Effect of early glycemic control on HbA1c tracking and development of vascular complications after 5 years of childhood onset type 1 diabetes: Systematic review and meta‐analysis

Abstract Objective A systematic review and meta‐analysis was conducted to investigate if glycemic control measured by glycated hemoglobin (HbA1c) levels near diagnosis are predictive of future glycemic outcomes and vascular complications in childhood onset type 1 diabetes (T1D). Methods Evidence was gathered using electronic databases (MEDLINE, EMBASE, Web of Science, CINAHL, Scopus, and Cochrane Library up to February 2017) and snowballing techniques. Studies investigating the association between the exposure “early glycemic control” and main outcome: “tracking of early control” and secondary outcome: risk of future complications; in children and young people aged 0 to 19 years at baseline; were systematically double‐reviewed, quality assessed, and outcome data extracted for synthesis and meta‐analysis. Findings Five studies (N = 4227 participants) were eligible. HbA1c levels were sub‐optimal throughout the study period but tended to stabilize in a “track” by 6 months after T1D diagnosis. The group with low HbA1c <53 mmol/mol (<7%) at baseline had lower long‐term HbA1c levels than the higher HbA1c group. The estimated standardized mean difference between the sub groups showed a reduction of HbA1c levels on average by 1.6% (range −0.95% to −2.28%) from baseline. Only one study investigated the association between early glycemic control and development of vascular complications in childhood onset T1D. Interpretations Glycemic control after the first few months of childhood onset T1D, remains stable but sub‐optimal for a decade. The low and high HbA1c levels at baseline seem to “track” in their respective tracks during the 10‐year follow‐up, however, the initial difference between groups narrows over time. PROSPERO: CRD42015024546 http://www.crd.york.ac.uk/PROSPERO/display_record.asp?ID=CRD42015024546

baseline seem to "track" in their respective tracks during the 10-year follow-up, however, the initial difference between groups narrows over time.

| INTRODUCTION
Glycated hemoglobin (HbA1c) levels, a measure for glycemic control is the main predictor of long-term type 1 diabetes (T1D) outcomes. [1][2][3] HbA1c levels are highest at diagnosis, but improve after insulin treatment and remain stable in most T1D patients. However, a few find it challenging to maintain good glycemic control despite targeted or intensive interventions, as they go through various stages in life. 4,5 Studies mainly in adults have shown a link between poor glycemic control in the early phase following T1D diagnosis and long-term HbA1c levels, with an increased risk of developing vascular complications and mortality. 6,7 The risk of vascular complications is likely to be greater for childhood onset T1D, because of a longer duration of glycemic exposure 8 and pathophysiological factors, such as reduced insulin sensitivity and psychosocial behaviors, such as insulin omission. [9][10][11] For childhood onset T1D, some observational studies indicate an association between poor glycemic control within 1 or 2 years of diagnosis and vascular complications in later life. [12][13][14] Others suggest that mean HbA1c levels nearer to diagnosis are predictive of HbA1c levels in the subsequent years, even lifetime, regardless of the type of insulin regimen. [15][16][17] This phenomenon, also known as glycemic "tracking," is poorly understood. 18 It is unclear exactly when and in whom the phenomenon of "tracking" of HbA1c occurs in childhood onset T1D and if it is because of the natural history of T1D. It is therefore important to investigate the evidence on this phenomenon to identify if there exists a window period in the initial phase of T1D diagnosis, during which appropriate resources could be mobilized to deliver targeted interventions to those at risk of developing poorer long-term glycemic outcomes and vascular complications.
The purpose of our study was to carry out a systematic review and meta-analysis of the evidence assessing the impact of early glycemic control in children (followed for at least 5 years from diagnosis) on tracking of early control and the risk of developing vascular complications.

| METHODS
This review is part of a series of systematic reviews of evidence on the effects of early glycemic control in childhood onset T1D. The review protocol was registered in PROSPERO (Registration number: CRD42015024546) and a detailed protocol published. 19 We followed the review methods for the rigorous conduct and reporting of systematic reviews for policy and practice as described by the Evidence for Policy and Practice Information (EPPI) Centre 20 which are as per PRI-SMA guidelines. 21

| Search strategy
A refined search strategy was designed after a number of initial iterative scoping searches, with input from experts in the field to maximize capturing of key publications. Three sets of search terms were used relating to population (children and young people diagnosed with T1D), exposure (terms to capture observational, intervention, qualitative studies, and review articles relating to early diabetes control) and outcome (complications, mortality, glycemic tracking i.e., metabolic memory) (Additional File 1).

