Understanding the relationship between the frequency of HbA1c monitoring, HbA1c changes over time, and the achievement of targets: a retrospective cohort study

Abstract

Background

The goal of post-diagnosis diabetes management is the achievement and maintenance of glycaemic control. Most clinical practice guidelines recommend 3–6 monthly HbA1c monitoring. Despite this guidance, there are few data supporting the impact of monitoring frequency on clinical outcomes, particularly from low- and middle-income country settings. This study evaluates the short-term impact of HbA1c monitoring intervals on changes in HbA1c levels, and the impact of adherence to monitoring recommendations on the achievement of HbA1c targets and HbA1c changes over time in a South African cohort.

Research design and methods

The study utilised routinely collected HbA1c test data on patients with diabetes aged ≥ 18 years in the Western and Northern Cape between 2015 and 2020. Two properties were calculated for each patient: the retest interval (the duration between consecutive HbA1c tests), and the monitoring adherence rate, which represents the proportion of retest intervals that met South Africa’s guideline recommendations. Mean changes in HbA1c by the retest interval were used to highlight the short-term impact of monitoring, while multistate modelling and linear mixed-effects modelling were then used to examine the effect of monitoring adherence on the achievement of glycaemic control targets and longitudinal changes in HbA1c.

Results

The 132,859 diabetes patients with repeat tests had a median of three HbA1c test requests, a median follow-up time of 2.3 years and a median retest interval of 10.3 months. A retest interval 2–4 months maximised the downward trajectory in HbA1c, while individuals with low adherence to the monitoring guidelines were the least likely to achieve glycaemic control in one year. Moreover, patients with low monitoring adherence had higher mean HbA1c levels compared to patients with moderate or high monitoring adherence.

Conclusions

The results from this study illustrate the importance of adherence to monitoring recommendations as adherent patients had better glycaemic control and lower mean HbA1c levels over time.

What is already known on this topic

There are few in-depth studies investigating the impact of HbA1c monitoring frequency on clinical outcomes with none of the studies coming from low-and-middle-income countries.

What this study adds

This study assesses the impact of current HbA1c monitoring recommendations in South Africa by evaluating associations between glycaemic control, HbA1c changes over time, and adherence to the recommended HbA1c monitoring frequencies.

How this study might affect research, practice or policy

The study may be used as supporting evidence for current HbA1c monitoring guidelines, which are largely based on expert opinion and serves as a clarion call for improving population HbA1c testing and reduce testing delays.

Introduction

Diabetes Mellitus (DM) has a rapidly increasing burden across low- and middle-income countries (LMICs). According to the International Diabetes Federation (IDF) [1], up to 81% of the 537 million adults living with diabetes in 2021 were in LMICs. Given the epidemiological disease burden transition taking place in LMICs, along with rapid urbanisation and associated harmful lifestyle changes, the greatest increases in diabetes prevalence are expected to occur in these settings [1, 2]. In addition, health system resource scarcities lead to substantial excess mortality and poorer disease management [3, 4]. Given how diabetes interacts with susceptibility to some infectious diseases (e.g., COVID-19 and tuberculosis) and worsens outcomes [5], effectively managing diabetes in LMICs is imperative.

Managing diabetes includes components related to prevention, effective screening and diagnosis, regular follow-up, and lifelong pharmacological and non-pharmacological treatment (physical activity, diet and education) [5]. The goal of post-diagnosis management strategies is to achieve and maintain glycaemic control. Poor diabetes management, and consequently, glycaemic control, is associated with increased risk of both microvascular and macrovascular complications, which impose a major drain on health resources, as well as worse clinical outcomes including mortality [6,7,8]. While glycaemic control goals should be individualised the threshold for long-term glycaemic control for most adults living with diabetes a glycated haemoglobin (HbA1c) level < 7% (< 53 mmol/mol). HbA1c is a reliable measure of average blood glucose concentration over the preceding 2–3 months. It is generally regarded as the gold standard for long-term blood glucose management as it correlates well with the risk of long-term diabetes complications [9]. Routinely monitoring HbA1c can therefore support treatment adherence, improve selection and titration of treatments, and provide an entry point for patient education about factors that may alter glycaemic status [5]. Despite this, the impact of monitoring, as a proxy for comprehensive patient care, including engagement and access to healthcare services, remains under-explored particularly in low- and middle-income countries.

