Continuous glucose monitoring assessment of metabolic control in east African children and young adults with type 1 diabetes: A pilot and feasibility study

Abstract Background For individuals with type 1 diabetes (T1D) in East Africa and other low‐income regions, the last decade has seen substantial gains in access to insulin and trained healthcare providers, yet metabolic control remains poor. Methods The objective was to determine the feasibility of continuous glucose monitoring (CGM) and to gather baseline metabolic data for future power analysis in Ugandan and Kenyan youth with T1D using a Freestyle Libre Pro blinded CGM. Results Of 78 participants recruited, four sensors fell off and six patients did not return, leaving 68 evaluable subjects. Average age was 16 ± 5 (range 4‐26) years, 43% female. Average diabetes duration was 7 ± 5 years, insulin dose 0.9 ± 0.3 U/kg/d, and number of fingerstick glucose levels per day 2.1 ± 1.1. All were on human insulin. Point‐of‐care HbA1c was 10.9 ± 2.7% (96 ± 30 mmol/mol). Mean number of sensor days was 13 ± 3; >90% wore the sensor for ≥10 days. Mean glucose was 231 ± 86 mg/dL (12.8 ± 4.8 mmol/L). Only 30 ± 19% of time was spent in the target range (70‐180 mg/dL; 3.9‐10 mmol/L), and 7 ± 8% of time was spent in hypoglycaemia (glucose <55 mg/dL, 3.0 mmol/L). Hypoglycaemia occurred in 81% of participants, averaging five events/wk with an average duration of 140 ± 79 minutes/event. Conclusions Despite significant diabetes care improvements, East African youth with T1D have poor metabolic control with chronic hyper‐ and hypoglycaemia, placing them at high risk for serious acute and chronic complications. This study demonstrates the feasibility of CGM use in this population and provides baseline metabolic data that will be used to inform a future intervention study.


| INTRODUC TI ON
In 2017, the International Diabetes Federation reported that 50 600 children aged 0-19 were known to be living with type 1 diabetes (T1D) in Africa, with about 18 300 new cases diagnosed each year. 1 Major obstacles to providing T1D care have included lack of trained healthcare professionals, poor availability of insulin and testing supplies, and limited patient education materials. These same issues are common in much of the resource-poor world. While there are still problems in East Africa, all of these issues have substantially improved over the last decade. Despite these tremendous improvements in diabetes care delivery, the available HbA1c data show that the expected positive impact on glycaemic control has not occurred, and HbA1c levels remain unacceptably high in sub-Saharan Africa. This suggests that a fresh intervention approach is needed. [3][4][5][6] While HbA1c is a well-accepted measure of overall diabetes control, it does not provide other important measures such as the per cent time spent in hypoglycaemic or hyperglycaemic glucose ranges, glycaemic variability or daily glycaemia patterns. The current study was performed to explore these measures of glucose metabolic control, using blinded continuous glucose monitoring (CGM) in African patients with T1D attending the paediatric diabetes clinics at national referral hospitals in Uganda and Kenya. These novel data will inform the power analysis for a planned intervention study. An additional goal was to determine the feasibility of wearing CGM in this population. The equatorial climate is hot and humid, raising concerns about sensor skin adhesion or skin irritation. There were also concerns about whether patients would accept wearing this visible evidence of diabetes in a region where the disease is a social stigma.

| Study design
The primary purpose of this cross-sectional observational study was to characterize glucose metabolic control in children and youth with T1D in East Africa using the blinded Freestyle Libre Pro CGM system (Abbott Diabetes Care), in order to obtain data for power analysis for a future intervention study. We hypothesized that time-in-range would be lower than that recommended, and time-in-hypoglycaemia higher. A second objective was to determine the feasibility of CGM in this population. The Uganda and Kenya sites were chosen because of their long-standing working relationships with the paediatric diabetes team at the University of Minnesota.

