Marked improvement in HbA_1c following commencement of flash glucose monitoring in people with type 1 diabetes

We conducted a prospective observational study of the first 900 patients commenced on National Health Service (NHS)-funded flash monitoring (Freestyle Libre, Abbott, Witney, UK) in two University hospital clinics (Royal Infirmary of Edinburgh [RIE] and Western General Hospital) during February and March 2018. Prior to February 2018, flash monitor use was limited to those able to self-fund the purchase of sensors. From February 2018 onwards, people with type 1 diabetes were eligible for NHS-funded flash monitor use, conditional upon fulfilling all Scottish Diabetes Group criteria, namely that they: (1) were using intensive insulin therapy; (2) agreed to attend a flash monitoring education session; (3) agreed to scan glucose levels at least six times per day; (4) agreed to share glucose data with their clinic; and (5) had attended a diabetes structured education programme or demonstrated equivalent diabetes self-management knowledge. In February 2018, all people with type 1 diabetes attending our clinics (n = 2910) were sent a letter detailing these criteria and, if eligible, how to obtain NHS-funded sensors. All individuals who commenced NHS-funded flash monitoring attended a 1 h education session [5] and completed a form providing the start date and extent of any previous self-funded flash monitor use.

An additional cohort of all individuals with type 1 diabetes attending RIE clinics (where complete HbA_1c data were available for each of the past 5 years) was also created (n = 1351), for the purpose of tracking longitudinal changes in HbA_1c, in relation to exposure to flash monitoring. This included a large comparator population with no prior or current flash monitoring exposure (n = 518) and also all RIE flash monitor users from the main cohort (described above) where 5 continuous years of HbA_1c data were available. This study was entirely observational (with no deviation from standard clinical care) and ethics approval was not required.

The primary outcome was change in HbA_1c, defined as the difference between HbA_1c prior to commencement of any flash monitoring and the next available value after the flash monitoring education session. We also report the proportion of individuals achieving the Scottish HbA_1c target (<58 mmol/mol [7.5%]) and UK National Institute for Health and Care Excellence (NICE) target (≤48 mmol/mol [6.5%]) [6]. We obtained hospital admission and emergency department attendance data for the 6 months following NHS-funded flash monitor use and the corresponding 6 month period in the preceding 2 years. National prescribing database data were obtained for collected prescriptions for glucose test strips and sensors, over the same timescale described above. HbA_1c, admission and prescribing data are presented for the entire cohort of flash monitor users from both participating hospitals (n = 900). Scottish Index of Multiple Deprivation 2016 (SIMD) rank and quintile were determined [7]. The structured education programme offered in our centre is Dose Adjustment for Normal Eating (DAFNE) [8] and previous participation was discerned from our national clinic database system, SCI-Diabetes (https://​www.​sci-diabetes.​scot.​nhs.​uk). Mode of insulin delivery (multiple daily injection [MDI] or continuous subcutaneous insulin infusion [CSII]) was also obtained from SCI-Diabetes.

Additional data were collected in the subgroup of flash monitor users attending the RIE (n = 589): these included change in BMI, clinic questionnaire data, online questionnaire data and flash monitoring data. All individuals attending RIE diabetes clinics are asked to complete a form at each attendance (electronic supplementary material [ESM] Questionnaire 1) which includes hypoglycaemia questions (including Gold score and a modification of the Clarke assessment [9]), frequency of self-monitoring of blood glucose (SMBG), timing of bolus insulin and the Hospital Anxiety and Depression Scale (HADS) [10]. We report changes in hypoglycaemia and HADS score in those where paired pre- and post-flash monitoring questionnaires were available. In addition, all individuals attending RIE flash monitoring education events were sent an online questionnaire invitation 1 month after attendance (ESM Questionnaire 2). This included questions on satisfaction with flash monitoring and a modified version of the diabetes distress scale (DDS) [11]. Flash glucose data was obtained from the ‘LibreView’ portal (Freestyle Libre).

