Review article: the diagnosis of non-alcoholic fatty liver disease – availability and accuracy of non-invasive methods

Authors


  • This uncommissioned review article was subject to full peer-review.

Correspondence to:

Prof. D. Festi, Department of Medical and Surgical Science, S. Orsola-Malpighi Hospital, Via Massarenti 9, 40138 Bologna, Italy.

E-mail: davide.festi@unibo.it

Summary

Background

Non-alcoholic fatty liver disease (NAFLD) encompasses a wide spectrum of clinical conditions, actually representing an emerging disease of great clinical interest. Currently, its diagnosis requires liver biopsy, an invasive procedure not free from potential complications. However, several non-invasive diagnostic strategies have been proposed as potential diagnostic alternatives, each with different sensitivities and accuracies.

Aim

To review non-invasive diagnostic parameters and tools for NAFLD diagnosis and to formulate a diagnostic and prognostic algorithm for a better classification of patients.

Methods

A literature search was carried out on MEDLINE, EMBASE, Web of Science and Scopus for articles and abstracts in English. The search terms used included ‘NAFLD’, ‘non invasive method and NAFLD’, ‘transient elastography’ and ‘liver fibrosis’. The articles cited were selected based on their relevancy to the objective of the review.

Results

Ultrasonography still represents the first-line diagnostic tool for simple liver steatosis; its sensitivity could be enhanced by the complex biochemical score SteatoTest. Serum cytokeratin-18 is a promising and accurate non-invasive parameter (AUROCs: 0.83; 0.91) for the diagnosis of non-alcoholic steatohepatitis (NASH). The staging of liver fibrosis still represents the most important prognostic problem: the most accurate estimating methods are FibroMeter, FIB-4, NAFLD fibrosis score (AUROCs: 0.94; 0.86; 0.82) and transient elastography (AUROC: 0.84–1.00).

Conclusions

Different non-invasive parameters are available for the accurate diagnosis and prognostic stratification of non-alcoholic fatty liver disease which, if employed in a sequential algorithm, may lead to a reduced use of invasive methods, i.e. liver biopsy.

Introduction

Non-alcoholic fatty liver disease (NAFLD) encompasses a wide spectrum of conditions, ranging from a simple fatty liver to non-alcoholic steatohepatitis (NASH) with or without fibrosis, to cirrhosis and its complications (e.g. portal hypertension and hepatocellular carcinoma),[1, 2] occurring in patients with no or a low daily consumption of alcohol (≤30 g/day for men, ≤20 g/day for women).[3]

The prevalence of the disease varies in different epidemiological studies[4-7]; the majority report an average rate of NAFLD prevalence of 20–30% in Europe[3, 4] and in the Middle East,[3, 5] 15% in the Far East[3, 6] and 10–35% in the most of the United States[7] studies. However, these rates vary according to the technique used for establishing the diagnosis[7]; in fact, they range from frequencies <10% in studies, which employed elevated liver enzymes as non-invasive markers for NAFLD to 20–30% in ultrasound and autopsy studies.[7]

Although the incidence rates in the different epidemiological studies are probably underreported and vary widely,[7] NAFLD actually represents an emerging disease of great clinical interest. This is due both to its increasing incidence, mainly in western countries, according to the growing diffusion of unhealthy lifestyles and the increasing diagnostic accuracy in the identification of the disease; in fact, nearly 40% of previously diagnosed ‘cryptogenic cirrhosis’ is currently recognised as NAFLD.[3, 8]

Knowledge of the high NAFLD frequency, together with the demonstration of its evolutive nature in a significant proportion of patients, reinforces the need for a precise and early NAFLD diagnosis.

To simplify the NAFLD diagnosis, several studies have identified demographic and clinical risk factors for NAFLD such as advanced age,[9] male gender,[10] Hispanic ethnicity,[11] genetic predisposition (PNPLA3 gene)[12] and the presence of the main features of metabolic syndrome, namely obesity, type 2 diabetes and hyperlipidaemia.[4] These risk factors, as well as the knowledge of the natural history of NAFLD, could help the clinician to carry out a diagnostic work-up focused not only on NAFLD identification but also on staging and risk stratification regarding its evolution.

