Article type
Year
Abstract
Background: A review was performed under the Cochrane Airways group 'Pressurised metered dose inhalers versus all other hand-held inhaler devices to deliver corticosteroids for non-acute asthma ' (currently protocol on Cochrane Library). This shows no treatment effect between pressurised metered-dose inhaler (pMDI) (with or without spacer device) versus dry powder inhalers or CFC-free pMDI. There may be significant biases in the individual studies and subsequent meta-analysis leading to a type II error or failing to find a true treatment effect. These biases include the need for licensing purposes of new inhalers to show equivalence to existing devices in studies, recruitment of relatively well asthmatics, higher than the clinical 'normal' doses used and studies of a relatively short duration (four to twelve weeks) compared to the duration of action of inhaled corticosteroids. A method was sought to investigate and account for these factors without the need for advanced and proprietary statistical packages and techniques such as meta-regression. Objectives. To further substantiate the initial review findings for equivalence between different inhaler devices using readily accessible statistical analysis.
Methods: Factors that might be expected to influence the treatment effect, should there be one (duration, dose, severity, baseline difference, sample size) were investigated by simple correlation and multiple linear regression. Duration of the treatment period was tested as a linear, root and logarithmic function. Severity was tested as a categorical score (mild=1, mild to moderate=2, severe=3) as defined by the investigators and as an absolute measure of airway function (FEV1) at baseline. Sample size was tested as the square root, as a measure of study power. Baseline difference was calculated as the standardised mean difference (SMD) of the treatment groups in the same manner as treatment effect was calculated in the review. All were correlated with the SMD of treatment effect (based on absolute values rather than change from baseline) using a simple and widely available package (Microsoft Excel).
Results: For none of the factors was there any significant correlation except for baseline difference. Baseline difference was strongly correlated with treatment difference (effect), r=0.95, p=0.000001. The adjusted r squared was 0.87. Duration, whether absolute, logarithmic or square rooted, r=-0.34, p=0.12 to 0.18. Severity as FEV1 and grade was r=-0.08, p=0.77 and r=-0.09, p=0.80. Sample size (square rooted), r=-0.01, p=0.96
Conclusions: This validates the original review findings of equivalence by suggesting that nearly 90% of the treatment effects (the 'r squared' value)seen were due to differences at baseline and by the exclusion of other factors affecting treatment effect.
Methods: Factors that might be expected to influence the treatment effect, should there be one (duration, dose, severity, baseline difference, sample size) were investigated by simple correlation and multiple linear regression. Duration of the treatment period was tested as a linear, root and logarithmic function. Severity was tested as a categorical score (mild=1, mild to moderate=2, severe=3) as defined by the investigators and as an absolute measure of airway function (FEV1) at baseline. Sample size was tested as the square root, as a measure of study power. Baseline difference was calculated as the standardised mean difference (SMD) of the treatment groups in the same manner as treatment effect was calculated in the review. All were correlated with the SMD of treatment effect (based on absolute values rather than change from baseline) using a simple and widely available package (Microsoft Excel).
Results: For none of the factors was there any significant correlation except for baseline difference. Baseline difference was strongly correlated with treatment difference (effect), r=0.95, p=0.000001. The adjusted r squared was 0.87. Duration, whether absolute, logarithmic or square rooted, r=-0.34, p=0.12 to 0.18. Severity as FEV1 and grade was r=-0.08, p=0.77 and r=-0.09, p=0.80. Sample size (square rooted), r=-0.01, p=0.96
Conclusions: This validates the original review findings of equivalence by suggesting that nearly 90% of the treatment effects (the 'r squared' value)seen were due to differences at baseline and by the exclusion of other factors affecting treatment effect.