A newly noninvasive model for prediction of non-alcoholic fatty liver disease: utility of serum prolactin levels

Non-alcoholic fatty liver disease (NAFLD) refers to fat accumulation more than 5% of hepatocyte, in the absence of excess alcohol intake, virus hepatitis and drug induced liver injury, including simple steatosis, nonalcoholic steatohepatitis (NASH), fibrosis and, ultimately, cirrhosis [1]. It is estimated that the global prevalence of NAFLD was 25.2%, and NAFLD was associated with a series of metabolic comorbidities [2]. Early diagnosis of NAFLD and NASH is of great significance, since advanced stage of NAFLD and NASH had a higher carotid artery intima–media and thickness and overall mortality [1, 3]. Liver histology is regarded as the most reliable method of detecting NAFLD, however, the risk of biopsy-related complications including severe pain, peri-procedural hypotension and bleeding limit its use in clinical practice [4]. Although there are non-invasive techniques for assessing hepatic steatosis (ultrasound, ¹H-magnetic resonance spectroscopy, and computed tomography), these procedures were time-consuming and costly and therefore often unavailable for screening NALFD in large population-based studies [5]. Subsequent studies have proposed several indexes such as fatty liver index (FLI) for diagnosing NAFLD [6], yet the calculation of these markers were complicated. Prolactin (PRL) is a pituitary-derived hormone which was recently shown to be closely associated with the existence and progression of fatty liver [7]. Here we aimed to develop a method to predict the presence of NAFLD based on data from subjects enrolled in two separate cohort studies and evaluated whether the involvement of PRL would improve the diagnostic value than previous reported indexes.

In this study, we attempted to develop an approach for diagnosing NAFLD via common clinical and laboratory data. We have demonstrated that by using the following equation: 0.474*BMI (kg/m²) - 0.131 * PRL (μg/l) + 0.026*ALT (U/l) -2.139*HDL (mmol/l) - 8.758, and two cut-off points (− 0.79 and 1.71) in males, and 0.386*BMI (kg/m²) - 0.24 * PRL (μg/l) + 0.52*HbA1c (%) + 0.06*ALT (U/l) - 11.619, and two cut-off points (− 0.68 and 2.16) in females, NAFLD can be identified with a high sensitivity and specificity both in the estimation group and validation group. The ORs of PRL in the model of males and females is 0.877 and 0.786 (Table 2), indicating that an increase of 1 SD in PRL was associated with a reduced risk of 12.3% in males and 21.4% in females for NAFLD when the other variables in the model were kept constant.

NAFLD is a chronic liver disease that may lead to fibrosis and cirrhosis if without early intervention [20]. Our study included not only 873 well-characterized individuals in whom hepatic steatosis was identified through abdominal ultrasound, but also 147 patients who have received liver biopsy, the gold standard for diagnosing NAFLD and NASH. Our models consisted following parameters: BMI, HbA1c, PRL, ALT and HDL, in which BMI, HbA1c were shown to be risk factors for NAFLD, while HDL was shown to be negatively associated with NAFLD [21‐23]. ALT levels reflect the inflammation state of liver, and subjects with elevated liver enzymes (ALT) are recommended to be evaluated for presence of NASH [24, 25]. Moreover, an inverse association between PRL and NAFLD was observed in our model. PRL is a polypeptide hormone mainly produced from anterior pituitary, well known for its lactogenic properties [26]. Recent studies also suggested an important role of PRL in metabolic disease, and PRL was proven to be a protective factor against the existence and progress of NAFLD [27, 28], which were supported by our current data. PRL is a hormone that has diurnal variation and varies through the menstrual cycle. To exclude this variability, fasting serum samples were collected in all subjects on 8:00 to minimize the influence of environmental stress. In addition, we applied our model in both pre- and postmenopausal females and found that the AUC were higher than 0.7 in both groups.

Previous evidence described that serum PRL levels were beginning to decline with the growth of age. In males, a study carried out in middle-aged men suggested that PRL concentration was negatively correlated with age [29]. However, another study demonstrated that changes of PRL in males after 50 years old did not reach statistical difference [12]. Since aging is also a risk factor for NAFLD [30], we then inspected the influence of aging on PRL. Here we divided male subjects into two groups according to whether their ages were more than 50 years old (Additional file 8: Table S5). We found that there were no significant difference in PRL concentrations between two groups. Moreover, age did not enter the final model in multivariate logistic regression analysis in both genders, suggesting that after adjusting for other confounding factors, age was not independently associated with NAFLD. We also tested the diagnostic efficiency of our models in these subgroups and the AUC were higher in both younger and elder males (both> 0.8 and all P < 0.01) (Additional file 2: Figure S2a, b). In females, postmenopausal females exhibited a 40% decrease in PRL secretion compared with premenopausal women [12]. Therefore we have analyzed the performance of our model in females separately based on the menopausal status (Additional file 2: Figure S2c, d). The AUC were still higher than 0.8 in both groups (all P < 0.01). These data revealed that our model is efficient for identifying NAFLD regardless of aging in both genders.