The diagnosis of NAFLD, and mainly of NASH, requires a liver biopsy,[13] which is an invasive procedure and is not without potential risks or complications.[14] However, several non-invasive diagnostic strategies have currently been proposed as potential diagnostic alternatives to liver biopsy, each with different sensitivities, specificities and accuracies.

Thus, the aims of this review were the evaluation of non-invasive diagnostic parameters and tools for NAFLD diagnosis, and formulating a diagnostic and prognostic algorithm for a better classification of patients with respect to fibrosis.

NAFLD diagnosis: invasive and non-invasive methods

Invasive methods

Liver biopsy

Liver biopsy currently represents the diagnostic gold standard for NAFLD[13] as it allows the exclusion of other causes of liver damage and the estimation of the severity of the fatty infiltration into the hepatocyte, lobular inflammation, hepatocyte ballooning and fibrosis.[15, 16] Even if liver biopsy represents the diagnostic gold standard, it has some limits; it is an invasive procedure, not free from potential risks and its interpretation is influenced by the subjective judgment of the pathologist.[17, 18] It also explores only a small portion of the liver parenchyma, not always representative of the damage to the entire parenchyma.

Moreover, several non-invasive scores and diagnostic tools have been proposed for the diagnosis of liver steatosis and NASH, and for the estimation of liver fibrosis. The applicability and limits of each of these tools will be discussed below.

Non-invasive methods of fatty liver

Serum markers

Different scores have been proposed for non-invasive NAFLD recognition (Table S1). Poynard et al.[19] proposed the SteatoTest, which is a combination of the FibroTest (Biopredictive, Paris France), previously validated as a surrogate marker of fibrosis in chronic hepatitis C[20] and B[21] and in alcoholic liver disease,[22] and the Acti-Test (Biopredictive, Paris, France), a surrogate marker of hepatocyte necrosis in chronic hepatitis C[20] and B,[21] plus body mass index, serum cholesterol, triglycerides, and glucose adjusted for age and gender. A cut-off value of 0.30 had 91% sensitivity, while a cut-off of 0.70 had 89% specificity for the diagnosis of significant steatosis (steatosis grade >5%). The diagnostic performance is shown in Table S1. SteatoTest has been successfully validated for the diagnosis of NASH and fibrosis in a population of patients at risk for metabolic syndrome,[23] and for the diagnosis of advanced steatosis among morbidly obese patients.[24]

The fatty liver index (FLI)[25] is a complex algorithm, based on four simple and routine parameters, such as body mass index (BMI), waist circumference, triglycerides and gamma-GT, which lead to the prediction of the presence of a fatty liver with an accuracy of 84%. It varies from 0 to 100, and a FLI <30 (negative likelihood ratio = 0.2) excludes the presence of NAFLD, while a FLI ≥60 (positive likelihood ratio = 4.3) confirms the diagnostic suspicion. This score had previously been designed to identify patients at high risk for NAFLD, necessitating ultrasonography (US) and intensification of lifestyle counselling; successively, it was validated as a good prognostic marker, useful in predicting increased intima-media thickness, coronary heart disease risk and reduced insulin sensitivity,[26] as well as 9-year incident diabetes[27] and early atherosclerosis.[28]

The lipid accumulation product (LAP) is an index, which is calculated using waist circumference and fasting triglycerides; it has previously been employed as a predictor of cardiovascular risk,[29] diabetes[30] and overall mortality in patients with a high cardiovascular risk.[31] The natural logarithm of LAP, adjusted for gender, is a reasonably accurate approach for recognising individuals with liver steatosis, necessitating US.[32]

Moreover, to identify patients affected by definitive NASH, there are several other scores, which are calculated using clinical and biochemical markers[33-38]; their diagnostic performances are reported in Table S2.

Currently, the most promising serum marker is represented by cytokeratin-18 (CK-18), a marker of hepatocyte apoptosis, which can be employed alone[36] or combined with other parameters[39] and is highly accurate (AUROC: 0.83[36]; 0.91[39]) in diagnosing NASH patients; in a recent study, CK-18 was the most accurate parameter for the diagnosis of NAFLD and NASH, in particular, when it was used in combination with fibroblast growth factor (FGF)-21.[40] Moreover, CK-18 is the only externally validated marker for NASH[41] and, even if it has been suggested by recent available guidelines,[42] it is thought that its extensive use in clinical practice is too premature to be recommended.

