Introduction

Lung function measurements are essential for predicting outcomes of interstitial lung diseases (ILDs). Results from spirometry measurements (e.g., forced vital capacity, FVC) and gas transfer tests (e.g., lung transfer factor for carbon monoxide, TLCO) are used as components of well-known and frequently used scales, such as the ‘GAP Index’ or ‘du Bois Score’ in idiopathic pulmonary fibrosis [1,2]. However, it is the TLC (total lung capacity) value that determines the presence of restrictive disorders, and FEV₁ (forced expiratory volume in one second) is considered a universal indicator of the severity of disorders, regardless of whether they are of obstructive or restrictive origin. For many years following the ATS (American Thoracic Society) and ERS (European Respiratory Society) 2005 recommendations, a z-score threshold of −1.64 (level of the 5th percentile) was an established criterion for classifying results as abnormal, and this approach was well accepted within the respiratory community [3]. It functioned in parallel with the classification of lung function impairment severity based on % predicted values, where different % predicted thresholds were applied to ventilatory disorders (e.g., 50% for severe/non-severe) and different ones to TLCO (40% pred for severe/non-severe).

The recently published recommendations for routine lung function interpretative strategy suggest abandoning % predicted and use of z-scores as more appropriate also for severity assessment [4]. One of the arguments is to better match the risk of death [5,6]. However, the ATS/ERS experts in the rationale refer either to chronic obstructive pulmonary disease (COPD) [5], or to a cohort of several thousand patients, in which a clearly defined diagnosis was available for only 38% of cases, with interstitial lung diseases (ILD) accounting for just 9% [6].

In contrast to the referenced study, where more than half of the patients had no confirmed diagnosis and received no treatment, a key strength of our study is that in the analyzed large cohort (n = 6,808), all patients had a confirmed diagnosis belonging to the group of interstitial lung diseases. This enabled us to test our hypothesis in a well-defined population of patients with established diagnoses, representing a group of relatively rare diseases, but associated with a serious prognosis (excluding sarcoidosis), where the role of pulmonary function parameters has always been considered. Previously published data indicated, that changes in the cut-off points may lead to discordance in classification in some patients [7]. From a practical and clinical point of view, it seems important to answer the question of whether the change in the method of evaluating test results and thus reclassifying the severity in some patients will actually result in a better match to their prognosis.

The main objective of the analysis was to examine the relationship between the presence and severity of various lung function disorders (ventilation and gas exchange) with consideration of two different assessment methods (ATS/ERS 2005 and ATS/ERS 2022) in the context of prognosis in a large group of patients with ILD. We hypothesized that there were significant differences in prognosis in patients with inconsistent classification compared to their original severity grade.

Materials and methods

Lung function and outcome assessment

Only baseline (at first presentation) and pre-bronchodilator data was included. TLCO results were corrected for hemoglobin. Absolute FEV₁, FVC, TLC and TLCO values from these patients were transformed, and we used reference values from the GLI project for all measurements (calculation made in May 2023, after correction for TLCO) [13–16], defining the lower limit of normal (LLN) at predicted −1.645 SD, and grading severity according to the ATS 1991, ATS/ERS 2005 and 2022 PFT interpretation guidelines [3,4,17].

For clarity and simplification, the ATS/ERS 2005 category of “moderately severe” was merged into the “moderate” category (ranging from 50% to 70% of predicted for FEV₁ and FVC), and the category of “very severe” was merged into the “severe” category (<50% predicted for FEV₁ and FVC). The category of “moderately severe” for TLC (<60% of pred.) was assigned to the “severe” category – see Table 1. Mortality data were obtained from the Ministry of Digital Affairs. Date of censoring was the March 15th 2023 and time to event was time from the first accessible pulmonary function test to censoring or death.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Categorization of lung function impairment severity recommended by ATS/ERS 2022 [4] and previously in 2005 [3] and ATS 1991 [17] guidelines.

https://doi.org/10.1371/journal.pmed.1004619.t001

In ILDs, particularly in progressive fibrotic diseases with a significantly poorer prognosis, unlike in obstructive diseases, greater emphasis is placed on the FVC and TLCO indices [18], rather than on FEV₁, as is the case with asthma or COPD. Therefore, this analysis primarily focuses on these two indices. However, an examination of TLC and FEV₁, as recommended for the assessment of restrictive disorders and the extent of ventilatory disorders in general, was also conducted.

