Reducing NAFLD-Screening Time: A Comparative Study of Eight Diagnostic Methods Offering an Alternative to Ultrasound Scans

Filippo Procino; Giovanni Misciagna; Nicola Veronese; Maria G. Caruso; Marisa Chiloiro; Anna M. Cisternino; Maria Notarnicola; Caterina Bonfiglio; Irene Bruno; Claudia Buongiorno; Angelo Campanella; Valentina Deflorio; Isabella Franco; Rocco Guerra; Carla M. Leone; Antonella Mirizzi; Alessandro Nitti; Alberto R. Osella; MICOL GROUP

Disclosures

Liver International. 2019;39(1):187-196. 

In This Article

Results

Table 2 shows the general characteristics of participants. We found a prevalence of 31.6% of NAFLD, with a higher percentage of males (38.8%) than females (22.1%). Female patients were significantly older than males in the NAFLD (53.98 years for females and 54.1 years for males) vs NO-NAFLD group (59.73 years for females and 53.53 years for males) for the females but not for males. We registered significantly higher serum biomarkers values in the NAFLD vs NO-NAFLD group. All the anthropometric measures (except height) and scores were significantly higher in the NAFLD than in the NO-NAFLD group for both sexes.

Figure 1 represents the ROC curves for the eight formulas in females and males. The area under the ROC curve (AUROC) appears ampler in females than in males for all formulas. The AUROC of FLI and HIS seems wider than the others, and more markedly so in males. Table 3 shows the AUROC of each formula for predicting NAFLD in the subjects, divided by sex or adjusted by age (AUROC-age). FLI shows the best AUROC (0.85 in females and 0.79 in males) and BMI the worst (0.82 in females and 0.75 in males). Adjusting the AUROC for age, we obtained better results only for anthro-m (except AVI) in males but not in females.

Figure 1.

Receiver operating characteristic curves

Table 4 reports the optimal cut-off for each ROC curve using Youden's index. The data are shown as the optimal cut-off, sensitivity, specificity and AUROC at the optimal cut-off (AUROC-cut-off). We also calculated the optimal cut-offs for males <40 years (BMI and WC) and for males <35 years (WHtR, WHt_5R and BRI).

To compare the performance of the eight alternative hybrid methods in identifying NAFLD in a large population, we operated a simulation, applying the calculated cut-offs and verifying the outcomes. Table 5 shows the data as number and percentage of subjects identified as at risk by the formula, number and percentage of patients with unidentified NAFLD (NAFLD missed), PPV and NPV, constituting the first step. The second part of Table 5 illustrates the US results in those subjects identified as at risk, shown as percentage of US reduction, number and percentage of NAFLD cases identified in the group at risk (NAFLD found in risk group) and percentage identified over total affected NAFLD population. In parallel, Figure 2 represents the results of Table 5, shown as percentage identified as at risk, percentage NAFLD missed and percentage NAFLD identified over total affected population.

Figure 2.

Performance of formula and anthro-m

WHtR showed the best filter power, identifying 1420 subjects as at risk (47.8%), thus reducing the need for US by 52.2%. AVI resulted in the least predictive, reducing US by 45.2%.

As regards NAFLD missed, the best performance was shown by AVI (5.2%) and the worst by WHtR (8.5%). To confirm our findings, we calculated the NPV and PPV values. AVI showed the best NPV but not the best PPV, FLI and HIS showed the best PPV but not the best NPV. WHtR showed the lowest NPV and PPV values. The percentage NAFLD identified over the total affected population is shown in Table 5 and represented in Figure 2. AVI yielded the best performance (83.6%) and WHtR the worst (73%). These data are relative and applicable to the prevalence of NAFLD detected in our population (31.6%). To evaluate the applicability of these findings to different prevalence rates, we calculated the PPV and NPV trends vsa. prevalence of NAFLD ranging between 20% and 40% (the worldwide prevalence range of NAFLD). Graph A on Figure 3 shows that the trend of PPV vs prevalence percentage is higher for FLI and HIS for the whole range of prevalence percentage while graph B illustrates that AVI is characterized by higher NPV values for the whole range of prevalence percentage.

Figure 3.

Positive predictive value and negative predictive value vs prevalence (%)

Comments

3090D553-9492-4563-8681-AD288FA52ACE

processing....