To quantify the clinical contribution of PRL in the diagnosis of NAFLD, we computed two novel described metrics, IDI and NRI. The categorical NRI can determine the advancement in classification between two models by sum of the proportion of increased predicted risk in cases and the proportion of decreased predicted risk in controls. We used 0–30%, 30–60%, more than 60% to define the low-, middle-, and high-risk. The IDI can be interpreted as the difference in percentage of variance explained by the model with or without the new predictor [15, 19]. In addition, we also compared the our model with FLI in identifying hepatic steatosis, and the results showed that the efficiency of our model increased 31.95 and 26.7% in males and females, respectively (data not shown). These findings manifested that incorporation of PRL showed increased values of IDI and NRI, indicating a superiority of adding PRL within our model for predicting NAFLD.

We selected two cut-off points to improve the diagnostic accuracy, the lower cut-off point provides higher sensitivity and the higher cut-off point provides higher specificity (both higher than 90%) in diagnosing NAFLD (Table 3). This is clinically helpful because below the lower cut-off is appropriate for excluding NAFLD and subjects above the upper cut-off are more likely to present NAFLD. Although the lower cut-off (0.79 in males and 0.68 in females) provided over 90% sensitivity, the specificity in both genders were relatively low (54.4% for males and 61.3% for females). Here we would like to recommend that in subjects with risk factors for NAFLD such as elevated ALT levels or obesity, using this lower cut-off point of our model is valuable as a screening tool. On the other hand, the upper cut-off (1.71 in males and 2.16 in females) yielded higher specificity (91.1% for males and 91.4% for females) but lower sensitivity. When a clinician is about to giving a patient liver biopsy for further examination of liver lesions or treatment for NAFLD, this cut-off is useful because higher specificity can minimize the inclusion of false-positive cases. Importantly, the AUC of our model for predicting NAFLD in subjects with liver biopsy was still high: 0.71 (95%CI: 0.56–0.83) in men and 0.74 (95%CI: 0.56–0.92) in women. An AUC of more than 0.7 indicated sufficient predictive ability, of more than 0.8 indicated accurate diagnostic power [31], hence our model is optimal for the diagnosis of NAFLD in both genders. Therefore, when a patient’s score is over 1.71 in male or over 2.16 in female calculated by our model, ultrasound examination is recommended to identify these patients with high risk of developing fatty liver. However, in subjects with a score between this two cut-off points, further examination is recommended.

So far, several diagnostic algorithms have been developed to predict NAFLD, and among these the FLI has been validated in several population studies [32, 33]. FLI showed a good level of accuracy in detecting NAFLD (sensitivity of 0.84 and specificity > 0.86 for an FLI > 60) [34]. Using liver biopsy as the reference, we compared our model with FLI, and found that AUC of our model was significantly higher than that of FLI in both genders, indicating that our model was superior to FLI in terms of the predicting performance of NAFLD. Besides, the advantage of adding PRL into the model is that a positive correlation between the scores and the severity of NAFLD was observed (Fig. 2c, d), which may help the general practitioner in estimating the severe degree of NAFLD.

More importantly, since current imaging technology cannot differentiate NASH from simple steatosis, the diagnosis of NASH was still based on liver pathology [35]. It has been suggested that early liver pathology may be indicated in all NAFLD patients, since earlier intervention and more aggressive treatment can reduce overall mortality [36]. By adding the variate of PRL, our model provides an alternative method to easily identifying those with a high risk for NASH, thus may make liver biopsy unnecessary in a considerable proportion of patients.

The limitation of our study is that there is a lack of large cohort of biopsy-proven NAFLD patients, therefore our models need to be further validated in separate independent cohorts of NAFLD patients with liver pathology.

In conclusion, we have established a novel non-invasive model for diagnosing the existence and severities NAFLD, which is an effective screening tool in epidemiologic studies and may help physicians identify patients for ultrasound examination.