Establishing normal reference values in quantitative computed tomography of emphysema

Summary

Quantitative computed tomography (QCT) can provide reliable and valid measures of lung structure and volumes. Similar to lung function and volumes measured by spirometry, lung measures obtained by QCT vary by demographic and anthropomorphic factors including sex, race/ethnicity and height in asymptomatic non-smokers. Hence, some accounting for these factors is necessary to define abnormal from normal QCT values and disease severity. Similar to spirometry and cardiac volumes, prediction equations for QCT may be derived from a sample of asymptomatic individuals to estimate reference values.

This paper describes the methodology of reference equation development using, as an example, quantitative densitometry to detect pulmonary emphysema. The process described is generalizable to other QCT measures, including lung volumes, airway dimensions and gas-trapping. Pulmonary emphysema is defined morphologically by airspace enlargement with alveolar wall destruction and has been shown to correlate with low lung attenuation estimated by QCT. Deriving reference values for a normal quantity of low lung attenuation requires three steps: First, criteria that define normal must be established. Second, variables for inclusion must be selected based on an understanding of subject, scanner and protocol specific factors that influence lung attenuation. Finally, a reference sample of normal individuals must be selected that is representative of the population in which QCT will be used to detect pulmonary emphysema. Sources of bias and confounding inherent to reference values are also discussed.

Reference equation development is a multistep process that can define normal values for QCT measures such as lung attenuation. Normative reference values will increase the utility of QCT in both research and clinical practice.

Introduction

Quantitative computed tomography (QCT) can provide reliable and valid measures of lung structure including parenchymal attenuation and airway dimensions.^1–9 These structural changes are important to the pathophysiology of many lung disorders, such as chronic obstructive pulmonary disease (COPD), emphysema, and asthma.^10,11 An understanding of the normal variation of QCT measures of lung structure is necessary in order to maximize its utility in research and clinical practice. The methodology of reference value derivation will be discussed here using, as an example, lung attenuation for detection of pulmonary emphysema. The concepts, however, can be extended to the development of reference values for other QCT measures, including lung and lobar volumes, airway dimensions and gas-trapping.

Pulmonary Emphysema and QCT

Pulmonary emphysema is defined morphologically as permanent enlargement of air spaces distal to the terminal bronchiole, accompanied by destruction of their walls.¹² Although emphysema is defined by gross and microscopic anatomic appearances, QCT has been shown to correlate well with pathologic specimens permitting detection in vivo, and can estimate the total and regional burden of emphysema.^1,13–18 In addition, QCT has superior reproducibility characteristics compared to visual assessment of emphysema.^3,9

A widely used QCT measure of pulmonary emphysema is the percentage of lung voxels below a specific Hounsfield unit (HU) threshold (e.g., −910 HU or −950 HU).^5,19 First described by Müller and colleagues, percent emphysema is a continuous measure provides an estimate of the quantity of emphysema-like lung.¹ The term emphysema-like lung is emphasized because, despite significant statistical correlation with pathologic specimens,^1,15,16,18 percent emphysema can be influenced by several factors unrelated to emphysema and can be observed among individuals without emphysema assessed visually.^5,20,21

Variation in Lung Volumes and Percent Emphysema in Healthy Individuals

Lung function and volumes measured by spirometry and plethysmography have long been noted to vary substantially in apparently healthy individuals by demographic factors and body size.^22–30 Hankinson reference equations, which allow calculation of a predicted normal value and limit of normal for an individual of a given age, sex, race/ethnicity and height, are used widely in clinical and research settings to account for this variation.²⁷ Recently statements recommend that airflow limitation, for example, be defined as a value below the lower limit of normal based upon such equations.³¹ A similar approach has been adopted for cardiac volumes.^32,33

Total lung volume and other thoracic volumes measured on QCT are highly likely to exhibit similar variation by demographic factors and body size in asymptomatic individuals, as are ratios such as the ratio of low lung attenuation to total lung volume, a metric of emphysema severity.¹ Multiple studies confirm this supposition and demonstrate that percent emphysema varies in asymptomatic individuals by, at a minimum, age, sex and body size.^2,20,34–39 This variation in asymptomatic individuals is often large and, not infrequently, larger than that observed in individuals with disease. Hence, in defining normal values for percent emphysema from abnormal, it may be helpful to account for these factors and possibly others in order to permit accurate classification of emphysema status and severity using QCT.