Results

Baseline characteristics of the patients

During the 14-year study period, over twenty-two thousand patients were investigated in our PFT lab. From these, we identified 6,808 patients (in whom 3281–48.2% were females) with diagnosed ILD. All of them were Caucasians. For the purposes of this study, we grouped patients according to the latest available information on their diagnoses into the following categories: sarcoidosis (SAR), idiopathic pulmonary fibrosis (IPF), unclassifiable interstitial lung disease (u-ILD), idiopathic non-specific interstitial pneumonia (i-NSIP), hypersensitivity pneumonitis (HP), connective tissue diseases pulmonary related disorders (CTD) and other ILDs (o-ILDs). See Table 2 for detailed patient characteristics with the breakdown of the data for each diagnosis group.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Study group characteristics with the breakdown of the data for each diagnosis group.

All were Caucasian. Mean ± standard deviation is given for continuous variables. Airway obstruction: FEV₁/FVC < LLN, volume restriction: TLC < LLN, mixed: FEV₁/FVC < LLN and TLC < LLN, non-specific: FEV₁/FVC > LLN, TLC > LLN, FEV₁ < LLN, FVC < LLN.

https://doi.org/10.1371/journal.pmed.1004619.t002

Any pulmonary function disorders were identified in 54.6% of patients (see S1 Table for details for overlapping ventilatory and gas transfer abnormalities). In most patients, abnormalities in any lung function test results were observed in 82% to 99%, depending on the group, except for patients with sarcoidosis, where the percentage of any abnormalities did not exceed 33%.

Severity classification % pred. versus z-scores

Adopting the 2022 PFT interpretation guidelines (using z-score severity thresholds instead of % predicted) resulted in reclassifying the severity of 26.8% of patients with a low FEV₁, 24.6% of patients with a low FVC, 15.1% of patients with low TLC and 28.1% of those with a low TLCO.

Recategorization of spirometric indices (FEV₁ and FVC) led to a shift to a less severe category for most recategorized patients. Only 6 patients with low FEV₁ and 20 patients with low FVC were moved to a more severe category. Using z-score thresholds, 384 out of 980 patients with moderately low FEV₁ were reclassified as mild, and 118 out of 283 patients with severe FEV₁ were assigned as moderate. Similarly, 331 out of 876 patients with moderately low FVC were reclassified as mild, and 50 out of 173 patients with severe FVC were assigned as moderate (see Fig 1, upper panels).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Plots of FEV₁ (upper left), FVC (upper right), TLC (lower left) and TLCO (lower right) values expressed in % pred. (y-axis) against z-scores (x-axis) with subjects stratified by age group.

Arrows indicate areas of discordance between the 2005, 1991 and 2022 severity categories. Horizontal dotted lines are at 2005 or 1991 severity thresholds. Dashed vertical lines are at 2022 severity thresholds. Note that results with a z-score above −1.645 are considered normal. The number of patients in each category is provided. For example, 5,238 patients had a normal FVC. % pred. – % of predicted value; ATS – American Thoracic Society; ERS – European Respiratory Society; FEV₁ – forced expiratory volume in 1 second; FVC – forced vital capacity; TLC – total lung capacity from body plethysmography; TLCO – lung transfer factor for carbon monoxide; z-score – the number of standard deviations from the mean value of the healthy GLI reference population.

https://doi.org/10.1371/journal.pmed.1004619.g001

In terms of TLC recategorization, 28 out of 356 cases were shifted from moderate to mild, and 115 out of 217 cases were shifted from severe to moderate. However, 46 out of 688 cases were reclassified from mild to moderate. Among the 489 patients currently classified as moderate, 15 were previously classified as mild, and 115 as severe.

On the other hand, the use of z-score thresholds for TLCO often resulted in patients with low TLCO being shifted to a more severe category. Specifically, 433 out of 1,445 patients with a mildly low TLCO were reclassified as moderate, and 439 out of 1,130 patients previously classified as moderate were now regarded as severe (see Fig 1, lower right panel).