Data were largely non-normally distributed (as determined by Shapiro–Wilk test) and are presented as median and interquartile range (IQR). Paired data were analysed by Wilcoxon signed-rank test and unpaired data by Mann–Whitney U test. Comparisons across multiple groups were analysed by Kruskal–Wallis test. Categorical data were analysed by χ² or by McNemar test, when comparing paired repeated measurements. Logistic regression analysis was performed to identify predictors of HbA_1c response. One-proportion Z test was used to analyse change in modified DDS score responses. Correlations were analysed using Spearman’s rank correlation. A mixed effects model was used to assess the interaction between time (2014–2018) and exposure to flash monitoring on log-transformed HbA_1c, and paired Student’s t test was used to analyse difference in log-transformed HbA_1c between 2016 and 2018 for each flash monitoring exposure group. Significance was accepted at p < 0.05. All analyses were performed using RStudio version 1.0.153 (https://​www.​rstudio.​com).

Baseline characteristics are presented in ESM Table 1. Of the 354 (39.3%) individuals who had a history of previous flash monitoring self-funding, 64.0% reported greater than 50% use prior to NHS funding. Flash monitor use was commenced by 7.4% of self-funders in 2015, 29.9% in 2016, 56.6% in 2017 and 6.0% in 2018.

The median change in HbA_1c, between the last value prior to flash monitor use and the most recent value, was −4 mmol/mol (IQR −10 to 0, p < 0.001) (−0.4% [−0.9 to 0]). In individuals with baseline HbA_1c >75 mmol/mol (9.0%), the median change was −14 mmol/mol (IQR −22 to −7, p < 0.001) (−1.3% [−2.0 to −0.6]). In those with baseline HbA_1c 58–75 mmol/mol (7.5–9.0%), the median change was −5 mmol/mol (IQR −10 to −1, p < 0.001) (−0.5% [−0.9 to 0.1]). In those with HbA_1c <58 mmol/mol (7.5%) at baseline, the median change was −1 mmol/mol (IQR −5 to 3, p = 0.06) (−0.1% [−0.5 to 0.3]). In total, those with a starting HbA_1c that did not meet our national target (<58 mmol/mol [7.5%]) experienced a median −7 mmol/mol (−0.6%) change in HbA_1c (IQR −13 to −1 mmol/mol [−1.2 to −0.1%], p < 0.001). The median interval from baseline HbA_1c to final HbA_1c was 245 days (IQR 182 to 330), with no significant correlation between this interval and change in HbA_1c (r 0.044, p = 0.244) (ESM Fig. 1). Baseline HbA_1c and subsequent change in HbA_1c was strongly negatively correlated (r −0.479, p < 0.001) (Fig. 1). Overall there was a 48.8% increase in those achieving an HbA_1c <58 mmol/mol (7.5%) and a greater than twofold reduction in those with HbA_1c above 75 mmol/mol (9.0%) (p < 0.001, Fig. 2). The proportion of individuals achieving the NICE target of ≤48 mmol/mol (6.5%) rose from 10.1% to 18.7% (p < 0.001). HbA_1c since commencement of flash monitoring was not available in 17.4% (n = 157) of individuals in this cohort. The only significant differences in individuals with missing follow-up HbA_1c were greater proportion of men (20.4% vs 14.4% of women, p = 0.017) and more individuals using MDI (19.3% vs 12.4% CSII users, p = 0.016) (full data in ESM Table 2). Compared with the corresponding period the previous year, in the first 6 months following NHS-funded flash monitor use, the median number of prescribed (and collected) glucose test strip items (50 test strips per item) fell from 14 (IQR 7 to 21) to 2 (IQR 0 to 6) (p < 0.001). This equates to 3.8 test strips per day falling to 0.6 per day. A sufficient number of sensors were collected by 86.5% of flash monitor users to provide complete coverage over the first 6 months (from February or March 2018). Only 3.4% experienced ≤50% sensor coverage across this same period.

Median BMI increased from 26.0 kg/m² (IQR 23.4 to 29.0) to 26.3 kg/m² (23.5 to 29.8) after commencing flash monitoring (median interval 15 months [IQR 5 to 39], p < 0.001). Median BMI increase across a similar interval (2016–2018) in 396 non-flash monitor users was 0.1 kg/m² (IQR −0.8 to 1) (p = 0.027 comparing BMI change in flash monitor vs non-flash monitor users).

