Normal Spirometric Reference Equations – When the Best Fit May Not be the Best Solution

1. Normal Spirometric Reference Equations – When the Best Fit May Not be the Best Solution Allan Coates, B Eng (Elect) MDCM University of Toronto Hospital for Sick Children, Toronto 2011 Canadian Respiratory Conference

2. The “Holy Grail” of Reference Equations • Representative of the population of interest • One equation for all ages for each sex • Simple to program into the spirometers • Sufficient numbers to give confidence to the lower limit of normal (LLN)

3. Definitions of “Normal” Values • American Association of Clinical Chemistry • Based on “healthy” individuals • Plus/minus 2 standard deviations or 95% of the populations • Clearly the variability of a value in the general population whether or not associated with a “disease” will impact the range of values within 2 SD • How does this fit with our spirometry reference values?

4. Health vs Disease • If 1000 perfectly healthy individuals had spirometry preformed, 2.5% would be below 2 SD and 2.5% above • By definition, none would have disease • Hence any clinical decision based on spirometric values would depend on pre test probability

5. Pre Test Probability • Definition • Pretest Probability is defined as the probability of the target disorder before a diagnostic test result is known • In respiratory medicine, only extreme deviations from the reference values are pathognomonic for disease • Hence pretest probability is an essential part of diagnosis

6. Who is Healthy? • NHANES III rejection criteria • Smoking (cigarettes, cigars, pipe) • MD dx of asthma, chronic bronchitis, emphysema • Whistling or wheezing in chest (last 12 months and apart from colds) • Persistent cough for phlegm • Moderate shortness of breath • Of the 15,000 plus acceptable spirometry tracings where did this leave us? Hankinson et al Am J Resp Crit Care Med 1999

7. 15,503 Acceptable Adult Tests • Smokers 7115 Remaining • MD Dx asthma, COPD 6465 Remaining • Whistling or wheezing in chest 5934 Remaining • Persistent cough and/or phlegm 5651 Remaining • Moderate shortness of breath 4803 Remaining • Over 80 (too few observations) 4634 Remaining • In adults, the rejection rate was > 2/3 Hankinson et al Am J Resp Crit Care Med 1999

8. What about Children? • There were 3917 good test in 8-16 year olds • Rejection criteria • Smoking 3580 Remaining • Asthma, chronic bronchitis 3170 Remaining • Wheezing, cough, phlegm 2796 Remaining • In pediatric sample, the rejection rate was > 1/4

9. Lower Limit of Normal - Definition This is a plot of the FEV1 measured from a group of normal, non-smoking men who were all 60 years old and 180 cm tall. 200 - 150 - 100 - 50- - 0- 200 - 150 - 100 - 50- - 0- FEV1 values less than LLN are considered to be below normal The predicted value for FEV1 for someone in this group is 3.5L. The shaded area represents 5% of normal men, age 60, height 180 cm, with the lowest FEV1. This defines the Lower Limit of Normal (LLN). Number of Subjects 95 % 5% Number of Subjects LLN for FEV1 for this group is 2.6L 5% of the population with normal lungs have FEV1 below LLN 1 2 3 4 5 6 95% of the population with normal lungs have FEV1 above LLN 1 2 3 4 5 6 FEV1 in Liters Lower Limit of Normal FEV1 in Liters Predicted Value Ref: MR Miller – www.millermr.com

10. Controversies over LLN • Most of us were trained on percent predicted and the concept that FEV1 and FVC ≥ 80% was normal • In other words, we had our own concept of LLN • In fact, for NHANES III, for FVC, LLN is 84% predicted for a tall young male and 75% for a short elderly female • All of us use ± 2 SD for electrolytes with normal (95% of healthy) being inside 2 SD • We have better PFT data – Why not use it?