Older patients were more likely to be reclassified as less severe in spirometry (low FEV₁ and FVC) when the 2022 interpretation guidelines were applied. Conversely, there was a higher occurrence of recategorization of TLCO results in women, and this was associated with a younger age group. Male patients demonstrated a lower likelihood of experiencing changes in interpretation severity compared to female patients, except for TLC recategorization, where younger females were more frequently reclassified to a more severe category, while older males were more commonly recategorized to a less severe category. Groups with a small number of reclassified cases (≤20) were excluded from the analysis. Specific changes in interpretation were found to be more prevalent based on the diagnostic category, as presented in Table 3.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Characteristics of patients with abnormal lung function whose severity classification changed when the z-score guidelines from 2022 were applied in comparison to those who were not reclassified and stayed in 2005 category.

Data for groups n > 30 are presented. The statistically significant differences are shaded.

https://doi.org/10.1371/journal.pmed.1004619.t003

Survival analysis

According to data from the Ministry of Digital Affairs, the national register of population, births and deaths in Poland, 1,525 (22.4%) of our investigated patients died during the follow-up period. Patients with sarcoidosis had the best prognosis, and patients with IPF had the highest mortality rate (S1 Fig and S2–S4 Tables provide a detailed breakdown of the data).

To compare the impact on prognosis between different severity levels of pulmonary function disorders, a univariate (Kaplan–Meier) model was used to examine groups of patients with concordant and discordant severity classification for predicting all-cause mortality. The results are presented in Fig 2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Kaplan–Meier analyses for survival according to 2005 and 1991 vs. 2022 guidelines for FEV₁ (upper left), FVC (upper right), TLC (lower left) and TLCO (lower right) severity.

Fourteen years of follow-up = 5,200 days. FEV₁ – forced expiratory volume in 1 second; FVC – forced vital capacity; TLC – total lung capacity from body plethysmography; TLCO – lung transfer factor for carbon monoxide. Data for groups n > 30 are presented.

https://doi.org/10.1371/journal.pmed.1004619.g002

In relation to ventilatory indices (FEV₁, FVC, and TLC), an interesting observation draws attention: reclassified patients exhibited a poorer prognosis compared to those whose category remained unchanged, despite the reclassification being in the direction of milder severity. For instance, patients with a reclassified FVC from severe to moderate and from moderate to mild demonstrated a worse prognosis than those with consistently severe impairment according to both classification systems. There was a considerable overlap in prognosis among patients with moderate and severe ventilatory disorders. Regarding TLCO, the greatest separation in prognosis was observed within the groups. It was also observed that patients with a reclassified TLCO prognosis were similar to those whose baseline category did not change. The hazard ratios with 95% confidence intervals, using the group with indices in normal range as a reference, for concordant and discordant severity classes can be found in Table 4.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Kaplan–Meier estimation in groups of patients recategorized according to 2022 ATS/ERS guidelines with comparison to those with concordant classification for FEV₁, FVC, TLC and TLCO (first part of name is grade according to 2005 or 1991 classification based on % pred., the second is from ATS/ERS 2022 based on z-scores, results for recategorized patients are shaded).

Cox proportional hazards regression was used to calculate HRs, to quantify the effect of covariates on the hazard rate.

https://doi.org/10.1371/journal.pmed.1004619.t004

To examine how specified risk factors affect all-cause mortality, we used the Cox proportional hazards models which included sex, age, body mass index (BMI), the diagnosis group (sarcoidosis as the reference) and lung function. Table 5 presents models with a threshold approach to the severity of the lung function disorders for both classification methods for FVC and TLCO.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Hazard ratios (HR) for all-cause mortality with 95% confidence intervals from Cox regression models using the 2005 vs. the 2022 guidelines for grading severity, normal FVC (n = 5,238) and normal TLCO (n = 3,704) as reference for lung function severity, sarcoidosis as reference for other diagnosis groups.

Variables which were statistically significant predictors for that model are shaded.

https://doi.org/10.1371/journal.pmed.1004619.t005

In different models we used ventilatory indices and gas transfer expressed in % pred., z-scores and severity stages. Among the ventilation indices analyzed, FVC appeared to have the greatest impact on survival in the models (detailed data in, S1–S4 Models) with lung function as continuous values. The risk of death increase was 10.3% per one unit change of FVC z-score, compared to 8.1% and 8.2% for FEV₁ and TLC, respectively. However, the impact of FVC was significantly weaker than that of TLCO, which showed an increase of risk over 30% for a one unit change of the z-score in every model.