Guidelines from professional organisations, such as the UK National Institute for Health and Clinical Excellence (NICE) and the American Diabetes Association (ADA), recommend testing HbA1c at 6-monthly intervals (NICE) or at least two times a year (ADA) for patients meeting treatment goals, and with stable glycaemic control, while patients who are not meeting glycaemic control goals should be tested 3–6 monthly (NICE) or quarterly (ADA) [10, 11]. However, unlike recommendations for glycaemic control targets that are based on high quality evidence from randomized controlled trials [12, 13], recommendations for frequency of HbA1c monitoring are largely based on expert opinion, as there are few in-depth studies investigating the impact of testing frequency on clinical outcomes [12, 13]. Moreover, the evidence base for glycaemic control targets, and the available studies on the impact of testing frequency are all from high-income countries (HICs) which may not be generalisable to LMICs due to context-specific factors [12]. These factors, which are relevant for other aspects of T2D management, include variations in healthcare infrastructure, accessibility to healthcare services, socioeconomic disparities, competing health priorities, behavioural factors, and differing healthcare policies [13,14,15]. This study, therefore, evaluates the short-term impact of HbA1c monitoring intervals on changes in HbA1c levels. Additionally, the study aims to provide evidence supporting current testing recommendations in South Africa by examining associations between glycaemic control, longitudinal HbA1c changes, and adherence to recommended monitoring frequencies.

Methodology

Study setting and data source

Guidelines for the management of Type 2 diabetes (T2D), which accounts for more than 90% of all diabetes cases in South Africa, are provided by the Society for Endocrinology, Metabolism and Diabetes of South Africa (SEMDSA) [16], and the National Department of Health [17] Recommendations from SEMDSA for the glycaemic control target and HbA1c monitoring frequency are in line with those from NICE and the ADA. South Africa’s Adult Primary Care (APC) guidelines for 2019–2020, however, set the target for glycaemic control for most patients is < 8% (63 mmol/mol), while the recommended HbA1c monitoring frequency is annually in well-controlled patients, and three months in uncontrolled patients or whenever there is a change in treatment [17].

For this analysis, we reviewed individual-level data collected as part of routine clinical practice between October 2015 and February 2020 in the Western and Northern Cape, South Africa, by the National Health Laboratory Service (NHLS). The NHLS is the largest diagnostic pathology service in South Africa, providing laboratory testing to public sector patients (over 80% of the South African population attends public sector services) through a national network of laboratories [18]. Minimal clinical and individual characteristics were available: date of test, demographic information (age, sex, date of birth), the location of test request (healthcare facility), and a unique patient identifier.

Definitions

An HbA1c cut-off of ≥ 6.5% (48 mmol/mol) was used to indicate clinical diabetes.17 We define the index test as the first test with HbA1c ≥ 6.5%. Once an individual experienced “diabetes onset”, we ignore any data prior to the index test. In addition, individuals continued to be categorised as having diabetes even if they subsequently reduced their HbA1c below the threshold.

Inclusion/exclusion criteria

All individuals classified as having diabetes, and aged ≥ 18 years on the index test, were eligible for inclusion in the analysis. Individuals with only one test in the eligible dates are reported but excluded from subsequent analysis. Figure 1 presents a flow diagram indicating the numbers included and excluded from the analytical study cohort.

Fig. 1

Patient inclusion diagram for study cohort

Statistical analysis

We used the index HbA1c as a proxy for optimal/suboptimal control. Specifically, we split the cohort into two, based on the index HbA1c, where patients with an index HbA1c ≥ 7% were classified as “sub-optimally controlled” and those with index HbA1c < 7% as “optimally controlled”. Descriptive summary statistics were conducted by using frequencies and proportions for categorical variables and mean (standard deviation) or median (interquartile range) for continuous variables. Differences in baseline characteristics between patients with index HbA1c ≥ 7% / index HbA1c < 7% were analysed using the independent t-test or the Wilcoxon rank-sum test for continuous variables, and Chi-squared tests for categorical variables. The Shapiro Wilks test was used to assess the normality of continuous variables.

To aid in investigating the outcomes of interest, we adopted procedures previously used on HIC data [7, 12, 19]. We created two variables: the retest interval and the monitoring adherence rate. The retest interval is the interval between consecutive HbA1c requests and was categorised into 1-month blocks [12].

The monitoring adherence rate (%) represents the proportion of retest intervals that met the South African guideline recommended intervals. It was calculated for each patient by dividing the number of HbA1c tests performed within the recommended intervals (n) by the total number of the conducted tests (N) minus 1 (to reflect how the first test has no preceding test) during the study period and multiplying by 100. Therefore: (n/(N − 1) × 100). The adherence rate was then categorised into three equally spaced categories: Low, Moderate, High [19]. This definition of monitoring adherence accounts for the different monitoring frequency recommendations in both poorly controlled, and well controlled patients. A multi-state model, under Markov assumptions, was implemented to estimate the effect of adherence to recommended guidelines on the achievement of glycaemic control targets. A diabetic patient’s status was classified in one of two states: State 1: HbA1c < 7% and State 2: HbA1c ≥ 7%. Additional covariates assessed were age, sex and baseline HBA1c.Separate models were fit for sub-optimally controlled patients as classified at index test.

