Abstract
Rationale: Forced expiratory volume in one second (FEV1), an important measure of pulmonary disease severity in patients with cystic fibrosis (CF), is frequently expressed as a percentage of a predicted value derived from a healthy reference population. There are limitations to comparing the lung function of a patient with CF to that of healthy control subjects, and potential advantages to comparing it to that of other patients with CF.
Objective: To estimate CF-specific percentiles of FEV1 as functions of height, age, and sex.
Methods: We used 287,108 FEV1 observations among more than 21,000 patients with CF in the CF Foundation National Patient Registry between 1994 and 2001. The percentiles were estimated using quantile regression methods.
Results: FEV1 percentile “growth grids” are presented, allowing comparison of an individual's FEV1 to that of patients with CF of the same sex, age, and height. Their potential uses in clinical practice and research are illustrated.
Conclusions: CF-specific reference equations allow individual patients' FEV1 to be placed in the context of the distribution of lung function of their peers with CF, and should improve generalizability of CF clinical trials by setting entry criteria that are equitable across sex and age ranges. They may serve as a useful adjunct to conventional reference equations.
The hallmark of cystic fibrosis (CF) is chronic, progressive obstructive lung disease. Pulmonary function, specifically the FEV1, is an important marker of CF lung disease severity (1, 2), widely employed in both the clinical and research arenas. Serial monitoring of lung function at clinic visits is used to assess disease progression, inform clinical management, and aid in lung transplant referral decisions (3, 4). FEV1 has served as the primary outcome measure in most clinical trials of therapies for CF pulmonary disease (5–8), and entry criteria for CF clinical trials generally include a range of FEV1 values.
Pulmonary function is generally expressed as a percentage of a predicted value based on sex, age, and height, calculated from regression equations derived from a healthy reference population. Expressing pulmonary function in relation to a predicted value facilitates the comparison of values over time in an individual growing subject, and allows comparison of lung function between subjects. However, there are important limitations to comparing the lung function of patients with CF to that of a healthy reference population. First, in order for percent-predicted values to be valid in a given population, the reference population must be similar with respect to age and height (9, 10). This criterion is frequently not met for patients with CF, who, on average, have markedly lower height-for-age values than those of the general population: between 6 and 20 years of age, the median height of patients with CF is between the 20th and 30th percentiles for the general population (11). Thus, for patients with CF with low height-for-age values, reference equations must be extrapolated beyond the data that generated them, a practice discouraged by the American Thoracic Society because it substantially decreases accuracy (9, 10). In addition, patients with CF frequently have delayed puberty, altering the relationship between height and lung function and rendering comparison with normal control subjects particularly problematic during adolescence (12–14).
A number of reference equations are widely employed (15–17), without consensus as to the most appropriate equation. The choice of reference equation applied to patients with CF has been shown to have important effects on the value of percent-predicted FEV1, the apparent change in lung function over time, and the classification of lung function as normal or abnormal; this is due, in part, to the poor comparability of height-for-age between patients with CF and normal subjects (14, 18). The reference equations (15) most widely employed by the CF research community, and currently used in the Cystic Fibrosis Foundation National Patient Registry, have relatively low predictive accuracy in short and young children due to the modeling of FEV1 as a linear function of height, resulting in underestimation of the true normal predicted values in this age range (19).
Despite its widespread use, the percent-predicted method of comparing an individual's lung function to that of a reference population has drawbacks. In expressing lung function as a percent of a predicted value, two assumptions are made, both of which have been shown to be invalid (20, 21). First, it is assumed that a given percent-predicted value is comparable in terms of the degree of abnormality (i.e., deviation from the mean) across multiple lung function indices (e.g., FEV1, FVC). Second, it is assumed that, for each lung function index, a given percent-predicted value indicates the same degree of abnormality for persons of different sex, ages, and heights. The use of percentiles or Z scores, such as are employed in National Center for Health Statistics growth curves for height and weight (22), avoids both these limitations. The American Thoracic Society has recommended that normal ranges for lung function be calculated based on percentiles (9, 10).
We propose an alternative paradigm for evaluating the lung function of patients with CF: CF-specific reference equations, by which the FEV1 of an individual with CF is compared not to that of a healthy reference population but rather to that of a CF population. These equations describe the distribution of FEV1 among patients with CF in terms of percentiles of FEV1 for a given sex, age, and height. In both the clinical and research arenas, CF-specific reference equations may serve as a useful adjunct to conventional reference equations derived from a healthy control population. The primary aim of this study was to develop percentiles for FEV1 among patients with CF, and to provide a simple tool for the calculation of CF-specific FEV1 percentiles. The secondary aim was to demonstrate the utility of CF-specific FEV1 percentiles in clinical practice and research. Part of this work has previously been presented in abstract form (23).
