INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

In 1986 our research group derived reference values for slow vital capacity (VC) and flow-volume curves from "normal" subjects participating in the baseline survey of the Po River Delta epidemiological study (northern Italy) (1). The mathematical technique proposed by Knudson and colleagues (2) was applied to the distribution of VC, Forced Vital Capacity (FVC), and FEV₁, separately in males and females. Stratified linear regressions were fitted on the three age intervals corresponding to the growth, plateau, and decline phases of lung function. The same age intervals were used for all spirometric indexes. This approach introduces conflicting estimates at the points of transition between equations that fit different lung function phases (3, 4). Further, using the same age intervals for flows and volumes may cause an inadequate fitting of lung function data (4).

Modeling the complex dependence of lung function on age by fitting continuous nonlinear functions might represent an alternative approach (4). However, the functional form usually required involves the estimation of a large number of parameters, which may yield poor precision. On the other hand, an overly simplified functional form would lead to biased estimates and incorrect inferences. Piecewise polynomial regressions may provide enough flexibility yet keep the number of parameters relatively small. In particular, natural cubic splines provide curves that are continuous over the entire age span up to the second derivative, yielding smooth predictions that are linear in the extreme age intervals (7).

The aim of this study was to derive reference values for VC and flow-volume curve indexes from the "normal" subjects who participated in one or both cross-sectional surveys of the Po river delta study by applying natural cubic splines. The advantages of newly proposed reference equations (cubic splines models) over the old ones (stratified linear regressions), in modeling lung function data, have also been assessed by comparing the present with the previously published reference values.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Po River Delta Study

The Po River Delta epidemiological study on obstructive pulmonary disease is a prospective study conducted on the general population of this rural area in northern Italy (near Venice). A baseline survey was carried out during 1980-1982 on 3,284 subjects of whom 2,136 (65%) were followed in 1988-1991, to whom 705 new subjects were added (8). Characteristics of the sample, the questionnaire, and the lung function tests protocols are reported elsewhere (1, 9).

Selection Criteria of "Normal" Subjects

Only the "normal" subjects were selected, using information obtained by the interviewer-administered CNR questionnaire (a modified version of the National Heart, Lung, and Blood Institute questionnaire), on the basis of the absence of: past or current active respiratory symptoms; past or current cardiorespiratory and neurologic diseases, and hypertension; childhood respiratory diseases; past or current active tobacco smoking history (including occasional); past or current allergic rhinitis; past or current occupational exposure to known bronchial irritants; acute respiratory diseases within 30 d before the filling out of the questionnaire. The selection of normal subjects from the total baseline sample and during the follow-up time according to the criteria reported above is shown in Table 1.

Lung Function Measurements

In both surveys, the same computerized pneumotachograph (Fleisch No. 3) (47804/s Pulmonary System; Hewlett-Packard, Waltham, MA) was used for measuring flow and volume. The protocol fulfilled the ATS recommendations (13, 14), except for the end-point criterion of the FVC maneuver (15). At least two trials were repeated to obtain a satisfactory VC value. The highest VC was used for statistical analyses. Afterwards, as many as eight FVC maneuvers were performed to obtain at least three acceptable trials. Among them, the two largest FVC and FEV₁ values should not vary more than 5%. The largest FVC and FEV₁ values were selected, regardless of the maneuver. From the flow-volume curve with the largest sum FVC + FEV₁, the following flows were measured: peak expiratory flow (PEF); forced expiratory flow between 25 and 50%, and between 75 and 85% of FVC (FEF_25-75 and FEF_75-85); maximal expiratory flow at 50 and 75% of FVC (MEF₅₀ and MEF₇₅).

Height (in centimeters) and weight (in kilograms) were measured in standing position without shoes in subjects wearing clothes. Age at last birthday was recorded.

Statistical Analysis

The nonlinear dependence of lung function on age was modeled by natural cubic splines for all indexes except for FEV₁/FVC and FEV₁/ VC whose dependence on age appeared to be linear. They provide smooth function and continuous first and second derivatives over the whole age range (16). In the extreme intervals, where fewer observations are available, natural cubic splines are constrained to be linear. Second-order polynomials for height and body mass index (BMI = weight/height², expressed in kg/m²) were also introduced in the models as covariates. Smooth continuous prediction curves were obtained for volumes, flows, and the ratios FEV₁/VC and FEV₁/FVC, separately for men and women. The "normal 95th percentile" in Tables 3 and 4 corresponds to the percent predicted above which values from 95% of the study population fell (2).

Data from individual subjects were independent, but repeated measures could be correlated. Thus, estimates for the parameters and standard errors were obtained by random-effects models where all the independent variables had fixed and random components (17, 18). All pairwise interactions were tested and were not significant at 0.05 level and were dropped out of the models.