Linear mixed-effects models with random intercepts were also fitted to HbA1c levels. The regression models included the adherence rate, a categorical variable for time from baseline, and other available predictors of HbA1c (sex, age). As we expected departures from the assumption of linearity between HbA1c levels and age, we used natural cubic splines, with interior knots at the 25th, 50th and 75th quantiles, to model the non-linear effects of age. As with the multi-state models, separate models were fitted based on grouping by the index test. All analyses were conducted using R version 4.04 (R Foundation for Statistical Computing, Vienna, Austria). The specific packages used were msm to fit multi-state models and nlme for the linear mixed-effects models [20, 21].

Sensitivity analysis

As part of a sensitivity analysis, we explored an alternative cut-off for optimal/suboptimal control. This analysis classified patients with an index HbA1c ≥ 8% (64 mmol/mol) as “sub-optimally controlled” and those with index HbA1c < 8% as “optimally controlled”, corresponding to South Africa’s Adult Primary Care (APC) guidelines. The monitoring adherence rate (%) was adjusted accordingly. The analyses described above were then checked to see if results differed.

Results

The characteristics of the 132,859 PLWD included in the study are given in Table 1. Overall, the mean (SD) age at the index test was 54.1 (12.7) years with 65.6% of the patients being female. The median (IQR) HbA1c level of the index test was 9.1% (7.5–11.1). The cohort had a median (IQR) of 3 (2–5) tests per patient, 10.3(5.5, 14.7) months between tests, and 2.3(1.4,3.2) years of follow-up. Adherence to South Africa’s HbA1c testing guidelines was classified as high in only 5,864(4.4%) of patients, or while close to 85% of the cohort had little to no monitoring, a percentage which excludes PLWD with no follow up test available. The proportion of patients with high monitoring adherence was higher for patients with an index HbA1c < 7% compared to those with an index HbA1c ≥ 7% (9.7% vs 3.6%, respectively).

Table 1 Patient and test characteristics

Effect of the retest interval on percentage change in HbA1c

Figure 2A and Fig. 2B provide the frequency of monthly retest intervals, and the relative percentage rise or fall in HbA1c by retest interval, respectively. The most common retest interval for the cohort was > 18 months which ranges from 18.5 months to 53.4 months. From Fig. 2B, we see that retest intervals < 6 months are generally associated with relative reductions in HbA1c, with a retest interval between two and four months appearing to maximise the downward trajectory in HbA1c, while a retest interval greater than nine months suggests increases in HbA1c levels. Pairwise comparisons suggest no statistically significant difference between retest intervals of 1–2 months, 2–3 months, and 3–4 months.

Fig. 2

A The frequency distribution of HbA1c retest intervals among diabetic patients in the study cohort and B the relative percentage change in HbA1c by the retest interval (1-month blocks)

Achievement of targets

The results from the multi-state model to estimate associations of covariates on the transition from HbA1c ≥ 7% to HbA1c < 7% (indicating the achievement of glycaemic control) are presented as hazard ratios in Table 2 for the entire study cohort, and for patients with an index HbA1c ≥ 7%. In both the full cohort and for patients with an index HbA1c ≥ 7%, patients aged 25–34 years, 35–44 years and 65 + years were more likely to achieve glycaemic control targets when compared with patients aged 18–24 years, while there were no differences in the likelihood of achieving glycaemic control targets between the 18–24 years age group and the remaining age groups. Males were 1.33(1.29–1.38) times more likely to achieve glycaemic control, in the full cohort, and 1.28 (1.23–1.33) times more likely to achieve glycaemic control for patients with an index HbA1c ≥ 7%, when compared to females.

Table 2 Results from the multi-state model for the transition from HbA1c ≥ 7% to HbA1c < 7% indicating the achievement of glycaemic control for the entire cohort and patients with index HbA1c ≥ 7%

For monitoring adherence, patients with moderate and high adherence were 2.72 (2.61–2.83) and 7.82 (7.28–8.4) times more likely to achieve glycaemic control, respectively, compared to patients with low monitoring adherence classification. For patients with index HbA1c ≥ 7%, patients classified as having moderate, and high monitoring adherence are more than two times (HR 2.7; 95% CI 2.58–2.82), and close to eight times (HR 7.49; 95% CI 6.96–8.07), more likely to achieve glycaemic control, compared with those with a low monitoring adherence classification, respectively.

