Selecting and using reference values

Reference values are important in the interpretation of lung function tests. The technologist and laboratory can assure their testing equipment and techniques are performed according to international recommendations; however, if they do not select an appropriate reference equation for their specific patient population, the results of the test and the effect on the patient’s outcome may be compromised. After all, a clinician does not always review specific “numbered” data but whether the results are deemed normal or abnormal. The American Thoracic Society-European Respiratory Society (ATS-ERS) recommend a laboratory select its reference values based on a “like population” to the subjects it tests, and reference sets that used similar instrumentation and testing protocols. The laboratory should also assure the reference equations used are consistent throughout its organization to reduce intra-laboratory variability. An example is a 2011 survey conducted in the greater Cleveland area noted three different reference sets were used for spirometry alone. Another issue associated with reference sets is they are 30-40 years old, and the instrumentation and testing techniques have changed significantly (e.g., volume versus flow spirometry, Dlco analyzer technology). The ATS-ERS has assembled a Global Lungs Initiative (GLI) task force and charged them with establishing improved lung function reference values.

Reference values for pulmonary function tests are derived by statistical analysis of a population of healthy subjects. These subjects are classified as healthy because they have no history of lung disease in themselves or in their families. Minimal exposure to risk factors, such as smoking or environmental pollution, is usually considered in selecting these individuals.

All lung function measurements vary in healthy individuals. Some tests vary much more than others. Arterial pH and Paco₂ have a very narrow range in healthy individuals. However, FEF_25%–75% may vary by as much as ±2 L/sec. This variability becomes important when measured values are compared with reference values. Most measurements regress; that is, they vary in a predictable way in relation to one or more physical factors. The physical characteristics that most influence pulmonary function include the following:

By analyzing each pulmonary function variable in regard to the individual’s physical characteristics, regression equations can be generated to predict the expected value.

Race or ethnic origin influences stature and body proportions. Lung function, particularly lung volumes, differ significantly among races. Some computerized pulmonary function systems apply a “correction factor” to reference values for whites to adjust for different races. Although differences in lung function among races are well documented, no single correction factor is applicable to all measurements. Some laboratories reduce reference values for volumes (e.g., FVC, TLC) by factors of 10%–15% for African-Americans. Separate regression equations derived from healthy individuals of each race tested are preferred. Race-specific reference values should be used if they are representative of the population the laboratory tests. Self-identification is the accepted standard for defining race with no adjustments for mixed percentages.

Diffusing Capacity

The ATS-ERS did not recommend a specific set of reference equations for diffusing capacity, citing inter-laboratory variability as their reason. Published data have shown even in a well controlled clinical trial that intersession variability can range from 10%-25%. The ATS-ERS statement did recommend that predicted values for alveolar volume (V_A), inspired volume (V_I), and Dlco should come from the same source. Figure 13-1 demonstrates the difference in a subject between various Dlco predicted equations. A 60-year-old female of average height can have a predicted value ranging from approximately 21-27 mL/min/mm Hg, depending on the reference set selected. Thompson and others published a reference set in middle aged to older subjects (ages 45-71), which complied entirely with the 2005 ATS-ERS recommendations for testing technique and quality assurance. Their equations compared favorably with those previously published by Miller. Several common reference authors are listed in Table 13-2.

Table 13-2

Common Reference Authors for Diffusing Capacity

Adult: Author	Year	Journal
Crapo	1986	Am Rev Respir Dis 1986; 134:856
Cotes	1993	Eur Respir J 1993; 6(Suppl 16):41–52
Knutson	1987	Am Rev Respir Dis 1987; 135:805-811
Miller	1983	Am Rev Respir Dis 1983; 127:270-277
Paoletti	1985	Am Rev Respir Dis 1985; 132:806-813
Thompson	2008	Thorax 2008; 63:889-893
Pediatric:
Hsu	1979	J Pediatr 1979; 95:14-23
Nasr	1991	Pediatr Pulmonol 1991; 10:267-272

Several methods for applying reference values are used:

• Tables of reference values

• Nomograms

• Graphs

• Regression equations used in calculators or software

Tables, nomograms, and graphs are rarely used because of the widespread availability of computerized systems. Peak flowmeters and other simple devices designed for use outside of the clinic or laboratory sometimes use a nomogram or printed graph to allow the user to look up a predicted value. The use of computers (or calculators) allows regression equations to be available in software. In most automated systems, the user selects sets of prediction equations best suited to the population being tested. Some software allows users to enter their own equations or modify published equations. This provides a means of adding new reference equations as they become available.

Establishing what is abnormal

Determining the lower limit of normal (LLN) should be done by analyzing some measure (e.g., FVC, FEV₁) in healthy subjects and then determining the variability of that measurement. In clinical medicine, the 5th percentile is often defined as the LLN because it represents the segment of healthy subjects farthest below the average. Even though subjects in the 5th percentile are healthy, they are arbitrarily defined as “abnormal” for clinical purposes. Figure 13-2 depicts the predicted and the LLN for white females from ages 8–80 years (NHANES III). It is noteworthy that the statistical LLN is approximately the same across the adult age range.