Ultrasonography (US)

Ultrasonography represents a non-invasive, inexpensive and widely available method, useful for the detection of liver steatosis with a sensitivity of 60–94% and a specificity of 66–95%[43]; it also leads to a subjective estimation of the entity of fatty infiltration on a three or more point scoring system (mild, moderate and severe).

This method is limited by its low sensitivity for mild steatosis, its inability to differentiate mild fibrosis from steatosis and to accurately quantify fatty infiltration,[43] its operator dependency with an inter-observer agreement of 72%[44] as well as its scarce applicability for obese patients and those with an excess of gas in the intestine.

A further application of US in the study of NAFLD is represented by contrast-enhanced US (CEUS) with Levovist (Schering, Berlin, Germany), a specific agent for postvascular liver-specific imaging. However, even if the results of this application have been promising in differentiating between simple steatosis and NASH, only one study has reported this possible application[45] and, furthermore, Levovist is not currently commercially available.

Transient elastography

Transient elastography (TE) (FibroScan; Echosens, Paris, France) is a new tool, designed for the non-invasive study of liver stiffness. TE uses an ultrasound transducer probe, mounted on the axis of a vibrator. Vibrations of mild amplitude and low frequency are transmitted by the transducer, inducing an elastic shear wave, which propagates through the underlying tissue. Pulse-echo ultrasound acquisitions are used to follow the propagation of the shear wave and to measure its velocity, which is directly related to tissue stiffness. Transient elastography measurements of liver stiffness are expressed in kilopascals (kPa).

In the diagnosis of NAFLD, TE has documented conflicting results, mostly due to the fact that TE is difficult to perform in case of obesity, since subcutaneous fatty tissue attenuates elastic share wave, reducing success rate and diagnostic performance.[46]

A promising new application of TE (Echosens, Paris, France) is the vibration-controlled transient elastography (VCTE) device, which allows the calculation of a new parameter, the controlled attenuation parameter (CAP), useful for the non-invasive and accurate estimation of liver steatosis.[47] It estimates the entity of ultrasonic attenuation generated by ultrasound propagation through a fatty liver; thus, it expresses the total ultrasonic attenuation at the central frequency of the M probe (3.5 MHz) in dB.m-1. In a preliminary study,[48] the CAP significantly correlated with steatosis grade (Spearman ρ = 0.81), and it was able to satisfactorily differentiate between the different degrees of severity of steatosis and to detect it even at early stages (≥11%). Thus, it is a non-invasive, immediate, objective and efficient method of detecting and quantifying liver steatosis, but it requires additional validation on larger populations.[47]

Computed tomography (CT)

Noncontrast-enhanced CT is characterised by a good diagnostic performance in the qualitative diagnosis of greater degrees of steatosis (>30%) with a sensitivity of 82% and a specificity of 100%.[49] Three main criteria could be employed for the detection of hepatic steatosis: the absolute measurement of x-ray hepatic attenuation <40 Hounsfield units (HU) (n.v. 50–57 HU), a hepatic attenuation value <10 HU compared with that of the spleen and a spleen to liver attenuation ratio <0.8.[43] However, the radiation exposure for patients, its scarce accuracy in the presence of other underlying diffuse liver diseases, which may alter liver attenuation values, and the significant variation in absolute liver attenuation values obtained using different machines represent important limitations, which could limit its wide use for the study of NAFLD.[43]

Magnetic resonance (MR)

There are two main techniques available for the assessment of hepatic steatosis: proton magnetic resonance spectroscopy with 1H (1H MRS) and magnetic resonance imaging (MRI) for the measurement of the signal-fat fraction or the proton density fat-fraction.[43]

1H MRS allows the study of the molecular composition of tissue in vivo of a selected volume of interest; 1H spectra from liver tissue usually shows two dominant signal portions, the water signal and the signal of methylene and methyl protons of fatty acids whose peaks reflect liver fat content.[43] The quantification of liver steatosis by 1H MRS is closely correlated with liver fat content as assessed by liver biopsy and with the entity of its decrease after weight loss[50]; thus, it represents a promising non-invasive tool for the quantification of steatosis and for the monitoring of patients after diet or pharmacological intervention.

