This population-based study showed distinct behavior of each class of serum lipids according to HbA1c levels in individuals with type 1 diabetes. The risk of having low HDL-cholesterol did not show a homogeneous inverse relation to HbA1c, being lower in the middle range of HbA1c (8.6 to 11.4%) than in HbA1c levels equal or above 11.5%. The risk of high LDL-cholesterol and triglycerides levels both showed increase with worsening glycemic control, although in different thresholds of HbA1c. LDL-cholesterol worsened starting at 9.7% and triglycerides only above an 11.5% HbA1c level. Regarding association with other variables, HDL-cholesterol showed marked inverse association with insulin dose. Rather than associating around a constant lipid factor, lipid variables correlated to glycemic control diversely according to HbA1c level.
Not all previous studies have shown homogeneous behavior of HDL-cholesterol in type 1 diabetes. Some have reported worsening of HDL-cholesterol with poorer glycemic control [4, 29], whereas others have demonstrated better HDL-cholesterol levels in type 1 diabetes when compared to non-diabetic controls [10, 30]. Somehow in comparisons performed only among individuals with type 1 diabetes, patients with poorer glycemic control have also been shown to have a paradoxical elevation of HDL-cholesterol compared to well-controlled individuals [11, 31]. Different stratification of HbA1c levels may account for differences between our findings and existing literature. Some studies have stratified subgroups in HbA1c levels as low as 7.5% , therefore being unable to assess differences in lipid behavior in the heterogeneous group of individuals with HbA1c levels above 7.5. While a lower threshold of HbA1c around 8% for changing behavior of HDL-cholesterol may be inferred from previous studies, an upper threshold is unclear. This also can account for the finding of worsening HDL-cholesterol with poor glycemic control. The grouping of individuals with intermediate HbA1c levels, who would supposedly have higher HDL, with individuals with higher HbA1c, who would have lower HDL-cholesterol values, could result in average worsening of HDL-cholesterol without accounting for the heterogeneity in the whole spectrum of glycemic control. This differentiation is important since there is evidence that these higher HDL-cholesterol levels can be associated with higher cardiovascular risk in individuals with type 1 diabetes, lacking the protective effect of high HDL-cholesterol in non-diabetic individuals . Another important aspect of low HDL-cholesterol is its relationship to higher insulin daily doses in type 1 diabetes therapy, which may indicate a higher insulin resistance background. Previous studies have shown IR to be a CVRF in individuals with type 1 diabetes . Low HDL-cholesterol is traditionally associated to IR and hyperinsulinemia in non-diabetic individuals, albeit the causal relationship among these three alterations is unclear. Our findings show association of higher insulin dose with low HDL-cholesterol consistently, independently of glycemic control. In type 1 diabetes, IR is not an established etiological factor but it can progressively develop after clinical diagnosis . Moreover, current insulin therapy methods are themselves a potential cause of hyperinsulinemia in these patients. Although the cross-sectional design of our study precludes any assumption of causality, one could postulate there is direct relationship between daily insulin doses and low HDL, without going through IR. The association of low-HDL-cholesterol with total daily insulin dose, corrected for BMI in the multivariable models (both regression and FA), could point to a role of IR by augmenting insulin needs, thus isolating the roles of obesity and exogenous insulin on HDL, rather than merging them both by using insulin dose per body weight instead. In this regard, a recent article has shown the insulin concentration required for 50% suppression of hepatic glucose production in a hyperinsulinemic/euglycemic clamp to be almost two times higher in type 1 diabetes than in controls adjusted for age, gender, and HbA1c. The authors suggest that hepatic and skeletal muscle IR in type 1 diabetes is not explained only by previously known factors . Previous studies have already suggested the CV risk conferred by IR to be detached from lipid variables .
The finding of higher triglyceride levels with worse glycemic control appears to depend also on the mode patients are stratified for glycemic control. Studies that have divided patients in two groups with low HbA1c thresholds have both showed no difference  or higher triglycerides in the higher HbA1c group [11, 29, 31]. Again, the stratification of HbA1c in lower levels can group together individuals with normal and high triglycerides, without necessarily establishing a threshold. Our data show a HbA1c threshold for increase in the probability of hypertrigliceridemia above 11%. This value has not been clearly established by previous studies and possibly varies according to diet and population differences.
The inverse relationship of triglycerides and HbA1c in normal controls and individuals with type 1 diabetes observed with HbA1c below 7.5 is less well explained although it has been previously demonstrated [11, 30, 33]. Previous studies hypothesized that more intensive insulin therapy, which is usually the case in well-controlled patients, can bring lipids to values below those shown by normal controls. Insulin has effects upon lipid metabolism by stimulating enzymes such as hormone-sensitive lipase . It is possible that in the normal (non-diabetic controls) or near-normal (good glycemic control) glucose range the effects of hyperinsulinemia can be more effective in controlling glucose as a compensatory mechanism than normal insulin values, thus mild degrees of hyperglycemia being associated with better triglyceride metabolism. A role for portal insulinopenia, as opposed to systemic hyperinsulinemia, cannot be excluded as well . Nevertheless, the correlation of triglycerides with the Hyperinsulinemia/IR factor hypothesized in our FA still leaves the order in the causal relationship of IR and hyperinsulinemia open to questioning.
LDL-cholesterol metabolism showed a pattern similar to triglycerides, although with a lower threshold. This finding furthermore supports the view that lipids are influenced by glycemic control of type 1 diabetes in a complex manner, not interacting with other cardiovascular risk factors such as blood pressure and obesity analogously to the CVRF cluster of type 2 diabetes.
Clustering of CVRF has been analyzed by factor analysis in various contexts. Its reproducibility has been criticized, since many reported models have yielded different results. Since this statistical technique (or any other, for that matter) is unable to assess biological plausibility of the models, one has to take especial care on previous planning of the analyses rather than in interpreting them. One-factor models of CVRF clustering are pathophysiologically implausible, given the multifactorial nature of involved conditions. The heterogeneity of previous results, therefore, can be attributed to differences in the populations analyzed and in the variables entered in the FAs. Glycemic control has not been frequently assessed previously in the clustering of CVRF in type 1 diabetes . We have previously demonstrated correlation of lipids and HbA1c in type 1 diabetes by means of factor analysis, but without the necessary statistical power to divide the 520 patients in subgroups according to HbA1c levels . From this point of view, the present study is adequately powered to perform the analyses, once the subgroups have samples above 200 hundred individuals, considered adequate by most authors . Regarding reproducibility, FA is an adequate method to test our hypothesis of different clustering of CVRF according to HbA1c level, since individuals form the same population have been compared using exactly the same variables.
The most important limitation of our study is its cross-sectional design, unable to assess temporal relationship among the various factors studied. Another limitation is that FA was not performed separately by gender. Gender significantly influenced the frequency of low-HDL-cholesterol, high-LDL-cholesterol, and hypertriglyceridemia in the logistic regression models, but dividing the five quintiles of HbA1c in ten groups by gender would impair sample power for performing FA. Another alternative approach would be using wider intervals of HbA1c to avoid excessive number of subgroups. In our view, this approach would generate more heterogeneous groups regarding glycemic control and would be inadequate to test our hypothesis. Nevertheless, gender differences in CV risk seem to be attenuated or even erased in type 1 diabetes , making our FA model without subdividing by gender valid to assess the main hypothesis of this paper. The absence of direct measurements of IR is also an important limitation, although it wouldn’t necessarily be feasible in such a large sample.