Abstract
Objective:
Non-alcoholic fatty liver disease (NAFLD) is common in obese adolescents with Polycystic ovary syndrome (PCOS), but there are no inexpensive ways to accurately identify NAFLD in PCOS. The objective was to develop a simple clinical score to screen for NAFLD risk in obese adolescents with PCOS.
Design:
This is a secondary analysis of 3 cross-sectional studies on metabolic characterization of obese adolescents with PCOS. 108 overweight and obese adolescents with PCOS (BMI >90th percentile, age 12–19 years) were enrolled from 2012–2018.
Methods:
Magnetic resonance imaging was used to quantify hepatic fat fraction (HFF). A development cohort of 87 girls were divided by presence of NAFLD (HFF > 5.5%). A logistic regression model with the outcome of NAFLD and candidate predictor variables was fit. A simplified model (PCOS-HS index) was created using backwards stepdown elimination. Validation was performed using 200 bootstrapped sample and in a second cohort of 21 PCOS participants.
Results:
52% of the development cohort had NAFLD. The PCOS-HS index that included BMI percentile, waist circumference, ALT and SHBG had an AUCROC of 0.81, sensitivity 82%, specificity 69%, negative predictive value (NPV) 78% and positive predictive value 74%, using a threshold of 0.43 to predict HS. A threshold of 0.15 ruled out NAFLD with a NPV 90%. In the validation cohort, the model showed an accuracy of 81%, sensitivity 91% and specificity 70%.
Conclusions:
We developed a clinical index to identify NAFLD in girls with PCOS who would need further evaluation and treatment.
Clinical trial registration:
Keywords: PCOS, Non-alcoholic fatty liver disease, Adolescence, screening
INTRODUCTION
Non-alcoholic fatty liver disease (NAFLD) comprises a spectrum of disease states ranging from simple hepatic steatosis (HS) without inflammation or fibrosis (NAFL) to steatohepatitis (NASH) or cirrhosis. About 20% of patients will progress to NASH. Given the current rise of obesity and type 2 diabetes (T2D), the prevalence of NASH is predicted to increase by 63% by 2030, leading to a 178% increase in liver-related deaths (1). Moreover, the clinical burden of NAFLD is not restricted to hepatic disease but includes the development of cardiovascular diseases and metabolic complications (2).
Polycystic ovary syndrome (PCOS) is a common disorder in women, affecting 15–20% of women with obesity. PCOS is associated with increased peripheral insulin resistance (IR) in obese and non-obese adolescents, and adipose and hepatic IR in obese adolescents (3, 4). Several studies have shown an increased prevalence of NAFLD in women with PCOS, in comparison to women without PCOS of similar BMI. A recent meta-analysis calculated an OR of NAFLD of 3.0 in obese women and 2.0 in non-obese women with PCOS relative to women without PCOS (5). We demonstrated that excess NAFLD is also present in obese adolescent girls with PCOS with a prevalence of 50% when measured by the gold standard MRI-proton density fat fraction (MRI-PDFF), compared to only 13% in equally obese adolescents without PCOS (3).
Endocrine Society Guidelines recommend screening for NAFLD in women with PCOS and metabolic risk factors by using serum transaminases, followed by noninvasive quantification of fibrosis if elevated (6). Furthermore, the North American Society of Pediatric Gastroenterology, Hepatology and Nutrition (NASPGHAN) guidelines recommend universal screening in obese and overweight children with metabolic risk factors after the age of 9 years old. The recommended screening test is ALT, using a threshold 2 times higher than sex-specific upper limit normal (ULN) from NHANES data (ULN= 22.1 IU/L for girls) (7). The sensitivity of ALT in detecting hepatic steatosis is lower in girls than boys, even at a low threshold (>20 UI/L) although data could not be calculated for female specific cut-offs due to sample size limitations (8). European Association for the Study of Liver (EASL) also favor universal screening for NAFLD in patients with metabolic risk factors, which includes obese adolescent with PCOS using steatosis biomarkers or ultrasound (9). Several non-invasive NAFLD risk indices have been assessed (8, 10–16), but most of those are not readily applicable in a clinical setting and none were developed specifically for a population with PCOS, who appear to have a unique predisposition for NAFLD.
Obese adolescents with PCOS are likely at risk of progressive NAFLD because of the high risk of T2D and IR and development of NAFLD at an early age, all risk factors for progressive disease (17). Several new treatments for NAFLD are currently under investigation and may be clinically available in the near future (18). Consequently, the development of a non-invasive, simple and inexpensive screening tool to identify patients at high risk for NAFLD who would benefit from additional more invasive or costly methods of testing is critical.
The aim of this study was to develop and validate a simple, predictive model using clinically available data, including hormonal markers specific to PCOS such as testosterone or sex-hormone binding globulin (SHBG), to accurately predict NAFLD in obese adolescents with PCOS. This predictive model is an easy screening tool for clinicians allowing identification of probability of NAFLD in adolescents with obesity and PCOS requiring further investigation.