Within the RIE type 1 diabetes population, those who had not used flash monitoring were older, more likely to be male, had greater social deprivation and were less likely to have an HbA_1c <58 mmol/mol (7.5%) at baseline (Table 1).

Trends in HbA_1c were compared in every individual with type 1 diabetes attending the RIE, assessing the influence of flash monitoring exposure. Those who had never used flash monitoring and those who started NHS-funded flash monitoring after the initial February and March 2018 sessions had significantly higher baseline HbA_1c than those with a history of self-funded flash monitor use and those commencing flash monitoring for the first time in February and March 2018 (Fig. 3). Between 2016 and 2018 there was a significant fall in HbA_1c in all flash monitoring exposed groups but no significant change in those not using flash monitoring (Fig. 3 and ESM Table 3).

48.1% (361/750) individuals achieved an HbA_1c reduction of 5 mmol/mol (0.5%) or greater following commencement of flash monitoring. Univariate analysis identified younger age at diagnosis (18 years [IQR 11 to 29] vs 20 years [12 to 24], p = 0.03), higher baseline HbA_1c (68 mmol/mol [60 to 78] vs 58 mmol/mol [51 to 65], p < 0.001) (8.4% [7.6 to 9.3] vs 7.5% [6.8 to 8.1]) and prior SMBG fewer than four times per day (67.7% vs 45.2%, p < 0.001) as predictive of a fall in HbA_1c of 5 mmol/mol (0.5%) or greater. Age, diabetes duration, sex, previous self-funding, SIMD quintile, CSII use and DAFNE attendance were not associated with a greater likelihood of achieving a 5 mmol/mol (0.5%) fall in HbA_1c with flash monitoring (ESM Table 4). Logistic regression analysis identified higher baseline HbA_1c (OR 1.07 [95% CI 1.05, 1.08] per mmol/mol increment [0.1%]; p < 0.001) as a predictor of response, whilst belonging to the lower three (most deprived) quintiles of SIMD (OR 0.68 [95% CI 0.49, 0.95], p = 0.024) and older age at diagnosis (OR 0.973 [95% CI 0.966, 0.993] per year increment, p = 0.003) were independently associated with non-response (full model in ESM Table 5). In a separate model including low frequency of prior SMBG (<4/day), there was a borderline independent association with response (OR 1.70 [95% CI 0.99, 2.93], p = 0.055) (ESM Table 6).

Frequency of symptomatic hypoglycaemia (defined as glucose <3.5 mmol/l) increased, with those reporting two to three episodes per week or more rising from 25.8% to 48.4% (p < 0.001) following flash monitor use (ESM Fig. 2). Similarly, the proportion of people experiencing any asymptomatic hypoglycaemia (<3.5 mmol/l) rose from 20.4% to 29.5% (p < 0.001) (ESM Fig. 3). No significant change in severe hypoglycaemia was observed (7.3% vs 8.8%, p = 0.499) although there was a non-significant increase in the number of episodes where unconsciousness or seizure occurred (2.1% vs 4.5%, p = 0.099). No significant difference was observed in Gold score (2 [IQR 1 to 2] vs 2 [1 to 3], p = 0.104) and the proportion of people with impaired awareness of hypoglycaemia (IAH, defined as Gold score ≥4) did not change (12.5% vs 13.1%, p = 0.867), although prior IAH resolved in 42.5% (17/40) of people following flash monitor use.

Median HADS depression scores increased from 1 (IQR 0 to 3) to 2 (1 to 5, p < 0.001) with the proportion of those with an elevated score (>7) rising from 7.6% to 15.0% (p < 0.001). In the 8.2% of individuals (26/317) with newly elevated HADS depression score after flash monitoring commencement, the only difference observed was lower SIMD rank (2686 [IQR 1715 to 3959] vs 4857 [2884 to 6445], p < 0.001).