When lung function was expressed as a percentage of predicted values, the risk of death increased by 8% and 35% for every 10% decrease in FVC and TLCO, respectively. Due to these findings, only the results of FVC and TLCO were considered in further analyses, which aligns with commonly accepted clinical practice. For the FVC severity stages model, TLCO severity (expressed as z-score) was included as a confounder, and similarly, for the TLCO severity stages model, FVC severity expressed using z-score was included as a confounder.

As suggested by the Kaplan–Meier analyses, both models using TLCO to grade severity provided equivalent mortality prediction. However, the 2005 classification of mildly low FVC (between 70% predicted and the LLN) was not associated with a significant increase in mortality when compared to patients with ILD with a normal FVC (see Table 5). The older the patient and the male sex, the higher the risk of mortality. The presence of airway obstruction was not an independent mortality risk factor except borderline result in model with TLCO and FVC expressed as z-scores. In all models, diagnosis of ILD other than sarcoidosis strongly affected the prognosis.

Using the same models, we compared consistent and non-consistent levels of severity grading with “normal” FVC and TLCO as reference. Results are presented in Fig 3, left and right panel respectively.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Cox proportional hazards model for severity of FVC (left panel) and TLCO (right panel) levels.

Recategorized groups have the arrow in the middle of the name (ATS/ERS 2005 → ATS/ERS 2022). ATS – American Thoracic Society; ERS – European Respiratory Society; FVC – forced vital capacity; TLCO – lung transfer factor for carbon monoxide.

https://doi.org/10.1371/journal.pmed.1004619.g003

In both Cox regression models, age, male sex, all ILD diseases, and lung function (as continuous variables) were significant predictors of mortality (detailed data in S5 and S6 Models). In the model with FVC severity, survival was not significantly lower for patients classified as „mild” when compared to those with a normal FVC. There was a considerable overlap observed in the confidence intervals for the mild to moderate classes, with only the “severe → moderate” category significantly deviating from the rest and had HR bigger than “severe/severe”. In contrast, all TLCO severity classes were associated with increased risk of mortality. Both functional indices were significant predictors of mortality in Cox regression models when included as continuous variables (% pred. or z-score), however TLCO had much greater impact than FVC.

Discussion

Our study is the large-scale analysis touching on the problem of reclassifying pulmonary dysfunction in patients with interstitial lung diseases in the context of their prognostic value. The comprehensive nature of the study (simultaneous spirometry, plethysmography and TLCO testing) performed in a single laboratory by experienced and trained in clinical trial staff is also an advantage.

The results of the pulmonary function tests obtained in different diagnostic groups are consistent with previous observations in the literature regarding the prevalence of obstruction in interstitial diseases [19,20]. It is noteworthy that restrictive disorders were not as prevalent as indicated by the reduced FVC value in most patients, except for those with IPF, i-NSIP, and u-ILD. The relatively reduced FVC compared to preserved TLC may be attributed to the phenotype of combined pulmonary fibrosis and emphysema (CPFE) [21].

PFTs are also used as eligibility criteria in many clinical trials, and, in some countries, as criteria for receiving financial support from national health systems for treatment.

The 2022 update of ATS/ERS PFT interpretation guidelines recommended the use of z-scores to define the LLNs, and also suggested z-score thresholds for mild, moderate, and severe categories for results below the LLN [4]. However, the 2022 severity thresholds were not disease specific and not based on morbidity or mortality. For example, a TLCO z-score below −4.0 is far below the predicted value, but this “severe” abnormality category was again not based on morbidity (such as dyspnea on exertion) and not based on an incremental risk of subsequent mortality for a group of patients with ILD. The relationship between % predicted values, the lower limit of normal expressed as % predicted, and z-scores has been extensively discussed in the literature, particularly in the context of the development of modern lung function reference equations [13–15] and their recommended application in clinical practice [4]. The aim of our study, however, was not to revisit these well-established theoretical concepts, but rather to assess the practical implications of transitioning from % predicted thresholds to z-score-based thresholds in risk stratification and mortality prediction in patients with interstitial lung diseases.

