Associations of sedentary time and self-reported television time during pregnancy with incident gestational diabetes and plasma glucose levels in women at risk of gestational diabetes in the UK

The target sample size was calculated in 2015 using G*Power [21] based on incident gestational diabetes as the key outcome variable. The required sample size at recruitment was 326 with significance set at 0.05, power at 0.80, and the effect size as OR 1.73, derived from a meta-analysis of studies that included a variety of measures sedentary time [22]. The calculated sample size anticipated a drop-out rate of 30%.

Participants were recruited from two antenatal clinics within hospitals in the North East of England when they attended for their 12-week ultrasound scan between February and August 2017. All participants had to have at least one risk factor for GDM to be eligible for the study (BMI ≥30 kg/m², first-degree relative with diabetes, previous GDM, minority ethnic origin with a high prevalence of diabetes (i.e., South Asian, Black Caribbean), or previous macrosomic baby (≥4.5 kg) [23]). Additional inclusion criteria included that participants were at least 18 years old, pregnant with only one baby, fluent in English, and did not have pre-existing diabetes (because, by definition, one who has pre-existing diabetes cannot develop GDM). All participants provided written consent prior to engagement in any research activities. Ethical approval was provided by the NHS (REC reference: 16/SC/0355).

Participants were asked to wear an activPAL3 (PAL Technologies, Glasgow, UK) continuously for seven days during their second trimester. The activPAL has been validated for the measurement of total sitting time [24] and the detection of transitions from sitting to standing (‘breaks’) [25] in free-living contexts. The activPAL was fitted immediately following the participant’s 20-week anomaly ultrasound scan, upon confirmation from the sonographer that no fetal problems were detected. Prior to fitting, the device was covered with a nitrile sleeve and piece of waterproof Tegaderm to allow for a continuous wear protocol. The activPAL was affixed by a trained member of the research team to the anterior midline of the participant’s right thigh using an 8x10cm piece of Tegaderm. Verbal and written instructions for use were given to each participant; participants were instructed to leave the device on for 24-h per day, removing it only if the leg was going to be fully submerged underwater (e.g., swimming and bathing) or if skin irritation occurred. Participants were instructed to reaffix the activPAL immediately following any periods of removal (on the opposite thigh, if necessary, unless skin irritation was too severe) using additional pieces of Tegaderm that were provided. Participants were also given diaries to record any instances of removal, as well as the start and stop times of all night-time sleep. At the end of the wear period, the devices were collected from participants at their homes or workplaces, or postal arrangements were made as needed. Accelerometry data sets were considered valid if participants provided at least four complete (24-h) days of measurement [6]. We did not require that one of the four days be a weekend day, although 97% of participants who provided four days of measurement provided data for at least one weekend day.

After downloading the raw data from the devices into EventsXYZ files, the activPAL data were processed using a validated automated algorithm in STATA [26] using the algorithm’s default criteria for identifying sleep and non-wear. These criteria defined the main bout of sleep as a continuous bout of non-movement lasting ≥5 h, or ≥ 2 h of non-movement that was also the longest per noon-to-noon 24-h period. Bouts on either side of the main sleep bout were also classified as sleep if they met any or all of the following three criteria: non-movement bouts lasting ≥2 h, ≥30 min non-movement with ≤20 steps, or only postural changes without steps. Days were considered invalid if one posture accounted for ≥95% of waking wear, if there were < 500 steps in a day, or if there were < 10 h of waking wear. The day of fitting or removal were also manually classed as invalid even if they contained > 10 h of waking wear, thus only 24-h days of wear were included in analyses. In accordance with best practice for activPAL data processing [27], heatmaps were generated to visually cross-check the algorithm’s identification of sleep start and stop times against participants’ sleep diaries; in cases of large disparity (i.e., at least an hour’s difference between the two), the algorithm’s sleep identification was manually corrected to align with the diary.