Prediction plots of the lung function measures versus age are shown for "average" subjects whose height and BMI were computed using Lowess smoothing (robust locally weighted regressions) of height and BMI on age (19).

A detailed description of the natural cubic splines regression models is reported in the Appendix , along with an example of calculation of reference values for FEV₁.

All statistical analyses were performed using the 1997 Stata Statistical Software release 5.0 (Stata Corporation, College Station, TX).

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Descriptive statistics of study population are shown in Table 2. Six hundred forty-five normal subjects (225 males 8 to 64 yr of age and 420 females 8 to 63 yr of age) participated in the baseline survey only, 146 (34 males aged 8 to 56 yr of age and 112 females 8 to 62 yr of age) in both surveys, and 248 (110 males 8 to 64 yr of age, and 138 females 8 to 70 yr of age) in the second survey only. Thus, 1,039 subjects provided 1,185 observations. No significant difference was observed at baseline for age, height, BMI, or spirometric measurements between participants in the first survey only and participants in both surveys (p values > 0.10 by Wilcoxon's rank-sum test).

Estimates for coefficients of the reference equations for VC, FEV₁, FVC, FEV₁/FVC, and FEV₁/VC, are reported in Table 3. Nearly all the covariates' estimates were significantly different from zero in both sexes for all the equations, the main exception being FEV₁/VC in males. The directions of the effects of BMI and height were all the same across sexes. The negative estimates for the coefficients of BMI-squared imply that all lung function predictions increase as BMI increases up to a certain value (e.g., for FEV₁ it was 25.7 kg/m² in males and 29.9 kg/m² in females) and decrease afterwards according to a parabolic trend. All lung function reference equations increased in a slightly curvilinear fashion as height increased. For VC, FEV₁, and FVC, the splines break points that maximized the goodness of fit were estimated at older ages in males than in females. For the two ratios, age was a significant linear predictor. The "normal 95th percentile" was 82 to 83% for the volumes and 86 to 88% for the ratios.

The effect of age on VC, FEV₁, and FVC as modeled by the natural cubic splines with two break points is shown in Figures 1-3, separately for males and females. The predictions computed for "average" height and BMI subjects (as described in METHODS) are represented by solid lines. The continuous curves smoothly fit the scatterplots. As expected, males show larger expired volumes and, in addition, they peaked earlier than females, whose growths and declines appeared flatter.

View larger version (14K): [in this window] [in a new window]   View larger version (20K): [in this window] [in a new window]   Figure 1.   Prediction curves for VC versus age for males (left panel  ) and females (right panel  ) with "average" height and body mass index. Solid lines, natural cubic splines; scattered open circles, raw measurements.

View larger version (15K): [in this window] [in a new window]   View larger version (23K): [in this window] [in a new window]   Figure 2.   Prediction curves for FEV1 versus age for males (left panel  ) and females (right panel  ) with "average" height and body mass index. Solid lines, natural cubic splines; scattered open circles, raw measurements.

View larger version (14K): [in this window] [in a new window]   View larger version (21K): [in this window] [in a new window]   Figure 3.   Prediction curves for FVC versus age for males (left panel  ) and females (right panel  ) with "average" height and body mass index. Solid lines, natural cubic splines; scattered open circles, raw measurements.

Estimates for the coefficients of the reference equations for flows are reported in Table 4. Compared with the indexes in Table 3, somewhat weaker associations were found between the pulmonary function indexes and the predictors. Fewer estimates were significantly different from zero, for the variability of PEF, FEF_25-75, FEF_75-85, MEF₅₀, and MEF₇₅ was higher than the variability of VC, FEV₁, and FVC. Nonetheless, the signs of the coefficients of BMI and height for PEF, FEF_25-75, and MEF₅₀ were consistent with those of VC, FEV₁, and FVC. As in Table 3, the splines break points that maximized the goodness of fit were estimated at older ages in males than in females. The "normal 95th percentile" was 77% for PEF in both sexes and within 50 to 66% for the other flows.

Reference values for VC in males (Figure 4, top panel) and females (Figure 4, bottom panel) are plotted versus age. They were obtained on the same subjects by the natural cubic splines proposed in this report and by the stratified regressions derived by Paoletti and colleagues (1). Equations obtained by the splines were smooth and rid of "jumps." The splines showed slightly steeper increases in the younger members in both sexes, and flatter decreases in the older males only. They peaked at 29 yr of age when VC was equal to 5.26 L (28 yr, VC = 5.75 L by stratified regressions) in "average" males, and at 32 yr when VC was equal to 3.77 L (23 yr, VC = 3.85 L by stratified regressions) in "average" females. Further, the predicted decline of VC from its maximum to its value at 65 yr of age was 1.05 L by the splines and 1.67 L by the stratified regressions in males, and 0.75 L by the splines and the stratified regressions in females. Similar plots, not shown, were also observed for FEV₁ and FVC in both sexes.

