The Dutch HbA1c Lifestyle Study (DAF-Study): Seasonal Variation of HbA1c in the Dutch Diabetes Population—Associations with Macronutrient Intake and Physical Activity

Abstract

Background/Objectives: Seasonal variation in hemoglobin A1c (HbA1c) values has been previously documented, with physical activity (PA) and macronutrient intake (MNI) suggested as potential drivers. This study combines seasonal mean HbA1c values from the Dutch (pre)diabetes population with national survey data on PA and MNI from 2018 to 2021 to identify key associations. Methods: HbA1c data were collected from 24 laboratory organizations in the Netherlands and Dutch Caribbean. MNI and total energy intake data were extracted from the Dutch National Food Consumption Survey, while PA data came from the Dutch National Sports Participation Index Survey. Weighting factors were applied to align PA and MNI data with HbA1c data. Seasonal averages were analyzed for significant differences, and a prediction model compared PA and MNI with actual HbA1c values. Results: Among 5,635,920 HbA1c results, the average HbA1c increased by 0.71 mmol/mol (NGSP 0.06%) over four years, with an overall mean of 52.4 mmol/mol (NGSP 7.0%). Seasonal HbA1c variation showed a dip in summer–autumn and a peak in winter–spring (1.2 mmol/mol; NGSP 0.11%; p < 0.0001). MNI, except for total energy intake (which peaked in summer; p < 0.001), showed no significant trends or association with HbA1c (p = 0.157). PA decreased by 7.2% over the study period, with seasonal peaks in summer–autumn, showing an inverse relationship with HbA1c (p < 0.0001). During the COVID-19 lockdowns, PA significantly decreased, and mean HbA1c values increased more markedly than in previous years. The prediction model confirmed PA as a significant driver of seasonal HbA1c variation (p = 0.004). Conclusions: These findings suggest that PA is the strongest driver of seasonal variation in HbA1c. Public health initiatives and support programs promoting physical activity are essential for improving HbA1c regulation.

1. Introduction

Diabetes mellitus (DM) is one of the most common chronic diseases, with a rising global prevalence [1]. Hallmark studies from the Diabetes Control and Complications Trail Group and UK Prospective Diabetes Study show that adequate glycemic control is essential for preventing or delaying both acute and chronic complications of diabetes [2,3]. Much more insight has been gained on diabetes prevention and control since then, and is now included in guidelines and clinical recommendations from the ADA and EASD [4,5]. Hemoglobin A1c (HbA1c) is a key biomarker that reflects average blood glucose levels over the preceding two months and is therefore used to assess long-term glucose control and predict diabetes-related complications.

Successful standardization [6], combined with increasingly accurate and precise measurement techniques, has enabled the use of HbA1c not only as a prognostic indicator but also as a diagnostic parameter for DM [7]. In the late 1980s, the first reports emerged describing seasonal variability in HbA1c levels, typically showing a circannual pattern with higher values in winter and lower values in summer [8,9,10,11,12,13]. This pattern is observed both above and below the equator, with greater peak-to-trough amplitudes at higher latitudes. In contrast, smaller or negligible differences are seen in equatorial regions [14,15].

Key factors contributing to this variation include physical activity (PA) and macronutrient intake (MNI; carbohydrate, fat, and protein), both of which can vary seasonally [16]. Carbohydrate intake, in particular, has a direct and well-established impact on HbA1c levels [17,18]. The roles of fat and protein intake are less clear [19,20], though evidence suggests that fat intake may have a stronger association with HbA1c than protein intake [21]. While the precise relationships remain unclear and may differ between individuals, seasonal variation in glucose control has important implications for interpreting HbA1c results and, consequently, for managing patients with DM—especially when measurements are taken at varying intervals throughout the year.

In the present study, the Dutch HbA1c Lifestyle Study (DAF-Study), we collected HbA1c data on a national scale over a four-year period including several lockdown periods during the COVID-19 pandemic and compared the results to national survey data on PA and MNI in the Dutch population. To our knowledge, this is the first study to compare these datasets on a national level to investigate whether seasonal variation in HbA1c exists in the Netherlands. The secondary objective was to explore potential associations between HbA1c, PA, and MNI.