MATERIAL AND METHODS
Study design, Setting and Participants
The current study is a secondary analysis of participants enrolled between 2012 and 2018 in one of our three cross-sectional studies including metabolic characterization of adolescents with PCOS: 1) androgens and insulin sensitivity (AIRS, prior to NCT, N=76), 2) Liver and Fat Regulation in Overweight Adolescent Girls (APPLE, NCT02157974, N=92) and 3) Post-Prandial Liver Glucose Metabolism in PCOS (PLUM, NCT03041129, N=17). The details of the first study have been published (3, 4, 19). Participants were recruited through our multidisciplinary PCOS, general endocrine clinic and lifestyle intervention obesity clinics at Children’s Hospital Colorado. Inclusion criteria for AIRS were: obese females (BMI ≥ 95th percentile) or lean (BMI< 85th percentile), with or without PCOS, who were physically inactive (exercising less than 3.0 hours a week) and age 12–21 years old. APPLE and PLUM had identical criteria except included only overweight or obese females (BMI ≥ 90th percentile). APPLE included girls with and without PCOS and PLUM PCOS only. PCOS was classified according to the NIH definition and more recent international PCOS 2018 guidelines as the presence of irregular menstrual cycles for at least 1.5 years (AIRS) or 2 years (APPLE and PLUM) after menarche and either clinical or biochemical hyperandrogenism (6, 20). Participants are thus representative of the typical obese population of adolescents presenting for care with PCOS. A physical exam was performed by an endocrinologist at the screening visit in order to assess clinical hyperandrogenism. Hirsutism was rated according to the Ferriman-Gallaway scale (FGS) (21).
110 overweight/obese participants with PCOS were included in this study according to these further inclusion criteria: overweight or obese (BMI≥ 90th percentile) and liver MRI performed. Exclusion criteria included: use of oral contraceptives, use of medication known to affect insulin sensitivity (except metformin), diabetes (defined as HbA1c ≥ 6.5%), liver disease other than hepatic steatosis and AST or ALT > 125 IU/L. Two participants were excluded because of missing data included in candidate variables. Eighty-seven participants were included in the model development cohort 1 (AIRS N= 37, APPLE N=50). Another sample of all consecutive participants (21 girls) with identical criteria who were enrolled chronologically after the development of the score served as an external validation sample, cohort 2 (APPLE N=4, PLUM N=17).
The University of Colorado Anschutz Medical Campus institutional review board and the Children’s Hospital Colorado Research Institute approved the studies. All participants aged 18 to 21 years provided written informed consent, and the parents and participants provided consent and assent for all participants aged <18 years.
Data collection
Hepatic fat fraction
Hepatic fat fraction was assessed using MRI-PDFF with the DIXON technique, by one radiologist (AS) blinded to clinical data (22). The imaging was obtained on a 3 Tesla Magnet (Siemens Magnetom Skyra, Tarrytown, NY or GE Healthcare, Milwaukee, WI). Up to 18 abdominal axial 10 mm slices to cover the whole liver were obtained using a multibreath-hold gradient echo sequence, using 6 different echo-times: 1.15, 2.3, 3.45, 4.6, 5.75 and 6.9. Regions of interest were manually drawn, avoiding liver edges and large vessels. Hepatic fat fraction maps were generated in Osirix by using the Lipoquant plug-in as previously described (23). The weighted average of the mean fat fraction was calculated for each subject. NAFLD was defined as HFF ≥5.5%. MRI-PDFF is recognized as an accurate, repeatable and reproducible quantitative assessment of HFF, the best non-invasive surrogate to liver biopsy for NAFLD in adults and children (24–26).
Anthropometric data
Waist circumference, BMI (kg/m2) and BMI percentile per Center for Disease Control and Prevention BMI growth charts (27) were obtained within 24 hours of the liver MRI.
Lab measurements
All laboratory measurements were obtained the day after the liver MRI, following a 12 hour fast. Analyses were performed by the University of Colorado Anschutz Research core laboratory or the Children’s Hospital Colorado clinical laboratory. Plasma total cholesterol, high-density lipoprotein cholesterol (HDL-C) (Hitachi 917 autoanalyzer; Boehringer Mannheim Diagnostics, Indianapolis, IN) and triglycerides (Beckman Coulter, Brea, CA) were analyzed enzymatically. Insulin was analyzed with radioimmunoassay (Millipore, Billerica, MA). Plasma glucose was measured at the bedside using the StatStrip® Hospital Glucose Monitoring System (Novo Biomedical, Waltham, MA, USA). Hemoglobin A1c (HbA1c), was measured with HbA1c immunoassay analyzer (Siemens DCA Vantage, Siemens Medical Solutions, CA). Alanine aminotransferase (ALT) was determined via VITROS 5600 (Ortho Clinical Diagnostics, Rochester, NY) and was obtained the day of the screening visit, within 1 month of the MRI. SHBG was measured using an electrochemiluminescence immunoassay (Esoterix, Calbassas Hills, CA). Total testosterone was analyzed using a liquid chromatography–tandem mass spectrometry and free testosterone with equilibrium dialysis (Esoterix, Calbassas Hills, CA). Hepatic insulin resistance was estimated using HOMA-IR and calculated as (fasting plasma glucose [mg/dL] x fasting plasma insulin [mg/dL] / 405) (28).
