. Accordingly, in a study of obese short sleepers, Lucassen and colleagues found that only individuals with an evening chronotype were more likely to have an unhealthy eating pattern characterized by larger food portions, less frequent meals, and greater energy intake later in the day [33
]. This may imply the need for controlling for chronotype in nutrition epidemiological studies investigating the association between sleep duration and diet.
In the present study, no significant association between shorter sleep duration and the ‘snacks’ dietary pattern was observed. This was in contrast to intervention studies, which have reported increased intake of snacks [34
], particularly carbohydrate-rich snacks and dessert, following sleep restriction [35
]. This was also in contrast to epidemiological studies, which have described a positive association between short sleep duration and snack intake after dinner [1
]. Future analyses of the NDNS RP should attempt to examine differences in food intake, according to time-of-day, using methods such as correspondence analysis [38
]. Differences in eating occasion-based dietary patterns may likewise warrant exploration [21
A novel finding of the current study was the inverted u-shape relationship between social jetlag, defined in the present study as the difference between sleep duration on weekends and weekdays, and the health dietary pattern. We found that beyond a 1 h 45 min positive social jetlag on weekends, scores for the healthy dietary pattern declined. These findings were interesting as they, albeit speculatively, indicated that sleep compensation on weekends was associated with an improved diet, up to a certain threshold. Although we cannot speculate on the nature of this relationship, this analysis was further supported by the sensitivity analyses we conducted, wherein the negative relationship between long sleep duration on weekdays and the health dietary pattern was attenuated, when sleep duration on weekends was not accounted for. Such findings suggest the importance of collecting data on sleep duration on weekends, to capture both a negative and a positive social jetlag. Moreover, our findings highlighted the need to characterize potential factors influencing social jetlag, to gain an understanding of the characteristics of the individuals who exhibit a long positive social jetlag. In particular, work patterns; mental illness; and psychosocial aspects such as stress, fatigue, and mood, as potential factors underlying social jetlag warrant investigation.
Although not the primary objective of our analyses, differences in sleep duration according to BMI were noted. In particular, individuals with a lower BMI were more likely to be in the highest tertile of sleep duration. These findings were in line with the observations made previously, in a cross-sectional analysis of NHANES 2007–2008 [27
]. They were also consistent with an earlier analysis of the NDNS RP [9
], wherein Potter and colleagues found that for every additional hour of sleep, there was a 0.46 kg/m2
and 0.9 cm reduction in BMI and waist circumference, respectively. However, in our study, we observed that despite the inverse association between sleep duration and BMI in the univariate models, long sleep duration on weekdays was associated with a lower healthy dietary pattern score, compared to normal sleep duration.
To date, inconsistencies remain about the best categorization of sleep duration. These discrepancies have been highlighted by a recent review by Al-Khatib and colleagues [4
]. The current National Sleep Foundation defines normal sleep for adults as a range of 7–9 h, while sleep duration equal or below 6 h is deemed to be short sleep, and 10 h and above is defined as long sleep [39
]. Other studies utilize 9 h as the cut-off, for defining long sleep duration [1
]. In the present study, we assessed model fit based on multiple approaches including categorizing sleep duration into tertiles and use of a quadratic term. We found that the model that included tertiles of sleep duration on weekdays vs. weekends provided the best fit when examining the association between sleep duration and diet. This was despite the unequal sample distribution within the tertiles, which arose as a result of the Stata xtile command being confronted with ties [40
]. Redefining the lowest and highest tertile using 6 h and 9 h, respectively, as cut-offs did not improve the model fit or change the value of the coefficient. In contrast, when analyzing social jetlag data, categorizing the social jetlag variable into tertiles did not provide the best model fit. Moreover, the categorization was difficult to interpret, as it was unclear as to what constitutes a normal range of social jetlag. Consequently, the final model with the quadratic term was selected, as it provided a better fit and interpretation. The implication of using the different methodologies to define sleep duration or social jetlag warrant further investigation. Such research may be important in understanding potential differences in the definition of short, normal, and long sleep duration or social jetlag in different global regions.
4.2. Strengths and Weaknesses of the Study
A main strength of the current study is the national representativeness of the survey sample, and the use of detailed dietary data, which was collected over 3–4 days. Data on sleep duration on weekends were also available, which allowed us to adjust for sleep duration on weekends and explore how sleep compensation on weekends may confound or modify the association between sleep during weekdays, and overall diet. Previously such analysis had not been possible, as most nutritional surveys did not collect data on sleep during weekends [1
]. This is an important strength, as research has shown that the relationship between sleep duration and diet may differ when considering sleep duration on weekends [27
]. A further strength of the present study was the use of multiple imputation to account for missing data, compared to previous studies, which focused only on individuals with complete data [9
]. Multiple imputation has the advantage of providing a narrower confidence interval, and reducing bias associated with the use of a selective sample [41
Concerning limitations; one of the main limitations of NDNS RP is the cross-sectional design, which does not permit investigating the direction of causality in the association between sleep and dietary patterns. Indeed, whilst sleep duration has been shown to influence eating behavior and nutritional status in a number of experimental studies, certain foods or nutrients may equally impact sleep duration and quality, as reviewed by Dashti [7
] and Pot [42
]. In addition, the NDNS RP included data from all participants, regardless as to whether the dietary data collected during the three to four days of food recordings, was reflective of habitual intake [43
]. This aspect of the study design meant that differentiating between under-reporters and under-consumers (for instance due to ill health), may not be possible [43
]. Nevertheless, mis-reporting is a common limitation of dietary surveys and cohorts. A further limitation was the absence of data on physical activity, shift work, timing of sleep onset, or sleep quality in the adult population. Finally, in the NDNS RP, data on sleep duration were self-reported, and no objective measures on sleep duration and quality were available in the adult population. Hence, findings from our study may differ from studies that utilize more objective measures of sleep duration [27