Longitudinal lung function in idiopathic NSIP: what are we missing?

In a recent article, PARK et al. [1] described the clinical course and changes in lung function of 83 patients classified as idiopathic nonspecific interstitial pneumonia (NSIP). PARK et al. [1] found that 81% of patients with fibrotic NSIP had improved or stable lung function over time after initial treatment. Specifically, the authors showed an increase in the percentage of patients with an improvement in forced vital capacity and diffusing capacity of the lung for carbon monoxide over time, as well as an increase in the magnitude of change in observed lung function. However, due to the lack of complete longitudinal data for all patients, such results must be interpreted with caution. Missing data is a problem that plagues all statistical analyses, particularly those involving longitudinal data from an observational study. When ''missingness'' of a variable depends on the value of that variable, this can lead to systematic bias [2]. In the case of the study by PARK et al. [1], those patients with worse lung function were more likely to die, thereby precluding any subsequent assessment of their lung function. This may explain, in part, the continued ''improvement'' in lung function observed over time, as these sicker patients drop out of the cohort. This effect has been widely described, both in medicine [3] and in other fields of study [4], and is sometimes referred to as ''survivor bias'' or ''survivorship bias''. An alternative data analysis approach to that presented by PARK et al. [1] would be to perform a ''complete case'' analysis, excluding those patients with incomplete lung function data. Although complete case analysis is not immune to bias [2], it may be helpful in testing the robustness of the reported findings. Another alternative would be to create a separate category for those patients with missing data, and then report the percentage of patients that were improved, stable, worsened and missing at each time-point with respect to the total number of patients at baseline. This would give a more conservative assessment of the clinical course of disease and illustrate the degree to which missing lung function data could influence the results. Assuming that deaths accounted for the majority of missing lung function data, it might also be reasonable to simply categorise the missing group together with those whose lung function worsened, resulting in a much lower percentage of subjects with improved lung function. Regardless of the analytical approach used, …