Magnetic resonance spectroscopy exploits the principle of chemical shift, that is the different resonance frequency of hydrogen, depending on the chemical bond (methylene groups of triglycerides vs. water). The fat-signal fraction represents the ratio of the signal from hydrogen nuclei in fat to the sum of the signals from hydrogen nuclei in free water and fat; thus, it is a biomarker of liver triglyceride concentration, while the proton density fat fraction is defined as the density of hydrogen protons from fat normalised from the total hydrogen proton density from all mobile proton species.[51] Employing a fat fraction threshold of 5.56% to differentiate the normal from the abnormal fat fraction, MR imaging achieves an accuracy of 100% in the detection of liver steatosis.[51]

Even if it represents an accurate tool for the detection and dynamic quantification of liver steatosis, it is expensive and not widely available.

In conclusion, taking into consideration the several methods available for hepatic steatosis diagnosis, US represents the first-line diagnostic tool because it permits not only the evaluation of liver steatosis but also a comprehensive study of the other abdominal organs. However, in case of a discrepancy between the clinical suspicion of liver steatosis and a negative US examination, a more sensitive test, even if more complex to use in clinical practice, such as the SteatoTest, can be used. Furthermore, both MR and CT represent additional available tools, but they are expensive and not widely available or applicable in clinical practice.

Non-invasive diagnosis of fibrosis

Serum markers

The main available serum markers for the detection of fibrosis in NAFLD are shown in Table S3.[52-58] Comparative studies on the diagnostic performance of different fibrosis markers for NAFLD have reported the superior performance of the FibroMeter[54] (AUROC; 0.94) or FIB-4[58] (AUROC: 0.86) or complex scores, such as the Hepascore, Fibrotest (AUROC: 0.802–0.858) as compared with simple scores (APRI, BARD), in particular for the diagnosis of advanced fibrosis.[57]

Moreover, FIB-4 has recently been validated as a surrogate marker of fibrosis in a Japanese population of patients affected by NAFLD, reporting the best accuracy (AUROC: 0.87) and the best negative predictive value (NPV) (98%) for the non-invasive diagnosis of fibrosis as compared with other scores.[59]

A recent meta-analysis[41] has reported that the FibroTest, ELF and NAFLD fibrosis scores had significantly better diagnostic accuracy than the BARD score, and their AUROCs did not significantly differ from each other. Moreover, the most validated score for the diagnosis of significant fibrosis was the NAFLD fibrosis score.[41] This score can easily be calculated by employing routine clinical parameters and has a good diagnostic performance; however, a nonnegligible percentage of patients (from 20 to 58%)[41] fall into the ‘grey area’. FibroMeter testing used for diagnosing NAFLD could also easily be calculated from simple parameters and employing the cut-off values with the best (≥90%) NPV and positive predictive value (PPV) (≤0.611 and ≥0.715 respectively) for the diagnosis of significant fibrosis; 97.4% of patients would be correctly classified, thus avoiding liver biopsy.[55] The better accuracy of the FibroMeter for the diagnosis of fibrosis F ≥ 2, as compared with other scores and with the NAFLD fibrosis score, was confirmed in a second study, reporting an AUROC of 0.95.[60] This score could thus easily be employed for the accurate differentiation of patients with significant fibrosis, combining it with a second non-invasive test, such as transient elastography, for patients who fall into the ‘grey area’, to create a diagnostic algorithm characterised by high diagnostic accuracy.

Transient elastography

Table 1 illustrates the diagnostic performance of transient elastography for the estimation of liver fibrosis in NAFLD. The different studies[46, 61-64] report various failure rates, ranging from 3[61] to 16%,[53] due to a high body mass index (BMI ≥ 30) or waist circumference,[63] which could interfere with the transmission of the ultrasound and the elastic impulses, thus limiting the correct estimation of liver stiffness.

Table 1. Diagnostic performance of transient elastography for the detection of fibrosis in non-alcoholic fatty-liver disease (NAFLD)
StudyPatientsFibrosis stageCut-offs (kPa)AUROCSe (%)Sp (%)PPV (%)NPV (%)
  1. AUROC, area under ROC curve; Se, sensitivity; Sp, specificity; PPV, positive predictive value; NPV, negative predictive value; NAFLD, non-alcoholic fatty liver disease; NASH, non-alcoholic steatohepatitis.