2. Materials and Methods

2.1. National HbA1c Data

Retrospective data from January 2018 to December 2021 were anonymized and extracted from the Laboratory Information Systems of 24 laboratory organizations across the Netherlands and the Dutch Caribbean islands of Curaçao and Aruba. Data included HbA1c values, date of sampling, patient age at the time of sampling, and sex. The diabetes status and number of unique individuals were unknown, including pregnancy. HbA1c values are expressed in IFCC units (mmol/mol) ± 95% confidence interval, with NGSP values given in parentheses (%).

For each organization, data were cleaned by removing control samples, incomplete entries, and non-numeric results. Non-numeric entries such as “<“ or “>” were replaced with the corresponding numerical threshold values. Values below 20 mmol/mol (NGSP < 4.0%) were excluded as a result of analytical incorrect measurement or increased erythrocyte turnover. Age values exceeding 110 years were also excluded because of an artificial birthday linkage to control samples. Due to extensive missing data, the complete dataset from Aruba was excluded. Additionally, data from a pancreas transplantation centre were omitted due to a non-representative patient population. All data from Curaçao were excluded from the main dataset due to selection bias (anomalous sex ratio compared to the official data of The Health Institute of Curaçao [22]) and analytical bias, based on annual external quality reports (For details see Section 2.2 HbA1c Analysis). In total, 10% of the initial HbA1c data were excluded.

2.2. HbA1c Analysis

HbA1c measurements were performed using 33 analysers representing four analytical methods: HPLC ion-exchange chromatography (65%), capillary electrophoresis (4%), immunochemistry (22%), and enzymatic assays (9%).

All laboratories participated in the national External Quality Assessment (EQA) Program, which uses whole blood samples of human origin. The production and value assignment of EQA samples, traceable to the IFCC reference measurement procedure, were performed by the European Reference Laboratory for HbA1c in Winterswijk, the Netherlands. Annual EQA reports for all participants were reviewed for analytical bias, following IFCC Task Force criteria [23]. Only the reports from Curaçao exceeded the biological variation limits and were therefore excluded.

2.3. National Food Consumption Data

MNI data were obtained from the Dutch National Food Consumption Survey (DNFCS) 2019–2021, conducted by the National Institute for Public Health and the Environment (RIVM). Detailed methodology is described by Sanderman-Nawijn et al. [24]. In short, it is a cross-sectional study performed on children and adults aged 1–79 living in the Netherlands. Annually 3570 participants completed online questionnaires and two non-consecutive 24-h dietary recalls. From this dataset, total energy intake (TEI) was extracted, along with macronutrient-specific data: total carbohydrate intake (TCI), total fat intake (TFI), and total protein intake (TPI). TEI is expressed in kilocalories, while MNI components are reported as a percentage of energy ± 95% confidence interval. Due to the study design, the extracted data had an equal distribution on age and sex ratio. Weighting for age and sex was applied to align with the HbA1c dataset.

2.4. National Sports Participation Data

PA data were retrieved from the Dutch National Sports Participation Index Survey (2018–2021), conducted in collaboration with the Dutch Olympic Committee and Dutch Sports Federation (NOC*NSF). In short, it is a cross-sectional study monthly performed on children and adults aged 5–80 living in the Netherlands. Monthly, about 3250 participants completed online questionnaires concerning sports activity, behaviour, frequency, etc. From this data, the Sports Participation Index (SPI) was derived, representing a weighted distribution across four categories: (1) no sports activity, (2) active 1–3 times/month, (3) active 4–9 times/month, and (4) active 10+ times/month [25]. Due to the study design, the extracted data had an equal distribution on age and sex ratio. Therefore, weighting factors were used to adjust for these variables to ensure comparability with the HbA1c data.

2.5. COVID-19

Information concerning lockdown periods during the COVID-19 pandemic in The Netherlands and Curaçao were retrieved from The Dutch Institute of Public Safety (NIPV), and the Dutch Ministry of Public Health (VWS) [26,27].

2.6. Data Transformation

Data were transformed into monthly and seasonal averages based on meteorological definitions (e.g., spring = 1 March–31 May).

