Reply to: Interpreting lung oscillometry results: Z-scores instead of fixed cut-off values?

Reply to Sundeep Salvi and co-workers:

We thank S. Salvi and co-workers for their interest in our original article [1], which analysed the characteristics and diagnostic performance of respiratory oscillometry in respiratory diseases. S. Salvi and co-workers proposed that fixed cut-off values, instead of Z-scores, should be used to differentiate between healthy and disease states.

Since lung function tests have been applied in diagnosing respiratory diseases in clinical practice, what kind of cut-offs to use to interpretate lung function results has always been a hot issue. Many studies have compared the diagnostic value of fixed cut-offs, per cent of the predicted value or z-scores of lung function indices in differentiating between healthy and disease states [2–4]. While fixed cut-offs or per cent of the predicted value are commonly used in clinical practice at present and in the past, Z-score was proposed to be used in interpreting lung function results in the past decade [5]. The Z-score of a result is the number of standardised residuals away from the predicted value a measurement is. As it takes into account the impact of anthropometric factors on variation, Z-scores can more accurately reflect how far an observed lung function value is from the predicted value; therefore, it is recommended to be used in interpreting lung function results in the latest American Thoracic Society/European Respiratory Society (ERS) technical standard on interpretive strategies for routine lung function tests [6] and the ERS technical standard on respiratory oscillometry [7].

S. Salvi and co-workers argued that Z-scores are not appropriate to be used in interpreting respiratory oscillometry results for the following reasons. First, oscillometric indices are not normally distributed; thus, the Z-score may not accurately reflect the probability of being within the healthy distribution. Secondly, although anthropometric factors (such as height and gender) have an impact on impulse oscillometry indices, their impact is not as large as that of spirometric indices (the highest determination coefficient of the predictive equation is ∼0.20). In this case, using fixed cut-offs for diagnosis may be a simple and feasible alternative.

For lung function indices, the Z-score is calculated by dividing the difference between the measured value and the predicted value by the residual standard deviation, which is the standard deviation of residuals of the regression equation. Our data on predictive equations of oscillometry we published previously [8] showed that, although the raw data of oscillometric indices were not normally distributed, the residuals of the regression equations were approximately normally distributed in raw or logarithmically transformed data. Under the condition that residuals of the regression equation are approximately normal distribution, we believe that Z-scores can still reflect the distribution probability of the measured value. Nevertheless, we agree with S. Salvi and co-workers that due to the low determination coefficients of predictive equations of oscillometry, Z-scores of oscillometric indices may present less efficacy when compared with Z-scores of spirometric indices. However, though the correction effect of Z-scores on anthropometric factors is relatively small in oscillometry, this correction effect still makes sense. In our prior analysis, we compared the diagnostic value of different kinds of cut-offs of oscillometric indices. As shown in figure 1, the R₅ z-score displays a superior diagnostic capacity than the absolute value of R₅ in diagnosing respiratory diseases (area under the curve (AUC) 0.81 versus 0.77, p<0.01) and obstructive airway diseases (AUC 0.79 versus 0.75, p<0.01). When we choose the best indicator for diagnosis, the accuracy of diagnosis is still the most important aspect to consider. Although absolute cut-off values are simple to use, they are not as accurate as Z-scores in diagnosis. The reason why some previous studies [9, 10] used absolute cut-off values rather than Z-scores of oscillometric indices in diagnosis is the lack of available adult reference values and Z-scores of oscillometry. Although the calculation of Z-scores is relatively complicated, it is believed that with the development of computer calculation and artificial intelligence interpretation, the interpretation of lung function results based on Z-scores is no longer a troublesome problem. Therefore, consistent with the recommendation of the current technical standard, we think that the Z-score is preferable in the interpretation of respiratory oscillometry results.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1

Comparisons of the diagnostic capacity between the actual value and Z-score of resistance measured at 5 Hz (R₅) in diagnosing respiratory diseases and obstructive airway diseases.

In conclusion, based on the current evidence, we believe that Z-score is still the best indicator to interpret the oscillometry results. With the inspiration of S. Salvi and co-workers, we will pay more attention to the comparisons between Z-scores and fixed cut-offs of oscillometric indices in disease diagnosis and assessment.