Obesity and dyslipidemia used to be considered the main risk factors for NAFLD, as well as increasing T2DM and cardiovascular disease risk in NAFLD [25]. However, owing to lean individuals, particularly from Asian populations, being diagnosed with NAFLD [15, 26], the concept of lean NAFLD has received increasing attention. In fact, surprisingly, lean individuals with NAFLD are at greater risk for T2DM, and have higher mortality rates, compared to obese individuals [17, 18]. The association between T2DM and NAFLD is not surprising, as NAFLD has been found to be an independent risk factor for T2DM, playing a significant role in its development; indeed, a meta-analysis from Ballestri et al. found that NAFLD was associated with an almost 2-fold increased risk for T2DM during a 5-year follow-up period [27]. There is also a reciprocal relationship between T2DM presence and NAFLD risk, in which NAFLD is more prevalent among T2DM individuals, compared to the overall global prevalence of NAFLD [10]. However, while the link between T2DM and NAFLD has been well-established, the association between NAFLD and pre-diabetes is poorly-defined. In this study, it was demonstrated that among lean individuals with normal blood lipid levels, prediabetics were more at risk for NAFLD, compared to normal individuals, as shown in the results from Study 2, where 7897 initially NAFLD-free, lean Chinese individuals with normal blood lipid levels were divided into normal, pre-diabetic, and diabetic groups. There, NAFLD incidence at the end of the 5-year follow-up period increased from as low as 3.7% among normal, to 9.7% in pre-diabetics and 15.3% in diabetics, suggesting that the latter 2 categories were more likely to develop NAFLD, compared to normal. This was further supported by Kaplan-Meier analysis demonstrating a positive correlation between diabetes progression and increased NAFLD risk; indeed, pre-diabetics had higher cumulative NAFLD risk after the 5-year follow-up period, compared to normal. This increased risk for NAFLD among pre-diabetic lean Chinese individuals, with normal blood lipid levels, was associated with 6 risk factors: BMI, TC, AAR, THR, FPG and GGT, which were then incorporated as part of a predictive nomogram, yielding an AUC ~ 0.8 among the 4 sets tested, indicating that it had a high discriminatory capability; this nomogram was further validated by calibration curves. The clinical utility of the nomogram was confirmed by DCA, indicating that it could be used for screening of high-risk individuals, allowing earlier and more effective interventions against NAFLD.
“Pre-diabetes” describes a condition where blood glucose levels are higher than normal, but lower than those associated with T2DM diagnosis. The pathogenesis underlying this condition stems from impaired β-cell function and increased insulin resistance (IR). Prediabetic cases have been increasing worldwide in an alarming trend [28], and is even more prevalent among Asians, compared to Westerners [29]. As a result, prediabetes is regarded as a critical stage, as early screening and intervention could reduce, or even reverse, the risk of progressing to diabetes. Such screening and intervention during prediabetes could also potentially aid in reducing NAFLD risk [30, 31]; however, the prediabetic population, particularly those who are lean, are often ignored in clinical practice until they had already progressed to T2DM or NAFLD.
To meet this unmet need, we thus developed a novel predictive nomogram to determine NAFLD risk among pre-diabetic lean Chinese individuals with normal blood lipid levels. There have already been several predictive models developed to determine the risk of individuals ending up with T2DM and NAFLD, such as from Zhang et al. and Xue et al. [20, 32], both of which focused on assessing NAFLD risk among Chinese T2DM. Another model from Cai et al. [33] is able to estimate 8-year incidence of T2DM among NAFLD populations. However, these models did not deal with predicting NAFLD risk among pre-diabetics, unlike our nomogram. In this study, BMI, TC, AAR, THR, FPG and GGT were used as the basis for the nomogram, as they were considered the most predictive under LASSO and logistic regression analyses. Both LASSO and logistic regression are able to solve all kinds of problems involving multicollinearity and confounding factors, providing more accurate results compared to other analytical methods. Additionally, compared with the traditional prediction model, the nomogram model is more accurate, easier to visualize, and more convenient for clinical decision-making. This was then verified by establishing 4 data sets: training, validation, longitudinal internal validation, and external validation. Furthermore, ROC and calibration curve results, as well as DCA, confirmed that our nomogram was highly accurate with respect to its predictions when compared to actual outcomes, as well as providing greater utility in clinical settings for prognosticating future NAFLD. It has been noted, though, that current non-invasive diagnostic techniques, such as ultrasonic liver imaging and measurement of serum biomarkers, have already been proven to be useful for diagnosing NAFLD. However, both of these methods have limitations, in that “gold standard” cut-off values for serum biomarkers, such as AAR, have not been fully defined and validated. Furthermore, these biomarkers are not liver-specific, meaning that they could be influenced by co-morbidities, resulting in misleading measurements [34]. Additionally, ultrasound detection is less effective in the extremely early stages of NAFLD, limiting its utility for facilitating early intervention against this disease [35]. By contrast, the predictive nomogram established in this study was able to predict NAFLD onset long before its occurrence, even before it was detectable by either serum biomarkers or ultrasound.