| The Freestyle Libre pro flash CGM system
The Freestyle Libre is an intermittently scanned (or 'flash') CGM system that is factory calibrated and can be worn for up to 14 days without the need for additional calibrations. 7 The Freestyle Libre Pro is a blinded CGM. Glucose readings are stored on the sensor with no real-time data available to the patient, who does not need to do anything but wear the device. Data are automatically stored and then uploaded once the sensor is removed, and a retrospective report is generated for the medical professional. It has been suggested that this system is less accurate in the low glucose range than in normal or high glucose ranges, with one group reporting a hypoglycaemia MARD (mean of the absolute value of the difference between the CGM value and the reference value, divided by the reference value) of up to 24%. 8,9 This means a CGM value of 55 mg/dL (3.0 mmol/L) could be on average 24% lower or higher than the laboratory assessment, or between 42 and 68 mg/dL (2.3-3.8 mmol/L), with just as much chance that the number could be off in one direction as the other. However, that was a small study with only 56 comparisons of CMG vs measured glucose levels. Other studies have suggested a high degree of accuracy with the FreeStyle Libre CGM, including at low glucose ranges. 10-12

| Participants
Participants were recruited from the paediatric endocrinology clinics at Mulago Hospital in Kampala, Uganda and Kenyatta Hospital in Nairobi, Kenya, using convenience single-stage sampling methods.

| Procedure
At enrolment, participants completed a survey including questions about demographics, diabetes history and current diabetes management. Participants reported any history of severe hypoglycaemia events in the last year and the frequency. Severe hypoglycaemia was defined as an event where individuals were cognitively impaired to the point that they were unable to treat themselves; were unable to verbalize their needs; were incoherent, disoriented and/or combative; or experienced seizure or loss of consciousness.
After completing the survey, all participants had point-of-care HbA1c levels measured. The analyser at Mulago Hospital was a Hemocue HbA1c 501 System, with a reported coefficient of variation (CV) of 1.7%-5.5%. 13 The analyser at Kenyatta Hospital was a Siemens DCA Vantage, CV 2.5%. 14 Analyser quality controls were run on each day of use. Participants were asked to wear the FreeStyle Libre Pro sensor for 2 weeks. CGM blood glucose levels were blinded to participants and investigators during this time. Participants were instructed to manage their diabetes normally, administering their daily insulin doses and checking their blood glucose level using home glucose metres per their usual routine, eating their typical diet and participating in their usual activities. They received a notebook and pens and were asked to record daily capillary blood glucose concentrations whenever they measured them, time and amount of insulin doses received, and to log other events such as hypoglycaemia symptoms, severe hypoglycaemia, hospitalization, or additional health visits or treatment.
At the end of 2 weeks, participants returned to clinic. Researchers removed the sensor (or subjects brought it with them if it had fallen off at home) and uploaded the glucose data to LibreView, an online website that stores the sensor data and creates a report. The reports were then downloaded for data analysis by the research team, and for clinical decision-making by the local teams.

| Data analysis
Analysis of demographics and subject characteristics (age, gender and glycaemic metrics) was descriptive and used the data analysis program IBM SPSS 25. Summary statistics were calculated, including mean and standard deviation, and because the data were nonnormally distributed, median and interquartile range (IQR). HbA1c results that were >14% (130 mmol/mol, the upper limit of measurement for the testing devices) were recorded as 15% (140 mmol/mol) for analysis purposes. A single-factor analysis of variance (ANOVA) was completed to evaluate for differences in mean blood glucose levels between age groups, duration of diabetes, insulin type and gender. Significance was accepted at P < .05. As there were no gender differences in age (P = .81), duration of diabetes (P = .75), HbA1c (P = .54) or glucose time-in-range (P = .47), the data were not separated by gender.
Regression analysis was performed using CGM outcomes as the dependent variable to determine whether duration of diabetes, insulin regimen, age, duration of diabetes and age at diagnosis were associated with the glycaemic data.

| Patient demographics
In Uganda, 62 participants were enrolled. Five did not return their sensors (8%), and one fell off within 72 hours and was not included in the data analysis, leaving 56 subjects with evaluable data. In Kenya, 16 sensors were placed on eligible subjects. Three fell off within 1-2 days due to adhesive problems and were not included in the analysis. One patient (6%) did not return the sensor, leaving 12 subjects with evaluable data.
Participant characteristic are listed in Table 1. In Uganda, the average age of participants was 16 ± 5 years. Fifty-four per cent were female, 23% were age 4-11 years, 32% age 11-17, and 45% age 18-26. This age distribution is roughly representative of the Ugandan paediatric diabetes clinic population. 3 In Kenya, there were five children age 4-17 and seven youth age 18-26, with one-third of subjects male. Patients were normally nourished in both countries.
Characteristics of the 10 subjects (13%) who did not complete the study were analysed and compared to completers. No statistically significant difference was found in age, gender, time with diabetes, age at diagnosis or point-of-care HbA1c compared to participants who completed the study (all P values >.05).