Our study demonstrates that the PFT interpretation of severity will change for a substantial number of patients with an ILD if the 2022 guidelines are followed instead of “sticking with” the 2005 guidelines. A previous study demonstrated this phenomenon for patients with airway obstruction [7]. In large populations of healthy adults, the average values of both FVC and TLCO decline linearly with age [14,15]. However, the absolute difference between the predicted value and the lower limit of normal (LLN), e.g., expressed in liters for FVC, remains relatively stable across age groups. This means that the range of values considered “normal” (i.e., the spread between the predicted value and the LLN) does not widen substantially with aging in absolute terms. When expressed as a percentage of predicted, however, the LLN declines from approximately 80% in younger adults (up to around 40 years of age) to approximately 70% at 80 years of age [15]. This increasing spread in % predicted values with age reflects the natural physiological decline in lung function over time.

Because z-scores adjust for this age-related change, they provide a more standardized method for expressing pulmonary function relative to age-appropriate norms. This is the primary reason ATS/ERS guidelines recommend the use of z-scores rather than % predicted when assessing pulmonary function, particularly in populations spanning a wide age range.

The recategorization of spirometric indices (FEV₁ and FVC) resulted in a shift to a less severe category for the majority of recategorized patients, and this shift was associated with older age. The reason for this phenomenon is that the percent predicted lower limit of normal decreases with age which causes the same % predicted value to have a less negative z-score in older compared to younger subjects. The range of values considered normal (z-scores between −1.64 and +1.64) remains relatively constant across age groups when assessed in absolute values. Fig 1 effectively illustrates the phenomenon, demonstrating a considerable spread of results presented as z-scores for the same percentage value. For instance, an FVC of 50% predicted may correspond to a z-score of approximately −2.75 for an 80-year-old individual but as low as −4.2 for a 30-year-old individual. Conversely, a z-score −4 corresponds to approximately 50% of predicted in younger patients, but can be as low as 35% of predicted in older patients. Assessing severity based on z-score thresholds appear to align better with the clinical perspective. A 30-year-old with a 50% FVC value raises greater concerns compared to an 80-year-old with the same FVC percentage.

The relationship between % predicted and z-scores for TLCO follows a somewhat different pattern. The result cloud clearly has an arched shape (unlike in the case of FVC, where the dispersion is sheaf-shaped), although the distribution of cases by age is similar (younger individuals are shifted up/left). For this reason, the recategorization was for the younger ones, but in this case, sex also mattered (women were more frequently recategorized).

Our study confirms statements in current ILD guidelines: PFTs are an important element in the clinical assessment of these patients [18,22–24]. Both FVC and TLCO independently predict mortality in patients with IPF [1,2], however our study revealed that TLCO seems to be stronger predictor than FVC (35% versus 8% change in HR for every 10% difference in index and 30% versus 10% change for HR for every 1 z-score difference, respectively). We are aware that we did not include all other predictors of survival identified, for example, in IPF studies, such as 6-min walk distance, and desaturation during the 6-min walk test [25–27], but our intention was to show the phenomenon occurring in the entire group of diseases characterized by the potential of developing similar type of functional disorders (lung volume restriction and gas transfer disorders). It is also well known, that a change in functional parameters over time is a better predictor of survival than a single score in ILDs [28–31], but the main goal of this study is to test the hypothesis about how severity at first presentation is assessed and its relation to prognosis.

Kaplan–Meier survival analysis demonstrated a significant association between abnormally low lung function and all-cause mortality. Notably, our patients with ILD exhibited a distinct grouping of results consistent with the 2005 ATS/ERS severity categories for TLCO. However, caution must be exercised when drawing simplistic conclusions from these findings, as they may be misleading and lack full justification.

For instance, it may be tempting to conclude that patients with TLCO > 60% of the predicted value share a similar prognosis, regardless of whether their TLCO value expressed as a z-score falls above or below −2.5. Likewise, moderately low TLCO values (40% to 60% predicted) with a z-score below −4.0 (presently classified as “severe”), did not show a significant increase in mortality.