Figure 3 provides a graphical representation of the probability of achieving the HbA1c target in one year for the entire cohort, and patients with Index HbA1c ≥ 7%. Overall, the probability that an individual with poor index glycaemic control (HbA1c ≥ 7%) will achieve glycaemic control in a year’s time is 0.11. Holding the other explanatory variables fixed at their mean value, males have a slightly higher probability of achieving control in one year than females (0.17 vs 0.14), while individuals aged 25–35 years have the highest probability of achieving glycaemic control (0.24). In terms of adherence to South Africa’s testing guidelines, individuals with little to no adherence to the monitoring guidelines (“Low monitoring adherence”) are the least likely to achieve glycaemic control (probability of 0.14), while those with “high monitoring adherence” have a much higher probability of achieving glycaemic control (0.5). This probability increases to 0.53 when the focus is solely on patients with an index HbA1c ≥ 7%, as shown in Fig. 3.

Fig. 3

Estimated 1-year transition probabilities: The plot gives the probability of transitioning from HbA1c ≥ 7 to HbA1c < 7 (95% CI) in one year overall and by each of the explanatory variables holding the values of other explanatory variables at their mean value

Longitudinal change in HbA1c by monitoring adherence

Results from the mixed-effects models estimating the mean differences in HbA1c are presented in Table 3. These results are presented as mean absolute differences in HbA1c levels (%). Mean HbA1c levels (%) for males were lower than females for both the full cohort and the group with index HbA1c ≥ 7%, by 0.13% (p < 0.001) and 0.14% (p < 0.001), respectively. In patients with index HbA1c < 7%, mean HbA1c levels for males were 0.06% (p < 0.001) higher than for females. As interpreting spline coefficients is not straight forward, we used effect plots to show the change in HbA1c by age. The effect plots are found in Figures S2-S4. Overall, HbA1c levels are higher in younger patients and gradually decrease with age. However, the rate of decrease becomes smaller with age. For patients with index HbA1c < 7%, HbA1c is approximately constant up to age 60 years, gradually decreases between the ages 60 years and 80 years, then increases slightly thereafter.

Table 3 Estimated mean differences from linear random effects model for HbA1c(%) levels

For the full cohort, patients with low monitoring adherence have 0.34% (p < 0.001) and 0.92% (p < 0.001) higher mean HbA1c compared to patients with moderate and high monitoring adherence, respectively. Patients with moderate or high monitoring adherence also have lower mean HbA1c when compared to patients with low monitoring adherence for the group with index HbA1c ≥ 7% (Moderate: 0.19%, High: 0.5%) and the group with index HbA1c < 7% (Moderate: 0.13%, High: 0.28%).

Results from the sensitivity analyses are available in Figure S6-S11, and Tables S2-S5 of the Supplementary material. Overall, the results were consistent with the main results as using different cutoff points for the definition of optimal/suboptimal control did not affect the trend of the results obtained, though there were differences in the values.

Discussion

As HbA1c testing guidelines are largely based on expert consensus, validation of the frequency of monitoring, and the benefits of repeated monitoring on patient outcomes in different settings, has been the subject of ongoing research[13, 20] Our study extends the work conducted by Duff et al., Driskell et al., and Imai et al. [7, 12, 19], who all examined the association between HbA1c testing frequency and glycaemic control, albeit using different methodologies. Our results illustrate that patients with moderate or high adherence to testing recommendations have a higher probability of achieving glycaemic control targets, and lower mean HbA1c over time, when compared to patients with low adherence. Moreover, we identified that a retest interval of between two and four months, is associated with the greatest reduction in HbA1c.

Our results are in keeping with those obtained from the studies mentioned above. Driskell et al. similarly found that testing quarterly was the optimum testing frequency to maximise the downward trajectory in HbA1c for PLWD, particularly among those starting with HbA1c > 7% [12]. In the study by Imai et al., data collected from Australian general practices was used to examine the association between adherence to the recommended HbA1c testing frequency and patient outcomes [19]. Applying generalised additive mixed models (GAMMs) to the data, they found that HbA1c levels for patients with low monitoring adherence gradually increased or remained inadequately controlled, while HbA1c levels for patients with high adherence remained controlled or improved over time.

Duff et al. examined the effect of the number of HbA1c tests in a year on the achievement of glycaemic control targets and on HbA1c changes over time, on sub-optimally controlled patients [7]. They found that having one test per year was associated with lower likelihoods of achieving glycaemic control and relatively higher mean HbA1c over time, while testing two or three times a year was equally as effective as testing four times a year. In addition, they estimated an overall probability of achieving glycaemic control of 0.20, which is much higher than the 0.097 estimate from our study for a similar population. The higher median HbA1c for sub-optimally controlled patients in our cohort potentially explains this difference. However, it may also be indicative of other issues including barriers to diabetes-care access, and delays to initiation or intensification of diabetes treatment in South Africa [22].