The Cystic Fibrosis Foundation has maintained a registry of U.S. patients with CF since 1966 (11). From 1994 to 2001, the registry included over 28,000 subjects. For each quarter, CF care centers were instructed to record the best prebronchodilator raw pulmonary function tests available, reported in liters. Although not specified, it was presumed that sites would record the best FEV1 per quarter. We extracted quarterly data on patients with CF who were older than 6 years of age in 2001, had at least one quarterly measurement of pulmonary function between 1994 and 2001, and had a recorded height between 105 and 190 cm for males and between 105 and 180 cm for females. To improve the quality of height data, we compared successive height measurements for each patient, searching for major inconsistencies in recorded heights, such as implausible, sudden positive changes or large negative changes. Patients who had inconsistencies in height measurements that were likely to result in errors exceeding 5 cm were excluded from all analyses. The details on height data quality checks are provided in the online supplement. Patients of all races were included in the study. Percent-predicted FEV1 values were calculated using the reference equations of Knudson and colleagues (15). The study was approved by the Children's Hospital and Regional Medical Center Human Subjects Review Board. Permission to use CF registry data was obtained from the Cystic Fibrosis Foundation.
All analyses were done separately by sex. Quantile regression methods (24) were used to estimate percentiles for FEV1 given height and/or age. This method models a given percentile of the outcome variable as a linear function of independent variables, allowing fitting and smoothing of the percentiles in one step without assuming normality of the outcome. Cubic B-spline bases for height and age were used as independent variables in the quantile regression model. The bases are piecewise cubic functions; the model combines them linearly to obtain best fitting smooth curves. We fitted 99 separate models for the 1st–99th percentiles and interpolated linearly between them. No adjustments were made for correlations between multiple FEV1 measurements on the same subject. This lack of adjustment for repeated measures does not affect the estimated percentiles. Data manipulations were performed using SAS (SAS Institute, Cary, NC) and the analyses were performed in R (R Foundation for Statistical Computing, Vienna, Austria), an open-source statistical package (25). Details are provided in the online supplement.
Reference curves for FEV1 were calculated from 287,108 FEV1 measurements made among 21,646 subjects (53.1% male) between 1994 and 2001. The average number of FEV1 measurements per subject was 13.3 (range, 1–47). Approximately 0.7% of all measurements (∼ 2,000) were excluded due to inconsistencies in height. The size of the study population decreased with increasing age (Table 1)
Male Patients | Female Patients | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| n* | Mean FEV1 (L) | Mean Height (cm) | n* | Mean FEV1 (L) | Mean Height (cm) | |||||
| Age, yr | ||||||||||
| 6–10 | 5,057 | 1.39 | 126.9 | 4,667 | 1.28 | 126.2 | ||||
| 11–15 | 4,826 | 2.14 | 152.8 | 4,364 | 1.91 | 151.3 | ||||
| 16–20 | 3,820 | 2.71 | 170.8 | 3,395 | 2.05 | 159.8 | ||||
| 21–25 | 2,579 | 2.33 | 172.5 | 2,242 | 1.86 | 160.8 | ||||
| 26–35 | 2,421 | 2.06 | 173.0 | 2,059 | 1.67 | 161.1 | ||||
| > 35 | 1,560 | 1.91 | 174.1 | 1,173 | 1.48 | 161.6 | ||||
| All ages | 11,485 | 2.04 | 155.4 | 10,161 | 1.71 | 149.2 | ||||
Figure 1. Selected cystic fibrosis (CF)–specific FEV1 percentile reference curves by age in years: (A) male patients with CF; (B) female patients with CF.
[More] [Minimize]CF-specific FEV1-for-height percentiles for male and female patients with CF are shown in Figure 2
Figure 2. Selected CF-specific FEV1 percentile reference curves by height: (A) male patients with CF; (B) female patients with CF.
[More] [Minimize]
Figure 3. Examples of CF-specific FEV1 percentile reference curves for specific ages for male patients with CF ([A] 12 years and [B] 18 years) and female patients with CF ([C] 12 years and [D] 18 years).
[More] [Minimize]CF-specific FEV1 percentiles may be especially useful in monitoring changes in lung function in an individual patient over time. Note that the curves are by necessity based on data from subjects alive at each age. Thus, the change in individuals' CF-specific percentiles with advancing age reflects their status relative to others surviving to each of these ages, not to patients who may have had steeper declines in lung function and then died. Figure 4
Figure 4. Trends in pulmonary function of four selected patients. (A and B): females; (C and D): males. Solid lines represent percent-predicted FEV1 compared with a healthy reference population, based on the reference equations of Knudson and coworkers (15). Dashed lines indicate CF-specific FEV1 percentiles; dotted lines indicate fitted linear trends.