Statistical analysis
Descriptive statistics were expressed as mean ± standard deviation (SD) (normally distributed) and median (interquartile range) (non-normally distributed). Student t-tests (normally distributed) or Wilcoxon tests (non-normally distributed) were used to compare variables between groups (NAFLD vs no NAFLD). Candidate variables were chosen to develop the model, based on clinical availability, relevance in other populations with NAFLD, and univariate comparisons between the 2 groups. Candidate variables were limited to nine in order to limit to 1 variable per approximately 10 participants. A logistic regression model with the outcome of NAFLD was fit to the model development cohort with all nine candidate predictor variables (full model): BMI percentile, waist circumference, HOMA-IR, HDL-C, triglycerides, ALT, free testosterone and SHBG. Next, backwards stepdown elimination using α=0.5 as the criterion to remain in the model and with total residual Aikaike Information Criterion (AIC) as the stopping criteria was used. This simplified model (PCOS-HS) was compared to the full model. The bootstrap (with 200 bootstrapped samples) was used to validate and to correct the over-optimism of the models. Predicted probabilities of NAFLD were obtained from the simplified model. Receiver Operation Curve (ROC) analysis was used to identify the cutoffs that maximized the Youden Index (high risk) and that excluded significant NAFLD (low risk) (negative predictive value >90%). Sensitivity (Se), specificity (Sp), positive and negative predictive values (PPV and NPV) were calculated. A nomogram to estimate the probability of NAFLD for a patient was obtained using the rms package in R(29). A web application was created using the shiny package in R(30), to provide an easy way to calculate the probability of NAFLD (https://childhealthbiostatscore.shinyapps.io/pcos-hs/). To validate the model using the external validation cohort, the PCOS-HS was fit to the data from the validation cohort, using the coefficients obtained from the development cohort. Se, Sp, PPV, NPV, were calculated using the same cutoff that maximized the Youden Index in the development cohort.
Statistical analyses were performed using R software version 3.4.3(31) and GraphPad Prism v. 7.04 (GraphPad Software, San Diego, CA)
RESULTS
Participant characteristics
The characteristics of the participants according to their NAFLD status are reported in Table 1. 52% of the participants in cohort 1 and in cohort 2 had a HFF ≥ 5.5% (total N=56 with NAFLD). The ethnicity of the two combined cohorts was 38% White Non-Hispanic, 44% White Hispanic, 9% Black and 2% Asian.
Table 1.
Participant characteristics.
| Development Cohort 1 |
Cohort 1 NAFLD HHF ≥5.5% |
Cohort 1 No NAFLD HFF < 5.5% |
Validation Cohort 2 |
|
|---|---|---|---|---|
| Number | 87 | 45 | 42 | 21 |
| Liver fat (%) | 5.8 (2.9–9.3) | 9.2 (7.3–12.1)a | 2.9 (2.1–4.0) | 5.6 (4.0–14.5) |
| Age (years) | 15 (14–17) | 15 (14– 17) | 16 (15– 17) | 16 (16–17) |
| Ethnicity (n (%)) | ||||
| White, non-Hispanic | 38 (44%) | 16 (36%) | 22 (52%) | 3 (14%) |
| White, Hispanic | 37 (43%) | 24 (53%) | 13 (31%) | 10 (48%) |
| Black | 7 (8%) | 2 (4%) | 5 (12%) | 3 (14%) |
| Asian | 0 (0%) | 0 (0%) | 0 (0%) | 2 (10%) |
| More than One | 5 (6%) | 3 (7%) | 2 (5%) | 3 (14%) |
| Treatment (n) | ||||
| Untreated | 83 | 42 | 41 | 19 |
| Metformin | 4 | 3 | 1 | 2 |
| Menarche Age (years) | 11 (11–12) | 12 (11–13) | 11 (11– 12) | 12 (10–12) |
| Hirsutism (FGS) | 6 (2–11) | 7 (4–13) | 5 (1–10) | 7 (3–11) |
| Anthropometric measurements | ||||
| BMI percentile | 98.4 (96.8–99.1) | 98.6 (97.0– 99.1) | 98.4 (96.2– 99.1) | 98.3 (97.6–99.0) |
| Waist circumference (cm) | 105 ± 11 | 106 ± 11 | 102 ± 10 | 108± 14 |
| Fasting laboratory measurements | ||||
| Free testosterone (nmol/l) | 0.26 (0.20– 0.38) | 0.29 (0.22– 0.44)c | 0.24 (0.18– 0.31) | 0.32 (0.25– 0.43) |
