Discussion
The international guidelines recommend HbA1c for monitoring glycemic control and blood glucose test where HbA1c is unavailable. Despite this guidance, there remains concerns about the accuracy of HbA1c in populations with high frequency of other medical conditions that may alter its reliability. In this study, we used CGM to compare the accuracy of HbA1c, FPG and RPG tests in assessing glycemic control among patients with diabetes under conditions of everyday life in low-resource settings. The prevalence of other medical conditions that may alter HbA1c reliability was remarkably high. However, we found that HbA1c remained the most accurate test of average glucose control, despite the high prevalence of hemoglobinopathies, anemia and renal impairment. Similarly, FPG and RPG demonstrated reasonable accuracy as measures of average glycemic control, providing confidence that glucose tests provide a good measure of glycemia where HbA1c is not available. Furthermore, the very modest loss of diagnostic test performance using RPG provides some reassurance for use of this test in situations where a RPG is the only or most practical measure available.
In the current study, we have compared FPG, RPG and HbA1c in the same study and more importantly against an independent measure of day-to-day glycemic burden. CGM was used as an independent marker of glycemic burden to allow assessment of the relative performance of HbA1c, FPG and RPG in assessing glycemic burden. This is a major strength of our analysis in contrast to previous studies which have compared between measures such as HbA1c and FPG, with no independent comparison. Further, we assessed performance of HbA1c in the presence of other medical conditions that may alter its effect. This gave us the opportunity to assess the overall impact on HbA1c reliability.
However, the present study has some limitations that should be taken into consideration. First, although CGM was the best available option for direct measurement of glucose in day-to-day living and allowed us to compare the relative performance of HbA1c and glucose tests, it should be noted that glycemia was measured using a CGM sensor over median 14 (IQR 13–14) days and yet HbA1c estimates glycemia over a longer duration.20 Second, we used HbA1c immunoassay, one of the most widely used HbA1c assays, particularly in low-resource settings. However, our results for the performance might not apply to other HbA1c assay types, which are known to have different susceptibility to the effects of hemoglobinopathies.21 Furthermore, although we screened for a number of potential comorbidities thought to alter HbA1c, with the available sample size and very modest subgroup numbers, we were unable to do further subgroup analyses to assess the impact of other individual underlying non-glycemic conditions.22 In addition, the impact of glucose-6-phosphate dehydrogenase variants, another common condition that may affect HbA1c results reliability, was not assessed.23
Our results showing a strong relationship between HbA1c and mean glucose from CGM are consistent with studies that have compared these two measures in high-income settings. The Diabetes Control and Complications Trial (DCCT) of participants in the USA with type 1 diabetes showed a strong relationship between the mean plasma glucose and HbA1c with a Pearson correlation (r) of 0.82.24 Similarly, results from the ADAG (A1c Derived Average Glucose) study, which included 507 participants with and without diabetes predominantly from the USA and Europe, and excluded participants with other medical conditions thought to alter HbA1c reliability, showed HbA1c and mean glucose were closely correlated (r=0.89, p<0.0001).25 Our similar results (r=0.88) in an African population, and without exclusion of participants with analytical concerns for HbA1c measurement, is reassuring for the use of HbA1c testing in this region.
Our results are broadly consistent with previous studies that have reported the relationship between glucose tests and HbA1c. El-Kebbi et al showed, in 1827 predominantly African–American living in the USA, that RPG collected 1–4 hour post meal was correlated strongly with HbA1c, although in this predominantly insulin-treated population, the correlation (r=0.63) was lower than observed in our study (0.74).26 In a study that compared both FPG and RPG to HbA1c among 1000 patients with diabetes living in India, FPG showed a better correlation with HbA1c than RPG (0.739 vs 0.601).27 In contrast, in studies where a fixed post meal time point was used, RPG was a slightly better correlate of HbA1c than FPG.18 Unfortunately, studies comparing performances of glucose tests against HbA1c in Africa are very few, with small sample sizes, and in these studies, the impact of common medical conditions that may alter HbA1c reliability was not assessed.17 28
Our data suggest that there is a high prevalence of other medical conditions that may alter HbA1c reliability justifying the questioning of HbA1c utility. However, even with these comorbidities, HbA1c, when measured with an immunoassay method, correlated strongly with mean glucose, outperforming glucose measures, and only displayed a modest improvement when patients with comorbidities were excluded. This suggests that HbA1c remains the optimal laboratory method of monitoring glucose burden even where prevalence of conditions that may affect its reliability is high. The strong correlation of HbA1c with glucose despite the prevalence of other medical conditions that may alter HbA1c reliability deserves further exploration. However, there are some reasons why the impact of these conditions on HbA1c reliability may be modest in this setting. First, in line with the National Glycohemoglobin Standardization Program (NGSP) recommendation, modern HbA1c immunoassays are not directly affected by the presence of hemoglobin variants like HbAS.21 Second, while comorbidities that affect red cell life will alter the accuracy of any HbA1c method, the predominant hemoglobinopathy in our study population was HbAS (sickle cell) trait, and previous research has been conflicting as to whether this meaningfully alters red cell lifespan.29
While our results support the use of HbA1c (where available) rather than glucose measures in LMIC populations, the small subgroup numbers in our study limited the power to definitively determine the impact of some of these comorbidities on HbA1c performance. To accurately determine the impact of individual comorbidities, larger multinational studies involving other regions in Africa and LMICs with enrichment for these comorbidities would be needed. Furthermore, while our data show that HbA1c (measured using an immunoassay method) has the closest relationship with average glucose, even with comorbidities, it is possible that the overall relationship between glucose and HbA1c is different in this population, therefore the thresholds used internationally are not appropriate, and bespoke HbA1c thresholds are needed for different populations. This further underscores the need for much larger studies, ideally incorporating risk of microvascular complications, to determine whether the HbA1c targets used internationally are appropriate for LMIC populations.
In conclusion, our results suggest that HbA1c is the optimal test for monitoring glucose control even in LMICs where medical conditions that may alter its reliability are prevalent; FPG and RPG are valuable alternatives where HbA1c is not available.