Yoneda et al.[61]97 NAFLD

F ≥ 2

F ≥ 3

F4

6.6

9.8

17.5

0.86

0.90

0.99

88

85

100

74

81

97

79

64

75

85

93

100

Nobili et al.[62]52 NASH

F ≥ 2

F ≥ 3

7.4

10.2

0.99

1

100

100

92

100

80

100

100

100

Wong et al.[63]246 NAFLD

F ≥ 2

F ≥ 3

F4

7.0

8.7

10.3

0.84

0.93

0.95

79

84

92

76

83

88

70

59

46

84

95

99

Lupsor et al.[64]72 NASH

F ≥ 2

F ≥ 3

6.8

10.4

0.78

0.98

67

100

84

97

69

71

88

100

Petta et al.[46]169 NAFLD

F ≥ 2

F ≥ 3

7.25

8.75

0.79

0.87

69

76

70

78

67

50

72

92

Moreover, a new probe (XL probe), specifically designed for obese patients (BMI ≥ 30), has been proposed for the evaluation of liver stiffness.[65, 66] The new XL probe is characterised by a lower central ultrasound frequency (2.5 vs. 3.5 MHz) as compared with the standard M probe, a lower vibration amplitude (3 mm vs. 2 mm), a smaller tip diameter (9 mm vs. 12 mm) and, in particular, the explored region of interest (ROI) of the new probe is located more deeply (3.5 cm vs. 2.5 cm from the skin surface), thus overcoming the diagnostic difficulty due to the interposition of thickened subcutaneous adipose tissue. Comparing the diagnostic performances of the two probes, the XL probe obtains more reliable results as compared with the M probe (failure rates of 7% vs. 35%[65] and 24% vs. 55%,[66] respectively, defined as failure to obtain 10 valid measurements). In fact, the new probe obtained more successful results in patients with a skin-to-liver capsule distance >2.5 cm in which the standard probe failed in 77% of cases.[66] The main limiting factors for the new probe were a skin-to-liver capsule distance ≥3.4 cm[66] and extreme obesity (BMI ≥40 kg/m2).[65]

The two probes reached comparable accuracy in the estimation of fibrosis compared with liver biopsy, even if the optimal reference cut-offs for the XL probe were lower than those obtained by the standard probe.[65] The reason for this bias was probably the interposition of adipose tissue or the hepatic liver capsule in the region of interest explored by the M probe, which could have led to an overestimation of liver stiffness.[65]

Moreover, the discordance rate between liver fibrosis estimated by biopsy and transient elastography by XL probe was low (11.4% of patients)[67]; severe obesity and elevated liver stiffness represented the independent risk factors associated with histological discordance.[67]

Acoustic radiation force impulse (ARFI)

Acoustic radiation force impulse imaging (Virtual Touch Tissue Quantification, Siemens ACUSON S2000) is a new device, which explores the elastic properties of a region of interest (ROI) while performing real-time B-mode imaging. Liver tissue of the ROI is mechanically excited using short-duration acoustic pulses of 2.67 MHz. The displacement tissue generated produces a propagating shear wave whose velocity is calculated (m/sec) and which is proportional to tissue elasticity. The advantage of this device is that the ROI can be freely placed at different depths and in different parts of the liver, evaluating large parts of the liver in one examination. Furthermore, it does not suffer from the same limitations as TE in obese patients and in the presence of thickened adipose tissue.

There is only one study that analyses the diagnostic performance of ARFI imaging, as compared with TE, for the evaluation of liver fibrosis in a population of patients affected by NAFLD.[68] No significant difference was found when comparing the diagnostic accuracies of TE with M and XL probes and ARFI imaging of the right and the left lobe for the diagnosis of significant fibrosis, severe fibrosis and cirrhosis. Moreover, the correlation with the histological degree of liver fibrosis was statistically significant for TE, but not for ARFI imaging.[60]

The possible limitations of this diagnostic tool are that the precise depth and site of the measurements as well as the required numbers of measurements are not standardised; in addition, the reference parameters for the precise definitions of ARFI failure have yet to be determined.[68] Moreover, ARFI imaging has a very narrow range of values separating the cut-offs for significant fibrosis and cirrhosis and is thus less accurate than TE in the detection of significant fibrosis[69]; finally, ARFI imaging has not yet been approved for diagnostic use by the Food and Drug Administration (FDA).