| Study selection
Interventional and observational studies with a follow-up of ≥5 years from diagnosis of T1D which described and quantified the association between early glycemic control (defined as glycemic control within 2 years of diagnosis of T1D) AND long-term glycemic tracking (defined as settling of HbA1c levels into long-term tracks of either > or <7% ie, 53 mmol/mol) and risk of future complications in children and young people aged 0 to 19 years at baseline were included (Additional File 3).
In addition to running electronic database searches in parallel (HC and VMP), sub-samples of papers were double-reviewed (DC and VMP), at each stage of the review process (title and abstract screening, data extraction and quality assessment). The interrater reliability for study selection was substantial. 22 Full texts of abstracts appearing to meet the inclusion criteria were retrieved and their status was recorded in a pre-piloted excel spread-sheet, which included specific study details and reasons for exclusion (for excluded studies). No foreign language papers were identified. Articles were re-examined (DC and VMP) if there was uncertainty about inclusion criteria and disagreements were resolved at team meetings.

| Data extraction
Data from included studies were extracted, analyzed, and synthesized by one reviewer (VMP). A proportion of shortlisted studies were also independently double reviewed and data extracted (DC and RA).

| Quality assessment
The quality of included studies was assessed independently by two reviewers (DC and VMP) using the quality assessment criteria by the EPPI Centre. 20 Any disagreements were resolved by consensus.
Scores were based on six items focusing on both internal and external validity (Additional File 5). Observational studies were classified as high (≥5), intermediate 3,4 or low (≤2) quality based on the number of quality criteria met out of a maximum assessment score of six.

| Statistical analysis
Information extracted from included studies were summarized through descriptive narrative synthesis and meta-analysis. 23 All statistical analyses were conducted by one reviewer (VMP) and were verified by a second reviewer (JB). The sample size, mean HbA1c measurements and SD or SE were available at population level and/or for categorized low and high HbA1c groups. Where not reported, the SE of the study at each time point was calculated using the reported SD and the group sample sizes. Baseline period included 3 to 6 months from T1D diagnosis. Mean HbA1c levels at diagnosis was not included in the main meta-analysis as by definition they were measured prior to exposure of glycemic control with insulin therapy.
The effect sizes and their SE were divided with SD to obtain standardized mean differences (SMD). 24 The primary outcome was the population mean HbA1c level at baseline (0, 3, and 6 months of diagnosis), 1, 2, 3, 5, 7, and 10 years follow-up. A further primary outcome was the difference in HbA1c levels between the low HbA1c (<7% at baseline) group (considered the "treated/exposed" group) and the high HbA1c group (≥7% at baseline) (the "control" group), reported as standardized mean differences. If multiple measurements of HbA1c were reported at follow-up then these measures were combined within each study before metaanalysis. Heterogeneity between studies was expected and therefore both fixed effects (FE, inverse variance) and random effects (RE, Dersimonian, and Laird) models were used to pool the effect sizes and reported using forest plots. 25 The heterogeneity between studies was assessed using the χ 2 test for heterogeneity and I 2 statistics. 26 The meta-analyses were carried using the metan command in STATA 15, StataCorp, College Station, Texas.
For glycated hemoglobin, the estimated pooled standardized mean differences were converted into absolute units, to facilitate clinical interpretation, by multiplying the estimate by the pooled SD of all included studies of the meta-analysis.
Furthermore, the long-term population average HbA1c trajectory from each study was plotted alongside the overall estimate at all-time points of follow-up obtained from the meta-analysis. The trajectories of HbA1c sub groups (low v/s high) in each study were also plotted.
The robustness of the meta-analysis to the choice of metaanalysis model was assessed by comparing FE and RE pooled standardized effect sizes. In a sensitivity analysis we excluded studies in pre-school children.
Assessing publication bias using the funnel plots, the Begg's rank correlation test or the Egger's linear regression test was deemed inappropriate as there were insufficient studies included in the review.
Because of the small number of included studies, meta-regression was not appropriate to explore heterogeneity between studies or to investigate if there were other potential factors that could be independently associated with long-term glycemic control. A minimum of 10 studies per study level parameter would be needed for meta-regression.
Only one included study assessed the association of micro and macro-vascular complications with early glycemic control, which precluded a meta-analysis and results of which were narrated separately.