Measurements of total sedentary time (hours per day), prolonged sedentary time (uninterrupted sedentary time (hours per day) accumulated in bouts lasting ≥30 min [6]), breaks in sedentary time (number of sit-to-stand transitions), and time spent stepping (hours per day) were based on the activPAL’s default outputs and were calculated as the sum of each variable on valid days, divided by the number of valid days.

At the time of accelerometer fitting (20 weeks’ gestation), participants were also asked to report the amount of time they usually spent watching television per day in the second trimester (none, < 30 min, 30 min to less than 2 h, 2 h to less than 4 h, 4 h to less than 6 h, ≥6 h). Responses were dichotomized as less than or ≥ 2 h per day [15].

Due to an increased risk status for GDM, participants had a standard 2-h 75 g oral glucose tolerance test (OGTT) after an overnight fast at between 24 and 28 weeks’ gestation as part of their routine pathway of care, providing fasting plasma glucose levels, 2-h glucose levels, and confirming cases of GDM. Twelve eligible participants did not have the OGTT due to health reasons that precluded the glucose test; thus for these participants, no fasting or 2-h glucose values are available, but it is known whether they developed GDM during the pregnancy (four of them did) through other forms of monitoring. GDM was diagnosed in accordance with NICE guidelines (fasting glucose ≥5.6 mmol/litre and/or 2-h ≥ 7.8 mmol/litre) [23].

Participants provided basic details about themselves on an enrolment form after giving consent at their 12-week scan, including whether they had been diagnosed with GDM before. Participants’ BMI recorded at their booking appointment was extracted from medical records.

All statistical analyses were conducted using R. Fasting and 2-h glucose values were positively skewed and were thus log-transformed for all analyses which resulted in a normal distribution. Independent t-tests and chi-square analyses were used to compare characteristics of those who did and did not withdraw from the study, as well as to compare those who did and did not provide valid accelerometry data. Logistic regression was used to test associations with gestational diabetes diagnosis, and linear regression was used to test associations with fasting and 2-h glucose. Although the study took place at two study sites, mixed models were not used because two sites is too few to robustly estimate the random effect, thus recruitment site was included as a control variable in all analyses. All models additionally controlled for age, BMI, and time spent stepping as measured by the activPAL; waking wear time was also controlled in accelerometry models. GDM logistic regression models additionally controlled for previous GDM. Because of the possibility that associations between television time and health outcomes may be confounded by socioeconomic status [28], television time models are shown with and without additional adjustment for household income category. For completeness, we also show the total sedentary time models with and without adjustment for income. Because of abnormalities in glucose metabolism in GDM, we expected the association between sedentary time and glucose levels might be different for those without and without GDM. Thus, we constructed additional linear regression models which included interaction terms between GDM status and centred accelerometry variables; if the interaction was significant and improved the fit of the model significantly based on likelihood ratio tests, estimated marginal means of linear trends were calculated (emmeans R package). The estimated marginal means of linear trends for each subgroup were interpreted as significant if the confidence intervals did not cross zero. Complete-case analysis was used for each model; as there are different missing variables within each model, the final sample size per model is specified within the table.

Of those who were approached to take part in the study and were eligible, 326 consented to take part at the 12-week scan (54.9% response rate). No information about those who declined to take part is available. Sixty-six (20.2%) of those who initially consented withdrew from the study prior to wearing the accelerometer, either through choice (n = 46) or because their continued participation was considered inappropriate for medical reasons (e.g., due to fetal anomaly, n = 20). Those who were withdrawn (all reasons) were significantly younger than those who were retained in the study (28.6 ± 0.7 vs 30.2 ± 0.3 years, p < 0.001); there were no significant differences in other characteristics such as BMI or parity (p > 0.05). Of the 260 women who were fitted with the accelerometer, 192 provided valid data sets (≥4 days of 24 h). A substantial proportion of insufficient wear was attributed to skin reactions to the activPAL and/or the dressing with which it was attached; 50 participants indicated either directly to the research team or on their wear diaries that they experienced at least some degree of skin irritation underneath the device. We were unable to retrieve BMI data for four participants; as BMI is a covariate in all models, these participants were excluded from all analyses, resulting in a final analytical sample size of 188.