Some clinicians use a fixed percentage (measured value divided by the reference value × 100) of the reference value to determine the degree of abnormality. Eighty percent (80%) is often used as the limit of normal. Unfortunately, this method leads to errors because the variability around the predicted value is relatively constant in adults. In other words, the scatter of normal values does not vary with the size of the predicted value. Figure 13-3 illustrates why using fixed percentages, such as 80% of the predicted, can lead to misclassification. In tall, young subjects 80% of the predicted is often less than the 5th percentile; using 80% as the limit can allow a patient who really does have decreased lung function (in the 5th percentile or lower) to be misclassified as normal. This situation is a false-negative result; the patient has disease but the test does not indicate abnormality. Similarly, an elderly patient who is short may have a lung function parameter that is less than 80% of predicted but well within the statistically normal range (above the 5th percentile). This short elderly subject would be misclassified as having lung disease when in fact she is within the “normal” range (i.e., a false-positive result). Using percents of predicted introduces both age and height biases. The situation is slightly different in children because the variability of lung function measures tends to change proportionately with the size of the predicted value. For this reason, percents of predicted values may be appropriate for classifying lung function in children.

A more statistically sound approach for classifying abnormality is to compute the z score or standard deviation score (SDS). If lung function varies in a normal fashion (a Gaussian or bell-shaped distribution curve; Figure 13-4), the mean ± 1.96 standard deviation (SD) defines the 95% confidence interval. Statistically, 95% of the healthy population falls within approximately 2 SD of the mean. The remaining subjects fall into either the highest or lowest 2.5% of the distribution. The z score or SDS can be calculated easily if the variability (residual standard deviation (RSD)) of the reference population is known:

< ?xml:namespace prefix = "mml" />zscore=(measured−predicted)RSD

where:

RSD = residual standard deviation

The RSD is the normal variability that remains when all other sources of variability have been accounted for in the regression. If an individual’s z score is less than −1.65, there is only a 5% chance that the test result is normal. If the z score is less than −1.96, the measured value is found in only 2.5% of healthy subjects.

For example, consider a male subject who is 70 years old and 69 inches (175 cm) tall. His FEV₁ is measured as 2.40 L; his predicted FEV₁ is 3.12 L. His FEV₁ is 77% of predicted; is this abnormal? Using 80% as the cutoff suggests that this patient has mild lung disease. However, if the patient’s z score is calculated as:

zscore=(2.40−3.12)0.468 =−0.720.468 =−1.53

where 0.468 is the residual standard deviation from the reference population, the z score of −1.53 suggests that this subject is above the 5th percentile and likely has normal lung function. The advantage of z scores is that they can be used for any index that is normally distributed. Because the z score accounts for the variability occurring in healthy subjects, it tells how common, or uncommon, the finding may be in the patient being studied.

For many pulmonary function variables, only the LLN (i.e., below the mean) is significant. For example, it is not usually clinically significant if FVC is greater than predicted, only if it is lower. For normally distributed variables, 1.645 × RSD can be considered the LLN. Variables that can be abnormally high or low (e.g., RV, TLC, Paco₂) must consider the upper limit of normal (ULN) in a similar manner.

The LLN can be easily calculated when the variable of interest (e.g., FEV₁, FVC) is normally distributed in the population. Using the 5th percentile to define the LLN, however, does not require the pulmonary function variable to be normally distributed in the population. Simple counts can determine the level for a specific variable that separates the lowest 5% of the subjects from the remainder. Lower limits of normal using the 5th percentile are sometimes defined for specific groupings of age or gender.

There are several areas in which the definition of lung function abnormality may have important clinical consequences. One such area is the use of a fixed ratio to define airway obstruction, as is frequently done with the FEV₁/FVC (FEV₁/VC). The World Health Organization’s Global Initiative for Obstructive Lung Disease (GOLD) recommends the use of 70% as a cutoff, with ratios less than this value defining the presence of airway obstruction. However, because the FEV₁/FVC ratio falls with age (sex, height, and ethnicity also may play a role), using a fixed ratio may misclassify younger subjects as normal (false negative) and older subjects as obstructed (false positive) (Figure 13-5). Similarly, using fixed percentages of predicted (e.g., 80%, 50%) to categorize the severity of obstruction may misclassify subjects who are young and tall or old and short (as discussed in a preceding paragraph). These misapplications of fixed ratios and fixed percents of predicted can have serious consequences for individual patients and for large groups of subjects when research is involved. Misclassifying an elderly subject as having COPD may mean the inappropriate prescription of drugs that can have serious side effects. Using an inappropriate classification, such as an FEV₁/FVC ratio of 70%, to exclude subjects from a clinical trial (because they are incorrectly classified as “obstructed”) means that otherwise healthy subjects are not exposed to the treatment or drug being evaluated.

Pulmonary function laboratories should try to choose reference studies from a population similar to the patients they test. The following factors may be considerations in selecting reference values:

Individual laboratories may wish to perform measurements on subjects who represent a healthy cross-section of the population that the laboratory usually tests. Doing this in a statistically meaningful way may require testing a large number of subjects. However, measured values from these individuals can then be compared with expected values using various reference equations. Equations that produce the smallest average differences (measured – predicted) may be preferable. Evaluation of a small number of individuals may not show much difference between equations for FVC and FEV₁. However, there may be noticeable discrepancies for Dlco or maximal flows. Equations for spirometry, lung volumes, and Dlco should be taken from a single reference, if possible. If healthy individuals fall outside of the limits of normal, the laboratory should examine its test methods, how the individuals were selected, and the prediction equations being used. Table 13-3 lists “typical” normal values for pulmonary function and blood gas parameters.