Re “Inhaled Corticosteroid Particle Size and Risk of Hospitalization Rue to Exacerbations and All-Cause Mortality in Patients With Chronic Obstructive Respiratory Disease. A Nationwide Cohort Study”. Heerfordt et al [Letter]

In response to the study titled "Inhaled Corticosteroid particle size and risk of hospitalization rue to exacerbations and all-cause mortality in patients with chronic obstructive respiratory disease.A Nationwide cohort study".Heerfordt et al. 1 Having previously reported that the use of extrafine particle ICS, compared to standard size preparations, is associated with reduced hospital admission for pneumonia Heerfordt el al now report that extrafine particle ICS do not reduce the risk of exacerbations or all-cause mortality in the same database. 1In a subgroup analysis of the more recent paper the authors suggest a potential benefit if the extra fine particles come from an MDI inhaler.We feel that the current paper has a number of limitations.The number of people using extrafine particle ICS in this dataset is low at 4.2%, and it seems likely to us that this represents a highly selected group of patients.Although the baseline demographic table (Table 1 in the Heertfordt paper), 1 suggests the patient groups are similar it seems likely to us that patients who received extrafine particle ICS inhalers may well have been moved to these treatments because of problems with standard inhalers.The impact of this selection bias would be to have a more severe group of patients in the extrafine group and to bias the results away from a beneficial impact of these drugs.There is no ideal way to circumvent this problem of selection bias, and the authors can be commended for trying several different approaches including prospensity score matching and multivariate adjusted models. 1 Both of these approaches moved the unadjusted models, which tended to show a protective effect of extrafine particle ICS, towards unity.In order to understand the associations more fully it would be informative to know which of the covariables were responsible for these changes by undertaking a series of bivariate models, and by inspecting the effect estimates as more variables are adjusted using the "changein-estimates" method. 2,3In simply adding all of the variables together at once we have lost an opportunity to explain these findings.It is not clear to us from the methods exactly when the exposure period was defined -and if exposures occurred after the study start date it is possible that issues with immortal time bias will also exist.As with all cohort studies a full description of the start and stop dates, and the period of exposure considered would make the paper more informative. 4ecause exposure to ICS particle type was defined based on highest cumulative exposure from the year prior to cohort entry, adjustment for covariates such as ICS equivalent dose from the same period may introduce overadjustment bias, whereby there is control for intermediate variables on a causal path from exposure to outcome. 5