Comparisons with other studies and what the current work add to existing knowledge
The inclusion of the 6 parameters, BMI, TC, THR, AAR, FPG and GGT, in the predictive nomogram was in line with findings from previous studies. Obesity has been considered a significant independent risk factor for T2DM and NAFLD [36,37,38]; moreover, even in non-obese individuals, defined as having BMI < 25 kg/m2, increases in NAFLD risk has been found to be positively associated with BMI increases [39]. This is consistent with what was found in this study among lean Chinese individuals, whose BMI were < 23 kg/m2. Lipid-based metabolic disorders and adipose tissue dysfunction also play important roles in NAFLD onset, and close associations have been found between NAFLD occurrence and increased levels of TC, TG, HDL-C, and other lipid components [24, 40]. In particular, THR has been shown to be independently associated with NAFLD onset in healthy individuals, which is in line with it being a surrogate indicator of IR, and thus the progression of an individual towards prediabetic and diabetic stages, as well as NAFLD. This association, in turn, enables THR to serve as a NAFLD predictor, in which the higher the ratio, the higher the risk for developing NAFLD and diabetes [41, 42]. Both lipid-associated parameters and THR correlating to NAFLD risk was supported by this study, which demonstrated that higher NAFLD risk was present among those with higher TC and THR, even if they otherwise had overall normal lipid levels. As for GGT, ALT, and AST, they have long been used in China as liver functional indicators to evaluate hepatobiliary diseases. GGT is found on the surface of multiple cell types and is highly active in the liver, where it is involved in reducing oxidative stress. It is believed to be closely related to liver steatosis and fat deposition, and could possibly serve as a surrogate marker for NAFLD [43]. Additionally, epidemiological studies have confirmed that serum GGT is closely related to T2DM, possibly serving as an important predictive risk indicator [44]. ALT and AST are both specific markers of liver inflammation and cell damage, and are also closely related to NAFLD, likely owing to higher ALT and AST contributing to chronic liver inflammation, IR, and hepatic steatosis [45]. Compared to ALT and AST alone, though, AAR is more strongly predictive for NAFLD onset, which has led to its increased prevalence as a predictive indicator [46]; this was further supported by a study of 12,127 initially non-obese, NAFLD-free individuals, where AAR was found to be an independent risk factor for NAFLD onset in obese individuals [47]. FPG levels have also been found to reflect the level of secretion and functioning of basal insulin, leading to it being considered an independent predictor of DM [48, 49]. The current study has extended this observation to NAFLD, in that higher FPG has been found to correspond to greater NAFLD risk.
Furthermore, this study demonstrated that among lean Chinese individuals with normal blood lipid levels, prediabetics were more at risk for NAFLD, compared to normal individuals. This higher risk was able to be predicted, with high discriminatory capability, using a nomogram incorporating 6 factors: BMI, TC, AAR, THR, FPG and GGT. This nomogram could be thus used as a screening tool for identifying high-risk individuals, allowing earlier and more effective interventions against NAFLD.
Study strengths and limitations
To the best of our knowledge, the present study is the first to develop and evaluate a predictive nomogram for NAFLD risk, among a lean Chinese population with normal lipid levels in the pre-diabetic stages. This nomogram was based on, and confirmed by, representative large sample populations obtained from different medical institutions in different regions of China and Japan, demonstrating its validity for a variety of different population groups. It was also based on findings from both cross-sectional and longitudinal studies, providing greater reliability in predicting NAFLD in a long-term time scale. Furthermore, the measurements for the 6 parameters in the predictive nomogram can be obtained simply and non-invasively, facilitating widespread ease in its adoption in clinical practice. However, there are a number of limitations to this study, one of which is that this was a secondary retrospective analysis, based on data collected from 3 previous studies, resulting in limitations in the data collected despite the large sample sizes. These limitations in the collected data included the number of times that FPG and HbA1c measurements were taken among the patients in those studies, as different conditions could affect FPG and HbA1c measurements, and thus patient categorization as normal, pre-diabetic, or diabetic. Additionally, NAFLD staging data was not available, even though it had been previously documented that a number of factors could have varying impacts at different NAFLD stages. For instance, it has been documented that the negative impact of dyslipidemia is less significant in later stages of NAFLD, when cirrhosis develops, due to the failure of hepatic lipid-synthesizing mechanisms at that stage [50]. Another limitation is that diabetes and pre-diabetes diagnostic criteria was mainly based on FPG, which could lead to undercounting, as FPG may miss some individuals who could otherwise be caught by other tests, such as the oral glucose tolerance test (OGTT), which was not carried out by the studies included in this paper. Future investigations should take OGTT, as well as multiple FPG and HbA1c measurements under multiple different conditions, to ensure that the overall values are fully reflective of patient glycemic statuses. Furthermore, the associations between the 6 factors incorporated into the predictive nomogram with different stages of NAFLD should be examined.