[More] [Minimize]Although the linear modeling of the Knudson reference equations (15) is known to produce inaccuracies in the slopes of percent-predicted FEV1 as children grow in height (19), these inaccuracies do not explain the observed discrepancies between the CF-specific percentiles and percent-predicted FEV1. Use of the Lebecque equations (26) (which employ a log-linear model of the association between FEV1 and height) to generate percent-predicted FEV1 resulted in plots for these four subjects with very similar trends (data not shown).
Because lung function generally declines with increasing age in patients with CF, the frequency of a particular value of percent-predicted FEV1 among patients with CF varies with age. CF-specific FEV1 percentiles can be applied to calculate frequencies of percent-predicted FEV1 among patients with CF. Figure 5
Figure 5. Percentages of male (A) and female (B) patients with CF whose FEV1 exceeds a given percent-predicted FEV1 (based on Knudson equations [15]) at four selected ages. The intersections of the horizontal and vertical grid lines at each age-specific curve identify the proportion of patients of that age whose FEV1 % predicted exceeds a given threshold. For example, for males, the 50th percentile (median) horizontal grid line intersects the 6-year-old curve at 97% predicted and the 12-year-old curve at 90% predicted. Thus, the median FEV1 among 6-year-old males is 97% predicted and among 12-year-old males is 90% predicted. Similarly, for males, the 80% predicted FEV1 vertical grid line intersects with the 6-year-old age curve at the 76th percentile and with the 12-year-old age curve at the 67th percentile. Thus, 76% of 6-year-old males with CF and 67% of 12-year-old males with CF have an FEV1 of at least 80% predicted.
[More] [Minimize]In Figure 5, the shift of the curves to the left with increasing age reflects the decline of pulmonary function in patients with CF compared with that in healthy subjects. The figure reveals a large difference in the rate of percent-predicted FEV1 decline between male and female patients with CF. Female patients have a large drop in percent-predicted FEV1 between ages 6 and 12 years, followed by much smaller drops between ages 12, 18, and 24 years. In male patients, the decline between ages 6 and 12 years is much smaller, but is followed by much larger drops between ages 12, 18, and 24 years. The large shift in the male curve between ages 18 and 24 years, and the lower percentages of males than females above any threshold, at least partly reflect the larger number of females who have died before age 24. Taken together, these curves suggest that males are more likely to survive for extended periods when they reach a critically low level of FEV1.
Most CF clinical trials include eligibility criteria based on FEV1, almost always defined in terms of the range of percent-predicted FEV1 (e.g., 30–80% predicted). Because the proportion of patients that falls within a given range of percent-predicted FEV1 varies markedly by sex, age, and height, entry criteria based on percent-predicted FEV1 can result in highly variable representation of groups of patients with CF depending on their sex, age, and height. For example, in a trial that enrolls subjects between 6 and 12 years of age, an entry criterion of an FEV1 30 to 80% predicted will result in an unequal representation of patients in different sex, age, and height categories (Table 2)
Height Percentiles | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Boys | Girls | ||||||||||
| 10th | 50th | 90th | 10th | 50th | 90th | ||||||
| Age, yr | |||||||||||
| 6 | 15.4 | 23.8 | 32.6 | 9.8 | 22.9 | 24.2 | |||||
| 8 | 39.4 | 32.4 | 24.9 | 36.7 | 31.4 | 26.6 | |||||
| 10 | 40.9 | 36.3 | 30.1 | 45.7 | 37.2 | 31.9 | |||||
| 12 | 34.1 | 31.7 | 32.3 | 54.9 | 45.8 | 30.8 | |||||
We have developed CF-specific reference equations for FEV1 that, for the first time, allow comparison of lung function of an individual patient with CF with that of his or her peers with CF. We hope that, in both the clinical and research arenas, these novel CF-specific reference equations may serve as a useful adjunct to conventional reference equations. CF-specific reference equations: (1) allow an individual patient's lung function to be placed in the context of the average lung function for patients with CF of a given sex, age, and height; (2) overcome the limitations of comparing patients with CF to a reference population that has a very different distribution of height-for-age; and (3) describe lung function in terms of percentiles, as recommended by the American Thoracic Society (9, 10), rather than percent-predicted values, as is typical for established reference equations (15–17). In addition, the use of CF-specific reference equations avoids the inconsistencies introduced by switching from pediatric to adult reference equations as patients mature.