There are a few additional studies that have attempted to provide some validation of HbA1c testing guidelines and establish a relationship between frequency of monitoring and glycaemic control [23,24,25,26,27]. Overall, there is consensus that there is an inverse association between frequency of monitoring and glycaemic control. However, some studies propose longer retest intervals, particularly among well-controlled PLWD, citing that excessively frequent testing may lead to unnecessary regimen changes, which may contribute to adverse effects including hypoglycaemia, a greater treatment burden, and higher healthcare costs [23, 26]. In this study, we found that both moderate and high monitoring adherence among “controlled” PLWD were associated with reductions in mean HbA1c when compared to low adherence. This was the case under the SEMDSA guidelines (6-monthly testing), and South Africa’s APC guidelines (annual testing), corroborating the proposals by Ohde et al. [26]. Further research on this, however, is required.

There are some limitations to the analysis that warrant consideration. Firstly, the laboratory data used do not differentiate between type 1, type 2, or gestational diabetes. However, as we excluded patients under 18 years (majority of type 1 diabetes cases) and since the most diabetes cases for patients 18 years or older are T2D, we can assume that most patients were people living with type 2 diabetes. Additionally, routine monitoring data from public sector laboratory records provide minimal additional information, lacking details on co-morbidities, medications, and other clinical factors. As such, laboratory data do not account for any lifestyle or pharmacological interventions, other risk factors (e.g., ethnicity, duration of diabetes, smoking status), or other unmeasured confounding which may influence HbA1c progression, the frequency of monitoring, and the achievement of therapeutic goals. Secondly, there are issues with generalisability of the results, as we used data from 2 out of 9 provinces in the country. Moreover, selection bias may have occurred due to non-representative patient inclusion, variability in healthcare access or provider practices, and loss to follow-up, potentially limiting the generalisability of the findings to the broader population. Thirdly, splitting the cohort into the optimally/sub-optimally controlled groups based on the index HbA1c is an oversimplification as it is possible for some patients who were sub-optimally controlled before the study period started to become optimally controlled by the time the study started, and vice versa. Moreover, given how glycaemic control targets for PLWD in South Africa are largely individualised, defining glycaemic control as HbA1c < 7% is slightly misleading. To offset this potential source of bias, we ran a sensitivity analysis with different definitions of glycaemic control. Lastly, there are inherent methodological limitations associated with the selected study design and the use of hazard ratios in the analysis. Furthermore, the models employed did not adequately account for clustering by healthcare facilities, which may have influenced our findings based on facility-specific factors.

Despite these limitations, our study has several major strengths. To our knowledge, this analysis is the first attempt to provide evidence on the impact of adherence to recommended HbA1c testing intervals in an African setting. In addition, by using routinely collected data, without requiring additional or specialised surveys, we were able to leverage an important source of real-world data.

This study demonstrates how conformity to recommended testing intervals is associated with glycaemic control. While the frequency of HbA1c monitoring itself does not necessarily improve diabetes control, it can be viewed as a proxy for the interaction with the health care system where interventions that contribute to achieving glycaemic control can take place. Though our findings do not imply causality, they may be used as supporting evidence for current HbA1c testing guidelines, which are largely based on expert opinion and clinical consensus [12, 13]. Low monitoring adherence, coupled with low rates of follow-up testing, signified by the more than 41% of the study’s diabetes patients who had only one test, are causes for concern, as they illustrate the underuse of HbA1c testing among diabetes patients in South Africa. In addition, the proportion of male diabetes patients receiving HbA1c tests was much lower than among females, reinforcing the notion of lower rates of health-care utilisation among males in South Africa [28]. As there are many key factors that can influence patient engagement in routine care, including healthcare system factors such as accessibility, and care coordination, as well as patient-related factors like financial constraints, health literacy, and competing priorities [29], the design of multifaceted interventions targeting both patients and providers is imperative to improve population HbA1c testing rates and reduce testing delays [25].

Conclusions

Our findings contribute to a growing body of literature on the impact of adherence to current monitoring recommendations. We observed that adherence to HbA1c monitoring is associated with better glycaemic control, suggesting that frequent monitoring reflects broader aspects of care, including patient engagement and clinician interaction. These data underscore the importance of regular HbA1c testing as part of a holistic approach to diabetes management in middle-income settings such as South Africa.