2.7. HbA1c Prediction by PA and TEI

To correlate PA and MNI with HbA1c, only data from the overlapping period (2019–2021) were used.

Assuming that MNI (via TEI) positively influences HbA1c levels and PA negatively influences them, an algorithm was developed to predict monthly relative HbA1c values. Because MNI data are expressed as energy percentages (always totalling 100%), TEI was used instead as a single positive driver, reflecting the cumulative effect of all macronutrients, and expressed as an absolute number.

The absolute difference between the monthly averages and the average of the entire study period was calculated. By dividing these monthly differences by the study period average, the numbers become relative and therefore more comparable.

SPI values were multiplied by −1 to emphasize their hypothesized inverse effect on HbA1c formation. The prediction formula is as follows:

pHbA1c_x = %TEI_x + (−1 × %SPI_x)

where

pHbA1c_x = predicted HbA1c for month x;

%TEI_x = relative TEI for month x;

%SPI_x = relative SPI for month x.

Monthly predicted values were subsequently converted into seasonal averages.

2.8. Statistical Analysis and Calculations

Calculations were performed using Microsoft Excel 365 (Microsoft Corp., Seattle, WA, USA). Statistical analyses were conducted using GraphPad Prism 10.2.0 for Windows (GraphPad Software, San Diego, CA, USA). Group differences were evaluated using one-way ANOVA (Kruskal–Wallis test) or the non-parametric Mann–Whitney U test. Relationships were estimated by Pearson correlation coefficient. p values < 0.05 were considered statistically significant.

3. Results

3.1. HbA1c Results over Time and Seasonal Variation

A total of 5,635,920 HbA1c results were included, comprising 2,755,965 data points from females (48.9%) and 2,879,955 from males (51.1%). The average age was 65.2 years (range: 0.9–105 years). Of the total dataset, 0.7% represented individuals aged 1–17 years, while 99.3% corresponded to individuals aged 18 years and older. The mean HbA1c was 52.4 ± 0.01 mmol/mol (NGSP 6.95 ± 0.001%) (

Table 1. Number of HbA1c results reported per season per year.

		Spring	Summer	Autumn	Winter	Annual Mean/ Total N
2018	XHbA1c				52.9 (6.99) *
	N				227,117
	XHbA1c	52.3 (6.94)	51.6 (6.87)	51.5 (6.73)	52.9 (6.99) **	52.0 (6.91)
	N	348,010	326,414	368,228	347,245
2019	XHbA1c	52.7 (6.97)	52.3 (6.94)	51.6 (6.87)	52.7 (6.97)	52.3 (6.94)
	N	376,682	338,322	386,652	368,016
2020	XHbA1c	53.2 (7.02)	51.8 (6.89)	52.2 (6.93)	53.7 (7.06)	52.5 (6.95)
	N	213,464	358,450	371,330	352,560
2021	XHbA1c	53.3 (7.03)	52.5 (6.95)	52.2 (6.93)	53.2 (7.02) ***	52.9 (6.99)
	N	374,936	363,433	396,230	118,631
	Ntotal	1,313,092	1,386,619	1,522,640	1,413,569	5,635,920
	Xtotal HbA1c	52.9 (6.99)	52.1 (6.92)	51.9 (6.90)	53.1 (7.01)	52.4 (6.95)
	95% CI	+/− 0.03 (0.003)	+/− 0.02 (0.002)	+/− 0.02 (0.002)	+/− 0.02 (0.002)	+/− 0.01 (0.001)

X_HbA1c: Mean HbA1c value expressed in mmol/mol; % value between brackets. N: number of results. 95% CI: 95% confidence limit. * January and February 2018; ** December 2018, January 2019 and February 2019; *** December 2021.

).

All HbA1c results were grouped by year and converted into monthly mean values as shown in Figure 1A. A significant and consistent increase in average HbA1c levels was observed over the four-year study period, with a total increase of 0.71 mmol/mol (NGSP 0.06%). Among males, this increase was 0.97 mmol/mol (NGSP 0.09%), which was notably higher than the 0.38 mmol/mol (NGSP 0.04%) increase observed in females (Figure 1A). These increases were statistically significant across all populations, exceeding the 95% confidence intervals.