| SHBG (nmol/l) | 15.9 (12.2–23.2) | 13.8 (12.0– 20.0)b | 19.5 (13.3– 27.3) | 18.8 (12.3–22.5) |
| Total cholesterol (mmol/l) | 3.9 ± 0.8 | 3.9 ± 0.8 | 3.9 ± 0.8 | 4.0 ± 0.9 |
| HDL-C (mmol/l) | 0.9 ± 0.2 | 0.9 ± 0.2 | 1.0 ± 0.2 | 1.0 ± 0.2 |
| TG (mmol/l) | 1.33 (0.9– 1.78) | 1.46 (1.13– 2.06)b | 1.11 (0.85– 1.46) | 1.53 (1.19–1.85) |
| LDL-C (mmol/l) | 2.3 (1.8–3.1) | 2.1 (1.7– 3.1) | 2.4 (1.9– 3.2) | 2.1 (1.7–2.7) |
| ALT (U/L) | 35 (28–42) | 32 (24– 43)a | 21 (15– 31) | 35 (31–49)* |
| AST (U/L) | 38 (33–45) | 27 (20– 37) | 22 (15– 32) | 40 (32–52) |
| HbA1c (%) | 5.3 (5.2–5.6) | 5.4 (5.2– 5.6)c | 5.2 (5.1– 5.4) | 5.6 (5.4–5.8)* |
| HOMA-IR | 5.7 (4.0–7.5) | 6.7 (5.1– 8.5)a | 4.7 (3.4– 6.0) | 5.1 (3.6–9.5) |
| Fasting glucose (mmol/l) | 4.9 ± 0.4 | 5.0 ± 0.4 | 4.9 ± 0.4 | 4.7 ± 0.6* |
| Fasting insulin (pmol/l) | 181 (125–236) | 215 (167– 264)a | 139 (111– 201) | 222 (125–319) |
p< 0.001
p<0.01
p<0.05 between the group with NAFLD and the group without NAFLD within cohort 1.
p < 0.05 between cohort 1 and cohort 2.
Laboratory measurements associated with NAFLD
The group with NAFLD had higher ALT, free testosterone, triglycerides, fasting insulin, HOMA-IR, and HbA1c, and lower SHBG. Cohort 2 had higher ALT and HbA1c, but lower fasting glucose than the development cohort 1.
Full model for identification of NAFLD
The index based on all 9 candidate variables was: Probability of NAFLD =1/1+(exp (-(26.01+(−0.3761* BMI percentile + 0.05781* waist circumference (cm) + 0.0448* HOMA-IR + 0.00095519* HDL (mmol/L) + 0.00005892*TG (mmol/L) + 0.0964*ALT (IU/L) + 0.001548*free testosterone (nmol/L) −0.06806*SHBG (nmol/L)))). The odds ratio from the full logistic model were (OR (95%IC), P value): BMI percentile: 0.69 (0.44, 0.96), P =0.05; waist circumference: 1.06 (1, 1.14), P =0.08; HOMA-IR: 1.05 (0.89, 1.29), P = 0.62; HDL-C: 1.04 (0.96, 1.13), P =0.37; TG: 1.01 (0.99, 1.02),P =0.39; ALT: 1.1 (1.04, 1.18), P =0.003; free testosterone: 1.05 (0.92, 1.2), P = 0.49; SHBG: 0.93 (0.87, 0.99), P = 0.04.
Table 2 shows performance indices of the model corrected for optimism after 200 bootstrapped samples. Overall, the full model performed well, even after correction for optimism. The full model had a Somers’ D index of 0.65, an R2 of 0.41, a Brier index of 0.17 and an area under the ROC (AUCROC) of 0.83.
Table 2.
Performance indices of the logistic regressions models.
| Logistic Regression Model | Test dataset | Optimism- corrected value |
|---|---|---|
| A. Full model | ||
| dxy (Somers’ D) | 0.655 | 0.526 |
| R2 | 0.413 | 0.253 |
| Brier | 0.171 | 0.214 |
| B.Simplified Model | ||
| dxy (Somers’ D) | 0.624 | 0.500 |
| R2 | 0.374 | 0.223 |
| Brier | 0.178 | 0.217 |
In the original dataset and after correction for optimism with 200 bootstrap samples. For A) The full model and B) the simplified model (PCOS-HS index)
Simplified model for identification of NAFLD (PCOS-HS index)
The simplified logistic regression model included 4 variables: BMI percentile, waist circumference, ALT and SHBG. The equation was: Probability of NAFLD = 1/(1+(exp(−(25.19 +(−0.3411* BMI percentile) + (0.06149* waist circumference (cm)) + (0.09374*ALT (U/L)) + (−0.07954*SHBG (nmol/L))). The simplified model performed almost as well as the full model (Table 2). For the simplicity of its clinical use, with similar performance, we decided to use the simplified model (PCOS-HS index) as the model of choice.
ROC’s were computed with the PCOS-HS index (Figure 1). PCOS-HS index had an AUCROC of 0.81 (95% CI 0.72–0.90). The predicted probability that maximized the Youden index (≥ 0.44) was used as a high-risk cut-off with Se 82.2%, Sp 69.0%, NPV 78.4% and PPV 74.0%. 59% of cohort 1 fell into this category (Table 3). The cut-off for low-risk of NAFLD (NPV 90%) was ≤ 0.15 and 12% fell into this category.