Characteristics of participants in the analytical sample (n = 188) are shown in Table 1.

Those who provided valid data sets (n = 188) were significantly older (31.0 ± 5.1 vs 28.3 ± 5.2 years, p < 0.001), were more likely to be in the highest household income group (29.3% vs 13.0%, p < 0.001), and were more likely to be married/cohabiting (86.6% vs 69.4%, p < 0.001) and employed (81.3% vs 65.3%, p < 0.01) than those who did not. There was no difference in GDM prevalence between those who did and did not provide valid data sets (p = 0.17). All but seven participants who provided valid accelerometry data provided self-reported television time (n = 181).

Mean accelerometry variables are shown in Table 2. On average, participants spent 65.1% of waking hours in sedentary time. Sixty-eight (37.6%) of those who provided data on television time reported watching ≥2 h of television per day in the second trimester.

Total sedentary time was not associated with risk of developing GDM (Table 3). Total sedentary time had positive but non-significant associations with fasting and 2-h glucose levels (Table 3). Additional adjustment for income did not substantially impact the associations between sedentary time and GDM (OR 1.00 (95%CI 1.00, 1.01)), fasting glucose (β = 0.14 (95%CI -0.02, 0.30)), or 2-h glucose (β = 0.11 (95%CI -0.06, 0.27)).

The interaction terms between sedentary time and GDM status in relation to fasting and 2-h glucose were significant (p < 0.05) and near-significant (p = 0.06), respectively. Estimated marginal means of linear trends were applied to the linear regression model which indicated that, for those who did not have GDM, sedentary time was significantly associated with fasting (β = 0.16 (95%CI 0.01, 0.31), SE = 0.08) and 2-h glucose (β = 0.15 (95%CI 0.01, 0.30), SE = 0.07). Sedentary time was not significantly associated with fasting (β = − 0.21 (95%CI-0.50, 0.09), SE = 0.15) or 2-h glucose (β = − 0.15 (95%CI -0.43, 0.14), SE = 0.14) among those with GDM.

Prolonged sedentary time was not significantly associated with risk of developing GDM (Table 3). Prolonged sedentary time had a significant, positive association with fasting glucose but not 2-h glucose (Table 3). The interaction terms between GDM status and prolonged sedentary time in relation to fasting and 2-h glucose levels were not significant (both p > 0.05).

Breaks in sedentary time were not significantly associated with risk of developing GDM (Table 3). Breaks in sedentary time were not significantly associated with fasting or 2-h glucose levels (Table 3). The interaction terms between GDM status and breaks in relation to fasting and 2-h glucose were significant (both p < 0.05). Estimated marginal means of linear trends indicated that breaks were associated with significantly lower fasting glucose (β = − 0.55 (95%CI –0.92, − 0.17), SE = 0.19) and lower 2-h glucose (β = − 0.40 (95%CI -0.77, − 0.03), SE = 0.19) among those with GDM. Breaks had no significant effect on fasting (β = − 0.05 (95%CI -0.20, 0.09), SE = 0.07) or 2-h glucose (β = 0.13 (95%CI -0.01, 0.26), SE = 0.07) among those without GDM.

Television time (less than or ≥ 2 h per day) was significantly associated with incident GDM (Table 3) The association between television time and GDM remained significant after additional adjustment for household income category (OR 2.93 (95%CI 1.15, 7.89), p = 0.03).

Television time was not significantly associated with fasting glucose levels (Table 3); additional adjustment for household income had a negligible effect (β = 0.11 (95%CI -0.05, 0.26), p = 0.18). Television time was also not associated with 2-h glucose levels (Table 3); additional adjustment for household income did not substantially affect the association (β = 0.05 (95%CI -0.11, 0.21), p = 0.51). Interaction terms between GDM status and television time in relation to fasting and 2-h glucose levels were not significant (both p > 0.05).