| Baseline diabetes characteristics
In both countries, the average diabetes duration was about 7 years and the average insulin dose was 0.9 ± 0.3 units per kg body weight. All participants received human insulin, including NPH (isophane), regular (soluble) and Mixtard (premixed 70% NPH and 30% regular insulin). Insulin usage was similar between the two countries with the majority of patients (n = 43, 63%) on Mixtard at the time of the study, with some of these (n = 11, 16%) receiving an extra noon injection of regular insulin. The remaining 25 subjects (37%) received twice daily NPH insulin plus two (n = 12), or three (n = 3) injections of regular insulin. There was no statistically significant relation between insulin type (separate NPH and regular versus premixed insulin) and either mean glucose (P = .76) or HbA1c (P = .78) level. There was also no significant relation between the duration of diabetes and mean glucose (P = .34) or HbA1c level (P = .72). Patients tested blood glucose levels at home an average of 2.2 ± 1.1 times per day in Uganda and 1.6 ± 1.0 times per day in Kenya. Across both countries, 78% of patients reported being able to test blood glucose at least twice daily, while 13% reported no daily blood glucose testing.

| Glucose metabolic data
Glucose metabolic data are shown in Regression analyses using CGM outcomes as the dependent variables were completed using patient demographics as the independent variables. In all cases, the data were too widely distributed to allow meaningful interpretation.
Levels appeared to be slightly higher in the 12-to 17-year age group and lower in the 18-to 26-year-old group, but this did not achieve statistical significance. As noted in Figure 2, for each HbA1c level, there was a wide range of CGM mean glucose levels. Fifteen per cent of patients had HbA1c levels >14% (130 mmol/mol), the upper limit of measurement accuracy, and these were recorded as 15% (140 mmol/mol) for analysis purposes which may at least partially explain the glucose spread seen in Figure 2 at this highest HbA1c level.

| Blood glucose variability
Blood glucose variability, measured by CGM as the coefficient of variation (CV), has been increasingly shown to be related to morbidity. The recommended goal for CV is <36%. 16

| Glucose time-in-range and time-above-range
Consensus guidelines recommend that at least 70% of time be spent with glucose levels between 70 and 180 mg/dL (3.9-10 mmol/L), considered 'time-in-range'. 16  greater percentage of time spent in hypoglycaemia. These trends did not achieve statistical significance. Figure 4 shows the data plotted as median and interquartile range. Review of individual CGM daytime patterns revealed large postmeal spikes in almost all subjects (data not shown as it was not composite data). Notably, the morning NPH dose did not usually appear to be preventing postlunch glucose elevation.

| Adverse events and feasibility
No major or minor adverse events related to the sensor either at insertion or during the study period were reported. Despite the fact that these are warm equatorial countries, the majority of patients did not have difficulty with sensor adhesion. They did not report significant skin irritation. The sensor was well tolerated, and many patients expressed the hope that CGM would become available to them clinically in the future.