Figure 1. Receiving Operating Curve.
A ROC Curve of the simplified model (PCOS-HS) (AUCROC 0.81). B ROC Curve of the full model (AUCROC 0.83). C ROC Curve of ALT (AUCROC 0.74).
Table 3.
Accuracy of the PCOS-HS Score in predicting NAFLD.
|
PCOS- HS |
Score | N (%) | Sensitivity | Specificity | NPV | PPV | Significance |
|---|---|---|---|---|---|---|---|
| Cut-off | |||||||
| ≤ 0.15 | ≤90 | 10/87 (12%) | 97.8% | 21.4% | 90% | 57.1% | Low Risk |
| 0.16-0.43 | 91–109 | 27/87 (31%) | Indeterminate | ||||
| ≥0.44 | ≥110 | 50/87 (57%) | 82.2% | 69.0% | 78.4% | 74.0% | High Risk |
Probability cut-off obtained from the calculation of PCOS-HS and corresponding clinical significance. The coresponding total score obtained from the nomogram (Figure 2) is in column 2. Number of patients from the development cohort, sensitivity, specificity, negative predictive value and positive predictive value are provided for the two cut-offs.
A nomogram (Figure 2) and web-based application (https://childhealthbiostatscore.shinyapps.io/pcos-hs/) were developed for ease of calculation. Cut-off scores corresponding to the different levels of probabilities and risk stratification are shown in Table 2.
Figure 2. Nomogram for probability calculation.
Points obtained for each variables are on the first line. By drawing a line corresponding to the value of each variable to this first line allow to give the according number of points. Points from each variable are added. The total number of points (Score) corresponds to the probability of HS and determines the risk category (Table 2). ≤ 90 points, represents a low risk for NAFLD, and ≥110, a high risk (Table 2).
ALT only
A ROC using ALT values alone (Figure 1) was assessed (7). The AUCROC was 0.74 (95% CI 0.64–0.85). The cut-off to maximize Youden index was 35.3 IU/L (ULN at Children’s Hospital Colorado: 35 IU/L) with a Se of 69%, a Sp of 76%, a NPV 70% and PPV 76%. Using the cut-off of >44 U/L for girls suggested by NASPGHAN guidelines, we obtained a low sensitivity (29%), but a very high specificity (93%).
External Validation sample
By applying the PCOS-HS score to cohort 2 (Table 4), we obtained a Se of 91%, a Sp 70%, a NPV 88% and a PPV 77% and an accuracy of 81% (p= 0.007).
Table 4.
Validation cohort 2.
| PCOS-HS score |
Outcome + (NAFLD) |
Outcome – (No NAFLD) |
Total (N (%)) |
Accuracy |
|---|---|---|---|---|
| ≥ 0.44 (High risk) | 10 | 3 | 13 (62%) | Sensitivity: 91% Specificity: 70% PPV: 77% NPV: 88% |
| < 0.44 | 1 | 7 | 8 (38%) | |
| Total | 11 (52%) | 10 (48%) | 21 (100%) |
Contingency table and accuracy obtained from the application of PCOS-HS index in cohort 2.
DISCUSSION
Even though PCOS is very common in adolescents, and that girls with PCOS and obesity have a prevalence of NAFLD as high as 50%, no screening tools exists to identify probable NAFLD in PCOS. In this study, we developed and validated a simple index allowing clinicians to screen for NAFLD in obese/overweight adolescents with PCOS. The PCOS-HS score uses simple clinical variables: BMI percentile, waist circumference, SHBG, and ALT. It has been developed as a first step tool for identifying adolescents who would benefit from further testing for NAFLD and NASH (e.g. MRI or Fibroscan©),) and follow-up surveillance of progression toward NASH and/or fibrosis, when identified as high risk.
Our PCOS-HS score has a similar accuracy as widely used scores in adults such as FIB-4 (AUCROC 0.80) or NAFLD fibrosis score (AUCROC 0.77) (32) which identify advanced fibrosis in adults s of both sexes. The PCOS-HS index is aimed at identifying not only patients with advanced fibrosis but all the patients with steatosis as they are at higher risk for cardiometabolic complications and rapid evolution to more severe liver disease (1, 2, 33). The Fatty Liver Index, routinely used in adults to screen for NAFLD, has been suggested as screening tool in EASL/EASD guidelines (9) with similar performance to our index (AUCROC 0.83; Se 76%, Sp 87%) (15, 34).
