Of the need to reconcile discrepancies between two different reference equations for combined single-breath D_LNO–D_LCO in systemic sclerosis

To the Editor:

Systemic sclerosis (SSc) is a chronic disease of the connective tissue with an estimated prevalence of 5.8 per 100 000 in a multiethnic population living in Europe [1]. The disease mechanisms are highly complex, affecting multiple organs, including the pulmonary system. Severe pulmonary complications include SSc-associated interstitial lung disease (ILD) and pulmonary hypertension (PH), which are also associated with premature mortality [2, 3]. Early diagnosis and monitoring of pulmonary complications by spirometry, body plethysmography and pulmonary diffusing capacity are critical to lung health in SSc patients [4, 5]. Diffusing capacity of the lung for carbon monoxide (D_LCO) is a well-established and frequently applied test in SSc, and is used as a surrogate of SSc-ILD and SSc-PH progression [4, 6]. Recently, diffusing capacity of the lung for nitric oxide (D_LNO) has received growing attention in the diagnostic setup of SSc patients [7, 8]; yet, its clinical value with respect to disease management and prognosis has yet to be shown. In this respect, accurate interpretation of test results using robust reference equations [5] is imperative to both clinicians and researchers. While D_LNO is a promising method, uncertainty in the accuracy of available references equations limit its clinical implementation on a larger scale.

The measurement of D_LNO has recently been standardised by a European Respiratory Society (ERS) Task Force [9]. The Task Force provided the largest dataset (n=490, age range 18–93 years) of normal values for combined single-breath D_LNO–D_LCO tests for white, European and Northern American adults by pooling datasets from previously published studies [10–12]. Despite some technical and methodological inconsistencies between the different underlying studies [10–12], these reference equations are an important step towards the interpretation of D_LNO tests in clinical research and practice. Recently, new reference equations for the combined single-breath D_LNO–D_LCO were published for European adults [13], using ERS technical standard methodology [9] and rigorous quality control in a relatively large population (n=282) covering a broad age range of 18–97 years.

To date, the impact of different reference equations on the interpretation of combined single-breath D_LNO–D_LCO values in patient populations with pulmonary involvement is unknown. We therefore sought to compare the two most recently published reference equations for D_LNO–D_LCO [9, 13] using data from a large cohort of SSc patients with a broad range of disease severity. Those two equations were chosen as the Zavorsky et al. [9] equation is the official reference equation of the ERS Task Force and the Munkholm et al. [13] equation is the first equation that was built upon D_LNO measurements following ERS Task Force standard methodology [9].

We retrospectively analysed data from an unselected cohort of clinically stable SSc patients assessed at Hôpital Cochin, Université Paris Descartes, Paris, France, between February 2007 and October 2016. 337 patients (84% females) with a mean±sd age of 54.4±13.6 years, a body mass index of 23.7±4.4 kg·m⁻²_, a forced expiratory volume in 1 s of 86±18% pred and a forced vital capacity of 85±18% pred were included. Of those, 133 (39%) and 19 (6%) had a diagnosis of ILD and PH, respectively, and 14 (4%) had both ILD and PH. Combined single-breath D_LNO–D_LCO tests were performed using a commercially available device (HypAir; Medisoft, Dinant, Belgium) using a breath-hold time of 6 s. The mean of two technically satisfactory tests fulfilling intertest reliability criteria [9] was used for statistical analysis. To compare D_LNO and D_LCO values, we calculated z-scores using the Zavorsky et al. [9] and Munkholm et al. [13] equations.

The study complied with our institutional rules and the World Medical Association Declaration of Helsinki that deemed the study to be observational and therefore waived the need for informed consent. All functional tests were routinely performed for diagnostic and therapeutic purposes.

Our analysis revealed systematic differences (i.e. a proportional error) in D_LNO and D_LCO z-scores between the two equations [9, 13] (figure 1). Moreover, comparing both equations [9, 13], we noticed differential agreement of D_LNO and D_LCO z-scores between sexes, with much wider limits of agreement for females (figure 1b and d).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1

Scatter and Bland–Altman plots for comparisons between z-scores for a and b) diffusing capacity of the lung for nitric oxide (D_LNO) c and d) and carbon monoxide (D_LCO) between reference equations published by Zavorsky et al. [9] and Munkholm et al. [13]. Solid black circles represent males and open white circles represent females. The solid lines (females) and dashed lines (males) in b and d represent the mean bias and upper and lower limits of agreement (mean±1.96sd).

The magnitudes of the difference in D_LNO and D_LCO z-scores between the two reference equations [9, 13] are likely to be multifactorial, including disparities at the population level as well as technical and methodological discrepancies. However, both D_LNO and D_LCO were consistently different across the entire spectrum of z-scores, which points toward a systematic methodological difference. We were not able to compute z-scores for the reference equation by Aguilaniu et al. [12], because the authors did not report the residual standard deviation for their equation. Consequently, we could not compare z-scores from this equation with those from Munkholm et al. [13] to assess the relative contribution of this study in the reference equation from Zavorsky et al. [9]. Nevertheless, per cent predicted values for D_LNO and D_LCO were, on average, 30% and 24% lower comparing Aguilaniu et al. [12] with Munkholm et al. [13] equations (data not shown), and this study [12] contributed about half (54%) of the measurements to the pooled dataset from Zavorsky et al. [9]. This suggests that the differences are not primarily due to differences in equipment (e.g. different devices and nitric oxide analysers) and/or breath-hold times but partially due to inclusion of data from Aguilaniu et al. [12]. For their analysis, Aguilaniu et al. [12] chose the D_LNO value from the highest D_LCO test and not a mean value as the other authors did [10, 11, 13]. This apparently resulted in reference equations with the highest D_LNO predicted values [13]. However, to what extent the differences in per cent predicted values between the two equations impact on z-scores cannot be determined.

The use of different analytical approaches is another reason for the discrepancies between the two reference equations [9, 13]. Munkholm et al. [13] reported sex-specific equations with residual standard errors (RSEs) separately for males and females, whereas Zavorsky et al. [9] did not apply sex-specific RSEs in their equation. For example, Munkholm et al. [13] uses an RSE for D_LNO of 11.4 for females and 16.6 for males, while Zavorsky et al. [9] uses an RSE of 20.0 for both sexes. The smaller difference in standard errors between males results in different regression slopes between sexes (figure 1a and b) with an overall better agreement between D_LNO z-scores for male patients [9, 13]. The same observation applies for D_LCO. Since sex is an independent explanatory variable of D_LNO [9, 13], and a significant and independent predictor of D_LCO [14], sex-specific reference equations for combined D_LNO–D_LCO [10, 13] are appropriate.

In conclusion, our data confirm considerable differences in D_LNO and D_LCO z-scores between different references equations [9, 13] in a large cohort of SSc patients. In an attempt to improve the validity of reference equations, pooling of available datasets [9, 13] would significantly strengthen the robustness of D_LNO and D_LCO equations, and facilitate interpretation of pulmonary diffusing capacity measurements for both clinicians and researchers. Future work on D_LNO–D_LCO reference values is required to allow for calculation of sex-specific, precise lower and upper limits of normal based on a large population including people of different ethnic backgrounds, a prerequisite to evaluate ultimately and exploit the clinical potential of this promising technique.

• Received November 5, 2018.
• Accepted January 8, 2019.

• Copyright ©ERS 2019