When comparing annual averages, the years 2018 and 2019 ranged from 52.0 to 52.3 mmol/mol (NGSP 6.91 to 6.94%), while 2020 and 2021 increased to 52.5 to 52.9 mmol/mol (NGSP 6.95 to 6.99%), coinciding with several COVID-19 lockdown periods.

Conversion to seasonal values revealed a distinct pattern of seasonal variation as is shown in Figure 1B. All seasonal means differed significantly from each other (p < 0.0001). The lowest mean HbA1c level was observed in autumn (51.9 ± 0.02 mmol/mol; NGSP 6.90 ± 0.0018%), and the highest in winter (53.1 ± 0.02 mmol/mol; NGSP 7.01 ± 0.0018%), resulting in a mean seasonal difference of 1.2 mmol/mol (NGSP 0.11%).

Although data from Curaçao were excluded from the main dataset due to discrepancies in the male/female diabetes ratio and analytical bias, seasonal variation should not be influenced by these anomalies and was therefore separately evaluated. The purpose of this evaluation was to assess whether seasonal fluctuations were less or absent, given the region’s proximity to the equator, in comparison to The Netherlands (Figure 2A,B). The dataset contained 66,575 HbA1c results, and the observed seasonal difference was much smaller (0.4 mmol/mol; NGSP 0.037%) compared to the main dataset, and not statistically significant.

Notably, in accordance with the main dataset, mean values from 2018–2021 showed an increase of 1.4 mmol/mol (NGSP 0.128%) in the last two years, which overlapped with COVID-19 lockdown periods.

3.2. Macronutrient Intake (MNI) and Seasonal Variation

Data from the Dutch National Food Consumption Survey showed only minimal fluctuations over time in carbohydrate, protein, fat, and total energy intake (TEI): +1.25% Energy, +0.1% Energy, −0.5% Energy, and −63.5 kcal, respectively (Figure 3A). These variations were not statistically significant and remained within confidence intervals.

During the lockdown periods in 2020 and 2021, no clear changes were observed in the intake patterns of individual macronutrients. However, TEI reached its highest values during the summer of 2020, aligning with the first lockdown period. This increase was not observed during the second lockdown in 2021.

Seasonal variation was observed for all three macronutrients, and TEI, with peak values occurring in different seasons. Although some seasonal differences were statistically significant, absolute differences were marginal and therefore clinically negligible (Figure 3B–E).

3.3. Physical Activity (PA) and Seasonal Variation

Data from the National Sports Participation Database showed a 7.2% decrease in physical activity over the study period. Significant reductions were observed in 2020 and 2021, coinciding with national COVID-19 lockdowns. Notably, 2021 recorded the lowest overall PA levels (Figure 4A).

The highest activity levels were reported in summer/autumn, while the lowest were seen in winter/spring (Figure 4B). Seasonal means differed significantly between spring and summer and between autumn and winter (p < 0.0001).

3.4. HbA1c Prediction

Data from the HbA1c database, the National Food Consumption Survey, and the National Sports Participation Database were combined for the period June 2019 to July 2021.

Figure 5 presents the relative deviations from the mean for PA, TEI, and TEI+PA (predicted HbA1c), compared with the measured HbA1c seasonal values. TEI alone followed a curve that did not align with actual HbA1c values. Correlation analysis confirmed no significant association (p = 0.157; Pearson r = 0.286).

In contrast, both PA and TEI+PA showed curves that aligned with the seasonal HbA1c trend. Statistically significant correlations were observed for PA (p < 0.0001; Pearson r = 0.700) and for TEI+PA (p = 0.004; Pearson r = 0.541), suggesting that physical activity may be the primary driver of seasonal variability in HbA1c.

4. Discussion

This large-scale study is the first to suggest a relationship between HbA1c, physical activity (PA), and macronutrient intake (MNI) in the Dutch diabetes population, aiming to identify which of these two factors contributes most to the observed seasonal variation in average HbA1c levels. Although seasonal fluctuations in HbA1c have been described in several countries, few studies have analysed them at a national scale or explicitly compared lifestyle-related drivers. This DAF-study therefore addresses a clear knowledge gab in linking nationwide laboratory and lifestyle data.