PCOS-HS complements other scores for NAFLD in pediatric populations. Saad V et al, developed a model that included metabolic syndrome, sex and ALT levels in a study of 129 overweight adolescents from 13–18 years old (8). This model performed similarly to our model to predict NAFLD (AUCROC of 0.85, a Se 72%, a Sp 82%, PPV 61% and NPV 89%). However, the sample size has a predominance of males and the authors were able to quantify cut-offs for females, thus it is not applicable in a female population with high prevalence of metabolic syndrome such as PCOS. Another score was developed in a small cohort of 56 obese white prepubertal children (29 with NAFLD) (13), using clinically accessible measurements including waist-to-height ratio, ALT and HOMA-IR. It had an AUCROC 0.88, a Se 89%, Sp 76%, PPV 77.5%, NPV 88%. However, the small sample size, lack of pubertal participants who are more insulin resistant that youth prior to puberty, and the ethnically homogenous population makes generalization limited, and it did not perform well in external validation (12). In a cross-sectional study of 134 girls 11–22 years old, a decision tree aiming to exclude girls with a low risk of NAFLD was developed(11). It included fasting insulin, total cholesterol, Hispanic ethnicity, and waist circumference. It had an excellent accuracy with an AUCROC 0.93 and an excellent NPV, PPV and Sp of 93%, 93% and 99% respectively. However, it had a Se of only 64%, which is problematic as a screening tool in a population with high prevalence of NAFLD. Therefore, our PCOS-HS score fills the need for a screening tool in pubertal adolescent girls with obesity and PCOS who have a uniquely higher pre-test probability for NAFLD, and is superior to all existing scores proposed in youth.
In the absence of better screening tools, hepatic transaminases are still recommended by various guidelines as a screening test (6, 7, 9). Unfortunately, transaminases are not sensitive enough to identify adolescents with NAFLD(35). The predictive value of ALT is even lower in overweight adolescent girls than boys(8, 35). The Screening ALT for Elevation in Today’s Youth (SAFETY) study raised awareness for values of ALT considered “normal” (35). By using the ULN for ALT in multiple US pediatric centers (median 53 IU/L ranging from 30–90 IU/L), NAFLD was detected with a Se of only 36% in girls. However, SAFETY determined a new ALT cut-off value (22.1 IU/L in girls), according to the 95th percentile of healthy female NHANES participants age 12–17 years old without any risk factors for liver disease. This cut-off had Se of 80% for NAFLD. In our study, the cut-off of ALT that maximized the Youden index was very close to the ULN of our lab (35 U/L, not sex-specific). However, using ALT alone was less sensitive and specific than the PCOS-HS model. Furthermore, applying the suggested cut-off from NASPGHAN guidelines (ALT >44 U/L) allowed to rule-in the presence of NAFLD (Sp 93%), but was not a sensitive (Se 26%) screening tool in our sample.
PCOS-HS index was developed to identify girls who need further testing for NAFLD with optimal sensitivity/specificity ratio. Within our development cohort with a high proportion of NAFLD (50%), as much as 43% were not considered at high risk for NAFLD, eliminating most of the girls that does not require further testing for NAFLD. However, of those, a third of our sample (N=26) felt in the indeterminate risk category and we could not exclude with high confidence the presence of NAFLD. In the context where NAFLD is a disease that tends to progress over years to more severe form, we consider it appropriate to consider the indeterminate category as a low risk, as these girls could be reassessed regularly with this non-invasive index. For the moment, using a cut-off of ≥ 0.44 for the PCOS-HS for those at high risk, not only provides a very good sensitivity (82–91%), but also a good specificity (70%) that avoids unnecessary further testing in a large proportion of girls. In the future, when new pharmacologic therapies may be available to prevent or treat NASH, it could be considered to reassess this analysis and use a cut-off with higher sensitivity and lower specificity when the benefits of identification are greater than the excessive testing associated with false positives.
Screening tools such as the PCOS-HS index to identify case of NAFLD in high-risk populations are needed to help target clinical intervention and follow-up and identifying cases at an early stage of the disease. Currently, it is well proven that lifestyle intervention using a target of 5% weight reduction is associated with a decrease of hepatic fat of approximately 30%, and a target of 7–10%, reduces hepatic fat by 40–50% and reduces inflammation (33). Identification of NAFLD in obese adolescents with PCOS should prompt more stringent lifestyle modifications (20). Multiple treatments are currently under investigation for the treatment of NASH and future development of markers allowing prediction of patient with NAFLD at high risk of progression to NASH or fibrosis may change clinical practice and patients eligible for treatment.
The PCOS-HS is readily applicable in clinical settings. We developed this index in an obese and overweight PCOS population, referred to a multidisciplinary PCOS clinic, that was unselected for the risk of NAFLD, typical of patients referred to an endocrine clinic. Our index is also generalizable to primary care practice, and necessitates only standard laboratory measurements obtained during the recommended initial assessment for PCOS. The population included in this study was multi-ethnic with equal proportions of White Non-Hispanic and White-Hispanic, but did include a low proportion of Asian or African American ethnicities. This has to be taken into account while generalizing to other clinical settings with higher proportion of patients from these ethnicities. Also, our index was developed in only one center, and whereas validation was done with a second sample, we did not assess validity in an external clinical center. Therefore, we propose that the PCOS-HS index be next validated in a multi-center setting with diverse ethnicities to show reproducibility. Even if ALT and SHBG are relatively standard assays (35), the exact calculation may be affected by the assay used for these measurements.
CONCLUSION
The PCOS-HS index can help identify obese/overweight adolescents with PCOS with probable NAFLD, using ALT, SHBG, BMI percentile, and waist circumference measurments. This easy screening tool will allow clinicians to appropriately select patients that would benefit from imaging assessment, follow-up, and treatment for NAFLD.