View larger version (8K): [in this window] [in a new window]   View larger version (8K): [in this window] [in a new window]   Figure 4.   Prediction curves for VC versus age for males (left panel  ) and females (right panel  ) with "average" height and body mass index. Solid line, natural cubic splines; open circles, stratified linear regressions, by Paoletti and colleagues (1).

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

We derived smooth continuous prediction equations for VC, FVC, and derived indexes. Estimates for the equations' coefficients were obtained from the "normal" subjects participating in one or both cross-sectional surveys of the Po River Delta study.

The growth and decline of lung function over the lifetimes present several issues as to how to choose appropriate prediction equations. Fitting nonlinear models over the entire age span may involve complicated mathematical functions. To avoid biased estimates, they usually require the estimation of a large number of parameters. Instead, smoothing techniques often provide useful tools for deriving prediction equations. In particular, regression splines were successfully applied to modeling lung function growth and decline over time by reasonable and parsimonious equations (7). We fitted natural cubic splines for age over the three phases (growth, plateau, and decline) by setting two break points. This technique provides curves that are preferable to stratified regressions because they are rid of "jumps" and are smooth (continuous up to the second-order derivatives). Natural cubic splines also improve on regular cubic splines in that they are constrained to be linear in the extreme age intervals, where fewer observations are available and erratic predictions could be obtained. This feature was particularly useful in dealing with our few subjects, especially males, older than 60 yr of age. These new reference equations show an improvement on continuity and smoothness compared with the stratified regressions applied by Paoletti and colleagues (1), although the former may tend to somewhat flatten out the highly variable peak of lung function over age. The use of longitudinal observations may partly explain the fact that the decline of lung function in the elderly appears flatter than in the reference equations derived by Paoletti and colleagues from the first cross-sectional sample (1, 4, 20).

Very restrictive selection criteria of normality were applied to our sample, including absence of childhood respiratory diseases, past or current neurologic diseases, hypertension, and occupational exposure, which are seldom applied for selecting "normal" subjects. Few of the baseline normal subjects returned to the second survey and were still normal (146 out of 1,139 subjects), mainly because of beginning to smoke or incidence of respiratory symptoms. The few longitudinal complete data taken at two points in time only were not sufficient to assess longitudinal changes in lung function. Further, the estimated equations were more affected by the cross-sectional observations than by the longitudinal ones. Nevertheless, samples taken at two points in time tended to attenuate the extent of the cohort effect.

Some selection bias might be present in our study because of the 267 subjects lost to follow-up (34% of the baseline normal subjects). In fact, the probability of returning to the second survey might vary across subjects with different risk factors and pulmonary function. However, no baseline characteristic appeared to be significantly different between normal subjects with different follow-up patterns (Table 2). The technical bias caused by lung function testing should have played only a minor role in our study. In fact, although the two surveys were carried out 8 yr apart, "normal" subjects were examined with the same instrument and after the same ATS standardized procedures.

High BMI values and weight gain were recently related to decline of FEV₁ and FVC in adults of both sexes (21). In our sample, BMI appeared to be an important predictor along with age and height. The increase of pulmonary function within low BMI values may reflect increasing physical complexion, and the decrease with high BMI values may be due to obesity which limits the mobility of the thorax, as observed by others (24).

The use of the newly proposed reference equations in clinical routine may contribute to improve the assessment of the patients' lung function. As of now, reference values are frequently obtained by stratified linear equations that cause insensible "jumps" in the predictions at particular ages, e.g., by the old equations the predicted VC value for a male 175 cm tall was 5.66 L at 27 yr, rising to 5.75 L at 28 yr, and sharply dropping to 5.42 L at 29 yr. Sharp changes of predictions may affect clinical interpretation of lung function measurements, particularly when patients are examined within a short period of time.

In this report, we provide reference values also for VC and FEV₁/VC, which play a key role in the interpretation of lung function testing. In fact, the current criteria of the European Respiratory Society for airway obstruction are based on the percent predicted value of FEV₁/VC (25), and ATS suggests the use of percent predicted of VC for assessing the severity of restrictive abnormalities (3).

In conclusion, we suggest the use of smooth continuous equation as reference for lung function indexes. Although calculation for splines variables might appear to be somewhat cumbersome, spirometer-linked computers easily provide predictions for any given patient. Further, we encourage the inclusion of BMI for the calculation of the reference values insofar as this simple measurement improves the predictions.