Firstly, a clear seasonal variation in mean HbA1c values at the population level was observed. The lowest levels occurred in autumn, and the highest in winter, with a maximum amplitude of 1.2 mmol/mol (NGSP 0.11%), which aligns with previously published findings [8,9,10,11,12,13,14].

Despite limitations in the Curaçao dataset, the absence of seasonal variation in this population supports earlier research, indicating that seasonal fluctuations in HbA1c decrease closer to the equator and increase with greater distance from it, regardless of hemisphere [8,15].

Both PA and MNI are commonly proposed as primary contributors to seasonal HbA1c variation. Seasonal changes in weather and temperature influence activity patterns and the availability of seasonally dependent food products.

Because the HbA1c dataset lacked corresponding individual-level MNI and PA data, national datasets representing the general Dutch population were used. These were partially matched to the HbA1c data using age- and sex-based weighting factors.

Although the proportion of individuals with (pre)diabetes within these national datasets is unknown, their documentation confirms that (pre)diabetics are included as a subset. To validate whether people with diabetes follow similar seasonal PA trends as the general population, we consulted longitudinal data from the Dutch National Institute for Public Health and the Environment (RIVM), covering 2001–2022. These data, which report the percentage of the population meeting national PA guidelines, showed that from 2018 to 2021 [28], the diabetes population followed the same annual trend as non-diabetics—albeit at lower absolute levels (approximately 35% vs. 50%, respectively). This suggests that the PA dataset is sufficiently representative for this study.

Our findings show that PA patterns also display seasonal variation. The highest levels were observed in summer and autumn, and the lowest in winter and spring. These results are consistent with the findings of Duijvestijn et al. (2021) [29] and a 2009 review by Shephard et al. [30]. In modern society, most seasonal variation in PA can be attributed to leisure-time physical activity, particularly during summer.

With respect to total energy intake (TEI), we observed only a significant difference between the highest values in summer and those in autumn. According to dietary reference intake values from the Dutch Health Council and the European Food Safety Authority [31,32,33], all seasonal averages for macronutrients remained within recommended ranges: carbohydrate (55.0% energy/day; reference: 45–60%), fat (23.3%; reference: 20–40%), protein (22.0%; reference: 10–25%), and TEI (2060 kcal; reference: up to 2400 kcal/day). Although significant changes were seen in the individual macronutrients (Figure 3), the absolute differences were marginal and therefore clinically negligible. These results indicate no excessive MNI during any particular season, but rather a stable dietary pattern throughout the year. The slight summer increase may be attributed to vacation periods. This effect is also reported from continuous glucose monitoring data suggesting that seasonal, holiday, and behavioral changes over extended periods can influence glycemic trends [34]. Next to this, the summer of 2020 shows the highest seasonal caloric intake value, probably as a result of the fist COVID-19 lockdown period.

The MNI dataset also includes an unknown proportion of individuals with (pre)diabetes. According to recent Dutch government figures, approximately 13% of the national population may fall into this category [35], which is substantially lower than the nearly 100% representation in our HbA1c dataset. Nevertheless, given the year-round availability of energy-dense foods and the likelihood that individuals with (pre)diabetes engage in lifestyle interventions, dietary modifications, and/or pharmacological treatment, this dataset remains adequately representative.

To assess the relative influence of PA and MNI on seasonal HbA1c variation, we applied a simulation model that assumed a decreasing effect of PA and an increasing effect of MNI on HbA1c. The role of macronutrients in glycemic control has been widely studied, but there is limited consensus regarding the specific impact of fat and protein [21]. The exact composition of nutrients appears to play a role in determining whether they have a positive or negative effect on postprandial glucose levels and, by extension, on HbA1c [36]. Since detailed nutrient composition data were unavailable, we assumed that all macronutrients contribute equally to HbA1c formation.

Combining the effects of PA and MNI should, in theory, generate a predicted HbA1c trend that reflects the actual seasonal values. Although this simplified model does not quantify absolute differences in amplitude because of the absence of specific weighting factors for every variable, the direction and shape of the trends are sufficiently informative. Our findings show that adding PA to the MNI curve produces a pattern that closely mirrors the actual HbA1c data, with statistical analysis confirming a significant correlation. This supports the hypothesis that PA is the primary driver of seasonal HbA1c variation.