MR elastography

Magnetic resonance elastography is a new tool, useful for studying the elastic properties of a region of interest (ROI) of the liver parenchyma. The device consists of an active driver, located outside the magnet room, which generates continuous low-frequency vibrations, transmitted via a flexible tube to a drum-like passive driver placed directly against the anterior right chest wall over the liver. After identifying the ROI, the waves generated on the parenchyma are visually transposed onto a quantitative elastogram image to calculate the stiffness values, expressed in kilopascals (kPas).

In a recent study carried out on a population of patients affected by NAFLD,[70] the mean values of liver stiffness, as assessed by MR elastography for simple steatosis, NASH without fibrosis and hepatic fibrosis were significantly different (P = 0.028; P = 0.030). Moreover, a cut-off value of 2.74 kPa differentiated patients with histologically confirmed NASH from those with simple steatosis with high accuracy (AUROC: 0.93; sensitivity: 94%; specificity: 73%).

Diagnostic algorithm for NAFLD diagnosis

Taking into consideration the different diagnostic performances of the non-invasive parameters available for the diagnosis of liver fibrosis in NAFLD, we tried to formulate a speculative sequential algorithm, selecting the diagnostic tool characterised by the best diagnostic performance, to reduce the number of false positives/negatives of each method. Figure 1 shows our proposed diagnostic algorithm for the correct diagnostic classification of patients with suspected NAFLD.

Figure 1.

Diagnostic algorithm for the management of suspected NAFLD. The algorithm illustrated is a proposal for the accurate diagnostic classification and management of patients with suspected NAFLD, employing different non-invasive tests in sequence to increase overall diagnostic accuracy and eliminate the need for liver biopsy.

The suspicion of NAFLD may derive from the presence of risk factors and the alteration of laboratory tests, such as the elevation of alanine aminotransferase (ALT), (usually greater than aspartate aminotransferase (AST), except in advanced liver disease), the elevation of gamma-glutamyltransferase (gGT), ferritin and transferrin saturation, in a specific clinical setting in which other potential causes of hepatic damage have been excluded.

The non-invasive diagnosis of a fatty liver could be confirmed by carrying out the SteatoTest[19] or by using the FLI[25] and an abdominal US. Patients with a documented fatty liver could then be stratified according to the risk of the presence of significant fibrosis, by carrying out a non-invasive test, for example, the FibroMeter for NAFLD.[55]

FibroMeter testing for NAFLD can easily be calculated from simple parameters, such as age, weight, platelet count, ferritin, glucose, AST and ALT, routinely evaluated in clinical practice. Moreover, two studies[55, 60] have reported good accuracy in the diagnosis of significant fibrosis (97% accuracy and AUROC: 0.90 respectively), higher than that of other non-invasive scores.[60] In this algorithm, two optimal cut-offs (≤0.611 and ≥0.715) are proposed, which are characterised by the best NPV and PPV to exclude or confirm, respectively, the risk of F ≥ 2.[55] For patients who fall into the ‘grey area’ (values between 0.612 and 0.714), a second non-invasive test could be carried out to increase overall diagnostic accuracy. The second test proposed is transient elastography, in which two reference cut-off values (7.9 kPa and 9.6 kPa) with the best sensitivity and specificity for ruling out and confirming, respectively, advanced fibrosis in patients affected by NAFLD[63] could be employed. Thus, in this proposed algorithm, liver biopsy would be suggested only for patients who fall into the ‘grey area’ also after the second test (values from 7.9 to 9.6 kPa).

In conclusion, different non-invasive parameters are available for the accurate diagnosis and prognostic stratification of NAFLD which, if employed in a sequential algorithm, may lead to a reduced use of invasive methods, i.e. liver biopsy. Thus, according to the natural history of NAFLD, all patients with a low risk of advanced disease, eventually diagnosed by one of above non-invasive parameters, could be referred to primary care, whereas subjects at high risk for advanced disease should be sent to specialists for the evaluation of the degree of fibrosis and the choice of specific NASH management.

Authorship

Guarantor of the article: Davide Festi.

Author contributions: RS, LM and DM conceptualised, searched and reviewed the literature. RS and ES drafted the manuscript. LM and GB constructed the tables and illustrations. ARDB and AC contributed to the data discussion. DF and GMR conceptualised and critically reviewed the manuscript. All authors contributed to and have approved the final manuscript.

Acknowledgement

We thank Gerald Goldsmith, MD, for reviewing the English language in the text.

Declaration of personal and funding interests: None.

Ancillary