Acknowledgements:
The authors would like to thank the participants and their families, the CTRC staff, and Radiology staff.
Funding:
MCG: NIH BIRCWH K12HD057022; NIH NIDDK K23DK107871; Children’s Hospital Colorado; Doris Duke Foundation 2015212
AMC: Diabetes Canada Post-doctoral Fellowship.
Institution: NIH/NCATS Colorado CTSA Grant Number UL1 TR002535. Contents are the authors’ sole responsibility and do not necessarily represent official NIH views.
Footnotes
Data availability: The data that support the findings of this study are available from the corresponding author upon reasonable request.
Declaration of interest: The authors declare no conflict of interest
REFERENCES
- 1.Estes C, Razavi H, Loomba R, Younossi Z, Sanyal AJ. Modeling the epidemic of nonalcoholic fatty liver disease demonstrates an exponential increase in burden of disease. Hepatology. 2018;67(1):123–33. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Targher G, Byrne CD, Lonardo A, Zoppini G, Barbui C. Non-alcoholic fatty liver disease and risk of incident cardiovascular disease: A meta-analysis. J Hepatol. 2016;65(3):589–600. [DOI] [PubMed] [Google Scholar]
- 3.Cree-Green M, Bergman BC, Coe GV, Newnes L, Baumgartner AD, Bacon S, et al. Hepatic Steatosis is Common in Adolescents with Obesity and PCOS and Relates to De Novo Lipogenesis but not Insulin Resistance. Obesity (Silver Spring). 2016;24(11):2399–406. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Cree-Green M, Rahat H, Newcomer BR, Bergman BC, Brown MS, Coe GV, et al. Insulin Resistance, Hyperinsulinemia, and Mitochondria Dysfunction in Nonobese Girls With Polycystic Ovarian Syndrome. J Endocr Soc. 2017;1(7):931–44. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Wu J, Yao XY, Shi RX, Liu SF, Wang XY. A potential link between polycystic ovary syndrome and non-alcoholic fatty liver disease: an update meta-analysis. Reprod Health. 2018;15(1):77. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Legro RS, Arslanian SA, Ehrmann DA, Hoeger KM, Murad MH, Pasquali R, et al. Diagnosis and treatment of polycystic ovary syndrome: an Endocrine Society clinical practice guideline. J Clin Endocrinol Metab. 2013;98(12):4565–92. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Vos MB, Abrams SH, Barlow SE, Caprio S, Daniels SR, Kohli R, et al. NASPGHAN Clinical Practice Guideline for the Diagnosis and Treatment of Nonalcoholic Fatty Liver Disease in Children: Recommendations from the Expert Committee on NAFLD (ECON) and the North American Society of Pediatric Gastroenterology, Hepatology and Nutrition (NASPGHAN). J Pediatr Gastroenterol Nutr. 2017;64(2):319–34. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Saad V, Wicklow B, Wittmeier K, Hay J, MacIntosh A, Venugopal N, et al. A clinically relevant method to screen for hepatic steatosis in overweight adolescents: a cross sectional study. BMC Pediatr. 2015;15:151. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.European Association for the Study of the L, European Association for the Study of D, European Association for the Study of O. EASL-EASD-EASO Clinical Practice Guidelines for the Management of Non-Alcoholic Fatty Liver Disease. Obes Facts. 2016;9(2):65–90. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Arulanandan A, Loomba R. Non-invasive Testing for NASH and NASH with Advanced Fibrosis: Are We There Yet? Curr Hepatol Rep. 2015;14(2):109–18. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Rehm JL, Connor EL, Wolfgram PM, Eickhoff JC, Reeder SB, Allen DB. Predicting hepatic steatosis in a racially and ethnically diverse cohort of adolescent girls. J Pediatr. 2014;165(2):319–25 e1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Koot BG, van der Baan-Slootweg OH, Bohte AE, Nederveen AJ, van Werven JR, Tamminga-Smeulders CL, et al. Accuracy of prediction scores and novel biomarkers for predicting nonalcoholic fatty liver disease in obese children. Obesity (Silver Spring). 2013;21(3):583–90. [DOI] [PubMed] [Google Scholar]
- 13.Maffeis C, Banzato C, Rigotti F, Nobili V, Valandro S, Manfredi R, et al. Biochemical parameters and anthropometry predict NAFLD in obese children. J Pediatr Gastroenterol Nutr. 2011;53(6):590–3. [DOI] [PubMed] [Google Scholar]