Approaches to Defining Disease

The optimal approach to defining a threshold of abnormality for diseases that are common in the general population is not obvious and varies considerably by disease. Arguably, the best approach to define such a disease, particularly an asymptomatic disease, is to pick a threshold above or below which treatment is proven to improve outcomes. Hypertension is an example of a disease defined following this approach. Randomized clinical trials have demonstrated a proven benefit (i.e., reduced cardiovascular events) of reducing systolic and diastolic blood pressure to 140 and 90 mm Hg,^40–43 respectively. Subsequent randomized clinical trials that targeted blood pressure reduction below these thresholds failed to show a benefit for clinical events,^44,45 hence hypertension is defined for most patients by these criteria.⁴⁶ Similarly, diabetes is defined by an absolute glycemic threshold at which specific complications (e.g., diabetic retinopathy) occur.⁴⁷ In the context of emphysema, a symptomatic disease for which effective, specific therapies are lacking,⁴⁸ a reasonable approach may be to define normal based on the distribution of the measure in asymptomatic individuals in the general population. This strategy has the goal of helping to define the relevance of QCT measures of emphysema to clinical practice and to facilitate the study of the pathobiology of emphysema and development of novel, emphysema-specific therapies.

Establishing Normal Reference Values for Percent Emphysema

Percent emphysema is a continuous measure that correlates with pathologic emphysema but also varies in healthy individuals. ^{1,15,16,18,20,21,36,37} Similar to spirometry,²⁷ development of reference equations for percent emphysema should attempt to minimize misclassification of disease (i.e., emphysema) by accounting for normal variation from subject and scanner factors. Once these factors have been measured in a representative sample of healthy individuals, regression modeling can be used to derive reference equations that minimize differences between predicted and observed values (i.e., residuals). Knowing the residual variance, disease presence or severity may then be defined in a study or clinical population based on observed deviation from normal reference values.^27,30,49 The process of reference equation development will be discussed in detail below.

Defining who is normal

Defining normal is not trivial. In the context of a QCT measure to detect pulmonary emphysema, the ideal definition of normal is anyone who is disease-free, i.e., without emphysema. In the absence of a gold-standard method of detecting emphysema pre-mortem, one might reasonably begin by excluding individuals with respiratory symptoms or diagnoses of chronic lower respiratory disease. Indeed, this approach has been used to develop reference equations for spirometry.^27,30 It should be noted that defining normal by the absence of respiratory symptoms or disease may introduce some misclassification by including individuals with emphysema who underreport symptoms or do not seek medical attention.⁵⁰ Similarly, respiratory diseases other than emphysema that influence lung attenuation may also be excluded from the reference sample. For example, gas-trapping is a physiologic consequence of airways obstruction that is associated with low lung attenuation that occurs in asthma.² Excluding individuals with asthma-related respiratory symptoms and gas-trapping may bias the derived lung attenuation reference values for detection of emphysema. This inherent limitation to defining normal must be kept in mind when using reference values to define disease.⁵¹

Knowledge of disease risk factors can also help define normal individuals. For example, smoking is the major risk factor for COPD and low lung function; hence reference equations used to define normal values for spirometry exclude individuals with any history of smoking.^27,30 Smoking is also a risk factor for centrilobular emphysema,⁵² although the relationship between smoking history (i.e., pack-years) and percent emphysema is much weaker than that for lung function.⁵³ Nonetheless, exclusion of smokers may help to minimize misclassification in the reference sample.