| RESULTS
The literature search strategy on glycemic control in childhood onset national, 24 regional, 27 Children's hospital, 29 academic medical centre 30 and clinic 28 level data.

| Characteristics of included studies
The Swedish cohort study 24  The Israeli study 28 was a retrospective observational study, investigating HbA1c tracking in 173 mainly Jewish (84.4%) preschool aged children (84 males) aged 0.5 to 6.5 years at diagnosis between 1993 and 2009 at a tertiary level diabetes clinic in Israel, with a median T1D duration of 4.3 years (range 1 to 11 years) and followed up for 7 years from T1D onset. All patients were advised on carbohydrate counting, required to perform >6 self-blood glucose measurements per day and both multiple daily injections and insulin pumps were used.
Further details of the data extracted from the five studies included in the systematic review are in Table 1.

| Study quality
The quality of the observational studies was intermediate to high.
Two studies were assessed to be "high" quality with a score of five each 24,29 Table 2).
The study in pre-school aged children (mean age at diagnosis 3.8 ± 1.6 years) showed better control than the other studies with older children. 28 The heterogeneity levels were significantly high (P = 0.001) at 1, 2, 3, and 5 years after diagnosis and were lower at follow-up time points 0.5, 7, and 10 years after diagnosis (P > 0.7) in the meta-analysis.
The meta-analysis was repeated after excluding the study in preschool aged children (Supplementary Figure 1). The pooled standardized mean difference in HbA1c levels between patients in the low HbA1c group and those in the high HbA1c group with 95% CI was slightly lower at −1.  (Table 2).
Comparing the long-term HbA1c trajectories between studies revealed that the Israeli study in pre-school children yielded better longterm control (Supplementary Figure 2). Individual study results suggest that early glycemic control tracks during the follow-up in the initially low and high HbA1c groups (Supplementary Figure 3).
Because there were only five studies in the review, we could not assess publication bias using the funnel plot, the Begg adjusted rank correlation test or the Egger test as there was insufficient power to distinguish real asymmetry from random chance.

| Association of early HbA1c levels and complications risk
Only one longitudinal study 24 investigated the association of early glycemic control and future complications and met the inclusion criteria for the systematic review. The study, adjusted for gender, T1D dura-