However, the aforementioned observations underwent notable changes when employing a multivariable Cox proportional hazards model. This model incorporated various factors, including age, sex, type of disease, and FVC expressed as a z-score. Utilizing univariate Kaplan–Meier survival analysis to compare severity scoring systems, particularly when one system retains age, sex, and size biases, can lead to erroneous conclusions.

S2 and S4 Models effectively illustrate this issue. When using percent of predicted values, the hazard ratio for lung function abnormality indicates minimal survival advantage for better function, and the hazards associated with sex and age are lower and significantly different from those observed using z-scores. Conversely, when assessing severity of function using z-scores, a stronger relationship with survival becomes apparent, with higher and more well-defined hazards for sex and age. This discrepancy arises because sex and age are encompassed within the percent predicted method, thus resulting in inherent biases. The authors are aware of other potential factors influencing the study results (e.g., comorbidities) that were not included in the models. It is also important to consider possible interactions between sex, age and diagnosis (IPF is known to affect older men more often, whereas sarcoidosis is a disease of a relatively younger population).

It is worth to note, that about 3/4th of patients had a normal FVC and about half had a normal TLCO, but most of them were diagnosed with sarcoidosis. Before the onset of their ILD, half of the patients had lung function above their predicted value. Change in lung function for an individual patient would be a much better predictor of subsequent morbidity and mortality [32], but very few patients have “baseline” PFTs measured as healthy young adults.

The presence of airway obstruction may be associated with a poor prognosis in some interstitial lung diseases [20] and therefore we have included the FEV₁/FVC ratio in the analyses. In patients with asthma or COPD, the presence of airflow limitation (a low FEV₁/FVC) and the percent predicted FEV₁ are important when estimating disease severity, but although 13.2% of our patients with an ILD also had airflow limitation, it was not an independent predictor of all-cause mortality. Male sex and age were independent predictors of mortality, consistent with previous observations in lung fibrotic diseases [1,2].

Our study analyzed previously collected data from a single center, and thus was retrospective. Some patients received treatment, in accordance with the indications and therapeutic recommendations applicable at the time. However, for the majority of IPF patients, antifibrotic treatment only became available in Poland in 2017, meaning that most patients with IPF included in our analysis were not treated with antifibrotic drugs.

During the study period, diagnostic standards for interstitial lung diseases, including IPF, did not undergo any major revisions that would have significantly affected diagnostic criteria or disease classification. Therefore, we believe that changes in diagnostic practice are unlikely to have introduced substantial time-dependent bias in our cohort.

Furthermore, due to the relatively low prevalence of interstitial lung diseases in the general population, there is a lack of high-quality evidence demonstrating a significant impact of therapeutic interventions (including antifibrotic therapy) on overall mortality in this patient population. As a result, although supportive care was available, there is no clear evidence that it systematically improved survival outcomes, particularly among patients who may have been misclassified based on absolute pulmonary function thresholds.

Additionally, our results may not be generalizable to younger patients or those from non-Caucasian populations with ILD. Our primary outcome was all-cause mortality, as data on specific causes of death (e.g., accidental death, suicide, or deaths due to comorbidities) were not available. Finally, the statistical models used in this study do not account for all potential confounding factors, which may introduce some degree of uncertainty into our estimates.

The implementation of newly recommended severity stages for lung function impairment utilizing z-scores results in the reclassification of a substantial number of patients with interstitial lung diseases. Age plays a major role in this reclassification, with older patients being shifted to less severe categories for spirometric indices, while younger patients tend to be classified into more severe categories for TLCO. The adoption of reclassification and the use of z-scores appear to be justified due to their stronger association with overall prognosis. Notably, severe impairment of TLCO emerges as a significantly stronger predictor of mortality in ILDs compared to FVC within the same category. This highlights the importance of considering TLCO as a valuable marker in assessing disease severity and prognosis. Given the above considerations, the reasonable message, in our opinion, is that perhaps future clinical trials should be conducted using z-score-based classification rather than % predicted thresholds.

Supporting information

Acknowledgments

We offer sincere thanks to our colleagues who provided medical care for these patients. We thank Paul Enright (retired from the University of Arizona) for his review and editing of the penultimate draft of our manuscript.