- 14.Poynard T, Ratziu V, Naveau S, Thabut D, Charlotte F, Messous D, et al. The diagnostic value of biomarkers (SteatoTest) for the prediction of liver steatosis. Comp Hepatol. 2005;4:10. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Bedogni G, Bellentani S, Miglioli L, Masutti F, Passalacqua M, Castiglione A, et al. The Fatty Liver Index: a simple and accurate predictor of hepatic steatosis in the general population. BMC Gastroenterol. 2006;6:33. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Lee JH, Kim D, Kim HJ, Lee CH, Yang JI, Kim W, et al. Hepatic steatosis index: a simple screening tool reflecting nonalcoholic fatty liver disease. Dig Liver Dis. 2010;42(7):503–8. [DOI] [PubMed] [Google Scholar]
- 17.Holterman AX, Guzman G, Fantuzzi G, Wang H, Aigner K, Browne A, et al. Nonalcoholic fatty liver disease in severely obese adolescent and adult patients. Obesity (Silver Spring). 2013;21(3):591–7. [DOI] [PubMed] [Google Scholar]
- 18.Younossi ZM, Loomba R, Rinella ME, Bugianesi E, Marchesini G, Neuschwander-Tetri BA, et al. Current and future therapeutic regimens for nonalcoholic fatty liver disease and nonalcoholic steatohepatitis. Hepatology. 2018;68(1):361–71. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Cree-Green M, Newcomer BR, Coe G, Newnes L, Baumgartner A, Brown MS, et al. Peripheral insulin resistance in obese girls with hyperandrogenism is related to oxidative phosphorylation and elevated serum free fatty acids. Am J Physiol Endocrinol Metab. 2015;308(9):E726–33. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Network IP. International evidence-based guideline for the assessment and management of polycystic ovary syndrome.; 2018. [Google Scholar]
- 21.Ferriman D, Gallwey JD. Clinical assessment of body hair growth in women. J Clin Endocrinol Metab. 1961;21:1440–7. [DOI] [PubMed] [Google Scholar]
- 22.Dixon WT. Simple proton spectroscopic imaging. Radiology. 1984;153(1):189–94. [DOI] [PubMed] [Google Scholar]
- 23.Smits LP, Coolen BF, Panno MD, Runge JH, Nijhof WH, Verheij J, et al. Noninvasive Differentiation between Hepatic Steatosis and Steatohepatitis with MR Imaging Enhanced with USPIOs in Patients with Nonalcoholic Fatty Liver Disease: A Proof-of-Concept Study. Radiology. 2016;278(3):782–91. [DOI] [PubMed] [Google Scholar]
- 24.Younossi ZM, Loomba R, Anstee QM, Rinella ME, Bugianesi E, Marchesini G, et al. Diagnostic modalities for nonalcoholic fatty liver disease, nonalcoholic steatohepatitis, and associated fibrosis. Hepatology. 2018;68(1):349–60. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Caussy C, Reeder SB, Sirlin CB, Loomba R. Non-invasive, quantitative assessment of liver fat by MRI-PDFF as an endpoint in NASH trials. Hepatology. 2018. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Middleton MS, Van Natta ML, Heba ER, Alazraki A, Trout AT, Masand P, et al. Diagnostic accuracy of magnetic resonance imaging hepatic proton density fat fraction in pediatric nonalcoholic fatty liver disease. Hepatology. 2018;67(3):858–72. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Kuczmarski RJ, Ogden CL, Guo SS, Grummer-Strawn LM, Flegal KM, Mei Z, et al. 2000 CDC Growth Charts for the United States: methods and development. Vital Health Stat 11 2002(246):1–190. [PubMed] [Google Scholar]
- 28.Wallace TM, Levy JC, Matthews DR. Use and abuse of HOMA modeling. Diabetes Care. 2004;27(6):1487–95. [DOI] [PubMed] [Google Scholar]
- 29.Harrell F Regression Modeling Strategies: Springers International Publishing; 2006. 582 p. [Google Scholar]
- 30.Winston Chang JC, Allaire JJ, Xie Yihui and McPherson Jonathan. Web Application Framework for R. R package 1.1.0. ed2018. [Google Scholar]
- 31.Team RC. R: A language and environment for statistical computing. R Foundation for Statistical Computing; 2017. [Google Scholar]
- 32.Shah AG, Lydecker A, Murray K, Tetri BN, Contos MJ, Sanyal AJ, et al. Comparison of noninvasive markers of fibrosis in patients with nonalcoholic fatty liver disease. Clin Gastroenterol Hepatol. 2009;7(10):1104–12. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Stefan N, Haring HU, Cusi K. Non-alcoholic fatty liver disease: causes, diagnosis, cardiometabolic consequences, and treatment strategies. Lancet Diabetes Endocrinol. 2018. [DOI] [PubMed] [Google Scholar]
- 34.Fedchuk L, Nascimbeni F, Pais R, Charlotte F, Housset C, Ratziu V, et al. Performance and limitations of steatosis biomarkers in patients with nonalcoholic fatty liver disease. Aliment Pharmacol Ther. 2014;40(10):1209–22. [DOI] [PubMed] [Google Scholar]
- 35.Schwimmer JB, Dunn W, Norman GJ, Pardee PE, Middleton MS, Kerkar N, et al. SAFETY study: alanine aminotransferase cutoff values are set too high for reliable detection of pediatric chronic liver disease. Gastroenterology. 2010;138(4):1357–64, 64 e1–2. [DOI] [PMC free article] [PubMed] [Google Scholar]