| DISCUSSION
We identified five longitudinal studies investigating the impact of early glycemic control on long-term glycemic control in children and adolescents (<19 years) followed from diagnosis of T1D. In the meta- The meta-analysis of the four studies comparing initially low v/s high HbA1c groups, indicates that the low HbA1c group showed overall slightly improved control than the high HbA1c group during the study period. In addition, the meta-analyses suggests that the overall glycemic control was stable in a "track" after 6 months of childhood onset T1D diagnosis. The low and high HbA1c levels at baseline also seem to "track" in their respective tracks during the 10-year followup. However, the initial difference between groups narrows over time.
The number of participants in the low HbA1c group was small and this may have influenced the power to detect group differences.
Three of the included studies were of intermediate quality while the remaining two were of high quality in reporting potential biases.
We adhered to strict systematic review procedures for study selection, data extraction and reporting to minimize reviewer related F I G U R E 3 Summary of fixed effects and random effects models: Estimated standardized mean difference of glycated hemoglobin (HbA1c) levels with 95% confidence interval between the low (exposed to glycaemic control) and high (unexposed to glycaemic control) HbA1c groups during various time-points of follow-up biases. The age ranges and sample sizes varied between studies which may have influenced the heterogeneity seen in the pooled estimates of long-term glycemic control. Heterogeneity was reduced when the study in pre-school children was excluded from the meta-analysis.
All studies included in the systematic review were conducted in developed countries, which had dissimilar health system models and this may have impacted the long-term glycemic outcomes. The sub-optimal HbA1c control estimated in the meta-analysis during the follow-up period may be because of more participants with higher HbA1c levels, age, 33 endogenous and exogenous factors or biological variation in the glycation phenotypes of children, [34][35][36] psychological factors particularly in older children. 37,38 These are all factors which may also have increased the risk of developing or progression of micro and macrovascular complications in those children as a consequence of those higher HbA1c levels. 39 The DCCT cohort were able to achieve HbA1c levels of 7% (53 mmol/mol) 40  Only one study provided evidence that albuminuria and retinopathy were associated with high mean HbA1c of ≥ 8.6% (≥70 mmol/mol) between 3 and 15 months after diagnosis of T1D. 24 This is consistent with findings by other studies, which did not meet our inclusion criteria. 6,17,43,44 It would be highly relevant for determining future prognosis, if these outcomes could be confirmed in future studies.
Cardiovascular disease is the major cause of death in T1D patients. Pre-symptomatic cardiovascular disease is evident in 100% of young adults with T1D 45 and there is evidence of accelerated atherosclerotic processes 46,47 and increased severity of cardiovascular disease 48 at an earlier age compared to the general population. Landmark trials show that intensive insulin therapy reduces cardiovascular events in adults. 6,49 Although differences in HbA1c account for most of this benefit, multivariate analyses suggest that part of the reduced risk is mediated by reduction in the incidence of diabetic renal disease. 50 In children and young people with T1D, atherosclerosis is present to a greater extent 51 and the prevalence of cardiovascular risk factors is greater 52,53 than in the general population. Diabetic nephropathy incidence accelerates during adolescence. 54 These are all T A B L E 2 Summary of pooled standardized mean differences in HbA1c levels between low and high HbA1c groups There is currently no evidence base for the effectiveness of ACE Inhibition or statin treatments in adolescents with T1D although, the important AdDIT Trial may inform practice in the coming years. 55 Therefore currently, in order to reduce vascular complications risk, the importance of achieving good glycemic control is arguably greater in childhood compared to adult T1D populations.
The meta-analysis indicates that the overall glycemic control stabilizes in a "track" after 6 months of childhood onset T1D diagnosis and pre-school aged children had better control throughout the follow-up period. Furthermore, the low and high HbA1c levels at baseline also seem to have metabolic memory, which shows HbA1c "tracking" during the 10-year follow-up despite differences between the high and low groups. This suggests there may be benefits of having good control during the initial few months of diagnosis. However, as these five studies report temporal associations, an experimental study of an intervention soon after diagnosis would be required to prove that better early control results in better later control. This review may also indicate a short window of opportunity to intervene and improve long-term glycemic outcomes. It may therefore be beneficial to develop clinical and educational strategies to identify and deliver targeted interventions during this early phase to those at risk of having poor glycemic control and to ensure that the HbA1c targets are maintained in the long-term. There is currently no evidence on effectiveness and timing of focused clinical interventions targeted at changing these tracks. 18 It would be useful to gather this evidence and to explore further the mechanisms of this phenomenon in order to deliver best care to newly diagnosed children and adolescents. The findings of this review would be useful to policy makers, health professionals and T1D patients to focus on designing interventions to prevent sub-optimal glycemic outcomes and decrease the risk of developing micro and macro vascular complications.

| Strengths and limitations of the review
The many strengths of this study include, being to our knowledge, the first systematic review and meta-analysis to rigorously investigate published and unpublished literature on the association of early glycemic control in childhood onset T1D with glycemic tracking and future risk of complications. Furthermore, this is the first review to rigorously and systematically search and review all available evidence as per preset inclusion/exclusion and quality assessment criteria. We have taken utmost care to minimize study selection, reviewer related and publication bias. All of the included studies were intermediate to high quality.
But, there are limitations to this systematic review which need to be considered. The diabetes diagnosis, care, and HbA1c outcome measures have evolved over the years and were not uniform across studies.
There was considerable heterogeneity between studies. The comparable follow-up data was not available beyond 10 years. We were unable to investigate if other factors may have confounded the findings. The small number of studies and the short duration of follow-up in studies may have masked the true association with long-term glycemic control.
Although we made every effort to search for unpublished and gray literature, we may have missed some that remain unreported because of unethical practices in reporting or publication bias. The results of our study may not be generalizable as they were mainly conducted in developed countries with varied health care system models.

| Review updating plans
The review will be updated if significant new evidence becomes available and results of the update review will be disseminated through peer-reviewed publications, conference presentations and at meetings.

CONFLICT OF INTEREST
No potential conflict of interest was reported by the authors.

AUTHORS CONTRIBUTION
VMP was the lead reviewer, designed the study, developed the study protocol, created the search strategy, searched electronic databases for literature, extracted the data, co-ordinated with authors of included studies for additional information, analyzed the evidence, drafted the report and is responsible for the article. JB and DTR par-