The models were based on 3,598 non-Hispanic white subjects aged 4 to 80 years, 2,182 (60.5%) of whom were younger than 20 years and 271 (7.5%) of whom were younger than 8 years. A description of the demographic characteristics of the study population can be found in . Children from the NHANES III reference were taller (height-for-age
z score) (
16) and heavier (weight-for-age
z score) (data not shown) than children in the other three datasets. On the basis of the original distributions of each of the three outcomes (FEV
_{1}, FVC, and FEF
_{25–75}) studied, height and age were nonlinear, the spread of values around the mean was nonuniform for both height and age, and there was also evidence that the distributions of all three outcomes were skewed. FEF
_{25–75} results were not available for the British data, so models for these data are based on 700 fewer subjects.
| TABLE 2.SUMMARY OF DATA INCLUDED IN ANALYSIS |
Median Trends
The models for all three outcomes were dependent on height and age, and logarithmic transformation of both the outcome and explanatory variables was necessary. The resulting models describe a multiplicative and allometric height relationship, where all three spirometric outcomes are proportional to height raised to the power 2.5. For example, a 1% increase in height corresponds to a 2.5% increase in spirometry. The median volumes for each of the outcomes, smoothed by age, are presented in . Despite age and height being highly correlated, there was a significant and independent effect of age after adjusting for height ().
Variability
The LMS method also quantifies the spread of values around the median, which is essential information when determining the range of expected lung function values in a normal population. After adjustment for the effects of height and age, the between-subject variability, characterized by the CV, demonstrated important age-related trends (). The between-subject variability was highly age dependent, being greatest in children younger than 11 years and increasing steadily with increasing age in adults after the age of 30. The variability of FEF_{25–75} was noticeably larger than for FEV_{1} and FVC. The commonly quoted “normal range” of 80 to 120% predicted assumes a CV of 10%; however, as can be seen from , even for FVC, this only occurs over a limited age range of 15 to 35 years. By contrast, at 5 to 6 years of age, the CV for FEV_{1} and FVC is 15%, corresponding to a normal range of 70 to 130% predicted. The CV for FEF_{25–75} at age 5 to 6 years is 20%, corresponding to 60 to 140% predicted, and by age 50, the CV for FEF_{25–75} has widened to 30%, a normal range of 40 to 160%.
Skewness
After adjustment for height and age, there was little evidence of skewness for FEV_{1} and FVC. By contrast, there was significant skewness in FEF_{25–75} and the FEV_{1}/FVC ratio for both sexes, which was incorporated into the prediction models.
The age-related changes in FEV_{1} and FVC were accompanied by age-related changes in the ratio (FEV_{1}/FVC) (). As can be seen, the frequently quoted predicted FEV_{1}/FVC of 0.7 is not in fact attained until around 50 years of age in males and considerably later in females, being noticeably higher during childhood and lower in the elderly. The range of “normal values” for this ratio is age dependent, being wider in both the young and the elderly, and sex differences are apparent, with females having greater predicted values of FEV_{1}/FVC than males at all ages and which are most marked in late puberty ().
Between-Center Differences
The models were further explored by evaluating the extent to which between-center differences affected the expected reference range. After adjustment for height and age, there were small but significant between-center differences in FEV_{1} for both males and females and in FVC for females. Compared with NHANES III, median values from Lebecque (12) and Corey (13) were 2 to 3% greater after adjustment, whereas those from Rosenthal and colleagues (7) were approximately 4% smaller. Interestingly, no between-center differences were observed for FVC in males or for FEF_{25–75} in either sex.
Ethnic Differences
The NHANES III African-American subjects had considerably lower FEV_{1} and FVC, but similar flows and FEV_{1}/FVC compared with non-Hispanic white subjects (). With the exception of FVC in females, Mexican Americans had similar values to non-Hispanic whites. Ethnic and racial differences varied according to sex, generally being more marked in females. Of significance is that the standard deviations for each of the sex-specific ethnic z scores were approximately 1, which could facilitate development of race- and sex-specific adjustment factors to account for the shift in values.
Comparison with the Original NHANES III Equations
compares the current model with the original NHANES III equations in terms of the median and the lower limit of normal. Although the new model is not dramatically different from the original, three major advantages of the current approach can be seen. First, the current models extend the reference down to 4 years of age, thereby improving the accuracy with which normal values can be predicted in very young children; it can be seen that the original NHANES III equations underpredict lung function in healthy children younger than 10 years and therefore fail to identify early lung disease. Second, smoothly changing curves describe the transition between childhood and early adulthood. Third, the age-dependent between-subject variability is quantified, thereby allowing improved precision with which to define the lower limits of normal at all ages.
Reference Equations
The methods used do not produce equations
per se but comprehensive look-up tables that can be applied in a Microsoft Excel add-in module. The module can be found at
www.growinglungs.org.uk (Pediatric Reference Ranges for Spirometry). The module can also be easily implemented into current commercial spirometers, upon request by manufacturers. The program facilitates prospective interpretation of a single observation or retrospective analysis of an entire dataset to calculate
z scores, % predicted, or centiles.