With the exception of BMD, all other analyzed CT body composition biomarkers demonstrated statistically significant changes in at least one analyzed sub-cohort from normal to prediabetic Hb A1c ranges, as well as from the prediabetic to diabetic Hb A1c range. The L1 vertebral body trabecular density did not vary with glycemic control, accepting the null hypothesis (Table 1, also Supplementary Fig. 1), and effectively acting as a control variable. Thus, macroscopic physiologic changes of metabolic syndrome observable with CT biomarkers appear to begin in the pre-diabetic phase.
We observed a significant trend of increasing CT biomarkers for measures of body composition bulk or amount, including visceral adipose tissue (VAT) area, kidney volume, liver volume, and muscle area, with increasing HbA1c (Figs. 3–6). Across all groups, the biomarker changes were generally stepwise with increasing HbA1c from normal to pre-diabetic to the diabetic range. However, in at least one sub-cohort, of all seven biomarkers that statistically significantly varied with HbA1c, the biomarker did not significantly change further when comparing the diabetic and poorly controlled diabetic groups. This could be due to the physiologic changes plateauing, biomarker variability from more labile glycemic control in the poorly-controlled group, or perhaps some changes require additional time. As one example, chronic poorly-controlled diabetic patients with renal involvement frequently develop renal atrophy, which would increase the data variance in the higher HbA1c categories.
The pattern of significantly increasing VAT area with increasing HbA1c from the normal range to the diabetic category is consistent, and within expectations. However, in all groups, the VAT area plateaus in the diabetic category and does not significantly differ in patients in the poorly controlled category. For men, only in contrast-enhanced CT exams, the plateau started in the pre-diabetic category (though the p-value of 0.011 is near the conservatively chosen alpha of 0.0083). The related biomarker ratio of VAT to SAT area related to glycemic control revealed sex-dependent changes in adipose tissue distribution, discussed in the Supplement. We observe a less strong but significant decrease in liver attenuation with increasing HbA1c, compatible with hepatic steatosis, also consistent with prior work22. The differences in liver density between groups were more accentuated after IV contrast as steatotic livers enhance less, in addition to starting at a lower HU level23. The ratio of VAT to SAT area related to glycemic control was sex dependent, explored further in the Supplement and illustrated on Supplementary Fig. 2.
We observe increased muscle area but decreased muscle density with increasing HbA1c, suggesting increased myosteatosis21. The best analogous prior study to this one compared hepatic steatosis to non-alcoholic steatohepatitis and agrees with our findings, observing that intermuscular adipose tissue (IMAT) as approximated by paraspinal CT muscle density at L3 level decreased as liver disease worsened24. This is further supported by empiric work with pathologic confirmation showing impaired lipolysis in both adipose tissue and skeletal muscle in the setting of diabetes mellitus and obesity25, also in agreement with recently proposed physiologic mechanisms underlying insulin resistance and ectopic lipid accumulation26. However, the increased muscle area we observe differs from some prior works, predominately using DXA to measure skeletal muscle in various body regions, which reported decreased skeletal muscle with poor glycemic control as an indicator for high cardiometabolic risk27,28. The divergence observed could be due to differences in technique, with partial volume averaging of increased IMAT decreasing the density seen by DXA, or differences in normalization.
We observe increased renal size with increasing HbA1c. This is concordant with complimentary literature, with the mechanism believed to be compensatory increased glomerular filtration as the kidneys excrete glucose. Eventually this compensatory effect fails, leading patients with chronic poor glycemic control to end-stage renal disease (ESRD) and renal atrophy. These atrophic changes may underlie the larger observed renal volume variance in the poorly controlled cohort.
The indications and thus populations of patients receiving contrast-enhanced CT exams versus unenhanced CT exams likely differ somewhat. To some degree, patients with unenhanced CT exams are more likely to have contrast deferred due to chronic kidney disease, which is associated with longstanding diabetes mellitus and poor glycemic control. We suspect this may account for the widened distribution of kidney volumes seen in the unenhanced CT groups for both women and men in the poorly controlled category shown in Fig. 5.
Additional limitations of this retrospective study include expected enrichment in patients with or with suspected metabolic syndrome. Due to retrospective analysis, our cohort is predicated on HbA1c laboratory orders from routine clinical practice. This allows norms to be established for different ranges of glycemic control, but we recognize this does not reflect the general population. Another limitation is related to the timing of the HbA1c measurement relative to the analyzed CT exam, which was a median of 3.5 months. This is reasonable for most patients, but in the setting of labile or poor glycemic control a tighter relationship between CT scan time and HbA1c may be beneficial. These limitations could be addressed in the future with prospective or multi-center studies. Additionally, all exams in this cohort are from unique patients. While this was a deliberate choice, future work analyzing pairwise changes in biomarkers from the same patients at different times with differing levels of glycemic control would be illustrative to confirm these relationships and explore if they are reversible. The present work also does not attempt to account for the length of time since diabetic diagnosis nor historic glycemic control.
The use of HbA1c as our serum biomarker introduces limitations due to lab variability or false elevation in certain settings such as iron-deficiency or sickle-cell anemia, renal or liver failure, sickle-cell anemia, or recent transfusion. As shown in Fig. 1, the glycemic control distribution in our cohort is non-parametric. This is to be expected, with the skewed peak in the well-managed range. However, the tail extends sufficiently into the higher HbA1c range for statistically useful comparison groups.
In summary, this work represents the largest cohort analysis to date demonstrating the multisystemic physiologic effects of diabetes mellitus and metabolic syndrome at differing levels of glycemic control. Through a suite of quantitative imaging body composition biomarkers extracted from CT exams, we show the effects of metabolic syndrome largely begin in pre-diabetes and continue into the diabetic range as measured by HbA1c. We confirm several imaging trends with high statistical power and further support the importance of inter- and intra-muscular adipose tissue as a biomarker in metabolic syndrome, which CT is ideal to measure. This and future related work offer the potential to augment and improve the definition and measurement of metabolic syndrome. All of these imaging biomarkers may be repurposed opportunistically from abdominal CT scans performed for any indication, adding value through understanding the physiologic effects of metabolic syndrome and glycemic control, potentially predicting which patients scanned for other reasons would benefit from biochemical evaluation and endocrinology referral, and providing insights into the physiology of new therapies such as glucagon-like peptide 1 (GLP-1) agonists and sodium glucose cotransporter 2 (SGLT-2) inhibitors29,30.