Median HADS anxiety scores increased from 4 (IQR 2 to 8) to 5 (2 to 8, p = 0.030) and the proportion with an elevated score rose from 24.9% to 30.9% (p = 0.028). HADS anxiety scores that were newly elevated after commencing flash monitoring developed in 12.3% (39/317) of people. This was associated with younger age (38 [IQR 26 to 47] vs 47 [35 to 58] years, p = 0.002), shorter duration of diabetes (13 [7 to 24] vs 22 [12 to 33] years, p = 0.002), prior history of self-funding (17.4% vs 9.2%, p = 0.031) and CSII use (21.7% vs 9.0%, p = 0.002).

These hypoglycaemia and HADS data were obtained from routinely collected clinic questionnaires. Paired pre- and post-flash monitoring questionnaires were available in 56.7% of eligible individuals (n = 334). Presence of paired questionnaire data was more likely in older individuals (45 years [IQR 33 to 57] vs 41 years [29 to 51], p = 0.001), those with lower baseline HbA_1c (63 mmol/mol [55 to 70] vs 65 mmol/mol [56 to 75], p = 0.023) (7.9% [7.2 to 8.6] vs 8.1% [7.3 to 9.00]) and prior DAFNE participants (62.8% vs 52.4%, p = 0.013) (ESM Table 7).

All individuals attending RIE flash monitoring education events were invited to complete a single online questionnaire which included a modified version of the DDS questionnaire, where potential responses were: ‘much more of a problem’ (assigned a value of −2), ‘more of a problem’ (value = −1), ‘unchanged’ (value = 0), ‘less of a problem’ (value = 1) and ‘much less of a problem’ (value = 2). The median responses were: 0 (IQR 0 to 0.3) for physician related distress, 0 (0 to 0.3) for interpersonal distress, 1 (0.4 to 1.4) for regimen-related distress and 0.6 (0.2 to 1.0) for the emotional component. Full results from the modified DDS are presented in ESM Table 8. The percentage of respondents with a net improvement in total DDS (90.0%), regimen-related distress (88.7%) and emotional distress (83.0%) were all significantly greater than 50.0% (p < 0.001). Respondents were also asked to provide a score from −5 (less control) to +5 (more control) with respect to the overall effect of flash monitoring on diabetes control; the median response was +5 (IQR 3 to 5). Of those who responded to the questionnaire, 70.6% reported taking their bolus insulin doses earlier in relation to meals following commencement of flash monitoring. The online questionnaire was completed by 54.1% of eligible individuals, with no demographic or diabetes-related differences between respondents and non-respondents (ESM Table 9).

When considering the entire cohort, where any flash monitoring data were available (n = 166), there were strong correlations with number of daily scans and both final HbA_1c (r −0.255, p < 0.001) and change in HbA_1c (r −0.279, p < 0.001). Paired flash monitoring data (i.e. within 7 days of first flash monitor use and at least 1 month later) was available in 53 of a possible 362 individuals with no prior self-funded flash monitor use. The only significant difference between those for whom this information was available and those for whom it was not was younger age at diagnosis (16 years [IQR 10 to 24] vs 21 years [12 to 33], p = 0.013) (ESM Table 10). The median interval between baseline and follow-up flash monitoring data was 6 months (IQR 5 to 7). No differences were observed in time between 3.9 and 10.0 mmol/l glucose, time above 10 mmol/l, time below 3.9 mmol/l, hypoglycaemic event number, hypoglycaemic event duration or mean glucose (ESM Table 11). The only significant difference was a fall in daily flash monitoring scans from baseline to follow-up (13 [IQR 9 to 19] vs 10 [6 to 14], p < 0.001).

There was a significant reduction in admissions for diabetic ketoacidosis (DKA), falling from 10 to 2 episodes (p = 0.043) in the 6 months following NHS-funded flash monitoring (compared with the corresponding period 2 years earlier [2016]). No differences were noted in the 6 months following NHS-funded flash monitoring with respect to number of hospital admissions (p = 0.539), duration of hospital admissions (p = 0.680) or emergency department attendances (p = 0.449).