11. LLN for the FEV1/FVC ratio NHANES III Hankinson, 1999 While the ratio clearly decreases with age,these data showed that the variance was not affected by age or height. ie, homoscedastic. Thanks to Bruce Culver

12. Concept of Homoscedasticity • For any given value of x (eg height) the standard deviation of y (eg FEV1) is the same • The standard deviation depends on both variability and n • Reference values from small samples may not meet this requirement

13. NHANES III Approach • Using a polynomial analysis for height and age, attempted to have one equation for FEV1 and FVC • Had to settle for separate equations that joined at 18 for females and 20 for males • Also included values for FEV1/FVC, PEF, FEV6 and FEF25-75 and LLN for all parameters • Reference values for Caucasian, Mexican Americans and African Americans between 8 and 80 years

14. Problems with NHANES III • Numbers small at either extremes of the ages giving rise to inhomoscedasticity • Extrapolation to ages less than 8 gave rise to significant over estimation in males • While the curves met at the 18 (♀s) and 20 years (♂s), the curves were discontinuous DESPITE THESE CONCERNS, IT WAS WIDELY ACCEPTED AND EASILY PROGRAMMED INTO SPIROMETRIC SOFTWARE

15. Solutions • The values from pediatric series down to age 5 (Corey et al, Lebeques et al and Rosenthal et al were found to over lap where ages overlapped with NHANES and added to the series • New data analysis by the LMS method • Resulting curves were “continuous”

16. LMS: lambda, mu, sigma Method . The distribution of the normal population at each point along the continuum is described by: mu the median sigma the coefficient of variance lambda an index of skewness. The result is a series of equations linked by “splines” with coefficients from a set of look up tables, read by computer. The method creates a smooth continuous predicted value (given by the median, mu ) Stanojevic et al Am J Resp Care Med 2008

17. The sigma and lambda terms allow for the 5th percentile LLN to be independently determined throughout the age-height spectrum FEV1/FVC ratio F M LLN Stanojevic2008

18. Stanojevic compared to NHANES III

19. Stanojevic vs NHANES • Mores sophisticated statistical approach (Coles et al 2008) with somewhat better “accuracy” overall • Solved the problem of age limitation of NHANES • Smoothed the 18 and 20 year transition points • NHANES uses simple polynomial equation, easy to program into a computer or hand calculator • The complex mathematical approach of Stanojevic has not been adapted (to date) in any commercial spirometric software

20. Reference Sources - Spirometry NHANES III v Knudson, Crapo, Glindmeyer

21. Does One Set Over Another Really Make a Difference? • The difference between NNANES, Stanojevic and older series in adults is too small to result is serious clinical errors • This is not the case in children

22. Differences Depending on Equations Hankinson breaks down when out of range Knudson equations just do not apply to young Subbarao et al Pediatr Pulmonol 2004

23. ERS Task Force – Global Lungs Initiative Project to collate available international lung function data to develop new reference equations. Unlike the 1983-93 ECSC compilation which merged equations, the current effort has collected raw dataand is using the LMS method to analyze it. Data from 150,000 individuals from 71 countries. Co-chairs: Janet Stocks – UK, Xvar Baur – Germany Graham Hall – ANZRS, Bruce Culver – ATS Steering Comm includes: Phil Quanjer, Sonja Stanojevik, John Hankinson, Paul Enright.

24. ERS GlobalLungs Initiative

25. Problems and Challenges • NHANES III is from one data set gathered on the same equipment under the same conditions • The Stanojevic data is a composite of 4 sets from different countries and different equipment • The ERS Task Force will have the same problems with multi site challenges • The challenge is enough numbers to have confidence in the LLN but have identical methodology and homogeneous sample

26. What to Do? • NHANES III is the largest data set to date and while the polynomial approach may not be as scientific as the LMU approach, few if any clinical errors would occur for patients ≥ 8 years • The Stanojevic analysis is the best available and while cumbersome, can be used for ≥ 5 years • New Canadian data is being analyzed and should be available in the next 18 months

27. Conclusions • We do not have a perfect data set yet so reference equations are less than absolute ESPECIALLY FOR NON CAUCASIANS • We have much more confidence and better data on the LLN • There will always be a certain inaccuracy in the application of the results of any pulmonary function test, especially near the LLN