However, current smoking appears to paradoxically influence lung attenuation independent of emphysema, possibly related to increased inflammatory cells in the lung of current smokers compared to former smokers.³⁸ Therefore, excluding current smokers completely will mean that the derived reference values for emphysema will be biased, potentially severely, when applied to current smokers. An alternative approach may be to include a secondary analysis in which current smokers are included in order to allow estimation of a term in the regression model that accounts for the difference in lung attenuation associated with current smoking or cigarettes per day.

Selecting predictors of variation in percent emphysema

Optimal use of reference values to detect emphysema by QCT requires that other sources of lung attenuation variation are known. Elsewhere in this issue of the Journal of Thoracic Imaging Coxson has reviewed sources of variation in QCT of the lungs. Briefly, with respect to lung attenuation these can be divided into scanner, protocol and subject-specific factors. Scanner-specific factors include model and manufacturer (particularly for newer scanners) whereas protocol-specific factors include tube current, slice thickness and number, reconstruction algorithm, and attenuation threshold. Provided a large enough reference sample with accurate ascertainment of these variables, a derived reference equation will account for much of the variability introduced by these factors. Scanner and protocol factors in the reference sample used to derive reference equations should be representative of those in the population where QCT will be applied. Although protocol-specific factors can be fixed by a highly standardized protocol, as used in most contemporary studies, scanner changes usually necessitate protocol changes. Scanner changes therefore represent a particular for challenge for reference equation development given the heterogeneity and rapid evolution of CT technology.^{(Ref Newell CT protocols from this issue)}

Subject-specific factors known to influence lung attenuation include age, gender, body size, socio-economic status, depth of inspiration, smoking, and gas-trapping.^2,20,34–39 Based on population-based studies of spirometry, one might speculate that QCT measures of lung structure also differ by race-ethnicity.^27,30,54 Inclusion of these variables in a regression model can, if necessary, account for differences in lung attenuation associated with these factors. It should be noted, however, that these predictors may directly or indirectly mask true causes (and presence) of emphysema. For example, low body weight may be causally related to emphysema.⁵⁵ Inclusion of a body weight term in reference equations would obscure this relationship. Furthermore, differences in lung attenuation associated with subject specific factors may reflect confounding by an unmeasured variable that causes emphysema. For example, exposure to biomass fuel combustion in developing countries may cause emphysema to differ by gender. If this exposure is not included in the reference regression model, the role of biomass fuel combustion would be masked by gender. Users of reference equations should be aware of the inputs used as predictors and the potential for over adjustment and residual confounding.⁵¹

Selecting a representative sample

The sample of asymptomatic individuals used to derive percent emphysema reference equations should be representative of the population in which QCT will be applied to detect emphysema. Specifically, the sample should reflect the range of scanner and subject variables encountered in the general population from which cases of disease (i.e., COPD or emphysema) are to be defined. Furthermore, the characteristics of the reference sample, including scanner and subject specific factors, should be reported. This will help to limit extrapolation of reference values beyond the observed data in the derivation sample. Finally, a large and diverse reference sample is essential in order to increase the precision of normal estimates by accounting for the numerous permutations of input predictors.

Summary

QCT can provide reliable and valid measures of lung structure in vivo. Application of this technology to study the biology and clinical significance of lung disease will be facilitated by normal reference values. Derivation of reference equations is a multistep process. First, one must define a normal sample population (i.e., free of the disease of interest). Second, one must identify scanner and subject specific factors that influence QCT measures independent of the disease of interest. Third, regression of QCT measures must be performed on a large reference sample representative of the population in which these measures will be applied. Developers and users of reference equations should have an understanding of the bias and confounding inherent in using normal reference values. Analogous to reference values for spirometry, establishing reference values for QCT will reduce misclassification of emphysema and help advance research and clinical management of lung disease.