CF-specific reference equations account for the fact that the lung function of patients with CF generally declines with age. Thus, an FEV1 at the 50th CF-specific percentile always means that half of patients with CF of the same age, sex, and height have an FEV1 greater than this individual's and half have an FEV1 that is less. On the other hand, an FEV1 of 80% predicted corresponds to the average FEV1 for a 12-year-old female with CF, but only 30% of 18-year-old females with CF have an FEV1 of at least 80% predicted (Figure 5). Clearly, the ultimate goal for our patients is to maintain normal lung function throughout their lives. Currently, however, patients with CF do generally lose lung function with advancing age. The objective of CF-specific reference equations is not to allow clinicians and patients to settle for less than normal health, but rather to provide an important additional context in which to better understand patients' lung function. CF-specific reference equations may be employed in clinical practice to aid in early identification of atypical declines in FEV1 that may warrant more aggressive evaluation or intervention, or alternatively to reassure an individual that his or her lung function is average or above-average for patients with CF of his/her age.
In the clinical trial arena, we hope that the use of CF-specific reference equations may improve the generalized applicability of results by ensuring a more equitable distribution of study subjects across ages and sex. Entry criteria that include a range of percent-predicted FEV1 values relative to a reference population can result in an uneven distribution of study subjects, with one sex or certain age ranges likely to be over- or underrepresented. If, instead, the entry criteria are based on CF-specific percentiles of FEV1, an equitable sex and age distribution is much more likely to be achieved. In addition, because our FEV1 percentiles were developed specifically for the population with CF, their use as outcome measures in both clinical trials and observational studies may improve the reliability of sample size calculations and power to detect significant effects.
Fortunately, therapeutic advances result in ongoing improvements in lung function and survival (11) among patients with CF. Our current analysis provides a snapshot of lung function in U.S. patients with CF between 1994 and 2001. These CF-specific reference equations will need to be updated regularly (perhaps every 5–10 years) using newly available data. A change in the CF-specific FEV1 percentiles over time will help to quantify improvements in pulmonary function and survival achieved by new therapies. Ultimately, as therapies continue to improve, CF-specific curves may no longer be necessary.
Lung function data, of course, can only be obtained from subjects surviving to a given age. Thus, particularly at the greatest ages, the population from which the CF-specific percentile data are derived is relatively enriched for those with “mild” mutations and other risk factors associated with above-average survival. This survivor bias, inherent in any comparison of a patient with CF to his or her living peers, should be kept in mind when utilizing the CF-specific reference curves. Although mortality attrition also leads to a declining number of subjects available for the analyses with increasing age (Table 1), the sample sizes in the current analysis (n = 21,000) nevertheless compare favorably with those of widely employed standard reference equations (n = 7,429) (27).
We derived CF-specific reference equations using data from patients of all races and ethnicities. In the 2002 registry, 96% of patients were white, 3% were African American, and 4% were Hispanic (black or white). Because CF is rare in non-white populations, the data did not allow development of race-specific FEV1 percentiles. Thus caution should be used in applying our reference equations to non-white patients with CF. Similarly, as for any reference equations, caution should be exercised when interpreting percentiles for patients at the extremes of age (< 6 years or > 35 years), height (very short or very tall for age among patients with CF), or FEV1 (the lowest and the highest percentiles are estimated with more error).
To our knowledge, we have developed the first disease-specific lung function reference equations. Potentially, such reference equations could also be useful for other chronic pulmonary diseases, such as Duchenne muscular dystrophy, in which lung function relative to a healthy control population is anticipated to slowly decline with advancing age.
The authors thank Preston Campbell, III, M.D., Executive Vice President of Medical Affairs, and Bruce Marshall, M.D., Director of Clinical Affairs, Cystic Fibrosis Foundation (CFF), for their support of this project and for making the CFF National Patient Registry data available to us.