| D ISCUSS I ON
Over the last decade, East Africa has seen major improvements in paediatric diabetes healthcare delivery, including the presence of Exchange data demonstrated that more frequent monitoring is associated with lower HbA1c levels in 1-to 26-year-olds. 18 ISPAD 2018 Guidelines recommend fingerstick blood glucose testing 6-10 times per day, 19 and the American Diabetes Association recommends glucose levels be tested at least four times per day (premeal and bedtime). 15 The current provision of about 2 test strips per day, while significantly better than in the past, is probably still not sufficient.
CGM is an alternative method of glucose monitoring that provides patients with extensive real-time data without requiring fingerpokes or test strips. It is gaining widespread use in high-income nations because it has been shown to decrease HbA1c and increase glucose time-in-range without causing hypoglycaemia. [20][21][22][23][24] In a meta-analysis of randomized controlled trials, the CGM-associated reduction in HbA1c was greatest in those with the highest HbA1c at baseline. All methods of blood glucose monitoring are expensive, but until recently CGM has been more expensive than fingerpoke capillary testing. A recent report from the United Kingdom suggested that flash CGM monitoring could be cost saving for people testing blood glucose levels 6 or more times per day, especially when taking into account potential reductions in the rate of severe hypoglycaemia. 25 While flash CGM systems are currently too expensive for common use in low-income settings, they are significantly less costly than other CGM systems and are expected to continue to come down in price. The current study shows that CGM is feasible and well accepted by patients in a resource-poor setting, suggesting potential future benefit both as a research tool and for clinical use.
Poor diabetes control in East Africa may be also related to problems with insulin. The average insulin dose of ~0.9 U/kg/d falls within the normal range observed for children with diabetes and does not suggest an underlying insulin resistance. 16  While recognizing the reality of financial limitations, it is imperative to find an appropriate balance between effective and safe treatment regimens and expense. Current practice standards arguably save costs in the short term, but the long-term implications are bleak.
With an average HbA1c ~11% (97 mmol/mol), East African children and young adults with T1D can be expected to experience acute and chronic diabetes complications which over time will result in high medical costs. There is a high cost to society as limited opportunities to obtain meaningful education and employment reduce their ability to become contributing members. 32,33 There are also high personal costs with an impact on the ability of people with T1D in these settings to marry, have children and live a satisfying life. 34 If analog insulin was shown to reduce the risk of life-threatening hypoglycaemia, then the ethical question would not be whether to provide this insulin in low-income countries, but how to make it affordable. Likewise, if CGM was shown to reduce HbA1c from the catastrophically high levels currently seen, this would also have to be figured into cost-benefit assessments. Similar cost-benefit questions were once asked about HIV/AIDS drugs, and due to a concerted international effort, these drugs are now widely available and affordable in low-income settings.
Such decisions must be guided by data obtained in children in the specific and unique settings found in low-income and low-middle income nations. More research in this area is necessary.
A range of possible mean glucose concentrations for a given HbA1c level has previously been reported in T1D, 35

especially in
African American subjects. 36 This is believed to be primarily due to individual variation in red blood cell life span. 37 We found that HbA1c may be a particularly unreliable outcome measure in East African populations, perhaps because haemolysis from malaria is common, there is a high frequency of sickle cell trait (~13%), 38 and there may be compromised protein nutritional status or iron deficiency, all of which can affect HbA1c values. 39 We plan further studies to explore these relations.
There are limitations to this study. While all participants were provided notebooks and pens and asked to keep a record of insu- In conclusion, children with T1D in many resource-poor regions including East Africa have seen remarkable progress in access to trained healthcare providers at dedicated diabetes clinics, engagement in diabetes education programmes, reliable availability of sufficient quantities of human insulin, and modest access to diabetes supplies. Despite these improvements, glucose metabolic control is still unacceptably poor, placing these patients at very high risk for serious acute and chronic diabetes complications.
Current methods of care are not adequately serving this disadvantaged population. As new therapeutic approaches are studied, our data suggest that CGM is a feasible tool for assessing research outcomes. Furthermore, as price comes down, CGM may be a viable clinical option.

ACK N OWLED G M ENTS
Abbott Diabetes Care, Alameda, California, donated FreeStyle Libre Pro flash glucose monitoring systems for this study.

CO N FLI C T O F I NTE R E S T
Nothing to declare.

AUTH O R CO NTR I B UTI O N S
AM and LMY designed the research study. All authors performed the research. LMY, EV and AM wrote the paper. All authors have edited and approved the final manuscript.

E TH I C A L A PPROVA L
The study protocol received IRB approval at the University of Minnesota and each of the East African sites prior to enrolling participants. Written informed consent was obtained from all participants. Parental consent was obtained for participants aged <18 years and assent for participants age 8-17. Consent and assent documents were provided in English and translated to the local language. Subjects were given $10 to cover travel expenses for study participation.