In addition to the seasonal variation, we also observed a steady increase in average HbA1c levels from 2018 to 2021. Calibration bias or drift as an interfering factor was ruled out by the check on external quality assessment reports from all the participating laboratories during this study. This increase, however, appears to be associated with a decrease in PA, while MNI seems to be a more consistent factor without much fluctuation. Evolutionarily, modern humans’ PA levels have decreased, and there is a continuous availability of energy-dense food, resulting in a positive energy balance. Over time this leads to the accumulation of body fat, increasing the risk of obesity and, consequently, diabetes, which poses a significant public health risk [37]. When examining the effects, particularly during the COVID-19 pandemic, it is notable that PA particularly decreased during lockdowns. Other studies have reported similar reductions in physical activity during lockdowns, especially in individuals over 60 years of age [38,39] which is exactly the age group in our DAF-study.

In contrast, the impact of the pandemic on energy intake remains unclear. Some studies report increased caloric intake [40], while others describe undernutrition [41]. Our data, however, do not show substantial changes in MNI, reinforcing the dominant role of PA in influencing HbA1c during this period.

The relatively sharp increase in average HbA1c during 2020–2021, compared to earlier years, strengthens the case for a link between declining PA and rising HbA1c levels.

Next to the seasonality, a steady increase in HbA1c is observed from 2018 to 2021 (Figure 1A). Although this increase is small in magnitude, it signals a worrisome trend that could reflect declining PA levels and worsening population-level glycemic control. Published longitudinal data are rare, dating from the pre-standardisation era, small in size, and lagging information on HbA1c analysis methods, and therefore unsuitable for trend evaluation with our DAF-study data [42].

This study has several limitations. Data from both the National Food Consumption Survey and the Dutch National Sports Participation Index Survey were obtained through self-reported questionnaires. While acceptable at a national survey level, they may not capture the nuanced behavioural changes most relevant for people with diabetes. Although this method has known drawbacks—such as recall bias and social desirability bias—previous reports have proven that self-reported data can still be sufficiently robust to support conclusions at the population level and remain valuable for policymakers and healthcare professionals.

The prediction model used to combine TEI, PA, and HbA1c is simplistic and likely oversimplifies the complex relationship. It assumes equal contribution of all macronutrients to HbA1c and treats PA as a uniform negative driver. More refined modeling, perhaps incorporating continuous glucose monitoring or dietary quality indices, would strengthen the argument. Future studies should consider multilevel modeling with stratification by age, sex, diabetes type, and treatment modality. Controls for confounders such as medication adherence, comorbidities, or socioeconomic status would also be valuable.

Our findings are particularly relevant for Western European countries with comparable climates, healthcare systems, and dietary habits. Similar national-level investigations in other Western European countries would allow cross-validation of results and support region-specific recommendations.

Furthermore, the study combined three distinct databases representing different population samples, which were aligned using weighting factors. As a result, the absence of individual-level linkage between HbA1c values and lifestyle variables introduces a major limitation. The study therefore cannot establish causality but only suggests associations at the population level. Nonetheless, our findings are consistent with, and strengthen, those of previously published studies in this field [21].

5. Conclusions

In conclusion, we demonstrate that the Dutch (pre)diabetes population exhibits a clear seasonal variation in mean HbA1c levels, which appears to be primarily driven by seasonal fluctuations in physical activity.

Events that significantly affect population-level physical activity—such as the COVID-19 lockdown periods—seem to have a direct impact on glycaemic control, as reflected by more profound increases in average HbA1c during those times. Therefore, policymakers should prioritize physical activity promotion during societal disruptions.

Lifestyle interventions have been shown to effectively prevent various forms of diabetes and improve both glycaemic and cardiometabolic outcomes, particularly through enhanced physical activity [43,44,45]. However, the success of such interventions depends heavily on sustained behavioural change and active promotion [46,47,48,49].

We therefore advocate for strong public health recommendations and supportive programs that emphasize the importance of initiating, maintaining, and intensifying physical activity across the population—particularly in winter months.