| 1. | Ramsey BW, Boat TF. Outcome measures for clinical trials in CF: summary of a cystic fibrosis conference. J Pediatr 1994;124:177–192. |
| 2. | Davis PB, Byard PJ, Konstan MW. Identifying treatments that halt progression of pulmonary disease in cystic fibrosis. Pediatr Res 1997;41:161–165. |
| 3. | Kerem E, Reisman J, Corey M, Canny GJ, Levison H. Prediction of mortality in patients with cystic fibrosis. N Engl J Med 1992;326:1187–1191. |
| 4. | American Society for Transplant Physicians, American Thoracic Society, European Respiratory Society, International Society for Heart and Lung Transplantation: international guidelines for the selection of lung transplant candidates. Am J Respir Crit Care Med 1998;158:335–339. |
| 5. | Ramsey BW, Astley SJ, Aitken ML. Efficacy and safety of short term administration of aerosolized recombinant human deoxyribonuclease in patients with cystic fibrosis. Am Rev Respir Dis 1993;148:145–151. |
| 6. | Konstan MW, Byard PJ, Hoppel CL, Davis PB. Effect of high-dose ibuprofen in patients with cystic fibrosis. N Engl J Med 1995;332:848–854. |
| 7. | Ramsey BW, Pepe MS, Quan JM, Otto KL, Montgomery AB, Williams-Warren J, Vasiljev-K M, Borowitz D, Bowman CM, Marshall BC, Marshall S, Smith AL. Efficacy and safety of chronic intermittent administration of inhaled tobramycin in patients with cystic fibrosis. Cystic Fibrosis Inhaled Tobramycin Study Group. N Engl J Med 1999;340:23–30. |
| 8. | Saiman L, Marshall BC, Mayer-Hamblett N, Burns JL, Quittner AL, Cibene DA, Coquillette S, Fieberg AY, Accurso FJ, Campbell PW III, Macrolide Study Group. Azithromycin in patients with cystic fibrosis chronically infected with Pseudomonas aeruginosa: a randomized controlled trial. JAMA 2003;290:1749–1756. |
| 9. | American Thoracic Society. Lung function testing: selection of reference values and interpretive strategies. Am Rev Respir Dis 1991;144:1202–1218. |
| 10. | American Thoracic Society. Standardization of spirometry, 1994 update. Am J Respir Crit Care Med 1995;152:1107–1136. |
| 11. | Cystic Fibrosis Foundation. Patient registry 2002 annual data report. Bethesda, Maryland: Cystic Fibrosis Foundation; 2003. |
| 12. | Lebowitz MD, Sherrill DL. The assessment of interpretation of spirometry during the transition from childhood to adulthood. Pediatr Pulmonol 1995;19:143–149. |
| 13. | Borsboom GJJM, Van Pelt W, Quanjer PH. Pubertal growth curves of ventilatory function: relationship with childhood respiratory symptoms. Am Rev Respir Dis 1993;147:372–378. |
| 14. | Merkus PJFM, Tiddens HAWM, de Jongste JC. Annual lung function changes in young patients with chronic lung disease. Eur Respir J 2002;19:886–891. |
| 15. | Knudson RJ, Lebowitz MD, Holberg CJ, Burrows B. Changes in the normal maximal expiratory flow–volume curve with growth and aging. Am Rev Respir Dis 1983;127:725–34. |
| 16. | Wang X, Dockery DW, Wypij D, Fay ME, Ferris BG. Pulmonary function between 6 and 18 years of age. Pediatr Pulmonol 1993;15:75–88. |
| 17. | Polgar G, Promadhat V. Pulmonary function testing in children: techniques and standards. Philadelphia: W.B. Saunders; 1971. pp. 170–180. |
| 18. | Rosenfeld M, Pepe MS, Emerson J, Longton G, FitzSimmons S. Effect of different reference equations on the analysis of pulmonary function data in cystic fibrosis. Pediatr Pulmonol 2001;31:227–237. |
| 19. | Subbarao P, Lebecque P, Corey M, Coates AL. Comparison of spirometric reference values. Pediatr Pulmonol 2004;36:515–522. |
| 20. | Glindmeyer HW. Predictable confusion. J Occup Med 1981;23:845. |
| 21. | Miller M, Pincock AC. Predicted values: how should we use them? Thorax 1988;43:265–267. |
| 22. | Kuczmarski RJ, Ogden CL, Grummer-Strawn LM, Flegel KM, Guo SS, Wei R, Mei Z, Curtin LR, Roche AF, Johnson CL. CDC growth charts: United States. Adv Data 2000;314:1–27. |
| 23. | Campbell JD, Kulich M, Rosenfeld M, Kronmal R. Pulmonary function reference curves for cystic fibrosis patients. Pediatr Pulmonol 2002;S34:332. |
| 24. | Koenker RW, Bassett GW. Regression quantiles. Econometrica 1978;46:33–50. |
| 25. | R Development Core Team. R: a language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing; 2003. |
| 26. | Lebecque P, Desmond K, Swartebroeckx Y, Dubois P, Lulling J, Coates A. Measurement of respiratory system resistance by forced oscillation in normal children: a comparison with spirometric values. Pediatr Pulmonol 1991;10:117–122. |
| 27. | Hankinson JL, Odencrantz JR, Fedan KB. Spirometric reference values from a sample of the general U.S. population. Am J Respir Crit Care Med 1999;159:179–187. |
